diff options
Diffstat (limited to 'src/mesa/drivers/dri/radeon')
57 files changed, 12772 insertions, 5284 deletions
diff --git a/src/mesa/drivers/dri/radeon/Makefile b/src/mesa/drivers/dri/radeon/Makefile index f223b2d9228..1f286776b5f 100644 --- a/src/mesa/drivers/dri/radeon/Makefile +++ b/src/mesa/drivers/dri/radeon/Makefile @@ -4,31 +4,53 @@ TOP = ../../../../.. include $(TOP)/configs/current +CFLAGS += $(RADEON_CFLAGS) + LIBNAME = radeon_dri.so MINIGLX_SOURCES = server/radeon_dri.c +ifeq ($(RADEON_LDFLAGS),) +CS_SOURCES = radeon_cs_space_drm.c +endif + +RADEON_COMMON_SOURCES = \ + radeon_bo_legacy.c \ + radeon_common_context.c \ + radeon_common.c \ + radeon_cs_legacy.c \ + radeon_dma.c \ + radeon_debug.c \ + radeon_fbo.c \ + radeon_lock.c \ + radeon_mipmap_tree.c \ + radeon_queryobj.c \ + radeon_span.c \ + radeon_texture.c + DRIVER_SOURCES = \ radeon_context.c \ radeon_ioctl.c \ - radeon_lock.c \ radeon_screen.c \ radeon_state.c \ radeon_state_init.c \ radeon_tex.c \ - radeon_texmem.c \ radeon_texstate.c \ radeon_tcl.c \ radeon_swtcl.c \ - radeon_span.c \ radeon_maos.c \ - radeon_sanity.c + radeon_sanity.c \ + $(RADEON_COMMON_SOURCES) C_SOURCES = \ $(COMMON_SOURCES) \ - $(DRIVER_SOURCES) + $(DRIVER_SOURCES) \ + $(CS_SOURCES) + +DRIVER_DEFINES = -DRADEON_COMMON=0 \ + -Wall -DRIVER_DEFINES = -DRADEON_COMMON=0 +DRI_LIB_DEPS += $(RADEON_LDFLAGS) X86_SOURCES = diff --git a/src/mesa/drivers/dri/radeon/radeon_bo_drm.h b/src/mesa/drivers/dri/radeon/radeon_bo_drm.h new file mode 100644 index 00000000000..71413716333 --- /dev/null +++ b/src/mesa/drivers/dri/radeon/radeon_bo_drm.h @@ -0,0 +1,219 @@ +/* + * Copyright © 2008 Jérôme Glisse + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES + * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS + * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ +/* + * Authors: + * Jérôme Glisse <glisse@freedesktop.org> + */ +#ifndef RADEON_BO_H +#define RADEON_BO_H + +#include <stdio.h> +#include <stdint.h> +//#include "radeon_track.h" + +/* bo object */ +#define RADEON_BO_FLAGS_MACRO_TILE 1 +#define RADEON_BO_FLAGS_MICRO_TILE 2 + +struct radeon_bo_manager; + +struct radeon_bo { + uint32_t alignment; + uint32_t handle; + uint32_t size; + uint32_t domains; + uint32_t flags; + unsigned cref; +#ifdef RADEON_BO_TRACK + struct radeon_track *track; +#endif + void *ptr; + struct radeon_bo_manager *bom; + uint32_t space_accounted; +}; + +/* bo functions */ +struct radeon_bo_funcs { + struct radeon_bo *(*bo_open)(struct radeon_bo_manager *bom, + uint32_t handle, + uint32_t size, + uint32_t alignment, + uint32_t domains, + uint32_t flags); + void (*bo_ref)(struct radeon_bo *bo); + struct radeon_bo *(*bo_unref)(struct radeon_bo *bo); + int (*bo_map)(struct radeon_bo *bo, int write); + int (*bo_unmap)(struct radeon_bo *bo); + int (*bo_wait)(struct radeon_bo *bo); + int (*bo_is_static)(struct radeon_bo *bo); + int (*bo_set_tiling)(struct radeon_bo *bo, uint32_t tiling_flags, + uint32_t pitch); + int (*bo_get_tiling)(struct radeon_bo *bo, uint32_t *tiling_flags, + uint32_t *pitch); + int (*bo_is_busy)(struct radeon_bo *bo, uint32_t *domain); +}; + +struct radeon_bo_manager { + struct radeon_bo_funcs *funcs; + int fd; + +#ifdef RADEON_BO_TRACK + struct radeon_tracker tracker; +#endif +}; + +static inline void _radeon_bo_debug(struct radeon_bo *bo, + const char *op, + const char *file, + const char *func, + int line) +{ + fprintf(stderr, "%s %p 0x%08X 0x%08X 0x%08X [%s %s %d]\n", + op, bo, bo->handle, bo->size, bo->cref, file, func, line); +} + +static inline struct radeon_bo *_radeon_bo_open(struct radeon_bo_manager *bom, + uint32_t handle, + uint32_t size, + uint32_t alignment, + uint32_t domains, + uint32_t flags, + const char *file, + const char *func, + int line) +{ + struct radeon_bo *bo; + + bo = bom->funcs->bo_open(bom, handle, size, alignment, domains, flags); + +#ifdef RADEON_BO_TRACK + if (bo) { + bo->track = radeon_tracker_add_track(&bom->tracker, bo->handle); + radeon_track_add_event(bo->track, file, func, "open", line); + } +#endif + return bo; +} + +static inline void _radeon_bo_ref(struct radeon_bo *bo, + const char *file, + const char *func, + int line) +{ + bo->cref++; +#ifdef RADEON_BO_TRACK + radeon_track_add_event(bo->track, file, func, "ref", line); +#endif + bo->bom->funcs->bo_ref(bo); +} + +static inline struct radeon_bo *_radeon_bo_unref(struct radeon_bo *bo, + const char *file, + const char *func, + int line) +{ + bo->cref--; +#ifdef RADEON_BO_TRACK + radeon_track_add_event(bo->track, file, func, "unref", line); + if (bo->cref <= 0) { + radeon_tracker_remove_track(&bo->bom->tracker, bo->track); + bo->track = NULL; + } +#endif + return bo->bom->funcs->bo_unref(bo); +} + +static inline int _radeon_bo_map(struct radeon_bo *bo, + int write, + const char *file, + const char *func, + int line) +{ + return bo->bom->funcs->bo_map(bo, write); +} + +static inline int _radeon_bo_unmap(struct radeon_bo *bo, + const char *file, + const char *func, + int line) +{ + return bo->bom->funcs->bo_unmap(bo); +} + +static inline int _radeon_bo_wait(struct radeon_bo *bo, + const char *file, + const char *func, + int line) +{ + return bo->bom->funcs->bo_wait(bo); +} + +static inline int _radeon_bo_is_busy(struct radeon_bo *bo, + uint32_t *domain, + const char *file, + const char *func, + int line) +{ + return bo->bom->funcs->bo_is_busy(bo, domain); +} + +static inline int radeon_bo_set_tiling(struct radeon_bo *bo, + uint32_t tiling_flags, uint32_t pitch) +{ + return bo->bom->funcs->bo_set_tiling(bo, tiling_flags, pitch); +} + +static inline int radeon_bo_get_tiling(struct radeon_bo *bo, + uint32_t *tiling_flags, uint32_t *pitch) +{ + return bo->bom->funcs->bo_get_tiling(bo, tiling_flags, pitch); +} + +static inline int radeon_bo_is_static(struct radeon_bo *bo) +{ + if (bo->bom->funcs->bo_is_static) + return bo->bom->funcs->bo_is_static(bo); + return 0; +} + +#define radeon_bo_open(bom, h, s, a, d, f)\ + _radeon_bo_open(bom, h, s, a, d, f, __FILE__, __FUNCTION__, __LINE__) +#define radeon_bo_ref(bo)\ + _radeon_bo_ref(bo, __FILE__, __FUNCTION__, __LINE__) +#define radeon_bo_unref(bo)\ + _radeon_bo_unref(bo, __FILE__, __FUNCTION__, __LINE__) +#define radeon_bo_map(bo, w)\ + _radeon_bo_map(bo, w, __FILE__, __FUNCTION__, __LINE__) +#define radeon_bo_unmap(bo)\ + _radeon_bo_unmap(bo, __FILE__, __FUNCTION__, __LINE__) +#define radeon_bo_debug(bo, opcode)\ + _radeon_bo_debug(bo, opcode, __FILE__, __FUNCTION__, __LINE__) +#define radeon_bo_wait(bo) \ + _radeon_bo_wait(bo, __FILE__, __func__, __LINE__) +#define radeon_bo_is_busy(bo, domain) \ + _radeon_bo_is_busy(bo, domain, __FILE__, __func__, __LINE__) + +#endif diff --git a/src/mesa/drivers/dri/radeon/radeon_bo_legacy.c b/src/mesa/drivers/dri/radeon/radeon_bo_legacy.c new file mode 100644 index 00000000000..3e7547d2f9d --- /dev/null +++ b/src/mesa/drivers/dri/radeon/radeon_bo_legacy.c @@ -0,0 +1,926 @@ +/* + * Copyright © 2008 Nicolai Haehnle + * Copyright © 2008 Dave Airlie + * Copyright © 2008 Jérôme Glisse + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ +/* + * Authors: + * Aapo Tahkola <aet@rasterburn.org> + * Nicolai Haehnle <prefect_@gmx.net> + * Dave Airlie + * Jérôme Glisse <glisse@freedesktop.org> + */ +#include <stdio.h> +#include <stddef.h> +#include <stdint.h> +#include <stdlib.h> +#include <string.h> +#include <errno.h> +#include <unistd.h> +#include <sys/mman.h> +#include <sys/ioctl.h> +#include "xf86drm.h" +#include "texmem.h" +#include "main/simple_list.h" + +#include "drm.h" +#include "radeon_drm.h" +#include "radeon_common.h" +#include "radeon_bocs_wrapper.h" +#include "radeon_macros.h" + +/* no seriously texmem.c is this screwed up */ +struct bo_legacy_texture_object { + driTextureObject base; + struct bo_legacy *parent; +}; + +struct bo_legacy { + struct radeon_bo base; + int map_count; + uint32_t pending; + int is_pending; + int static_bo; + uint32_t offset; + struct bo_legacy_texture_object *tobj; + int validated; + int dirty; + void *ptr; + struct bo_legacy *next, *prev; + struct bo_legacy *pnext, *pprev; +}; + +struct bo_manager_legacy { + struct radeon_bo_manager base; + unsigned nhandle; + unsigned nfree_handles; + unsigned cfree_handles; + uint32_t current_age; + struct bo_legacy bos; + struct bo_legacy pending_bos; + uint32_t fb_location; + uint32_t texture_offset; + unsigned dma_alloc_size; + uint32_t dma_buf_count; + unsigned cpendings; + driTextureObject texture_swapped; + driTexHeap *texture_heap; + struct radeon_screen *screen; + unsigned *free_handles; +}; + +static void bo_legacy_tobj_destroy(void *data, driTextureObject *t) +{ + struct bo_legacy_texture_object *tobj = (struct bo_legacy_texture_object *)t; + + if (tobj->parent) { + tobj->parent->tobj = NULL; + tobj->parent->validated = 0; + } +} + +static void inline clean_handles(struct bo_manager_legacy *bom) +{ + while (bom->cfree_handles > 0 && + !bom->free_handles[bom->cfree_handles - 1]) + bom->cfree_handles--; + +} +static int legacy_new_handle(struct bo_manager_legacy *bom, uint32_t *handle) +{ + uint32_t tmp; + + *handle = 0; + if (bom->nhandle == 0xFFFFFFFF) { + return -EINVAL; + } + if (bom->cfree_handles > 0) { + tmp = bom->free_handles[--bom->cfree_handles]; + clean_handles(bom); + } else { + bom->cfree_handles = 0; + tmp = bom->nhandle++; + } + assert(tmp); + *handle = tmp; + return 0; +} + +static int legacy_free_handle(struct bo_manager_legacy *bom, uint32_t handle) +{ + uint32_t *handles; + + if (!handle) { + return 0; + } + if (handle == (bom->nhandle - 1)) { + int i; + + bom->nhandle--; + for (i = bom->cfree_handles - 1; i >= 0; i--) { + if (bom->free_handles[i] == (bom->nhandle - 1)) { + bom->nhandle--; + bom->free_handles[i] = 0; + } + } + clean_handles(bom); + return 0; + } + if (bom->cfree_handles < bom->nfree_handles) { + bom->free_handles[bom->cfree_handles++] = handle; + return 0; + } + bom->nfree_handles += 0x100; + handles = (uint32_t*)realloc(bom->free_handles, bom->nfree_handles * 4); + if (handles == NULL) { + bom->nfree_handles -= 0x100; + return -ENOMEM; + } + bom->free_handles = handles; + bom->free_handles[bom->cfree_handles++] = handle; + return 0; +} + +static void legacy_get_current_age(struct bo_manager_legacy *boml) +{ + drm_radeon_getparam_t gp; + unsigned char *RADEONMMIO = NULL; + int r; + + if ( IS_R300_CLASS(boml->screen) + || IS_R600_CLASS(boml->screen) ) + { + gp.param = RADEON_PARAM_LAST_CLEAR; + gp.value = (int *)&boml->current_age; + r = drmCommandWriteRead(boml->base.fd, DRM_RADEON_GETPARAM, + &gp, sizeof(gp)); + if (r) { + fprintf(stderr, "%s: drmRadeonGetParam: %d\n", __FUNCTION__, r); + exit(1); + } + } + else { + RADEONMMIO = boml->screen->mmio.map; + boml->current_age = boml->screen->scratch[3]; + boml->current_age = INREG(RADEON_GUI_SCRATCH_REG3); + } +} + +static int legacy_is_pending(struct radeon_bo *bo) +{ + struct bo_manager_legacy *boml = (struct bo_manager_legacy *)bo->bom; + struct bo_legacy *bo_legacy = (struct bo_legacy*)bo; + + if (bo_legacy->is_pending <= 0) { + bo_legacy->is_pending = 0; + return 0; + } + if (boml->current_age >= bo_legacy->pending) { + if (boml->pending_bos.pprev == bo_legacy) { + boml->pending_bos.pprev = bo_legacy->pprev; + } + bo_legacy->pprev->pnext = bo_legacy->pnext; + if (bo_legacy->pnext) { + bo_legacy->pnext->pprev = bo_legacy->pprev; + } + assert(bo_legacy->is_pending <= bo->cref); + while (bo_legacy->is_pending--) { + bo = radeon_bo_unref(bo); + if (!bo) + break; + } + if (bo) + bo_legacy->is_pending = 0; + boml->cpendings--; + return 0; + } + return 1; +} + +static int legacy_wait_pending(struct radeon_bo *bo) +{ + struct bo_manager_legacy *boml = (struct bo_manager_legacy *)bo->bom; + struct bo_legacy *bo_legacy = (struct bo_legacy*)bo; + + if (!bo_legacy->is_pending) { + return 0; + } + /* FIXME: lockup and userspace busy looping that's all the folks */ + legacy_get_current_age(boml); + while (legacy_is_pending(bo)) { + usleep(10); + legacy_get_current_age(boml); + } + return 0; +} + +void legacy_track_pending(struct radeon_bo_manager *bom, int debug) +{ + struct bo_manager_legacy *boml = (struct bo_manager_legacy*) bom; + struct bo_legacy *bo_legacy; + struct bo_legacy *next; + + legacy_get_current_age(boml); + bo_legacy = boml->pending_bos.pnext; + while (bo_legacy) { + if (debug) + fprintf(stderr,"pending %p %d %d %d\n", bo_legacy, bo_legacy->base.size, + boml->current_age, bo_legacy->pending); + next = bo_legacy->pnext; + if (legacy_is_pending(&(bo_legacy->base))) { + } + bo_legacy = next; + } +} + +static int legacy_wait_any_pending(struct bo_manager_legacy *boml) +{ + struct bo_legacy *bo_legacy; + + legacy_get_current_age(boml); + bo_legacy = boml->pending_bos.pnext; + if (!bo_legacy) + return -1; + legacy_wait_pending(&bo_legacy->base); + return 0; +} + +static void legacy_kick_all_buffers(struct bo_manager_legacy *boml) +{ + struct bo_legacy *legacy; + + legacy = boml->bos.next; + while (legacy != &boml->bos) { + if (legacy->tobj) { + if (legacy->validated) { + driDestroyTextureObject(&legacy->tobj->base); + legacy->tobj = 0; + legacy->validated = 0; + } + } + legacy = legacy->next; + } +} + +static struct bo_legacy *bo_allocate(struct bo_manager_legacy *boml, + uint32_t size, + uint32_t alignment, + uint32_t domains, + uint32_t flags) +{ + struct bo_legacy *bo_legacy; + static int pgsize; + + if (pgsize == 0) + pgsize = getpagesize() - 1; + + size = (size + pgsize) & ~pgsize; + + bo_legacy = (struct bo_legacy*)calloc(1, sizeof(struct bo_legacy)); + if (bo_legacy == NULL) { + return NULL; + } + bo_legacy->base.bom = (struct radeon_bo_manager*)boml; + bo_legacy->base.handle = 0; + bo_legacy->base.size = size; + bo_legacy->base.alignment = alignment; + bo_legacy->base.domains = domains; + bo_legacy->base.flags = flags; + bo_legacy->base.ptr = NULL; + bo_legacy->map_count = 0; + bo_legacy->next = NULL; + bo_legacy->prev = NULL; + bo_legacy->pnext = NULL; + bo_legacy->pprev = NULL; + bo_legacy->next = boml->bos.next; + bo_legacy->prev = &boml->bos; + boml->bos.next = bo_legacy; + if (bo_legacy->next) { + bo_legacy->next->prev = bo_legacy; + } + + return bo_legacy; +} + +static int bo_dma_alloc(struct radeon_bo *bo) +{ + struct bo_manager_legacy *boml = (struct bo_manager_legacy *)bo->bom; + struct bo_legacy *bo_legacy = (struct bo_legacy*)bo; + drm_radeon_mem_alloc_t alloc; + unsigned size; + int base_offset; + int r; + + /* align size on 4Kb */ + size = (((4 * 1024) - 1) + bo->size) & ~((4 * 1024) - 1); + alloc.region = RADEON_MEM_REGION_GART; + alloc.alignment = bo_legacy->base.alignment; + alloc.size = size; + alloc.region_offset = &base_offset; + r = drmCommandWriteRead(bo->bom->fd, + DRM_RADEON_ALLOC, + &alloc, + sizeof(alloc)); + if (r) { + /* ptr is set to NULL if dma allocation failed */ + bo_legacy->ptr = NULL; + return r; + } + bo_legacy->ptr = boml->screen->gartTextures.map + base_offset; + bo_legacy->offset = boml->screen->gart_texture_offset + base_offset; + bo->size = size; + boml->dma_alloc_size += size; + boml->dma_buf_count++; + return 0; +} + +static int bo_dma_free(struct radeon_bo *bo) +{ + struct bo_manager_legacy *boml = (struct bo_manager_legacy *)bo->bom; + struct bo_legacy *bo_legacy = (struct bo_legacy*)bo; + drm_radeon_mem_free_t memfree; + int r; + + if (bo_legacy->ptr == NULL) { + /* ptr is set to NULL if dma allocation failed */ + return 0; + } + legacy_get_current_age(boml); + memfree.region = RADEON_MEM_REGION_GART; + memfree.region_offset = bo_legacy->offset; + memfree.region_offset -= boml->screen->gart_texture_offset; + r = drmCommandWrite(boml->base.fd, + DRM_RADEON_FREE, + &memfree, + sizeof(memfree)); + if (r) { + fprintf(stderr, "Failed to free bo[%p] at %08x\n", + &bo_legacy->base, memfree.region_offset); + fprintf(stderr, "ret = %s\n", strerror(-r)); + return r; + } + boml->dma_alloc_size -= bo_legacy->base.size; + boml->dma_buf_count--; + return 0; +} + +static void bo_free(struct bo_legacy *bo_legacy) +{ + struct bo_manager_legacy *boml; + + if (bo_legacy == NULL) { + return; + } + boml = (struct bo_manager_legacy *)bo_legacy->base.bom; + bo_legacy->prev->next = bo_legacy->next; + if (bo_legacy->next) { + bo_legacy->next->prev = bo_legacy->prev; + } + if (!bo_legacy->static_bo) { + legacy_free_handle(boml, bo_legacy->base.handle); + if (bo_legacy->base.domains & RADEON_GEM_DOMAIN_GTT) { + /* dma buffers */ + bo_dma_free(&bo_legacy->base); + } else { + driDestroyTextureObject(&bo_legacy->tobj->base); + bo_legacy->tobj = NULL; + /* free backing store */ + free(bo_legacy->ptr); + } + } + memset(bo_legacy, 0 , sizeof(struct bo_legacy)); + free(bo_legacy); +} + +static struct radeon_bo *bo_open(struct radeon_bo_manager *bom, + uint32_t handle, + uint32_t size, + uint32_t alignment, + uint32_t domains, + uint32_t flags) +{ + struct bo_manager_legacy *boml = (struct bo_manager_legacy *)bom; + struct bo_legacy *bo_legacy; + int r; + + if (handle) { + bo_legacy = boml->bos.next; + while (bo_legacy) { + if (bo_legacy->base.handle == handle) { + radeon_bo_ref(&(bo_legacy->base)); + return (struct radeon_bo*)bo_legacy; + } + bo_legacy = bo_legacy->next; + } + return NULL; + } + bo_legacy = bo_allocate(boml, size, alignment, domains, flags); + bo_legacy->static_bo = 0; + r = legacy_new_handle(boml, &bo_legacy->base.handle); + if (r) { + bo_free(bo_legacy); + return NULL; + } + if (bo_legacy->base.domains & RADEON_GEM_DOMAIN_GTT) + { +retry: + legacy_track_pending(&boml->base, 0); + /* dma buffers */ + + r = bo_dma_alloc(&(bo_legacy->base)); + if (r) + { + if (legacy_wait_any_pending(boml) == -1) + { + bo_free(bo_legacy); + return NULL; + } + goto retry; + return NULL; + } + } + else + { + bo_legacy->ptr = malloc(bo_legacy->base.size); + if (bo_legacy->ptr == NULL) { + bo_free(bo_legacy); + return NULL; + } + } + radeon_bo_ref(&(bo_legacy->base)); + + return (struct radeon_bo*)bo_legacy; +} + +static void bo_ref(struct radeon_bo *bo) +{ +} + +static struct radeon_bo *bo_unref(struct radeon_bo *bo) +{ + struct bo_legacy *bo_legacy = (struct bo_legacy*)bo; + + if (bo->cref <= 0) { + bo_legacy->prev->next = bo_legacy->next; + if (bo_legacy->next) { + bo_legacy->next->prev = bo_legacy->prev; + } + if (!bo_legacy->is_pending) { + bo_free(bo_legacy); + } + return NULL; + } + return bo; +} + +static int bo_map(struct radeon_bo *bo, int write) +{ + struct bo_manager_legacy *boml = (struct bo_manager_legacy *)bo->bom; + struct bo_legacy *bo_legacy = (struct bo_legacy*)bo; + + legacy_wait_pending(bo); + bo_legacy->validated = 0; + bo_legacy->dirty = 1; + bo_legacy->map_count++; + bo->ptr = bo_legacy->ptr; + /* Read the first pixel in the frame buffer. This should + * be a noop, right? In fact without this conform fails as reading + * from the framebuffer sometimes produces old results -- the + * on-card read cache gets mixed up and doesn't notice that the + * framebuffer has been updated. + * + * Note that we should probably be reading some otherwise unused + * region of VRAM, otherwise we might get incorrect results when + * reading pixels from the top left of the screen. + * + * I found this problem on an R420 with glean's texCube test. + * Note that the R200 span code also *writes* the first pixel in the + * framebuffer, but I've found this to be unnecessary. + * -- Nicolai Hähnle, June 2008 + */ + if (!(bo->domains & RADEON_GEM_DOMAIN_GTT)) { + int p; + volatile int *buf = (int*)boml->screen->driScreen->pFB; + p = *buf; + } + + return 0; +} + +static int bo_unmap(struct radeon_bo *bo) +{ + struct bo_legacy *bo_legacy = (struct bo_legacy*)bo; + + if (--bo_legacy->map_count > 0) + { + return 0; + } + + bo->ptr = NULL; + + return 0; +} + +static int bo_is_busy(struct radeon_bo *bo, uint32_t *domain) +{ + *domain = 0; + if (bo->domains & RADEON_GEM_DOMAIN_GTT) + *domain = RADEON_GEM_DOMAIN_GTT; + else + *domain = RADEON_GEM_DOMAIN_CPU; + if (legacy_is_pending(bo)) + return -EBUSY; + else + return 0; +} + +static int bo_is_static(struct radeon_bo *bo) +{ + struct bo_legacy *bo_legacy = (struct bo_legacy*)bo; + return bo_legacy->static_bo; +} + +static struct radeon_bo_funcs bo_legacy_funcs = { + bo_open, + bo_ref, + bo_unref, + bo_map, + bo_unmap, + NULL, + bo_is_static, + NULL, + NULL, + bo_is_busy +}; + +static int bo_vram_validate(struct radeon_bo *bo, + uint32_t *soffset, + uint32_t *eoffset) +{ + struct bo_manager_legacy *boml = (struct bo_manager_legacy *)bo->bom; + struct bo_legacy *bo_legacy = (struct bo_legacy*)bo; + int r; + int retry_count = 0, pending_retry = 0; + + if (!bo_legacy->tobj) { + bo_legacy->tobj = CALLOC(sizeof(struct bo_legacy_texture_object)); + bo_legacy->tobj->parent = bo_legacy; + make_empty_list(&bo_legacy->tobj->base); + bo_legacy->tobj->base.totalSize = bo->size; + retry: + r = driAllocateTexture(&boml->texture_heap, 1, + &bo_legacy->tobj->base); + if (r) { + pending_retry = 0; + while(boml->cpendings && pending_retry++ < 10000) { + legacy_track_pending(&boml->base, 0); + retry_count++; + if (retry_count > 2) { + free(bo_legacy->tobj); + bo_legacy->tobj = NULL; + fprintf(stderr, "Ouch! vram_validate failed %d\n", r); + return -1; + } + goto retry; + } + } + bo_legacy->offset = boml->texture_offset + + bo_legacy->tobj->base.memBlock->ofs; + bo_legacy->dirty = 1; + } + + assert(bo_legacy->tobj->base.memBlock); + + if (bo_legacy->tobj) + driUpdateTextureLRU(&bo_legacy->tobj->base); + + if (bo_legacy->dirty || bo_legacy->tobj->base.dirty_images[0]) { + if (IS_R600_CLASS(boml->screen)) { + drm_radeon_texture_t tex; + drm_radeon_tex_image_t tmp; + int ret; + + tex.offset = bo_legacy->offset; + tex.image = &tmp; + assert(!(tex.offset & 1023)); + + tmp.x = 0; + tmp.y = 0; + tmp.width = bo->size; + tmp.height = 1; + tmp.data = bo_legacy->ptr; + tex.format = RADEON_TXFORMAT_ARGB8888; + tex.width = tmp.width; + tex.height = tmp.height; + tex.pitch = bo->size; + do { + ret = drmCommandWriteRead(bo->bom->fd, + DRM_RADEON_TEXTURE, + &tex, + sizeof(drm_radeon_texture_t)); + if (ret) { + if (RADEON_DEBUG & RADEON_IOCTL) + fprintf(stderr, "DRM_RADEON_TEXTURE: again!\n"); + usleep(1); + } + } while (ret == -EAGAIN); + } else { + /* Copy to VRAM using a blit. + * All memory is 4K aligned. We're using 1024 pixels wide blits. + */ + drm_radeon_texture_t tex; + drm_radeon_tex_image_t tmp; + int ret; + + tex.offset = bo_legacy->offset; + tex.image = &tmp; + assert(!(tex.offset & 1023)); + + tmp.x = 0; + tmp.y = 0; + if (bo->size < 4096) { + tmp.width = (bo->size + 3) / 4; + tmp.height = 1; + } else { + tmp.width = 1024; + tmp.height = (bo->size + 4095) / 4096; + } + tmp.data = bo_legacy->ptr; + tex.format = RADEON_TXFORMAT_ARGB8888; + tex.width = tmp.width; + tex.height = tmp.height; + tex.pitch = MAX2(tmp.width / 16, 1); + do { + ret = drmCommandWriteRead(bo->bom->fd, + DRM_RADEON_TEXTURE, + &tex, + sizeof(drm_radeon_texture_t)); + if (ret) { + if (RADEON_DEBUG & RADEON_IOCTL) + fprintf(stderr, "DRM_RADEON_TEXTURE: again!\n"); + usleep(1); + } + } while (ret == -EAGAIN); + } + bo_legacy->dirty = 0; + bo_legacy->tobj->base.dirty_images[0] = 0; + } + return 0; +} + +/* + * radeon_bo_legacy_validate - + * returns: + * 0 - all good + * -EINVAL - mapped buffer can't be validated + * -EAGAIN - restart validation we've kicked all the buffers out + */ +int radeon_bo_legacy_validate(struct radeon_bo *bo, + uint32_t *soffset, + uint32_t *eoffset) +{ + struct bo_manager_legacy *boml = (struct bo_manager_legacy *)bo->bom; + struct bo_legacy *bo_legacy = (struct bo_legacy*)bo; + int r; + int retries = 0; + + if (bo_legacy->map_count) { + fprintf(stderr, "bo(%p, %d) is mapped (%d) can't valide it.\n", + bo, bo->size, bo_legacy->map_count); + return -EINVAL; + } + if (bo_legacy->static_bo || bo_legacy->validated) { + *soffset = bo_legacy->offset; + *eoffset = bo_legacy->offset + bo->size; + + return 0; + } + if (!(bo->domains & RADEON_GEM_DOMAIN_GTT)) { + + r = bo_vram_validate(bo, soffset, eoffset); + if (r) { + legacy_track_pending(&boml->base, 0); + legacy_kick_all_buffers(boml); + retries++; + if (retries == 2) { + fprintf(stderr,"legacy bo: failed to get relocations into aperture\n"); + assert(0); + exit(-1); + } + return -EAGAIN; + } + } + *soffset = bo_legacy->offset; + *eoffset = bo_legacy->offset + bo->size; + bo_legacy->validated = 1; + + return 0; +} + +void radeon_bo_legacy_pending(struct radeon_bo *bo, uint32_t pending) +{ + struct bo_manager_legacy *boml = (struct bo_manager_legacy *)bo->bom; + struct bo_legacy *bo_legacy = (struct bo_legacy*)bo; + + bo_legacy->pending = pending; + bo_legacy->is_pending++; + /* add to pending list */ + radeon_bo_ref(bo); + if (bo_legacy->is_pending > 1) { + return; + } + bo_legacy->pprev = boml->pending_bos.pprev; + bo_legacy->pnext = NULL; + bo_legacy->pprev->pnext = bo_legacy; + boml->pending_bos.pprev = bo_legacy; + boml->cpendings++; +} + +void radeon_bo_manager_legacy_dtor(struct radeon_bo_manager *bom) +{ + struct bo_manager_legacy *boml = (struct bo_manager_legacy *)bom; + struct bo_legacy *bo_legacy; + + if (bom == NULL) { + return; + } + bo_legacy = boml->bos.next; + while (bo_legacy) { + struct bo_legacy *next; + + next = bo_legacy->next; + bo_free(bo_legacy); + bo_legacy = next; + } + driDestroyTextureHeap(boml->texture_heap); + free(boml->free_handles); + free(boml); +} + +static struct bo_legacy *radeon_legacy_bo_alloc_static(struct bo_manager_legacy *bom, + int size, + uint32_t offset) +{ + struct bo_legacy *bo; + + bo = bo_allocate(bom, size, 0, RADEON_GEM_DOMAIN_VRAM, 0); + + if (bo == NULL) + return NULL; + bo->static_bo = 1; + bo->offset = offset + bom->fb_location; + bo->base.handle = bo->offset; + bo->ptr = bom->screen->driScreen->pFB + offset; + if (bo->base.handle > bom->nhandle) { + bom->nhandle = bo->base.handle + 1; + } + radeon_bo_ref(&(bo->base)); + return bo; +} + +struct radeon_bo_manager *radeon_bo_manager_legacy_ctor(struct radeon_screen *scrn) +{ + struct bo_manager_legacy *bom; + struct bo_legacy *bo; + unsigned size; + + bom = (struct bo_manager_legacy*) + calloc(1, sizeof(struct bo_manager_legacy)); + if (bom == NULL) { + return NULL; + } + + make_empty_list(&bom->texture_swapped); + + bom->texture_heap = driCreateTextureHeap(0, + bom, + scrn->texSize[0], + 12, + RADEON_NR_TEX_REGIONS, + (drmTextureRegionPtr)scrn->sarea->tex_list[0], + &scrn->sarea->tex_age[0], + &bom->texture_swapped, + sizeof(struct bo_legacy_texture_object), + &bo_legacy_tobj_destroy); + bom->texture_offset = scrn->texOffset[0]; + + bom->base.funcs = &bo_legacy_funcs; + bom->base.fd = scrn->driScreen->fd; + bom->bos.next = NULL; + bom->bos.prev = NULL; + bom->pending_bos.pprev = &bom->pending_bos; + bom->pending_bos.pnext = NULL; + bom->screen = scrn; + bom->fb_location = scrn->fbLocation; + bom->nhandle = 1; + bom->cfree_handles = 0; + bom->nfree_handles = 0x400; + bom->free_handles = (uint32_t*)malloc(bom->nfree_handles * 4); + if (bom->free_handles == NULL) { + radeon_bo_manager_legacy_dtor((struct radeon_bo_manager*)bom); + return NULL; + } + + /* biggest framebuffer size */ + size = 4096*4096*4; + + /* allocate front */ + bo = radeon_legacy_bo_alloc_static(bom, size, bom->screen->frontOffset); + + if (!bo) { + radeon_bo_manager_legacy_dtor((struct radeon_bo_manager*)bom); + return NULL; + } + if (scrn->sarea->tiling_enabled) { + bo->base.flags = RADEON_BO_FLAGS_MACRO_TILE; + } + + /* allocate back */ + bo = radeon_legacy_bo_alloc_static(bom, size, bom->screen->backOffset); + + if (!bo) { + radeon_bo_manager_legacy_dtor((struct radeon_bo_manager*)bom); + return NULL; + } + if (scrn->sarea->tiling_enabled) { + bo->base.flags = RADEON_BO_FLAGS_MACRO_TILE; + } + + /* allocate depth */ + bo = radeon_legacy_bo_alloc_static(bom, size, bom->screen->depthOffset); + + if (!bo) { + radeon_bo_manager_legacy_dtor((struct radeon_bo_manager*)bom); + return NULL; + } + bo->base.flags = 0; + if (scrn->sarea->tiling_enabled) { + bo->base.flags |= RADEON_BO_FLAGS_MACRO_TILE; + bo->base.flags |= RADEON_BO_FLAGS_MICRO_TILE; + } + return (struct radeon_bo_manager*)bom; +} + +void radeon_bo_legacy_texture_age(struct radeon_bo_manager *bom) +{ + struct bo_manager_legacy *boml = (struct bo_manager_legacy *)bom; + DRI_AGE_TEXTURES(boml->texture_heap); +} + +unsigned radeon_bo_legacy_relocs_size(struct radeon_bo *bo) +{ + struct bo_legacy *bo_legacy = (struct bo_legacy*)bo; + + if (bo_legacy->static_bo || (bo->domains & RADEON_GEM_DOMAIN_GTT)) { + return 0; + } + return bo->size; +} + +/* + * Fake up a bo for things like texture image_override. + * bo->offset already includes fb_location + */ +struct radeon_bo *radeon_legacy_bo_alloc_fake(struct radeon_bo_manager *bom, + int size, + uint32_t offset) +{ + struct bo_manager_legacy *boml = (struct bo_manager_legacy *)bom; + struct bo_legacy *bo; + + bo = bo_allocate(boml, size, 0, RADEON_GEM_DOMAIN_VRAM, 0); + + if (bo == NULL) + return NULL; + bo->static_bo = 1; + bo->offset = offset; + bo->base.handle = bo->offset; + bo->ptr = boml->screen->driScreen->pFB + (offset - boml->fb_location); + if (bo->base.handle > boml->nhandle) { + boml->nhandle = bo->base.handle + 1; + } + radeon_bo_ref(&(bo->base)); + return &(bo->base); +} + diff --git a/src/mesa/drivers/dri/radeon/radeon_bo_legacy.h b/src/mesa/drivers/dri/radeon/radeon_bo_legacy.h new file mode 100644 index 00000000000..2cf15dfaff3 --- /dev/null +++ b/src/mesa/drivers/dri/radeon/radeon_bo_legacy.h @@ -0,0 +1,50 @@ +/* + * Copyright © 2008 Nicolai Haehnle + * Copyright © 2008 Jérôme Glisse + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ +/* + * Authors: + * Aapo Tahkola <aet@rasterburn.org> + * Nicolai Haehnle <prefect_@gmx.net> + * Jérôme Glisse <glisse@freedesktop.org> + */ +#ifndef RADEON_BO_LEGACY_H +#define RADEON_BO_LEGACY_H + +#include "radeon_screen.h" + +void radeon_bo_legacy_pending(struct radeon_bo *bo, uint32_t pending); +int radeon_bo_legacy_validate(struct radeon_bo *bo, + uint32_t *soffset, + uint32_t *eoffset); +struct radeon_bo_manager *radeon_bo_manager_legacy_ctor(struct radeon_screen *scrn); +void radeon_bo_manager_legacy_dtor(struct radeon_bo_manager *bom); +void radeon_bo_legacy_texture_age(struct radeon_bo_manager *bom); +unsigned radeon_bo_legacy_relocs_size(struct radeon_bo *bo); +struct radeon_bo *radeon_legacy_bo_alloc_fake(struct radeon_bo_manager *bom, + int size, + uint32_t offset); +void legacy_track_pending(struct radeon_bo_manager *bom, int debug); + +#endif diff --git a/src/mesa/drivers/dri/radeon/radeon_bocs_wrapper.h b/src/mesa/drivers/dri/radeon/radeon_bocs_wrapper.h new file mode 100644 index 00000000000..4520a7d7d49 --- /dev/null +++ b/src/mesa/drivers/dri/radeon/radeon_bocs_wrapper.h @@ -0,0 +1,99 @@ +#ifndef RADEON_CS_WRAPPER_H +#define RADEON_CS_WRAPPER_H + +#ifdef HAVE_LIBDRM_RADEON + +#include "radeon_bo.h" +#include "radeon_bo_gem.h" +#include "radeon_cs.h" +#include "radeon_cs_gem.h" + +#else +#include <stdint.h> + +#define RADEON_GEM_DOMAIN_CPU 0x1 // Cached CPU domain +#define RADEON_GEM_DOMAIN_GTT 0x2 // GTT or cache flushed +#define RADEON_GEM_DOMAIN_VRAM 0x4 // VRAM domain + +#define RADEON_TILING_MACRO 0x1 +#define RADEON_TILING_MICRO 0x2 +#define RADEON_TILING_SWAP 0x4 +#define RADEON_TILING_SURFACE 0x8 /* this object requires a surface + * when mapped - i.e. front buffer */ + +/* to be used to build locally in mesa with no libdrm bits */ +#include "../radeon/radeon_bo_drm.h" +#include "../radeon/radeon_cs_drm.h" + +#ifndef DRM_RADEON_GEM_INFO +#define DRM_RADEON_GEM_INFO 0x1c + +struct drm_radeon_gem_info { + uint64_t gart_size; + uint64_t vram_size; + uint64_t vram_visible; +}; + +struct drm_radeon_info { + uint32_t request; + uint32_t pad; + uint32_t value; +}; +#endif + +#ifndef RADEON_PARAM_DEVICE_ID +#define RADEON_PARAM_DEVICE_ID 16 +#endif + +#ifndef RADEON_PARAM_NUM_Z_PIPES +#define RADEON_PARAM_NUM_Z_PIPES 17 +#endif + +#ifndef RADEON_INFO_DEVICE_ID +#define RADEON_INFO_DEVICE_ID 0 +#endif +#ifndef RADEON_INFO_NUM_GB_PIPES +#define RADEON_INFO_NUM_GB_PIPES 0 +#endif + +#ifndef RADEON_INFO_NUM_Z_PIPES +#define RADEON_INFO_NUM_Z_PIPES 0 +#endif + +#ifndef DRM_RADEON_INFO +#define DRM_RADEON_INFO 0x1 +#endif + + +static inline uint32_t radeon_gem_name_bo(struct radeon_bo *dummy) +{ + return 0; +} + +static inline void *radeon_bo_manager_gem_ctor(int fd) +{ + return NULL; +} + +static inline void radeon_bo_manager_gem_dtor(void *dummy) +{ +} + +static inline void *radeon_cs_manager_gem_ctor(int fd) +{ + return NULL; +} + +static inline void radeon_cs_manager_gem_dtor(void *dummy) +{ +} + +static inline void radeon_tracker_print(void *ptr, int io) +{ +} +#endif + +#include "radeon_bo_legacy.h" +#include "radeon_cs_legacy.h" + +#endif diff --git a/src/mesa/drivers/dri/radeon/radeon_buffer_objects.c b/src/mesa/drivers/dri/radeon/radeon_buffer_objects.c new file mode 100644 index 00000000000..a24b6dac265 --- /dev/null +++ b/src/mesa/drivers/dri/radeon/radeon_buffer_objects.c @@ -0,0 +1,222 @@ +/* + * Copyright 2009 Maciej Cencora <m.cencora@gmail.com> + * + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sublicense, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial + * portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + */ + +#include "radeon_buffer_objects.h" + +#include "main/imports.h" +#include "main/mtypes.h" +#include "main/bufferobj.h" + +#include "radeon_common.h" + +struct radeon_buffer_object * +get_radeon_buffer_object(struct gl_buffer_object *obj) +{ + return (struct radeon_buffer_object *) obj; +} + +static struct gl_buffer_object * +radeonNewBufferObject(GLcontext * ctx, + GLuint name, + GLenum target) +{ + struct radeon_buffer_object *obj = CALLOC_STRUCT(radeon_buffer_object); + + _mesa_initialize_buffer_object(&obj->Base, name, target); + + obj->bo = NULL; + + return &obj->Base; +} + +/** + * Called via glDeleteBuffersARB(). + */ +static void +radeonDeleteBufferObject(GLcontext * ctx, + struct gl_buffer_object *obj) +{ + struct radeon_buffer_object *radeon_obj = get_radeon_buffer_object(obj); + + if (obj->Pointer) { + radeon_bo_unmap(radeon_obj->bo); + } + + if (radeon_obj->bo) { + radeon_bo_unref(radeon_obj->bo); + } + + _mesa_free(radeon_obj); +} + + +/** + * Allocate space for and store data in a buffer object. Any data that was + * previously stored in the buffer object is lost. If data is NULL, + * memory will be allocated, but no copy will occur. + * Called via ctx->Driver.BufferData(). + * \return GL_TRUE for success, GL_FALSE if out of memory + */ +static GLboolean +radeonBufferData(GLcontext * ctx, + GLenum target, + GLsizeiptrARB size, + const GLvoid * data, + GLenum usage, + struct gl_buffer_object *obj) +{ + radeonContextPtr radeon = RADEON_CONTEXT(ctx); + struct radeon_buffer_object *radeon_obj = get_radeon_buffer_object(obj); + + radeon_obj->Base.Size = size; + radeon_obj->Base.Usage = usage; + + if (radeon_obj->bo != NULL) { + radeon_bo_unref(radeon_obj->bo); + radeon_obj->bo = NULL; + } + + if (size != 0) { + radeon_obj->bo = radeon_bo_open(radeon->radeonScreen->bom, + 0, + size, + 32, + RADEON_GEM_DOMAIN_GTT, + 0); + + if (!radeon_obj->bo) + return GL_FALSE; + + if (data != NULL) { + radeon_bo_map(radeon_obj->bo, GL_TRUE); + + _mesa_memcpy(radeon_obj->bo->ptr, data, size); + + radeon_bo_unmap(radeon_obj->bo); + } + } + return GL_TRUE; +} + +/** + * Replace data in a subrange of buffer object. If the data range + * specified by size + offset extends beyond the end of the buffer or + * if data is NULL, no copy is performed. + * Called via glBufferSubDataARB(). + */ +static void +radeonBufferSubData(GLcontext * ctx, + GLenum target, + GLintptrARB offset, + GLsizeiptrARB size, + const GLvoid * data, + struct gl_buffer_object *obj) +{ + struct radeon_buffer_object *radeon_obj = get_radeon_buffer_object(obj); + + radeon_bo_map(radeon_obj->bo, GL_TRUE); + + _mesa_memcpy(radeon_obj->bo->ptr + offset, data, size); + + radeon_bo_unmap(radeon_obj->bo); +} + +/** + * Called via glGetBufferSubDataARB() + */ +static void +radeonGetBufferSubData(GLcontext * ctx, + GLenum target, + GLintptrARB offset, + GLsizeiptrARB size, + GLvoid * data, + struct gl_buffer_object *obj) +{ + struct radeon_buffer_object *radeon_obj = get_radeon_buffer_object(obj); + + radeon_bo_map(radeon_obj->bo, GL_FALSE); + + _mesa_memcpy(data, radeon_obj->bo->ptr + offset, size); + + radeon_bo_unmap(radeon_obj->bo); +} + +/** + * Called via glMapBufferARB() + */ +static void * +radeonMapBuffer(GLcontext * ctx, + GLenum target, + GLenum access, + struct gl_buffer_object *obj) +{ + struct radeon_buffer_object *radeon_obj = get_radeon_buffer_object(obj); + + if (access == GL_WRITE_ONLY_ARB) { + ctx->Driver.Flush(ctx); + } + + if (radeon_obj->bo == NULL) { + obj->Pointer = NULL; + return NULL; + } + + radeon_bo_map(radeon_obj->bo, access == GL_WRITE_ONLY_ARB); + + return obj->Pointer = radeon_obj->bo->ptr; +} + + +/** + * Called via glUnmapBufferARB() + */ +static GLboolean +radeonUnmapBuffer(GLcontext * ctx, + GLenum target, + struct gl_buffer_object *obj) +{ + struct radeon_buffer_object *radeon_obj = get_radeon_buffer_object(obj); + + if (radeon_obj->bo != NULL) { + radeon_bo_unmap(radeon_obj->bo); + obj->Pointer = NULL; + } + + return GL_TRUE; +} + +void +radeonInitBufferObjectFuncs(struct dd_function_table *functions) +{ + functions->NewBufferObject = radeonNewBufferObject; + functions->DeleteBuffer = radeonDeleteBufferObject; + functions->BufferData = radeonBufferData; + functions->BufferSubData = radeonBufferSubData; + functions->GetBufferSubData = radeonGetBufferSubData; + functions->MapBuffer = radeonMapBuffer; + functions->UnmapBuffer = radeonUnmapBuffer; +} diff --git a/src/mesa/drivers/dri/radeon/radeon_buffer_objects.h b/src/mesa/drivers/dri/radeon/radeon_buffer_objects.h new file mode 100644 index 00000000000..d681960825a --- /dev/null +++ b/src/mesa/drivers/dri/radeon/radeon_buffer_objects.h @@ -0,0 +1,52 @@ +/* + * Copyright 2009 Maciej Cencora <m.cencora@gmail.com> + * + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sublicense, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial + * portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + */ + +#ifndef RADEON_BUFFER_OBJECTS_H +#define RADEON_BUFFER_OBJECTS_H + +#include "main/mtypes.h" + +struct radeon_bo; + +/** + * Radeon vertex/pixel buffer object, derived from Mesa's gl_buffer_object. + */ +struct radeon_buffer_object +{ + struct gl_buffer_object Base; + struct radeon_bo *bo; +}; + +struct radeon_buffer_object * +get_radeon_buffer_object(struct gl_buffer_object *obj); + +/** + * Hook the bufferobject implementation into mesa: + */ +void radeonInitBufferObjectFuncs(struct dd_function_table *functions); + +#endif diff --git a/src/mesa/drivers/dri/radeon/radeon_chipset.h b/src/mesa/drivers/dri/radeon/radeon_chipset.h index f6bd1eb83fb..46a9cd5ff84 100644 --- a/src/mesa/drivers/dri/radeon/radeon_chipset.h +++ b/src/mesa/drivers/dri/radeon/radeon_chipset.h @@ -255,6 +255,145 @@ #define PCI_CHIP_RS740_796E 0x796E #define PCI_CHIP_RS740_796F 0x796F +#define PCI_CHIP_R600_9400 0x9400 +#define PCI_CHIP_R600_9401 0x9401 +#define PCI_CHIP_R600_9402 0x9402 +#define PCI_CHIP_R600_9403 0x9403 +#define PCI_CHIP_R600_9405 0x9405 +#define PCI_CHIP_R600_940A 0x940A +#define PCI_CHIP_R600_940B 0x940B +#define PCI_CHIP_R600_940F 0x940F + +#define PCI_CHIP_RV610_94C0 0x94C0 +#define PCI_CHIP_RV610_94C1 0x94C1 +#define PCI_CHIP_RV610_94C3 0x94C3 +#define PCI_CHIP_RV610_94C4 0x94C4 +#define PCI_CHIP_RV610_94C5 0x94C5 +#define PCI_CHIP_RV610_94C6 0x94C6 +#define PCI_CHIP_RV610_94C7 0x94C7 +#define PCI_CHIP_RV610_94C8 0x94C8 +#define PCI_CHIP_RV610_94C9 0x94C9 +#define PCI_CHIP_RV610_94CB 0x94CB +#define PCI_CHIP_RV610_94CC 0x94CC +#define PCI_CHIP_RV610_94CD 0x94CD + +#define PCI_CHIP_RV630_9580 0x9580 +#define PCI_CHIP_RV630_9581 0x9581 +#define PCI_CHIP_RV630_9583 0x9583 +#define PCI_CHIP_RV630_9586 0x9586 +#define PCI_CHIP_RV630_9587 0x9587 +#define PCI_CHIP_RV630_9588 0x9588 +#define PCI_CHIP_RV630_9589 0x9589 +#define PCI_CHIP_RV630_958A 0x958A +#define PCI_CHIP_RV630_958B 0x958B +#define PCI_CHIP_RV630_958C 0x958C +#define PCI_CHIP_RV630_958D 0x958D +#define PCI_CHIP_RV630_958E 0x958E +#define PCI_CHIP_RV630_958F 0x958F + +#define PCI_CHIP_RV670_9500 0x9500 +#define PCI_CHIP_RV670_9501 0x9501 +#define PCI_CHIP_RV670_9504 0x9504 +#define PCI_CHIP_RV670_9505 0x9505 +#define PCI_CHIP_RV670_9506 0x9506 +#define PCI_CHIP_RV670_9507 0x9507 +#define PCI_CHIP_RV670_9508 0x9508 +#define PCI_CHIP_RV670_9509 0x9509 +#define PCI_CHIP_RV670_950F 0x950F +#define PCI_CHIP_RV670_9511 0x9511 +#define PCI_CHIP_RV670_9515 0x9515 +#define PCI_CHIP_RV670_9517 0x9517 +#define PCI_CHIP_RV670_9519 0x9519 + +#define PCI_CHIP_RV620_95C0 0x95C0 +#define PCI_CHIP_RV620_95C2 0x95C2 +#define PCI_CHIP_RV620_95C4 0x95C4 +#define PCI_CHIP_RV620_95C5 0x95C5 +#define PCI_CHIP_RV620_95C6 0x95C6 +#define PCI_CHIP_RV620_95C7 0x95C7 +#define PCI_CHIP_RV620_95C9 0x95C9 +#define PCI_CHIP_RV620_95CC 0x95CC +#define PCI_CHIP_RV620_95CD 0x95CD +#define PCI_CHIP_RV620_95CE 0x95CE +#define PCI_CHIP_RV620_95CF 0x95CF + +#define PCI_CHIP_RV635_9590 0x9590 +#define PCI_CHIP_RV635_9591 0x9591 +#define PCI_CHIP_RV635_9593 0x9593 +#define PCI_CHIP_RV635_9595 0x9595 +#define PCI_CHIP_RV635_9596 0x9596 +#define PCI_CHIP_RV635_9597 0x9597 +#define PCI_CHIP_RV635_9598 0x9598 +#define PCI_CHIP_RV635_9599 0x9599 +#define PCI_CHIP_RV635_959B 0x959B + +#define PCI_CHIP_RS780_9610 0x9610 +#define PCI_CHIP_RS780_9611 0x9611 +#define PCI_CHIP_RS780_9612 0x9612 +#define PCI_CHIP_RS780_9613 0x9613 +#define PCI_CHIP_RS780_9614 0x9614 +#define PCI_CHIP_RS780_9615 0x9615 +#define PCI_CHIP_RS780_9616 0x9616 + +#define PCI_CHIP_RS880_9710 0x9710 +#define PCI_CHIP_RS880_9711 0x9711 +#define PCI_CHIP_RS880_9712 0x9712 +#define PCI_CHIP_RS880_9713 0x9713 +#define PCI_CHIP_RS880_9714 0x9714 + +#define PCI_CHIP_RV770_9440 0x9440 +#define PCI_CHIP_RV770_9441 0x9441 +#define PCI_CHIP_RV770_9442 0x9442 +#define PCI_CHIP_RV770_9443 0x9443 +#define PCI_CHIP_RV770_9444 0x9444 +#define PCI_CHIP_RV770_9446 0x9446 +#define PCI_CHIP_RV770_944A 0x944A +#define PCI_CHIP_RV770_944B 0x944B +#define PCI_CHIP_RV770_944C 0x944C +#define PCI_CHIP_RV770_944E 0x944E +#define PCI_CHIP_RV770_9450 0x9450 +#define PCI_CHIP_RV770_9452 0x9452 +#define PCI_CHIP_RV770_9456 0x9456 +#define PCI_CHIP_RV770_945A 0x945A +#define PCI_CHIP_RV770_945B 0x945B +#define PCI_CHIP_RV790_9460 0x9460 +#define PCI_CHIP_RV790_9462 0x9462 +#define PCI_CHIP_RV770_946A 0x946A +#define PCI_CHIP_RV770_946B 0x946B +#define PCI_CHIP_RV770_947A 0x947A +#define PCI_CHIP_RV770_947B 0x947B + +#define PCI_CHIP_RV730_9480 0x9480 +#define PCI_CHIP_RV730_9487 0x9487 +#define PCI_CHIP_RV730_9488 0x9488 +#define PCI_CHIP_RV730_9489 0x9489 +#define PCI_CHIP_RV730_948F 0x948F +#define PCI_CHIP_RV730_9490 0x9490 +#define PCI_CHIP_RV730_9491 0x9491 +#define PCI_CHIP_RV730_9495 0x9495 +#define PCI_CHIP_RV730_9498 0x9498 +#define PCI_CHIP_RV730_949C 0x949C +#define PCI_CHIP_RV730_949E 0x949E +#define PCI_CHIP_RV730_949F 0x949F + +#define PCI_CHIP_RV710_9540 0x9540 +#define PCI_CHIP_RV710_9541 0x9541 +#define PCI_CHIP_RV710_9542 0x9542 +#define PCI_CHIP_RV710_954E 0x954E +#define PCI_CHIP_RV710_954F 0x954F +#define PCI_CHIP_RV710_9552 0x9552 +#define PCI_CHIP_RV710_9553 0x9553 +#define PCI_CHIP_RV710_9555 0x9555 +#define PCI_CHIP_RV710_9557 0x9557 + +#define PCI_CHIP_RV740_94A0 0x94A0 +#define PCI_CHIP_RV740_94A1 0x94A1 +#define PCI_CHIP_RV740_94A3 0x94A3 +#define PCI_CHIP_RV740_94B1 0x94B1 +#define PCI_CHIP_RV740_94B3 0x94B3 +#define PCI_CHIP_RV740_94B4 0x94B4 +#define PCI_CHIP_RV740_94B5 0x94B5 +#define PCI_CHIP_RV740_94B9 0x94B9 enum { CHIP_FAMILY_R100, @@ -282,6 +421,18 @@ enum { CHIP_FAMILY_R580, CHIP_FAMILY_RV560, CHIP_FAMILY_RV570, + CHIP_FAMILY_R600, + CHIP_FAMILY_RV610, + CHIP_FAMILY_RV630, + CHIP_FAMILY_RV670, + CHIP_FAMILY_RV620, + CHIP_FAMILY_RV635, + CHIP_FAMILY_RS780, + CHIP_FAMILY_RS880, + CHIP_FAMILY_RV770, + CHIP_FAMILY_RV730, + CHIP_FAMILY_RV710, + CHIP_FAMILY_RV740, CHIP_FAMILY_LAST }; @@ -289,6 +440,7 @@ enum { #define RADEON_CLASS_R100 (0 << 0) #define RADEON_CLASS_R200 (1 << 0) #define RADEON_CLASS_R300 (2 << 0) +#define RADEON_CLASS_R600 (3 << 0) #define RADEON_CLASS_MASK (3 << 0) #define RADEON_CHIPSET_TCL (1 << 2) /* tcl support - any radeon */ diff --git a/src/mesa/drivers/dri/radeon/radeon_cmdbuf.h b/src/mesa/drivers/dri/radeon/radeon_cmdbuf.h new file mode 100644 index 00000000000..6fcd1ce7ca6 --- /dev/null +++ b/src/mesa/drivers/dri/radeon/radeon_cmdbuf.h @@ -0,0 +1,121 @@ +#ifndef COMMON_CMDBUF_H +#define COMMON_CMDBUF_H + +#include "radeon_bocs_wrapper.h" + +GLboolean rcommonEnsureCmdBufSpace(radeonContextPtr rmesa, int dwords, const char *caller); +int rcommonFlushCmdBuf(radeonContextPtr rmesa, const char *caller); +int rcommonFlushCmdBufLocked(radeonContextPtr rmesa, const char *caller); +void rcommonInitCmdBuf(radeonContextPtr rmesa); +void rcommonDestroyCmdBuf(radeonContextPtr rmesa); + +void rcommonBeginBatch(radeonContextPtr rmesa, + int n, + int dostate, + const char *file, + const char *function, + int line); + +/* +r6/r7 : code here moved */ + +#define CP_PACKET2 (2 << 30) +#define CP_PACKET0(reg, n) (RADEON_CP_PACKET0 | ((n)<<16) | ((reg)>>2)) +#define CP_PACKET0_ONE(reg, n) (RADEON_CP_PACKET0 | RADEON_CP_PACKET0_ONE_REG_WR | ((n)<<16) | ((reg)>>2)) +#define CP_PACKET3(pkt, n) (RADEON_CP_PACKET3 | (pkt) | ((n) << 16)) + +/** + * Every function writing to the command buffer needs to declare this + * to get the necessary local variables. + */ +#define BATCH_LOCALS(rmesa) \ + const radeonContextPtr b_l_rmesa = rmesa + +/** + * Prepare writing n dwords to the command buffer, + * including producing any necessary state emits on buffer wraparound. + */ +#define BEGIN_BATCH(n) rcommonBeginBatch(b_l_rmesa, n, 1, __FILE__, __FUNCTION__, __LINE__) + +/** + * Same as BEGIN_BATCH, but do not cause automatic state emits. + */ +#define BEGIN_BATCH_NO_AUTOSTATE(n) rcommonBeginBatch(b_l_rmesa, n, 0, __FILE__, __FUNCTION__, __LINE__) + +/** + * Write one dword to the command buffer. + */ +#define OUT_BATCH(data) \ + do { \ + radeon_cs_write_dword(b_l_rmesa->cmdbuf.cs, data);\ + } while(0) + +/** + * Write a relocated dword to the command buffer. + */ +#define OUT_BATCH_RELOC(data, bo, offset, rd, wd, flags) \ + do { \ + int __offset = (offset); \ + if (0 && __offset) { \ + fprintf(stderr, "(%s:%s:%d) offset : %d\n", \ + __FILE__, __FUNCTION__, __LINE__, __offset); \ + } \ + radeon_cs_write_dword(b_l_rmesa->cmdbuf.cs, __offset); \ + radeon_cs_write_reloc(b_l_rmesa->cmdbuf.cs, \ + bo, rd, wd, flags); \ + if (!b_l_rmesa->radeonScreen->kernel_mm) \ + b_l_rmesa->cmdbuf.cs->section_cdw += 2; \ + } while(0) + + +/** + * Write n dwords from ptr to the command buffer. + */ +#define OUT_BATCH_TABLE(ptr,n) \ + do { \ + radeon_cs_write_table(b_l_rmesa->cmdbuf.cs, (ptr), (n));\ + } while(0) + +/** + * Finish writing dwords to the command buffer. + * The number of (direct or indirect) OUT_BATCH calls between the previous + * BEGIN_BATCH and END_BATCH must match the number specified at BEGIN_BATCH time. + */ +#define END_BATCH() \ + do { \ + radeon_cs_end(b_l_rmesa->cmdbuf.cs, __FILE__, __FUNCTION__, __LINE__);\ + } while(0) + +/** + * After the last END_BATCH() of rendering, this indicates that flushing + * the command buffer now is okay. + */ +#define COMMIT_BATCH() \ + do { \ + } while(0) + + +/** Single register write to command buffer; requires 2 dwords. */ +#define OUT_BATCH_REGVAL(reg, val) \ + OUT_BATCH(cmdpacket0(b_l_rmesa->radeonScreen, (reg), 1)); \ + OUT_BATCH((val)) + +/** Continuous register range write to command buffer; requires 1 dword, + * expects count dwords afterwards for register contents. */ +#define OUT_BATCH_REGSEQ(reg, count) \ + OUT_BATCH(cmdpacket0(b_l_rmesa->radeonScreen, (reg), (count))) + +/** Write a 32 bit float to the ring; requires 1 dword. */ +#define OUT_BATCH_FLOAT32(f) \ + OUT_BATCH(radeonPackFloat32((f))) + +/* +r6/r7 : code here moved */ + +/* Fire the buffered vertices no matter what. + */ +static INLINE void radeon_firevertices(radeonContextPtr radeon) +{ + if (radeon->cmdbuf.cs->cdw || radeon->dma.flush ) + radeon->glCtx->Driver.Flush(radeon->glCtx); /* +r6/r7 */ +} + +#endif diff --git a/src/mesa/drivers/dri/radeon/radeon_common.c b/src/mesa/drivers/dri/radeon/radeon_common.c new file mode 100644 index 00000000000..a4c7b40798a --- /dev/null +++ b/src/mesa/drivers/dri/radeon/radeon_common.c @@ -0,0 +1,1349 @@ +/************************************************************************** + +Copyright (C) The Weather Channel, Inc. 2002. All Rights Reserved. + +The Weather Channel (TM) funded Tungsten Graphics to develop the +initial release of the Radeon 8500 driver under the XFree86 license. +This notice must be preserved. + +Permission is hereby granted, free of charge, to any person obtaining +a copy of this software and associated documentation files (the +"Software"), to deal in the Software without restriction, including +without limitation the rights to use, copy, modify, merge, publish, +distribute, sublicense, and/or sell copies of the Software, and to +permit persons to whom the Software is furnished to do so, subject to +the following conditions: + +The above copyright notice and this permission notice (including the +next paragraph) shall be included in all copies or substantial +portions of the Software. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, +EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF +MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. +IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE +LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION +OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION +WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + +**************************************************************************/ + +/* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + +/* + - Scissor implementation + - buffer swap/copy ioctls + - finish/flush + - state emission + - cmdbuffer management +*/ + +#include <errno.h> +#include "main/glheader.h" +#include "main/imports.h" +#include "main/context.h" +#include "main/enums.h" +#include "main/framebuffer.h" +#include "main/renderbuffer.h" +#include "drivers/common/meta.h" + +#include "vblank.h" + +#include "radeon_common.h" +#include "radeon_bocs_wrapper.h" +#include "radeon_lock.h" +#include "radeon_drm.h" +#include "radeon_queryobj.h" + +/** + * Enable verbose debug output for emit code. + * 0 no output + * 1 most output + * 2 also print state alues + */ +#define RADEON_CMDBUF 0 + +/* ============================================================= + * Scissoring + */ + +static GLboolean intersect_rect(drm_clip_rect_t * out, + drm_clip_rect_t * a, drm_clip_rect_t * b) +{ + *out = *a; + if (b->x1 > out->x1) + out->x1 = b->x1; + if (b->y1 > out->y1) + out->y1 = b->y1; + if (b->x2 < out->x2) + out->x2 = b->x2; + if (b->y2 < out->y2) + out->y2 = b->y2; + if (out->x1 >= out->x2) + return GL_FALSE; + if (out->y1 >= out->y2) + return GL_FALSE; + return GL_TRUE; +} + +void radeonRecalcScissorRects(radeonContextPtr radeon) +{ + drm_clip_rect_t *out; + int i; + + /* Grow cliprect store? + */ + if (radeon->state.scissor.numAllocedClipRects < radeon->numClipRects) { + while (radeon->state.scissor.numAllocedClipRects < + radeon->numClipRects) { + radeon->state.scissor.numAllocedClipRects += 1; /* zero case */ + radeon->state.scissor.numAllocedClipRects *= 2; + } + + if (radeon->state.scissor.pClipRects) + FREE(radeon->state.scissor.pClipRects); + + radeon->state.scissor.pClipRects = + MALLOC(radeon->state.scissor.numAllocedClipRects * + sizeof(drm_clip_rect_t)); + + if (radeon->state.scissor.pClipRects == NULL) { + radeon->state.scissor.numAllocedClipRects = 0; + return; + } + } + + out = radeon->state.scissor.pClipRects; + radeon->state.scissor.numClipRects = 0; + + for (i = 0; i < radeon->numClipRects; i++) { + if (intersect_rect(out, + &radeon->pClipRects[i], + &radeon->state.scissor.rect)) { + radeon->state.scissor.numClipRects++; + out++; + } + } + + if (radeon->vtbl.update_scissor) + radeon->vtbl.update_scissor(radeon->glCtx); +} + +void radeon_get_cliprects(radeonContextPtr radeon, + struct drm_clip_rect **cliprects, + unsigned int *num_cliprects, + int *x_off, int *y_off) +{ + __DRIdrawablePrivate *dPriv = radeon_get_drawable(radeon); + struct radeon_framebuffer *rfb = dPriv->driverPrivate; + + if (radeon->constant_cliprect) { + radeon->fboRect.x1 = 0; + radeon->fboRect.y1 = 0; + radeon->fboRect.x2 = radeon->glCtx->DrawBuffer->Width; + radeon->fboRect.y2 = radeon->glCtx->DrawBuffer->Height; + + *cliprects = &radeon->fboRect; + *num_cliprects = 1; + *x_off = 0; + *y_off = 0; + } else if (radeon->front_cliprects || + rfb->pf_active || dPriv->numBackClipRects == 0) { + *cliprects = dPriv->pClipRects; + *num_cliprects = dPriv->numClipRects; + *x_off = dPriv->x; + *y_off = dPriv->y; + } else { + *num_cliprects = dPriv->numBackClipRects; + *cliprects = dPriv->pBackClipRects; + *x_off = dPriv->backX; + *y_off = dPriv->backY; + } +} + +/** + * Update cliprects and scissors. + */ +void radeonSetCliprects(radeonContextPtr radeon) +{ + __DRIdrawablePrivate *const drawable = radeon_get_drawable(radeon); + __DRIdrawablePrivate *const readable = radeon_get_readable(radeon); + struct radeon_framebuffer *const draw_rfb = drawable->driverPrivate; + struct radeon_framebuffer *const read_rfb = readable->driverPrivate; + int x_off, y_off; + + radeon_get_cliprects(radeon, &radeon->pClipRects, + &radeon->numClipRects, &x_off, &y_off); + + if ((draw_rfb->base.Width != drawable->w) || + (draw_rfb->base.Height != drawable->h)) { + _mesa_resize_framebuffer(radeon->glCtx, &draw_rfb->base, + drawable->w, drawable->h); + draw_rfb->base.Initialized = GL_TRUE; + } + + if (drawable != readable) { + if ((read_rfb->base.Width != readable->w) || + (read_rfb->base.Height != readable->h)) { + _mesa_resize_framebuffer(radeon->glCtx, &read_rfb->base, + readable->w, readable->h); + read_rfb->base.Initialized = GL_TRUE; + } + } + + if (radeon->state.scissor.enabled) + radeonRecalcScissorRects(radeon); + +} + + + +void radeonUpdateScissor( GLcontext *ctx ) +{ + radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + GLint x = ctx->Scissor.X, y = ctx->Scissor.Y; + GLsizei w = ctx->Scissor.Width, h = ctx->Scissor.Height; + int x1, y1, x2, y2; + int min_x, min_y, max_x, max_y; + + if (!ctx->DrawBuffer) + return; + min_x = min_y = 0; + max_x = ctx->DrawBuffer->Width - 1; + max_y = ctx->DrawBuffer->Height - 1; + + if ( !ctx->DrawBuffer->Name ) { + x1 = x; + y1 = ctx->DrawBuffer->Height - (y + h); + x2 = x + w - 1; + y2 = y1 + h - 1; + } else { + x1 = x; + y1 = y; + x2 = x + w - 1; + y2 = y + h - 1; + + } + if (!rmesa->radeonScreen->kernel_mm) { + /* Fix scissors for dri 1 */ + + __DRIdrawablePrivate *dPriv = radeon_get_drawable(rmesa); + x1 += dPriv->x; + x2 += dPriv->x + 1; + min_x += dPriv->x; + max_x += dPriv->x + 1; + y1 += dPriv->y; + y2 += dPriv->y + 1; + min_y += dPriv->y; + max_y += dPriv->y + 1; + } + + rmesa->state.scissor.rect.x1 = CLAMP(x1, min_x, max_x); + rmesa->state.scissor.rect.y1 = CLAMP(y1, min_y, max_y); + rmesa->state.scissor.rect.x2 = CLAMP(x2, min_x, max_x); + rmesa->state.scissor.rect.y2 = CLAMP(y2, min_y, max_y); + + radeonRecalcScissorRects( rmesa ); +} + +/* ============================================================= + * Scissoring + */ + +void radeonScissor(GLcontext* ctx, GLint x, GLint y, GLsizei w, GLsizei h) +{ + radeonContextPtr radeon = RADEON_CONTEXT(ctx); + if (ctx->Scissor.Enabled) { + /* We don't pipeline cliprect changes */ + radeon_firevertices(radeon); + radeonUpdateScissor(ctx); + } +} + +void radeonPolygonStipplePreKMS( GLcontext *ctx, const GLubyte *mask ) +{ + radeonContextPtr radeon = RADEON_CONTEXT(ctx); + GLuint i; + drm_radeon_stipple_t stipple; + + /* Must flip pattern upside down. + */ + for ( i = 0 ; i < 32 ; i++ ) { + stipple.mask[31 - i] = ((GLuint *) mask)[i]; + } + + /* TODO: push this into cmd mechanism + */ + radeon_firevertices(radeon); + LOCK_HARDWARE( radeon ); + + drmCommandWrite( radeon->dri.fd, DRM_RADEON_STIPPLE, + &stipple, sizeof(stipple) ); + UNLOCK_HARDWARE( radeon ); +} + + +/* ================================================================ + * SwapBuffers with client-side throttling + */ + +static uint32_t radeonGetLastFrame(radeonContextPtr radeon) +{ + drm_radeon_getparam_t gp; + int ret; + uint32_t frame = 0; + + gp.param = RADEON_PARAM_LAST_FRAME; + gp.value = (int *)&frame; + ret = drmCommandWriteRead(radeon->dri.fd, DRM_RADEON_GETPARAM, + &gp, sizeof(gp)); + if (ret) { + fprintf(stderr, "%s: drmRadeonGetParam: %d\n", __FUNCTION__, + ret); + exit(1); + } + + return frame; +} + +uint32_t radeonGetAge(radeonContextPtr radeon) +{ + drm_radeon_getparam_t gp; + int ret; + uint32_t age; + + gp.param = RADEON_PARAM_LAST_CLEAR; + gp.value = (int *)&age; + ret = drmCommandWriteRead(radeon->dri.fd, DRM_RADEON_GETPARAM, + &gp, sizeof(gp)); + if (ret) { + fprintf(stderr, "%s: drmRadeonGetParam: %d\n", __FUNCTION__, + ret); + exit(1); + } + + return age; +} + +static void radeonEmitIrqLocked(radeonContextPtr radeon) +{ + drm_radeon_irq_emit_t ie; + int ret; + + ie.irq_seq = &radeon->iw.irq_seq; + ret = drmCommandWriteRead(radeon->dri.fd, DRM_RADEON_IRQ_EMIT, + &ie, sizeof(ie)); + if (ret) { + fprintf(stderr, "%s: drmRadeonIrqEmit: %d\n", __FUNCTION__, + ret); + exit(1); + } +} + +static void radeonWaitIrq(radeonContextPtr radeon) +{ + int ret; + + do { + ret = drmCommandWrite(radeon->dri.fd, DRM_RADEON_IRQ_WAIT, + &radeon->iw, sizeof(radeon->iw)); + } while (ret && (errno == EINTR || errno == EBUSY)); + + if (ret) { + fprintf(stderr, "%s: drmRadeonIrqWait: %d\n", __FUNCTION__, + ret); + exit(1); + } +} + +static void radeonWaitForFrameCompletion(radeonContextPtr radeon) +{ + drm_radeon_sarea_t *sarea = radeon->sarea; + + if (radeon->do_irqs) { + if (radeonGetLastFrame(radeon) < sarea->last_frame) { + if (!radeon->irqsEmitted) { + while (radeonGetLastFrame(radeon) < + sarea->last_frame) ; + } else { + UNLOCK_HARDWARE(radeon); + radeonWaitIrq(radeon); + LOCK_HARDWARE(radeon); + } + radeon->irqsEmitted = 10; + } + + if (radeon->irqsEmitted) { + radeonEmitIrqLocked(radeon); + radeon->irqsEmitted--; + } + } else { + while (radeonGetLastFrame(radeon) < sarea->last_frame) { + UNLOCK_HARDWARE(radeon); + if (radeon->do_usleeps) + DO_USLEEP(1); + LOCK_HARDWARE(radeon); + } + } +} + +/* wait for idle */ +void radeonWaitForIdleLocked(radeonContextPtr radeon) +{ + int ret; + int i = 0; + + do { + ret = drmCommandNone(radeon->dri.fd, DRM_RADEON_CP_IDLE); + if (ret) + DO_USLEEP(1); + } while (ret && ++i < 100); + + if (ret < 0) { + UNLOCK_HARDWARE(radeon); + fprintf(stderr, "Error: R300 timed out... exiting\n"); + exit(-1); + } +} + +static void radeonWaitForIdle(radeonContextPtr radeon) +{ + if (!radeon->radeonScreen->driScreen->dri2.enabled) { + LOCK_HARDWARE(radeon); + radeonWaitForIdleLocked(radeon); + UNLOCK_HARDWARE(radeon); + } +} + +static void radeon_flip_renderbuffers(struct radeon_framebuffer *rfb) +{ + int current_page = rfb->pf_current_page; + int next_page = (current_page + 1) % rfb->pf_num_pages; + struct gl_renderbuffer *tmp_rb; + + /* Exchange renderbuffers if necessary but make sure their + * reference counts are preserved. + */ + if (rfb->color_rb[current_page] && + rfb->base.Attachment[BUFFER_FRONT_LEFT].Renderbuffer != + &rfb->color_rb[current_page]->base) { + tmp_rb = NULL; + _mesa_reference_renderbuffer(&tmp_rb, + rfb->base.Attachment[BUFFER_FRONT_LEFT].Renderbuffer); + tmp_rb = &rfb->color_rb[current_page]->base; + _mesa_reference_renderbuffer(&rfb->base.Attachment[BUFFER_FRONT_LEFT].Renderbuffer, tmp_rb); + _mesa_reference_renderbuffer(&tmp_rb, NULL); + } + + if (rfb->color_rb[next_page] && + rfb->base.Attachment[BUFFER_BACK_LEFT].Renderbuffer != + &rfb->color_rb[next_page]->base) { + tmp_rb = NULL; + _mesa_reference_renderbuffer(&tmp_rb, + rfb->base.Attachment[BUFFER_BACK_LEFT].Renderbuffer); + tmp_rb = &rfb->color_rb[next_page]->base; + _mesa_reference_renderbuffer(&rfb->base.Attachment[BUFFER_BACK_LEFT].Renderbuffer, tmp_rb); + _mesa_reference_renderbuffer(&tmp_rb, NULL); + } +} + +/* Copy the back color buffer to the front color buffer. + */ +void radeonCopyBuffer( __DRIdrawablePrivate *dPriv, + const drm_clip_rect_t *rect) +{ + radeonContextPtr rmesa; + struct radeon_framebuffer *rfb; + GLint nbox, i, ret; + + assert(dPriv); + assert(dPriv->driContextPriv); + assert(dPriv->driContextPriv->driverPrivate); + + rmesa = (radeonContextPtr) dPriv->driContextPriv->driverPrivate; + + LOCK_HARDWARE(rmesa); + + rfb = dPriv->driverPrivate; + + if ( RADEON_DEBUG & RADEON_IOCTL ) { + fprintf( stderr, "\n%s( %p )\n\n", __FUNCTION__, (void *) rmesa->glCtx ); + } + + nbox = dPriv->numClipRects; /* must be in locked region */ + + for ( i = 0 ; i < nbox ; ) { + GLint nr = MIN2( i + RADEON_NR_SAREA_CLIPRECTS , nbox ); + drm_clip_rect_t *box = dPriv->pClipRects; + drm_clip_rect_t *b = rmesa->sarea->boxes; + GLint n = 0; + + for ( ; i < nr ; i++ ) { + + *b = box[i]; + + if (rect) + { + if (rect->x1 > b->x1) + b->x1 = rect->x1; + if (rect->y1 > b->y1) + b->y1 = rect->y1; + if (rect->x2 < b->x2) + b->x2 = rect->x2; + if (rect->y2 < b->y2) + b->y2 = rect->y2; + + if (b->x1 >= b->x2 || b->y1 >= b->y2) + continue; + } + + b++; + n++; + } + rmesa->sarea->nbox = n; + + if (!n) + continue; + + ret = drmCommandNone( rmesa->dri.fd, DRM_RADEON_SWAP ); + + if ( ret ) { + fprintf( stderr, "DRM_RADEON_SWAP_BUFFERS: return = %d\n", ret ); + UNLOCK_HARDWARE( rmesa ); + exit( 1 ); + } + } + + UNLOCK_HARDWARE( rmesa ); +} + +static int radeonScheduleSwap(__DRIdrawablePrivate *dPriv, GLboolean *missed_target) +{ + radeonContextPtr rmesa; + + rmesa = (radeonContextPtr) dPriv->driContextPriv->driverPrivate; + radeon_firevertices(rmesa); + + LOCK_HARDWARE( rmesa ); + + if (!dPriv->numClipRects) { + UNLOCK_HARDWARE(rmesa); + usleep(10000); /* throttle invisible client 10ms */ + return 0; + } + + radeonWaitForFrameCompletion(rmesa); + + UNLOCK_HARDWARE(rmesa); + driWaitForVBlank(dPriv, missed_target); + + return 0; +} + +static GLboolean radeonPageFlip( __DRIdrawablePrivate *dPriv ) +{ + radeonContextPtr radeon; + GLint ret; + __DRIscreenPrivate *psp; + struct radeon_renderbuffer *rrb; + struct radeon_framebuffer *rfb; + + assert(dPriv); + assert(dPriv->driContextPriv); + assert(dPriv->driContextPriv->driverPrivate); + + radeon = (radeonContextPtr) dPriv->driContextPriv->driverPrivate; + rfb = dPriv->driverPrivate; + rrb = (void *)rfb->base.Attachment[BUFFER_FRONT_LEFT].Renderbuffer; + + psp = dPriv->driScreenPriv; + + LOCK_HARDWARE(radeon); + + if ( RADEON_DEBUG & RADEON_IOCTL ) { + fprintf(stderr, "%s: pfCurrentPage: %d %d\n", __FUNCTION__, + radeon->sarea->pfCurrentPage, radeon->sarea->pfState); + } + drm_clip_rect_t *box = dPriv->pClipRects; + drm_clip_rect_t *b = radeon->sarea->boxes; + b[0] = box[0]; + radeon->sarea->nbox = 1; + + ret = drmCommandNone( radeon->dri.fd, DRM_RADEON_FLIP ); + + UNLOCK_HARDWARE(radeon); + + if ( ret ) { + fprintf( stderr, "DRM_RADEON_FLIP: return = %d\n", ret ); + return GL_FALSE; + } + + if (!rfb->pf_active) + return GL_FALSE; + + rfb->pf_current_page = radeon->sarea->pfCurrentPage; + radeon_flip_renderbuffers(rfb); + radeon_draw_buffer(radeon->glCtx, &rfb->base); + + return GL_TRUE; +} + + +/** + * Swap front and back buffer. + */ +void radeonSwapBuffers(__DRIdrawablePrivate * dPriv) +{ + int64_t ust; + __DRIscreenPrivate *psp; + + if (dPriv->driContextPriv && dPriv->driContextPriv->driverPrivate) { + radeonContextPtr radeon; + GLcontext *ctx; + + radeon = (radeonContextPtr) dPriv->driContextPriv->driverPrivate; + ctx = radeon->glCtx; + + if (ctx->Visual.doubleBufferMode) { + GLboolean missed_target; + struct radeon_framebuffer *rfb = dPriv->driverPrivate; + _mesa_notifySwapBuffers(ctx);/* flush pending rendering comands */ + + radeonScheduleSwap(dPriv, &missed_target); + + if (rfb->pf_active) { + radeonPageFlip(dPriv); + } else { + radeonCopyBuffer(dPriv, NULL); + } + + psp = dPriv->driScreenPriv; + + rfb->swap_count++; + (*psp->systemTime->getUST)( & ust ); + if ( missed_target ) { + rfb->swap_missed_count++; + rfb->swap_missed_ust = ust - rfb->swap_ust; + } + + rfb->swap_ust = ust; + radeon->hw.all_dirty = GL_TRUE; + } + } else { + /* XXX this shouldn't be an error but we can't handle it for now */ + _mesa_problem(NULL, "%s: drawable has no context!", + __FUNCTION__); + } +} + +void radeonCopySubBuffer(__DRIdrawablePrivate * dPriv, + int x, int y, int w, int h ) +{ + if (dPriv->driContextPriv && dPriv->driContextPriv->driverPrivate) { + radeonContextPtr radeon; + GLcontext *ctx; + + radeon = (radeonContextPtr) dPriv->driContextPriv->driverPrivate; + ctx = radeon->glCtx; + + if (ctx->Visual.doubleBufferMode) { + drm_clip_rect_t rect; + rect.x1 = x + dPriv->x; + rect.y1 = (dPriv->h - y - h) + dPriv->y; + rect.x2 = rect.x1 + w; + rect.y2 = rect.y1 + h; + _mesa_notifySwapBuffers(ctx); /* flush pending rendering comands */ + radeonCopyBuffer(dPriv, &rect); + } + } else { + /* XXX this shouldn't be an error but we can't handle it for now */ + _mesa_problem(NULL, "%s: drawable has no context!", + __FUNCTION__); + } +} + +void radeon_draw_buffer(GLcontext *ctx, struct gl_framebuffer *fb) +{ + radeonContextPtr radeon = RADEON_CONTEXT(ctx); + struct radeon_renderbuffer *rrbDepth = NULL, *rrbStencil = NULL, + *rrbColor = NULL; + uint32_t offset = 0; + + + if (!fb) { + /* this can happen during the initial context initialization */ + return; + } + + /* radeons only handle 1 color draw so far */ + if (fb->_NumColorDrawBuffers != 1) { + radeon->vtbl.fallback(ctx, RADEON_FALLBACK_DRAW_BUFFER, GL_TRUE); + return; + } + + /* Do this here, note core Mesa, since this function is called from + * many places within the driver. + */ + if (ctx->NewState & (_NEW_BUFFERS | _NEW_COLOR | _NEW_PIXEL)) { + /* this updates the DrawBuffer->_NumColorDrawBuffers fields, etc */ + _mesa_update_framebuffer(ctx); + /* this updates the DrawBuffer's Width/Height if it's a FBO */ + _mesa_update_draw_buffer_bounds(ctx); + } + + if (fb->_Status != GL_FRAMEBUFFER_COMPLETE_EXT) { + /* this may occur when we're called by glBindFrameBuffer() during + * the process of someone setting up renderbuffers, etc. + */ + /*_mesa_debug(ctx, "DrawBuffer: incomplete user FBO\n");*/ + return; + } + + if (fb->Name) + ;/* do something depthy/stencily TODO */ + + + /* none */ + if (fb->Name == 0) { + if (fb->_ColorDrawBufferIndexes[0] == BUFFER_FRONT_LEFT) { + rrbColor = radeon_renderbuffer(fb->Attachment[BUFFER_FRONT_LEFT].Renderbuffer); + radeon->front_cliprects = GL_TRUE; + radeon->front_buffer_dirty = GL_TRUE; + } else { + rrbColor = radeon_renderbuffer(fb->Attachment[BUFFER_BACK_LEFT].Renderbuffer); + radeon->front_cliprects = GL_FALSE; + } + } else { + /* user FBO in theory */ + struct radeon_renderbuffer *rrb; + rrb = radeon_renderbuffer(fb->_ColorDrawBuffers[0]); + if (rrb) { + offset = rrb->draw_offset; + rrbColor = rrb; + } + radeon->constant_cliprect = GL_TRUE; + } + + if (rrbColor == NULL) + radeon->vtbl.fallback(ctx, RADEON_FALLBACK_DRAW_BUFFER, GL_TRUE); + else + radeon->vtbl.fallback(ctx, RADEON_FALLBACK_DRAW_BUFFER, GL_FALSE); + + + if (fb->_DepthBuffer && fb->_DepthBuffer->Wrapped) { + rrbDepth = radeon_renderbuffer(fb->_DepthBuffer->Wrapped); + if (rrbDepth && rrbDepth->bo) { + radeon->vtbl.fallback(ctx, RADEON_FALLBACK_DEPTH_BUFFER, GL_FALSE); + } else { + radeon->vtbl.fallback(ctx, RADEON_FALLBACK_DEPTH_BUFFER, GL_TRUE); + } + } else { + radeon->vtbl.fallback(ctx, RADEON_FALLBACK_DEPTH_BUFFER, GL_FALSE); + rrbDepth = NULL; + } + + if (fb->_StencilBuffer && fb->_StencilBuffer->Wrapped) { + rrbStencil = radeon_renderbuffer(fb->_StencilBuffer->Wrapped); + if (rrbStencil && rrbStencil->bo) { + radeon->vtbl.fallback(ctx, RADEON_FALLBACK_STENCIL_BUFFER, GL_FALSE); + /* need to re-compute stencil hw state */ + if (!rrbDepth) + rrbDepth = rrbStencil; + } else { + radeon->vtbl.fallback(ctx, RADEON_FALLBACK_STENCIL_BUFFER, GL_TRUE); + } + } else { + radeon->vtbl.fallback(ctx, RADEON_FALLBACK_STENCIL_BUFFER, GL_FALSE); + if (ctx->Driver.Enable != NULL) + ctx->Driver.Enable(ctx, GL_STENCIL_TEST, ctx->Stencil.Enabled); + else + ctx->NewState |= _NEW_STENCIL; + } + + /* Update culling direction which changes depending on the + * orientation of the buffer: + */ + if (ctx->Driver.FrontFace) + ctx->Driver.FrontFace(ctx, ctx->Polygon.FrontFace); + else + ctx->NewState |= _NEW_POLYGON; + + /* + * Update depth test state + */ + if (ctx->Driver.Enable) { + ctx->Driver.Enable(ctx, GL_DEPTH_TEST, + (ctx->Depth.Test && fb->Visual.depthBits > 0)); + /* Need to update the derived ctx->Stencil._Enabled first */ + ctx->Driver.Enable(ctx, GL_STENCIL_TEST, + (ctx->Stencil.Enabled && fb->Visual.stencilBits > 0)); + } else { + ctx->NewState |= (_NEW_DEPTH | _NEW_STENCIL); + } + + _mesa_reference_renderbuffer(&radeon->state.depth.rb, &rrbDepth->base); + _mesa_reference_renderbuffer(&radeon->state.color.rb, &rrbColor->base); + radeon->state.color.draw_offset = offset; + +#if 0 + /* update viewport since it depends on window size */ + if (ctx->Driver.Viewport) { + ctx->Driver.Viewport(ctx, ctx->Viewport.X, ctx->Viewport.Y, + ctx->Viewport.Width, ctx->Viewport.Height); + } else { + + } +#endif + ctx->NewState |= _NEW_VIEWPORT; + + /* Set state we know depends on drawable parameters: + */ + radeonUpdateScissor(ctx); + radeon->NewGLState |= _NEW_SCISSOR; + + if (ctx->Driver.DepthRange) + ctx->Driver.DepthRange(ctx, + ctx->Viewport.Near, + ctx->Viewport.Far); + + /* Update culling direction which changes depending on the + * orientation of the buffer: + */ + if (ctx->Driver.FrontFace) + ctx->Driver.FrontFace(ctx, ctx->Polygon.FrontFace); + else + ctx->NewState |= _NEW_POLYGON; +} + +/** + * Called via glDrawBuffer. + */ +void radeonDrawBuffer( GLcontext *ctx, GLenum mode ) +{ + if (RADEON_DEBUG & RADEON_DRI) + fprintf(stderr, "%s %s\n", __FUNCTION__, + _mesa_lookup_enum_by_nr( mode )); + + if (ctx->DrawBuffer->Name == 0) { + radeonContextPtr radeon = RADEON_CONTEXT(ctx); + + const GLboolean was_front_buffer_rendering = + radeon->is_front_buffer_rendering; + + radeon->is_front_buffer_rendering = (mode == GL_FRONT_LEFT) || + (mode == GL_FRONT); + + /* If we weren't front-buffer rendering before but we are now, make sure + * that the front-buffer has actually been allocated. + */ + if (!was_front_buffer_rendering && radeon->is_front_buffer_rendering) { + radeon_update_renderbuffers(radeon->dri.context, + radeon->dri.context->driDrawablePriv); + } + } + + radeon_draw_buffer(ctx, ctx->DrawBuffer); +} + +void radeonReadBuffer( GLcontext *ctx, GLenum mode ) +{ + if ((ctx->DrawBuffer != NULL) && (ctx->DrawBuffer->Name == 0)) { + struct radeon_context *const rmesa = RADEON_CONTEXT(ctx); + const GLboolean was_front_buffer_reading = rmesa->is_front_buffer_reading; + rmesa->is_front_buffer_reading = (mode == GL_FRONT_LEFT) + || (mode == GL_FRONT); + + if (!was_front_buffer_reading && rmesa->is_front_buffer_reading) { + radeon_update_renderbuffers(rmesa->dri.context, + rmesa->dri.context->driReadablePriv); + } + } + /* nothing, until we implement h/w glRead/CopyPixels or CopyTexImage */ + if (ctx->ReadBuffer == ctx->DrawBuffer) { + /* This will update FBO completeness status. + * A framebuffer will be incomplete if the GL_READ_BUFFER setting + * refers to a missing renderbuffer. Calling glReadBuffer can set + * that straight and can make the drawing buffer complete. + */ + radeon_draw_buffer(ctx, ctx->DrawBuffer); + } +} + + +/* Turn on/off page flipping according to the flags in the sarea: + */ +void radeonUpdatePageFlipping(radeonContextPtr radeon) +{ + struct radeon_framebuffer *rfb = radeon_get_drawable(radeon)->driverPrivate; + + rfb->pf_active = radeon->sarea->pfState; + rfb->pf_current_page = radeon->sarea->pfCurrentPage; + rfb->pf_num_pages = 2; + radeon_flip_renderbuffers(rfb); + radeon_draw_buffer(radeon->glCtx, radeon->glCtx->DrawBuffer); +} + +void radeon_window_moved(radeonContextPtr radeon) +{ + /* Cliprects has to be updated before doing anything else */ + radeonSetCliprects(radeon); + if (!radeon->radeonScreen->driScreen->dri2.enabled) { + radeonUpdatePageFlipping(radeon); + } +} + +void radeon_viewport(GLcontext *ctx, GLint x, GLint y, GLsizei width, GLsizei height) +{ + radeonContextPtr radeon = RADEON_CONTEXT(ctx); + __DRIcontext *driContext = radeon->dri.context; + void (*old_viewport)(GLcontext *ctx, GLint x, GLint y, + GLsizei w, GLsizei h); + + if (!driContext->driScreenPriv->dri2.enabled) + return; + + if (!radeon->meta.internal_viewport_call && ctx->DrawBuffer->Name == 0) { + if (radeon->is_front_buffer_rendering) { + ctx->Driver.Flush(ctx); + } + radeon_update_renderbuffers(driContext, driContext->driDrawablePriv); + if (driContext->driDrawablePriv != driContext->driReadablePriv) + radeon_update_renderbuffers(driContext, driContext->driReadablePriv); + } + + old_viewport = ctx->Driver.Viewport; + ctx->Driver.Viewport = NULL; + radeon_window_moved(radeon); + radeon_draw_buffer(ctx, radeon->glCtx->DrawBuffer); + ctx->Driver.Viewport = old_viewport; +} + +static void radeon_print_state_atom_prekmm(radeonContextPtr radeon, struct radeon_state_atom *state) +{ + int i, j, reg; + int dwords = (*state->check) (radeon->glCtx, state); + drm_r300_cmd_header_t cmd; + + fprintf(stderr, " emit %s %d/%d\n", state->name, dwords, state->cmd_size); + + if (radeon_is_debug_enabled(RADEON_STATE, RADEON_TRACE)) { + if (dwords > state->cmd_size) + dwords = state->cmd_size; + + for (i = 0; i < dwords;) { + cmd = *((drm_r300_cmd_header_t *) &state->cmd[i]); + reg = (cmd.packet0.reghi << 8) | cmd.packet0.reglo; + fprintf(stderr, " %s[%d]: cmdpacket0 (first reg=0x%04x, count=%d)\n", + state->name, i, reg, cmd.packet0.count); + ++i; + for (j = 0; j < cmd.packet0.count && i < dwords; j++) { + fprintf(stderr, " %s[%d]: 0x%04x = %08x\n", + state->name, i, reg, state->cmd[i]); + reg += 4; + ++i; + } + } + } +} + +static void radeon_print_state_atom(radeonContextPtr radeon, struct radeon_state_atom *state) +{ + int i, j, reg, count; + int dwords; + uint32_t packet0; + if (!radeon_is_debug_enabled(RADEON_STATE, RADEON_VERBOSE) ) + return; + + if (!radeon->radeonScreen->kernel_mm) { + radeon_print_state_atom_prekmm(radeon, state); + return; + } + + dwords = (*state->check) (radeon->glCtx, state); + + fprintf(stderr, " emit %s %d/%d\n", state->name, dwords, state->cmd_size); + + if (radeon_is_debug_enabled(RADEON_STATE, RADEON_TRACE)) { + if (dwords > state->cmd_size) + dwords = state->cmd_size; + for (i = 0; i < dwords;) { + packet0 = state->cmd[i]; + reg = (packet0 & 0x1FFF) << 2; + count = ((packet0 & 0x3FFF0000) >> 16) + 1; + fprintf(stderr, " %s[%d]: cmdpacket0 (first reg=0x%04x, count=%d)\n", + state->name, i, reg, count); + ++i; + for (j = 0; j < count && i < dwords; j++) { + fprintf(stderr, " %s[%d]: 0x%04x = %08x\n", + state->name, i, reg, state->cmd[i]); + reg += 4; + ++i; + } + } + } +} + +/** + * Count total size for next state emit. + **/ +GLuint radeonCountStateEmitSize(radeonContextPtr radeon) +{ + struct radeon_state_atom *atom; + GLuint dwords = 0; + /* check if we are going to emit full state */ + + if (radeon->cmdbuf.cs->cdw && !radeon->hw.all_dirty) { + if (!radeon->hw.is_dirty) + goto out; + foreach(atom, &radeon->hw.atomlist) { + if (atom->dirty) { + const GLuint atom_size = atom->check(radeon->glCtx, atom); + dwords += atom_size; + if (RADEON_CMDBUF && atom_size) { + radeon_print_state_atom(radeon, atom); + } + } + } + } else { + foreach(atom, &radeon->hw.atomlist) { + const GLuint atom_size = atom->check(radeon->glCtx, atom); + dwords += atom_size; + if (RADEON_CMDBUF && atom_size) { + radeon_print_state_atom(radeon, atom); + } + + } + } +out: + radeon_print(RADEON_STATE, RADEON_NORMAL, "%s %u\n", __func__, dwords); + return dwords; +} + +static INLINE void radeon_emit_atom(radeonContextPtr radeon, struct radeon_state_atom *atom) +{ + BATCH_LOCALS(radeon); + int dwords; + + dwords = (*atom->check) (radeon->glCtx, atom); + if (dwords) { + + radeon_print_state_atom(radeon, atom); + + if (atom->emit) { + (*atom->emit)(radeon->glCtx, atom); + } else { + BEGIN_BATCH_NO_AUTOSTATE(dwords); + OUT_BATCH_TABLE(atom->cmd, dwords); + END_BATCH(); + } + } else { + radeon_print(RADEON_STATE, RADEON_VERBOSE, " skip state %s\n", atom->name); + } + atom->dirty = GL_FALSE; + +} + +static INLINE void radeonEmitAtoms(radeonContextPtr radeon, GLboolean emitAll) +{ + struct radeon_state_atom *atom; + + if (radeon->vtbl.pre_emit_atoms) + radeon->vtbl.pre_emit_atoms(radeon); + + /* Emit actual atoms */ + if (radeon->hw.all_dirty || emitAll) { + foreach(atom, &radeon->hw.atomlist) + radeon_emit_atom( radeon, atom ); + } else { + foreach(atom, &radeon->hw.atomlist) { + if ( atom->dirty ) + radeon_emit_atom( radeon, atom ); + } + } + + COMMIT_BATCH(); +} + +static GLboolean radeon_revalidate_bos(GLcontext *ctx) +{ + radeonContextPtr radeon = RADEON_CONTEXT(ctx); + int ret; + + ret = radeon_cs_space_check(radeon->cmdbuf.cs); + if (ret == RADEON_CS_SPACE_FLUSH) + return GL_FALSE; + return GL_TRUE; +} + +void radeonEmitState(radeonContextPtr radeon) +{ + radeon_print(RADEON_STATE, RADEON_NORMAL, "%s\n", __FUNCTION__); + + if (radeon->vtbl.pre_emit_state) + radeon->vtbl.pre_emit_state(radeon); + + /* this code used to return here but now it emits zbs */ + if (radeon->cmdbuf.cs->cdw && !radeon->hw.is_dirty && !radeon->hw.all_dirty) + return; + + if (!radeon->cmdbuf.cs->cdw) { + if (RADEON_DEBUG & RADEON_STATE) + fprintf(stderr, "Begin reemit state\n"); + + radeonEmitAtoms(radeon, GL_TRUE); + } else { + + if (RADEON_DEBUG & RADEON_STATE) + fprintf(stderr, "Begin dirty state\n"); + + radeonEmitAtoms(radeon, GL_FALSE); + } + + radeon->hw.is_dirty = GL_FALSE; + radeon->hw.all_dirty = GL_FALSE; +} + + +void radeonFlush(GLcontext *ctx) +{ + radeonContextPtr radeon = RADEON_CONTEXT(ctx); + if (RADEON_DEBUG & RADEON_IOCTL) + fprintf(stderr, "%s %d\n", __FUNCTION__, radeon->cmdbuf.cs->cdw); + + /* okay if we have no cmds in the buffer && + we have no DMA flush && + we have no DMA buffer allocated. + then no point flushing anything at all. + */ + if (!radeon->dma.flush && !radeon->cmdbuf.cs->cdw && is_empty_list(&radeon->dma.reserved)) + return; + + if (radeon->dma.flush) + radeon->dma.flush( ctx ); + + radeonEmitState(radeon); + + if (radeon->cmdbuf.cs->cdw) + rcommonFlushCmdBuf(radeon, __FUNCTION__); + + if ((ctx->DrawBuffer->Name == 0) && radeon->front_buffer_dirty) { + __DRIscreen *const screen = radeon->radeonScreen->driScreen; + + if (screen->dri2.loader && (screen->dri2.loader->base.version >= 2) + && (screen->dri2.loader->flushFrontBuffer != NULL)) { + __DRIdrawablePrivate * drawable = radeon_get_drawable(radeon); + (*screen->dri2.loader->flushFrontBuffer)(drawable, drawable->loaderPrivate); + + /* Only clear the dirty bit if front-buffer rendering is no longer + * enabled. This is done so that the dirty bit can only be set in + * glDrawBuffer. Otherwise the dirty bit would have to be set at + * each of N places that do rendering. This has worse performances, + * but it is much easier to get correct. + */ + if (!radeon->is_front_buffer_rendering) { + radeon->front_buffer_dirty = GL_FALSE; + } + } + } + + make_empty_list(&radeon->query.not_flushed_head); + +} + +/* Make sure all commands have been sent to the hardware and have + * completed processing. + */ +void radeonFinish(GLcontext * ctx) +{ + radeonContextPtr radeon = RADEON_CONTEXT(ctx); + struct gl_framebuffer *fb = ctx->DrawBuffer; + int i; + + if (ctx->Driver.Flush) + ctx->Driver.Flush(ctx); /* +r6/r7 */ + + if (radeon->radeonScreen->kernel_mm) { + for (i = 0; i < fb->_NumColorDrawBuffers; i++) { + struct radeon_renderbuffer *rrb; + rrb = radeon_renderbuffer(fb->_ColorDrawBuffers[i]); + if (rrb && rrb->bo) + radeon_bo_wait(rrb->bo); + } + { + struct radeon_renderbuffer *rrb; + rrb = radeon_get_depthbuffer(radeon); + if (rrb && rrb->bo) + radeon_bo_wait(rrb->bo); + } + } else if (radeon->do_irqs) { + LOCK_HARDWARE(radeon); + radeonEmitIrqLocked(radeon); + UNLOCK_HARDWARE(radeon); + radeonWaitIrq(radeon); + } else { + radeonWaitForIdle(radeon); + } +} + +/* cmdbuffer */ +/** + * Send the current command buffer via ioctl to the hardware. + */ +int rcommonFlushCmdBufLocked(radeonContextPtr rmesa, const char *caller) +{ + int ret = 0; + + if (rmesa->cmdbuf.flushing) { + fprintf(stderr, "Recursive call into r300FlushCmdBufLocked!\n"); + exit(-1); + } + rmesa->cmdbuf.flushing = 1; + + if (RADEON_DEBUG & RADEON_IOCTL) { + fprintf(stderr, "%s from %s - %i cliprects\n", + __FUNCTION__, caller, rmesa->numClipRects); + } + + radeonEmitQueryEnd(rmesa->glCtx); + + if (rmesa->cmdbuf.cs->cdw) { + ret = radeon_cs_emit(rmesa->cmdbuf.cs); + rmesa->hw.all_dirty = GL_TRUE; + } + radeon_cs_erase(rmesa->cmdbuf.cs); + rmesa->cmdbuf.flushing = 0; + + if (radeon_revalidate_bos(rmesa->glCtx) == GL_FALSE) { + fprintf(stderr,"failed to revalidate buffers\n"); + } + + return ret; +} + +int rcommonFlushCmdBuf(radeonContextPtr rmesa, const char *caller) +{ + int ret; + + radeonReleaseDmaRegions(rmesa); + + LOCK_HARDWARE(rmesa); + ret = rcommonFlushCmdBufLocked(rmesa, caller); + UNLOCK_HARDWARE(rmesa); + + if (ret) { + fprintf(stderr, "drmRadeonCmdBuffer: %d. Kernel failed to " + "parse or rejected command stream. See dmesg " + "for more info.\n", ret); + _mesa_exit(ret); + } + + return ret; +} + +/** + * Make sure that enough space is available in the command buffer + * by flushing if necessary. + * + * \param dwords The number of dwords we need to be free on the command buffer + */ +GLboolean rcommonEnsureCmdBufSpace(radeonContextPtr rmesa, int dwords, const char *caller) +{ + if ((rmesa->cmdbuf.cs->cdw + dwords + 128) > rmesa->cmdbuf.size + || radeon_cs_need_flush(rmesa->cmdbuf.cs)) { + /* If we try to flush empty buffer there is too big rendering operation. */ + assert(rmesa->cmdbuf.cs->cdw); + rcommonFlushCmdBuf(rmesa, caller); + return GL_TRUE; + } + return GL_FALSE; +} + +void rcommonInitCmdBuf(radeonContextPtr rmesa) +{ + GLuint size; + /* Initialize command buffer */ + size = 256 * driQueryOptioni(&rmesa->optionCache, + "command_buffer_size"); + if (size < 2 * rmesa->hw.max_state_size) { + size = 2 * rmesa->hw.max_state_size + 65535; + } + if (size > 64 * 256) + size = 64 * 256; + + radeon_print(RADEON_CS, RADEON_VERBOSE, + "sizeof(drm_r300_cmd_header_t)=%zd\n", sizeof(drm_r300_cmd_header_t)); + radeon_print(RADEON_CS, RADEON_VERBOSE, + "sizeof(drm_radeon_cmd_buffer_t)=%zd\n", sizeof(drm_radeon_cmd_buffer_t)); + radeon_print(RADEON_CS, RADEON_VERBOSE, + "Allocating %d bytes command buffer (max state is %d bytes)\n", + size * 4, rmesa->hw.max_state_size * 4); + + if (rmesa->radeonScreen->kernel_mm) { + int fd = rmesa->radeonScreen->driScreen->fd; + rmesa->cmdbuf.csm = radeon_cs_manager_gem_ctor(fd); + } else { + rmesa->cmdbuf.csm = radeon_cs_manager_legacy_ctor(rmesa); + } + if (rmesa->cmdbuf.csm == NULL) { + /* FIXME: fatal error */ + return; + } + rmesa->cmdbuf.cs = radeon_cs_create(rmesa->cmdbuf.csm, size); + assert(rmesa->cmdbuf.cs != NULL); + rmesa->cmdbuf.size = size; + + radeon_cs_space_set_flush(rmesa->cmdbuf.cs, + (void (*)(void *))rmesa->glCtx->Driver.Flush, rmesa->glCtx); + + if (!rmesa->radeonScreen->kernel_mm) { + radeon_cs_set_limit(rmesa->cmdbuf.cs, RADEON_GEM_DOMAIN_VRAM, rmesa->radeonScreen->texSize[0]); + radeon_cs_set_limit(rmesa->cmdbuf.cs, RADEON_GEM_DOMAIN_GTT, rmesa->radeonScreen->gartTextures.size); + } else { + struct drm_radeon_gem_info mminfo = { 0 }; + + if (!drmCommandWriteRead(rmesa->dri.fd, DRM_RADEON_GEM_INFO, &mminfo, sizeof(mminfo))) + { + radeon_cs_set_limit(rmesa->cmdbuf.cs, RADEON_GEM_DOMAIN_VRAM, mminfo.vram_visible); + radeon_cs_set_limit(rmesa->cmdbuf.cs, RADEON_GEM_DOMAIN_GTT, mminfo.gart_size); + } + } + +} +/** + * Destroy the command buffer + */ +void rcommonDestroyCmdBuf(radeonContextPtr rmesa) +{ + radeon_cs_destroy(rmesa->cmdbuf.cs); + if (rmesa->radeonScreen->driScreen->dri2.enabled || rmesa->radeonScreen->kernel_mm) { + radeon_cs_manager_gem_dtor(rmesa->cmdbuf.csm); + } else { + radeon_cs_manager_legacy_dtor(rmesa->cmdbuf.csm); + } +} + +void rcommonBeginBatch(radeonContextPtr rmesa, int n, + int dostate, + const char *file, + const char *function, + int line) +{ + if (!rmesa->cmdbuf.cs->cdw && dostate) { + radeon_print(RADEON_STATE, RADEON_NORMAL, + "Reemit state after flush (from %s)\n", function); + radeonEmitState(rmesa); + } + radeon_cs_begin(rmesa->cmdbuf.cs, n, file, function, line); + + radeon_print(RADEON_CS, RADEON_VERBOSE, "BEGIN_BATCH(%d) at %d, from %s:%i\n", + n, rmesa->cmdbuf.cs->cdw, function, line); + +} + +void radeonUserClear(GLcontext *ctx, GLuint mask) +{ + _mesa_meta_clear(ctx, mask); +} diff --git a/src/mesa/drivers/dri/radeon/radeon_common.h b/src/mesa/drivers/dri/radeon/radeon_common.h new file mode 100644 index 00000000000..f3201911ac6 --- /dev/null +++ b/src/mesa/drivers/dri/radeon/radeon_common.h @@ -0,0 +1,87 @@ +#ifndef COMMON_MISC_H +#define COMMON_MISC_H + +#include "radeon_common_context.h" +#include "radeon_dma.h" +#include "radeon_texture.h" + +void radeonUserClear(GLcontext *ctx, GLuint mask); +void radeonRecalcScissorRects(radeonContextPtr radeon); +void radeonSetCliprects(radeonContextPtr radeon); +void radeonUpdateScissor( GLcontext *ctx ); +void radeonScissor(GLcontext* ctx, GLint x, GLint y, GLsizei w, GLsizei h); +void radeonPolygonStipplePreKMS( GLcontext *ctx, const GLubyte *mask ); + +void radeonWaitForIdleLocked(radeonContextPtr radeon); +extern uint32_t radeonGetAge(radeonContextPtr radeon); +void radeonCopyBuffer( __DRIdrawablePrivate *dPriv, + const drm_clip_rect_t *rect); +void radeonSwapBuffers(__DRIdrawablePrivate * dPriv); +void radeonCopySubBuffer(__DRIdrawablePrivate * dPriv, + int x, int y, int w, int h ); + +void radeonUpdatePageFlipping(radeonContextPtr rmesa); + +void radeonFlush(GLcontext *ctx); +void radeonFinish(GLcontext * ctx); +void radeonEmitState(radeonContextPtr radeon); +GLuint radeonCountStateEmitSize(radeonContextPtr radeon); + +void radeon_clear_tris(GLcontext *ctx, GLbitfield mask); + +void radeon_window_moved(radeonContextPtr radeon); +void radeon_draw_buffer(GLcontext *ctx, struct gl_framebuffer *fb); +void radeonDrawBuffer( GLcontext *ctx, GLenum mode ); +void radeonReadBuffer( GLcontext *ctx, GLenum mode ); +void radeon_viewport(GLcontext *ctx, GLint x, GLint y, GLsizei width, GLsizei height); +void radeon_get_cliprects(radeonContextPtr radeon, + struct drm_clip_rect **cliprects, + unsigned int *num_cliprects, + int *x_off, int *y_off); +void radeon_fbo_init(struct radeon_context *radeon); +void +radeon_renderbuffer_set_bo(struct radeon_renderbuffer *rb, + struct radeon_bo *bo); +struct radeon_renderbuffer * +radeon_create_renderbuffer(GLenum format, __DRIdrawablePrivate *driDrawPriv); +static inline struct radeon_renderbuffer *radeon_renderbuffer(struct gl_renderbuffer *rb) +{ + struct radeon_renderbuffer *rrb = (struct radeon_renderbuffer *)rb; + if (rrb && rrb->base.ClassID == RADEON_RB_CLASS) + return rrb; + else + return NULL; +} + +static inline struct radeon_renderbuffer *radeon_get_renderbuffer(struct gl_framebuffer *fb, int att_index) +{ + if (att_index >= 0) + return radeon_renderbuffer(fb->Attachment[att_index].Renderbuffer); + else + return NULL; +} + +static inline struct radeon_renderbuffer *radeon_get_depthbuffer(radeonContextPtr rmesa) +{ + struct radeon_renderbuffer *rrb; + rrb = radeon_renderbuffer(rmesa->state.depth.rb); + if (!rrb) + return NULL; + + return rrb; +} + +static inline struct radeon_renderbuffer *radeon_get_colorbuffer(radeonContextPtr rmesa) +{ + struct radeon_renderbuffer *rrb; + + rrb = radeon_renderbuffer(rmesa->state.color.rb); + if (!rrb) + return NULL; + return rrb; +} + +#include "radeon_cmdbuf.h" + + +#endif diff --git a/src/mesa/drivers/dri/radeon/radeon_common_context.c b/src/mesa/drivers/dri/radeon/radeon_common_context.c new file mode 100644 index 00000000000..71ee06d9a79 --- /dev/null +++ b/src/mesa/drivers/dri/radeon/radeon_common_context.c @@ -0,0 +1,805 @@ +/************************************************************************** + +Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and + VA Linux Systems Inc., Fremont, California. +Copyright (C) The Weather Channel, Inc. 2002. All Rights Reserved. + +The Weather Channel (TM) funded Tungsten Graphics to develop the +initial release of the Radeon 8500 driver under the XFree86 license. +This notice must be preserved. + +All Rights Reserved. + +Permission is hereby granted, free of charge, to any person obtaining +a copy of this software and associated documentation files (the +"Software"), to deal in the Software without restriction, including +without limitation the rights to use, copy, modify, merge, publish, +distribute, sublicense, and/or sell copies of the Software, and to +permit persons to whom the Software is furnished to do so, subject to +the following conditions: + +The above copyright notice and this permission notice (including the +next paragraph) shall be included in all copies or substantial +portions of the Software. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, +EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF +MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. +IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE +LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION +OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION +WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + +**************************************************************************/ + +#include "radeon_common.h" +#include "xmlpool.h" /* for symbolic values of enum-type options */ +#include "utils.h" +#include "vblank.h" +#include "drirenderbuffer.h" +#include "drivers/common/meta.h" +#include "main/context.h" +#include "main/framebuffer.h" +#include "main/renderbuffer.h" +#include "main/state.h" +#include "main/simple_list.h" +#include "swrast/swrast.h" +#include "swrast_setup/swrast_setup.h" +#include "tnl/tnl.h" + +#if RADEON_COMMON && defined(RADEON_COMMON_FOR_R600) /* +r6/r7 */ +#include "r600_context.h" +#endif + +#define DRIVER_DATE "20090101" + +#ifndef RADEON_DEBUG +int RADEON_DEBUG = (0); +#endif + + +static const char* get_chip_family_name(int chip_family) +{ + switch(chip_family) { + case CHIP_FAMILY_R100: return "R100"; + case CHIP_FAMILY_RV100: return "RV100"; + case CHIP_FAMILY_RS100: return "RS100"; + case CHIP_FAMILY_RV200: return "RV200"; + case CHIP_FAMILY_RS200: return "RS200"; + case CHIP_FAMILY_R200: return "R200"; + case CHIP_FAMILY_RV250: return "RV250"; + case CHIP_FAMILY_RS300: return "RS300"; + case CHIP_FAMILY_RV280: return "RV280"; + case CHIP_FAMILY_R300: return "R300"; + case CHIP_FAMILY_R350: return "R350"; + case CHIP_FAMILY_RV350: return "RV350"; + case CHIP_FAMILY_RV380: return "RV380"; + case CHIP_FAMILY_R420: return "R420"; + case CHIP_FAMILY_RV410: return "RV410"; + case CHIP_FAMILY_RS400: return "RS400"; + case CHIP_FAMILY_RS600: return "RS600"; + case CHIP_FAMILY_RS690: return "RS690"; + case CHIP_FAMILY_RS740: return "RS740"; + case CHIP_FAMILY_RV515: return "RV515"; + case CHIP_FAMILY_R520: return "R520"; + case CHIP_FAMILY_RV530: return "RV530"; + case CHIP_FAMILY_R580: return "R580"; + case CHIP_FAMILY_RV560: return "RV560"; + case CHIP_FAMILY_RV570: return "RV570"; + case CHIP_FAMILY_R600: return "R600"; + case CHIP_FAMILY_RV610: return "RV610"; + case CHIP_FAMILY_RV630: return "RV630"; + case CHIP_FAMILY_RV670: return "RV670"; + case CHIP_FAMILY_RV620: return "RV620"; + case CHIP_FAMILY_RV635: return "RV635"; + case CHIP_FAMILY_RS780: return "RS780"; + case CHIP_FAMILY_RS880: return "RS880"; + case CHIP_FAMILY_RV770: return "RV770"; + case CHIP_FAMILY_RV730: return "RV730"; + case CHIP_FAMILY_RV710: return "RV710"; + case CHIP_FAMILY_RV740: return "RV740"; + default: return "unknown"; + } +} + + +/* Return various strings for glGetString(). + */ +static const GLubyte *radeonGetString(GLcontext * ctx, GLenum name) +{ + radeonContextPtr radeon = RADEON_CONTEXT(ctx); + static char buffer[128]; + + switch (name) { + case GL_VENDOR: + if (IS_R600_CLASS(radeon->radeonScreen)) + return (GLubyte *) "Advanced Micro Devices, Inc."; + else if (IS_R300_CLASS(radeon->radeonScreen)) + return (GLubyte *) "DRI R300 Project"; + else + return (GLubyte *) "Tungsten Graphics, Inc."; + + case GL_RENDERER: + { + unsigned offset; + GLuint agp_mode = (radeon->radeonScreen->card_type==RADEON_CARD_PCI) ? 0 : + radeon->radeonScreen->AGPMode; + const char* chipclass; + char hardwarename[32]; + + if (IS_R600_CLASS(radeon->radeonScreen)) + chipclass = "R600"; + else if (IS_R300_CLASS(radeon->radeonScreen)) + chipclass = "R300"; + else if (IS_R200_CLASS(radeon->radeonScreen)) + chipclass = "R200"; + else + chipclass = "R100"; + + sprintf(hardwarename, "%s (%s %04X)", + chipclass, + get_chip_family_name(radeon->radeonScreen->chip_family), + radeon->radeonScreen->device_id); + + offset = driGetRendererString(buffer, hardwarename, DRIVER_DATE, + agp_mode); + + if (IS_R600_CLASS(radeon->radeonScreen)) { + sprintf(&buffer[offset], " TCL"); + } else if (IS_R300_CLASS(radeon->radeonScreen)) { + sprintf(&buffer[offset], " %sTCL", + (radeon->radeonScreen->chip_flags & RADEON_CHIPSET_TCL) + ? "" : "NO-"); + } else { + sprintf(&buffer[offset], " %sTCL", + !(radeon->TclFallback & RADEON_TCL_FALLBACK_TCL_DISABLE) + ? "" : "NO-"); + } + + if (radeon->radeonScreen->driScreen->dri2.enabled) + strcat(buffer, " DRI2"); + + return (GLubyte *) buffer; + } + + default: + return NULL; + } +} + +/* Initialize the driver's misc functions. + */ +static void radeonInitDriverFuncs(struct dd_function_table *functions) +{ + functions->GetString = radeonGetString; +} + +/** + * Create and initialize all common fields of the context, + * including the Mesa context itself. + */ +GLboolean radeonInitContext(radeonContextPtr radeon, + struct dd_function_table* functions, + const __GLcontextModes * glVisual, + __DRIcontextPrivate * driContextPriv, + void *sharedContextPrivate) +{ + __DRIscreenPrivate *sPriv = driContextPriv->driScreenPriv; + radeonScreenPtr screen = (radeonScreenPtr) (sPriv->private); + GLcontext* ctx; + GLcontext* shareCtx; + int fthrottle_mode; + + /* Fill in additional standard functions. */ + radeonInitDriverFuncs(functions); + + radeon->radeonScreen = screen; + /* Allocate and initialize the Mesa context */ + if (sharedContextPrivate) + shareCtx = ((radeonContextPtr)sharedContextPrivate)->glCtx; + else + shareCtx = NULL; + radeon->glCtx = _mesa_create_context(glVisual, shareCtx, + functions, (void *)radeon); + if (!radeon->glCtx) + return GL_FALSE; + + ctx = radeon->glCtx; + driContextPriv->driverPrivate = radeon; + + meta_init_metaops(ctx, &radeon->meta); + + _mesa_meta_init(ctx); + + /* DRI fields */ + radeon->dri.context = driContextPriv; + radeon->dri.screen = sPriv; + radeon->dri.hwContext = driContextPriv->hHWContext; + radeon->dri.hwLock = &sPriv->pSAREA->lock; + radeon->dri.hwLockCount = 0; + radeon->dri.fd = sPriv->fd; + radeon->dri.drmMinor = sPriv->drm_version.minor; + + radeon->sarea = (drm_radeon_sarea_t *) ((GLubyte *) sPriv->pSAREA + + screen->sarea_priv_offset); + + /* Setup IRQs */ + fthrottle_mode = driQueryOptioni(&radeon->optionCache, "fthrottle_mode"); + radeon->iw.irq_seq = -1; + radeon->irqsEmitted = 0; + if (IS_R600_CLASS(radeon->radeonScreen)) + radeon->do_irqs = 0; + else + radeon->do_irqs = (fthrottle_mode == DRI_CONF_FTHROTTLE_IRQS && + radeon->radeonScreen->irq); + + radeon->do_usleeps = (fthrottle_mode == DRI_CONF_FTHROTTLE_USLEEPS); + + if (!radeon->do_irqs) + fprintf(stderr, + "IRQ's not enabled, falling back to %s: %d %d\n", + radeon->do_usleeps ? "usleeps" : "busy waits", + fthrottle_mode, radeon->radeonScreen->irq); + + radeon->texture_depth = driQueryOptioni (&radeon->optionCache, + "texture_depth"); + if (radeon->texture_depth == DRI_CONF_TEXTURE_DEPTH_FB) + radeon->texture_depth = ( glVisual->rgbBits > 16 ) ? + DRI_CONF_TEXTURE_DEPTH_32 : DRI_CONF_TEXTURE_DEPTH_16; + + if (IS_R600_CLASS(radeon->radeonScreen)) { + radeon->texture_row_align = 256; + radeon->texture_rect_row_align = 256; + radeon->texture_compressed_row_align = 256; + } else if (IS_R200_CLASS(radeon->radeonScreen) || + IS_R100_CLASS(radeon->radeonScreen)) { + radeon->texture_row_align = 32; + radeon->texture_rect_row_align = 64; + radeon->texture_compressed_row_align = 32; + } else { /* R300 - not sure this is all correct */ + int chip_family = radeon->radeonScreen->chip_family; + if (chip_family == CHIP_FAMILY_RS600 || + chip_family == CHIP_FAMILY_RS690 || + chip_family == CHIP_FAMILY_RS740) + radeon->texture_row_align = 64; + else + radeon->texture_row_align = 32; + radeon->texture_rect_row_align = 64; + radeon->texture_compressed_row_align = 64; + } + + make_empty_list(&radeon->query.not_flushed_head); + radeon_init_dma(radeon); + + return GL_TRUE; +} + + + +/** + * Destroy the command buffer and state atoms. + */ +static void radeon_destroy_atom_list(radeonContextPtr radeon) +{ + struct radeon_state_atom *atom; + + foreach(atom, &radeon->hw.atomlist) { + FREE(atom->cmd); + if (atom->lastcmd) + FREE(atom->lastcmd); + } + +} + +/** + * Cleanup common context fields. + * Called by r200DestroyContext/r300DestroyContext + */ +void radeonDestroyContext(__DRIcontextPrivate *driContextPriv ) +{ +#ifdef RADEON_BO_TRACK + FILE *track; +#endif + GET_CURRENT_CONTEXT(ctx); + radeonContextPtr radeon = (radeonContextPtr) driContextPriv->driverPrivate; + radeonContextPtr current = ctx ? RADEON_CONTEXT(ctx) : NULL; + + assert(radeon); + + _mesa_meta_free(radeon->glCtx); + + if (radeon == current) { + radeon_firevertices(radeon); + _mesa_make_current(NULL, NULL, NULL); + } + + if (!is_empty_list(&radeon->dma.reserved)) { + rcommonFlushCmdBuf( radeon, __FUNCTION__ ); + } + + radeonFreeDmaRegions(radeon); + radeonReleaseArrays(radeon->glCtx, ~0); + meta_destroy_metaops(&radeon->meta); + if (radeon->vtbl.free_context) + radeon->vtbl.free_context(radeon->glCtx); + _swsetup_DestroyContext( radeon->glCtx ); + _tnl_DestroyContext( radeon->glCtx ); + _vbo_DestroyContext( radeon->glCtx ); + _swrast_DestroyContext( radeon->glCtx ); + + /* free atom list */ + /* free the Mesa context */ + _mesa_destroy_context(radeon->glCtx); + + /* _mesa_destroy_context() might result in calls to functions that + * depend on the DriverCtx, so don't set it to NULL before. + * + * radeon->glCtx->DriverCtx = NULL; + */ + /* free the option cache */ + driDestroyOptionCache(&radeon->optionCache); + + rcommonDestroyCmdBuf(radeon); + + radeon_destroy_atom_list(radeon); + + if (radeon->state.scissor.pClipRects) { + FREE(radeon->state.scissor.pClipRects); + radeon->state.scissor.pClipRects = 0; + } +#ifdef RADEON_BO_TRACK + track = fopen("/tmp/tracklog", "w"); + if (track) { + radeon_tracker_print(&radeon->radeonScreen->bom->tracker, track); + fclose(track); + } +#endif + FREE(radeon); +} + +/* Force the context `c' to be unbound from its buffer. + */ +GLboolean radeonUnbindContext(__DRIcontextPrivate * driContextPriv) +{ + radeonContextPtr radeon = (radeonContextPtr) driContextPriv->driverPrivate; + + if (RADEON_DEBUG & RADEON_DRI) + fprintf(stderr, "%s ctx %p\n", __FUNCTION__, + radeon->glCtx); + + return GL_TRUE; +} + + +static void +radeon_make_kernel_renderbuffer_current(radeonContextPtr radeon, + struct radeon_framebuffer *draw) +{ + /* if radeon->fake */ + struct radeon_renderbuffer *rb; + + if ((rb = (void *)draw->base.Attachment[BUFFER_FRONT_LEFT].Renderbuffer)) { + if (!rb->bo) { + rb->bo = radeon_bo_open(radeon->radeonScreen->bom, + radeon->radeonScreen->frontOffset, + 0, + 0, + RADEON_GEM_DOMAIN_VRAM, + 0); + } + rb->cpp = radeon->radeonScreen->cpp; + rb->pitch = radeon->radeonScreen->frontPitch * rb->cpp; + } + if ((rb = (void *)draw->base.Attachment[BUFFER_BACK_LEFT].Renderbuffer)) { + if (!rb->bo) { + rb->bo = radeon_bo_open(radeon->radeonScreen->bom, + radeon->radeonScreen->backOffset, + 0, + 0, + RADEON_GEM_DOMAIN_VRAM, + 0); + } + rb->cpp = radeon->radeonScreen->cpp; + rb->pitch = radeon->radeonScreen->backPitch * rb->cpp; + } + if ((rb = (void *)draw->base.Attachment[BUFFER_DEPTH].Renderbuffer)) { + if (!rb->bo) { + rb->bo = radeon_bo_open(radeon->radeonScreen->bom, + radeon->radeonScreen->depthOffset, + 0, + 0, + RADEON_GEM_DOMAIN_VRAM, + 0); + } + rb->cpp = radeon->radeonScreen->cpp; + rb->pitch = radeon->radeonScreen->depthPitch * rb->cpp; + } + if ((rb = (void *)draw->base.Attachment[BUFFER_STENCIL].Renderbuffer)) { + if (!rb->bo) { + rb->bo = radeon_bo_open(radeon->radeonScreen->bom, + radeon->radeonScreen->depthOffset, + 0, + 0, + RADEON_GEM_DOMAIN_VRAM, + 0); + } + rb->cpp = radeon->radeonScreen->cpp; + rb->pitch = radeon->radeonScreen->depthPitch * rb->cpp; + } +} + +static void +radeon_make_renderbuffer_current(radeonContextPtr radeon, + struct radeon_framebuffer *draw) +{ + int size = 4096*4096*4; + /* if radeon->fake */ + struct radeon_renderbuffer *rb; + + if (radeon->radeonScreen->kernel_mm) { + radeon_make_kernel_renderbuffer_current(radeon, draw); + return; + } + + + if ((rb = (void *)draw->base.Attachment[BUFFER_FRONT_LEFT].Renderbuffer)) { + if (!rb->bo) { + rb->bo = radeon_bo_open(radeon->radeonScreen->bom, + radeon->radeonScreen->frontOffset + + radeon->radeonScreen->fbLocation, + size, + 4096, + RADEON_GEM_DOMAIN_VRAM, + 0); + } + rb->cpp = radeon->radeonScreen->cpp; + rb->pitch = radeon->radeonScreen->frontPitch * rb->cpp; + } + if ((rb = (void *)draw->base.Attachment[BUFFER_BACK_LEFT].Renderbuffer)) { + if (!rb->bo) { + rb->bo = radeon_bo_open(radeon->radeonScreen->bom, + radeon->radeonScreen->backOffset + + radeon->radeonScreen->fbLocation, + size, + 4096, + RADEON_GEM_DOMAIN_VRAM, + 0); + } + rb->cpp = radeon->radeonScreen->cpp; + rb->pitch = radeon->radeonScreen->backPitch * rb->cpp; + } + if ((rb = (void *)draw->base.Attachment[BUFFER_DEPTH].Renderbuffer)) { + if (!rb->bo) { + rb->bo = radeon_bo_open(radeon->radeonScreen->bom, + radeon->radeonScreen->depthOffset + + radeon->radeonScreen->fbLocation, + size, + 4096, + RADEON_GEM_DOMAIN_VRAM, + 0); + } + rb->cpp = radeon->radeonScreen->cpp; + rb->pitch = radeon->radeonScreen->depthPitch * rb->cpp; + } + if ((rb = (void *)draw->base.Attachment[BUFFER_STENCIL].Renderbuffer)) { + if (!rb->bo) { + rb->bo = radeon_bo_open(radeon->radeonScreen->bom, + radeon->radeonScreen->depthOffset + + radeon->radeonScreen->fbLocation, + size, + 4096, + RADEON_GEM_DOMAIN_VRAM, + 0); + } + rb->cpp = radeon->radeonScreen->cpp; + rb->pitch = radeon->radeonScreen->depthPitch * rb->cpp; + } +} + +static unsigned +radeon_bits_per_pixel(const struct radeon_renderbuffer *rb) +{ + switch (rb->base._ActualFormat) { + case GL_RGB5: + case GL_DEPTH_COMPONENT16: + return 16; + case GL_RGB8: + case GL_RGBA8: + case GL_DEPTH_COMPONENT24: + case GL_DEPTH24_STENCIL8_EXT: + case GL_STENCIL_INDEX8_EXT: + return 32; + default: + return 0; + } +} + +void +radeon_update_renderbuffers(__DRIcontext *context, __DRIdrawable *drawable) +{ + unsigned int attachments[10]; + __DRIbuffer *buffers = NULL; + __DRIscreen *screen; + struct radeon_renderbuffer *rb; + int i, count; + struct radeon_framebuffer *draw; + radeonContextPtr radeon; + char *regname; + struct radeon_bo *depth_bo = NULL, *bo; + + if (RADEON_DEBUG & RADEON_DRI) + fprintf(stderr, "enter %s, drawable %p\n", __func__, drawable); + + draw = drawable->driverPrivate; + screen = context->driScreenPriv; + radeon = (radeonContextPtr) context->driverPrivate; + + if (screen->dri2.loader + && (screen->dri2.loader->base.version > 2) + && (screen->dri2.loader->getBuffersWithFormat != NULL)) { + struct radeon_renderbuffer *depth_rb; + struct radeon_renderbuffer *stencil_rb; + + i = 0; + if ((radeon->is_front_buffer_rendering || + radeon->is_front_buffer_reading || + !draw->color_rb[1]) + && draw->color_rb[0]) { + attachments[i++] = __DRI_BUFFER_FRONT_LEFT; + attachments[i++] = radeon_bits_per_pixel(draw->color_rb[0]); + } + + if (draw->color_rb[1]) { + attachments[i++] = __DRI_BUFFER_BACK_LEFT; + attachments[i++] = radeon_bits_per_pixel(draw->color_rb[1]); + } + + depth_rb = radeon_get_renderbuffer(&draw->base, BUFFER_DEPTH); + stencil_rb = radeon_get_renderbuffer(&draw->base, BUFFER_STENCIL); + + if ((depth_rb != NULL) && (stencil_rb != NULL)) { + attachments[i++] = __DRI_BUFFER_DEPTH_STENCIL; + attachments[i++] = radeon_bits_per_pixel(depth_rb); + } else if (depth_rb != NULL) { + attachments[i++] = __DRI_BUFFER_DEPTH; + attachments[i++] = radeon_bits_per_pixel(depth_rb); + } else if (stencil_rb != NULL) { + attachments[i++] = __DRI_BUFFER_STENCIL; + attachments[i++] = radeon_bits_per_pixel(stencil_rb); + } + + buffers = (*screen->dri2.loader->getBuffersWithFormat)(drawable, + &drawable->w, + &drawable->h, + attachments, i / 2, + &count, + drawable->loaderPrivate); + } else if (screen->dri2.loader) { + i = 0; + if (draw->color_rb[0]) + attachments[i++] = __DRI_BUFFER_FRONT_LEFT; + if (draw->color_rb[1]) + attachments[i++] = __DRI_BUFFER_BACK_LEFT; + if (radeon_get_renderbuffer(&draw->base, BUFFER_DEPTH)) + attachments[i++] = __DRI_BUFFER_DEPTH; + if (radeon_get_renderbuffer(&draw->base, BUFFER_STENCIL)) + attachments[i++] = __DRI_BUFFER_STENCIL; + + buffers = (*screen->dri2.loader->getBuffers)(drawable, + &drawable->w, + &drawable->h, + attachments, i, + &count, + drawable->loaderPrivate); + } + + if (buffers == NULL) + return; + + /* set one cliprect to cover the whole drawable */ + drawable->x = 0; + drawable->y = 0; + drawable->backX = 0; + drawable->backY = 0; + drawable->numClipRects = 1; + drawable->pClipRects[0].x1 = 0; + drawable->pClipRects[0].y1 = 0; + drawable->pClipRects[0].x2 = drawable->w; + drawable->pClipRects[0].y2 = drawable->h; + drawable->numBackClipRects = 1; + drawable->pBackClipRects[0].x1 = 0; + drawable->pBackClipRects[0].y1 = 0; + drawable->pBackClipRects[0].x2 = drawable->w; + drawable->pBackClipRects[0].y2 = drawable->h; + for (i = 0; i < count; i++) { + switch (buffers[i].attachment) { + case __DRI_BUFFER_FRONT_LEFT: + rb = draw->color_rb[0]; + regname = "dri2 front buffer"; + break; + case __DRI_BUFFER_FAKE_FRONT_LEFT: + rb = draw->color_rb[0]; + regname = "dri2 fake front buffer"; + break; + case __DRI_BUFFER_BACK_LEFT: + rb = draw->color_rb[1]; + regname = "dri2 back buffer"; + break; + case __DRI_BUFFER_DEPTH: + rb = radeon_get_renderbuffer(&draw->base, BUFFER_DEPTH); + regname = "dri2 depth buffer"; + break; + case __DRI_BUFFER_DEPTH_STENCIL: + rb = radeon_get_renderbuffer(&draw->base, BUFFER_DEPTH); + regname = "dri2 depth / stencil buffer"; + break; + case __DRI_BUFFER_STENCIL: + rb = radeon_get_renderbuffer(&draw->base, BUFFER_STENCIL); + regname = "dri2 stencil buffer"; + break; + case __DRI_BUFFER_ACCUM: + default: + fprintf(stderr, + "unhandled buffer attach event, attacment type %d\n", + buffers[i].attachment); + return; + } + + if (rb == NULL) + continue; + + if (rb->bo) { + uint32_t name = radeon_gem_name_bo(rb->bo); + if (name == buffers[i].name) + continue; + } + + if (RADEON_DEBUG & RADEON_DRI) + fprintf(stderr, + "attaching buffer %s, %d, at %d, cpp %d, pitch %d\n", + regname, buffers[i].name, buffers[i].attachment, + buffers[i].cpp, buffers[i].pitch); + + rb->cpp = buffers[i].cpp; + rb->pitch = buffers[i].pitch; + rb->base.Width = drawable->w; + rb->base.Height = drawable->h; + rb->has_surface = 0; + + if (buffers[i].attachment == __DRI_BUFFER_STENCIL && depth_bo) { + if (RADEON_DEBUG & RADEON_DRI) + fprintf(stderr, "(reusing depth buffer as stencil)\n"); + bo = depth_bo; + radeon_bo_ref(bo); + } else { + uint32_t tiling_flags = 0, pitch = 0; + int ret; + + bo = radeon_bo_open(radeon->radeonScreen->bom, + buffers[i].name, + 0, + 0, + RADEON_GEM_DOMAIN_VRAM, + buffers[i].flags); + + if (bo == NULL) { + + fprintf(stderr, "failed to attach %s %d\n", + regname, buffers[i].name); + + } + + ret = radeon_bo_get_tiling(bo, &tiling_flags, &pitch); + if (tiling_flags & RADEON_TILING_MACRO) + bo->flags |= RADEON_BO_FLAGS_MACRO_TILE; + if (tiling_flags & RADEON_TILING_MICRO) + bo->flags |= RADEON_BO_FLAGS_MICRO_TILE; + + } + + if (buffers[i].attachment == __DRI_BUFFER_DEPTH) { + if (draw->base.Visual.depthBits == 16) + rb->cpp = 2; + depth_bo = bo; + } + + radeon_renderbuffer_set_bo(rb, bo); + radeon_bo_unref(bo); + + if (buffers[i].attachment == __DRI_BUFFER_DEPTH_STENCIL) { + rb = radeon_get_renderbuffer(&draw->base, BUFFER_STENCIL); + if (rb != NULL) { + struct radeon_bo *stencil_bo = NULL; + + if (rb->bo) { + uint32_t name = radeon_gem_name_bo(rb->bo); + if (name == buffers[i].name) + continue; + } + + stencil_bo = bo; + radeon_bo_ref(stencil_bo); + radeon_renderbuffer_set_bo(rb, stencil_bo); + radeon_bo_unref(stencil_bo); + } + } + } + + driUpdateFramebufferSize(radeon->glCtx, drawable); +} + +/* Force the context `c' to be the current context and associate with it + * buffer `b'. + */ +GLboolean radeonMakeCurrent(__DRIcontextPrivate * driContextPriv, + __DRIdrawablePrivate * driDrawPriv, + __DRIdrawablePrivate * driReadPriv) +{ + radeonContextPtr radeon; + struct radeon_framebuffer *drfb; + struct gl_framebuffer *readfb; + + if (!driContextPriv) { + if (RADEON_DEBUG & RADEON_DRI) + fprintf(stderr, "%s ctx is null\n", __FUNCTION__); + _mesa_make_current(NULL, NULL, NULL); + return GL_TRUE; + } + + radeon = (radeonContextPtr) driContextPriv->driverPrivate; + drfb = driDrawPriv->driverPrivate; + readfb = driReadPriv->driverPrivate; + + if (driContextPriv->driScreenPriv->dri2.enabled) { + radeon_update_renderbuffers(driContextPriv, driDrawPriv); + if (driDrawPriv != driReadPriv) + radeon_update_renderbuffers(driContextPriv, driReadPriv); + _mesa_reference_renderbuffer(&radeon->state.color.rb, + &(radeon_get_renderbuffer(&drfb->base, BUFFER_BACK_LEFT)->base)); + _mesa_reference_renderbuffer(&radeon->state.depth.rb, + &(radeon_get_renderbuffer(&drfb->base, BUFFER_DEPTH)->base)); + } else { + radeon_make_renderbuffer_current(radeon, drfb); + } + + if (RADEON_DEBUG & RADEON_DRI) + fprintf(stderr, "%s ctx %p dfb %p rfb %p\n", __FUNCTION__, radeon->glCtx, drfb, readfb); + + driUpdateFramebufferSize(radeon->glCtx, driDrawPriv); + if (driReadPriv != driDrawPriv) + driUpdateFramebufferSize(radeon->glCtx, driReadPriv); + + _mesa_make_current(radeon->glCtx, &drfb->base, readfb); + + _mesa_update_state(radeon->glCtx); + + if (radeon->glCtx->DrawBuffer == &drfb->base) { + if (driDrawPriv->swap_interval == (unsigned)-1) { + int i; + driDrawPriv->vblFlags = + (radeon->radeonScreen->irq != 0) + ? driGetDefaultVBlankFlags(&radeon-> + optionCache) + : VBLANK_FLAG_NO_IRQ; + + driDrawableInitVBlank(driDrawPriv); + drfb->vbl_waited = driDrawPriv->vblSeq; + + for (i = 0; i < 2; i++) { + if (drfb->color_rb[i]) + drfb->color_rb[i]->vbl_pending = driDrawPriv->vblSeq; + } + + } + + radeon_window_moved(radeon); + radeon_draw_buffer(radeon->glCtx, &drfb->base); + } + + + if (RADEON_DEBUG & RADEON_DRI) + fprintf(stderr, "End %s\n", __FUNCTION__); + + return GL_TRUE; +} + diff --git a/src/mesa/drivers/dri/radeon/radeon_common_context.h b/src/mesa/drivers/dri/radeon/radeon_common_context.h new file mode 100644 index 00000000000..0309345393d --- /dev/null +++ b/src/mesa/drivers/dri/radeon/radeon_common_context.h @@ -0,0 +1,594 @@ + +#ifndef COMMON_CONTEXT_H +#define COMMON_CONTEXT_H + +#include "main/mm.h" +#include "math/m_vector.h" +#include "texmem.h" +#include "tnl/t_context.h" +#include "main/colormac.h" + +#include "radeon_debug.h" +#include "radeon_screen.h" +#include "radeon_drm.h" +#include "dri_util.h" +#include "tnl/t_vertex.h" + +#include "dri_metaops.h" +struct radeon_context; + +#include "radeon_bocs_wrapper.h" + +/* This union is used to avoid warnings/miscompilation + with float to uint32_t casts due to strict-aliasing */ +typedef union { GLfloat f; uint32_t ui32; } float_ui32_type; + +struct radeon_context; +typedef struct radeon_context radeonContextRec; +typedef struct radeon_context *radeonContextPtr; + + +#define TEX_0 0x1 +#define TEX_1 0x2 +#define TEX_2 0x4 +#define TEX_3 0x8 +#define TEX_4 0x10 +#define TEX_5 0x20 + +/* Rasterizing fallbacks */ +/* See correponding strings in r200_swtcl.c */ +#define RADEON_FALLBACK_TEXTURE 0x0001 +#define RADEON_FALLBACK_DRAW_BUFFER 0x0002 +#define RADEON_FALLBACK_STENCIL 0x0004 +#define RADEON_FALLBACK_RENDER_MODE 0x0008 +#define RADEON_FALLBACK_BLEND_EQ 0x0010 +#define RADEON_FALLBACK_BLEND_FUNC 0x0020 +#define RADEON_FALLBACK_DISABLE 0x0040 +#define RADEON_FALLBACK_BORDER_MODE 0x0080 +#define RADEON_FALLBACK_DEPTH_BUFFER 0x0100 +#define RADEON_FALLBACK_STENCIL_BUFFER 0x0200 + +#define R200_FALLBACK_TEXTURE 0x01 +#define R200_FALLBACK_DRAW_BUFFER 0x02 +#define R200_FALLBACK_STENCIL 0x04 +#define R200_FALLBACK_RENDER_MODE 0x08 +#define R200_FALLBACK_DISABLE 0x10 +#define R200_FALLBACK_BORDER_MODE 0x20 + +#define RADEON_TCL_FALLBACK_RASTER 0x1 /* rasterization */ +#define RADEON_TCL_FALLBACK_UNFILLED 0x2 /* unfilled tris */ +#define RADEON_TCL_FALLBACK_LIGHT_TWOSIDE 0x4 /* twoside tris */ +#define RADEON_TCL_FALLBACK_MATERIAL 0x8 /* material in vb */ +#define RADEON_TCL_FALLBACK_TEXGEN_0 0x10 /* texgen, unit 0 */ +#define RADEON_TCL_FALLBACK_TEXGEN_1 0x20 /* texgen, unit 1 */ +#define RADEON_TCL_FALLBACK_TEXGEN_2 0x40 /* texgen, unit 2 */ +#define RADEON_TCL_FALLBACK_TCL_DISABLE 0x80 /* user disable */ +#define RADEON_TCL_FALLBACK_FOGCOORDSPEC 0x100 /* fogcoord, sep. spec light */ + +/* The blit width for texture uploads + */ +#define BLIT_WIDTH_BYTES 1024 + +/* Use the templated vertex format: + */ +#define COLOR_IS_RGBA +#define TAG(x) radeon##x +#include "tnl_dd/t_dd_vertex.h" +#undef TAG + +#define RADEON_RB_CLASS 0xdeadbeef + +struct radeon_renderbuffer +{ + struct gl_renderbuffer base; + struct radeon_bo *bo; + unsigned int cpp; + /* unsigned int offset; */ + unsigned int pitch; + + uint32_t draw_offset; /* FBO */ + /* boo Xorg 6.8.2 compat */ + int has_surface; + + GLuint pf_pending; /**< sequence number of pending flip */ + GLuint vbl_pending; /**< vblank sequence number of pending flip */ + __DRIdrawablePrivate *dPriv; +}; + +struct radeon_framebuffer +{ + struct gl_framebuffer base; + + struct radeon_renderbuffer *color_rb[2]; + + GLuint vbl_waited; + + /* buffer swap */ + int64_t swap_ust; + int64_t swap_missed_ust; + + GLuint swap_count; + GLuint swap_missed_count; + + /* Drawable page flipping state */ + GLboolean pf_active; + GLint pf_current_page; + GLint pf_num_pages; + +}; + + +struct radeon_colorbuffer_state { + GLuint clear; + int roundEnable; + struct gl_renderbuffer *rb; + uint32_t draw_offset; /* offset into color renderbuffer - FBOs */ +}; + +struct radeon_depthbuffer_state { + GLuint clear; + struct gl_renderbuffer *rb; +}; + +struct radeon_scissor_state { + drm_clip_rect_t rect; + GLboolean enabled; + + GLuint numClipRects; /* Cliprects active */ + GLuint numAllocedClipRects; /* Cliprects available */ + drm_clip_rect_t *pClipRects; +}; + +struct radeon_stencilbuffer_state { + GLuint clear; /* rb3d_stencilrefmask value */ +}; + +struct radeon_state_atom { + struct radeon_state_atom *next, *prev; + const char *name; /* for debug */ + int cmd_size; /* size in bytes */ + GLuint idx; + GLuint is_tcl; + GLuint *cmd; /* one or more cmd's */ + GLuint *lastcmd; /* one or more cmd's */ + GLboolean dirty; /* dirty-mark in emit_state_list */ + int (*check) (GLcontext *, struct radeon_state_atom *atom); /* is this state active? */ + void (*emit) (GLcontext *, struct radeon_state_atom *atom); +}; + +struct radeon_hw_state { + /* Head of the linked list of state atoms. */ + struct radeon_state_atom atomlist; + int max_state_size; /* Number of bytes necessary for a full state emit. */ + int max_post_flush_size; /* Number of bytes necessary for post flushing emits */ + GLboolean is_dirty, all_dirty; +}; + + +/* Texture related */ +typedef struct _radeon_texture_image radeon_texture_image; + +struct _radeon_texture_image { + struct gl_texture_image base; + + /** + * If mt != 0, the image is stored in hardware format in the + * given mipmap tree. In this case, base.Data may point into the + * mapping of the buffer object that contains the mipmap tree. + * + * If mt == 0, the image is stored in normal memory pointed to + * by base.Data. + */ + struct _radeon_mipmap_tree *mt; + struct radeon_bo *bo; + + int mtlevel; /** if mt != 0, this is the image's level in the mipmap tree */ + int mtface; /** if mt != 0, this is the image's face in the mipmap tree */ +}; + + +static INLINE radeon_texture_image *get_radeon_texture_image(struct gl_texture_image *image) +{ + return (radeon_texture_image*)image; +} + + +typedef struct radeon_tex_obj radeonTexObj, *radeonTexObjPtr; + +#define RADEON_TXO_MICRO_TILE (1 << 3) + +/* Texture object in locally shared texture space. + */ +struct radeon_tex_obj { + struct gl_texture_object base; + struct _radeon_mipmap_tree *mt; + + /** + * This is true if we've verified that the mipmap tree above is complete + * and so on. + */ + GLboolean validated; + + GLuint override_offset; + GLboolean image_override; /* Image overridden by GLX_EXT_tfp */ + GLuint tile_bits; /* hw texture tile bits used on this texture */ + struct radeon_bo *bo; + + GLuint pp_txfilter; /* hardware register values */ + GLuint pp_txformat; + GLuint pp_txformat_x; + GLuint pp_txsize; /* npot only */ + GLuint pp_txpitch; /* npot only */ + GLuint pp_border_color; + GLuint pp_cubic_faces; /* cube face 1,2,3,4 log2 sizes */ + + GLuint pp_txfilter_1; /* r300 */ + + /* r700 texture states */ + GLuint SQ_TEX_RESOURCE0; + GLuint SQ_TEX_RESOURCE1; + GLuint SQ_TEX_RESOURCE2; + GLuint SQ_TEX_RESOURCE3; + GLuint SQ_TEX_RESOURCE4; + GLuint SQ_TEX_RESOURCE5; + GLuint SQ_TEX_RESOURCE6; + + GLuint SQ_TEX_SAMPLER0; + GLuint SQ_TEX_SAMPLER1; + GLuint SQ_TEX_SAMPLER2; + + GLuint TD_PS_SAMPLER0_BORDER_RED; + GLuint TD_PS_SAMPLER0_BORDER_GREEN; + GLuint TD_PS_SAMPLER0_BORDER_BLUE; + GLuint TD_PS_SAMPLER0_BORDER_ALPHA; + + GLboolean border_fallback; + + +}; + +static INLINE radeonTexObj* radeon_tex_obj(struct gl_texture_object *texObj) +{ + return (radeonTexObj*)texObj; +} + +/* occlusion query */ +struct radeon_query_object { + struct gl_query_object Base; + struct radeon_bo *bo; + int curr_offset; + GLboolean emitted_begin; + + /* Double linked list of not flushed query objects */ + struct radeon_query_object *prev, *next; +}; + +/* Need refcounting on dma buffers: + */ +struct radeon_dma_buffer { + int refcount; /* the number of retained regions in buf */ + drmBufPtr buf; +}; + +struct radeon_aos { + struct radeon_bo *bo; /** Buffer object where vertex data is stored */ + int offset; /** Offset into buffer object, in bytes */ + int components; /** Number of components per vertex */ + int stride; /** Stride in dwords (may be 0 for repeating) */ + int count; /** Number of vertices */ +}; + +#define DMA_BO_FREE_TIME 100 + +struct radeon_dma_bo { + struct radeon_dma_bo *next, *prev; + struct radeon_bo *bo; + int expire_counter; +}; + +struct radeon_dma { + /* Active dma region. Allocations for vertices and retained + * regions come from here. Also used for emitting random vertices, + * these may be flushed by calling flush_current(); + */ + struct radeon_dma_bo free; + struct radeon_dma_bo wait; + struct radeon_dma_bo reserved; + size_t current_used; /** Number of bytes allocated and forgotten about */ + size_t current_vertexptr; /** End of active vertex region */ + size_t minimum_size; + + /** + * If current_vertexptr != current_used then flush must be non-zero. + * flush must be called before non-active vertex allocations can be + * performed. + */ + void (*flush) (GLcontext *); +}; + +/* radeon_swtcl.c + */ +struct radeon_swtcl_info { + + GLuint RenderIndex; + GLuint vertex_size; + GLubyte *verts; + + /* Fallback rasterization functions + */ + GLuint hw_primitive; + GLenum render_primitive; + GLuint numverts; + + struct tnl_attr_map vertex_attrs[VERT_ATTRIB_MAX]; + GLuint vertex_attr_count; + + GLuint emit_prediction; +}; + +#define RADEON_MAX_AOS_ARRAYS 16 +struct radeon_tcl_info { + struct radeon_aos aos[RADEON_MAX_AOS_ARRAYS]; + GLuint aos_count; + struct radeon_bo *elt_dma_bo; /** Buffer object that contains element indices */ + int elt_dma_offset; /** Offset into this buffer object, in bytes */ +}; + +struct radeon_ioctl { + GLuint vertex_offset; + GLuint vertex_max; + struct radeon_bo *bo; + GLuint vertex_size; +}; + +#define RADEON_MAX_PRIMS 64 + +struct radeon_prim { + GLuint start; + GLuint end; + GLuint prim; +}; + +static INLINE GLuint radeonPackColor(GLuint cpp, + GLubyte r, GLubyte g, + GLubyte b, GLubyte a) +{ + switch (cpp) { + case 2: + return PACK_COLOR_565(r, g, b); + case 4: + return PACK_COLOR_8888(a, r, g, b); + default: + return 0; + } +} + +#define MAX_CMD_BUF_SZ (16*1024) + +#define MAX_DMA_BUF_SZ (64*1024) + +struct radeon_store { + GLuint statenr; + GLuint primnr; + char cmd_buf[MAX_CMD_BUF_SZ]; + int cmd_used; + int elts_start; +}; + +struct radeon_dri_mirror { + __DRIcontextPrivate *context; /* DRI context */ + __DRIscreenPrivate *screen; /* DRI screen */ + + drm_context_t hwContext; + drm_hw_lock_t *hwLock; + int hwLockCount; + int fd; + int drmMinor; +}; + +typedef void (*radeon_tri_func) (radeonContextPtr, + radeonVertex *, + radeonVertex *, radeonVertex *); + +typedef void (*radeon_line_func) (radeonContextPtr, + radeonVertex *, radeonVertex *); + +typedef void (*radeon_point_func) (radeonContextPtr, radeonVertex *); + +#define RADEON_MAX_BOS 32 +struct radeon_state { + struct radeon_colorbuffer_state color; + struct radeon_depthbuffer_state depth; + struct radeon_scissor_state scissor; + struct radeon_stencilbuffer_state stencil; + + struct radeon_cs_space_check bos[RADEON_MAX_BOS]; + int validated_bo_count; +}; + +/** + * This structure holds the command buffer while it is being constructed. + * + * The first batch of commands in the buffer is always the state that needs + * to be re-emitted when the context is lost. This batch can be skipped + * otherwise. + */ +struct radeon_cmdbuf { + struct radeon_cs_manager *csm; + struct radeon_cs *cs; + int size; /** # of dwords total */ + unsigned int flushing:1; /** whether we're currently in FlushCmdBufLocked */ +}; + +struct radeon_context { + GLcontext *glCtx; + radeonScreenPtr radeonScreen; /* Screen private DRI data */ + + /* Texture object bookkeeping + */ + int texture_depth; + float initialMaxAnisotropy; + uint32_t texture_row_align; + uint32_t texture_rect_row_align; + uint32_t texture_compressed_row_align; + + struct radeon_dma dma; + struct radeon_hw_state hw; + /* Rasterization and vertex state: + */ + GLuint TclFallback; + GLuint Fallback; + GLuint NewGLState; + DECLARE_RENDERINPUTS(tnl_index_bitset); /* index of bits for last tnl_install_attrs */ + + /* Drawable, cliprect and scissor information */ + GLuint numClipRects; /* Cliprects for the draw buffer */ + drm_clip_rect_t *pClipRects; + unsigned int lastStamp; + drm_radeon_sarea_t *sarea; /* Private SAREA data */ + + /* Mirrors of some DRI state */ + struct radeon_dri_mirror dri; + + /* Busy waiting */ + GLuint do_usleeps; + GLuint do_irqs; + GLuint irqsEmitted; + drm_radeon_irq_wait_t iw; + + /* Derived state - for r300 only */ + struct radeon_state state; + + struct radeon_swtcl_info swtcl; + struct radeon_tcl_info tcl; + /* Configuration cache + */ + driOptionCache optionCache; + + struct radeon_cmdbuf cmdbuf; + + struct radeon_debug debug; + + drm_clip_rect_t fboRect; + GLboolean constant_cliprect; /* use for FBO or DRI2 rendering */ + GLboolean front_cliprects; + + /** + * Set if rendering has occured to the drawable's front buffer. + * + * This is used in the DRI2 case to detect that glFlush should also copy + * the contents of the fake front buffer to the real front buffer. + */ + GLboolean front_buffer_dirty; + + /** + * Track whether front-buffer rendering is currently enabled + * + * A separate flag is used to track this in order to support MRT more + * easily. + */ + GLboolean is_front_buffer_rendering; + + /** + * Track whether front-buffer is the current read target. + * + * This is closely associated with is_front_buffer_rendering, but may + * be set separately. The DRI2 fake front buffer must be referenced + * either way. + */ + GLboolean is_front_buffer_reading; + + struct dri_metaops meta; + + struct { + struct radeon_query_object *current; + struct radeon_query_object not_flushed_head; + struct radeon_state_atom queryobj; + } query; + + struct { + void (*get_lock)(radeonContextPtr radeon); + void (*update_viewport_offset)(GLcontext *ctx); + void (*emit_cs_header)(struct radeon_cs *cs, radeonContextPtr rmesa); + void (*swtcl_flush)(GLcontext *ctx, uint32_t offset); + void (*pre_emit_atoms)(radeonContextPtr rmesa); + void (*pre_emit_state)(radeonContextPtr rmesa); + void (*fallback)(GLcontext *ctx, GLuint bit, GLboolean mode); + void (*free_context)(GLcontext *ctx); + void (*emit_query_finish)(radeonContextPtr radeon); + void (*update_scissor)(GLcontext *ctx); + } vtbl; +}; + +#define RADEON_CONTEXT(glctx) ((radeonContextPtr)(ctx->DriverCtx)) + +static inline __DRIdrawablePrivate* radeon_get_drawable(radeonContextPtr radeon) +{ + return radeon->dri.context->driDrawablePriv; +} + +static inline __DRIdrawablePrivate* radeon_get_readable(radeonContextPtr radeon) +{ + return radeon->dri.context->driReadablePriv; +} + +/** + * This function takes a float and packs it into a uint32_t + */ +static INLINE uint32_t radeonPackFloat32(float fl) +{ + union { + float fl; + uint32_t u; + } u; + + u.fl = fl; + return u.u; +} + +/* This is probably wrong for some values, I need to test this + * some more. Range checking would be a good idea also.. + * + * But it works for most things. I'll fix it later if someone + * else with a better clue doesn't + */ +static INLINE uint32_t radeonPackFloat24(float f) +{ + float mantissa; + int exponent; + uint32_t float24 = 0; + + if (f == 0.0) + return 0; + + mantissa = frexpf(f, &exponent); + + /* Handle -ve */ + if (mantissa < 0) { + float24 |= (1 << 23); + mantissa = mantissa * -1.0; + } + /* Handle exponent, bias of 63 */ + exponent += 62; + float24 |= (exponent << 16); + /* Kill 7 LSB of mantissa */ + float24 |= (radeonPackFloat32(mantissa) & 0x7FFFFF) >> 7; + + return float24; +} + +GLboolean radeonInitContext(radeonContextPtr radeon, + struct dd_function_table* functions, + const __GLcontextModes * glVisual, + __DRIcontextPrivate * driContextPriv, + void *sharedContextPrivate); + +void radeonCleanupContext(radeonContextPtr radeon); +GLboolean radeonUnbindContext(__DRIcontextPrivate * driContextPriv); +void radeon_update_renderbuffers(__DRIcontext *context, __DRIdrawable *drawable); +GLboolean radeonMakeCurrent(__DRIcontextPrivate * driContextPriv, + __DRIdrawablePrivate * driDrawPriv, + __DRIdrawablePrivate * driReadPriv); +extern void radeonDestroyContext(__DRIcontextPrivate * driContextPriv); + +#endif diff --git a/src/mesa/drivers/dri/radeon/radeon_compat.c b/src/mesa/drivers/dri/radeon/radeon_compat.c deleted file mode 100644 index 46b490d61f7..00000000000 --- a/src/mesa/drivers/dri/radeon/radeon_compat.c +++ /dev/null @@ -1,301 +0,0 @@ -/************************************************************************** - -Copyright 2002 ATI Technologies Inc., Ontario, Canada, and - Tungsten Graphics Inc., Austin, Texas. - -All Rights Reserved. - -Permission is hereby granted, free of charge, to any person obtaining a -copy of this software and associated documentation files (the "Software"), -to deal in the Software without restriction, including without limitation -on the rights to use, copy, modify, merge, publish, distribute, sub -license, and/or sell copies of the Software, and to permit persons to whom -the Software is furnished to do so, subject to the following conditions: - -The above copyright notice and this permission notice (including the next -paragraph) shall be included in all copies or substantial portions of the -Software. - -THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR -IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, -FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL -ATI, TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, -DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR -OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE -USE OR OTHER DEALINGS IN THE SOFTWARE. - -**************************************************************************/ - -/* - * Authors: - * Keith Whitwell <keith@tungstengraphics.com> - * - */ - -#include "main/glheader.h" -#include "main/imports.h" - -#include "radeon_context.h" -#include "radeon_state.h" -#include "radeon_ioctl.h" - - -static struct { - int start; - int len; - const char *name; -} packet[RADEON_MAX_STATE_PACKETS] = { - { RADEON_PP_MISC,7,"RADEON_PP_MISC" }, - { RADEON_PP_CNTL,3,"RADEON_PP_CNTL" }, - { RADEON_RB3D_COLORPITCH,1,"RADEON_RB3D_COLORPITCH" }, - { RADEON_RE_LINE_PATTERN,2,"RADEON_RE_LINE_PATTERN" }, - { RADEON_SE_LINE_WIDTH,1,"RADEON_SE_LINE_WIDTH" }, - { RADEON_PP_LUM_MATRIX,1,"RADEON_PP_LUM_MATRIX" }, - { RADEON_PP_ROT_MATRIX_0,2,"RADEON_PP_ROT_MATRIX_0" }, - { RADEON_RB3D_STENCILREFMASK,3,"RADEON_RB3D_STENCILREFMASK" }, - { RADEON_SE_VPORT_XSCALE,6,"RADEON_SE_VPORT_XSCALE" }, - { RADEON_SE_CNTL,2,"RADEON_SE_CNTL" }, - { RADEON_SE_CNTL_STATUS,1,"RADEON_SE_CNTL_STATUS" }, - { RADEON_RE_MISC,1,"RADEON_RE_MISC" }, - { RADEON_PP_TXFILTER_0,6,"RADEON_PP_TXFILTER_0" }, - { RADEON_PP_BORDER_COLOR_0,1,"RADEON_PP_BORDER_COLOR_0" }, - { RADEON_PP_TXFILTER_1,6,"RADEON_PP_TXFILTER_1" }, - { RADEON_PP_BORDER_COLOR_1,1,"RADEON_PP_BORDER_COLOR_1" }, - { RADEON_PP_TXFILTER_2,6,"RADEON_PP_TXFILTER_2" }, - { RADEON_PP_BORDER_COLOR_2,1,"RADEON_PP_BORDER_COLOR_2" }, - { RADEON_SE_ZBIAS_FACTOR,2,"RADEON_SE_ZBIAS_FACTOR" }, - { RADEON_SE_TCL_OUTPUT_VTX_FMT,11,"RADEON_SE_TCL_OUTPUT_VTX_FMT" }, - { RADEON_SE_TCL_MATERIAL_EMMISSIVE_RED,17,"RADEON_SE_TCL_MATERIAL_EMMISSIVE_RED" }, -}; - - -static void radeonCompatEmitPacket( radeonContextPtr rmesa, - struct radeon_state_atom *state ) -{ - drm_radeon_sarea_t *sarea = rmesa->sarea; - drm_radeon_context_regs_t *ctx = &sarea->context_state; - drm_radeon_texture_regs_t *tex0 = &sarea->tex_state[0]; - drm_radeon_texture_regs_t *tex1 = &sarea->tex_state[1]; - int i; - int *buf = state->cmd; - - for ( i = 0 ; i < state->cmd_size ; ) { - drm_radeon_cmd_header_t *header = (drm_radeon_cmd_header_t *)&buf[i++]; - - if (RADEON_DEBUG & DEBUG_STATE) - fprintf(stderr, "%s %d: %s\n", __FUNCTION__, header->packet.packet_id, - packet[(int)header->packet.packet_id].name); - - switch (header->packet.packet_id) { - case RADEON_EMIT_PP_MISC: - ctx->pp_misc = buf[i++]; - ctx->pp_fog_color = buf[i++]; - ctx->re_solid_color = buf[i++]; - ctx->rb3d_blendcntl = buf[i++]; - ctx->rb3d_depthoffset = buf[i++]; - ctx->rb3d_depthpitch = buf[i++]; - ctx->rb3d_zstencilcntl = buf[i++]; - sarea->dirty |= RADEON_UPLOAD_CONTEXT; - break; - case RADEON_EMIT_PP_CNTL: - ctx->pp_cntl = buf[i++]; - ctx->rb3d_cntl = buf[i++]; - ctx->rb3d_coloroffset = buf[i++]; - sarea->dirty |= RADEON_UPLOAD_CONTEXT; - break; - case RADEON_EMIT_RB3D_COLORPITCH: - ctx->rb3d_colorpitch = buf[i++]; - sarea->dirty |= RADEON_UPLOAD_CONTEXT; - break; - case RADEON_EMIT_RE_LINE_PATTERN: - ctx->re_line_pattern = buf[i++]; - ctx->re_line_state = buf[i++]; - sarea->dirty |= RADEON_UPLOAD_LINE; - break; - case RADEON_EMIT_SE_LINE_WIDTH: - ctx->se_line_width = buf[i++]; - sarea->dirty |= RADEON_UPLOAD_LINE; - break; - case RADEON_EMIT_PP_LUM_MATRIX: - ctx->pp_lum_matrix = buf[i++]; - sarea->dirty |= RADEON_UPLOAD_BUMPMAP; - break; - case RADEON_EMIT_PP_ROT_MATRIX_0: - ctx->pp_rot_matrix_0 = buf[i++]; - ctx->pp_rot_matrix_1 = buf[i++]; - sarea->dirty |= RADEON_UPLOAD_BUMPMAP; - break; - case RADEON_EMIT_RB3D_STENCILREFMASK: - ctx->rb3d_stencilrefmask = buf[i++]; - ctx->rb3d_ropcntl = buf[i++]; - ctx->rb3d_planemask = buf[i++]; - sarea->dirty |= RADEON_UPLOAD_MASKS; - break; - case RADEON_EMIT_SE_VPORT_XSCALE: - ctx->se_vport_xscale = buf[i++]; - ctx->se_vport_xoffset = buf[i++]; - ctx->se_vport_yscale = buf[i++]; - ctx->se_vport_yoffset = buf[i++]; - ctx->se_vport_zscale = buf[i++]; - ctx->se_vport_zoffset = buf[i++]; - sarea->dirty |= RADEON_UPLOAD_VIEWPORT; - break; - case RADEON_EMIT_SE_CNTL: - ctx->se_cntl = buf[i++]; - ctx->se_coord_fmt = buf[i++]; - sarea->dirty |= RADEON_UPLOAD_CONTEXT | RADEON_UPLOAD_VERTFMT; - break; - case RADEON_EMIT_SE_CNTL_STATUS: - ctx->se_cntl_status = buf[i++]; - sarea->dirty |= RADEON_UPLOAD_SETUP; - break; - case RADEON_EMIT_RE_MISC: - ctx->re_misc = buf[i++]; - sarea->dirty |= RADEON_UPLOAD_MISC; - break; - case RADEON_EMIT_PP_TXFILTER_0: - tex0->pp_txfilter = buf[i++]; - tex0->pp_txformat = buf[i++]; - tex0->pp_txoffset = buf[i++]; - tex0->pp_txcblend = buf[i++]; - tex0->pp_txablend = buf[i++]; - tex0->pp_tfactor = buf[i++]; - sarea->dirty |= RADEON_UPLOAD_TEX0; - break; - case RADEON_EMIT_PP_BORDER_COLOR_0: - tex0->pp_border_color = buf[i++]; - sarea->dirty |= RADEON_UPLOAD_TEX0; - break; - case RADEON_EMIT_PP_TXFILTER_1: - tex1->pp_txfilter = buf[i++]; - tex1->pp_txformat = buf[i++]; - tex1->pp_txoffset = buf[i++]; - tex1->pp_txcblend = buf[i++]; - tex1->pp_txablend = buf[i++]; - tex1->pp_tfactor = buf[i++]; - sarea->dirty |= RADEON_UPLOAD_TEX1; - break; - case RADEON_EMIT_PP_BORDER_COLOR_1: - tex1->pp_border_color = buf[i++]; - sarea->dirty |= RADEON_UPLOAD_TEX1; - break; - - case RADEON_EMIT_SE_ZBIAS_FACTOR: - i++; - i++; - break; - - case RADEON_EMIT_PP_TXFILTER_2: - case RADEON_EMIT_PP_BORDER_COLOR_2: - case RADEON_EMIT_SE_TCL_OUTPUT_VTX_FMT: - case RADEON_EMIT_SE_TCL_MATERIAL_EMMISSIVE_RED: - default: - /* These states aren't understood by radeon drm 1.1 */ - fprintf(stderr, "Tried to emit unsupported state\n"); - return; - } - } -} - - - -static void radeonCompatEmitStateLocked( radeonContextPtr rmesa ) -{ - struct radeon_state_atom *atom; - - if (RADEON_DEBUG & (DEBUG_STATE|DEBUG_PRIMS)) - fprintf(stderr, "%s\n", __FUNCTION__); - - if (!rmesa->hw.is_dirty && !rmesa->hw.all_dirty) - return; - - foreach(atom, &rmesa->hw.atomlist) { - if (rmesa->hw.all_dirty) - atom->dirty = GL_TRUE; - if (atom->is_tcl) - atom->dirty = GL_FALSE; - if (atom->dirty) - radeonCompatEmitPacket(rmesa, atom); - } - - rmesa->hw.is_dirty = GL_FALSE; - rmesa->hw.all_dirty = GL_FALSE; -} - - -static void radeonCompatEmitPrimitiveLocked( radeonContextPtr rmesa, - GLuint hw_primitive, - GLuint nverts, - drm_clip_rect_t *pbox, - GLuint nbox ) -{ - int i; - - for ( i = 0 ; i < nbox ; ) { - int nr = MIN2( i + RADEON_NR_SAREA_CLIPRECTS, nbox ); - drm_clip_rect_t *b = rmesa->sarea->boxes; - drm_radeon_vertex_t vtx; - - rmesa->sarea->dirty |= RADEON_UPLOAD_CLIPRECTS; - rmesa->sarea->nbox = nr - i; - - for ( ; i < nr ; i++) - *b++ = pbox[i]; - - if (RADEON_DEBUG & DEBUG_IOCTL) - fprintf(stderr, - "RadeonFlushVertexBuffer: prim %x buf %d verts %d " - "disc %d nbox %d\n", - hw_primitive, - rmesa->dma.current.buf->buf->idx, - nverts, - nr == nbox, - rmesa->sarea->nbox ); - - vtx.prim = hw_primitive; - vtx.idx = rmesa->dma.current.buf->buf->idx; - vtx.count = nverts; - vtx.discard = (nr == nbox); - - drmCommandWrite( rmesa->dri.fd, - DRM_RADEON_VERTEX, - &vtx, sizeof(vtx)); - } -} - - - -/* No 'start' for 1.1 vertices ioctl: only one vertex prim/buffer! - */ -void radeonCompatEmitPrimitive( radeonContextPtr rmesa, - GLuint vertex_format, - GLuint hw_primitive, - GLuint nrverts ) -{ - if (RADEON_DEBUG & DEBUG_IOCTL) - fprintf(stderr, "%s\n", __FUNCTION__); - - LOCK_HARDWARE( rmesa ); - - radeonCompatEmitStateLocked( rmesa ); - rmesa->sarea->vc_format = vertex_format; - - if (rmesa->state.scissor.enabled) { - radeonCompatEmitPrimitiveLocked( rmesa, - hw_primitive, - nrverts, - rmesa->state.scissor.pClipRects, - rmesa->state.scissor.numClipRects ); - } - else { - radeonCompatEmitPrimitiveLocked( rmesa, - hw_primitive, - nrverts, - rmesa->pClipRects, - rmesa->numClipRects ); - } - - - UNLOCK_HARDWARE( rmesa ); -} - diff --git a/src/mesa/drivers/dri/radeon/radeon_context.c b/src/mesa/drivers/dri/radeon/radeon_context.c index b67bda7d06a..8f4485aee7c 100644 --- a/src/mesa/drivers/dri/radeon/radeon_context.c +++ b/src/mesa/drivers/dri/radeon/radeon_context.c @@ -53,6 +53,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "drivers/common/driverfuncs.h" +#include "radeon_common.h" #include "radeon_context.h" #include "radeon_ioctl.h" #include "radeon_state.h" @@ -61,10 +62,13 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "radeon_swtcl.h" #include "radeon_tcl.h" #include "radeon_maos.h" +#include "radeon_queryobj.h" +#define need_GL_ARB_occlusion_query #define need_GL_EXT_blend_minmax #define need_GL_EXT_fog_coord #define need_GL_EXT_secondary_color +#define need_GL_EXT_framebuffer_object #include "extension_helper.h" #define DRIVER_DATE "20061018" @@ -72,46 +76,13 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "vblank.h" #include "utils.h" #include "xmlpool.h" /* for symbolic values of enum-type options */ -#ifndef RADEON_DEBUG -int RADEON_DEBUG = (0); -#endif - - -/* Return various strings for glGetString(). - */ -static const GLubyte *radeonGetString( GLcontext *ctx, GLenum name ) -{ - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - static char buffer[128]; - unsigned offset; - GLuint agp_mode = (rmesa->radeonScreen->card_type==RADEON_CARD_PCI) ? 0 : - rmesa->radeonScreen->AGPMode; - - switch ( name ) { - case GL_VENDOR: - return (GLubyte *)"Tungsten Graphics, Inc."; - - case GL_RENDERER: - offset = driGetRendererString( buffer, "Radeon", DRIVER_DATE, - agp_mode ); - - sprintf( & buffer[ offset ], " %sTCL", - !(rmesa->TclFallback & RADEON_TCL_FALLBACK_TCL_DISABLE) - ? "" : "NO-" ); - - return (GLubyte *)buffer; - - default: - return NULL; - } -} - /* Extension strings exported by the R100 driver. */ const struct dri_extension card_extensions[] = { { "GL_ARB_multitexture", NULL }, + { "GL_ARB_occlusion_query", GL_ARB_occlusion_query_functions}, { "GL_ARB_texture_border_clamp", NULL }, { "GL_ARB_texture_env_add", NULL }, { "GL_ARB_texture_env_combine", NULL }, @@ -121,6 +92,7 @@ const struct dri_extension card_extensions[] = { "GL_EXT_blend_logic_op", NULL }, { "GL_EXT_blend_subtract", GL_EXT_blend_minmax_functions }, { "GL_EXT_fog_coord", GL_EXT_fog_coord_functions }, + { "GL_EXT_packed_depth_stencil", NULL}, { "GL_EXT_secondary_color", GL_EXT_secondary_color_functions }, { "GL_EXT_stencil_wrap", NULL }, { "GL_EXT_texture_edge_clamp", NULL }, @@ -137,6 +109,11 @@ const struct dri_extension card_extensions[] = { NULL, NULL } }; +const struct dri_extension mm_extensions[] = { + { "GL_EXT_framebuffer_object", GL_EXT_framebuffer_object_functions }, + { NULL, NULL } +}; + extern const struct tnl_pipeline_stage _radeon_render_stage; extern const struct tnl_pipeline_stage _radeon_tcl_stage; @@ -160,47 +137,85 @@ static const struct tnl_pipeline_stage *radeon_pipeline[] = { NULL, }; +static void r100_get_lock(radeonContextPtr radeon) +{ + r100ContextPtr rmesa = (r100ContextPtr)radeon; + drm_radeon_sarea_t *sarea = radeon->sarea; + RADEON_STATECHANGE(rmesa, ctx); + if (rmesa->radeon.sarea->tiling_enabled) { + rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] |= + RADEON_COLOR_TILE_ENABLE; + } else { + rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] &= + ~RADEON_COLOR_TILE_ENABLE; + } + + if (sarea->ctx_owner != rmesa->radeon.dri.hwContext) { + sarea->ctx_owner = rmesa->radeon.dri.hwContext; + + if (!radeon->radeonScreen->kernel_mm) + radeon_bo_legacy_texture_age(radeon->radeonScreen->bom); + } +} -/* Initialize the driver's misc functions. - */ -static void radeonInitDriverFuncs( struct dd_function_table *functions ) +static void r100_vtbl_emit_cs_header(struct radeon_cs *cs, radeonContextPtr rmesa) { - functions->GetString = radeonGetString; } -static const struct dri_debug_control debug_control[] = +static void r100_vtbl_pre_emit_state(radeonContextPtr radeon) { - { "fall", DEBUG_FALLBACKS }, - { "tex", DEBUG_TEXTURE }, - { "ioctl", DEBUG_IOCTL }, - { "prim", DEBUG_PRIMS }, - { "vert", DEBUG_VERTS }, - { "state", DEBUG_STATE }, - { "code", DEBUG_CODEGEN }, - { "vfmt", DEBUG_VFMT }, - { "vtxf", DEBUG_VFMT }, - { "verb", DEBUG_VERBOSE }, - { "dri", DEBUG_DRI }, - { "dma", DEBUG_DMA }, - { "san", DEBUG_SANITY }, - { "sync", DEBUG_SYNC }, - { NULL, 0 } -}; + r100ContextPtr rmesa = (r100ContextPtr)radeon; + + /* r100 always needs to emit ZBS to avoid TCL lockups */ + rmesa->hw.zbs.dirty = 1; + radeon->hw.is_dirty = 1; +} + +static void r100_vtbl_free_context(GLcontext *ctx) +{ + r100ContextPtr rmesa = R100_CONTEXT(ctx); + _mesa_vector4f_free( &rmesa->tcl.ObjClean ); +} +static void r100_emit_query_finish(radeonContextPtr radeon) +{ + BATCH_LOCALS(radeon); + struct radeon_query_object *query = radeon->query.current; + + BEGIN_BATCH_NO_AUTOSTATE(4); + OUT_BATCH(CP_PACKET0(RADEON_RB3D_ZPASS_ADDR, 0)); + OUT_BATCH_RELOC(0, query->bo, query->curr_offset, 0, RADEON_GEM_DOMAIN_GTT, 0); + END_BATCH(); + query->curr_offset += sizeof(uint32_t); + assert(query->curr_offset < RADEON_QUERY_PAGE_SIZE); + query->emitted_begin = GL_FALSE; +} + +static void r100_init_vtbl(radeonContextPtr radeon) +{ + radeon->vtbl.get_lock = r100_get_lock; + radeon->vtbl.update_viewport_offset = radeonUpdateViewportOffset; + radeon->vtbl.emit_cs_header = r100_vtbl_emit_cs_header; + radeon->vtbl.swtcl_flush = r100_swtcl_flush; + radeon->vtbl.pre_emit_state = r100_vtbl_pre_emit_state; + radeon->vtbl.fallback = radeonFallback; + radeon->vtbl.free_context = r100_vtbl_free_context; + radeon->vtbl.emit_query_finish = r100_emit_query_finish; +} /* Create the device specific context. */ GLboolean -radeonCreateContext( const __GLcontextModes *glVisual, +r100CreateContext( const __GLcontextModes *glVisual, __DRIcontextPrivate *driContextPriv, void *sharedContextPrivate) { __DRIscreenPrivate *sPriv = driContextPriv->driScreenPriv; radeonScreenPtr screen = (radeonScreenPtr)(sPriv->private); struct dd_function_table functions; - radeonContextPtr rmesa; - GLcontext *ctx, *shareCtx; + r100ContextPtr rmesa; + GLcontext *ctx; int i; int tcl_mode, fthrottle_mode; @@ -209,10 +224,12 @@ radeonCreateContext( const __GLcontextModes *glVisual, assert(screen); /* Allocate the Radeon context */ - rmesa = (radeonContextPtr) CALLOC( sizeof(*rmesa) ); + rmesa = (r100ContextPtr) CALLOC( sizeof(*rmesa) ); if ( !rmesa ) return GL_FALSE; + r100_init_vtbl(&rmesa->radeon); + /* init exp fog table data */ radeonInitStaticFogData(); @@ -220,12 +237,12 @@ radeonCreateContext( const __GLcontextModes *glVisual, * Do this here so that initialMaxAnisotropy is set before we create * the default textures. */ - driParseConfigFiles (&rmesa->optionCache, &screen->optionCache, + driParseConfigFiles (&rmesa->radeon.optionCache, &screen->optionCache, screen->driScreen->myNum, "radeon"); - rmesa->initialMaxAnisotropy = driQueryOptionf(&rmesa->optionCache, + rmesa->radeon.initialMaxAnisotropy = driQueryOptionf(&rmesa->radeon.optionCache, "def_max_anisotropy"); - if ( driQueryOptionb( &rmesa->optionCache, "hyperz" ) ) { + if ( driQueryOptionb( &rmesa->radeon.optionCache, "hyperz" ) ) { if ( sPriv->drm_version.minor < 13 ) fprintf( stderr, "DRM version 1.%d too old to support HyperZ, " "disabling.\n", sPriv->drm_version.minor ); @@ -240,65 +257,18 @@ radeonCreateContext( const __GLcontextModes *glVisual, * (the texture functions are especially important) */ _mesa_init_driver_functions( &functions ); - radeonInitDriverFuncs( &functions ); radeonInitTextureFuncs( &functions ); + radeonInitQueryObjFunctions(&functions); - /* Allocate the Mesa context */ - if (sharedContextPrivate) - shareCtx = ((radeonContextPtr) sharedContextPrivate)->glCtx; - else - shareCtx = NULL; - rmesa->glCtx = _mesa_create_context(glVisual, shareCtx, - &functions, (void *) rmesa); - if (!rmesa->glCtx) { - FREE(rmesa); - return GL_FALSE; - } - driContextPriv->driverPrivate = rmesa; - - /* Init radeon context data */ - rmesa->dri.context = driContextPriv; - rmesa->dri.screen = sPriv; - rmesa->dri.drawable = NULL; - rmesa->dri.readable = NULL; - rmesa->dri.hwContext = driContextPriv->hHWContext; - rmesa->dri.hwLock = &sPriv->pSAREA->lock; - rmesa->dri.fd = sPriv->fd; - rmesa->dri.drmMinor = sPriv->drm_version.minor; - - rmesa->radeonScreen = screen; - rmesa->sarea = (drm_radeon_sarea_t *)((GLubyte *)sPriv->pSAREA + - screen->sarea_priv_offset); - - - rmesa->dma.buf0_address = rmesa->radeonScreen->buffers->list[0].address; - - (void) memset( rmesa->texture_heaps, 0, sizeof( rmesa->texture_heaps ) ); - make_empty_list( & rmesa->swapped ); - - rmesa->nr_heaps = screen->numTexHeaps; - for ( i = 0 ; i < rmesa->nr_heaps ; i++ ) { - rmesa->texture_heaps[i] = driCreateTextureHeap( i, rmesa, - screen->texSize[i], - 12, - RADEON_NR_TEX_REGIONS, - (drmTextureRegionPtr)rmesa->sarea->tex_list[i], - & rmesa->sarea->tex_age[i], - & rmesa->swapped, - sizeof( radeonTexObj ), - (destroy_texture_object_t *) radeonDestroyTexObj ); - - driSetTextureSwapCounterLocation( rmesa->texture_heaps[i], - & rmesa->c_textureSwaps ); + if (!radeonInitContext(&rmesa->radeon, &functions, + glVisual, driContextPriv, + sharedContextPrivate)) { + FREE(rmesa); + return GL_FALSE; } - rmesa->texture_depth = driQueryOptioni (&rmesa->optionCache, - "texture_depth"); - if (rmesa->texture_depth == DRI_CONF_TEXTURE_DEPTH_FB) - rmesa->texture_depth = ( screen->cpp == 4 ) ? - DRI_CONF_TEXTURE_DEPTH_32 : DRI_CONF_TEXTURE_DEPTH_16; - rmesa->swtcl.RenderIndex = ~0; - rmesa->hw.all_dirty = GL_TRUE; + rmesa->radeon.swtcl.RenderIndex = ~0; + rmesa->radeon.hw.all_dirty = GL_TRUE; /* Set the maximum texture size small enough that we can guarentee that * all texture units can bind a maximal texture and have all of them in @@ -306,26 +276,20 @@ radeonCreateContext( const __GLcontextModes *glVisual, * setting allow larger textures. */ - ctx = rmesa->glCtx; - ctx->Const.MaxTextureUnits = driQueryOptioni (&rmesa->optionCache, + ctx = rmesa->radeon.glCtx; + ctx->Const.MaxTextureUnits = driQueryOptioni (&rmesa->radeon.optionCache, "texture_units"); ctx->Const.MaxTextureImageUnits = ctx->Const.MaxTextureUnits; ctx->Const.MaxTextureCoordUnits = ctx->Const.MaxTextureUnits; - i = driQueryOptioni( &rmesa->optionCache, "allow_large_textures"); - - driCalculateMaxTextureLevels( rmesa->texture_heaps, - rmesa->nr_heaps, - & ctx->Const, - 4, - 11, /* max 2D texture size is 2048x2048 */ - 8, /* 256^3 */ - 9, /* \todo: max cube texture size seems to be 512x512(x6) */ - 11, /* max rect texture size is 2048x2048. */ - 12, - GL_FALSE, - i ); + i = driQueryOptioni( &rmesa->radeon.optionCache, "allow_large_textures"); + /* FIXME: When no memory manager is available we should set this + * to some reasonable value based on texture memory pool size */ + ctx->Const.MaxTextureLevels = 12; + ctx->Const.Max3DTextureLevels = 9; + ctx->Const.MaxCubeTextureLevels = 12; + ctx->Const.MaxTextureRectSize = 2048; ctx->Const.MaxTextureMaxAnisotropy = 16.0; @@ -390,38 +354,42 @@ radeonCreateContext( const __GLcontextModes *glVisual, } driInitExtensions( ctx, card_extensions, GL_TRUE ); - if (rmesa->radeonScreen->drmSupportsCubeMapsR100) + if (rmesa->radeon.radeonScreen->kernel_mm) + driInitExtensions(ctx, mm_extensions, GL_FALSE); + if (rmesa->radeon.radeonScreen->drmSupportsCubeMapsR100) _mesa_enable_extension( ctx, "GL_ARB_texture_cube_map" ); - if (rmesa->glCtx->Mesa_DXTn) { + if (rmesa->radeon.glCtx->Mesa_DXTn) { _mesa_enable_extension( ctx, "GL_EXT_texture_compression_s3tc" ); _mesa_enable_extension( ctx, "GL_S3_s3tc" ); } - else if (driQueryOptionb (&rmesa->optionCache, "force_s3tc_enable")) { + else if (driQueryOptionb (&rmesa->radeon.optionCache, "force_s3tc_enable")) { _mesa_enable_extension( ctx, "GL_EXT_texture_compression_s3tc" ); } - if (rmesa->dri.drmMinor >= 9) + if (rmesa->radeon.radeonScreen->kernel_mm || rmesa->radeon.dri.drmMinor >= 9) _mesa_enable_extension( ctx, "GL_NV_texture_rectangle"); + if (!rmesa->radeon.radeonScreen->kernel_mm) + _mesa_disable_extension(ctx, "GL_ARB_occlusion_query"); + /* XXX these should really go right after _mesa_init_driver_functions() */ - radeonInitIoctlFuncs( ctx ); - radeonInitStateFuncs( ctx ); + radeon_fbo_init(&rmesa->radeon); radeonInitSpanFuncs( ctx ); + radeonInitIoctlFuncs( ctx ); + radeonInitStateFuncs( ctx , rmesa->radeon.radeonScreen->kernel_mm ); radeonInitState( rmesa ); radeonInitSwtcl( ctx ); _mesa_vector4f_alloc( &rmesa->tcl.ObjClean, 0, ctx->Const.MaxArrayLockSize, 32 ); - fthrottle_mode = driQueryOptioni(&rmesa->optionCache, "fthrottle_mode"); - rmesa->iw.irq_seq = -1; - rmesa->irqsEmitted = 0; - rmesa->do_irqs = (rmesa->radeonScreen->irq != 0 && - fthrottle_mode == DRI_CONF_FTHROTTLE_IRQS); - - rmesa->do_usleeps = (fthrottle_mode == DRI_CONF_FTHROTTLE_USLEEPS); + fthrottle_mode = driQueryOptioni(&rmesa->radeon.optionCache, "fthrottle_mode"); + rmesa->radeon.iw.irq_seq = -1; + rmesa->radeon.irqsEmitted = 0; + rmesa->radeon.do_irqs = (rmesa->radeon.radeonScreen->irq != 0 && + fthrottle_mode == DRI_CONF_FTHROTTLE_IRQS); - (*sPriv->systemTime->getUST)( & rmesa->swap_ust ); + rmesa->radeon.do_usleeps = (fthrottle_mode == DRI_CONF_FTHROTTLE_USLEEPS); #if DO_DEBUG @@ -429,206 +397,21 @@ radeonCreateContext( const __GLcontextModes *glVisual, debug_control ); #endif - tcl_mode = driQueryOptioni(&rmesa->optionCache, "tcl_mode"); - if (driQueryOptionb(&rmesa->optionCache, "no_rast")) { + tcl_mode = driQueryOptioni(&rmesa->radeon.optionCache, "tcl_mode"); + if (driQueryOptionb(&rmesa->radeon.optionCache, "no_rast")) { fprintf(stderr, "disabling 3D acceleration\n"); FALLBACK(rmesa, RADEON_FALLBACK_DISABLE, 1); } else if (tcl_mode == DRI_CONF_TCL_SW || - !(rmesa->radeonScreen->chip_flags & RADEON_CHIPSET_TCL)) { - if (rmesa->radeonScreen->chip_flags & RADEON_CHIPSET_TCL) { - rmesa->radeonScreen->chip_flags &= ~RADEON_CHIPSET_TCL; + !(rmesa->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL)) { + if (rmesa->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL) { + rmesa->radeon.radeonScreen->chip_flags &= ~RADEON_CHIPSET_TCL; fprintf(stderr, "Disabling HW TCL support\n"); } - TCL_FALLBACK(rmesa->glCtx, RADEON_TCL_FALLBACK_TCL_DISABLE, 1); + TCL_FALLBACK(rmesa->radeon.glCtx, RADEON_TCL_FALLBACK_TCL_DISABLE, 1); } - if (rmesa->radeonScreen->chip_flags & RADEON_CHIPSET_TCL) { + if (rmesa->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL) { /* _tnl_need_dlist_norm_lengths( ctx, GL_FALSE ); */ } return GL_TRUE; } - - -/* Destroy the device specific context. - */ -/* Destroy the Mesa and driver specific context data. - */ -void radeonDestroyContext( __DRIcontextPrivate *driContextPriv ) -{ - GET_CURRENT_CONTEXT(ctx); - radeonContextPtr rmesa = (radeonContextPtr) driContextPriv->driverPrivate; - radeonContextPtr current = ctx ? RADEON_CONTEXT(ctx) : NULL; - - /* check if we're deleting the currently bound context */ - if (rmesa == current) { - RADEON_FIREVERTICES( rmesa ); - _mesa_make_current(NULL, NULL, NULL); - } - - /* Free radeon context resources */ - assert(rmesa); /* should never be null */ - if ( rmesa ) { - GLboolean release_texture_heaps; - - - release_texture_heaps = (rmesa->glCtx->Shared->RefCount == 1); - _swsetup_DestroyContext( rmesa->glCtx ); - _tnl_DestroyContext( rmesa->glCtx ); - _vbo_DestroyContext( rmesa->glCtx ); - _swrast_DestroyContext( rmesa->glCtx ); - - radeonDestroySwtcl( rmesa->glCtx ); - radeonReleaseArrays( rmesa->glCtx, ~0 ); - if (rmesa->dma.current.buf) { - radeonReleaseDmaRegion( rmesa, &rmesa->dma.current, __FUNCTION__ ); - radeonFlushCmdBuf( rmesa, __FUNCTION__ ); - } - - _mesa_vector4f_free( &rmesa->tcl.ObjClean ); - - if (rmesa->state.scissor.pClipRects) { - FREE(rmesa->state.scissor.pClipRects); - rmesa->state.scissor.pClipRects = NULL; - } - - if ( release_texture_heaps ) { - /* This share group is about to go away, free our private - * texture object data. - */ - int i; - - for ( i = 0 ; i < rmesa->nr_heaps ; i++ ) { - driDestroyTextureHeap( rmesa->texture_heaps[ i ] ); - rmesa->texture_heaps[ i ] = NULL; - } - - assert( is_empty_list( & rmesa->swapped ) ); - } - - /* free the Mesa context */ - rmesa->glCtx->DriverCtx = NULL; - _mesa_destroy_context( rmesa->glCtx ); - - /* free the option cache */ - driDestroyOptionCache (&rmesa->optionCache); - - FREE( rmesa ); - } -} - - - - -void -radeonSwapBuffers( __DRIdrawablePrivate *dPriv ) -{ - - if (dPriv->driContextPriv && dPriv->driContextPriv->driverPrivate) { - radeonContextPtr rmesa; - GLcontext *ctx; - rmesa = (radeonContextPtr) dPriv->driContextPriv->driverPrivate; - ctx = rmesa->glCtx; - if (ctx->Visual.doubleBufferMode) { - _mesa_notifySwapBuffers( ctx ); /* flush pending rendering comands */ - - if ( rmesa->doPageFlip ) { - radeonPageFlip( dPriv ); - } - else { - radeonCopyBuffer( dPriv, NULL ); - } - } - } - else { - /* XXX this shouldn't be an error but we can't handle it for now */ - _mesa_problem(NULL, "%s: drawable has no context!", __FUNCTION__); - } -} - -void radeonCopySubBuffer(__DRIdrawablePrivate * dPriv, - int x, int y, int w, int h ) -{ - if (dPriv->driContextPriv && dPriv->driContextPriv->driverPrivate) { - radeonContextPtr radeon; - GLcontext *ctx; - - radeon = (radeonContextPtr) dPriv->driContextPriv->driverPrivate; - ctx = radeon->glCtx; - - if (ctx->Visual.doubleBufferMode) { - drm_clip_rect_t rect; - rect.x1 = x + dPriv->x; - rect.y1 = (dPriv->h - y - h) + dPriv->y; - rect.x2 = rect.x1 + w; - rect.y2 = rect.y1 + h; - _mesa_notifySwapBuffers(ctx); /* flush pending rendering comands */ - radeonCopyBuffer(dPriv, &rect); - } - } else { - /* XXX this shouldn't be an error but we can't handle it for now */ - _mesa_problem(NULL, "%s: drawable has no context!", - __FUNCTION__); - } -} - -/* Make context `c' the current context and bind it to the given - * drawing and reading surfaces. - */ -GLboolean -radeonMakeCurrent( __DRIcontextPrivate *driContextPriv, - __DRIdrawablePrivate *driDrawPriv, - __DRIdrawablePrivate *driReadPriv ) -{ - if ( driContextPriv ) { - radeonContextPtr newCtx = - (radeonContextPtr) driContextPriv->driverPrivate; - - if (RADEON_DEBUG & DEBUG_DRI) - fprintf(stderr, "%s ctx %p\n", __FUNCTION__, (void *) newCtx->glCtx); - - newCtx->dri.readable = driReadPriv; - - if ( (newCtx->dri.drawable != driDrawPriv) || - newCtx->lastStamp != driDrawPriv->lastStamp ) { - if (driDrawPriv->swap_interval == (unsigned)-1) { - driDrawPriv->vblFlags = (newCtx->radeonScreen->irq != 0) - ? driGetDefaultVBlankFlags(&newCtx->optionCache) - : VBLANK_FLAG_NO_IRQ; - - driDrawableInitVBlank( driDrawPriv ); - } - - newCtx->dri.drawable = driDrawPriv; - - radeonSetCliprects(newCtx); - radeonUpdateViewportOffset( newCtx->glCtx ); - } - - _mesa_make_current( newCtx->glCtx, - (GLframebuffer *) driDrawPriv->driverPrivate, - (GLframebuffer *) driReadPriv->driverPrivate ); - - _mesa_update_state( newCtx->glCtx ); - } else { - if (RADEON_DEBUG & DEBUG_DRI) - fprintf(stderr, "%s ctx is null\n", __FUNCTION__); - _mesa_make_current( NULL, NULL, NULL ); - } - - if (RADEON_DEBUG & DEBUG_DRI) - fprintf(stderr, "End %s\n", __FUNCTION__); - return GL_TRUE; -} - -/* Force the context `c' to be unbound from its buffer. - */ -GLboolean -radeonUnbindContext( __DRIcontextPrivate *driContextPriv ) -{ - radeonContextPtr rmesa = (radeonContextPtr) driContextPriv->driverPrivate; - - if (RADEON_DEBUG & DEBUG_DRI) - fprintf(stderr, "%s ctx %p\n", __FUNCTION__, (void *) rmesa->glCtx); - - return GL_TRUE; -} diff --git a/src/mesa/drivers/dri/radeon/radeon_context.h b/src/mesa/drivers/dri/radeon/radeon_context.h index 53df766f8cb..4e2c52c835c 100644 --- a/src/mesa/drivers/dri/radeon/radeon_context.h +++ b/src/mesa/drivers/dri/radeon/radeon_context.h @@ -48,91 +48,23 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "drm.h" #include "radeon_drm.h" #include "texmem.h" - #include "main/macros.h" #include "main/mtypes.h" #include "main/colormac.h" - -struct radeon_context; -typedef struct radeon_context radeonContextRec; -typedef struct radeon_context *radeonContextPtr; - -/* This union is used to avoid warnings/miscompilation - with float to uint32_t casts due to strict-aliasing */ -typedef union { - GLfloat f; - uint32_t ui32; -} float_ui32_type; - -#include "radeon_lock.h" #include "radeon_screen.h" -#include "main/mm.h" - -#include "math/m_vector.h" - -#define TEX_0 0x1 -#define TEX_1 0x2 -#define TEX_2 0x4 -#define TEX_ALL 0x7 - -/* Rasterizing fallbacks */ -/* See correponding strings in r200_swtcl.c */ -#define RADEON_FALLBACK_TEXTURE 0x0001 -#define RADEON_FALLBACK_DRAW_BUFFER 0x0002 -#define RADEON_FALLBACK_STENCIL 0x0004 -#define RADEON_FALLBACK_RENDER_MODE 0x0008 -#define RADEON_FALLBACK_BLEND_EQ 0x0010 -#define RADEON_FALLBACK_BLEND_FUNC 0x0020 -#define RADEON_FALLBACK_DISABLE 0x0040 -#define RADEON_FALLBACK_BORDER_MODE 0x0080 - -/* The blit width for texture uploads - */ -#define BLIT_WIDTH_BYTES 1024 -/* Use the templated vertex format: - */ -#define COLOR_IS_RGBA -#define TAG(x) radeon##x -#include "tnl_dd/t_dd_vertex.h" -#undef TAG - -typedef void (*radeon_tri_func) (radeonContextPtr, - radeonVertex *, - radeonVertex *, radeonVertex *); - -typedef void (*radeon_line_func) (radeonContextPtr, - radeonVertex *, radeonVertex *); +#include "radeon_common.h" -typedef void (*radeon_point_func) (radeonContextPtr, radeonVertex *); - -struct radeon_colorbuffer_state { - GLuint clear; - int roundEnable; -}; -struct radeon_depthbuffer_state { - GLuint clear; - GLfloat scale; -}; +struct r100_context; +typedef struct r100_context r100ContextRec; +typedef struct r100_context *r100ContextPtr; -struct radeon_scissor_state { - drm_clip_rect_t rect; - GLboolean enabled; +#include "radeon_lock.h" - GLuint numClipRects; /* Cliprects active */ - GLuint numAllocedClipRects; /* Cliprects available */ - drm_clip_rect_t *pClipRects; -}; -struct radeon_stencilbuffer_state { - GLboolean hwBuffer; - GLuint clear; /* rb3d_stencilrefmask value */ -}; -struct radeon_stipple_state { - GLuint mask[32]; -}; +#define R100_TEX_ALL 0x7 /* used for both tcl_vtx and vc_frmt tex bits (they are identical) */ #define RADEON_ST_BIT(unit) \ @@ -141,42 +73,6 @@ struct radeon_stipple_state { #define RADEON_Q_BIT(unit) \ (unit == 0 ? RADEON_CP_VC_FRMT_Q0 : (RADEON_CP_VC_FRMT_Q1 >> 2) << (2 * unit)) -typedef struct radeon_tex_obj radeonTexObj, *radeonTexObjPtr; - -/* Texture object in locally shared texture space. - */ -struct radeon_tex_obj { - driTextureObject base; - - GLuint bufAddr; /* Offset to start of locally - shared texture block */ - - GLuint dirty_state; /* Flags (1 per texunit) for - whether or not this texobj - has dirty hardware state - (pp_*) that needs to be - brought into the - texunit. */ - - drm_radeon_tex_image_t image[6][RADEON_MAX_TEXTURE_LEVELS]; - /* Six, for the cube faces */ - - GLboolean image_override; /* Image overridden by GLX_EXT_tfp */ - - GLuint pp_txfilter; /* hardware register values */ - GLuint pp_txformat; - GLuint pp_txoffset; /* Image location in texmem. - All cube faces follow. */ - GLuint pp_txsize; /* npot only */ - GLuint pp_txpitch; /* npot only */ - GLuint pp_border_color; - GLuint pp_cubic_faces; /* cube face 1,2,3,4 log2 sizes */ - - GLboolean border_fallback; - - GLuint tile_bits; /* hw texture tile bits used on this texture */ -}; - struct radeon_texture_env_state { radeonTexObjPtr texobj; GLenum format; @@ -187,17 +83,6 @@ struct radeon_texture_state { struct radeon_texture_env_state unit[RADEON_MAX_TEXTURE_UNITS]; }; -struct radeon_state_atom { - struct radeon_state_atom *next, *prev; - const char *name; /* for debug */ - int cmd_size; /* size in bytes */ - GLuint is_tcl; - int *cmd; /* one or more cmd's */ - int *lastcmd; /* one or more cmd's */ - GLboolean dirty; /* dirty-mark in emit_state_list */ - GLboolean(*check) (GLcontext *); /* is this state active? */ -}; - /* Trying to keep these relatively short as the variables are becoming * extravagently long. Drop the driver name prefix off the front of * everything - I think we know which driver we're in by now, and keep the @@ -410,10 +295,16 @@ struct radeon_state_atom { #define SHN_SHININESS 1 #define SHN_STATE_SIZE 2 -struct radeon_hw_state { - /* Head of the linked list of state atoms. */ - struct radeon_state_atom atomlist; +#define R100_QUERYOBJ_CMD_0 0 +#define R100_QUERYOBJ_DATA_0 1 +#define R100_QUERYOBJ_CMDSIZE 2 + +#define STP_CMD_0 0 +#define STP_DATA_0 1 +#define STP_CMD_1 2 +#define STP_STATE_SIZE 35 +struct r100_hw_state { /* Hardware state, stored as cmdbuf commands: * -- Need to doublebuffer for * - eliding noop statechange loops? (except line stipple count) @@ -437,90 +328,19 @@ struct radeon_hw_state { struct radeon_state_atom fog; struct radeon_state_atom glt; struct radeon_state_atom txr[3]; /* for NPOT */ - - int max_state_size; /* Number of bytes necessary for a full state emit. */ - GLboolean is_dirty, all_dirty; -}; - -struct radeon_state { - /* Derived state for internal purposes: - */ - struct radeon_colorbuffer_state color; - struct radeon_depthbuffer_state depth; - struct radeon_scissor_state scissor; - struct radeon_stencilbuffer_state stencil; - struct radeon_stipple_state stipple; - struct radeon_texture_state texture; -}; - -/* Need refcounting on dma buffers: - */ -struct radeon_dma_buffer { - int refcount; /* the number of retained regions in buf */ - drmBufPtr buf; -}; - -#define GET_START(rvb) (rmesa->radeonScreen->gart_buffer_offset + \ - (rvb)->address - rmesa->dma.buf0_address + \ - (rvb)->start) - -/* A retained region, eg vertices for indexed vertices. - */ -struct radeon_dma_region { - struct radeon_dma_buffer *buf; - char *address; /* == buf->address */ - int start, end, ptr; /* offsets from start of buf */ - int aos_start; - int aos_stride; - int aos_size; -}; - -struct radeon_dma { - /* Active dma region. Allocations for vertices and retained - * regions come from here. Also used for emitting random vertices, - * these may be flushed by calling flush_current(); - */ - struct radeon_dma_region current; - - void (*flush) (radeonContextPtr); - - char *buf0_address; /* start of buf[0], for index calcs */ - GLuint nr_released_bufs; /* flush after so many buffers released */ + struct radeon_state_atom stp; }; -struct radeon_dri_mirror { - __DRIcontextPrivate *context; /* DRI context */ - __DRIscreenPrivate *screen; /* DRI screen */ - - /** - * DRI drawable bound to this context for drawing. - */ - __DRIdrawablePrivate *drawable; - /** - * DRI drawable bound to this context for reading. - */ - __DRIdrawablePrivate *readable; - - drm_context_t hwContext; - drm_hw_lock_t *hwLock; - int fd; - int drmMinor; +struct r100_state { + struct radeon_texture_state texture; }; #define RADEON_CMD_BUF_SZ (8*1024) - -struct radeon_store { - GLuint statenr; - GLuint primnr; - char cmd_buf[RADEON_CMD_BUF_SZ]; - int cmd_used; - int elts_start; -}; - +#define R200_ELT_BUF_SZ (8*1024) /* radeon_tcl.c */ -struct radeon_tcl_info { +struct r100_tcl_info { GLuint vertex_format; GLuint hw_primitive; @@ -529,30 +349,18 @@ struct radeon_tcl_info { */ GLvector4f ObjClean; - struct radeon_dma_region *aos_components[8]; - GLuint nr_aos_components; - GLuint *Elts; - struct radeon_dma_region indexed_verts; - struct radeon_dma_region obj; - struct radeon_dma_region rgba; - struct radeon_dma_region spec; - struct radeon_dma_region fog; - struct radeon_dma_region tex[RADEON_MAX_TEXTURE_UNITS]; - struct radeon_dma_region norm; + int elt_cmd_offset; + int elt_cmd_start; + int elt_used; }; /* radeon_swtcl.c */ -struct radeon_swtcl_info { - GLuint RenderIndex; - GLuint vertex_size; +struct r100_swtcl_info { GLuint vertex_format; - struct tnl_attr_map vertex_attrs[VERT_ATTRIB_MAX]; - GLuint vertex_attr_count; - GLubyte *verts; /* Fallback rasterization functions @@ -561,10 +369,6 @@ struct radeon_swtcl_info { radeon_line_func draw_line; radeon_tri_func draw_tri; - GLuint hw_primitive; - GLenum render_primitive; - GLuint numverts; - /** * Offset of the 4UB color data within a hardware (swtcl) vertex. */ @@ -576,22 +380,9 @@ struct radeon_swtcl_info { GLuint specoffset; GLboolean needproj; - - struct radeon_dma_region indexed_verts; }; -struct radeon_ioctl { - GLuint vertex_offset; - GLuint vertex_size; -}; -#define RADEON_MAX_PRIMS 64 - -struct radeon_prim { - GLuint start; - GLuint end; - GLuint prim; -}; /* A maximum total of 20 elements per vertex: 3 floats for position, 3 * floats for normal, 4 floats for color, 4 bytes for secondary color, @@ -602,59 +393,18 @@ struct radeon_prim { */ #define RADEON_MAX_VERTEX_SIZE 20 -struct radeon_context { - GLcontext *glCtx; /* Mesa context */ +struct r100_context { + struct radeon_context radeon; /* Driver and hardware state management */ - struct radeon_hw_state hw; - struct radeon_state state; - - /* Texture object bookkeeping - */ - unsigned nr_heaps; - driTexHeap *texture_heaps[RADEON_NR_TEX_HEAPS]; - driTextureObject swapped; - int texture_depth; - float initialMaxAnisotropy; - - /* Rasterization and vertex state: - */ - GLuint TclFallback; - GLuint Fallback; - GLuint NewGLState; - DECLARE_RENDERINPUTS(tnl_index_bitset); /* index of bits for last tnl_install_attrs */ + struct r100_hw_state hw; + struct r100_state state; /* Vertex buffers */ struct radeon_ioctl ioctl; - struct radeon_dma dma; struct radeon_store store; - /* A full state emit as of the first state emit in the main store, in case - * the context is lost. - */ - struct radeon_store backup_store; - - /* Page flipping - */ - GLuint doPageFlip; - - /* Busy waiting - */ - GLuint do_usleeps; - GLuint do_irqs; - GLuint irqsEmitted; - drm_radeon_irq_wait_t iw; - - /* Drawable, cliprect and scissor information - */ - GLuint numClipRects; /* Cliprects for the draw buffer */ - drm_clip_rect_t *pClipRects; - unsigned int lastStamp; - GLboolean lost_context; - GLboolean save_on_next_emit; - radeonScreenPtr radeonScreen; /* Screen private DRI data */ - drm_radeon_sarea_t *sarea; /* Private SAREA data */ /* TCL stuff */ @@ -667,29 +417,13 @@ struct radeon_context { GLmatrix tmpmat[RADEON_MAX_TEXTURE_UNITS]; GLuint last_ReallyEnabled; - /* VBI - */ - int64_t swap_ust; - int64_t swap_missed_ust; - - GLuint swap_count; - GLuint swap_missed_count; - /* radeon_tcl.c */ - struct radeon_tcl_info tcl; + struct r100_tcl_info tcl; /* radeon_swtcl.c */ - struct radeon_swtcl_info swtcl; - - /* Mirrors of some DRI state - */ - struct radeon_dri_mirror dri; - - /* Configuration cache - */ - driOptionCache optionCache; + struct r100_swtcl_info swtcl; GLboolean using_hyperz; GLboolean texmicrotile; @@ -703,61 +437,19 @@ struct radeon_context { GLuint c_textureSwaps; GLuint c_textureBytes; GLuint c_vertexBuffers; + }; -#define RADEON_CONTEXT(ctx) ((radeonContextPtr)(ctx->DriverCtx)) - -static INLINE GLuint radeonPackColor(GLuint cpp, - GLubyte r, GLubyte g, - GLubyte b, GLubyte a) -{ - switch (cpp) { - case 2: - return PACK_COLOR_565(r, g, b); - case 4: - return PACK_COLOR_8888(a, r, g, b); - default: - return 0; - } -} + +#define R100_CONTEXT(ctx) ((r100ContextPtr)(ctx->DriverCtx)) + #define RADEON_OLD_PACKETS 1 -extern void radeonDestroyContext(__DRIcontextPrivate * driContextPriv); -extern GLboolean radeonCreateContext(const __GLcontextModes * glVisual, - __DRIcontextPrivate * driContextPriv, - void *sharedContextPrivate); -extern void radeonSwapBuffers(__DRIdrawablePrivate * dPriv); -extern void radeonCopySubBuffer(__DRIdrawablePrivate * dPriv, - int x, int y, int w, int h); -extern GLboolean radeonMakeCurrent(__DRIcontextPrivate * driContextPriv, - __DRIdrawablePrivate * driDrawPriv, - __DRIdrawablePrivate * driReadPriv); -extern GLboolean radeonUnbindContext(__DRIcontextPrivate * driContextPriv); - -/* ================================================================ - * Debugging: - */ -#define DO_DEBUG 1 - -#if DO_DEBUG -extern int RADEON_DEBUG; -#else -#define RADEON_DEBUG 0 -#endif - -#define DEBUG_TEXTURE 0x0001 -#define DEBUG_STATE 0x0002 -#define DEBUG_IOCTL 0x0004 -#define DEBUG_PRIMS 0x0008 -#define DEBUG_VERTS 0x0010 -#define DEBUG_FALLBACKS 0x0020 -#define DEBUG_VFMT 0x0040 -#define DEBUG_CODEGEN 0x0080 -#define DEBUG_VERBOSE 0x0100 -#define DEBUG_DRI 0x0200 -#define DEBUG_DMA 0x0400 -#define DEBUG_SANITY 0x0800 -#define DEBUG_SYNC 0x1000 +extern GLboolean r100CreateContext( const __GLcontextModes *glVisual, + __DRIcontextPrivate *driContextPriv, + void *sharedContextPrivate); + + #endif /* __RADEON_CONTEXT_H__ */ diff --git a/src/mesa/drivers/dri/radeon/radeon_cs_drm.h b/src/mesa/drivers/dri/radeon/radeon_cs_drm.h new file mode 100644 index 00000000000..ab4eca31a3c --- /dev/null +++ b/src/mesa/drivers/dri/radeon/radeon_cs_drm.h @@ -0,0 +1,246 @@ +/* + * Copyright © 2008 Nicolai Haehnle + * Copyright © 2008 Jérôme Glisse + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ +/* + * Authors: + * Aapo Tahkola <aet@rasterburn.org> + * Nicolai Haehnle <prefect_@gmx.net> + * Jérôme Glisse <glisse@freedesktop.org> + */ +#ifndef RADEON_CS_H +#define RADEON_CS_H + +#include <stdint.h> +#include <string.h> +#include "drm.h" +#include "radeon_drm.h" + +struct radeon_cs_reloc { + struct radeon_bo *bo; + uint32_t read_domain; + uint32_t write_domain; + uint32_t flags; +}; + + +#define RADEON_CS_SPACE_OK 0 +#define RADEON_CS_SPACE_OP_TO_BIG 1 +#define RADEON_CS_SPACE_FLUSH 2 + +struct radeon_cs_space_check { + struct radeon_bo *bo; + uint32_t read_domains; + uint32_t write_domain; + uint32_t new_accounted; +}; + +#define MAX_SPACE_BOS (32) + +struct radeon_cs_manager; + +struct radeon_cs { + struct radeon_cs_manager *csm; + void *relocs; + uint32_t *packets; + unsigned crelocs; + unsigned relocs_total_size; + unsigned cdw; + unsigned ndw; + int section; + unsigned section_ndw; + unsigned section_cdw; + const char *section_file; + const char *section_func; + int section_line; + struct radeon_cs_space_check bos[MAX_SPACE_BOS]; + int bo_count; + void (*space_flush_fn)(void *); + void *space_flush_data; +}; + +/* cs functions */ +struct radeon_cs_funcs { + struct radeon_cs *(*cs_create)(struct radeon_cs_manager *csm, + uint32_t ndw); + int (*cs_write_reloc)(struct radeon_cs *cs, + struct radeon_bo *bo, + uint32_t read_domain, + uint32_t write_domain, + uint32_t flags); + int (*cs_begin)(struct radeon_cs *cs, + uint32_t ndw, + const char *file, + const char *func, + int line); + int (*cs_end)(struct radeon_cs *cs, + const char *file, + const char *func, + int line); + int (*cs_emit)(struct radeon_cs *cs); + int (*cs_destroy)(struct radeon_cs *cs); + int (*cs_erase)(struct radeon_cs *cs); + int (*cs_need_flush)(struct radeon_cs *cs); + void (*cs_print)(struct radeon_cs *cs, FILE *file); +}; + +struct radeon_cs_manager { + struct radeon_cs_funcs *funcs; + int fd; + int32_t vram_limit, gart_limit; + int32_t vram_write_used, gart_write_used; + int32_t read_used; +}; + +static inline struct radeon_cs *radeon_cs_create(struct radeon_cs_manager *csm, + uint32_t ndw) +{ + return csm->funcs->cs_create(csm, ndw); +} + +static inline int radeon_cs_write_reloc(struct radeon_cs *cs, + struct radeon_bo *bo, + uint32_t read_domain, + uint32_t write_domain, + uint32_t flags) +{ + return cs->csm->funcs->cs_write_reloc(cs, + bo, + read_domain, + write_domain, + flags); +} + +static inline int radeon_cs_begin(struct radeon_cs *cs, + uint32_t ndw, + const char *file, + const char *func, + int line) +{ + return cs->csm->funcs->cs_begin(cs, ndw, file, func, line); +} + +static inline int radeon_cs_end(struct radeon_cs *cs, + const char *file, + const char *func, + int line) +{ + return cs->csm->funcs->cs_end(cs, file, func, line); +} + +static inline int radeon_cs_emit(struct radeon_cs *cs) +{ + return cs->csm->funcs->cs_emit(cs); +} + +static inline int radeon_cs_destroy(struct radeon_cs *cs) +{ + return cs->csm->funcs->cs_destroy(cs); +} + +static inline int radeon_cs_erase(struct radeon_cs *cs) +{ + return cs->csm->funcs->cs_erase(cs); +} + +static inline int radeon_cs_need_flush(struct radeon_cs *cs) +{ + return cs->csm->funcs->cs_need_flush(cs); +} + +static inline void radeon_cs_print(struct radeon_cs *cs, FILE *file) +{ + cs->csm->funcs->cs_print(cs, file); +} + +static inline void radeon_cs_set_limit(struct radeon_cs *cs, uint32_t domain, uint32_t limit) +{ + + if (domain == RADEON_GEM_DOMAIN_VRAM) + cs->csm->vram_limit = limit; + else + cs->csm->gart_limit = limit; +} + +static inline void radeon_cs_write_dword(struct radeon_cs *cs, uint32_t dword) +{ + cs->packets[cs->cdw++] = dword; + if (cs->section) { + cs->section_cdw++; + } +} + +static inline void radeon_cs_write_qword(struct radeon_cs *cs, uint64_t qword) +{ + + memcpy(cs->packets + cs->cdw, &qword, sizeof(qword)); + cs->cdw+=2; + if (cs->section) { + cs->section_cdw+=2; + } +} + +static inline void radeon_cs_write_table(struct radeon_cs *cs, void *data, uint32_t size) +{ + memcpy(cs->packets + cs->cdw, data, size * 4); + cs->cdw += size; + if (cs->section) { + cs->section_cdw += size; + } +} + +static inline void radeon_cs_space_set_flush(struct radeon_cs *cs, void (*fn)(void *), void *data) +{ + cs->space_flush_fn = fn; + cs->space_flush_data = data; +} + + +/* + * add a persistent BO to the list + * a persistent BO is one that will be referenced across flushes, + * i.e. colorbuffer, textures etc. + * They get reset when a new "operation" happens, where an operation + * is a state emission with a color/textures etc followed by a bunch of vertices. + */ +void radeon_cs_space_add_persistent_bo(struct radeon_cs *cs, + struct radeon_bo *bo, + uint32_t read_domains, + uint32_t write_domain); + +/* reset the persistent BO list */ +void radeon_cs_space_reset_bos(struct radeon_cs *cs); + +/* do a space check with the current persistent BO list */ +int radeon_cs_space_check(struct radeon_cs *cs); + +/* do a space check with the current persistent BO list and a temporary BO + * a temporary BO is like a DMA buffer, which gets flushed with the + * command buffer */ +int radeon_cs_space_check_with_bo(struct radeon_cs *cs, + struct radeon_bo *bo, + uint32_t read_domains, + uint32_t write_domain); + +#endif diff --git a/src/mesa/drivers/dri/radeon/radeon_cs_legacy.c b/src/mesa/drivers/dri/radeon/radeon_cs_legacy.c new file mode 100644 index 00000000000..f1addb299e2 --- /dev/null +++ b/src/mesa/drivers/dri/radeon/radeon_cs_legacy.c @@ -0,0 +1,409 @@ +/* + * Copyright © 2008 Nicolai Haehnle + * Copyright © 2008 Jérôme Glisse + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ +/* + * Authors: + * Aapo Tahkola <aet@rasterburn.org> + * Nicolai Haehnle <prefect_@gmx.net> + * Jérôme Glisse <glisse@freedesktop.org> + */ +#include <errno.h> + +#include "radeon_bocs_wrapper.h" +#include "radeon_common.h" + +struct cs_manager_legacy { + struct radeon_cs_manager base; + struct radeon_context *ctx; + /* hack for scratch stuff */ + uint32_t pending_age; + uint32_t pending_count; + + +}; + +struct cs_reloc_legacy { + struct radeon_cs_reloc base; + uint32_t cindices; + uint32_t *indices; +}; + + +static struct radeon_cs *cs_create(struct radeon_cs_manager *csm, + uint32_t ndw) +{ + struct radeon_cs *cs; + + cs = (struct radeon_cs*)calloc(1, sizeof(struct radeon_cs)); + if (cs == NULL) { + return NULL; + } + cs->csm = csm; + cs->ndw = (ndw + 0x3FF) & (~0x3FF); + cs->packets = (uint32_t*)malloc(4*cs->ndw); + if (cs->packets == NULL) { + free(cs); + return NULL; + } + cs->relocs_total_size = 0; + return cs; +} + +static int cs_write_reloc(struct radeon_cs *cs, + struct radeon_bo *bo, + uint32_t read_domain, + uint32_t write_domain, + uint32_t flags) +{ + struct cs_reloc_legacy *relocs; + int i; + + relocs = (struct cs_reloc_legacy *)cs->relocs; + /* check domains */ + if ((read_domain && write_domain) || (!read_domain && !write_domain)) { + /* in one CS a bo can only be in read or write domain but not + * in read & write domain at the same sime + */ + return -EINVAL; + } + if (read_domain == RADEON_GEM_DOMAIN_CPU) { + return -EINVAL; + } + if (write_domain == RADEON_GEM_DOMAIN_CPU) { + return -EINVAL; + } + /* check if bo is already referenced */ + for(i = 0; i < cs->crelocs; i++) { + uint32_t *indices; + + if (relocs[i].base.bo->handle == bo->handle) { + /* Check domains must be in read or write. As we check already + * checked that in argument one of the read or write domain was + * set we only need to check that if previous reloc as the read + * domain set then the read_domain should also be set for this + * new relocation. + */ + if (relocs[i].base.read_domain && !read_domain) { + return -EINVAL; + } + if (relocs[i].base.write_domain && !write_domain) { + return -EINVAL; + } + relocs[i].base.read_domain |= read_domain; + relocs[i].base.write_domain |= write_domain; + /* save indice */ + relocs[i].cindices++; + indices = (uint32_t*)realloc(relocs[i].indices, + relocs[i].cindices * 4); + if (indices == NULL) { + relocs[i].cindices -= 1; + return -ENOMEM; + } + relocs[i].indices = indices; + relocs[i].indices[relocs[i].cindices - 1] = cs->cdw - 1; + return 0; + } + } + /* add bo to reloc */ + relocs = (struct cs_reloc_legacy*) + realloc(cs->relocs, + sizeof(struct cs_reloc_legacy) * (cs->crelocs + 1)); + if (relocs == NULL) { + return -ENOMEM; + } + cs->relocs = relocs; + relocs[cs->crelocs].base.bo = bo; + relocs[cs->crelocs].base.read_domain = read_domain; + relocs[cs->crelocs].base.write_domain = write_domain; + relocs[cs->crelocs].base.flags = flags; + relocs[cs->crelocs].indices = (uint32_t*)malloc(4); + if (relocs[cs->crelocs].indices == NULL) { + return -ENOMEM; + } + relocs[cs->crelocs].indices[0] = cs->cdw - 1; + relocs[cs->crelocs].cindices = 1; + cs->relocs_total_size += radeon_bo_legacy_relocs_size(bo); + cs->crelocs++; + radeon_bo_ref(bo); + return 0; +} + +static int cs_begin(struct radeon_cs *cs, + uint32_t ndw, + const char *file, + const char *func, + int line) +{ + if (cs->section) { + fprintf(stderr, "CS already in a section(%s,%s,%d)\n", + cs->section_file, cs->section_func, cs->section_line); + fprintf(stderr, "CS can't start section(%s,%s,%d)\n", + file, func, line); + return -EPIPE; + } + cs->section = 1; + cs->section_ndw = ndw; + cs->section_cdw = 0; + cs->section_file = file; + cs->section_func = func; + cs->section_line = line; + + + if (cs->cdw + ndw > cs->ndw) { + uint32_t tmp, *ptr; + int num = (ndw > 0x3FF) ? ndw : 0x3FF; + + tmp = (cs->cdw + 1 + num) & (~num); + ptr = (uint32_t*)realloc(cs->packets, 4 * tmp); + if (ptr == NULL) { + return -ENOMEM; + } + cs->packets = ptr; + cs->ndw = tmp; + } + + return 0; +} + +static int cs_end(struct radeon_cs *cs, + const char *file, + const char *func, + int line) + +{ + if (!cs->section) { + fprintf(stderr, "CS no section to end at (%s,%s,%d)\n", + file, func, line); + return -EPIPE; + } + cs->section = 0; + if (cs->section_ndw != cs->section_cdw) { + fprintf(stderr, "CS section size missmatch start at (%s,%s,%d) %d vs %d\n", + cs->section_file, cs->section_func, cs->section_line, cs->section_ndw, cs->section_cdw); + fprintf(stderr, "CS section end at (%s,%s,%d)\n", + file, func, line); + return -EPIPE; + } + return 0; +} + +static int cs_process_relocs(struct radeon_cs *cs) +{ + struct cs_manager_legacy *csm = (struct cs_manager_legacy*)cs->csm; + struct cs_reloc_legacy *relocs; + int i, j, r; + + csm = (struct cs_manager_legacy*)cs->csm; + relocs = (struct cs_reloc_legacy *)cs->relocs; +restart: + for (i = 0; i < cs->crelocs; i++) + { + for (j = 0; j < relocs[i].cindices; j++) + { + uint32_t soffset, eoffset; + + r = radeon_bo_legacy_validate(relocs[i].base.bo, + &soffset, &eoffset); + if (r == -EAGAIN) + { + goto restart; + } + if (r) + { + fprintf(stderr, "validated %p [0x%08X, 0x%08X]\n", + relocs[i].base.bo, soffset, eoffset); + return r; + } + cs->packets[relocs[i].indices[j]] += soffset; + if (cs->packets[relocs[i].indices[j]] >= eoffset) + { + /* radeon_bo_debug(relocs[i].base.bo, 12); */ + fprintf(stderr, "validated %p [0x%08X, 0x%08X]\n", + relocs[i].base.bo, soffset, eoffset); + fprintf(stderr, "above end: %p 0x%08X 0x%08X\n", + relocs[i].base.bo, + cs->packets[relocs[i].indices[j]], + eoffset); + exit(0); + return -EINVAL; + } + } + } + return 0; +} + +static int cs_set_age(struct radeon_cs *cs) +{ + struct cs_manager_legacy *csm = (struct cs_manager_legacy*)cs->csm; + struct cs_reloc_legacy *relocs; + int i; + + relocs = (struct cs_reloc_legacy *)cs->relocs; + for (i = 0; i < cs->crelocs; i++) { + radeon_bo_legacy_pending(relocs[i].base.bo, csm->pending_age); + radeon_bo_unref(relocs[i].base.bo); + } + return 0; +} + +static int cs_emit(struct radeon_cs *cs) +{ + struct cs_manager_legacy *csm = (struct cs_manager_legacy*)cs->csm; + drm_radeon_cmd_buffer_t cmd; + drm_r300_cmd_header_t age; + uint64_t ull; + int r; + + csm->ctx->vtbl.emit_cs_header(cs, csm->ctx); + + /* append buffer age */ + if ( IS_R300_CLASS(csm->ctx->radeonScreen) ) + { + age.scratch.cmd_type = R300_CMD_SCRATCH; + /* Scratch register 2 corresponds to what radeonGetAge polls */ + csm->pending_age = 0; + csm->pending_count = 1; + ull = (uint64_t) (intptr_t) &csm->pending_age; + age.scratch.reg = 2; + age.scratch.n_bufs = 1; + age.scratch.flags = 0; + radeon_cs_write_dword(cs, age.u); + radeon_cs_write_qword(cs, ull); + radeon_cs_write_dword(cs, 0); + } + + r = cs_process_relocs(cs); + if (r) { + return 0; + } + + cmd.buf = (char *)cs->packets; + cmd.bufsz = cs->cdw * 4; + if (csm->ctx->state.scissor.enabled) { + cmd.nbox = csm->ctx->state.scissor.numClipRects; + cmd.boxes = (drm_clip_rect_t *) csm->ctx->state.scissor.pClipRects; + } else { + cmd.nbox = csm->ctx->numClipRects; + cmd.boxes = (drm_clip_rect_t *) csm->ctx->pClipRects; + } + + //dump_cmdbuf(cs); + + r = drmCommandWrite(cs->csm->fd, DRM_RADEON_CMDBUF, &cmd, sizeof(cmd)); + if (r) { + return r; + } + if ((!IS_R300_CLASS(csm->ctx->radeonScreen)) && + (!IS_R600_CLASS(csm->ctx->radeonScreen))) { /* +r6/r7 : No irq for r6/r7 yet. */ + drm_radeon_irq_emit_t emit_cmd; + emit_cmd.irq_seq = (int*)&csm->pending_age; + r = drmCommandWrite(cs->csm->fd, DRM_RADEON_IRQ_EMIT, &emit_cmd, sizeof(emit_cmd)); + if (r) { + return r; + } + } + cs_set_age(cs); + + cs->csm->read_used = 0; + cs->csm->vram_write_used = 0; + cs->csm->gart_write_used = 0; + return 0; +} + +static void inline cs_free_reloc(void *relocs_p, int crelocs) +{ + struct cs_reloc_legacy *relocs = relocs_p; + int i; + if (!relocs_p) + return; + for (i = 0; i < crelocs; i++) + free(relocs[i].indices); +} + +static int cs_destroy(struct radeon_cs *cs) +{ + cs_free_reloc(cs->relocs, cs->crelocs); + free(cs->relocs); + free(cs->packets); + free(cs); + return 0; +} + +static int cs_erase(struct radeon_cs *cs) +{ + cs_free_reloc(cs->relocs, cs->crelocs); + free(cs->relocs); + cs->relocs_total_size = 0; + cs->relocs = NULL; + cs->crelocs = 0; + cs->cdw = 0; + cs->section = 0; + return 0; +} + +static int cs_need_flush(struct radeon_cs *cs) +{ + /* this function used to flush when the BO usage got to + * a certain size, now the higher levels handle this better */ + return 0; +} + +static void cs_print(struct radeon_cs *cs, FILE *file) +{ +} + +static struct radeon_cs_funcs radeon_cs_legacy_funcs = { + cs_create, + cs_write_reloc, + cs_begin, + cs_end, + cs_emit, + cs_destroy, + cs_erase, + cs_need_flush, + cs_print, +}; + +struct radeon_cs_manager *radeon_cs_manager_legacy_ctor(struct radeon_context *ctx) +{ + struct cs_manager_legacy *csm; + + csm = (struct cs_manager_legacy*) + calloc(1, sizeof(struct cs_manager_legacy)); + if (csm == NULL) { + return NULL; + } + csm->base.funcs = &radeon_cs_legacy_funcs; + csm->base.fd = ctx->dri.fd; + csm->ctx = ctx; + csm->pending_age = 1; + return (struct radeon_cs_manager*)csm; +} + +void radeon_cs_manager_legacy_dtor(struct radeon_cs_manager *csm) +{ + free(csm); +} + diff --git a/src/mesa/drivers/dri/radeon/radeon_cs_legacy.h b/src/mesa/drivers/dri/radeon/radeon_cs_legacy.h new file mode 100644 index 00000000000..cafbc9e576e --- /dev/null +++ b/src/mesa/drivers/dri/radeon/radeon_cs_legacy.h @@ -0,0 +1,40 @@ +/* + * Copyright © 2008 Nicolai Haehnle + * Copyright © 2008 Jérôme Glisse + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ +/* + * Authors: + * Aapo Tahkola <aet@rasterburn.org> + * Nicolai Haehnle <prefect_@gmx.net> + * Jérôme Glisse <glisse@freedesktop.org> + */ +#ifndef RADEON_CS_LEGACY_H +#define RADEON_CS_LEGACY_H + +struct radeon_context; + +struct radeon_cs_manager *radeon_cs_manager_legacy_ctor(struct radeon_context *ctx); +void radeon_cs_manager_legacy_dtor(struct radeon_cs_manager *csm); + +#endif diff --git a/src/mesa/drivers/dri/radeon/radeon_cs_space_drm.c b/src/mesa/drivers/dri/radeon/radeon_cs_space_drm.c new file mode 100644 index 00000000000..89cbbb5a6b9 --- /dev/null +++ b/src/mesa/drivers/dri/radeon/radeon_cs_space_drm.c @@ -0,0 +1,234 @@ +/* + * Copyright © 2009 Red Hat Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES + * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS + * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ +/* + */ +#include <assert.h> +#include <errno.h> +#include <stdlib.h> +#include "radeon_bocs_wrapper.h" + +struct rad_sizes { + int32_t op_read; + int32_t op_gart_write; + int32_t op_vram_write; +}; + +static inline int radeon_cs_setup_bo(struct radeon_cs_space_check *sc, struct rad_sizes *sizes) +{ + uint32_t read_domains, write_domain; + struct radeon_bo *bo; + + bo = sc->bo; + sc->new_accounted = 0; + read_domains = sc->read_domains; + write_domain = sc->write_domain; + + /* legacy needs a static check */ + if (radeon_bo_is_static(bo)) { + bo->space_accounted = sc->new_accounted = (read_domains << 16) | write_domain; + return 0; + } + + /* already accounted this bo */ + if (write_domain && (write_domain == bo->space_accounted)) { + sc->new_accounted = bo->space_accounted; + return 0; + } + if (read_domains && ((read_domains << 16) == bo->space_accounted)) { + sc->new_accounted = bo->space_accounted; + return 0; + } + + if (bo->space_accounted == 0) { + if (write_domain == RADEON_GEM_DOMAIN_VRAM) + sizes->op_vram_write += bo->size; + else if (write_domain == RADEON_GEM_DOMAIN_GTT) + sizes->op_gart_write += bo->size; + else + sizes->op_read += bo->size; + sc->new_accounted = (read_domains << 16) | write_domain; + } else { + uint16_t old_read, old_write; + + old_read = bo->space_accounted >> 16; + old_write = bo->space_accounted & 0xffff; + + if (write_domain && (old_read & write_domain)) { + sc->new_accounted = write_domain; + /* moving from read to a write domain */ + if (write_domain == RADEON_GEM_DOMAIN_VRAM) { + sizes->op_read -= bo->size; + sizes->op_vram_write += bo->size; + } else if (write_domain == RADEON_GEM_DOMAIN_GTT) { + sizes->op_read -= bo->size; + sizes->op_gart_write += bo->size; + } + } else if (read_domains & old_write) { + sc->new_accounted = bo->space_accounted & 0xffff; + } else { + /* rewrite the domains */ + if (write_domain != old_write) + fprintf(stderr,"WRITE DOMAIN RELOC FAILURE 0x%x %d %d\n", bo->handle, write_domain, old_write); + if (read_domains != old_read) + fprintf(stderr,"READ DOMAIN RELOC FAILURE 0x%x %d %d\n", bo->handle, read_domains, old_read); + return RADEON_CS_SPACE_FLUSH; + } + } + return 0; +} + +static int radeon_cs_do_space_check(struct radeon_cs *cs, struct radeon_cs_space_check *new_tmp) +{ + struct radeon_cs_manager *csm = cs->csm; + int i; + struct radeon_bo *bo; + struct rad_sizes sizes; + int ret; + + /* check the totals for this operation */ + + if (cs->bo_count == 0 && !new_tmp) + return 0; + + memset(&sizes, 0, sizeof(struct rad_sizes)); + + /* prepare */ + for (i = 0; i < cs->bo_count; i++) { + ret = radeon_cs_setup_bo(&cs->bos[i], &sizes); + if (ret) + return ret; + } + + if (new_tmp) { + ret = radeon_cs_setup_bo(new_tmp, &sizes); + if (ret) + return ret; + } + + if (sizes.op_read < 0) + sizes.op_read = 0; + + /* check sizes - operation first */ + if ((sizes.op_read + sizes.op_gart_write > csm->gart_limit) || + (sizes.op_vram_write > csm->vram_limit)) { + return RADEON_CS_SPACE_OP_TO_BIG; + } + + if (((csm->vram_write_used + sizes.op_vram_write) > csm->vram_limit) || + ((csm->read_used + csm->gart_write_used + sizes.op_gart_write + sizes.op_read) > csm->gart_limit)) { + return RADEON_CS_SPACE_FLUSH; + } + + csm->gart_write_used += sizes.op_gart_write; + csm->vram_write_used += sizes.op_vram_write; + csm->read_used += sizes.op_read; + /* commit */ + for (i = 0; i < cs->bo_count; i++) { + bo = cs->bos[i].bo; + bo->space_accounted = cs->bos[i].new_accounted; + } + if (new_tmp) + new_tmp->bo->space_accounted = new_tmp->new_accounted; + + return RADEON_CS_SPACE_OK; +} + +void radeon_cs_space_add_persistent_bo(struct radeon_cs *cs, struct radeon_bo *bo, uint32_t read_domains, uint32_t write_domain) +{ + int i; + for (i = 0; i < cs->bo_count; i++) { + if (cs->bos[i].bo == bo && + cs->bos[i].read_domains == read_domains && + cs->bos[i].write_domain == write_domain) + return; + } + radeon_bo_ref(bo); + i = cs->bo_count; + cs->bos[i].bo = bo; + cs->bos[i].read_domains = read_domains; + cs->bos[i].write_domain = write_domain; + cs->bos[i].new_accounted = 0; + cs->bo_count++; + + assert(cs->bo_count < MAX_SPACE_BOS); +} + +static int radeon_cs_check_space_internal(struct radeon_cs *cs, struct radeon_cs_space_check *tmp_bo) +{ + int ret; + int flushed = 0; + +again: + ret = radeon_cs_do_space_check(cs, tmp_bo); + if (ret == RADEON_CS_SPACE_OP_TO_BIG) + return -1; + if (ret == RADEON_CS_SPACE_FLUSH) { + (*cs->space_flush_fn)(cs->space_flush_data); + if (flushed) + return -1; + flushed = 1; + goto again; + } + return 0; +} + +int radeon_cs_space_check_with_bo(struct radeon_cs *cs, + struct radeon_bo *bo, + uint32_t read_domains, uint32_t write_domain) +{ + struct radeon_cs_space_check temp_bo; + int ret = 0; + + if (bo) { + temp_bo.bo = bo; + temp_bo.read_domains = read_domains; + temp_bo.write_domain = write_domain; + temp_bo.new_accounted = 0; + } + + ret = radeon_cs_check_space_internal(cs, bo ? &temp_bo : NULL); + return ret; +} + +int radeon_cs_space_check(struct radeon_cs *cs) +{ + return radeon_cs_check_space_internal(cs, NULL); +} + +void radeon_cs_space_reset_bos(struct radeon_cs *cs) +{ + int i; + for (i = 0; i < cs->bo_count; i++) { + radeon_bo_unref(cs->bos[i].bo); + cs->bos[i].bo = NULL; + cs->bos[i].read_domains = 0; + cs->bos[i].write_domain = 0; + cs->bos[i].new_accounted = 0; + } + cs->bo_count = 0; +} + + diff --git a/src/mesa/drivers/dri/radeon/radeon_debug.c b/src/mesa/drivers/dri/radeon/radeon_debug.c new file mode 100644 index 00000000000..3b6f0038037 --- /dev/null +++ b/src/mesa/drivers/dri/radeon/radeon_debug.c @@ -0,0 +1,107 @@ +/* + * Copyright © 2009 Pauli Nieminen + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ +/* + * Authors: + * Pauli Nieminen <suokkos@gmail.com> + */ + +#include "utils.h" + +#include "radeon_debug.h" +#include "radeon_common_context.h" + +#include <stdarg.h> +#include <stdio.h> + +static const struct dri_debug_control debug_control[] = { + {"fall", RADEON_FALLBACKS}, + {"tex", RADEON_TEXTURE}, + {"ioctl", RADEON_IOCTL}, + {"verts", RADEON_RENDER}, + {"render", RADEON_RENDER}, + {"swrender", RADEON_SWRENDER}, + {"state", RADEON_STATE}, + {"shader", RADEON_SHADER}, + {"vfmt", RADEON_VFMT}, + {"vtxf", RADEON_VFMT}, + {"dri", RADEON_DRI}, + {"dma", RADEON_DMA}, + {"sanity", RADEON_SANITY}, + {"sync", RADEON_SYNC}, + {"pixel", RADEON_PIXEL}, + {"mem", RADEON_MEMORY}, + {"cs", RADEON_CS}, + {"allmsg", ~RADEON_SYNC}, /* avoid the term "sync" because the parser uses strstr */ + {NULL, 0} +}; + +radeon_debug_type_t radeon_enabled_debug_types; + +void radeon_init_debug(void) +{ + radeon_enabled_debug_types = driParseDebugString(getenv("RADEON_DEBUG"), debug_control); + + radeon_enabled_debug_types |= RADEON_GENERAL; +} + +void _radeon_debug_add_indent(void) +{ + GET_CURRENT_CONTEXT(ctx); + radeonContextPtr radeon = RADEON_CONTEXT(ctx); + const size_t length = sizeof(radeon->debug.indent) + / sizeof(radeon->debug.indent[0]); + if (radeon->debug.indent_depth < length - 1) { + radeon->debug.indent[radeon->debug.indent_depth] = '\t'; + ++radeon->debug.indent_depth; + }; +} + +void _radeon_debug_remove_indent(void) +{ + GET_CURRENT_CONTEXT(ctx); + radeonContextPtr radeon = RADEON_CONTEXT(ctx); + if (radeon->debug.indent_depth > 0) { + radeon->debug.indent[radeon->debug.indent_depth] = '\0'; + --radeon->debug.indent_depth; + } +} + +void _radeon_print(const radeon_debug_type_t type, + const radeon_debug_level_t level, + const char* message, + ...) +{ + GET_CURRENT_CONTEXT(ctx); + if (ctx) { + radeonContextPtr radeon = RADEON_CONTEXT(ctx); + // FIXME: Make this multi thread safe + if (radeon->debug.indent_depth) + fprintf(stderr, "%s", radeon->debug.indent); + } + va_list values; + va_start( values, message ); + vfprintf(stderr, message, values); + va_end( values ); +} diff --git a/src/mesa/drivers/dri/radeon/radeon_debug.h b/src/mesa/drivers/dri/radeon/radeon_debug.h new file mode 100644 index 00000000000..2a8302293b2 --- /dev/null +++ b/src/mesa/drivers/dri/radeon/radeon_debug.h @@ -0,0 +1,169 @@ +/* + * Copyright © 2009 Pauli Nieminen + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ +/* + * Authors: + * Pauli Nieminen <suokkos@gmail.com> + */ + +#ifndef RADEON_DEBUG_H_INCLUDED +#define RADEON_DEBUG_H_INCLUDED + +#include <stdlib.h> + +typedef enum radeon_debug_levels { + RADEON_CRITICAL = 0, /* Only errors */ + RADEON_IMPORTANT = 1, /* Important warnings and messages */ + RADEON_NORMAL = 2, /* Normal log messages usefull for debugging */ + RADEON_VERBOSE = 3, /* Extra details to debugging */ + RADEON_TRACE = 4 /* Log about everything that happens */ +} radeon_debug_level_t; + +/** + * Compile time option to change level of debugging compiled to dri driver. + * Selecting critical level is not recommended because perfromance gains are + * going to minimal but you will lose a lot of important warnings in case of + * errors. + */ +#ifndef RADEON_DEBUG_LEVEL +#define RADEON_DEBUG_LEVEL RADEON_VERBOSE +#endif + +typedef enum radeon_debug_types { + RADEON_TEXTURE = 0x00001, + RADEON_STATE = 0x00002, + RADEON_IOCTL = 0x00004, + RADEON_RENDER = 0x00008, + RADEON_SWRENDER = 0x00010, + RADEON_FALLBACKS = 0x00020, + RADEON_VFMT = 0x00040, + RADEON_SHADER = 0x00080, + RADEON_CS = 0x00100, + RADEON_DRI = 0x00200, + RADEON_DMA = 0x00400, + RADEON_SANITY = 0x00800, + RADEON_SYNC = 0x01000, + RADEON_PIXEL = 0x02000, + RADEON_MEMORY = 0x04000, + RADEON_VERTS = 0x08000, + RADEON_GENERAL = 0x10000 /* Used for errors and warnings */ +} radeon_debug_type_t; + +#define RADEON_MAX_INDENT 5 + +struct radeon_debug { + size_t indent_depth; + char indent[RADEON_MAX_INDENT]; +}; + +extern radeon_debug_type_t radeon_enabled_debug_types; + +/** + * Compabibility layer for old debug code + **/ +#define RADEON_DEBUG radeon_enabled_debug_types + +static inline int radeon_is_debug_enabled(const radeon_debug_type_t type, + const radeon_debug_level_t level) +{ + return RADEON_DEBUG_LEVEL >= level + && (type & radeon_enabled_debug_types); +} +/* + * define macro for gcc specific __attribute__ if using alternative compiler + */ +#ifndef __GNUC__ +#define __attribute__(x) /*empty*/ +#endif + + +extern void _radeon_print(const radeon_debug_type_t type, + const radeon_debug_level_t level, + const char* message, + ...) __attribute__((format(printf,3,4))); +/** + * Print out debug message if channel specified by type is enabled + * and compile time debugging level is at least as high as level parameter + */ +#define radeon_print(type, level, message, ...) do { \ + const radeon_debug_level_t _debug_level = (level); \ + const radeon_debug_type_t _debug_type = (type); \ + /* Compile out if level of message is too high */ \ + if (radeon_is_debug_enabled(type, level)) { \ + _radeon_print(_debug_type, _debug_level, \ + (message), ## __VA_ARGS__); \ + } \ +} while(0) + +/** + * printf style function for writing error messages. + */ +#define radeon_error(message, ...) do { \ + radeon_print(RADEON_GENERAL, RADEON_CRITICAL, \ + (message), ## __VA_ARGS__); \ +} while(0) + +/** + * printf style function for writing warnings. + */ +#define radeon_warning(message, ...) do { \ + radeon_print(RADEON_GENERAL, RADEON_IMPORTANT, \ + (message), ## __VA_ARGS__); \ +} while(0) + +extern void radeon_init_debug(void); +extern void _radeon_debug_add_indent(void); +extern void _radeon_debug_remove_indent(void); + +static inline void radeon_debug_add_indent(void) +{ + if (RADEON_DEBUG_LEVEL >= RADEON_VERBOSE) { + _radeon_debug_add_indent(); + } +} +static inline void radeon_debug_remove_indent(void) +{ + if (RADEON_DEBUG_LEVEL >= RADEON_VERBOSE) { + _radeon_debug_remove_indent(); + } +} + +/* From http://gcc. gnu.org/onlinedocs/gcc-3.2.3/gcc/Variadic-Macros.html . + I suppose we could inline this and use macro to fetch out __LINE__ and stuff in case we run into trouble + with other compilers ... GLUE! +*/ +#define WARN_ONCE(a, ...) { \ + static int warn##__LINE__=1; \ + if(warn##__LINE__){ \ + radeon_warning("*********************************WARN_ONCE*********************************\n"); \ + radeon_warning("File %s function %s line %d\n", \ + __FILE__, __FUNCTION__, __LINE__); \ + radeon_warning( (a), ## __VA_ARGS__);\ + radeon_warning("***************************************************************************\n"); \ + warn##__LINE__=0;\ + } \ + } + + +#endif diff --git a/src/mesa/drivers/dri/radeon/radeon_dma.c b/src/mesa/drivers/dri/radeon/radeon_dma.c new file mode 100644 index 00000000000..c9a32c808be --- /dev/null +++ b/src/mesa/drivers/dri/radeon/radeon_dma.c @@ -0,0 +1,476 @@ +/************************************************************************** + +Copyright (C) 2004 Nicolai Haehnle. +Copyright (C) The Weather Channel, Inc. 2002. All Rights Reserved. + +The Weather Channel (TM) funded Tungsten Graphics to develop the +initial release of the Radeon 8500 driver under the XFree86 license. +This notice must be preserved. + +All Rights Reserved. + +Permission is hereby granted, free of charge, to any person obtaining a +copy of this software and associated documentation files (the "Software"), +to deal in the Software without restriction, including without limitation +on the rights to use, copy, modify, merge, publish, distribute, sub +license, and/or sell copies of the Software, and to permit persons to whom +the Software is furnished to do so, subject to the following conditions: + +The above copyright notice and this permission notice (including the next +paragraph) shall be included in all copies or substantial portions of the +Software. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL +ATI, VA LINUX SYSTEMS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, +DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR +OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE +USE OR OTHER DEALINGS IN THE SOFTWARE. + +**************************************************************************/ + +#include <errno.h> +#include "radeon_common.h" +#include "main/simple_list.h" + +#if defined(USE_X86_ASM) +#define COPY_DWORDS( dst, src, nr ) \ +do { \ + int __tmp; \ + __asm__ __volatile__( "rep ; movsl" \ + : "=%c" (__tmp), "=D" (dst), "=S" (__tmp) \ + : "0" (nr), \ + "D" ((long)dst), \ + "S" ((long)src) ); \ +} while (0) +#else +#define COPY_DWORDS( dst, src, nr ) \ +do { \ + int j; \ + for ( j = 0 ; j < nr ; j++ ) \ + dst[j] = ((int *)src)[j]; \ + dst += nr; \ +} while (0) +#endif + +void radeonEmitVec4(uint32_t *out, const GLvoid * data, int stride, int count) +{ + int i; + + if (RADEON_DEBUG & RADEON_VERTS) + fprintf(stderr, "%s count %d stride %d out %p data %p\n", + __FUNCTION__, count, stride, (void *)out, (void *)data); + + if (stride == 4) + COPY_DWORDS(out, data, count); + else + for (i = 0; i < count; i++) { + out[0] = *(int *)data; + out++; + data += stride; + } +} + +void radeonEmitVec8(uint32_t *out, const GLvoid * data, int stride, int count) +{ + int i; + + if (RADEON_DEBUG & RADEON_VERTS) + fprintf(stderr, "%s count %d stride %d out %p data %p\n", + __FUNCTION__, count, stride, (void *)out, (void *)data); + + if (stride == 8) + COPY_DWORDS(out, data, count * 2); + else + for (i = 0; i < count; i++) { + out[0] = *(int *)data; + out[1] = *(int *)(data + 4); + out += 2; + data += stride; + } +} + +void radeonEmitVec12(uint32_t *out, const GLvoid * data, int stride, int count) +{ + int i; + + if (RADEON_DEBUG & RADEON_VERTS) + fprintf(stderr, "%s count %d stride %d out %p data %p\n", + __FUNCTION__, count, stride, (void *)out, (void *)data); + + if (stride == 12) { + COPY_DWORDS(out, data, count * 3); + } + else + for (i = 0; i < count; i++) { + out[0] = *(int *)data; + out[1] = *(int *)(data + 4); + out[2] = *(int *)(data + 8); + out += 3; + data += stride; + } +} + +void radeonEmitVec16(uint32_t *out, const GLvoid * data, int stride, int count) +{ + int i; + + if (RADEON_DEBUG & RADEON_VERTS) + fprintf(stderr, "%s count %d stride %d out %p data %p\n", + __FUNCTION__, count, stride, (void *)out, (void *)data); + + if (stride == 16) + COPY_DWORDS(out, data, count * 4); + else + for (i = 0; i < count; i++) { + out[0] = *(int *)data; + out[1] = *(int *)(data + 4); + out[2] = *(int *)(data + 8); + out[3] = *(int *)(data + 12); + out += 4; + data += stride; + } +} + +void rcommon_emit_vector(GLcontext * ctx, struct radeon_aos *aos, + const GLvoid * data, int size, int stride, int count) +{ + radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + uint32_t *out; + + if (stride == 0) { + radeonAllocDmaRegion(rmesa, &aos->bo, &aos->offset, size * 4, 32); + count = 1; + aos->stride = 0; + } else { + radeonAllocDmaRegion(rmesa, &aos->bo, &aos->offset, size * count * 4, 32); + aos->stride = size; + } + + aos->components = size; + aos->count = count; + + out = (uint32_t*)((char*)aos->bo->ptr + aos->offset); + switch (size) { + case 1: radeonEmitVec4(out, data, stride, count); break; + case 2: radeonEmitVec8(out, data, stride, count); break; + case 3: radeonEmitVec12(out, data, stride, count); break; + case 4: radeonEmitVec16(out, data, stride, count); break; + default: + assert(0); + break; + } +} + +void radeon_init_dma(radeonContextPtr rmesa) +{ + make_empty_list(&rmesa->dma.free); + make_empty_list(&rmesa->dma.wait); + make_empty_list(&rmesa->dma.reserved); + rmesa->dma.minimum_size = MAX_DMA_BUF_SZ; +} + +void radeonRefillCurrentDmaRegion(radeonContextPtr rmesa, int size) +{ + struct radeon_dma_bo *dma_bo = NULL; + /* we set minimum sizes to at least requested size + aligned to next 16 bytes. */ + if (size > rmesa->dma.minimum_size) + rmesa->dma.minimum_size = (size + 15) & (~15); + + radeon_print(RADEON_DMA, RADEON_NORMAL, "%s size %d minimum_size %d\n", + __FUNCTION__, size, rmesa->dma.minimum_size); + + + /* unmap old reserved bo */ + if (!is_empty_list(&rmesa->dma.reserved)) + radeon_bo_unmap(first_elem(&rmesa->dma.reserved)->bo); + + if (is_empty_list(&rmesa->dma.free) + || last_elem(&rmesa->dma.free)->bo->size < size) { + dma_bo = CALLOC_STRUCT(radeon_dma_bo); + assert(dma_bo); + +again_alloc: + dma_bo->bo = radeon_bo_open(rmesa->radeonScreen->bom, + 0, rmesa->dma.minimum_size, 4, + RADEON_GEM_DOMAIN_GTT, 0); + + if (!dma_bo->bo) { + rcommonFlushCmdBuf(rmesa, __FUNCTION__); + goto again_alloc; + } + insert_at_head(&rmesa->dma.reserved, dma_bo); + } else { + /* We push and pop buffers from end of list so we can keep + counter on unused buffers for later freeing them from + begin of list */ + dma_bo = last_elem(&rmesa->dma.free); + assert(dma_bo->bo->cref == 1); + remove_from_list(dma_bo); + insert_at_head(&rmesa->dma.reserved, dma_bo); + } + + rmesa->dma.current_used = 0; + rmesa->dma.current_vertexptr = 0; + + if (radeon_cs_space_check_with_bo(rmesa->cmdbuf.cs, + first_elem(&rmesa->dma.reserved)->bo, + RADEON_GEM_DOMAIN_GTT, 0)) + fprintf(stderr,"failure to revalidate BOs - badness\n"); + + if (is_empty_list(&rmesa->dma.reserved)) { + /* Cmd buff have been flushed in radeon_revalidate_bos */ + goto again_alloc; + } + + radeon_bo_map(first_elem(&rmesa->dma.reserved)->bo, 1); +} + +/* Allocates a region from rmesa->dma.current. If there isn't enough + * space in current, grab a new buffer (and discard what was left of current) + */ +void radeonAllocDmaRegion(radeonContextPtr rmesa, + struct radeon_bo **pbo, int *poffset, + int bytes, int alignment) +{ + if (RADEON_DEBUG & RADEON_IOCTL) + fprintf(stderr, "%s %d\n", __FUNCTION__, bytes); + + if (rmesa->dma.flush) + rmesa->dma.flush(rmesa->glCtx); + + assert(rmesa->dma.current_used == rmesa->dma.current_vertexptr); + + alignment--; + rmesa->dma.current_used = (rmesa->dma.current_used + alignment) & ~alignment; + + if (is_empty_list(&rmesa->dma.reserved) + || rmesa->dma.current_used + bytes > first_elem(&rmesa->dma.reserved)->bo->size) + radeonRefillCurrentDmaRegion(rmesa, bytes); + + *poffset = rmesa->dma.current_used; + *pbo = first_elem(&rmesa->dma.reserved)->bo; + radeon_bo_ref(*pbo); + + /* Always align to at least 16 bytes */ + rmesa->dma.current_used = (rmesa->dma.current_used + bytes + 15) & ~15; + rmesa->dma.current_vertexptr = rmesa->dma.current_used; + + assert(rmesa->dma.current_used <= first_elem(&rmesa->dma.reserved)->bo->size); +} + +void radeonFreeDmaRegions(radeonContextPtr rmesa) +{ + struct radeon_dma_bo *dma_bo; + struct radeon_dma_bo *temp; + if (RADEON_DEBUG & RADEON_DMA) + fprintf(stderr, "%s\n", __FUNCTION__); + + foreach_s(dma_bo, temp, &rmesa->dma.free) { + remove_from_list(dma_bo); + radeon_bo_unref(dma_bo->bo); + FREE(dma_bo); + } + + foreach_s(dma_bo, temp, &rmesa->dma.wait) { + remove_from_list(dma_bo); + radeon_bo_unref(dma_bo->bo); + FREE(dma_bo); + } + + foreach_s(dma_bo, temp, &rmesa->dma.reserved) { + remove_from_list(dma_bo); + radeon_bo_unmap(dma_bo->bo); + radeon_bo_unref(dma_bo->bo); + FREE(dma_bo); + } +} + +void radeonReturnDmaRegion(radeonContextPtr rmesa, int return_bytes) +{ + if (is_empty_list(&rmesa->dma.reserved)) + return; + + if (RADEON_DEBUG & RADEON_IOCTL) + fprintf(stderr, "%s %d\n", __FUNCTION__, return_bytes); + rmesa->dma.current_used -= return_bytes; + rmesa->dma.current_vertexptr = rmesa->dma.current_used; +} + +static int radeon_bo_is_idle(struct radeon_bo* bo) +{ + uint32_t domain; + int ret = radeon_bo_is_busy(bo, &domain); + if (ret == -EINVAL) { + WARN_ONCE("Your libdrm or kernel doesn't have support for busy query.\n" + "This may cause small performance drop for you.\n"); + } + return ret != -EBUSY; +} + +void radeonReleaseDmaRegions(radeonContextPtr rmesa) +{ + struct radeon_dma_bo *dma_bo; + struct radeon_dma_bo *temp; + const int expire_at = ++rmesa->dma.free.expire_counter + DMA_BO_FREE_TIME; + const int time = rmesa->dma.free.expire_counter; + + if (RADEON_DEBUG & RADEON_DMA) { + size_t free = 0, + wait = 0, + reserved = 0; + foreach(dma_bo, &rmesa->dma.free) + ++free; + + foreach(dma_bo, &rmesa->dma.wait) + ++wait; + + foreach(dma_bo, &rmesa->dma.reserved) + ++reserved; + + fprintf(stderr, "%s: free %zu, wait %zu, reserved %zu, minimum_size: %zu\n", + __FUNCTION__, free, wait, reserved, rmesa->dma.minimum_size); + } + + if (!rmesa->radeonScreen->driScreen->dri2.enabled) { + /* request updated cs processing information from kernel */ + legacy_track_pending(rmesa->radeonScreen->bom, 0); + } + /* move waiting bos to free list. + wait list provides gpu time to handle data before reuse */ + foreach_s(dma_bo, temp, &rmesa->dma.wait) { + if (dma_bo->expire_counter == time) { + WARN_ONCE("Leaking dma buffer object!\n"); + radeon_bo_unref(dma_bo->bo); + remove_from_list(dma_bo); + FREE(dma_bo); + continue; + } + /* free objects that are too small to be used because of large request */ + if (dma_bo->bo->size < rmesa->dma.minimum_size) { + radeon_bo_unref(dma_bo->bo); + remove_from_list(dma_bo); + FREE(dma_bo); + continue; + } + if (!radeon_bo_is_idle(dma_bo->bo)) + continue; + remove_from_list(dma_bo); + dma_bo->expire_counter = expire_at; + insert_at_tail(&rmesa->dma.free, dma_bo); + } + + /* unmap the last dma region */ + if (!is_empty_list(&rmesa->dma.reserved)) + radeon_bo_unmap(first_elem(&rmesa->dma.reserved)->bo); + /* move reserved to wait list */ + foreach_s(dma_bo, temp, &rmesa->dma.reserved) { + /* free objects that are too small to be used because of large request */ + if (dma_bo->bo->size < rmesa->dma.minimum_size) { + radeon_bo_unref(dma_bo->bo); + remove_from_list(dma_bo); + FREE(dma_bo); + continue; + } + remove_from_list(dma_bo); + dma_bo->expire_counter = expire_at; + insert_at_tail(&rmesa->dma.wait, dma_bo); + } + + /* free bos that have been unused for some time */ + foreach_s(dma_bo, temp, &rmesa->dma.free) { + if (dma_bo->expire_counter != time) + break; + remove_from_list(dma_bo); + radeon_bo_unref(dma_bo->bo); + FREE(dma_bo); + } + +} + + +/* Flush vertices in the current dma region. + */ +void rcommon_flush_last_swtcl_prim( GLcontext *ctx ) +{ + radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + struct radeon_dma *dma = &rmesa->dma; + + + if (RADEON_DEBUG & RADEON_IOCTL) + fprintf(stderr, "%s\n", __FUNCTION__); + dma->flush = NULL; + + if (!is_empty_list(&dma->reserved)) { + GLuint current_offset = dma->current_used; + + assert (dma->current_used + + rmesa->swtcl.numverts * rmesa->swtcl.vertex_size * 4 == + dma->current_vertexptr); + + if (dma->current_used != dma->current_vertexptr) { + dma->current_used = dma->current_vertexptr; + + rmesa->vtbl.swtcl_flush(ctx, current_offset); + } + rmesa->swtcl.numverts = 0; + } +} +/* Alloc space in the current dma region. + */ +void * +rcommonAllocDmaLowVerts( radeonContextPtr rmesa, int nverts, int vsize ) +{ + GLuint bytes = vsize * nverts; + void *head; + if (RADEON_DEBUG & RADEON_IOCTL) + fprintf(stderr, "%s\n", __FUNCTION__); + if(is_empty_list(&rmesa->dma.reserved) + ||rmesa->dma.current_vertexptr + bytes > first_elem(&rmesa->dma.reserved)->bo->size) { + if (rmesa->dma.flush) { + rmesa->dma.flush(rmesa->glCtx); + } + + radeonRefillCurrentDmaRegion(rmesa, bytes); + + return NULL; + } + + if (!rmesa->dma.flush) { + /* if cmdbuf flushed DMA restart */ + rmesa->glCtx->Driver.NeedFlush |= FLUSH_STORED_VERTICES; + rmesa->dma.flush = rcommon_flush_last_swtcl_prim; + } + + ASSERT( vsize == rmesa->swtcl.vertex_size * 4 ); + ASSERT( rmesa->dma.flush == rcommon_flush_last_swtcl_prim ); + ASSERT( rmesa->dma.current_used + + rmesa->swtcl.numverts * rmesa->swtcl.vertex_size * 4 == + rmesa->dma.current_vertexptr ); + + head = (first_elem(&rmesa->dma.reserved)->bo->ptr + rmesa->dma.current_vertexptr); + rmesa->dma.current_vertexptr += bytes; + rmesa->swtcl.numverts += nverts; + return head; +} + +void radeonReleaseArrays( GLcontext *ctx, GLuint newinputs ) +{ + radeonContextPtr radeon = RADEON_CONTEXT( ctx ); + int i; + if (RADEON_DEBUG & RADEON_IOCTL) + fprintf(stderr, "%s\n", __FUNCTION__); + + if (radeon->dma.flush) { + radeon->dma.flush(radeon->glCtx); + } + for (i = 0; i < radeon->tcl.aos_count; i++) { + if (radeon->tcl.aos[i].bo) { + radeon_bo_unref(radeon->tcl.aos[i].bo); + radeon->tcl.aos[i].bo = NULL; + + } + } +} diff --git a/src/mesa/drivers/dri/radeon/radeon_dma.h b/src/mesa/drivers/dri/radeon/radeon_dma.h new file mode 100644 index 00000000000..74e653fd189 --- /dev/null +++ b/src/mesa/drivers/dri/radeon/radeon_dma.h @@ -0,0 +1,58 @@ +/************************************************************************** + +Copyright (C) 2004 Nicolai Haehnle. +Copyright (C) The Weather Channel, Inc. 2002. All Rights Reserved. + +The Weather Channel (TM) funded Tungsten Graphics to develop the +initial release of the Radeon 8500 driver under the XFree86 license. +This notice must be preserved. + +All Rights Reserved. + +Permission is hereby granted, free of charge, to any person obtaining a +copy of this software and associated documentation files (the "Software"), +to deal in the Software without restriction, including without limitation +on the rights to use, copy, modify, merge, publish, distribute, sub +license, and/or sell copies of the Software, and to permit persons to whom +the Software is furnished to do so, subject to the following conditions: + +The above copyright notice and this permission notice (including the next +paragraph) shall be included in all copies or substantial portions of the +Software. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL +ATI, VA LINUX SYSTEMS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, +DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR +OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE +USE OR OTHER DEALINGS IN THE SOFTWARE. + +**************************************************************************/ + +#ifndef RADEON_DMA_H +#define RADEON_DMA_H + +void radeonEmitVec4(uint32_t *out, const GLvoid * data, int stride, int count); +void radeonEmitVec8(uint32_t *out, const GLvoid * data, int stride, int count); +void radeonEmitVec12(uint32_t *out, const GLvoid * data, int stride, int count); +void radeonEmitVec16(uint32_t *out, const GLvoid * data, int stride, int count); + +void rcommon_emit_vector(GLcontext * ctx, struct radeon_aos *aos, + const GLvoid * data, int size, int stride, int count); + +void radeonReturnDmaRegion(radeonContextPtr rmesa, int return_bytes); +void radeonRefillCurrentDmaRegion(radeonContextPtr rmesa, int size); +void radeon_init_dma(radeonContextPtr rmesa); +void radeonReturnDmaRegion(radeonContextPtr rmesa, int return_bytes); +void radeonAllocDmaRegion(radeonContextPtr rmesa, + struct radeon_bo **pbo, int *poffset, + int bytes, int alignment); +void radeonReleaseDmaRegions(radeonContextPtr rmesa); + +void rcommon_flush_last_swtcl_prim(GLcontext *ctx); + +void *rcommonAllocDmaLowVerts(radeonContextPtr rmesa, int nverts, int vsize); +void radeonFreeDmaRegions(radeonContextPtr rmesa); +void radeonReleaseArrays( GLcontext *ctx, GLuint newinputs ); +#endif diff --git a/src/mesa/drivers/dri/radeon/radeon_fbo.c b/src/mesa/drivers/dri/radeon/radeon_fbo.c new file mode 100644 index 00000000000..d83b166742c --- /dev/null +++ b/src/mesa/drivers/dri/radeon/radeon_fbo.c @@ -0,0 +1,599 @@ +/************************************************************************** + * + * Copyright 2008 Red Hat Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "main/imports.h" +#include "main/macros.h" +#include "main/mtypes.h" +#include "main/fbobject.h" +#include "main/framebuffer.h" +#include "main/renderbuffer.h" +#include "main/context.h" +#include "main/texformat.h" +#include "main/texrender.h" +#include "drivers/common/meta.h" + +#include "radeon_common.h" +#include "radeon_mipmap_tree.h" + +#define FILE_DEBUG_FLAG RADEON_TEXTURE +#define DBG(...) do { \ + if (RADEON_DEBUG & FILE_DEBUG_FLAG) \ + _mesa_printf(__VA_ARGS__); \ +} while(0) + +static struct gl_framebuffer * +radeon_new_framebuffer(GLcontext *ctx, GLuint name) +{ + return _mesa_new_framebuffer(ctx, name); +} + +static void +radeon_delete_renderbuffer(struct gl_renderbuffer *rb) +{ + struct radeon_renderbuffer *rrb = radeon_renderbuffer(rb); + + ASSERT(rrb); + + if (rrb && rrb->bo) { + radeon_bo_unref(rrb->bo); + } + _mesa_free(rrb); +} + +static void * +radeon_get_pointer(GLcontext *ctx, struct gl_renderbuffer *rb, + GLint x, GLint y) +{ + return NULL; +} + +/** + * Called via glRenderbufferStorageEXT() to set the format and allocate + * storage for a user-created renderbuffer. + */ +static GLboolean +radeon_alloc_renderbuffer_storage(GLcontext * ctx, struct gl_renderbuffer *rb, + GLenum internalFormat, + GLuint width, GLuint height) +{ + struct radeon_context *radeon = RADEON_CONTEXT(ctx); + struct radeon_renderbuffer *rrb = radeon_renderbuffer(rb); + GLboolean software_buffer = GL_FALSE; + int cpp; + + ASSERT(rb->Name != 0); + switch (internalFormat) { + case GL_R3_G3_B2: + case GL_RGB4: + case GL_RGB5: + rb->_ActualFormat = GL_RGB5; + rb->DataType = GL_UNSIGNED_BYTE; + rb->RedBits = 5; + rb->GreenBits = 6; + rb->BlueBits = 5; + cpp = 2; + break; + case GL_RGB: + case GL_RGB8: + case GL_RGB10: + case GL_RGB12: + case GL_RGB16: + rb->_ActualFormat = GL_RGB8; + rb->DataType = GL_UNSIGNED_BYTE; + rb->RedBits = 8; + rb->GreenBits = 8; + rb->BlueBits = 8; + rb->AlphaBits = 0; + cpp = 4; + break; + case GL_RGBA: + case GL_RGBA2: + case GL_RGBA4: + case GL_RGB5_A1: + case GL_RGBA8: + case GL_RGB10_A2: + case GL_RGBA12: + case GL_RGBA16: + rb->_ActualFormat = GL_RGBA8; + rb->DataType = GL_UNSIGNED_BYTE; + rb->RedBits = 8; + rb->GreenBits = 8; + rb->BlueBits = 8; + rb->AlphaBits = 8; + cpp = 4; + break; + case GL_STENCIL_INDEX: + case GL_STENCIL_INDEX1_EXT: + case GL_STENCIL_INDEX4_EXT: + case GL_STENCIL_INDEX8_EXT: + case GL_STENCIL_INDEX16_EXT: + /* alloc a depth+stencil buffer */ + rb->_ActualFormat = GL_DEPTH24_STENCIL8_EXT; + rb->DataType = GL_UNSIGNED_INT_24_8_EXT; + rb->StencilBits = 8; + cpp = 4; + break; + case GL_DEPTH_COMPONENT16: + rb->_ActualFormat = GL_DEPTH_COMPONENT16; + rb->DataType = GL_UNSIGNED_SHORT; + rb->DepthBits = 16; + cpp = 2; + break; + case GL_DEPTH_COMPONENT: + case GL_DEPTH_COMPONENT24: + case GL_DEPTH_COMPONENT32: + rb->_ActualFormat = GL_DEPTH_COMPONENT24; + rb->DataType = GL_UNSIGNED_INT; + rb->DepthBits = 24; + cpp = 4; + break; + case GL_DEPTH_STENCIL_EXT: + case GL_DEPTH24_STENCIL8_EXT: + rb->_ActualFormat = GL_DEPTH24_STENCIL8_EXT; + rb->DataType = GL_UNSIGNED_INT_24_8_EXT; + rb->DepthBits = 24; + rb->StencilBits = 8; + cpp = 4; + break; + default: + _mesa_problem(ctx, + "Unexpected format in intel_alloc_renderbuffer_storage"); + return GL_FALSE; + } + + if (ctx->Driver.Flush) + ctx->Driver.Flush(ctx); /* +r6/r7 */ + + if (rrb->bo) + radeon_bo_unref(rrb->bo); + + + if (software_buffer) { + return _mesa_soft_renderbuffer_storage(ctx, rb, internalFormat, + width, height); + } + else { + uint32_t size; + uint32_t pitch = ((cpp * width + 63) & ~63) / cpp; + + fprintf(stderr,"Allocating %d x %d radeon RBO (pitch %d)\n", width, + height, pitch); + + size = pitch * height * cpp; + rrb->pitch = pitch * cpp; + rrb->cpp = cpp; + rrb->bo = radeon_bo_open(radeon->radeonScreen->bom, + 0, + size, + 0, + RADEON_GEM_DOMAIN_VRAM, + 0); + rb->Width = width; + rb->Height = height; + return GL_TRUE; + } + +} + + +/** + * Called for each hardware renderbuffer when a _window_ is resized. + * Just update fields. + * Not used for user-created renderbuffers! + */ +static GLboolean +radeon_alloc_window_storage(GLcontext * ctx, struct gl_renderbuffer *rb, + GLenum internalFormat, GLuint width, GLuint height) +{ + ASSERT(rb->Name == 0); + rb->Width = width; + rb->Height = height; + rb->_ActualFormat = internalFormat; + + return GL_TRUE; +} + + +static void +radeon_resize_buffers(GLcontext *ctx, struct gl_framebuffer *fb, + GLuint width, GLuint height) +{ + struct radeon_framebuffer *radeon_fb = (struct radeon_framebuffer*)fb; + int i; + + _mesa_resize_framebuffer(ctx, fb, width, height); + + fb->Initialized = GL_TRUE; /* XXX remove someday */ + + if (fb->Name != 0) { + return; + } + + /* Make sure all window system renderbuffers are up to date */ + for (i = 0; i < 2; i++) { + struct gl_renderbuffer *rb = &radeon_fb->color_rb[i]->base; + + /* only resize if size is changing */ + if (rb && (rb->Width != width || rb->Height != height)) { + rb->AllocStorage(ctx, rb, rb->InternalFormat, width, height); + } + } +} + + +/** Dummy function for gl_renderbuffer::AllocStorage() */ +static GLboolean +radeon_nop_alloc_storage(GLcontext * ctx, struct gl_renderbuffer *rb, + GLenum internalFormat, GLuint width, GLuint height) +{ + _mesa_problem(ctx, "radeon_op_alloc_storage should never be called."); + return GL_FALSE; +} + +struct radeon_renderbuffer * +radeon_create_renderbuffer(GLenum format, __DRIdrawablePrivate *driDrawPriv) +{ + struct radeon_renderbuffer *rrb; + + rrb = CALLOC_STRUCT(radeon_renderbuffer); + if (!rrb) + return NULL; + + _mesa_init_renderbuffer(&rrb->base, 0); + rrb->base.ClassID = RADEON_RB_CLASS; + + /* XXX format junk */ + switch (format) { + case GL_RGB5: + rrb->base._ActualFormat = GL_RGB5; + rrb->base._BaseFormat = GL_RGBA; + rrb->base.RedBits = 5; + rrb->base.GreenBits = 6; + rrb->base.BlueBits = 5; + rrb->base.DataType = GL_UNSIGNED_BYTE; + break; + case GL_RGB8: + rrb->base._ActualFormat = GL_RGB8; + rrb->base._BaseFormat = GL_RGB; + rrb->base.RedBits = 8; + rrb->base.GreenBits = 8; + rrb->base.BlueBits = 8; + rrb->base.AlphaBits = 0; + rrb->base.DataType = GL_UNSIGNED_BYTE; + break; + case GL_RGBA8: + rrb->base._ActualFormat = GL_RGBA8; + rrb->base._BaseFormat = GL_RGBA; + rrb->base.RedBits = 8; + rrb->base.GreenBits = 8; + rrb->base.BlueBits = 8; + rrb->base.AlphaBits = 8; + rrb->base.DataType = GL_UNSIGNED_BYTE; + break; + case GL_STENCIL_INDEX8_EXT: + rrb->base._ActualFormat = GL_STENCIL_INDEX8_EXT; + rrb->base._BaseFormat = GL_STENCIL_INDEX; + rrb->base.StencilBits = 8; + rrb->base.DataType = GL_UNSIGNED_BYTE; + break; + case GL_DEPTH_COMPONENT16: + rrb->base._ActualFormat = GL_DEPTH_COMPONENT16; + rrb->base._BaseFormat = GL_DEPTH_COMPONENT; + rrb->base.DepthBits = 16; + rrb->base.DataType = GL_UNSIGNED_SHORT; + break; + case GL_DEPTH_COMPONENT24: + rrb->base._ActualFormat = GL_DEPTH_COMPONENT24; + rrb->base._BaseFormat = GL_DEPTH_COMPONENT; + rrb->base.DepthBits = 24; + rrb->base.DataType = GL_UNSIGNED_INT; + break; + case GL_DEPTH24_STENCIL8_EXT: + rrb->base._ActualFormat = GL_DEPTH24_STENCIL8_EXT; + rrb->base._BaseFormat = GL_DEPTH_STENCIL_EXT; + rrb->base.DepthBits = 24; + rrb->base.StencilBits = 8; + rrb->base.DataType = GL_UNSIGNED_INT_24_8_EXT; + break; + default: + fprintf(stderr, "%s: Unknown format 0x%04x\n", __FUNCTION__, format); + _mesa_delete_renderbuffer(&rrb->base); + return NULL; + } + + rrb->dPriv = driDrawPriv; + rrb->base.InternalFormat = format; + + rrb->base.Delete = radeon_delete_renderbuffer; + rrb->base.AllocStorage = radeon_alloc_window_storage; + rrb->base.GetPointer = radeon_get_pointer; + + rrb->bo = NULL; + return rrb; +} + +static struct gl_renderbuffer * +radeon_new_renderbuffer(GLcontext * ctx, GLuint name) +{ + struct radeon_renderbuffer *rrb; + + rrb = CALLOC_STRUCT(radeon_renderbuffer); + if (!rrb) + return NULL; + + _mesa_init_renderbuffer(&rrb->base, name); + rrb->base.ClassID = RADEON_RB_CLASS; + + rrb->base.Delete = radeon_delete_renderbuffer; + rrb->base.AllocStorage = radeon_alloc_renderbuffer_storage; + rrb->base.GetPointer = radeon_get_pointer; + + return &rrb->base; +} + +static void +radeon_bind_framebuffer(GLcontext * ctx, GLenum target, + struct gl_framebuffer *fb, struct gl_framebuffer *fbread) +{ + if (target == GL_FRAMEBUFFER_EXT || target == GL_DRAW_FRAMEBUFFER_EXT) { + radeon_draw_buffer(ctx, fb); + } + else { + /* don't need to do anything if target == GL_READ_FRAMEBUFFER_EXT */ + } +} + +static void +radeon_framebuffer_renderbuffer(GLcontext * ctx, + struct gl_framebuffer *fb, + GLenum attachment, struct gl_renderbuffer *rb) +{ + + if (ctx->Driver.Flush) + ctx->Driver.Flush(ctx); /* +r6/r7 */ + + _mesa_framebuffer_renderbuffer(ctx, fb, attachment, rb); + radeon_draw_buffer(ctx, fb); +} + + +static GLboolean +radeon_update_wrapper(GLcontext *ctx, struct radeon_renderbuffer *rrb, + struct gl_texture_image *texImage) +{ + int retry = 0; +restart: + if (texImage->TexFormat == &_mesa_texformat_argb8888) { + rrb->cpp = 4; + rrb->base._ActualFormat = GL_RGBA8; + rrb->base._BaseFormat = GL_RGBA; + rrb->base.DataType = GL_UNSIGNED_BYTE; + DBG("Render to RGBA8 texture OK\n"); + } + else if (texImage->TexFormat == &_mesa_texformat_rgb565) { + rrb->cpp = 2; + rrb->base._ActualFormat = GL_RGB5; + rrb->base._BaseFormat = GL_RGB; + rrb->base.DataType = GL_UNSIGNED_BYTE; + DBG("Render to RGB5 texture OK\n"); + } + else if (texImage->TexFormat == &_mesa_texformat_argb1555) { + rrb->cpp = 2; + rrb->base._ActualFormat = GL_RGB5_A1; + rrb->base._BaseFormat = GL_RGBA; + rrb->base.DataType = GL_UNSIGNED_BYTE; + DBG("Render to ARGB1555 texture OK\n"); + } + else if (texImage->TexFormat == &_mesa_texformat_argb4444) { + rrb->cpp = 2; + rrb->base._ActualFormat = GL_RGBA4; + rrb->base._BaseFormat = GL_RGBA; + rrb->base.DataType = GL_UNSIGNED_BYTE; + DBG("Render to ARGB1555 texture OK\n"); + } + else if (texImage->TexFormat == &_mesa_texformat_z16) { + rrb->cpp = 2; + rrb->base._ActualFormat = GL_DEPTH_COMPONENT16; + rrb->base._BaseFormat = GL_DEPTH_COMPONENT; + rrb->base.DataType = GL_UNSIGNED_SHORT; + DBG("Render to DEPTH16 texture OK\n"); + } + else if (texImage->TexFormat == &_mesa_texformat_s8_z24) { + rrb->cpp = 4; + rrb->base._ActualFormat = GL_DEPTH24_STENCIL8_EXT; + rrb->base._BaseFormat = GL_DEPTH_STENCIL_EXT; + rrb->base.DataType = GL_UNSIGNED_INT_24_8_EXT; + DBG("Render to DEPTH_STENCIL texture OK\n"); + } + else { + /* try redoing the FBO */ + if (retry == 1) { + DBG("Render to texture BAD FORMAT %d\n", + texImage->TexFormat->MesaFormat); + return GL_FALSE; + } + texImage->TexFormat = radeonChooseTextureFormat(ctx, texImage->InternalFormat, 0, + texImage->TexFormat->DataType, + 1); + + retry++; + goto restart; + } + + rrb->pitch = texImage->Width * rrb->cpp; + rrb->base.InternalFormat = rrb->base._ActualFormat; + rrb->base.Width = texImage->Width; + rrb->base.Height = texImage->Height; + rrb->base.RedBits = texImage->TexFormat->RedBits; + rrb->base.GreenBits = texImage->TexFormat->GreenBits; + rrb->base.BlueBits = texImage->TexFormat->BlueBits; + rrb->base.AlphaBits = texImage->TexFormat->AlphaBits; + rrb->base.DepthBits = texImage->TexFormat->DepthBits; + rrb->base.StencilBits = texImage->TexFormat->StencilBits; + + rrb->base.Delete = radeon_delete_renderbuffer; + rrb->base.AllocStorage = radeon_nop_alloc_storage; + + return GL_TRUE; +} + + +static struct radeon_renderbuffer * +radeon_wrap_texture(GLcontext * ctx, struct gl_texture_image *texImage) +{ + const GLuint name = ~0; /* not significant, but distinct for debugging */ + struct radeon_renderbuffer *rrb; + + /* make an radeon_renderbuffer to wrap the texture image */ + rrb = CALLOC_STRUCT(radeon_renderbuffer); + if (!rrb) { + _mesa_error(ctx, GL_OUT_OF_MEMORY, "glFramebufferTexture"); + return NULL; + } + + _mesa_init_renderbuffer(&rrb->base, name); + rrb->base.ClassID = RADEON_RB_CLASS; + + if (!radeon_update_wrapper(ctx, rrb, texImage)) { + _mesa_free(rrb); + return NULL; + } + + return rrb; + +} +static void +radeon_render_texture(GLcontext * ctx, + struct gl_framebuffer *fb, + struct gl_renderbuffer_attachment *att) +{ + struct gl_texture_image *newImage + = att->Texture->Image[att->CubeMapFace][att->TextureLevel]; + struct radeon_renderbuffer *rrb = radeon_renderbuffer(att->Renderbuffer); + radeon_texture_image *radeon_image; + GLuint imageOffset; + + (void) fb; + + ASSERT(newImage); + + if (newImage->Border != 0) { + /* Fallback on drawing to a texture with a border, which won't have a + * miptree. + */ + _mesa_reference_renderbuffer(&att->Renderbuffer, NULL); + _mesa_render_texture(ctx, fb, att); + return; + } + else if (!rrb) { + rrb = radeon_wrap_texture(ctx, newImage); + if (rrb) { + /* bind the wrapper to the attachment point */ + _mesa_reference_renderbuffer(&att->Renderbuffer, &rrb->base); + } + else { + /* fallback to software rendering */ + _mesa_render_texture(ctx, fb, att); + return; + } + } + + if (!radeon_update_wrapper(ctx, rrb, newImage)) { + _mesa_reference_renderbuffer(&att->Renderbuffer, NULL); + _mesa_render_texture(ctx, fb, att); + return; + } + + DBG("Begin render texture tid %x tex=%u w=%d h=%d refcount=%d\n", + _glthread_GetID(), + att->Texture->Name, newImage->Width, newImage->Height, + rrb->base.RefCount); + + /* point the renderbufer's region to the texture image region */ + radeon_image = (radeon_texture_image *)newImage; + if (rrb->bo != radeon_image->mt->bo) { + if (rrb->bo) + radeon_bo_unref(rrb->bo); + rrb->bo = radeon_image->mt->bo; + radeon_bo_ref(rrb->bo); + } + + /* compute offset of the particular 2D image within the texture region */ + imageOffset = radeon_miptree_image_offset(radeon_image->mt, + att->CubeMapFace, + att->TextureLevel); + + if (att->Texture->Target == GL_TEXTURE_3D) { + GLuint offsets[6]; + radeon_miptree_depth_offsets(radeon_image->mt, att->TextureLevel, + offsets); + imageOffset += offsets[att->Zoffset]; + } + + /* store that offset in the region */ + rrb->draw_offset = imageOffset; + + /* update drawing region, etc */ + radeon_draw_buffer(ctx, fb); +} + +static void +radeon_finish_render_texture(GLcontext * ctx, + struct gl_renderbuffer_attachment *att) +{ + +} +static void +radeon_validate_framebuffer(GLcontext *ctx, struct gl_framebuffer *fb) +{ +} + +void radeon_fbo_init(struct radeon_context *radeon) +{ + radeon->glCtx->Driver.NewFramebuffer = radeon_new_framebuffer; + radeon->glCtx->Driver.NewRenderbuffer = radeon_new_renderbuffer; + radeon->glCtx->Driver.BindFramebuffer = radeon_bind_framebuffer; + radeon->glCtx->Driver.FramebufferRenderbuffer = radeon_framebuffer_renderbuffer; + radeon->glCtx->Driver.RenderTexture = radeon_render_texture; + radeon->glCtx->Driver.FinishRenderTexture = radeon_finish_render_texture; + radeon->glCtx->Driver.ResizeBuffers = radeon_resize_buffers; + radeon->glCtx->Driver.ValidateFramebuffer = radeon_validate_framebuffer; + radeon->glCtx->Driver.BlitFramebuffer = _mesa_meta_blit_framebuffer; +} + + +void radeon_renderbuffer_set_bo(struct radeon_renderbuffer *rb, + struct radeon_bo *bo) +{ + struct radeon_bo *old; + old = rb->bo; + rb->bo = bo; + radeon_bo_ref(bo); + if (old) + radeon_bo_unref(old); +} diff --git a/src/mesa/drivers/dri/radeon/radeon_ioctl.c b/src/mesa/drivers/dri/radeon/radeon_ioctl.c index 09acf6b4f85..a0106d00fa2 100644 --- a/src/mesa/drivers/dri/radeon/radeon_ioctl.c +++ b/src/mesa/drivers/dri/radeon/radeon_ioctl.c @@ -35,7 +35,21 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */ #include <sched.h> -#include <errno.h> +#include <errno.h> + +#include "main/attrib.h" +#include "main/enable.h" +#include "main/blend.h" +#include "main/bufferobj.h" +#include "main/buffers.h" +#include "main/depth.h" +#include "main/shaders.h" +#include "main/texstate.h" +#include "main/varray.h" +#include "glapi/dispatch.h" +#include "swrast/swrast.h" +#include "main/stencil.h" +#include "main/matrix.h" #include "main/glheader.h" #include "main/imports.h" @@ -43,6 +57,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "swrast/swrast.h" #include "radeon_context.h" +#include "radeon_common.h" #include "radeon_state.h" #include "radeon_ioctl.h" #include "radeon_tcl.h" @@ -58,75 +73,6 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #define RADEON_IDLE_RETRY 16 -static void radeonWaitForIdle( radeonContextPtr rmesa ); -static int radeonFlushCmdBufLocked( radeonContextPtr rmesa, - const char * caller ); - -static void print_state_atom( struct radeon_state_atom *state ) -{ - int i; - - fprintf(stderr, "emit %s/%d\n", state->name, state->cmd_size); - - if (RADEON_DEBUG & DEBUG_VERBOSE) - for (i = 0 ; i < state->cmd_size ; i++) - fprintf(stderr, "\t%s[%d]: %x\n", state->name, i, state->cmd[i]); - -} - -static void radeonSaveHwState( radeonContextPtr rmesa ) -{ - struct radeon_state_atom *atom; - char * dest = rmesa->backup_store.cmd_buf; - - if (RADEON_DEBUG & DEBUG_STATE) - fprintf(stderr, "%s\n", __FUNCTION__); - - rmesa->backup_store.cmd_used = 0; - - foreach( atom, &rmesa->hw.atomlist ) { - if ( atom->check( rmesa->glCtx ) ) { - int size = atom->cmd_size * 4; - memcpy( dest, atom->cmd, size); - dest += size; - rmesa->backup_store.cmd_used += size; - if (RADEON_DEBUG & DEBUG_STATE) - print_state_atom( atom ); - } - } - - assert( rmesa->backup_store.cmd_used <= RADEON_CMD_BUF_SZ ); - if (RADEON_DEBUG & DEBUG_STATE) - fprintf(stderr, "Returning to radeonEmitState\n"); -} - -/* At this point we were in FlushCmdBufLocked but we had lost our context, so - * we need to unwire our current cmdbuf, hook the one with the saved state in - * it, flush it, and then put the current one back. This is so commands at the - * start of a cmdbuf can rely on the state being kept from the previous one. - */ -static void radeonBackUpAndEmitLostStateLocked( radeonContextPtr rmesa ) -{ - GLuint nr_released_bufs; - struct radeon_store saved_store; - - if (rmesa->backup_store.cmd_used == 0) - return; - - if (RADEON_DEBUG & DEBUG_STATE) - fprintf(stderr, "Emitting backup state on lost context\n"); - - rmesa->lost_context = GL_FALSE; - - nr_released_bufs = rmesa->dma.nr_released_bufs; - saved_store = rmesa->store; - rmesa->dma.nr_released_bufs = 0; - rmesa->store = rmesa->backup_store; - radeonFlushCmdBufLocked( rmesa, __FUNCTION__ ); - rmesa->dma.nr_released_bufs = nr_released_bufs; - rmesa->store = saved_store; -} - /* ============================================================= * Kernel command buffer handling */ @@ -134,965 +80,382 @@ static void radeonBackUpAndEmitLostStateLocked( radeonContextPtr rmesa ) /* The state atoms will be emitted in the order they appear in the atom list, * so this step is important. */ -void radeonSetUpAtomList( radeonContextPtr rmesa ) +void radeonSetUpAtomList( r100ContextPtr rmesa ) { - int i, mtu = rmesa->glCtx->Const.MaxTextureUnits; - - make_empty_list(&rmesa->hw.atomlist); - rmesa->hw.atomlist.name = "atom-list"; - - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.ctx); - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.set); - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.lin); - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.msk); - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.vpt); - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.tcl); - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.msc); + int i, mtu = rmesa->radeon.glCtx->Const.MaxTextureUnits; + + make_empty_list(&rmesa->radeon.hw.atomlist); + rmesa->radeon.hw.atomlist.name = "atom-list"; + + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.ctx); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.set); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.lin); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.msk); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.vpt); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.tcl); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.msc); for (i = 0; i < mtu; ++i) { - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.tex[i]); - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.txr[i]); - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.cube[i]); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.tex[i]); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.txr[i]); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.cube[i]); } - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.zbs); - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.mtl); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.zbs); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.mtl); for (i = 0; i < 3 + mtu; ++i) - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.mat[i]); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.mat[i]); for (i = 0; i < 8; ++i) - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.lit[i]); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.lit[i]); for (i = 0; i < 6; ++i) - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.ucp[i]); - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.eye); - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.grd); - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.fog); - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.glt); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.ucp[i]); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.eye); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.grd); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.fog); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.glt); } -void radeonEmitState( radeonContextPtr rmesa ) +static void radeonEmitScissor(r100ContextPtr rmesa) { - struct radeon_state_atom *atom; - char *dest; - - if (RADEON_DEBUG & (DEBUG_STATE|DEBUG_PRIMS)) - fprintf(stderr, "%s\n", __FUNCTION__); - - if (rmesa->save_on_next_emit) { - radeonSaveHwState(rmesa); - rmesa->save_on_next_emit = GL_FALSE; - } - - /* this code used to return here but now it emits zbs */ - - /* To avoid going across the entire set of states multiple times, just check - * for enough space for the case of emitting all state, and inline the - * radeonAllocCmdBuf code here without all the checks. - */ - radeonEnsureCmdBufSpace(rmesa, rmesa->hw.max_state_size); - dest = rmesa->store.cmd_buf + rmesa->store.cmd_used; - - /* We always always emit zbs, this is due to a bug found by keithw in - the hardware and rediscovered after Erics changes by me. - if you ever touch this code make sure you emit zbs otherwise - you get tcl lockups on at least M7/7500 class of chips - airlied */ - rmesa->hw.zbs.dirty=1; - - if (RADEON_DEBUG & DEBUG_STATE) { - foreach(atom, &rmesa->hw.atomlist) { - if (atom->dirty || rmesa->hw.all_dirty) { - if (atom->check(rmesa->glCtx)) - print_state_atom(atom); - else - fprintf(stderr, "skip state %s\n", atom->name); - } - } - } - - foreach(atom, &rmesa->hw.atomlist) { - if (rmesa->hw.all_dirty) - atom->dirty = GL_TRUE; - if (!(rmesa->radeonScreen->chip_flags & RADEON_CHIPSET_TCL) && - atom->is_tcl) - atom->dirty = GL_FALSE; - if (atom->dirty) { - if (atom->check(rmesa->glCtx)) { - int size = atom->cmd_size * 4; - memcpy(dest, atom->cmd, size); - dest += size; - rmesa->store.cmd_used += size; - atom->dirty = GL_FALSE; - } - } - } - - assert(rmesa->store.cmd_used <= RADEON_CMD_BUF_SZ); - - rmesa->hw.is_dirty = GL_FALSE; - rmesa->hw.all_dirty = GL_FALSE; + BATCH_LOCALS(&rmesa->radeon); + if (!rmesa->radeon.radeonScreen->kernel_mm) { + return; + } + if (rmesa->radeon.state.scissor.enabled) { + BEGIN_BATCH(6); + OUT_BATCH(CP_PACKET0(RADEON_PP_CNTL, 0)); + OUT_BATCH(rmesa->hw.ctx.cmd[CTX_PP_CNTL] | RADEON_SCISSOR_ENABLE); + OUT_BATCH(CP_PACKET0(RADEON_RE_TOP_LEFT, 0)); + OUT_BATCH((rmesa->radeon.state.scissor.rect.y1 << 16) | + rmesa->radeon.state.scissor.rect.x1); + OUT_BATCH(CP_PACKET0(RADEON_RE_WIDTH_HEIGHT, 0)); + OUT_BATCH(((rmesa->radeon.state.scissor.rect.y2) << 16) | + (rmesa->radeon.state.scissor.rect.x2)); + END_BATCH(); + } else { + BEGIN_BATCH(2); + OUT_BATCH(CP_PACKET0(RADEON_PP_CNTL, 0)); + OUT_BATCH(rmesa->hw.ctx.cmd[CTX_PP_CNTL] & ~RADEON_SCISSOR_ENABLE); + END_BATCH(); + } } /* Fire a section of the retained (indexed_verts) buffer as a regular - * primtive. + * primtive. */ -extern void radeonEmitVbufPrim( radeonContextPtr rmesa, +extern void radeonEmitVbufPrim( r100ContextPtr rmesa, GLuint vertex_format, GLuint primitive, GLuint vertex_nr ) { - drm_radeon_cmd_header_t *cmd; - + BATCH_LOCALS(&rmesa->radeon); assert(!(primitive & RADEON_CP_VC_CNTL_PRIM_WALK_IND)); - - radeonEmitState( rmesa ); - - if (RADEON_DEBUG & DEBUG_IOCTL) - fprintf(stderr, "%s cmd_used/4: %d\n", __FUNCTION__, - rmesa->store.cmd_used/4); - - cmd = (drm_radeon_cmd_header_t *)radeonAllocCmdBuf( rmesa, VBUF_BUFSZ, - __FUNCTION__ ); + + radeonEmitState(&rmesa->radeon); + radeonEmitScissor(rmesa); + #if RADEON_OLD_PACKETS - cmd[0].i = 0; - cmd[0].header.cmd_type = RADEON_CMD_PACKET3_CLIP; - cmd[1].i = RADEON_CP_PACKET3_3D_RNDR_GEN_INDX_PRIM | (3 << 16); - cmd[2].i = rmesa->ioctl.vertex_offset; - cmd[3].i = vertex_nr; - cmd[4].i = vertex_format; - cmd[5].i = (primitive | - RADEON_CP_VC_CNTL_PRIM_WALK_LIST | - RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA | - RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE | - (vertex_nr << RADEON_CP_VC_CNTL_NUM_SHIFT)); - - if (RADEON_DEBUG & DEBUG_PRIMS) - fprintf(stderr, "%s: header 0x%x offt 0x%x vfmt 0x%x vfcntl %x \n", - __FUNCTION__, - cmd[1].i, cmd[2].i, cmd[4].i, cmd[5].i); + BEGIN_BATCH(8); + OUT_BATCH_PACKET3_CLIP(RADEON_CP_PACKET3_3D_RNDR_GEN_INDX_PRIM, 3); + if (!rmesa->radeon.radeonScreen->kernel_mm) { + OUT_BATCH_RELOC(rmesa->ioctl.vertex_offset, rmesa->ioctl.bo, rmesa->ioctl.vertex_offset, RADEON_GEM_DOMAIN_GTT, 0, 0); + } else { + OUT_BATCH(rmesa->ioctl.vertex_offset); + } + + OUT_BATCH(vertex_nr); + OUT_BATCH(vertex_format); + OUT_BATCH(primitive | RADEON_CP_VC_CNTL_PRIM_WALK_LIST | + RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA | + RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE | + (vertex_nr << RADEON_CP_VC_CNTL_NUM_SHIFT)); + + if (rmesa->radeon.radeonScreen->kernel_mm) { + radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs, + rmesa->ioctl.bo, + RADEON_GEM_DOMAIN_GTT, + 0, 0); + } + + END_BATCH(); + #else - cmd[0].i = 0; - cmd[0].header.cmd_type = RADEON_CMD_PACKET3_CLIP; - cmd[1].i = RADEON_CP_PACKET3_3D_DRAW_VBUF | (1 << 16); - cmd[2].i = vertex_format; - cmd[3].i = (primitive | - RADEON_CP_VC_CNTL_PRIM_WALK_LIST | - RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA | - RADEON_CP_VC_CNTL_MAOS_ENABLE | - RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE | - (vertex_nr << RADEON_CP_VC_CNTL_NUM_SHIFT)); - - - if (RADEON_DEBUG & DEBUG_PRIMS) - fprintf(stderr, "%s: header 0x%x vfmt 0x%x vfcntl %x \n", - __FUNCTION__, - cmd[1].i, cmd[2].i, cmd[3].i); + BEGIN_BATCH(4); + OUT_BATCH_PACKET3_CLIP(RADEON_CP_PACKET3_3D_DRAW_VBUF, 1); + OUT_BATCH(vertex_format); + OUT_BATCH(primitive | + RADEON_CP_VC_CNTL_PRIM_WALK_LIST | + RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA | + RADEON_CP_VC_CNTL_MAOS_ENABLE | + RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE | + (vertex_nr << RADEON_CP_VC_CNTL_NUM_SHIFT)); + END_BATCH(); #endif } - -void radeonFlushElts( radeonContextPtr rmesa ) +void radeonFlushElts( GLcontext *ctx ) { - int *cmd = (int *)(rmesa->store.cmd_buf + rmesa->store.elts_start); - int dwords; -#if RADEON_OLD_PACKETS - int nr = (rmesa->store.cmd_used - (rmesa->store.elts_start + 24)) / 2; -#else - int nr = (rmesa->store.cmd_used - (rmesa->store.elts_start + 16)) / 2; -#endif + r100ContextPtr rmesa = R100_CONTEXT(ctx); + BATCH_LOCALS(&rmesa->radeon); + int nr; + uint32_t *cmd = (uint32_t *)(rmesa->radeon.cmdbuf.cs->packets + rmesa->tcl.elt_cmd_start); + int dwords = (rmesa->radeon.cmdbuf.cs->section_ndw - rmesa->radeon.cmdbuf.cs->section_cdw); - if (RADEON_DEBUG & DEBUG_IOCTL) + if (RADEON_DEBUG & RADEON_IOCTL) fprintf(stderr, "%s\n", __FUNCTION__); - assert( rmesa->dma.flush == radeonFlushElts ); - rmesa->dma.flush = NULL; + assert( rmesa->radeon.dma.flush == radeonFlushElts ); + rmesa->radeon.dma.flush = NULL; - /* Cope with odd number of elts: - */ - rmesa->store.cmd_used = (rmesa->store.cmd_used + 2) & ~2; - dwords = (rmesa->store.cmd_used - rmesa->store.elts_start) / 4; + nr = rmesa->tcl.elt_used; #if RADEON_OLD_PACKETS - cmd[1] |= (dwords - 3) << 16; + if (rmesa->radeon.radeonScreen->kernel_mm) { + dwords -= 2; + } +#endif + +#if RADEON_OLD_PACKETS + cmd[1] |= (dwords + 3) << 16; cmd[5] |= nr << RADEON_CP_VC_CNTL_NUM_SHIFT; #else - cmd[1] |= (dwords - 3) << 16; + cmd[1] |= (dwords + 2) << 16; cmd[3] |= nr << RADEON_CP_VC_CNTL_NUM_SHIFT; #endif - if (RADEON_DEBUG & DEBUG_SYNC) { + rmesa->radeon.cmdbuf.cs->cdw += dwords; + rmesa->radeon.cmdbuf.cs->section_cdw += dwords; + +#if RADEON_OLD_PACKETS + if (rmesa->radeon.radeonScreen->kernel_mm) { + radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs, + rmesa->ioctl.bo, + RADEON_GEM_DOMAIN_GTT, + 0, 0); + } +#endif + + END_BATCH(); + + if (RADEON_DEBUG & RADEON_SYNC) { fprintf(stderr, "%s: Syncing\n", __FUNCTION__); - radeonFinish( rmesa->glCtx ); + radeonFinish( rmesa->radeon.glCtx ); } -} +} -GLushort *radeonAllocEltsOpenEnded( radeonContextPtr rmesa, +GLushort *radeonAllocEltsOpenEnded( r100ContextPtr rmesa, GLuint vertex_format, GLuint primitive, GLuint min_nr ) { - drm_radeon_cmd_header_t *cmd; GLushort *retval; + int align_min_nr; + BATCH_LOCALS(&rmesa->radeon); - if (RADEON_DEBUG & DEBUG_IOCTL) - fprintf(stderr, "%s %d\n", __FUNCTION__, min_nr); + if (RADEON_DEBUG & RADEON_IOCTL) + fprintf(stderr, "%s %d prim %x\n", __FUNCTION__, min_nr, primitive); assert((primitive & RADEON_CP_VC_CNTL_PRIM_WALK_IND)); - - radeonEmitState( rmesa ); - - cmd = (drm_radeon_cmd_header_t *)radeonAllocCmdBuf( rmesa, - ELTS_BUFSZ(min_nr), - __FUNCTION__ ); + + radeonEmitState(&rmesa->radeon); + radeonEmitScissor(rmesa); + + rmesa->tcl.elt_cmd_start = rmesa->radeon.cmdbuf.cs->cdw; + + /* round up min_nr to align the state */ + align_min_nr = (min_nr + 1) & ~1; + #if RADEON_OLD_PACKETS - cmd[0].i = 0; - cmd[0].header.cmd_type = RADEON_CMD_PACKET3_CLIP; - cmd[1].i = RADEON_CP_PACKET3_3D_RNDR_GEN_INDX_PRIM; - cmd[2].i = rmesa->ioctl.vertex_offset; - cmd[3].i = 0xffff; - cmd[4].i = vertex_format; - cmd[5].i = (primitive | - RADEON_CP_VC_CNTL_PRIM_WALK_IND | - RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA | - RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE); - - retval = (GLushort *)(cmd+6); -#else - cmd[0].i = 0; - cmd[0].header.cmd_type = RADEON_CMD_PACKET3_CLIP; - cmd[1].i = RADEON_CP_PACKET3_3D_DRAW_INDX; - cmd[2].i = vertex_format; - cmd[3].i = (primitive | - RADEON_CP_VC_CNTL_PRIM_WALK_IND | - RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA | - RADEON_CP_VC_CNTL_MAOS_ENABLE | - RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE); - - retval = (GLushort *)(cmd+4); + BEGIN_BATCH_NO_AUTOSTATE(2+ELTS_BUFSZ(align_min_nr)/4); + OUT_BATCH_PACKET3_CLIP(RADEON_CP_PACKET3_3D_RNDR_GEN_INDX_PRIM, 0); + if (!rmesa->radeon.radeonScreen->kernel_mm) { + OUT_BATCH_RELOC(rmesa->ioctl.vertex_offset, rmesa->ioctl.bo, rmesa->ioctl.vertex_offset, RADEON_GEM_DOMAIN_GTT, 0, 0); + } else { + OUT_BATCH(rmesa->ioctl.vertex_offset); + } + OUT_BATCH(rmesa->ioctl.vertex_max); + OUT_BATCH(vertex_format); + OUT_BATCH(primitive | + RADEON_CP_VC_CNTL_PRIM_WALK_IND | + RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA | + RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE); +#else + BEGIN_BATCH_NO_AUTOSTATE(ELTS_BUFSZ(align_min_nr)/4); + OUT_BATCH_PACKET3_CLIP(RADEON_CP_PACKET3_DRAW_INDX, 0); + OUT_BATCH(vertex_format); + OUT_BATCH(primitive | + RADEON_CP_VC_CNTL_PRIM_WALK_IND | + RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA | + RADEON_CP_VC_CNTL_MAOS_ENABLE | + RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE); #endif - if (RADEON_DEBUG & DEBUG_PRIMS) - fprintf(stderr, "%s: header 0x%x vfmt 0x%x prim %x \n", - __FUNCTION__, - cmd[1].i, vertex_format, primitive); - assert(!rmesa->dma.flush); - rmesa->glCtx->Driver.NeedFlush |= FLUSH_STORED_VERTICES; - rmesa->dma.flush = radeonFlushElts; + rmesa->tcl.elt_cmd_offset = rmesa->radeon.cmdbuf.cs->cdw; + rmesa->tcl.elt_used = min_nr; - rmesa->store.elts_start = ((char *)cmd) - rmesa->store.cmd_buf; + retval = (GLushort *)(rmesa->radeon.cmdbuf.cs->packets + rmesa->tcl.elt_cmd_offset); - return retval; -} + if (RADEON_DEBUG & RADEON_RENDER) + fprintf(stderr, "%s: header prim %x \n", + __FUNCTION__, primitive); + assert(!rmesa->radeon.dma.flush); + rmesa->radeon.glCtx->Driver.NeedFlush |= FLUSH_STORED_VERTICES; + rmesa->radeon.dma.flush = radeonFlushElts; + return retval; +} -void radeonEmitVertexAOS( radeonContextPtr rmesa, +void radeonEmitVertexAOS( r100ContextPtr rmesa, GLuint vertex_size, + struct radeon_bo *bo, GLuint offset ) { #if RADEON_OLD_PACKETS - rmesa->ioctl.vertex_size = vertex_size; rmesa->ioctl.vertex_offset = offset; + rmesa->ioctl.bo = bo; #else - drm_radeon_cmd_header_t *cmd; + BATCH_LOCALS(&rmesa->radeon); - if (RADEON_DEBUG & (DEBUG_PRIMS|DEBUG_IOCTL)) + if (RADEON_DEBUG & (RADEON_PRIMS|DEBUG_IOCTL)) fprintf(stderr, "%s: vertex_size 0x%x offset 0x%x \n", __FUNCTION__, vertex_size, offset); - cmd = (drm_radeon_cmd_header_t *)radeonAllocCmdBuf( rmesa, VERT_AOS_BUFSZ, - __FUNCTION__ ); + BEGIN_BATCH(7); + OUT_BATCH_PACKET3(RADEON_CP_PACKET3_3D_LOAD_VBPNTR, 2); + OUT_BATCH(1); + OUT_BATCH(vertex_size | (vertex_size << 8)); + OUT_BATCH_RELOC(offset, bo, offset, RADEON_GEM_DOMAIN_GTT, 0, 0); + END_BATCH(); - cmd[0].i = 0; - cmd[0].header.cmd_type = RADEON_CMD_PACKET3; - cmd[1].i = RADEON_CP_PACKET3_3D_LOAD_VBPNTR | (2 << 16); - cmd[2].i = 1; - cmd[3].i = vertex_size | (vertex_size << 8); - cmd[4].i = offset; #endif } - -void radeonEmitAOS( radeonContextPtr rmesa, - struct radeon_dma_region **component, + +void radeonEmitAOS( r100ContextPtr rmesa, GLuint nr, GLuint offset ) { #if RADEON_OLD_PACKETS assert( nr == 1 ); - assert( component[0]->aos_size == component[0]->aos_stride ); - rmesa->ioctl.vertex_size = component[0]->aos_size; - rmesa->ioctl.vertex_offset = - (component[0]->aos_start + offset * component[0]->aos_stride * 4); + rmesa->ioctl.bo = rmesa->radeon.tcl.aos[0].bo; + rmesa->ioctl.vertex_offset = + (rmesa->radeon.tcl.aos[0].offset + offset * rmesa->radeon.tcl.aos[0].stride * 4); + rmesa->ioctl.vertex_max = rmesa->radeon.tcl.aos[0].count; #else - drm_radeon_cmd_header_t *cmd; - int sz = AOS_BUFSZ(nr); + BATCH_LOCALS(&rmesa->radeon); + uint32_t voffset; + // int sz = AOS_BUFSZ(nr); + int sz = 1 + (nr >> 1) * 3 + (nr & 1) * 2; int i; - int *tmp; - if (RADEON_DEBUG & DEBUG_IOCTL) + if (RADEON_DEBUG & RADEON_IOCTL) fprintf(stderr, "%s\n", __FUNCTION__); - - cmd = (drm_radeon_cmd_header_t *)radeonAllocCmdBuf( rmesa, sz, - __FUNCTION__ ); - cmd[0].i = 0; - cmd[0].header.cmd_type = RADEON_CMD_PACKET3; - cmd[1].i = RADEON_CP_PACKET3_3D_LOAD_VBPNTR | (((sz / sizeof(int))-3) << 16); - cmd[2].i = nr; - tmp = &cmd[0].i; - cmd += 3; - - for (i = 0 ; i < nr ; i++) { - if (i & 1) { - cmd[0].i |= ((component[i]->aos_stride << 24) | - (component[i]->aos_size << 16)); - cmd[2].i = (component[i]->aos_start + - offset * component[i]->aos_stride * 4); - cmd += 3; - } - else { - cmd[0].i = ((component[i]->aos_stride << 8) | - (component[i]->aos_size << 0)); - cmd[1].i = (component[i]->aos_start + - offset * component[i]->aos_stride * 4); + BEGIN_BATCH(sz+2+(nr * 2)); + OUT_BATCH_PACKET3(RADEON_CP_PACKET3_3D_LOAD_VBPNTR, sz - 1); + OUT_BATCH(nr); + + if (!rmesa->radeon.radeonScreen->kernel_mm) { + for (i = 0; i + 1 < nr; i += 2) { + OUT_BATCH((rmesa->radeon.tcl.aos[i].components << 0) | + (rmesa->radeon.tcl.aos[i].stride << 8) | + (rmesa->radeon.tcl.aos[i + 1].components << 16) | + (rmesa->radeon.tcl.aos[i + 1].stride << 24)); + + voffset = rmesa->radeon.tcl.aos[i + 0].offset + + offset * 4 * rmesa->radeon.tcl.aos[i + 0].stride; + OUT_BATCH_RELOC(voffset, + rmesa->radeon.tcl.aos[i].bo, + voffset, + RADEON_GEM_DOMAIN_GTT, + 0, 0); + voffset = rmesa->radeon.tcl.aos[i + 1].offset + + offset * 4 * rmesa->radeon.tcl.aos[i + 1].stride; + OUT_BATCH_RELOC(voffset, + rmesa->radeon.tcl.aos[i+1].bo, + voffset, + RADEON_GEM_DOMAIN_GTT, + 0, 0); } - } - if (RADEON_DEBUG & DEBUG_VERTS) { - fprintf(stderr, "%s:\n", __FUNCTION__); - for (i = 0 ; i < sz ; i++) - fprintf(stderr, " %d: %x\n", i, tmp[i]); - } -#endif -} - -/* using already shifted color_fmt! */ -void radeonEmitBlit( radeonContextPtr rmesa, /* FIXME: which drmMinor is required? */ - GLuint color_fmt, - GLuint src_pitch, - GLuint src_offset, - GLuint dst_pitch, - GLuint dst_offset, - GLint srcx, GLint srcy, - GLint dstx, GLint dsty, - GLuint w, GLuint h ) -{ - drm_radeon_cmd_header_t *cmd; - - if (RADEON_DEBUG & DEBUG_IOCTL) - fprintf(stderr, "%s src %x/%x %d,%d dst: %x/%x %d,%d sz: %dx%d\n", - __FUNCTION__, - src_pitch, src_offset, srcx, srcy, - dst_pitch, dst_offset, dstx, dsty, - w, h); - - assert( (src_pitch & 63) == 0 ); - assert( (dst_pitch & 63) == 0 ); - assert( (src_offset & 1023) == 0 ); - assert( (dst_offset & 1023) == 0 ); - assert( w < (1<<16) ); - assert( h < (1<<16) ); - - cmd = (drm_radeon_cmd_header_t *)radeonAllocCmdBuf( rmesa, 8 * sizeof(int), - __FUNCTION__ ); - - - cmd[0].i = 0; - cmd[0].header.cmd_type = RADEON_CMD_PACKET3; - cmd[1].i = RADEON_CP_PACKET3_CNTL_BITBLT_MULTI | (5 << 16); - cmd[2].i = (RADEON_GMC_SRC_PITCH_OFFSET_CNTL | - RADEON_GMC_DST_PITCH_OFFSET_CNTL | - RADEON_GMC_BRUSH_NONE | - color_fmt | - RADEON_GMC_SRC_DATATYPE_COLOR | - RADEON_ROP3_S | - RADEON_DP_SRC_SOURCE_MEMORY | - RADEON_GMC_CLR_CMP_CNTL_DIS | - RADEON_GMC_WR_MSK_DIS ); - - cmd[3].i = ((src_pitch/64)<<22) | (src_offset >> 10); - cmd[4].i = ((dst_pitch/64)<<22) | (dst_offset >> 10); - cmd[5].i = (srcx << 16) | srcy; - cmd[6].i = (dstx << 16) | dsty; /* dst */ - cmd[7].i = (w << 16) | h; -} - - -void radeonEmitWait( radeonContextPtr rmesa, GLuint flags ) -{ - drm_radeon_cmd_header_t *cmd; - - assert( !(flags & ~(RADEON_WAIT_2D|RADEON_WAIT_3D)) ); - - cmd = (drm_radeon_cmd_header_t *)radeonAllocCmdBuf( rmesa, 1 * sizeof(int), - __FUNCTION__ ); - cmd[0].i = 0; - cmd[0].wait.cmd_type = RADEON_CMD_WAIT; - cmd[0].wait.flags = flags; -} - - -static int radeonFlushCmdBufLocked( radeonContextPtr rmesa, - const char * caller ) -{ - int ret, i; - drm_radeon_cmd_buffer_t cmd; - - if (rmesa->lost_context) - radeonBackUpAndEmitLostStateLocked(rmesa); - - if (RADEON_DEBUG & DEBUG_IOCTL) { - fprintf(stderr, "%s from %s\n", __FUNCTION__, caller); - - if (RADEON_DEBUG & DEBUG_VERBOSE) - for (i = 0 ; i < rmesa->store.cmd_used ; i += 4 ) - fprintf(stderr, "%d: %x\n", i/4, - *(int *)(&rmesa->store.cmd_buf[i])); - } - - if (RADEON_DEBUG & DEBUG_DMA) - fprintf(stderr, "%s: Releasing %d buffers\n", __FUNCTION__, - rmesa->dma.nr_released_bufs); - - - if (RADEON_DEBUG & DEBUG_SANITY) { - if (rmesa->state.scissor.enabled) - ret = radeonSanityCmdBuffer( rmesa, - rmesa->state.scissor.numClipRects, - rmesa->state.scissor.pClipRects); - else - ret = radeonSanityCmdBuffer( rmesa, - rmesa->numClipRects, - rmesa->pClipRects); - if (ret) { - fprintf(stderr, "drmSanityCommandWrite: %d\n", ret); - goto out; + if (nr & 1) { + OUT_BATCH((rmesa->radeon.tcl.aos[nr - 1].components << 0) | + (rmesa->radeon.tcl.aos[nr - 1].stride << 8)); + voffset = rmesa->radeon.tcl.aos[nr - 1].offset + + offset * 4 * rmesa->radeon.tcl.aos[nr - 1].stride; + OUT_BATCH_RELOC(voffset, + rmesa->radeon.tcl.aos[nr - 1].bo, + voffset, + RADEON_GEM_DOMAIN_GTT, + 0, 0); } - } - - - cmd.bufsz = rmesa->store.cmd_used; - cmd.buf = rmesa->store.cmd_buf; - - if (rmesa->state.scissor.enabled) { - cmd.nbox = rmesa->state.scissor.numClipRects; - cmd.boxes = rmesa->state.scissor.pClipRects; } else { - cmd.nbox = rmesa->numClipRects; - cmd.boxes = rmesa->pClipRects; - } - - ret = drmCommandWrite( rmesa->dri.fd, - DRM_RADEON_CMDBUF, - &cmd, sizeof(cmd) ); - - if (ret) - fprintf(stderr, "drmCommandWrite: %d\n", ret); - - if (RADEON_DEBUG & DEBUG_SYNC) { - fprintf(stderr, "\nSyncing in %s\n\n", __FUNCTION__); - radeonWaitForIdleLocked( rmesa ); - } - - out: - rmesa->store.primnr = 0; - rmesa->store.statenr = 0; - rmesa->store.cmd_used = 0; - rmesa->dma.nr_released_bufs = 0; - rmesa->save_on_next_emit = 1; - - return ret; -} - - -/* Note: does not emit any commands to avoid recursion on - * radeonAllocCmdBuf. - */ -void radeonFlushCmdBuf( radeonContextPtr rmesa, const char *caller ) -{ - int ret; - - - LOCK_HARDWARE( rmesa ); - - ret = radeonFlushCmdBufLocked( rmesa, caller ); - - UNLOCK_HARDWARE( rmesa ); - - if (ret) { - fprintf(stderr, "drm_radeon_cmd_buffer_t: %d (exiting)\n", ret); - exit(ret); - } -} - -/* ============================================================= - * Hardware vertex buffer handling - */ - - -void radeonRefillCurrentDmaRegion( radeonContextPtr rmesa ) -{ - struct radeon_dma_buffer *dmabuf; - int fd = rmesa->dri.fd; - int index = 0; - int size = 0; - drmDMAReq dma; - int ret; - - if (RADEON_DEBUG & (DEBUG_IOCTL|DEBUG_DMA)) - fprintf(stderr, "%s\n", __FUNCTION__); - - if (rmesa->dma.flush) { - rmesa->dma.flush( rmesa ); - } - - if (rmesa->dma.current.buf) - radeonReleaseDmaRegion( rmesa, &rmesa->dma.current, __FUNCTION__ ); - - if (rmesa->dma.nr_released_bufs > 4) - radeonFlushCmdBuf( rmesa, __FUNCTION__ ); - - dma.context = rmesa->dri.hwContext; - dma.send_count = 0; - dma.send_list = NULL; - dma.send_sizes = NULL; - dma.flags = 0; - dma.request_count = 1; - dma.request_size = RADEON_BUFFER_SIZE; - dma.request_list = &index; - dma.request_sizes = &size; - dma.granted_count = 0; - - LOCK_HARDWARE(rmesa); /* no need to validate */ - - ret = drmDMA( fd, &dma ); - - if (ret != 0) { - /* Free some up this way? - */ - if (rmesa->dma.nr_released_bufs) { - radeonFlushCmdBufLocked( rmesa, __FUNCTION__ ); - } - - if (RADEON_DEBUG & DEBUG_DMA) - fprintf(stderr, "Waiting for buffers\n"); - - radeonWaitForIdleLocked( rmesa ); - ret = drmDMA( fd, &dma ); - - if ( ret != 0 ) { - UNLOCK_HARDWARE( rmesa ); - fprintf( stderr, "Error: Could not get dma buffer... exiting\n" ); - exit( -1 ); - } - } - - UNLOCK_HARDWARE(rmesa); - - if (RADEON_DEBUG & DEBUG_DMA) - fprintf(stderr, "Allocated buffer %d\n", index); - - dmabuf = CALLOC_STRUCT( radeon_dma_buffer ); - dmabuf->buf = &rmesa->radeonScreen->buffers->list[index]; - dmabuf->refcount = 1; - - rmesa->dma.current.buf = dmabuf; - rmesa->dma.current.address = dmabuf->buf->address; - rmesa->dma.current.end = dmabuf->buf->total; - rmesa->dma.current.start = 0; - rmesa->dma.current.ptr = 0; - - rmesa->c_vertexBuffers++; -} - -void radeonReleaseDmaRegion( radeonContextPtr rmesa, - struct radeon_dma_region *region, - const char *caller ) -{ - if (RADEON_DEBUG & DEBUG_IOCTL) - fprintf(stderr, "%s from %s\n", __FUNCTION__, caller); - - if (!region->buf) - return; - - if (rmesa->dma.flush) - rmesa->dma.flush( rmesa ); - - if (--region->buf->refcount == 0) { - drm_radeon_cmd_header_t *cmd; - - if (RADEON_DEBUG & (DEBUG_IOCTL|DEBUG_DMA)) - fprintf(stderr, "%s -- DISCARD BUF %d\n", __FUNCTION__, - region->buf->buf->idx); - - cmd = (drm_radeon_cmd_header_t *)radeonAllocCmdBuf( rmesa, sizeof(*cmd), - __FUNCTION__ ); - cmd->dma.cmd_type = RADEON_CMD_DMA_DISCARD; - cmd->dma.buf_idx = region->buf->buf->idx; - FREE(region->buf); - rmesa->dma.nr_released_bufs++; - } - - region->buf = NULL; - region->start = 0; -} - -/* Allocates a region from rmesa->dma.current. If there isn't enough - * space in current, grab a new buffer (and discard what was left of current) - */ -void radeonAllocDmaRegion( radeonContextPtr rmesa, - struct radeon_dma_region *region, - int bytes, - int alignment ) -{ - if (RADEON_DEBUG & DEBUG_IOCTL) - fprintf(stderr, "%s %d\n", __FUNCTION__, bytes); - - if (rmesa->dma.flush) - rmesa->dma.flush( rmesa ); - - if (region->buf) - radeonReleaseDmaRegion( rmesa, region, __FUNCTION__ ); - - alignment--; - rmesa->dma.current.start = rmesa->dma.current.ptr = - (rmesa->dma.current.ptr + alignment) & ~alignment; - - if ( rmesa->dma.current.ptr + bytes > rmesa->dma.current.end ) - radeonRefillCurrentDmaRegion( rmesa ); - - region->start = rmesa->dma.current.start; - region->ptr = rmesa->dma.current.start; - region->end = rmesa->dma.current.start + bytes; - region->address = rmesa->dma.current.address; - region->buf = rmesa->dma.current.buf; - region->buf->refcount++; - - rmesa->dma.current.ptr += bytes; /* bug - if alignment > 7 */ - rmesa->dma.current.start = - rmesa->dma.current.ptr = (rmesa->dma.current.ptr + 0x7) & ~0x7; -} - -/* ================================================================ - * SwapBuffers with client-side throttling - */ - -static uint32_t radeonGetLastFrame (radeonContextPtr rmesa) -{ - drm_radeon_getparam_t gp; - int ret; - uint32_t frame; - - gp.param = RADEON_PARAM_LAST_FRAME; - gp.value = (int *)&frame; - ret = drmCommandWriteRead( rmesa->dri.fd, DRM_RADEON_GETPARAM, - &gp, sizeof(gp) ); - - if ( ret ) { - fprintf( stderr, "%s: drm_radeon_getparam_t: %d\n", __FUNCTION__, ret ); - exit(1); - } - - return frame; -} - -static void radeonEmitIrqLocked( radeonContextPtr rmesa ) -{ - drm_radeon_irq_emit_t ie; - int ret; - - ie.irq_seq = &rmesa->iw.irq_seq; - ret = drmCommandWriteRead( rmesa->dri.fd, DRM_RADEON_IRQ_EMIT, - &ie, sizeof(ie) ); - if ( ret ) { - fprintf( stderr, "%s: drm_radeon_irq_emit_t: %d\n", __FUNCTION__, ret ); - exit(1); - } -} - - -static void radeonWaitIrq( radeonContextPtr rmesa ) -{ - int ret; - - do { - ret = drmCommandWrite( rmesa->dri.fd, DRM_RADEON_IRQ_WAIT, - &rmesa->iw, sizeof(rmesa->iw) ); - } while (ret && (errno == EINTR || errno == EBUSY)); - - if ( ret ) { - fprintf( stderr, "%s: drmRadeonIrqWait: %d\n", __FUNCTION__, ret ); - exit(1); - } -} - - -static void radeonWaitForFrameCompletion( radeonContextPtr rmesa ) -{ - drm_radeon_sarea_t *sarea = rmesa->sarea; - - if (rmesa->do_irqs) { - if (radeonGetLastFrame(rmesa) < sarea->last_frame) { - if (!rmesa->irqsEmitted) { - while (radeonGetLastFrame (rmesa) < sarea->last_frame) - ; - } - else { - UNLOCK_HARDWARE( rmesa ); - radeonWaitIrq( rmesa ); - LOCK_HARDWARE( rmesa ); - } - rmesa->irqsEmitted = 10; + for (i = 0; i + 1 < nr; i += 2) { + OUT_BATCH((rmesa->radeon.tcl.aos[i].components << 0) | + (rmesa->radeon.tcl.aos[i].stride << 8) | + (rmesa->radeon.tcl.aos[i + 1].components << 16) | + (rmesa->radeon.tcl.aos[i + 1].stride << 24)); + + voffset = rmesa->radeon.tcl.aos[i + 0].offset + + offset * 4 * rmesa->radeon.tcl.aos[i + 0].stride; + OUT_BATCH(voffset); + voffset = rmesa->radeon.tcl.aos[i + 1].offset + + offset * 4 * rmesa->radeon.tcl.aos[i + 1].stride; + OUT_BATCH(voffset); } - if (rmesa->irqsEmitted) { - radeonEmitIrqLocked( rmesa ); - rmesa->irqsEmitted--; + if (nr & 1) { + OUT_BATCH((rmesa->radeon.tcl.aos[nr - 1].components << 0) | + (rmesa->radeon.tcl.aos[nr - 1].stride << 8)); + voffset = rmesa->radeon.tcl.aos[nr - 1].offset + + offset * 4 * rmesa->radeon.tcl.aos[nr - 1].stride; + OUT_BATCH(voffset); } - } - else { - while (radeonGetLastFrame (rmesa) < sarea->last_frame) { - UNLOCK_HARDWARE( rmesa ); - if (rmesa->do_usleeps) - DO_USLEEP( 1 ); - LOCK_HARDWARE( rmesa ); + for (i = 0; i + 1 < nr; i += 2) { + voffset = rmesa->radeon.tcl.aos[i + 0].offset + + offset * 4 * rmesa->radeon.tcl.aos[i + 0].stride; + radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs, + rmesa->radeon.tcl.aos[i+0].bo, + RADEON_GEM_DOMAIN_GTT, + 0, 0); + voffset = rmesa->radeon.tcl.aos[i + 1].offset + + offset * 4 * rmesa->radeon.tcl.aos[i + 1].stride; + radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs, + rmesa->radeon.tcl.aos[i+1].bo, + RADEON_GEM_DOMAIN_GTT, + 0, 0); } - } -} - -/* Copy the back color buffer to the front color buffer. - */ -void radeonCopyBuffer( __DRIdrawablePrivate *dPriv, - const drm_clip_rect_t *rect) -{ - radeonContextPtr rmesa; - GLint nbox, i, ret; - GLboolean missed_target; - int64_t ust; - __DRIscreenPrivate *psp; - - assert(dPriv); - assert(dPriv->driContextPriv); - assert(dPriv->driContextPriv->driverPrivate); - - rmesa = (radeonContextPtr) dPriv->driContextPriv->driverPrivate; - - if ( RADEON_DEBUG & DEBUG_IOCTL ) { - fprintf( stderr, "\n%s( %p )\n\n", __FUNCTION__, (void *) rmesa->glCtx ); - } - - RADEON_FIREVERTICES( rmesa ); - LOCK_HARDWARE( rmesa ); - - /* Throttle the frame rate -- only allow one pending swap buffers - * request at a time. - */ - radeonWaitForFrameCompletion( rmesa ); - if (!rect) - { - UNLOCK_HARDWARE( rmesa ); - driWaitForVBlank( dPriv, & missed_target ); - LOCK_HARDWARE( rmesa ); - } - - nbox = dPriv->numClipRects; /* must be in locked region */ - - for ( i = 0 ; i < nbox ; ) { - GLint nr = MIN2( i + RADEON_NR_SAREA_CLIPRECTS , nbox ); - drm_clip_rect_t *box = dPriv->pClipRects; - drm_clip_rect_t *b = rmesa->sarea->boxes; - GLint n = 0; - - for ( ; i < nr ; i++ ) { - - *b = box[i]; - - if (rect) - { - if (rect->x1 > b->x1) - b->x1 = rect->x1; - if (rect->y1 > b->y1) - b->y1 = rect->y1; - if (rect->x2 < b->x2) - b->x2 = rect->x2; - if (rect->y2 < b->y2) - b->y2 = rect->y2; - - if (b->x1 >= b->x2 || b->y1 >= b->y2) - continue; - } - - b++; - n++; - } - rmesa->sarea->nbox = n; - - if (!n) - continue; - - ret = drmCommandNone( rmesa->dri.fd, DRM_RADEON_SWAP ); - - if ( ret ) { - fprintf( stderr, "DRM_RADEON_SWAP_BUFFERS: return = %d\n", ret ); - UNLOCK_HARDWARE( rmesa ); - exit( 1 ); + if (nr & 1) { + voffset = rmesa->radeon.tcl.aos[nr - 1].offset + + offset * 4 * rmesa->radeon.tcl.aos[nr - 1].stride; + radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs, + rmesa->radeon.tcl.aos[nr-1].bo, + RADEON_GEM_DOMAIN_GTT, + 0, 0); } } + END_BATCH(); - UNLOCK_HARDWARE( rmesa ); - if (!rect) - { - psp = dPriv->driScreenPriv; - rmesa->swap_count++; - (*psp->systemTime->getUST)( & ust ); - if ( missed_target ) { - rmesa->swap_missed_count++; - rmesa->swap_missed_ust = ust - rmesa->swap_ust; - } - - rmesa->swap_ust = ust; - rmesa->hw.all_dirty = GL_TRUE; - } -} - -void radeonPageFlip( __DRIdrawablePrivate *dPriv ) -{ - radeonContextPtr rmesa; - GLint ret; - GLboolean missed_target; - __DRIscreenPrivate *psp; - - assert(dPriv); - assert(dPriv->driContextPriv); - assert(dPriv->driContextPriv->driverPrivate); - - rmesa = (radeonContextPtr) dPriv->driContextPriv->driverPrivate; - psp = dPriv->driScreenPriv; - - if ( RADEON_DEBUG & DEBUG_IOCTL ) { - fprintf(stderr, "%s: pfCurrentPage: %d\n", __FUNCTION__, - rmesa->sarea->pfCurrentPage); - } - - RADEON_FIREVERTICES( rmesa ); - LOCK_HARDWARE( rmesa ); - - /* Need to do this for the perf box placement: - */ - if (dPriv->numClipRects) - { - drm_clip_rect_t *box = dPriv->pClipRects; - drm_clip_rect_t *b = rmesa->sarea->boxes; - b[0] = box[0]; - rmesa->sarea->nbox = 1; - } - - /* Throttle the frame rate -- only allow a few pending swap buffers - * request at a time. - */ - radeonWaitForFrameCompletion( rmesa ); - UNLOCK_HARDWARE( rmesa ); - driWaitForVBlank( dPriv, & missed_target ); - if ( missed_target ) { - rmesa->swap_missed_count++; - (void) (*psp->systemTime->getUST)( & rmesa->swap_missed_ust ); - } - LOCK_HARDWARE( rmesa ); - - ret = drmCommandNone( rmesa->dri.fd, DRM_RADEON_FLIP ); - - UNLOCK_HARDWARE( rmesa ); - - if ( ret ) { - fprintf( stderr, "DRM_RADEON_FLIP: return = %d\n", ret ); - exit( 1 ); - } - - rmesa->swap_count++; - (void) (*psp->systemTime->getUST)( & rmesa->swap_ust ); - - /* Get ready for drawing next frame. Update the renderbuffers' - * flippedOffset/Pitch fields so we draw into the right place. - */ - driFlipRenderbuffers(rmesa->glCtx->WinSysDrawBuffer, - rmesa->sarea->pfCurrentPage); - - radeonUpdateDrawBuffer(rmesa->glCtx); +#endif } - /* ================================================================ * Buffer clear */ #define RADEON_MAX_CLEARS 256 -static void radeonClear( GLcontext *ctx, GLbitfield mask ) +static void radeonKernelClear(GLcontext *ctx, GLuint flags) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - __DRIdrawablePrivate *dPriv = rmesa->dri.drawable; - drm_radeon_sarea_t *sarea = rmesa->sarea; + r100ContextPtr rmesa = R100_CONTEXT(ctx); + __DRIdrawablePrivate *dPriv = radeon_get_drawable(&rmesa->radeon); + drm_radeon_sarea_t *sarea = rmesa->radeon.sarea; uint32_t clear; - GLuint flags = 0; - GLuint color_mask = 0; GLint ret, i; GLint cx, cy, cw, ch; - if ( RADEON_DEBUG & DEBUG_IOCTL ) { - fprintf( stderr, "radeonClear\n"); - } - - { - LOCK_HARDWARE( rmesa ); - UNLOCK_HARDWARE( rmesa ); - if ( dPriv->numClipRects == 0 ) - return; - } - - radeonFlush( ctx ); - - if ( mask & BUFFER_BIT_FRONT_LEFT ) { - flags |= RADEON_FRONT; - color_mask = rmesa->hw.msk.cmd[MSK_RB3D_PLANEMASK]; - mask &= ~BUFFER_BIT_FRONT_LEFT; - } - - if ( mask & BUFFER_BIT_BACK_LEFT ) { - flags |= RADEON_BACK; - color_mask = rmesa->hw.msk.cmd[MSK_RB3D_PLANEMASK]; - mask &= ~BUFFER_BIT_BACK_LEFT; - } - - if ( mask & BUFFER_BIT_DEPTH ) { - flags |= RADEON_DEPTH; - mask &= ~BUFFER_BIT_DEPTH; - } - - if ( (mask & BUFFER_BIT_STENCIL) && rmesa->state.stencil.hwBuffer ) { - flags |= RADEON_STENCIL; - mask &= ~BUFFER_BIT_STENCIL; - } - - if ( mask ) { - if (RADEON_DEBUG & DEBUG_FALLBACKS) - fprintf(stderr, "%s: swrast clear, mask: %x\n", __FUNCTION__, mask); - _swrast_Clear( ctx, mask ); - } - - if ( !flags ) - return; - - if (rmesa->using_hyperz) { - flags |= RADEON_USE_COMP_ZBUF; -/* if (rmesa->radeonScreen->chipset & RADEON_CHIPSET_TCL) - flags |= RADEON_USE_HIERZ; */ - if (!(rmesa->state.stencil.hwBuffer) || - ((flags & RADEON_DEPTH) && (flags & RADEON_STENCIL) && - ((rmesa->state.stencil.clear & RADEON_STENCIL_WRITE_MASK) == RADEON_STENCIL_WRITE_MASK))) { - flags |= RADEON_CLEAR_FASTZ; - } - } - - LOCK_HARDWARE( rmesa ); + LOCK_HARDWARE( &rmesa->radeon ); /* compute region after locking: */ cx = ctx->DrawBuffer->_Xmin; @@ -1112,7 +475,7 @@ static void radeonClear( GLcontext *ctx, GLbitfield mask ) gp.param = RADEON_PARAM_LAST_CLEAR; gp.value = (int *)&clear; - ret = drmCommandWriteRead( rmesa->dri.fd, + ret = drmCommandWriteRead( rmesa->radeon.dri.fd, DRM_RADEON_GETPARAM, &gp, sizeof(gp) ); if ( ret ) { @@ -1124,20 +487,20 @@ static void radeonClear( GLcontext *ctx, GLbitfield mask ) break; } - if ( rmesa->do_usleeps ) { - UNLOCK_HARDWARE( rmesa ); + if ( rmesa->radeon.do_usleeps ) { + UNLOCK_HARDWARE( &rmesa->radeon ); DO_USLEEP( 1 ); - LOCK_HARDWARE( rmesa ); + LOCK_HARDWARE( &rmesa->radeon ); } } /* Send current state to the hardware */ - radeonFlushCmdBufLocked( rmesa, __FUNCTION__ ); + rcommonFlushCmdBufLocked( &rmesa->radeon, __FUNCTION__ ); for ( i = 0 ; i < dPriv->numClipRects ; ) { GLint nr = MIN2( i + RADEON_NR_SAREA_CLIPRECTS, dPriv->numClipRects ); drm_clip_rect_t *box = dPriv->pClipRects; - drm_clip_rect_t *b = rmesa->sarea->boxes; + drm_clip_rect_t *b = rmesa->radeon.sarea->boxes; drm_radeon_clear_t clear; drm_radeon_clear_rect_t depth_boxes[RADEON_NR_SAREA_CLIPRECTS]; GLint n = 0; @@ -1172,105 +535,107 @@ static void radeonClear( GLcontext *ctx, GLbitfield mask ) } } - rmesa->sarea->nbox = n; + rmesa->radeon.sarea->nbox = n; clear.flags = flags; - clear.clear_color = rmesa->state.color.clear; - clear.clear_depth = rmesa->state.depth.clear; + clear.clear_color = rmesa->radeon.state.color.clear; + clear.clear_depth = rmesa->radeon.state.depth.clear; clear.color_mask = rmesa->hw.msk.cmd[MSK_RB3D_PLANEMASK]; - clear.depth_mask = rmesa->state.stencil.clear; + clear.depth_mask = rmesa->radeon.state.stencil.clear; clear.depth_boxes = depth_boxes; n--; - b = rmesa->sarea->boxes; + b = rmesa->radeon.sarea->boxes; for ( ; n >= 0 ; n-- ) { depth_boxes[n].f[CLEAR_X1] = (float)b[n].x1; depth_boxes[n].f[CLEAR_Y1] = (float)b[n].y1; depth_boxes[n].f[CLEAR_X2] = (float)b[n].x2; depth_boxes[n].f[CLEAR_Y2] = (float)b[n].y2; - depth_boxes[n].f[CLEAR_DEPTH] = - (float)rmesa->state.depth.clear; + depth_boxes[n].f[CLEAR_DEPTH] = + (float)rmesa->radeon.state.depth.clear; } - ret = drmCommandWrite( rmesa->dri.fd, DRM_RADEON_CLEAR, + ret = drmCommandWrite( rmesa->radeon.dri.fd, DRM_RADEON_CLEAR, &clear, sizeof(drm_radeon_clear_t)); if ( ret ) { - UNLOCK_HARDWARE( rmesa ); + UNLOCK_HARDWARE( &rmesa->radeon ); fprintf( stderr, "DRM_RADEON_CLEAR: return = %d\n", ret ); exit( 1 ); } } - - UNLOCK_HARDWARE( rmesa ); - rmesa->hw.all_dirty = GL_TRUE; + UNLOCK_HARDWARE( &rmesa->radeon ); } - -void radeonWaitForIdleLocked( radeonContextPtr rmesa ) +static void radeonClear( GLcontext *ctx, GLbitfield mask ) { - int fd = rmesa->dri.fd; - int to = 0; - int ret, i = 0; - - rmesa->c_drawWaits++; - - do { - do { - ret = drmCommandNone( fd, DRM_RADEON_CP_IDLE); - } while ( ret && errno == EBUSY && i++ < RADEON_IDLE_RETRY ); - } while ( ( ret == -EBUSY ) && ( to++ < RADEON_TIMEOUT ) ); - - if ( ret < 0 ) { - UNLOCK_HARDWARE( rmesa ); - fprintf( stderr, "Error: Radeon timed out... exiting\n" ); - exit( -1 ); - } -} + r100ContextPtr rmesa = R100_CONTEXT(ctx); + __DRIdrawablePrivate *dPriv = radeon_get_drawable(&rmesa->radeon); + GLuint flags = 0; + GLuint color_mask = 0; + GLuint orig_mask = mask; + if ( RADEON_DEBUG & RADEON_IOCTL ) { + fprintf( stderr, "radeonClear\n"); + } -static void radeonWaitForIdle( radeonContextPtr rmesa ) -{ - LOCK_HARDWARE(rmesa); - radeonWaitForIdleLocked( rmesa ); - UNLOCK_HARDWARE(rmesa); -} + { + LOCK_HARDWARE( &rmesa->radeon ); + UNLOCK_HARDWARE( &rmesa->radeon ); + if ( dPriv->numClipRects == 0 ) + return; + } + radeon_firevertices(&rmesa->radeon); -void radeonFlush( GLcontext *ctx ) -{ - radeonContextPtr rmesa = RADEON_CONTEXT( ctx ); + if ( mask & BUFFER_BIT_FRONT_LEFT ) { + flags |= RADEON_FRONT; + color_mask = rmesa->hw.msk.cmd[MSK_RB3D_PLANEMASK]; + mask &= ~BUFFER_BIT_FRONT_LEFT; + } - if (RADEON_DEBUG & DEBUG_IOCTL) - fprintf(stderr, "%s\n", __FUNCTION__); + if ( mask & BUFFER_BIT_BACK_LEFT ) { + flags |= RADEON_BACK; + color_mask = rmesa->hw.msk.cmd[MSK_RB3D_PLANEMASK]; + mask &= ~BUFFER_BIT_BACK_LEFT; + } - if (rmesa->dma.flush) - rmesa->dma.flush( rmesa ); + if ( mask & BUFFER_BIT_DEPTH ) { + flags |= RADEON_DEPTH; + mask &= ~BUFFER_BIT_DEPTH; + } - radeonEmitState( rmesa ); - - if (rmesa->store.cmd_used) - radeonFlushCmdBuf( rmesa, __FUNCTION__ ); -} + if ( (mask & BUFFER_BIT_STENCIL) ) { + flags |= RADEON_STENCIL; + mask &= ~BUFFER_BIT_STENCIL; + } -/* Make sure all commands have been sent to the hardware and have - * completed processing. - */ -void radeonFinish( GLcontext *ctx ) -{ - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - radeonFlush( ctx ); - - if (rmesa->do_irqs) { - LOCK_HARDWARE( rmesa ); - radeonEmitIrqLocked( rmesa ); - UNLOCK_HARDWARE( rmesa ); - radeonWaitIrq( rmesa ); + if ( mask ) { + if (RADEON_DEBUG & RADEON_FALLBACKS) + fprintf(stderr, "%s: swrast clear, mask: %x\n", __FUNCTION__, mask); + _swrast_Clear( ctx, mask ); } - else - radeonWaitForIdle( rmesa ); -} + if ( !flags ) + return; + + if (rmesa->using_hyperz) { + flags |= RADEON_USE_COMP_ZBUF; +/* if (rmesa->radeon.radeonScreen->chipset & RADEON_CHIPSET_TCL) + flags |= RADEON_USE_HIERZ; */ + if (((flags & RADEON_DEPTH) && (flags & RADEON_STENCIL) && + ((rmesa->radeon.state.stencil.clear & RADEON_STENCIL_WRITE_MASK) == RADEON_STENCIL_WRITE_MASK))) { + flags |= RADEON_CLEAR_FASTZ; + } + } + + if (rmesa->radeon.radeonScreen->kernel_mm) + radeonUserClear(ctx, orig_mask); + else { + radeonKernelClear(ctx, flags); + rmesa->radeon.hw.all_dirty = GL_TRUE; + } +} void radeonInitIoctlFuncs( GLcontext *ctx ) { diff --git a/src/mesa/drivers/dri/radeon/radeon_ioctl.h b/src/mesa/drivers/dri/radeon/radeon_ioctl.h index 4e3a44df075..deb53ae3138 100644 --- a/src/mesa/drivers/dri/radeon/radeon_ioctl.h +++ b/src/mesa/drivers/dri/radeon/radeon_ioctl.h @@ -38,31 +38,32 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "main/simple_list.h" #include "radeon_lock.h" +#include "radeon_bocs_wrapper.h" - -extern void radeonEmitState( radeonContextPtr rmesa ); -extern void radeonEmitVertexAOS( radeonContextPtr rmesa, +extern void radeonEmitVertexAOS( r100ContextPtr rmesa, GLuint vertex_size, + struct radeon_bo *bo, GLuint offset ); -extern void radeonEmitVbufPrim( radeonContextPtr rmesa, +extern void radeonEmitVbufPrim( r100ContextPtr rmesa, GLuint vertex_format, GLuint primitive, GLuint vertex_nr ); -extern void radeonFlushElts( radeonContextPtr rmesa ); +extern void radeonFlushElts( GLcontext *ctx ); + -extern GLushort *radeonAllocEltsOpenEnded( radeonContextPtr rmesa, +extern GLushort *radeonAllocEltsOpenEnded( r100ContextPtr rmesa, GLuint vertex_format, GLuint primitive, GLuint min_nr ); -extern void radeonEmitAOS( radeonContextPtr rmesa, - struct radeon_dma_region **regions, + +extern void radeonEmitAOS( r100ContextPtr rmesa, GLuint n, GLuint offset ); -extern void radeonEmitBlit( radeonContextPtr rmesa, +extern void radeonEmitBlit( r100ContextPtr rmesa, GLuint color_fmt, GLuint src_pitch, GLuint src_offset, @@ -72,30 +73,15 @@ extern void radeonEmitBlit( radeonContextPtr rmesa, GLint dstx, GLint dsty, GLuint w, GLuint h ); -extern void radeonEmitWait( radeonContextPtr rmesa, GLuint flags ); - -extern void radeonFlushCmdBuf( radeonContextPtr rmesa, const char * ); -extern void radeonRefillCurrentDmaRegion( radeonContextPtr rmesa ); +extern void radeonEmitWait( r100ContextPtr rmesa, GLuint flags ); -extern void radeonAllocDmaRegion( radeonContextPtr rmesa, - struct radeon_dma_region *region, - int bytes, - int alignment ); +extern void radeonFlushCmdBuf( r100ContextPtr rmesa, const char * ); -extern void radeonReleaseDmaRegion( radeonContextPtr rmesa, - struct radeon_dma_region *region, - const char *caller ); - -extern void radeonCopyBuffer( __DRIdrawablePrivate *drawable, - const drm_clip_rect_t *rect); -extern void radeonPageFlip( __DRIdrawablePrivate *drawable ); extern void radeonFlush( GLcontext *ctx ); extern void radeonFinish( GLcontext *ctx ); -extern void radeonWaitForIdleLocked( radeonContextPtr rmesa ); -extern void radeonWaitForVBlank( radeonContextPtr rmesa ); extern void radeonInitIoctlFuncs( GLcontext *ctx ); -extern void radeonGetAllParams( radeonContextPtr rmesa ); -extern void radeonSetUpAtomList( radeonContextPtr rmesa ); +extern void radeonGetAllParams( r100ContextPtr rmesa ); +extern void radeonSetUpAtomList( r100ContextPtr rmesa ); /* ================================================================ * Helper macros: @@ -105,33 +91,33 @@ extern void radeonSetUpAtomList( radeonContextPtr rmesa ); */ #define RADEON_NEWPRIM( rmesa ) \ do { \ - if ( rmesa->dma.flush ) \ - rmesa->dma.flush( rmesa ); \ + if ( rmesa->radeon.dma.flush ) \ + rmesa->radeon.dma.flush( rmesa->radeon.glCtx ); \ } while (0) /* Can accomodate several state changes and primitive changes without * actually firing the buffer. */ + #define RADEON_STATECHANGE( rmesa, ATOM ) \ do { \ RADEON_NEWPRIM( rmesa ); \ rmesa->hw.ATOM.dirty = GL_TRUE; \ - rmesa->hw.is_dirty = GL_TRUE; \ + rmesa->radeon.hw.is_dirty = GL_TRUE; \ } while (0) -#define RADEON_DB_STATE( ATOM ) \ +#define RADEON_DB_STATE( ATOM ) \ memcpy( rmesa->hw.ATOM.lastcmd, rmesa->hw.ATOM.cmd, \ rmesa->hw.ATOM.cmd_size * 4) -static INLINE int RADEON_DB_STATECHANGE( - radeonContextPtr rmesa, - struct radeon_state_atom *atom ) +static INLINE int RADEON_DB_STATECHANGE(r100ContextPtr rmesa, + struct radeon_state_atom *atom ) { if (memcmp(atom->cmd, atom->lastcmd, atom->cmd_size*4)) { - int *tmp; + GLuint *tmp; RADEON_NEWPRIM( rmesa ); atom->dirty = GL_TRUE; - rmesa->hw.is_dirty = GL_TRUE; + rmesa->radeon.hw.is_dirty = GL_TRUE; tmp = atom->cmd; atom->cmd = atom->lastcmd; atom->lastcmd = tmp; @@ -141,62 +127,55 @@ static INLINE int RADEON_DB_STATECHANGE( return 0; } - -/* Fire the buffered vertices no matter what. - */ -#define RADEON_FIREVERTICES( rmesa ) \ -do { \ - if ( rmesa->store.cmd_used || rmesa->dma.flush ) { \ - radeonFlush( rmesa->glCtx ); \ - } \ -} while (0) - /* Command lengths. Note that any time you ensure ELTS_BUFSZ or VBUF_BUFSZ * are available, you will also be adding an rmesa->state.max_state_size because * r200EmitState is called from within r200EmitVbufPrim and r200FlushElts. */ #if RADEON_OLD_PACKETS -#define AOS_BUFSZ(nr) ((3 + ((nr / 2) * 3) + ((nr & 1) * 2)) * sizeof(int)) +#define AOS_BUFSZ(nr) ((3 + ((nr / 2) * 3) + ((nr & 1) * 2))+nr*2) #define VERT_AOS_BUFSZ (0) #define ELTS_BUFSZ(nr) (24 + nr * 2) -#define VBUF_BUFSZ (6 * sizeof(int)) +#define VBUF_BUFSZ (8) #else -#define AOS_BUFSZ(nr) ((3 + ((nr / 2) * 3) + ((nr & 1) * 2)) * sizeof(int)) -#define VERT_AOS_BUFSZ (5 * sizeof(int)) +#define AOS_BUFSZ(nr) ((3 + ((nr / 2) * 3) + ((nr & 1) * 2) + nr*2)) +#define VERT_AOS_BUFSZ (5) #define ELTS_BUFSZ(nr) (16 + nr * 2) -#define VBUF_BUFSZ (4 * sizeof(int)) +#define VBUF_BUFSZ (4) #endif +#define SCISSOR_BUFSZ (8) +#define INDEX_BUFSZ (7) -/* Ensure that a minimum amount of space is available in the command buffer. - * This is used to ensure atomicity of state updates with the rendering requests - * that rely on them. - * - * An alternative would be to implement a "soft lock" such that when the buffer - * wraps at an inopportune time, we grab the lock, flush the current buffer, - * and hang on to the lock until the critical section is finished and we flush - * the buffer again and unlock. - */ -static INLINE void radeonEnsureCmdBufSpace( radeonContextPtr rmesa, - int bytes ) -{ - if (rmesa->store.cmd_used + bytes > RADEON_CMD_BUF_SZ) - radeonFlushCmdBuf( rmesa, __FUNCTION__ ); - assert( bytes <= RADEON_CMD_BUF_SZ ); -} -/* Alloc space in the command buffer - */ -static INLINE char *radeonAllocCmdBuf( radeonContextPtr rmesa, - int bytes, const char *where ) +static inline uint32_t cmdpacket3(int cmd_type) { - if (rmesa->store.cmd_used + bytes > RADEON_CMD_BUF_SZ) - radeonFlushCmdBuf( rmesa, __FUNCTION__ ); + drm_radeon_cmd_header_t cmd; + + cmd.i = 0; + cmd.header.cmd_type = cmd_type; + + return (uint32_t)cmd.i; - { - char *head = rmesa->store.cmd_buf + rmesa->store.cmd_used; - rmesa->store.cmd_used += bytes; - return head; - } } +#define OUT_BATCH_PACKET3(packet, num_extra) do { \ + if (!b_l_rmesa->radeonScreen->kernel_mm) { \ + OUT_BATCH(cmdpacket3(RADEON_CMD_PACKET3)); \ + OUT_BATCH(CP_PACKET3((packet), (num_extra))); \ + } else { \ + OUT_BATCH(CP_PACKET2); \ + OUT_BATCH(CP_PACKET3((packet), (num_extra))); \ + } \ + } while(0) + +#define OUT_BATCH_PACKET3_CLIP(packet, num_extra) do { \ + if (!b_l_rmesa->radeonScreen->kernel_mm) { \ + OUT_BATCH(cmdpacket3(RADEON_CMD_PACKET3_CLIP)); \ + OUT_BATCH(CP_PACKET3((packet), (num_extra))); \ + } else { \ + OUT_BATCH(CP_PACKET2); \ + OUT_BATCH(CP_PACKET3((packet), (num_extra))); \ + } \ + } while(0) + + #endif /* __RADEON_IOCTL_H__ */ diff --git a/src/mesa/drivers/dri/radeon/radeon_lighting.c b/src/mesa/drivers/dri/radeon/radeon_lighting.c index ac3b94e4a68..ba444f2b100 100644 --- a/src/mesa/drivers/dri/radeon/radeon_lighting.c +++ b/src/mesa/drivers/dri/radeon/radeon_lighting.c @@ -195,7 +195,7 @@ void radeonUpdateMaterial( GLcontext *ctx ) if (ctx->Light.ColorMaterialEnabled) mask &= ~ctx->Light.ColorMaterialBitmask; - if (RADEON_DEBUG & DEBUG_STATE) + if (RADEON_DEBUG & RADEON_STATE) fprintf(stderr, "%s\n", __FUNCTION__); @@ -234,7 +234,7 @@ void radeonUpdateMaterial( GLcontext *ctx ) check_twoside_fallback( ctx ); update_global_ambient( ctx ); } - else if (RADEON_DEBUG & (DEBUG_PRIMS|DEBUG_STATE)) + else if (RADEON_DEBUG & (RADEON_PRIMS|DEBUG_STATE)) fprintf(stderr, "%s: Elided noop material call\n", __FUNCTION__); } @@ -624,7 +624,7 @@ static void radeonLightingSpaceChange( GLcontext *ctx ) GLboolean tmp; RADEON_STATECHANGE( rmesa, tcl ); - if (RADEON_DEBUG & DEBUG_STATE) + if (RADEON_DEBUG & RADEON_STATE) fprintf(stderr, "%s %d\n", __FUNCTION__, ctx->_NeedEyeCoords); if (ctx->_NeedEyeCoords) diff --git a/src/mesa/drivers/dri/radeon/radeon_lock.c b/src/mesa/drivers/dri/radeon/radeon_lock.c index 64bb3ca103f..02de8e5fd1d 100644 --- a/src/mesa/drivers/dri/radeon/radeon_lock.c +++ b/src/mesa/drivers/dri/radeon/radeon_lock.c @@ -41,30 +41,13 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "main/glheader.h" #include "main/mtypes.h" -#include "radeon_context.h" +#include "main/colormac.h" +#include "dri_util.h" +#include "radeon_screen.h" +#include "radeon_common.h" #include "radeon_lock.h" -#include "radeon_tex.h" -#include "radeon_state.h" -#include "radeon_ioctl.h" - #include "drirenderbuffer.h" -#if DEBUG_LOCKING -char *prevLockFile = NULL; -int prevLockLine = 0; -#endif - -/* Turn on/off page flipping according to the flags in the sarea: - */ -static void radeonUpdatePageFlipping(radeonContextPtr rmesa) -{ - rmesa->doPageFlip = rmesa->sarea->pfState; - if (rmesa->glCtx->WinSysDrawBuffer) { - driFlipRenderbuffers(rmesa->glCtx->WinSysDrawBuffer, - rmesa->sarea->pfCurrentPage); - } -} - /* Update the hardware state. This is called if another context has * grabbed the hardware lock, which includes the X server. This * function also updates the driver's window state after the X server @@ -75,10 +58,11 @@ static void radeonUpdatePageFlipping(radeonContextPtr rmesa) */ void radeonGetLock(radeonContextPtr rmesa, GLuint flags) { - __DRIdrawablePrivate *const drawable = rmesa->dri.drawable; - __DRIdrawablePrivate *const readable = rmesa->dri.readable; + __DRIdrawablePrivate *const drawable = radeon_get_drawable(rmesa); + __DRIdrawablePrivate *const readable = radeon_get_readable(rmesa); __DRIscreenPrivate *sPriv = rmesa->dri.screen; - drm_radeon_sarea_t *sarea = rmesa->sarea; + + assert(drawable != NULL); drmGetLock(rmesa->dri.fd, rmesa->dri.hwContext, flags); @@ -96,29 +80,89 @@ void radeonGetLock(radeonContextPtr rmesa, GLuint flags) } if (rmesa->lastStamp != drawable->lastStamp) { - radeonUpdatePageFlipping(rmesa); - radeonSetCliprects(rmesa); - radeonUpdateViewportOffset(rmesa->glCtx); - driUpdateFramebufferSize(rmesa->glCtx, drawable); + radeon_window_moved(rmesa); + rmesa->lastStamp = drawable->lastStamp; } - RADEON_STATECHANGE(rmesa, ctx); - if (rmesa->sarea->tiling_enabled) { - rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] |= - RADEON_COLOR_TILE_ENABLE; - } else { - rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] &= - ~RADEON_COLOR_TILE_ENABLE; - } + rmesa->vtbl.get_lock(rmesa); +} +#ifndef NDEBUG +struct lock_debug { + const char* function; + const char* file; + int line; +}; + +static struct lock_debug ldebug = {0}; +#endif + +#if 0 +/** TODO: use atomic operations for reference counting **/ +/** gcc 4.2 has builtin functios for this **/ +#define ATOMIC_INC_AND_FETCH(atomic) __sync_add_and_fetch(&atomic, 1) +#define ATOMIC_DEC_AND_FETCH(atomic) __sync_sub_and_fetch(&atomic, 1) +#else +#define ATOMIC_INC_AND_FETCH(atomic) (++atomic) +#define ATOMIC_DEC_AND_FETCH(atomic) (--atomic) +#endif + - if (sarea->ctx_owner != rmesa->dri.hwContext) { - int i; - sarea->ctx_owner = rmesa->dri.hwContext; +void radeon_lock_hardware(radeonContextPtr radeon +#ifndef NDEBUG + ,const char* function + ,const char* file + ,const int line +#endif + ) +{ + char ret = 0; + struct radeon_framebuffer *rfb = NULL; + struct radeon_renderbuffer *rrb = NULL; + + if (radeon_get_drawable(radeon)) { + rfb = radeon_get_drawable(radeon)->driverPrivate; + + if (rfb) + rrb = radeon_get_renderbuffer(&rfb->base, + rfb->base._ColorDrawBufferIndexes[0]); + } - for (i = 0; i < rmesa->nr_heaps; i++) { - DRI_AGE_TEXTURES(rmesa->texture_heaps[i]); + if (!radeon->radeonScreen->driScreen->dri2.enabled) { + if (ATOMIC_INC_AND_FETCH(radeon->dri.hwLockCount) > 1) + { +#ifndef NDEBUG + if ( RADEON_DEBUG & RADEON_SANITY ) + fprintf(stderr, "*** %d times of recursive call to %s ***\n" + "Original call was from %s (file: %s line: %d)\n" + "Now call is coming from %s (file: %s line: %d)\n" + , radeon->dri.hwLockCount, __FUNCTION__ + , ldebug.function, ldebug.file, ldebug.line + , function, file, line + ); +#endif + return; } + DRM_CAS(radeon->dri.hwLock, radeon->dri.hwContext, + (DRM_LOCK_HELD | radeon->dri.hwContext), ret ); + if (ret) + radeonGetLock(radeon, 0); +#ifndef NDEBUG + ldebug.function = function; + ldebug.file = file; + ldebug.line = line; +#endif } +} - rmesa->lost_context = GL_TRUE; +void radeon_unlock_hardware(radeonContextPtr radeon) +{ + if (!radeon->radeonScreen->driScreen->dri2.enabled) { + if (ATOMIC_DEC_AND_FETCH(radeon->dri.hwLockCount) > 0) + { + return; + } + DRM_UNLOCK( radeon->dri.fd, + radeon->dri.hwLock, + radeon->dri.hwContext ); + } } diff --git a/src/mesa/drivers/dri/radeon/radeon_lock.h b/src/mesa/drivers/dri/radeon/radeon_lock.h index 86e96aa7d2c..da5a5b43715 100644 --- a/src/mesa/drivers/dri/radeon/radeon_lock.h +++ b/src/mesa/drivers/dri/radeon/radeon_lock.h @@ -39,74 +39,31 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. * Kevin E. Martin <martin@valinux.com> */ -#ifndef __RADEON_LOCK_H__ -#define __RADEON_LOCK_H__ +#ifndef COMMON_LOCK_H +#define COMMON_LOCK_H -extern void radeonGetLock(radeonContextPtr rmesa, GLuint flags); - -/* Turn DEBUG_LOCKING on to find locking conflicts. - */ -#define DEBUG_LOCKING 0 - -#if DEBUG_LOCKING -extern char *prevLockFile; -extern int prevLockLine; - -#define DEBUG_LOCK() \ - do { \ - prevLockFile = (__FILE__); \ - prevLockLine = (__LINE__); \ - } while (0) - -#define DEBUG_RESET() \ - do { \ - prevLockFile = 0; \ - prevLockLine = 0; \ - } while (0) - -#define DEBUG_CHECK_LOCK() \ - do { \ - if ( prevLockFile ) { \ - fprintf( stderr, \ - "LOCK SET!\n\tPrevious %s:%d\n\tCurrent: %s:%d\n", \ - prevLockFile, prevLockLine, __FILE__, __LINE__ ); \ - exit( 1 ); \ - } \ - } while (0) - -#else +#include "main/colormac.h" +#include "radeon_screen.h" +#include "radeon_common.h" -#define DEBUG_LOCK() -#define DEBUG_RESET() -#define DEBUG_CHECK_LOCK() +extern void radeonGetLock(radeonContextPtr rmesa, GLuint flags); +void radeon_lock_hardware(radeonContextPtr rmesa +#ifndef NDEBUG + ,const char* function + ,const char* file + ,const int line #endif - -/* - * !!! We may want to separate locks from locks with validation. This - * could be used to improve performance for those things commands that - * do not do any drawing !!! - */ + ); +void radeon_unlock_hardware(radeonContextPtr rmesa); /* Lock the hardware and validate our state. */ -#define LOCK_HARDWARE( rmesa ) \ - do { \ - char __ret = 0; \ - DEBUG_CHECK_LOCK(); \ - DRM_CAS( (rmesa)->dri.hwLock, (rmesa)->dri.hwContext, \ - (DRM_LOCK_HELD | (rmesa)->dri.hwContext), __ret ); \ - if ( __ret ) \ - radeonGetLock( (rmesa), 0 ); \ - DEBUG_LOCK(); \ - } while (0) - -#define UNLOCK_HARDWARE( rmesa ) \ - do { \ - DRM_UNLOCK( (rmesa)->dri.fd, \ - (rmesa)->dri.hwLock, \ - (rmesa)->dri.hwContext ); \ - DEBUG_RESET(); \ - } while (0) +#ifdef NDEBUG +#define LOCK_HARDWARE( rmesa ) radeon_lock_hardware(rmesa) +#else +#define LOCK_HARDWARE( rmesa ) radeon_lock_hardware(rmesa, __FUNCTION__, __FILE__, __LINE__) +#endif +#define UNLOCK_HARDWARE( rmesa ) radeon_unlock_hardware(rmesa) -#endif /* __RADEON_LOCK_H__ */ +#endif diff --git a/src/mesa/drivers/dri/radeon/radeon_maos.h b/src/mesa/drivers/dri/radeon/radeon_maos.h index b8935e84a05..b88eb198d57 100644 --- a/src/mesa/drivers/dri/radeon/radeon_maos.h +++ b/src/mesa/drivers/dri/radeon/radeon_maos.h @@ -38,6 +38,5 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "radeon_context.h" extern void radeonEmitArrays( GLcontext *ctx, GLuint inputs ); -extern void radeonReleaseArrays( GLcontext *ctx, GLuint newinputs ); #endif diff --git a/src/mesa/drivers/dri/radeon/radeon_maos_arrays.c b/src/mesa/drivers/dri/radeon/radeon_maos_arrays.c index 31eea13f4ef..08e1c5d00df 100644 --- a/src/mesa/drivers/dri/radeon/radeon_maos_arrays.c +++ b/src/mesa/drivers/dri/radeon/radeon_maos_arrays.c @@ -48,160 +48,35 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "radeon_maos.h" #include "radeon_tcl.h" -#if 0 -/* Usage: - * - from radeon_tcl_render - * - call radeonEmitArrays to ensure uptodate arrays in dma - * - emit primitives (new type?) which reference the data - * -- need to use elts for lineloop, quads, quadstrip/flat - * -- other primitives are all well-formed (need tristrip-1,fake-poly) - * - */ -static void emit_ubyte_rgba3( GLcontext *ctx, - struct radeon_dma_region *rvb, - char *data, - int stride, - int count ) -{ - int i; - radeon_color_t *out = (radeon_color_t *)(rvb->start + rvb->address); - - if (RADEON_DEBUG & DEBUG_VERTS) - fprintf(stderr, "%s count %d stride %d out %p\n", - __FUNCTION__, count, stride, (void *)out); - - for (i = 0; i < count; i++) { - out->red = *data; - out->green = *(data+1); - out->blue = *(data+2); - out->alpha = 0xFF; - out++; - data += stride; - } -} - -static void emit_ubyte_rgba4( GLcontext *ctx, - struct radeon_dma_region *rvb, - char *data, - int stride, - int count ) +static void emit_vecfog(GLcontext *ctx, struct radeon_aos *aos, + GLvoid *data, int stride, int count) { int i; - int *out = (int *)(rvb->address + rvb->start); + uint32_t *out; + int size = 1; + radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - if (RADEON_DEBUG & DEBUG_VERTS) + if (RADEON_DEBUG & RADEON_VERTS) fprintf(stderr, "%s count %d stride %d\n", __FUNCTION__, count, stride); - if (stride == 4) - COPY_DWORDS( out, data, count ); - else - for (i = 0; i < count; i++) { - *out++ = LE32_TO_CPU(*(int *)data); - data += stride; - } -} - - -static void emit_ubyte_rgba( GLcontext *ctx, - struct radeon_dma_region *rvb, - char *data, - int size, - int stride, - int count ) -{ - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - - if (RADEON_DEBUG & DEBUG_VERTS) - fprintf(stderr, "%s %d/%d\n", __FUNCTION__, count, size); - - assert (!rvb->buf); - if (stride == 0) { - radeonAllocDmaRegion( rmesa, rvb, 4, 4 ); + radeonAllocDmaRegion( rmesa, &aos->bo, &aos->offset, size * 4, 32 ); count = 1; - rvb->aos_start = GET_START(rvb); - rvb->aos_stride = 0; - rvb->aos_size = 1; + aos->stride = 0; } else { - radeonAllocDmaRegion( rmesa, rvb, 4 * count, 4 ); /* alignment? */ - rvb->aos_start = GET_START(rvb); - rvb->aos_stride = 1; - rvb->aos_size = 1; - } - - /* Emit the data - */ - switch (size) { - case 3: - emit_ubyte_rgba3( ctx, rvb, data, stride, count ); - break; - case 4: - emit_ubyte_rgba4( ctx, rvb, data, stride, count ); - break; - default: - assert(0); - exit(1); - break; + radeonAllocDmaRegion(rmesa, &aos->bo, &aos->offset, size * 4, 32); + aos->stride = size; } -} -#endif - -#if defined(USE_X86_ASM) -#define COPY_DWORDS( dst, src, nr ) \ -do { \ - int __tmp; \ - __asm__ __volatile__( "rep ; movsl" \ - : "=%c" (__tmp), "=D" (dst), "=S" (__tmp) \ - : "0" (nr), \ - "D" ((long)dst), \ - "S" ((long)src) ); \ -} while (0) -#else -#define COPY_DWORDS( dst, src, nr ) \ -do { \ - int j; \ - for ( j = 0 ; j < nr ; j++ ) \ - dst[j] = ((int *)src)[j]; \ - dst += nr; \ -} while (0) -#endif - -static void emit_vecfog( GLcontext *ctx, - struct radeon_dma_region *rvb, - char *data, - int stride, - int count ) -{ - int i; - GLfloat *out; - - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - if (RADEON_DEBUG & DEBUG_VERTS) - fprintf(stderr, "%s count %d stride %d\n", - __FUNCTION__, count, stride); + aos->components = size; + aos->count = count; - assert (!rvb->buf); - - if (stride == 0) { - radeonAllocDmaRegion( rmesa, rvb, 4, 4 ); - count = 1; - rvb->aos_start = GET_START(rvb); - rvb->aos_stride = 0; - rvb->aos_size = 1; - } - else { - radeonAllocDmaRegion( rmesa, rvb, count * 4, 4 ); /* alignment? */ - rvb->aos_start = GET_START(rvb); - rvb->aos_stride = 1; - rvb->aos_size = 1; - } /* Emit the data */ - out = (GLfloat *)(rvb->address + rvb->start); + out = (uint32_t*)((char*)aos->bo->ptr + aos->offset); for (i = 0; i < count; i++) { out[0] = radeonComputeFogBlendFactor( ctx, *(GLfloat *)data ); out++; @@ -209,170 +84,10 @@ static void emit_vecfog( GLcontext *ctx, } } -static void emit_vec4( GLcontext *ctx, - struct radeon_dma_region *rvb, - char *data, - int stride, - int count ) -{ - int i; - int *out = (int *)(rvb->address + rvb->start); - - if (RADEON_DEBUG & DEBUG_VERTS) - fprintf(stderr, "%s count %d stride %d\n", - __FUNCTION__, count, stride); - - if (stride == 4) - COPY_DWORDS( out, data, count ); - else - for (i = 0; i < count; i++) { - out[0] = *(int *)data; - out++; - data += stride; - } -} - - -static void emit_vec8( GLcontext *ctx, - struct radeon_dma_region *rvb, - char *data, - int stride, - int count ) -{ - int i; - int *out = (int *)(rvb->address + rvb->start); - - if (RADEON_DEBUG & DEBUG_VERTS) - fprintf(stderr, "%s count %d stride %d\n", - __FUNCTION__, count, stride); - - if (stride == 8) - COPY_DWORDS( out, data, count*2 ); - else - for (i = 0; i < count; i++) { - out[0] = *(int *)data; - out[1] = *(int *)(data+4); - out += 2; - data += stride; - } -} - -static void emit_vec12( GLcontext *ctx, - struct radeon_dma_region *rvb, - char *data, - int stride, - int count ) -{ - int i; - int *out = (int *)(rvb->address + rvb->start); - - if (RADEON_DEBUG & DEBUG_VERTS) - fprintf(stderr, "%s count %d stride %d out %p data %p\n", - __FUNCTION__, count, stride, (void *)out, (void *)data); - - if (stride == 12) - COPY_DWORDS( out, data, count*3 ); - else - for (i = 0; i < count; i++) { - out[0] = *(int *)data; - out[1] = *(int *)(data+4); - out[2] = *(int *)(data+8); - out += 3; - data += stride; - } -} - -static void emit_vec16( GLcontext *ctx, - struct radeon_dma_region *rvb, - char *data, - int stride, - int count ) -{ - int i; - int *out = (int *)(rvb->address + rvb->start); - - if (RADEON_DEBUG & DEBUG_VERTS) - fprintf(stderr, "%s count %d stride %d\n", - __FUNCTION__, count, stride); - - if (stride == 16) - COPY_DWORDS( out, data, count*4 ); - else - for (i = 0; i < count; i++) { - out[0] = *(int *)data; - out[1] = *(int *)(data+4); - out[2] = *(int *)(data+8); - out[3] = *(int *)(data+12); - out += 4; - data += stride; - } -} - - -static void emit_vector( GLcontext *ctx, - struct radeon_dma_region *rvb, - char *data, - int size, - int stride, - int count ) -{ - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - - if (RADEON_DEBUG & DEBUG_VERTS) - fprintf(stderr, "%s count %d size %d stride %d\n", - __FUNCTION__, count, size, stride); - - assert (!rvb->buf); - - if (stride == 0) { - radeonAllocDmaRegion( rmesa, rvb, size * 4, 4 ); - count = 1; - rvb->aos_start = GET_START(rvb); - rvb->aos_stride = 0; - rvb->aos_size = size; - } - else { - radeonAllocDmaRegion( rmesa, rvb, size * count * 4, 4 ); /* alignment? */ - rvb->aos_start = GET_START(rvb); - rvb->aos_stride = size; - rvb->aos_size = size; - } - - /* Emit the data - */ - switch (size) { - case 1: - emit_vec4( ctx, rvb, data, stride, count ); - break; - case 2: - emit_vec8( ctx, rvb, data, stride, count ); - break; - case 3: - emit_vec12( ctx, rvb, data, stride, count ); - break; - case 4: - emit_vec16( ctx, rvb, data, stride, count ); - break; - default: - assert(0); - exit(1); - break; - } - -} - - - -static void emit_s0_vec( GLcontext *ctx, - struct radeon_dma_region *rvb, - char *data, - int stride, - int count ) +static void emit_s0_vec(uint32_t *out, GLvoid *data, int stride, int count) { int i; - int *out = (int *)(rvb->address + rvb->start); - - if (RADEON_DEBUG & DEBUG_VERTS) + if (RADEON_DEBUG & RADEON_VERTS) fprintf(stderr, "%s count %d stride %d\n", __FUNCTION__, count, stride); @@ -384,16 +99,11 @@ static void emit_s0_vec( GLcontext *ctx, } } -static void emit_stq_vec( GLcontext *ctx, - struct radeon_dma_region *rvb, - char *data, - int stride, - int count ) +static void emit_stq_vec(uint32_t *out, GLvoid *data, int stride, int count) { int i; - int *out = (int *)(rvb->address + rvb->start); - if (RADEON_DEBUG & DEBUG_VERTS) + if (RADEON_DEBUG & RADEON_VERTS) fprintf(stderr, "%s count %d stride %d\n", __FUNCTION__, count, stride); @@ -409,21 +119,16 @@ static void emit_stq_vec( GLcontext *ctx, -static void emit_tex_vector( GLcontext *ctx, - struct radeon_dma_region *rvb, - char *data, - int size, - int stride, - int count ) +static void emit_tex_vector(GLcontext *ctx, struct radeon_aos *aos, + GLvoid *data, int size, int stride, int count) { radeonContextPtr rmesa = RADEON_CONTEXT(ctx); int emitsize; + uint32_t *out; - if (RADEON_DEBUG & DEBUG_VERTS) + if (RADEON_DEBUG & RADEON_VERTS) fprintf(stderr, "%s %d/%d\n", __FUNCTION__, count, size); - assert (!rvb->buf); - switch (size) { case 4: emitsize = 3; break; case 3: emitsize = 3; break; @@ -432,34 +137,33 @@ static void emit_tex_vector( GLcontext *ctx, if (stride == 0) { - radeonAllocDmaRegion( rmesa, rvb, 4 * emitsize, 4 ); + radeonAllocDmaRegion(rmesa, &aos->bo, &aos->offset, emitsize * 4, 32); count = 1; - rvb->aos_start = GET_START(rvb); - rvb->aos_stride = 0; - rvb->aos_size = emitsize; + aos->stride = 0; } else { - radeonAllocDmaRegion( rmesa, rvb, 4 * emitsize * count, 4 ); - rvb->aos_start = GET_START(rvb); - rvb->aos_stride = emitsize; - rvb->aos_size = emitsize; + radeonAllocDmaRegion(rmesa, &aos->bo, &aos->offset, emitsize * count * 4, 32); + aos->stride = emitsize; } + aos->components = emitsize; + aos->count = count; /* Emit the data */ + out = (uint32_t*)((char*)aos->bo->ptr + aos->offset); switch (size) { case 1: - emit_s0_vec( ctx, rvb, data, stride, count ); + emit_s0_vec( out, data, stride, count ); break; case 2: - emit_vec8( ctx, rvb, data, stride, count ); + radeonEmitVec8( out, data, stride, count ); break; case 3: - emit_vec12( ctx, rvb, data, stride, count ); + radeonEmitVec12( out, data, stride, count ); break; case 4: - emit_stq_vec( ctx, rvb, data, stride, count ); + emit_stq_vec( out, data, stride, count ); break; default: assert(0); @@ -476,27 +180,26 @@ static void emit_tex_vector( GLcontext *ctx, */ void radeonEmitArrays( GLcontext *ctx, GLuint inputs ) { - radeonContextPtr rmesa = RADEON_CONTEXT( ctx ); + r100ContextPtr rmesa = R100_CONTEXT( ctx ); struct vertex_buffer *VB = &TNL_CONTEXT( ctx )->vb; - struct radeon_dma_region **component = rmesa->tcl.aos_components; GLuint nr = 0; GLuint vfmt = 0; GLuint count = VB->Count; GLuint vtx, unit; #if 0 - if (RADEON_DEBUG & DEBUG_VERTS) + if (RADEON_DEBUG & RADEON_VERTS) _tnl_print_vert_flags( __FUNCTION__, inputs ); #endif if (1) { if (!rmesa->tcl.obj.buf) - emit_vector( ctx, - &rmesa->tcl.obj, - (char *)VB->ObjPtr->data, - VB->ObjPtr->size, - VB->ObjPtr->stride, - count); + rcommon_emit_vector( ctx, + &(rmesa->tcl.aos[nr]), + (char *)VB->ObjPtr->data, + VB->ObjPtr->size, + VB->ObjPtr->stride, + count); switch( VB->ObjPtr->size ) { case 4: vfmt |= RADEON_CP_VC_FRMT_W0; @@ -505,21 +208,21 @@ void radeonEmitArrays( GLcontext *ctx, GLuint inputs ) default: break; } - component[nr++] = &rmesa->tcl.obj; + nr++; } if (inputs & VERT_BIT_NORMAL) { if (!rmesa->tcl.norm.buf) - emit_vector( ctx, - &(rmesa->tcl.norm), - (char *)VB->NormalPtr->data, - 3, - VB->NormalPtr->stride, - count); + rcommon_emit_vector( ctx, + &(rmesa->tcl.aos[nr]), + (char *)VB->NormalPtr->data, + 3, + VB->NormalPtr->stride, + count); vfmt |= RADEON_CP_VC_FRMT_N0; - component[nr++] = &rmesa->tcl.norm; + nr++; } if (inputs & VERT_BIT_COLOR0) { @@ -537,31 +240,30 @@ void radeonEmitArrays( GLcontext *ctx, GLuint inputs ) } if (!rmesa->tcl.rgba.buf) - emit_vector( ctx, - &(rmesa->tcl.rgba), - (char *)VB->ColorPtr[0]->data, - emitsize, - VB->ColorPtr[0]->stride, - count); - - - component[nr++] = &rmesa->tcl.rgba; + rcommon_emit_vector( ctx, + &(rmesa->tcl.aos[nr]), + (char *)VB->ColorPtr[0]->data, + emitsize, + VB->ColorPtr[0]->stride, + count); + + nr++; } if (inputs & VERT_BIT_COLOR1) { if (!rmesa->tcl.spec.buf) { - emit_vector( ctx, - &rmesa->tcl.spec, - (char *)VB->SecondaryColorPtr[0]->data, - 3, - VB->SecondaryColorPtr[0]->stride, - count); + rcommon_emit_vector( ctx, + &(rmesa->tcl.aos[nr]), + (char *)VB->SecondaryColorPtr[0]->data, + 3, + VB->SecondaryColorPtr[0]->stride, + count); } vfmt |= RADEON_CP_VC_FRMT_FPSPEC; - component[nr++] = &rmesa->tcl.spec; + nr++; } /* FIXME: not sure if this is correct. May need to stitch this together with @@ -570,13 +272,13 @@ void radeonEmitArrays( GLcontext *ctx, GLuint inputs ) if (inputs & VERT_BIT_FOG) { if (!rmesa->tcl.fog.buf) emit_vecfog( ctx, - &(rmesa->tcl.fog), + &(rmesa->tcl.aos[nr]), (char *)VB->FogCoordPtr->data, VB->FogCoordPtr->stride, count); vfmt |= RADEON_CP_VC_FRMT_FPFOG; - component[nr++] = &rmesa->tcl.fog; + nr++; } @@ -587,11 +289,12 @@ void radeonEmitArrays( GLcontext *ctx, GLuint inputs ) if (inputs & VERT_BIT_TEX(unit)) { if (!rmesa->tcl.tex[unit].buf) emit_tex_vector( ctx, - &(rmesa->tcl.tex[unit]), + &(rmesa->tcl.aos[nr]), (char *)VB->TexCoordPtr[unit]->data, VB->TexCoordPtr[unit]->size, VB->TexCoordPtr[unit]->stride, count ); + nr++; vfmt |= RADEON_ST_BIT(unit); /* assume we need the 3rd coord if texgen is active for r/q OR at least @@ -609,7 +312,6 @@ void radeonEmitArrays( GLcontext *ctx, GLuint inputs ) (swaptexmatcol != ((rmesa->TexMatColSwap >> unit) & 1))) radeonUploadTexMatrix( rmesa, unit, swaptexmatcol ) ; } - component[nr++] = &rmesa->tcl.tex[unit]; } } @@ -622,34 +324,3 @@ void radeonEmitArrays( GLcontext *ctx, GLuint inputs ) rmesa->tcl.vertex_format = vfmt; } - -void radeonReleaseArrays( GLcontext *ctx, GLuint newinputs ) -{ - radeonContextPtr rmesa = RADEON_CONTEXT( ctx ); - GLuint unit; - -#if 0 - if (RADEON_DEBUG & DEBUG_VERTS) - _tnl_print_vert_flags( __FUNCTION__, newinputs ); -#endif - - if (newinputs & VERT_BIT_POS) - radeonReleaseDmaRegion( rmesa, &rmesa->tcl.obj, __FUNCTION__ ); - - if (newinputs & VERT_BIT_NORMAL) - radeonReleaseDmaRegion( rmesa, &rmesa->tcl.norm, __FUNCTION__ ); - - if (newinputs & VERT_BIT_COLOR0) - radeonReleaseDmaRegion( rmesa, &rmesa->tcl.rgba, __FUNCTION__ ); - - if (newinputs & VERT_BIT_COLOR1) - radeonReleaseDmaRegion( rmesa, &rmesa->tcl.spec, __FUNCTION__ ); - - if (newinputs & VERT_BIT_FOG) - radeonReleaseDmaRegion( rmesa, &rmesa->tcl.fog, __FUNCTION__ ); - - for (unit = 0 ; unit < ctx->Const.MaxTextureUnits; unit++) { - if (newinputs & VERT_BIT_TEX(unit)) - radeonReleaseDmaRegion( rmesa, &rmesa->tcl.tex[unit], __FUNCTION__ ); - } -} diff --git a/src/mesa/drivers/dri/radeon/radeon_maos_vbtmp.h b/src/mesa/drivers/dri/radeon/radeon_maos_vbtmp.h index 034cda8a65b..515783135d6 100644 --- a/src/mesa/drivers/dri/radeon/radeon_maos_vbtmp.h +++ b/src/mesa/drivers/dri/radeon/radeon_maos_vbtmp.h @@ -54,8 +54,7 @@ static void TAG(emit)( GLcontext *ctx, union emit_union *v = (union emit_union *)dest; - if (RADEON_DEBUG & DEBUG_VERTS) - fprintf(stderr, "%s\n", __FUNCTION__); + radeon_print(RADEON_SWRENDER, RADEON_VERBOSE, "%s\n", __FUNCTION__); coord = (GLuint (*)[4])VB->ObjPtr->data; coord_stride = VB->ObjPtr->stride; diff --git a/src/mesa/drivers/dri/radeon/radeon_maos_verts.c b/src/mesa/drivers/dri/radeon/radeon_maos_verts.c index 126d0727c63..78ec1193026 100644 --- a/src/mesa/drivers/dri/radeon/radeon_maos_verts.c +++ b/src/mesa/drivers/dri/radeon/radeon_maos_verts.c @@ -310,7 +310,7 @@ static void init_tcl_verts( void ) void radeonEmitArrays( GLcontext *ctx, GLuint inputs ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb; GLuint req = 0; GLuint unit; @@ -374,14 +374,15 @@ void radeonEmitArrays( GLcontext *ctx, GLuint inputs ) break; if (rmesa->tcl.vertex_format == setup_tab[i].vertex_format && - rmesa->tcl.indexed_verts.buf) + rmesa->radeon.tcl.aos[0].bo) return; - if (rmesa->tcl.indexed_verts.buf) + if (rmesa->radeon.tcl.aos[0].bo) radeonReleaseArrays( ctx, ~0 ); - radeonAllocDmaRegion( rmesa, - &rmesa->tcl.indexed_verts, + radeonAllocDmaRegion( &rmesa->radeon, + &rmesa->radeon.tcl.aos[0].bo, + &rmesa->radeon.tcl.aos[0].offset, VB->Count * setup_tab[i].vertex_size * 4, 4); @@ -421,29 +422,12 @@ void radeonEmitArrays( GLcontext *ctx, GLuint inputs ) setup_tab[i].emit( ctx, 0, VB->Count, - rmesa->tcl.indexed_verts.address + - rmesa->tcl.indexed_verts.start ); + rmesa->radeon.tcl.aos[0].bo->ptr + rmesa->radeon.tcl.aos[0].offset); + // rmesa->radeon.tcl.aos[0].size = setup_tab[i].vertex_size; + rmesa->radeon.tcl.aos[0].stride = setup_tab[i].vertex_size; rmesa->tcl.vertex_format = setup_tab[i].vertex_format; - rmesa->tcl.indexed_verts.aos_start = GET_START( &rmesa->tcl.indexed_verts ); - rmesa->tcl.indexed_verts.aos_size = setup_tab[i].vertex_size; - rmesa->tcl.indexed_verts.aos_stride = setup_tab[i].vertex_size; - - rmesa->tcl.aos_components[0] = &rmesa->tcl.indexed_verts; - rmesa->tcl.nr_aos_components = 1; + rmesa->radeon.tcl.aos_count = 1; } - -void radeonReleaseArrays( GLcontext *ctx, GLuint newinputs ) -{ - radeonContextPtr rmesa = RADEON_CONTEXT( ctx ); - -#if 0 - if (RADEON_DEBUG & DEBUG_VERTS) - _tnl_print_vert_flags( __FUNCTION__, newinputs ); -#endif - - if (newinputs) - radeonReleaseDmaRegion( rmesa, &rmesa->tcl.indexed_verts, __FUNCTION__ ); -} diff --git a/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.c b/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.c new file mode 100644 index 00000000000..38db305e2a7 --- /dev/null +++ b/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.c @@ -0,0 +1,421 @@ +/* + * Copyright (C) 2008 Nicolai Haehnle. + * + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sublicense, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial + * portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + */ + +#include "radeon_mipmap_tree.h" + +#include <errno.h> +#include <unistd.h> + +#include "main/simple_list.h" +#include "main/texcompress.h" +#include "main/texformat.h" + +static GLuint radeon_compressed_texture_size(GLcontext *ctx, + GLsizei width, GLsizei height, GLsizei depth, + GLuint mesaFormat) +{ + GLuint size = _mesa_compressed_texture_size(ctx, width, height, depth, mesaFormat); + + if (mesaFormat == MESA_FORMAT_RGB_DXT1 || + mesaFormat == MESA_FORMAT_RGBA_DXT1) { + if (width + 3 < 8) /* width one block */ + size = size * 4; + else if (width + 3 < 16) + size = size * 2; + } else { + /* DXT3/5, 16 bytes per block */ + // WARN_ONCE("DXT 3/5 suffers from multitexturing problems!\n"); + if (width + 3 < 8) + size = size * 2; + } + + return size; +} + + +static int radeon_compressed_num_bytes(GLuint mesaFormat) +{ + int bytes = 0; + switch(mesaFormat) { + + case MESA_FORMAT_RGB_FXT1: + case MESA_FORMAT_RGBA_FXT1: + case MESA_FORMAT_RGB_DXT1: + case MESA_FORMAT_RGBA_DXT1: + bytes = 2; + break; + + case MESA_FORMAT_RGBA_DXT3: + case MESA_FORMAT_RGBA_DXT5: + bytes = 4; + default: + break; + } + + return bytes; +} + +/** + * Compute sizes and fill in offset and blit information for the given + * image (determined by \p face and \p level). + * + * \param curOffset points to the offset at which the image is to be stored + * and is updated by this function according to the size of the image. + */ +static void compute_tex_image_offset(radeonContextPtr rmesa, radeon_mipmap_tree *mt, + GLuint face, GLuint level, GLuint* curOffset) +{ + radeon_mipmap_level *lvl = &mt->levels[level]; + uint32_t row_align; + + /* Find image size in bytes */ + if (mt->compressed) { + /* TODO: Is this correct? Need test cases for compressed textures! */ + row_align = rmesa->texture_compressed_row_align - 1; + lvl->rowstride = (lvl->width * mt->bpp + row_align) & ~row_align; + lvl->size = radeon_compressed_texture_size(mt->radeon->glCtx, + lvl->width, lvl->height, lvl->depth, mt->compressed); + } else if (mt->target == GL_TEXTURE_RECTANGLE_NV) { + row_align = rmesa->texture_rect_row_align - 1; + lvl->rowstride = (lvl->width * mt->bpp + row_align) & ~row_align; + lvl->size = lvl->rowstride * lvl->height; + } else if (mt->tilebits & RADEON_TXO_MICRO_TILE) { + /* tile pattern is 16 bytes x2. mipmaps stay 32 byte aligned, + * though the actual offset may be different (if texture is less than + * 32 bytes width) to the untiled case */ + lvl->rowstride = (lvl->width * mt->bpp * 2 + 31) & ~31; + lvl->size = lvl->rowstride * ((lvl->height + 1) / 2) * lvl->depth; + } else { + row_align = rmesa->texture_row_align - 1; + lvl->rowstride = (lvl->width * mt->bpp + row_align) & ~row_align; + lvl->size = lvl->rowstride * lvl->height * lvl->depth; + } + assert(lvl->size > 0); + + /* All images are aligned to a 32-byte offset */ + *curOffset = (*curOffset + 0x1f) & ~0x1f; + lvl->faces[face].offset = *curOffset; + *curOffset += lvl->size; + + if (RADEON_DEBUG & RADEON_TEXTURE) + fprintf(stderr, + "level %d, face %d: rs:%d %dx%d at %d\n", + level, face, lvl->rowstride, lvl->width, lvl->height, lvl->faces[face].offset); +} + +static GLuint minify(GLuint size, GLuint levels) +{ + size = size >> levels; + if (size < 1) + size = 1; + return size; +} + + +static void calculate_miptree_layout_r100(radeonContextPtr rmesa, radeon_mipmap_tree *mt) +{ + GLuint curOffset; + GLuint numLevels; + GLuint i; + GLuint face; + + numLevels = mt->lastLevel - mt->firstLevel + 1; + assert(numLevels <= rmesa->glCtx->Const.MaxTextureLevels); + + curOffset = 0; + for(face = 0; face < mt->faces; face++) { + + for(i = 0; i < numLevels; i++) { + mt->levels[i].width = minify(mt->width0, i); + mt->levels[i].height = minify(mt->height0, i); + mt->levels[i].depth = minify(mt->depth0, i); + compute_tex_image_offset(rmesa, mt, face, i, &curOffset); + } + } + + /* Note the required size in memory */ + mt->totalsize = (curOffset + RADEON_OFFSET_MASK) & ~RADEON_OFFSET_MASK; +} + +static void calculate_miptree_layout_r300(radeonContextPtr rmesa, radeon_mipmap_tree *mt) +{ + GLuint curOffset; + GLuint numLevels; + GLuint i; + + numLevels = mt->lastLevel - mt->firstLevel + 1; + assert(numLevels <= rmesa->glCtx->Const.MaxTextureLevels); + + curOffset = 0; + for(i = 0; i < numLevels; i++) { + GLuint face; + + mt->levels[i].width = minify(mt->width0, i); + mt->levels[i].height = minify(mt->height0, i); + mt->levels[i].depth = minify(mt->depth0, i); + + for(face = 0; face < mt->faces; face++) + compute_tex_image_offset(rmesa, mt, face, i, &curOffset); + } + + /* Note the required size in memory */ + mt->totalsize = (curOffset + RADEON_OFFSET_MASK) & ~RADEON_OFFSET_MASK; +} + +/** + * Create a new mipmap tree, calculate its layout and allocate memory. + */ +radeon_mipmap_tree* radeon_miptree_create(radeonContextPtr rmesa, radeonTexObj *t, + GLenum target, GLenum internal_format, GLuint firstLevel, GLuint lastLevel, + GLuint width0, GLuint height0, GLuint depth0, + GLuint bpp, GLuint tilebits, GLuint compressed) +{ + radeon_mipmap_tree *mt = CALLOC_STRUCT(_radeon_mipmap_tree); + + mt->radeon = rmesa; + mt->internal_format = internal_format; + mt->refcount = 1; + mt->t = t; + mt->target = target; + mt->faces = (target == GL_TEXTURE_CUBE_MAP) ? 6 : 1; + mt->firstLevel = firstLevel; + mt->lastLevel = lastLevel; + mt->width0 = width0; + mt->height0 = height0; + mt->depth0 = depth0; + mt->bpp = compressed ? radeon_compressed_num_bytes(compressed) : bpp; + mt->tilebits = tilebits; + mt->compressed = compressed; + + if (rmesa->radeonScreen->chip_family >= CHIP_FAMILY_R300) + calculate_miptree_layout_r300(rmesa, mt); + else + calculate_miptree_layout_r100(rmesa, mt); + + mt->bo = radeon_bo_open(rmesa->radeonScreen->bom, + 0, mt->totalsize, 1024, + RADEON_GEM_DOMAIN_VRAM, + 0); + + return mt; +} + +void radeon_miptree_reference(radeon_mipmap_tree *mt) +{ + mt->refcount++; + assert(mt->refcount > 0); +} + +void radeon_miptree_unreference(radeon_mipmap_tree *mt) +{ + if (!mt) + return; + + assert(mt->refcount > 0); + mt->refcount--; + if (!mt->refcount) { + radeon_bo_unref(mt->bo); + free(mt); + } +} + + +/** + * Calculate first and last mip levels for the given texture object, + * where the dimensions are taken from the given texture image at + * the given level. + * + * Note: level is the OpenGL level number, which is not necessarily the same + * as the first level that is actually present. + * + * The base level image of the given texture face must be non-null, + * or this will fail. + */ +static void calculate_first_last_level(struct gl_texture_object *tObj, + GLuint *pfirstLevel, GLuint *plastLevel, + GLuint face, GLuint level) +{ + const struct gl_texture_image * const baseImage = + tObj->Image[face][level]; + + assert(baseImage); + + /* These must be signed values. MinLod and MaxLod can be negative numbers, + * and having firstLevel and lastLevel as signed prevents the need for + * extra sign checks. + */ + int firstLevel; + int lastLevel; + + /* Yes, this looks overly complicated, but it's all needed. + */ + switch (tObj->Target) { + case GL_TEXTURE_1D: + case GL_TEXTURE_2D: + case GL_TEXTURE_3D: + case GL_TEXTURE_CUBE_MAP: + if (tObj->MinFilter == GL_NEAREST || tObj->MinFilter == GL_LINEAR) { + /* GL_NEAREST and GL_LINEAR only care about GL_TEXTURE_BASE_LEVEL. + */ + firstLevel = lastLevel = tObj->BaseLevel; + } else { + firstLevel = tObj->BaseLevel + (GLint)(tObj->MinLod + 0.5); + firstLevel = MAX2(firstLevel, tObj->BaseLevel); + firstLevel = MIN2(firstLevel, level + baseImage->MaxLog2); + lastLevel = tObj->BaseLevel + (GLint)(tObj->MaxLod + 0.5); + lastLevel = MAX2(lastLevel, tObj->BaseLevel); + lastLevel = MIN2(lastLevel, level + baseImage->MaxLog2); + lastLevel = MIN2(lastLevel, tObj->MaxLevel); + lastLevel = MAX2(firstLevel, lastLevel); /* need at least one level */ + } + break; + case GL_TEXTURE_RECTANGLE_NV: + case GL_TEXTURE_4D_SGIS: + firstLevel = lastLevel = 0; + break; + default: + return; + } + + /* save these values */ + *pfirstLevel = firstLevel; + *plastLevel = lastLevel; +} + + +/** + * Checks whether the given miptree can hold the given texture image at the + * given face and level. + */ +GLboolean radeon_miptree_matches_image(radeon_mipmap_tree *mt, + struct gl_texture_image *texImage, GLuint face, GLuint level) +{ + radeon_mipmap_level *lvl; + + if (face >= mt->faces || level < mt->firstLevel || level > mt->lastLevel) + return GL_FALSE; + + if (texImage->InternalFormat != mt->internal_format || + texImage->IsCompressed != mt->compressed) + return GL_FALSE; + + if (!texImage->IsCompressed && + !mt->compressed && + texImage->TexFormat->TexelBytes != mt->bpp) + return GL_FALSE; + + lvl = &mt->levels[level - mt->firstLevel]; + if (lvl->width != texImage->Width || + lvl->height != texImage->Height || + lvl->depth != texImage->Depth) + return GL_FALSE; + + return GL_TRUE; +} + + +/** + * Checks whether the given miptree has the right format to store the given texture object. + */ +GLboolean radeon_miptree_matches_texture(radeon_mipmap_tree *mt, struct gl_texture_object *texObj) +{ + struct gl_texture_image *firstImage; + GLuint compressed; + GLuint numfaces = 1; + GLuint firstLevel, lastLevel; + + calculate_first_last_level(texObj, &firstLevel, &lastLevel, 0, texObj->BaseLevel); + if (texObj->Target == GL_TEXTURE_CUBE_MAP) + numfaces = 6; + + firstImage = texObj->Image[0][firstLevel]; + compressed = firstImage->IsCompressed ? firstImage->TexFormat->MesaFormat : 0; + + return (mt->firstLevel == firstLevel && + mt->lastLevel == lastLevel && + mt->width0 == firstImage->Width && + mt->height0 == firstImage->Height && + mt->depth0 == firstImage->Depth && + mt->compressed == compressed && + (!mt->compressed ? (mt->bpp == firstImage->TexFormat->TexelBytes) : 1)); +} + + +/** + * Try to allocate a mipmap tree for the given texture that will fit the + * given image in the given position. + */ +void radeon_try_alloc_miptree(radeonContextPtr rmesa, radeonTexObj *t, + radeon_texture_image *image, GLuint face, GLuint level) +{ + GLuint compressed = image->base.IsCompressed ? image->base.TexFormat->MesaFormat : 0; + GLuint numfaces = 1; + GLuint firstLevel, lastLevel; + + assert(!t->mt); + + calculate_first_last_level(&t->base, &firstLevel, &lastLevel, face, level); + if (t->base.Target == GL_TEXTURE_CUBE_MAP) + numfaces = 6; + + if (level != firstLevel || face >= numfaces) + return; + + t->mt = radeon_miptree_create(rmesa, t, t->base.Target, + image->base.InternalFormat, + firstLevel, lastLevel, + image->base.Width, image->base.Height, image->base.Depth, + image->base.TexFormat->TexelBytes, t->tile_bits, compressed); +} + +/* Although we use the image_offset[] array to store relative offsets + * to cube faces, Mesa doesn't know anything about this and expects + * each cube face to be treated as a separate image. + * + * These functions present that view to mesa: + */ +void +radeon_miptree_depth_offsets(radeon_mipmap_tree *mt, GLuint level, GLuint *offsets) +{ + if (mt->target != GL_TEXTURE_3D || mt->faces == 1) + offsets[0] = 0; + else { + int i; + for (i = 0; i < 6; i++) + offsets[i] = mt->levels[level].faces[i].offset; + } +} + +GLuint +radeon_miptree_image_offset(radeon_mipmap_tree *mt, + GLuint face, GLuint level) +{ + if (mt->target == GL_TEXTURE_CUBE_MAP_ARB) + return (mt->levels[level].faces[face].offset); + else + return mt->levels[level].faces[0].offset; +} diff --git a/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.h b/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.h new file mode 100644 index 00000000000..db28252da37 --- /dev/null +++ b/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.h @@ -0,0 +1,101 @@ +/* + * Copyright (C) 2008 Nicolai Haehnle. + * + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sublicense, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial + * portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + */ + +#ifndef __RADEON_MIPMAP_TREE_H_ +#define __RADEON_MIPMAP_TREE_H_ + +#include "radeon_common.h" + +typedef struct _radeon_mipmap_tree radeon_mipmap_tree; +typedef struct _radeon_mipmap_level radeon_mipmap_level; +typedef struct _radeon_mipmap_image radeon_mipmap_image; + +struct _radeon_mipmap_image { + GLuint offset; /** Offset of this image from the start of mipmap tree buffer, in bytes */ +}; + +struct _radeon_mipmap_level { + GLuint width; + GLuint height; + GLuint depth; + GLuint size; /** Size of each image, in bytes */ + GLuint rowstride; /** in bytes */ + radeon_mipmap_image faces[6]; +}; + +/* store the max possible in the miptree */ +#define RADEON_MIPTREE_MAX_TEXTURE_LEVELS 13 + +/** + * A mipmap tree contains texture images in the layout that the hardware + * expects. + * + * The meta-data of mipmap trees is immutable, i.e. you cannot change the + * layout on-the-fly; however, the texture contents (i.e. texels) can be + * changed. + */ +struct _radeon_mipmap_tree { + radeonContextPtr radeon; + radeonTexObj *t; + struct radeon_bo *bo; + GLuint refcount; + + GLuint totalsize; /** total size of the miptree, in bytes */ + + GLenum target; /** GL_TEXTURE_xxx */ + GLenum internal_format; + GLuint faces; /** # of faces: 6 for cubemaps, 1 otherwise */ + GLuint firstLevel; /** First mip level stored in this mipmap tree */ + GLuint lastLevel; /** Last mip level stored in this mipmap tree */ + + GLuint width0; /** Width of firstLevel image */ + GLuint height0; /** Height of firstLevel image */ + GLuint depth0; /** Depth of firstLevel image */ + + GLuint bpp; /** Bytes per texel */ + GLuint tilebits; /** RADEON_TXO_xxx_TILE */ + GLuint compressed; /** MESA_FORMAT_xxx indicating a compressed format, or 0 if uncompressed */ + + radeon_mipmap_level levels[RADEON_MIPTREE_MAX_TEXTURE_LEVELS]; +}; + +radeon_mipmap_tree* radeon_miptree_create(radeonContextPtr rmesa, radeonTexObj *t, + GLenum target, GLenum internal_format, GLuint firstLevel, GLuint lastLevel, + GLuint width0, GLuint height0, GLuint depth0, + GLuint bpp, GLuint tilebits, GLuint compressed); +void radeon_miptree_reference(radeon_mipmap_tree *mt); +void radeon_miptree_unreference(radeon_mipmap_tree *mt); + +GLboolean radeon_miptree_matches_image(radeon_mipmap_tree *mt, + struct gl_texture_image *texImage, GLuint face, GLuint level); +GLboolean radeon_miptree_matches_texture(radeon_mipmap_tree *mt, struct gl_texture_object *texObj); +void radeon_try_alloc_miptree(radeonContextPtr rmesa, radeonTexObj *t, + radeon_texture_image *texImage, GLuint face, GLuint level); +GLuint radeon_miptree_image_offset(radeon_mipmap_tree *mt, + GLuint face, GLuint level); +void radeon_miptree_depth_offsets(radeon_mipmap_tree *mt, GLuint level, GLuint *offsets); +#endif /* __RADEON_MIPMAP_TREE_H_ */ diff --git a/src/mesa/drivers/dri/radeon/radeon_queryobj.c b/src/mesa/drivers/dri/radeon/radeon_queryobj.c new file mode 100644 index 00000000000..b79d864ba29 --- /dev/null +++ b/src/mesa/drivers/dri/radeon/radeon_queryobj.c @@ -0,0 +1,235 @@ +/* + * Copyright © 2008-2009 Maciej Cencora <m.cencora@gmail.com> + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS + * IN THE SOFTWARE. + * + * Authors: + * Maciej Cencora <m.cencora@gmail.com> + * + */ +#include "radeon_common.h" +#include "radeon_queryobj.h" +#include "radeon_debug.h" + +#include "main/imports.h" +#include "main/simple_list.h" + +static int radeonQueryIsFlushed(GLcontext *ctx, struct gl_query_object *q) +{ + radeonContextPtr radeon = RADEON_CONTEXT(ctx); + struct radeon_query_object *tmp, *query = (struct radeon_query_object *)q; + + foreach(tmp, &radeon->query.not_flushed_head) { + if (tmp == query) { + return 0; + } + } + + return 1; +} + +static void radeonQueryGetResult(GLcontext *ctx, struct gl_query_object *q) +{ + struct radeon_query_object *query = (struct radeon_query_object *)q; + uint32_t *result; + int i; + + radeon_print(RADEON_STATE, RADEON_VERBOSE, + "%s: query id %d, result %d\n", + __FUNCTION__, query->Base.Id, (int) query->Base.Result); + + radeon_bo_map(query->bo, GL_FALSE); + + result = query->bo->ptr; + + query->Base.Result = 0; + for (i = 0; i < query->curr_offset/sizeof(uint32_t); ++i) { + query->Base.Result += result[i]; + radeon_print(RADEON_STATE, RADEON_TRACE, "result[%d] = %d\n", i, result[i]); + } + + radeon_bo_unmap(query->bo); +} + +static struct gl_query_object * radeonNewQueryObject(GLcontext *ctx, GLuint id) +{ + struct radeon_query_object *query; + + query = _mesa_calloc(sizeof(struct radeon_query_object)); + + query->Base.Id = id; + query->Base.Result = 0; + query->Base.Active = GL_FALSE; + query->Base.Ready = GL_TRUE; + + radeon_print(RADEON_STATE, RADEON_VERBOSE,"%s: query id %d\n", __FUNCTION__, query->Base.Id); + + return &query->Base; +} + +static void radeonDeleteQuery(GLcontext *ctx, struct gl_query_object *q) +{ + struct radeon_query_object *query = (struct radeon_query_object *)q; + + radeon_print(RADEON_STATE, RADEON_NORMAL, "%s: query id %d\n", __FUNCTION__, q->Id); + + if (query->bo) { + radeon_bo_unref(query->bo); + } + + _mesa_free(query); +} + +static void radeonWaitQuery(GLcontext *ctx, struct gl_query_object *q) +{ + struct radeon_query_object *query = (struct radeon_query_object *)q; + + /* If the cmdbuf with packets for this query hasn't been flushed yet, do it now */ + if (!radeonQueryIsFlushed(ctx, q)) + ctx->Driver.Flush(ctx); + + radeon_print(RADEON_STATE, RADEON_VERBOSE, "%s: query id %d, bo %p, offset %d\n", __FUNCTION__, q->Id, query->bo, query->curr_offset); + + radeonQueryGetResult(ctx, q); + + query->Base.Ready = GL_TRUE; +} + + +static void radeonBeginQuery(GLcontext *ctx, struct gl_query_object *q) +{ + radeonContextPtr radeon = RADEON_CONTEXT(ctx); + struct radeon_query_object *query = (struct radeon_query_object *)q; + + radeon_print(RADEON_STATE, RADEON_NORMAL, "%s: query id %d\n", __FUNCTION__, q->Id); + + assert(radeon->query.current == NULL); + + if (radeon->dma.flush) + radeon->dma.flush(radeon->glCtx); + + if (!query->bo) { + query->bo = radeon_bo_open(radeon->radeonScreen->bom, 0, RADEON_QUERY_PAGE_SIZE, RADEON_QUERY_PAGE_SIZE, RADEON_GEM_DOMAIN_GTT, 0); + } + query->curr_offset = 0; + + radeon->query.current = query; + + radeon->query.queryobj.dirty = GL_TRUE; + radeon->hw.is_dirty = GL_TRUE; + insert_at_tail(&radeon->query.not_flushed_head, query); + +} + +void radeonEmitQueryEnd(GLcontext *ctx) +{ + radeonContextPtr radeon = RADEON_CONTEXT(ctx); + struct radeon_query_object *query = radeon->query.current; + + if (!query) + return; + + if (query->emitted_begin == GL_FALSE) + return; + + radeon_print(RADEON_STATE, RADEON_NORMAL, "%s: query id %d, bo %p, offset %d\n", __FUNCTION__, query->Base.Id, query->bo, query->curr_offset); + + radeon_cs_space_check_with_bo(radeon->cmdbuf.cs, + query->bo, + 0, RADEON_GEM_DOMAIN_GTT); + + radeon->vtbl.emit_query_finish(radeon); +} + +static void radeonEndQuery(GLcontext *ctx, struct gl_query_object *q) +{ + radeonContextPtr radeon = RADEON_CONTEXT(ctx); + + radeon_print(RADEON_STATE, RADEON_NORMAL, "%s: query id %d\n", __FUNCTION__, q->Id); + + if (radeon->dma.flush) + radeon->dma.flush(radeon->glCtx); + radeonEmitQueryEnd(ctx); + + radeon->query.current = NULL; +} + +static void radeonCheckQuery(GLcontext *ctx, struct gl_query_object *q) +{ + radeon_print(RADEON_STATE, RADEON_TRACE, "%s: query id %d\n", __FUNCTION__, q->Id); + +#ifdef DRM_RADEON_GEM_BUSY + radeonContextPtr radeon = RADEON_CONTEXT(ctx); + + if (radeon->radeonScreen->kernel_mm) { + struct radeon_query_object *query = (struct radeon_query_object *)q; + uint32_t domain; + + /* Need to perform a flush, as per ARB_occlusion_query spec */ + if (!radeonQueryIsFlushed(ctx, q)) { + ctx->Driver.Flush(ctx); + } + + if (radeon_bo_is_busy(query->bo, &domain) == 0) { + radeonQueryGetResult(ctx, q); + query->Base.Ready = GL_TRUE; + } + } else { + radeonWaitQuery(ctx, q); + } +#else + radeonWaitQuery(ctx, q); +#endif +} + +void radeonInitQueryObjFunctions(struct dd_function_table *functions) +{ + functions->NewQueryObject = radeonNewQueryObject; + functions->DeleteQuery = radeonDeleteQuery; + functions->BeginQuery = radeonBeginQuery; + functions->EndQuery = radeonEndQuery; + functions->CheckQuery = radeonCheckQuery; + functions->WaitQuery = radeonWaitQuery; +} + +int radeon_check_query_active(GLcontext *ctx, struct radeon_state_atom *atom) +{ + radeonContextPtr radeon = RADEON_CONTEXT(ctx); + struct radeon_query_object *query = radeon->query.current; + + if (!query || query->emitted_begin) + return 0; + return atom->cmd_size; +} + +void radeon_emit_queryobj(GLcontext *ctx, struct radeon_state_atom *atom) +{ + radeonContextPtr radeon = RADEON_CONTEXT(ctx); + BATCH_LOCALS(radeon); + int dwords; + + dwords = (*atom->check) (ctx, atom); + + BEGIN_BATCH_NO_AUTOSTATE(dwords); + OUT_BATCH_TABLE(atom->cmd, dwords); + END_BATCH(); + + radeon->query.current->emitted_begin = GL_TRUE; +} diff --git a/src/mesa/drivers/dri/radeon/radeon_queryobj.h b/src/mesa/drivers/dri/radeon/radeon_queryobj.h new file mode 100644 index 00000000000..19374dc76b7 --- /dev/null +++ b/src/mesa/drivers/dri/radeon/radeon_queryobj.h @@ -0,0 +1,55 @@ +/* + * Copyright © 2008 Maciej Cencora <m.cencora@gmail.com> + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS + * IN THE SOFTWARE. + * + * Authors: + * Maciej Cencora <m.cencora@gmail.com> + * + */ + +#include "main/imports.h" +#include "main/simple_list.h" +#include "radeon_common_context.h" + +extern void radeonEmitQueryBegin(GLcontext *ctx); +extern void radeonEmitQueryEnd(GLcontext *ctx); + +extern void radeonInitQueryObjFunctions(struct dd_function_table *functions); + +#define RADEON_QUERY_PAGE_SIZE 4096 + +int radeon_check_query_active(GLcontext *ctx, struct radeon_state_atom *atom); +void radeon_emit_queryobj(GLcontext *ctx, struct radeon_state_atom *atom); + +static inline void radeon_init_query_stateobj(radeonContextPtr radeon, int SZ) +{ + radeon->query.queryobj.cmd_size = (SZ); + radeon->query.queryobj.cmd = (uint32_t*)CALLOC((SZ) * sizeof(uint32_t)); + radeon->query.queryobj.name = "queryobj"; + radeon->query.queryobj.idx = 0; + radeon->query.queryobj.check = radeon_check_query_active; + radeon->query.queryobj.dirty = GL_FALSE; + radeon->query.queryobj.emit = radeon_emit_queryobj; + + radeon->hw.max_state_size += (SZ); + insert_at_tail(&radeon->hw.atomlist, &radeon->query.queryobj); +} + diff --git a/src/mesa/drivers/dri/radeon/radeon_sanity.c b/src/mesa/drivers/dri/radeon/radeon_sanity.c index 6613757fcea..1ab570f5071 100644 --- a/src/mesa/drivers/dri/radeon/radeon_sanity.c +++ b/src/mesa/drivers/dri/radeon/radeon_sanity.c @@ -44,11 +44,11 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. #define MORE_VERBOSE 1 #if MORE_VERBOSE -#define VERBOSE (RADEON_DEBUG & DEBUG_VERBOSE) +#define VERBOSE (RADEON_DEBUG & RADEON_VERBOSE) #define NORMAL (1) #else #define VERBOSE 0 -#define NORMAL (RADEON_DEBUG & DEBUG_VERBOSE) +#define NORMAL (RADEON_DEBUG & RADEON_VERBOSE) #endif @@ -973,7 +973,7 @@ static int radeon_emit_packet3_cliprect( drm_radeon_cmd_buffer_t *cmdbuf ) } -int radeonSanityCmdBuffer( radeonContextPtr rmesa, +int radeonSanityCmdBuffer( r100ContextPtr rmesa, int nbox, drm_clip_rect_t *boxes ) { diff --git a/src/mesa/drivers/dri/radeon/radeon_sanity.h b/src/mesa/drivers/dri/radeon/radeon_sanity.h index 1ec06bc586b..f30eb1c4f15 100644 --- a/src/mesa/drivers/dri/radeon/radeon_sanity.h +++ b/src/mesa/drivers/dri/radeon/radeon_sanity.h @@ -1,7 +1,7 @@ #ifndef RADEON_SANITY_H #define RADEON_SANITY_H -extern int radeonSanityCmdBuffer( radeonContextPtr rmesa, +extern int radeonSanityCmdBuffer( r100ContextPtr rmesa, int nbox, drm_clip_rect_t *boxes ); diff --git a/src/mesa/drivers/dri/radeon/radeon_screen.c b/src/mesa/drivers/dri/radeon/radeon_screen.c index a977bedbc28..5ffb55db5ef 100644 --- a/src/mesa/drivers/dri/radeon/radeon_screen.c +++ b/src/mesa/drivers/dri/radeon/radeon_screen.c @@ -35,6 +35,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. * \author Gareth Hughes <gareth@valinux.com> */ +#include <errno.h> #include "main/glheader.h" #include "main/imports.h" #include "main/mtypes.h" @@ -45,32 +46,42 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "radeon_chipset.h" #include "radeon_macros.h" #include "radeon_screen.h" +#include "radeon_common.h" +#include "radeon_span.h" #if !RADEON_COMMON #include "radeon_context.h" -#include "radeon_span.h" #include "radeon_tex.h" #elif RADEON_COMMON && defined(RADEON_COMMON_FOR_R200) #include "r200_context.h" #include "r200_ioctl.h" -#include "r200_span.h" #include "r200_tex.h" #elif RADEON_COMMON && defined(RADEON_COMMON_FOR_R300) #include "r300_context.h" -#include "r300_fragprog.h" #include "r300_tex.h" -#include "radeon_span.h" +#elif RADEON_COMMON && defined(RADEON_COMMON_FOR_R600) +#include "r600_context.h" +#include "r700_driconf.h" /* +r6/r7 */ +#include "r600_tex.h" /* +r6/r7 */ #endif #include "utils.h" #include "vblank.h" #include "drirenderbuffer.h" +#include "radeon_bocs_wrapper.h" + #include "GL/internal/dri_interface.h" /* Radeon configuration */ #include "xmlpool.h" +#define DRI_CONF_COMMAND_BUFFER_SIZE(def,min,max) \ +DRI_CONF_OPT_BEGIN_V(command_buffer_size,int,def, # min ":" # max ) \ + DRI_CONF_DESC(en,"Size of command buffer (in KB)") \ + DRI_CONF_DESC(de,"Grösse des Befehlspuffers (in KB)") \ +DRI_CONF_OPT_END + #if !RADEON_COMMON /* R100 */ PUBLIC const char __driConfigOptions[] = DRI_CONF_BEGIN @@ -80,6 +91,7 @@ DRI_CONF_BEGIN DRI_CONF_VBLANK_MODE(DRI_CONF_VBLANK_DEF_INTERVAL_0) DRI_CONF_MAX_TEXTURE_UNITS(3,2,3) DRI_CONF_HYPERZ(false) + DRI_CONF_COMMAND_BUFFER_SIZE(8, 8, 32) DRI_CONF_SECTION_END DRI_CONF_SECTION_QUALITY DRI_CONF_TEXTURE_DEPTH(DRI_CONF_TEXTURE_DEPTH_FB) @@ -95,7 +107,7 @@ DRI_CONF_BEGIN DRI_CONF_NO_RAST(false) DRI_CONF_SECTION_END DRI_CONF_END; -static const GLuint __driNConfigOptions = 14; +static const GLuint __driNConfigOptions = 15; #elif RADEON_COMMON && defined(RADEON_COMMON_FOR_R200) @@ -107,6 +119,7 @@ DRI_CONF_BEGIN DRI_CONF_VBLANK_MODE(DRI_CONF_VBLANK_DEF_INTERVAL_0) DRI_CONF_MAX_TEXTURE_UNITS(6,2,6) DRI_CONF_HYPERZ(false) + DRI_CONF_COMMAND_BUFFER_SIZE(8, 8, 32) DRI_CONF_SECTION_END DRI_CONF_SECTION_QUALITY DRI_CONF_TEXTURE_DEPTH(DRI_CONF_TEXTURE_DEPTH_FB) @@ -126,7 +139,7 @@ DRI_CONF_BEGIN DRI_CONF_NV_VERTEX_PROGRAM(false) DRI_CONF_SECTION_END DRI_CONF_END; -static const GLuint __driNConfigOptions = 16; +static const GLuint __driNConfigOptions = 17; extern const struct dri_extension blend_extensions[]; extern const struct dri_extension ARB_vp_extension[]; @@ -134,7 +147,10 @@ extern const struct dri_extension NV_vp_extension[]; extern const struct dri_extension ATI_fs_extension[]; extern const struct dri_extension point_extensions[]; -#elif RADEON_COMMON && defined(RADEON_COMMON_FOR_R300) +#elif RADEON_COMMON && (defined(RADEON_COMMON_FOR_R300) || defined(RADEON_COMMON_FOR_R600)) + +#define DRI_CONF_FP_OPTIMIZATION_SPEED 0 +#define DRI_CONF_FP_OPTIMIZATION_QUALITY 1 /* TODO: integrate these into xmlpool.h! */ #define DRI_CONF_MAX_TEXTURE_IMAGE_UNITS(def,min,max) \ @@ -149,11 +165,7 @@ DRI_CONF_OPT_BEGIN_V(texture_coord_units,int,def, # min ":" # max ) \ DRI_CONF_DESC(de,"Anzahl der Texturkoordinateneinheiten") \ DRI_CONF_OPT_END -#define DRI_CONF_COMMAND_BUFFER_SIZE(def,min,max) \ -DRI_CONF_OPT_BEGIN_V(command_buffer_size,int,def, # min ":" # max ) \ - DRI_CONF_DESC(en,"Size of command buffer (in KB)") \ - DRI_CONF_DESC(de,"Grösse des Befehlspuffers (in KB)") \ -DRI_CONF_OPT_END + #define DRI_CONF_DISABLE_S3TC(def) \ DRI_CONF_OPT_BEGIN(disable_s3tc,bool,def) \ @@ -208,47 +220,42 @@ static const GLuint __driNConfigOptions = 17; extern const struct dri_extension gl_20_extension[]; -#ifndef RADEON_DEBUG -int RADEON_DEBUG = 0; - -static const struct dri_debug_control debug_control[] = { - {"fall", DEBUG_FALLBACKS}, - {"tex", DEBUG_TEXTURE}, - {"ioctl", DEBUG_IOCTL}, - {"prim", DEBUG_PRIMS}, - {"vert", DEBUG_VERTS}, - {"state", DEBUG_STATE}, - {"code", DEBUG_CODEGEN}, - {"vfmt", DEBUG_VFMT}, - {"vtxf", DEBUG_VFMT}, - {"verb", DEBUG_VERBOSE}, - {"dri", DEBUG_DRI}, - {"dma", DEBUG_DMA}, - {"san", DEBUG_SANITY}, - {"sync", DEBUG_SYNC}, - {"pix", DEBUG_PIXEL}, - {"mem", DEBUG_MEMORY}, - {"allmsg", ~DEBUG_SYNC}, /* avoid the term "sync" because the parser uses strstr */ - {NULL, 0} -}; -#endif /* RADEON_DEBUG */ - #endif /* RADEON_COMMON && defined(RADEON_COMMON_FOR_R300) */ extern const struct dri_extension card_extensions[]; +extern const struct dri_extension mm_extensions[]; static int getSwapInfo( __DRIdrawablePrivate *dPriv, __DRIswapInfo * sInfo ); static int -radeonGetParam(int fd, int param, void *value) +radeonGetParam(__DRIscreenPrivate *sPriv, int param, void *value) { int ret; - drm_radeon_getparam_t gp; - - gp.param = param; - gp.value = value; + drm_radeon_getparam_t gp = { 0 }; + struct drm_radeon_info info = { 0 }; + + if (sPriv->drm_version.major >= 2) { + info.value = (uint64_t)(uintptr_t)value; + switch (param) { + case RADEON_PARAM_DEVICE_ID: + info.request = RADEON_INFO_DEVICE_ID; + break; + case RADEON_PARAM_NUM_GB_PIPES: + info.request = RADEON_INFO_NUM_GB_PIPES; + break; + case RADEON_PARAM_NUM_Z_PIPES: + info.request = RADEON_INFO_NUM_Z_PIPES; + break; + default: + return -EINVAL; + } + ret = drmCommandWriteRead(sPriv->fd, DRM_RADEON_INFO, &info, sizeof(info)); + } else { + gp.param = param; + gp.value = value; - ret = drmCommandWriteRead( fd, DRM_RADEON_GETPARAM, &gp, sizeof(gp)); + ret = drmCommandWriteRead(sPriv->fd, DRM_RADEON_GETPARAM, &gp, sizeof(gp)); + } return ret; } @@ -335,6 +342,12 @@ static const __DRItexOffsetExtension radeonTexOffsetExtension = { { __DRI_TEX_OFFSET, __DRI_TEX_OFFSET_VERSION }, radeonSetTexOffset, }; + +static const __DRItexBufferExtension radeonTexBufferExtension = { + { __DRI_TEX_BUFFER, __DRI_TEX_BUFFER_VERSION }, + radeonSetTexBuffer, + radeonSetTexBuffer2, +}; #endif #if RADEON_COMMON && defined(RADEON_COMMON_FOR_R200) @@ -349,6 +362,12 @@ static const __DRItexOffsetExtension r200texOffsetExtension = { { __DRI_TEX_OFFSET, __DRI_TEX_OFFSET_VERSION }, r200SetTexOffset, }; + +static const __DRItexBufferExtension r200TexBufferExtension = { + { __DRI_TEX_BUFFER, __DRI_TEX_BUFFER_VERSION }, + r200SetTexBuffer, + r200SetTexBuffer2, +}; #endif #if RADEON_COMMON && defined(RADEON_COMMON_FOR_R300) @@ -356,137 +375,32 @@ static const __DRItexOffsetExtension r300texOffsetExtension = { { __DRI_TEX_OFFSET, __DRI_TEX_OFFSET_VERSION }, r300SetTexOffset, }; -#endif -/* Create the device specific screen private data struct. - */ -static radeonScreenPtr -radeonCreateScreen( __DRIscreenPrivate *sPriv ) -{ - radeonScreenPtr screen; - RADEONDRIPtr dri_priv = (RADEONDRIPtr)sPriv->pDevPriv; - unsigned char *RADEONMMIO; - int i; - int ret; - uint32_t temp = 0; - - if (sPriv->devPrivSize != sizeof(RADEONDRIRec)) { - fprintf(stderr,"\nERROR! sizeof(RADEONDRIRec) does not match passed size from device driver\n"); - return GL_FALSE; - } - - /* Allocate the private area */ - screen = (radeonScreenPtr) CALLOC( sizeof(*screen) ); - if ( !screen ) { - __driUtilMessage("%s: Could not allocate memory for screen structure", - __FUNCTION__); - return NULL; - } - -#if DO_DEBUG && RADEON_COMMON && defined(RADEON_COMMON_FOR_R300) - RADEON_DEBUG = driParseDebugString(getenv("RADEON_DEBUG"), debug_control); +static const __DRItexBufferExtension r300TexBufferExtension = { + { __DRI_TEX_BUFFER, __DRI_TEX_BUFFER_VERSION }, + r300SetTexBuffer, + r300SetTexBuffer2, +}; #endif - /* parse information in __driConfigOptions */ - driParseOptionInfo (&screen->optionCache, - __driConfigOptions, __driNConfigOptions); - - /* This is first since which regions we map depends on whether or - * not we are using a PCI card. - */ - screen->card_type = (dri_priv->IsPCI ? RADEON_CARD_PCI : RADEON_CARD_AGP); - { - int ret; - ret = radeonGetParam( sPriv->fd, RADEON_PARAM_GART_BUFFER_OFFSET, - &screen->gart_buffer_offset); - - if (ret) { - FREE( screen ); - fprintf(stderr, "drm_radeon_getparam_t (RADEON_PARAM_GART_BUFFER_OFFSET): %d\n", ret); - return NULL; - } - - ret = radeonGetParam( sPriv->fd, RADEON_PARAM_GART_BASE, - &screen->gart_base); - if (ret) { - FREE( screen ); - fprintf(stderr, "drm_radeon_getparam_t (RADEON_PARAM_GART_BASE): %d\n", ret); - return NULL; - } - - ret = radeonGetParam( sPriv->fd, RADEON_PARAM_IRQ_NR, - &screen->irq); - if (ret) { - FREE( screen ); - fprintf(stderr, "drm_radeon_getparam_t (RADEON_PARAM_IRQ_NR): %d\n", ret); - return NULL; - } - screen->drmSupportsCubeMapsR200 = (sPriv->drm_version.minor >= 7); - screen->drmSupportsBlendColor = (sPriv->drm_version.minor >= 11); - screen->drmSupportsTriPerf = (sPriv->drm_version.minor >= 16); - screen->drmSupportsFragShader = (sPriv->drm_version.minor >= 18); - screen->drmSupportsPointSprites = (sPriv->drm_version.minor >= 13); - screen->drmSupportsCubeMapsR100 = (sPriv->drm_version.minor >= 15); - screen->drmSupportsVertexProgram = (sPriv->drm_version.minor >= 25); - } - - screen->mmio.handle = dri_priv->registerHandle; - screen->mmio.size = dri_priv->registerSize; - if ( drmMap( sPriv->fd, - screen->mmio.handle, - screen->mmio.size, - &screen->mmio.map ) ) { - FREE( screen ); - __driUtilMessage("%s: drmMap failed\n", __FUNCTION__ ); - return NULL; - } - - RADEONMMIO = screen->mmio.map; - - screen->status.handle = dri_priv->statusHandle; - screen->status.size = dri_priv->statusSize; - if ( drmMap( sPriv->fd, - screen->status.handle, - screen->status.size, - &screen->status.map ) ) { - drmUnmap( screen->mmio.map, screen->mmio.size ); - FREE( screen ); - __driUtilMessage("%s: drmMap (2) failed\n", __FUNCTION__ ); - return NULL; - } - screen->scratch = (__volatile__ uint32_t *) - ((GLubyte *)screen->status.map + RADEON_SCRATCH_REG_OFFSET); - - screen->buffers = drmMapBufs( sPriv->fd ); - if ( !screen->buffers ) { - drmUnmap( screen->status.map, screen->status.size ); - drmUnmap( screen->mmio.map, screen->mmio.size ); - FREE( screen ); - __driUtilMessage("%s: drmMapBufs failed\n", __FUNCTION__ ); - return NULL; - } - - if ( dri_priv->gartTexHandle && dri_priv->gartTexMapSize ) { - screen->gartTextures.handle = dri_priv->gartTexHandle; - screen->gartTextures.size = dri_priv->gartTexMapSize; - if ( drmMap( sPriv->fd, - screen->gartTextures.handle, - screen->gartTextures.size, - (drmAddressPtr)&screen->gartTextures.map ) ) { - drmUnmapBufs( screen->buffers ); - drmUnmap( screen->status.map, screen->status.size ); - drmUnmap( screen->mmio.map, screen->mmio.size ); - FREE( screen ); - __driUtilMessage("%s: drmMap failed for GART texture area\n", __FUNCTION__); - return NULL; - } +#if RADEON_COMMON && defined(RADEON_COMMON_FOR_R600) +static const __DRItexOffsetExtension r600texOffsetExtension = { + { __DRI_TEX_OFFSET, __DRI_TEX_OFFSET_VERSION }, + r600SetTexOffset, /* +r6/r7 */ +}; - screen->gart_texture_offset = dri_priv->gartTexOffset + screen->gart_base; - } +static const __DRItexBufferExtension r600TexBufferExtension = { + { __DRI_TEX_BUFFER, __DRI_TEX_BUFFER_VERSION }, + r600SetTexBuffer, /* +r6/r7 */ + r600SetTexBuffer2, /* +r6/r7 */ +}; +#endif +static int radeon_set_screen_flags(radeonScreenPtr screen, int device_id) +{ + screen->device_id = device_id; screen->chip_flags = 0; - /* XXX: add more chipsets */ - switch ( dri_priv->deviceID ) { + switch ( device_id ) { case PCI_CHIP_RADEON_LY: case PCI_CHIP_RADEON_LZ: case PCI_CHIP_RADEON_QY: @@ -561,7 +475,6 @@ radeonCreateScreen( __DRIscreenPrivate *sPriv ) screen->chip_family = CHIP_FAMILY_RS300; break; - case PCI_CHIP_R300_AD: case PCI_CHIP_R300_AE: case PCI_CHIP_R300_AF: @@ -819,11 +732,325 @@ radeonCreateScreen( __DRIscreenPrivate *sPriv ) screen->chip_flags = RADEON_CHIPSET_TCL; break; + case PCI_CHIP_R600_9400: + case PCI_CHIP_R600_9401: + case PCI_CHIP_R600_9402: + case PCI_CHIP_R600_9403: + case PCI_CHIP_R600_9405: + case PCI_CHIP_R600_940A: + case PCI_CHIP_R600_940B: + case PCI_CHIP_R600_940F: + screen->chip_family = CHIP_FAMILY_R600; + screen->chip_flags = RADEON_CHIPSET_TCL; + break; + + case PCI_CHIP_RV610_94C0: + case PCI_CHIP_RV610_94C1: + case PCI_CHIP_RV610_94C3: + case PCI_CHIP_RV610_94C4: + case PCI_CHIP_RV610_94C5: + case PCI_CHIP_RV610_94C6: + case PCI_CHIP_RV610_94C7: + case PCI_CHIP_RV610_94C8: + case PCI_CHIP_RV610_94C9: + case PCI_CHIP_RV610_94CB: + case PCI_CHIP_RV610_94CC: + case PCI_CHIP_RV610_94CD: + screen->chip_family = CHIP_FAMILY_RV610; + screen->chip_flags = RADEON_CHIPSET_TCL; + break; + + case PCI_CHIP_RV630_9580: + case PCI_CHIP_RV630_9581: + case PCI_CHIP_RV630_9583: + case PCI_CHIP_RV630_9586: + case PCI_CHIP_RV630_9587: + case PCI_CHIP_RV630_9588: + case PCI_CHIP_RV630_9589: + case PCI_CHIP_RV630_958A: + case PCI_CHIP_RV630_958B: + case PCI_CHIP_RV630_958C: + case PCI_CHIP_RV630_958D: + case PCI_CHIP_RV630_958E: + case PCI_CHIP_RV630_958F: + screen->chip_family = CHIP_FAMILY_RV630; + screen->chip_flags = RADEON_CHIPSET_TCL; + break; + + case PCI_CHIP_RV670_9500: + case PCI_CHIP_RV670_9501: + case PCI_CHIP_RV670_9504: + case PCI_CHIP_RV670_9505: + case PCI_CHIP_RV670_9506: + case PCI_CHIP_RV670_9507: + case PCI_CHIP_RV670_9508: + case PCI_CHIP_RV670_9509: + case PCI_CHIP_RV670_950F: + case PCI_CHIP_RV670_9511: + case PCI_CHIP_RV670_9515: + case PCI_CHIP_RV670_9517: + case PCI_CHIP_RV670_9519: + screen->chip_family = CHIP_FAMILY_RV670; + screen->chip_flags = RADEON_CHIPSET_TCL; + break; + + case PCI_CHIP_RV620_95C0: + case PCI_CHIP_RV620_95C2: + case PCI_CHIP_RV620_95C4: + case PCI_CHIP_RV620_95C5: + case PCI_CHIP_RV620_95C6: + case PCI_CHIP_RV620_95C7: + case PCI_CHIP_RV620_95C9: + case PCI_CHIP_RV620_95CC: + case PCI_CHIP_RV620_95CD: + case PCI_CHIP_RV620_95CE: + case PCI_CHIP_RV620_95CF: + screen->chip_family = CHIP_FAMILY_RV620; + screen->chip_flags = RADEON_CHIPSET_TCL; + break; + + case PCI_CHIP_RV635_9590: + case PCI_CHIP_RV635_9591: + case PCI_CHIP_RV635_9593: + case PCI_CHIP_RV635_9595: + case PCI_CHIP_RV635_9596: + case PCI_CHIP_RV635_9597: + case PCI_CHIP_RV635_9598: + case PCI_CHIP_RV635_9599: + case PCI_CHIP_RV635_959B: + screen->chip_family = CHIP_FAMILY_RV635; + screen->chip_flags = RADEON_CHIPSET_TCL; + break; + + case PCI_CHIP_RS780_9610: + case PCI_CHIP_RS780_9611: + case PCI_CHIP_RS780_9612: + case PCI_CHIP_RS780_9613: + case PCI_CHIP_RS780_9614: + case PCI_CHIP_RS780_9615: + case PCI_CHIP_RS780_9616: + screen->chip_family = CHIP_FAMILY_RS780; + screen->chip_flags = RADEON_CHIPSET_TCL; + break; + case PCI_CHIP_RS880_9710: + case PCI_CHIP_RS880_9711: + case PCI_CHIP_RS880_9712: + case PCI_CHIP_RS880_9713: + case PCI_CHIP_RS880_9714: + screen->chip_family = CHIP_FAMILY_RS880; + screen->chip_flags = RADEON_CHIPSET_TCL; + break; + + case PCI_CHIP_RV770_9440: + case PCI_CHIP_RV770_9441: + case PCI_CHIP_RV770_9442: + case PCI_CHIP_RV770_9443: + case PCI_CHIP_RV770_9444: + case PCI_CHIP_RV770_9446: + case PCI_CHIP_RV770_944A: + case PCI_CHIP_RV770_944B: + case PCI_CHIP_RV770_944C: + case PCI_CHIP_RV770_944E: + case PCI_CHIP_RV770_9450: + case PCI_CHIP_RV770_9452: + case PCI_CHIP_RV770_9456: + case PCI_CHIP_RV770_945A: + case PCI_CHIP_RV770_945B: + case PCI_CHIP_RV790_9460: + case PCI_CHIP_RV790_9462: + case PCI_CHIP_RV770_946A: + case PCI_CHIP_RV770_946B: + case PCI_CHIP_RV770_947A: + case PCI_CHIP_RV770_947B: + screen->chip_family = CHIP_FAMILY_RV770; + screen->chip_flags = RADEON_CHIPSET_TCL; + break; + + case PCI_CHIP_RV730_9480: + case PCI_CHIP_RV730_9487: + case PCI_CHIP_RV730_9488: + case PCI_CHIP_RV730_9489: + case PCI_CHIP_RV730_948F: + case PCI_CHIP_RV730_9490: + case PCI_CHIP_RV730_9491: + case PCI_CHIP_RV730_9495: + case PCI_CHIP_RV730_9498: + case PCI_CHIP_RV730_949C: + case PCI_CHIP_RV730_949E: + case PCI_CHIP_RV730_949F: + screen->chip_family = CHIP_FAMILY_RV730; + screen->chip_flags = RADEON_CHIPSET_TCL; + break; + + case PCI_CHIP_RV710_9540: + case PCI_CHIP_RV710_9541: + case PCI_CHIP_RV710_9542: + case PCI_CHIP_RV710_954E: + case PCI_CHIP_RV710_954F: + case PCI_CHIP_RV710_9552: + case PCI_CHIP_RV710_9553: + case PCI_CHIP_RV710_9555: + case PCI_CHIP_RV710_9557: + screen->chip_family = CHIP_FAMILY_RV710; + screen->chip_flags = RADEON_CHIPSET_TCL; + break; + + case PCI_CHIP_RV740_94A0: + case PCI_CHIP_RV740_94A1: + case PCI_CHIP_RV740_94A3: + case PCI_CHIP_RV740_94B1: + case PCI_CHIP_RV740_94B3: + case PCI_CHIP_RV740_94B4: + case PCI_CHIP_RV740_94B5: + case PCI_CHIP_RV740_94B9: + screen->chip_family = CHIP_FAMILY_RV740; + screen->chip_flags = RADEON_CHIPSET_TCL; + break; + default: fprintf(stderr, "unknown chip id 0x%x, can't guess.\n", - dri_priv->deviceID); + device_id); + return -1; + } + + return 0; +} + + +/* Create the device specific screen private data struct. + */ +static radeonScreenPtr +radeonCreateScreen( __DRIscreenPrivate *sPriv ) +{ + radeonScreenPtr screen; + RADEONDRIPtr dri_priv = (RADEONDRIPtr)sPriv->pDevPriv; + unsigned char *RADEONMMIO = NULL; + int i; + int ret; + uint32_t temp = 0; + + if (sPriv->devPrivSize != sizeof(RADEONDRIRec)) { + fprintf(stderr,"\nERROR! sizeof(RADEONDRIRec) does not match passed size from device driver\n"); + return GL_FALSE; + } + + /* Allocate the private area */ + screen = (radeonScreenPtr) CALLOC( sizeof(*screen) ); + if ( !screen ) { + __driUtilMessage("%s: Could not allocate memory for screen structure", + __FUNCTION__); return NULL; } + + radeon_init_debug(); + + /* parse information in __driConfigOptions */ + driParseOptionInfo (&screen->optionCache, + __driConfigOptions, __driNConfigOptions); + + /* This is first since which regions we map depends on whether or + * not we are using a PCI card. + */ + screen->card_type = (dri_priv->IsPCI ? RADEON_CARD_PCI : RADEON_CARD_AGP); + { + int ret; + + ret = radeonGetParam(sPriv, RADEON_PARAM_GART_BUFFER_OFFSET, + &screen->gart_buffer_offset); + + if (ret) { + FREE( screen ); + fprintf(stderr, "drm_radeon_getparam_t (RADEON_PARAM_GART_BUFFER_OFFSET): %d\n", ret); + return NULL; + } + + ret = radeonGetParam(sPriv, RADEON_PARAM_GART_BASE, + &screen->gart_base); + if (ret) { + FREE( screen ); + fprintf(stderr, "drm_radeon_getparam_t (RADEON_PARAM_GART_BASE): %d\n", ret); + return NULL; + } + + ret = radeonGetParam(sPriv, RADEON_PARAM_IRQ_NR, + &screen->irq); + if (ret) { + FREE( screen ); + fprintf(stderr, "drm_radeon_getparam_t (RADEON_PARAM_IRQ_NR): %d\n", ret); + return NULL; + } + screen->drmSupportsCubeMapsR200 = (sPriv->drm_version.minor >= 7); + screen->drmSupportsBlendColor = (sPriv->drm_version.minor >= 11); + screen->drmSupportsTriPerf = (sPriv->drm_version.minor >= 16); + screen->drmSupportsFragShader = (sPriv->drm_version.minor >= 18); + screen->drmSupportsPointSprites = (sPriv->drm_version.minor >= 13); + screen->drmSupportsCubeMapsR100 = (sPriv->drm_version.minor >= 15); + screen->drmSupportsVertexProgram = (sPriv->drm_version.minor >= 25); + screen->drmSupportsOcclusionQueries = (sPriv->drm_version.minor >= 30); + } + + ret = radeon_set_screen_flags(screen, dri_priv->deviceID); + if (ret == -1) + return NULL; + + screen->mmio.handle = dri_priv->registerHandle; + screen->mmio.size = dri_priv->registerSize; + if ( drmMap( sPriv->fd, + screen->mmio.handle, + screen->mmio.size, + &screen->mmio.map ) ) { + FREE( screen ); + __driUtilMessage("%s: drmMap failed\n", __FUNCTION__ ); + return NULL; + } + + RADEONMMIO = screen->mmio.map; + + screen->status.handle = dri_priv->statusHandle; + screen->status.size = dri_priv->statusSize; + if ( drmMap( sPriv->fd, + screen->status.handle, + screen->status.size, + &screen->status.map ) ) { + drmUnmap( screen->mmio.map, screen->mmio.size ); + FREE( screen ); + __driUtilMessage("%s: drmMap (2) failed\n", __FUNCTION__ ); + return NULL; + } + if (screen->chip_family < CHIP_FAMILY_R600) + screen->scratch = (__volatile__ uint32_t *) + ((GLubyte *)screen->status.map + RADEON_SCRATCH_REG_OFFSET); + else + screen->scratch = (__volatile__ uint32_t *) + ((GLubyte *)screen->status.map + R600_SCRATCH_REG_OFFSET); + + screen->buffers = drmMapBufs( sPriv->fd ); + if ( !screen->buffers ) { + drmUnmap( screen->status.map, screen->status.size ); + drmUnmap( screen->mmio.map, screen->mmio.size ); + FREE( screen ); + __driUtilMessage("%s: drmMapBufs failed\n", __FUNCTION__ ); + return NULL; + } + + if ( dri_priv->gartTexHandle && dri_priv->gartTexMapSize ) { + screen->gartTextures.handle = dri_priv->gartTexHandle; + screen->gartTextures.size = dri_priv->gartTexMapSize; + if ( drmMap( sPriv->fd, + screen->gartTextures.handle, + screen->gartTextures.size, + (drmAddressPtr)&screen->gartTextures.map ) ) { + drmUnmapBufs( screen->buffers ); + drmUnmap( screen->status.map, screen->status.size ); + drmUnmap( screen->mmio.map, screen->mmio.size ); + FREE( screen ); + __driUtilMessage("%s: drmMap failed for GART texture area\n", __FUNCTION__); + return NULL; + } + + screen->gart_texture_offset = dri_priv->gartTexOffset + screen->gart_base; + } + if ((screen->chip_family == CHIP_FAMILY_R350 || screen->chip_family == CHIP_FAMILY_R300) && sPriv->ddx_version.minor < 2) { fprintf(stderr, "xf86-video-ati-6.6.2 or newer needed for Radeon 9500/9700/9800 cards.\n"); @@ -836,35 +1063,57 @@ radeonCreateScreen( __DRIscreenPrivate *sPriv ) } if (getenv("R300_NO_TCL")) - screen->chip_flags &= ~RADEON_CHIPSET_TCL; + screen->chip_flags &= ~RADEON_CHIPSET_TCL; if (screen->chip_family <= CHIP_FAMILY_RS200) - screen->chip_flags |= RADEON_CLASS_R100; + screen->chip_flags |= RADEON_CLASS_R100; else if (screen->chip_family <= CHIP_FAMILY_RV280) - screen->chip_flags |= RADEON_CLASS_R200; + screen->chip_flags |= RADEON_CLASS_R200; + else if (screen->chip_family <= CHIP_FAMILY_RV570) + screen->chip_flags |= RADEON_CLASS_R300; else - screen->chip_flags |= RADEON_CLASS_R300; + screen->chip_flags |= RADEON_CLASS_R600; screen->cpp = dri_priv->bpp / 8; screen->AGPMode = dri_priv->AGPMode; - ret = radeonGetParam( sPriv->fd, RADEON_PARAM_FB_LOCATION, - &temp); - if (ret) { - if (screen->chip_family < CHIP_FAMILY_RS600) - screen->fbLocation = ( INREG( RADEON_MC_FB_LOCATION ) & 0xffff) << 16; - else { - FREE( screen ); - fprintf(stderr, "Unable to get fb location need newer drm\n"); - return NULL; + ret = radeonGetParam(sPriv, RADEON_PARAM_FB_LOCATION, &temp); + + /* +r6/r7 */ + if(screen->chip_family >= CHIP_FAMILY_R600) + { + if (ret) + { + FREE( screen ); + fprintf(stderr, "Unable to get fb location need newer drm\n"); + return NULL; + } + else + { + screen->fbLocation = (temp & 0xffff) << 24; } - } else { - screen->fbLocation = (temp & 0xffff) << 16; + } + else + { + if (ret) + { + if (screen->chip_family < CHIP_FAMILY_RS600 && !screen->kernel_mm) + screen->fbLocation = ( INREG( RADEON_MC_FB_LOCATION ) & 0xffff) << 16; + else + { + FREE( screen ); + fprintf(stderr, "Unable to get fb location need newer drm\n"); + return NULL; + } + } + else + { + screen->fbLocation = (temp & 0xffff) << 16; + } } - if (screen->chip_family >= CHIP_FAMILY_R300) { - ret = radeonGetParam( sPriv->fd, RADEON_PARAM_NUM_GB_PIPES, - &temp); + if (IS_R300_CLASS(screen)) { + ret = radeonGetParam(sPriv, RADEON_PARAM_NUM_GB_PIPES, &temp); if (ret) { fprintf(stderr, "Unable to get num_pipes, need newer drm\n"); switch (screen->chip_family) { @@ -902,6 +1151,14 @@ radeonCreateScreen( __DRIscreenPrivate *sPriv ) break; } + if ( sPriv->drm_version.minor >= 31 ) { + ret = radeonGetParam(sPriv, RADEON_PARAM_NUM_Z_PIPES, &temp); + if (ret) + screen->num_z_pipes = 2; + else + screen->num_z_pipes = temp; + } else + screen->num_z_pipes = 2; } if ( sPriv->drm_version.minor >= 10 ) { @@ -971,7 +1228,7 @@ radeonCreateScreen( __DRIscreenPrivate *sPriv ) #if RADEON_COMMON && defined(RADEON_COMMON_FOR_R200) if (IS_R200_CLASS(screen)) - screen->extensions[i++] = &r200AllocateExtension.base; + screen->extensions[i++] = &r200AllocateExtension.base; screen->extensions[i++] = &r200texOffsetExtension.base; #endif @@ -980,11 +1237,173 @@ radeonCreateScreen( __DRIscreenPrivate *sPriv ) screen->extensions[i++] = &r300texOffsetExtension.base; #endif +#if RADEON_COMMON && defined(RADEON_COMMON_FOR_R600) + screen->extensions[i++] = &r600texOffsetExtension.base; +#endif + screen->extensions[i++] = NULL; sPriv->extensions = screen->extensions; screen->driScreen = sPriv; screen->sarea_priv_offset = dri_priv->sarea_priv_offset; + screen->sarea = (drm_radeon_sarea_t *) ((GLubyte *) sPriv->pSAREA + + screen->sarea_priv_offset); + + screen->bom = radeon_bo_manager_legacy_ctor(screen); + if (screen->bom == NULL) { + free(screen); + return NULL; + } + + return screen; +} + +static radeonScreenPtr +radeonCreateScreen2(__DRIscreenPrivate *sPriv) +{ + radeonScreenPtr screen; + int i; + int ret; + uint32_t device_id = 0; + uint32_t temp = 0; + + /* Allocate the private area */ + screen = (radeonScreenPtr) CALLOC( sizeof(*screen) ); + if ( !screen ) { + __driUtilMessage("%s: Could not allocate memory for screen structure", + __FUNCTION__); + fprintf(stderr, "leaving here\n"); + return NULL; + } + + radeon_init_debug(); + + /* parse information in __driConfigOptions */ + driParseOptionInfo (&screen->optionCache, + __driConfigOptions, __driNConfigOptions); + + screen->kernel_mm = 1; + screen->chip_flags = 0; + + /* if we have kms we can support all of these */ + screen->drmSupportsCubeMapsR200 = 1; + screen->drmSupportsBlendColor = 1; + screen->drmSupportsTriPerf = 1; + screen->drmSupportsFragShader = 1; + screen->drmSupportsPointSprites = 1; + screen->drmSupportsCubeMapsR100 = 1; + screen->drmSupportsVertexProgram = 1; + screen->drmSupportsOcclusionQueries = 1; + screen->irq = 1; + + ret = radeonGetParam(sPriv, RADEON_PARAM_DEVICE_ID, &device_id); + if (ret) { + FREE( screen ); + fprintf(stderr, "drm_radeon_getparam_t (RADEON_PARAM_DEVICE_ID): %d\n", ret); + return NULL; + } + + ret = radeon_set_screen_flags(screen, device_id); + if (ret == -1) + return NULL; + + if (getenv("R300_NO_TCL")) + screen->chip_flags &= ~RADEON_CHIPSET_TCL; + + if (screen->chip_family <= CHIP_FAMILY_RS200) + screen->chip_flags |= RADEON_CLASS_R100; + else if (screen->chip_family <= CHIP_FAMILY_RV280) + screen->chip_flags |= RADEON_CLASS_R200; + else if (screen->chip_family <= CHIP_FAMILY_RV570) + screen->chip_flags |= RADEON_CLASS_R300; + else + screen->chip_flags |= RADEON_CLASS_R600; + + if (IS_R300_CLASS(screen)) { + ret = radeonGetParam(sPriv, RADEON_PARAM_NUM_GB_PIPES, &temp); + if (ret) { + fprintf(stderr, "Unable to get num_pipes, need newer drm\n"); + switch (screen->chip_family) { + case CHIP_FAMILY_R300: + case CHIP_FAMILY_R350: + screen->num_gb_pipes = 2; + break; + case CHIP_FAMILY_R420: + case CHIP_FAMILY_R520: + case CHIP_FAMILY_R580: + case CHIP_FAMILY_RV560: + case CHIP_FAMILY_RV570: + screen->num_gb_pipes = 4; + break; + case CHIP_FAMILY_RV350: + case CHIP_FAMILY_RV515: + case CHIP_FAMILY_RV530: + case CHIP_FAMILY_RV410: + default: + screen->num_gb_pipes = 1; + break; + } + } else { + screen->num_gb_pipes = temp; + } + + /* pipe overrides */ + switch (device_id) { + case PCI_CHIP_R300_AD: /* 9500 with 1 quadpipe verified by: Reid Linnemann <lreid@cs.okstate.edu> */ + case PCI_CHIP_RV410_5E4C: /* RV410 SE only have 1 quadpipe */ + case PCI_CHIP_RV410_5E4F: /* RV410 SE only have 1 quadpipe */ + screen->num_gb_pipes = 1; + break; + default: + break; + } + + ret = radeonGetParam(sPriv, RADEON_PARAM_NUM_Z_PIPES, &temp); + if (ret) + screen->num_z_pipes = 2; + else + screen->num_z_pipes = temp; + + } + + i = 0; + screen->extensions[i++] = &driCopySubBufferExtension.base; + screen->extensions[i++] = &driFrameTrackingExtension.base; + screen->extensions[i++] = &driReadDrawableExtension; + + if ( screen->irq != 0 ) { + screen->extensions[i++] = &driSwapControlExtension.base; + screen->extensions[i++] = &driMediaStreamCounterExtension.base; + } + +#if !RADEON_COMMON + screen->extensions[i++] = &radeonTexBufferExtension.base; +#endif + +#if RADEON_COMMON && defined(RADEON_COMMON_FOR_R200) + if (IS_R200_CLASS(screen)) + screen->extensions[i++] = &r200AllocateExtension.base; + + screen->extensions[i++] = &r200TexBufferExtension.base; +#endif + +#if RADEON_COMMON && defined(RADEON_COMMON_FOR_R300) + screen->extensions[i++] = &r300TexBufferExtension.base; +#endif + +#if RADEON_COMMON && defined(RADEON_COMMON_FOR_R600) + screen->extensions[i++] = &r600TexBufferExtension.base; +#endif + + screen->extensions[i++] = NULL; + sPriv->extensions = screen->extensions; + + screen->driScreen = sPriv; + screen->bom = radeon_bo_manager_gem_ctor(sPriv->fd); + if (screen->bom == NULL) { + free(screen); + return NULL; + } return screen; } @@ -993,23 +1412,32 @@ radeonCreateScreen( __DRIscreenPrivate *sPriv ) static void radeonDestroyScreen( __DRIscreenPrivate *sPriv ) { - radeonScreenPtr screen = (radeonScreenPtr)sPriv->private; + radeonScreenPtr screen = (radeonScreenPtr)sPriv->private; - if (!screen) - return; + if (!screen) + return; - if ( screen->gartTextures.map ) { - drmUnmap( screen->gartTextures.map, screen->gartTextures.size ); - } - drmUnmapBufs( screen->buffers ); - drmUnmap( screen->status.map, screen->status.size ); - drmUnmap( screen->mmio.map, screen->mmio.size ); + if (screen->kernel_mm) { +#ifdef RADEON_BO_TRACK + radeon_tracker_print(&screen->bom->tracker, stderr); +#endif + radeon_bo_manager_gem_dtor(screen->bom); + } else { + radeon_bo_manager_legacy_dtor(screen->bom); + + if ( screen->gartTextures.map ) { + drmUnmap( screen->gartTextures.map, screen->gartTextures.size ); + } + drmUnmapBufs( screen->buffers ); + drmUnmap( screen->status.map, screen->status.size ); + drmUnmap( screen->mmio.map, screen->mmio.size ); + } - /* free all option information */ - driDestroyOptionInfo (&screen->optionCache); + /* free all option information */ + driDestroyOptionInfo (&screen->optionCache); - FREE( screen ); - sPriv->private = NULL; + FREE( screen ); + sPriv->private = NULL; } @@ -1018,16 +1446,21 @@ radeonDestroyScreen( __DRIscreenPrivate *sPriv ) static GLboolean radeonInitDriver( __DRIscreenPrivate *sPriv ) { - sPriv->private = (void *) radeonCreateScreen( sPriv ); - if ( !sPriv->private ) { - radeonDestroyScreen( sPriv ); - return GL_FALSE; - } + if (sPriv->dri2.enabled) { + sPriv->private = (void *) radeonCreateScreen2( sPriv ); + } else { + sPriv->private = (void *) radeonCreateScreen( sPriv ); + } + if ( !sPriv->private ) { + radeonDestroyScreen( sPriv ); + return GL_FALSE; + } - return GL_TRUE; + return GL_TRUE; } + /** * Create the Mesa framebuffer and renderbuffers for a given window/drawable. * @@ -1040,132 +1473,112 @@ radeonCreateBuffer( __DRIscreenPrivate *driScrnPriv, const __GLcontextModes *mesaVis, GLboolean isPixmap ) { - radeonScreenPtr screen = (radeonScreenPtr) driScrnPriv->private; + radeonScreenPtr screen = (radeonScreenPtr) driScrnPriv->private; - if (isPixmap) { + const GLboolean swDepth = GL_FALSE; + const GLboolean swAlpha = GL_FALSE; + const GLboolean swAccum = mesaVis->accumRedBits > 0; + const GLboolean swStencil = mesaVis->stencilBits > 0 && + mesaVis->depthBits != 24; + GLenum rgbFormat; + struct radeon_framebuffer *rfb; + + if (isPixmap) return GL_FALSE; /* not implemented */ - } - else { - const GLboolean swDepth = GL_FALSE; - const GLboolean swAlpha = GL_FALSE; - const GLboolean swAccum = mesaVis->accumRedBits > 0; - const GLboolean swStencil = mesaVis->stencilBits > 0 && - mesaVis->depthBits != 24; - struct gl_framebuffer *fb = _mesa_create_framebuffer(mesaVis); - - /* front color renderbuffer */ - { - driRenderbuffer *frontRb - = driNewRenderbuffer(GL_RGBA, - driScrnPriv->pFB + screen->frontOffset, - screen->cpp, - screen->frontOffset, screen->frontPitch, - driDrawPriv); - radeonSetSpanFunctions(frontRb, mesaVis); - _mesa_add_renderbuffer(fb, BUFFER_FRONT_LEFT, &frontRb->Base); - } - /* back color renderbuffer */ - if (mesaVis->doubleBufferMode) { - driRenderbuffer *backRb - = driNewRenderbuffer(GL_RGBA, - driScrnPriv->pFB + screen->backOffset, - screen->cpp, - screen->backOffset, screen->backPitch, - driDrawPriv); - radeonSetSpanFunctions(backRb, mesaVis); - _mesa_add_renderbuffer(fb, BUFFER_BACK_LEFT, &backRb->Base); - } + rfb = CALLOC_STRUCT(radeon_framebuffer); + if (!rfb) + return GL_FALSE; - /* depth renderbuffer */ - if (mesaVis->depthBits == 16) { - driRenderbuffer *depthRb - = driNewRenderbuffer(GL_DEPTH_COMPONENT16, - driScrnPriv->pFB + screen->depthOffset, - screen->cpp, - screen->depthOffset, screen->depthPitch, - driDrawPriv); - radeonSetSpanFunctions(depthRb, mesaVis); - _mesa_add_renderbuffer(fb, BUFFER_DEPTH, &depthRb->Base); - depthRb->depthHasSurface = screen->depthHasSurface; - } - else if (mesaVis->depthBits == 24) { - driRenderbuffer *depthRb - = driNewRenderbuffer(GL_DEPTH_COMPONENT24, - driScrnPriv->pFB + screen->depthOffset, - screen->cpp, - screen->depthOffset, screen->depthPitch, - driDrawPriv); - radeonSetSpanFunctions(depthRb, mesaVis); - _mesa_add_renderbuffer(fb, BUFFER_DEPTH, &depthRb->Base); - depthRb->depthHasSurface = screen->depthHasSurface; - } + _mesa_initialize_framebuffer(&rfb->base, mesaVis); + + if (mesaVis->redBits == 5) + rgbFormat = GL_RGB5; + else if (mesaVis->alphaBits == 0) + rgbFormat = GL_RGB8; + else + rgbFormat = GL_RGBA8; + + /* front color renderbuffer */ + rfb->color_rb[0] = radeon_create_renderbuffer(rgbFormat, driDrawPriv); + _mesa_add_renderbuffer(&rfb->base, BUFFER_FRONT_LEFT, &rfb->color_rb[0]->base); + rfb->color_rb[0]->has_surface = 1; + + /* back color renderbuffer */ + if (mesaVis->doubleBufferMode) { + rfb->color_rb[1] = radeon_create_renderbuffer(rgbFormat, driDrawPriv); + _mesa_add_renderbuffer(&rfb->base, BUFFER_BACK_LEFT, &rfb->color_rb[1]->base); + rfb->color_rb[1]->has_surface = 1; + } - /* stencil renderbuffer */ - if (mesaVis->stencilBits > 0 && !swStencil) { - driRenderbuffer *stencilRb - = driNewRenderbuffer(GL_STENCIL_INDEX8_EXT, - driScrnPriv->pFB + screen->depthOffset, - screen->cpp, - screen->depthOffset, screen->depthPitch, - driDrawPriv); - radeonSetSpanFunctions(stencilRb, mesaVis); - _mesa_add_renderbuffer(fb, BUFFER_STENCIL, &stencilRb->Base); - stencilRb->depthHasSurface = screen->depthHasSurface; + if (mesaVis->depthBits == 24) { + if (mesaVis->stencilBits == 8) { + struct radeon_renderbuffer *depthStencilRb = radeon_create_renderbuffer(GL_DEPTH24_STENCIL8_EXT, driDrawPriv); + _mesa_add_renderbuffer(&rfb->base, BUFFER_DEPTH, &depthStencilRb->base); + _mesa_add_renderbuffer(&rfb->base, BUFFER_STENCIL, &depthStencilRb->base); + depthStencilRb->has_surface = screen->depthHasSurface; + } else { + /* depth renderbuffer */ + struct radeon_renderbuffer *depth = radeon_create_renderbuffer(GL_DEPTH_COMPONENT24, driDrawPriv); + _mesa_add_renderbuffer(&rfb->base, BUFFER_DEPTH, &depth->base); + depth->has_surface = screen->depthHasSurface; } + } else if (mesaVis->depthBits == 16) { + /* just 16-bit depth buffer, no hw stencil */ + struct radeon_renderbuffer *depth = radeon_create_renderbuffer(GL_DEPTH_COMPONENT16, driDrawPriv); + _mesa_add_renderbuffer(&rfb->base, BUFFER_DEPTH, &depth->base); + depth->has_surface = screen->depthHasSurface; + } - _mesa_add_soft_renderbuffers(fb, - GL_FALSE, /* color */ - swDepth, - swStencil, - swAccum, - swAlpha, - GL_FALSE /* aux */); - driDrawPriv->driverPrivate = (void *) fb; + _mesa_add_soft_renderbuffers(&rfb->base, + GL_FALSE, /* color */ + swDepth, + swStencil, + swAccum, + swAlpha, + GL_FALSE /* aux */); + driDrawPriv->driverPrivate = (void *) rfb; - return (driDrawPriv->driverPrivate != NULL); - } + return (driDrawPriv->driverPrivate != NULL); } -static void -radeonDestroyBuffer(__DRIdrawablePrivate *driDrawPriv) +static void radeon_cleanup_renderbuffers(struct radeon_framebuffer *rfb) { - _mesa_reference_framebuffer((GLframebuffer **)(&(driDrawPriv->driverPrivate)), NULL); -} + struct radeon_renderbuffer *rb; -#if RADEON_COMMON && defined(RADEON_COMMON_FOR_R300) -/** - * Choose the appropriate CreateContext function based on the chipset. - * Eventually, all drivers will go through this process. - */ -static GLboolean radeonCreateContext(const __GLcontextModes * glVisual, - __DRIcontextPrivate * driContextPriv, - void *sharedContextPriv) -{ - __DRIscreenPrivate *sPriv = driContextPriv->driScreenPriv; - radeonScreenPtr screen = (radeonScreenPtr) (sPriv->private); - - if (IS_R300_CLASS(screen)) - return r300CreateContext(glVisual, driContextPriv, sharedContextPriv); - return GL_FALSE; + rb = rfb->color_rb[0]; + if (rb && rb->bo) { + radeon_bo_unref(rb->bo); + rb->bo = NULL; + } + rb = rfb->color_rb[1]; + if (rb && rb->bo) { + radeon_bo_unref(rb->bo); + rb->bo = NULL; + } + rb = radeon_get_renderbuffer(&rfb->base, BUFFER_DEPTH); + if (rb && rb->bo) { + radeon_bo_unref(rb->bo); + rb->bo = NULL; + } } -/** - * Choose the appropriate DestroyContext function based on the chipset. - */ -static void radeonDestroyContext(__DRIcontextPrivate * driContextPriv) +void +radeonDestroyBuffer(__DRIdrawablePrivate *driDrawPriv) { - radeonContextPtr radeon = (radeonContextPtr) driContextPriv->driverPrivate; - - if (IS_R300_CLASS(radeon->radeonScreen)) - return r300DestroyContext(driContextPriv); + struct radeon_framebuffer *rfb; + if (!driDrawPriv) + return; + + rfb = (void*)driDrawPriv->driverPrivate; + if (!rfb) + return; + radeon_cleanup_renderbuffers(rfb); + _mesa_reference_framebuffer((GLframebuffer **)(&(driDrawPriv->driverPrivate)), NULL); } -#endif - - /** * This is the driver specific part of the createNewScreen entry point. * @@ -1191,6 +1604,11 @@ radeonInitScreen(__DRIscreenPrivate *psp) static const __DRIutilversion2 ddx_expected = { 4, 5, 0, 0 }; static const __DRIversion dri_expected = { 4, 0, 0 }; static const __DRIversion drm_expected = { 1, 24, 0 }; +#elif RADEON_COMMON && defined(RADEON_COMMON_FOR_R600) + static const char *driver_name = "R600"; + static const __DRIutilversion2 ddx_expected = { 4, 5, 0, 0 }; + static const __DRIversion dri_expected = { 4, 0, 0 }; + static const __DRIversion drm_expected = { 1, 24, 0 }; #endif RADEONDRIPtr dri_priv = (RADEONDRIPtr) psp->pDevPriv; @@ -1218,20 +1636,112 @@ radeonInitScreen(__DRIscreenPrivate *psp) driInitSingleExtension( NULL, NV_vp_extension ); driInitSingleExtension( NULL, ATI_fs_extension ); driInitExtensions( NULL, point_extensions, GL_FALSE ); -#elif defined(RADEON_COMMON_FOR_R300) +#elif (defined(RADEON_COMMON_FOR_R300) || defined(RADEON_COMMON_FOR_R600)) driInitSingleExtension( NULL, gl_20_extension ); #endif if (!radeonInitDriver(psp)) return NULL; + /* for now fill in all modes */ return radeonFillInModes( psp, dri_priv->bpp, (dri_priv->bpp == 16) ? 16 : 24, - (dri_priv->bpp == 16) ? 0 : 8, - (dri_priv->backOffset != dri_priv->depthOffset) ); + (dri_priv->bpp == 16) ? 0 : 8, 1); } +#define ARRAY_SIZE(a) (sizeof (a) / sizeof ((a)[0])) +/** + * This is the driver specific part of the createNewScreen entry point. + * Called when using DRI2. + * + * \return the __GLcontextModes supported by this driver + */ +static const +__DRIconfig **radeonInitScreen2(__DRIscreenPrivate *psp) +{ + GLenum fb_format[3]; + GLenum fb_type[3]; + /* GLX_SWAP_COPY_OML is only supported because the Intel driver doesn't + * support pageflipping at all. + */ + static const GLenum back_buffer_modes[] = { + GLX_NONE, GLX_SWAP_UNDEFINED_OML, /*, GLX_SWAP_COPY_OML*/ + }; + uint8_t depth_bits[4], stencil_bits[4], msaa_samples_array[1]; + int color; + __DRIconfig **configs = NULL; + + /* Calling driInitExtensions here, with a NULL context pointer, + * does not actually enable the extensions. It just makes sure + * that all the dispatch offsets for all the extensions that + * *might* be enables are known. This is needed because the + * dispatch offsets need to be known when _mesa_context_create + * is called, but we can't enable the extensions until we have a + * context pointer. + * + * Hello chicken. Hello egg. How are you two today? + */ + driInitExtensions( NULL, card_extensions, GL_FALSE ); + driInitExtensions( NULL, mm_extensions, GL_FALSE ); +#if RADEON_COMMON && defined(RADEON_COMMON_FOR_R200) + driInitExtensions( NULL, blend_extensions, GL_FALSE ); + driInitSingleExtension( NULL, ARB_vp_extension ); + driInitSingleExtension( NULL, NV_vp_extension ); + driInitSingleExtension( NULL, ATI_fs_extension ); + driInitExtensions( NULL, point_extensions, GL_FALSE ); +#elif (defined(RADEON_COMMON_FOR_R300) || defined(RADEON_COMMON_FOR_R600)) + driInitSingleExtension( NULL, gl_20_extension ); +#endif + + if (!radeonInitDriver(psp)) { + return NULL; + } + depth_bits[0] = 0; + stencil_bits[0] = 0; + depth_bits[1] = 16; + stencil_bits[1] = 0; + depth_bits[2] = 24; + stencil_bits[2] = 0; + depth_bits[3] = 24; + stencil_bits[3] = 8; + + msaa_samples_array[0] = 0; + + fb_format[0] = GL_RGB; + fb_type[0] = GL_UNSIGNED_SHORT_5_6_5; + + fb_format[1] = GL_BGR; + fb_type[1] = GL_UNSIGNED_INT_8_8_8_8_REV; + + fb_format[2] = GL_BGRA; + fb_type[2] = GL_UNSIGNED_INT_8_8_8_8_REV; + + for (color = 0; color < ARRAY_SIZE(fb_format); color++) { + __DRIconfig **new_configs; + + new_configs = driCreateConfigs(fb_format[color], fb_type[color], + depth_bits, + stencil_bits, + ARRAY_SIZE(depth_bits), + back_buffer_modes, + ARRAY_SIZE(back_buffer_modes), + msaa_samples_array, + ARRAY_SIZE(msaa_samples_array)); + if (configs == NULL) + configs = new_configs; + else + configs = driConcatConfigs(configs, new_configs); + } + + if (configs == NULL) { + fprintf(stderr, "[%s:%u] Error creating FBConfig!\n", __func__, + __LINE__); + return NULL; + } + + return (const __DRIconfig **)configs; +} /** * Get information about previous buffer swaps. @@ -1239,36 +1749,42 @@ radeonInitScreen(__DRIscreenPrivate *psp) static int getSwapInfo( __DRIdrawablePrivate *dPriv, __DRIswapInfo * sInfo ) { -#if !RADEON_COMMON || (RADEON_COMMON && defined(RADEON_COMMON_FOR_R300)) - radeonContextPtr rmesa; -#elif RADEON_COMMON && defined(RADEON_COMMON_FOR_R200) - r200ContextPtr rmesa; -#endif + struct radeon_framebuffer *rfb; - if ( (dPriv == NULL) || (dPriv->driContextPriv == NULL) - || (dPriv->driContextPriv->driverPrivate == NULL) - || (sInfo == NULL) ) { - return -1; + if ( (dPriv == NULL) || (dPriv->driContextPriv == NULL) + || (dPriv->driContextPriv->driverPrivate == NULL) + || (sInfo == NULL) ) { + return -1; } - rmesa = dPriv->driContextPriv->driverPrivate; - sInfo->swap_count = rmesa->swap_count; - sInfo->swap_ust = rmesa->swap_ust; - sInfo->swap_missed_count = rmesa->swap_missed_count; + rfb = dPriv->driverPrivate; + sInfo->swap_count = rfb->swap_count; + sInfo->swap_ust = rfb->swap_ust; + sInfo->swap_missed_count = rfb->swap_missed_count; sInfo->swap_missed_usage = (sInfo->swap_missed_count != 0) - ? driCalculateSwapUsage( dPriv, 0, rmesa->swap_missed_ust ) + ? driCalculateSwapUsage( dPriv, 0, rfb->swap_missed_ust ) : 0.0; return 0; } -#if !RADEON_COMMON || (RADEON_COMMON && defined(RADEON_COMMON_FOR_R300)) const struct __DriverAPIRec driDriverAPI = { .InitScreen = radeonInitScreen, .DestroyScreen = radeonDestroyScreen, - .CreateContext = radeonCreateContext, +#if RADEON_COMMON && defined(RADEON_COMMON_FOR_R200) + .CreateContext = r200CreateContext, + .DestroyContext = r200DestroyContext, +#elif RADEON_COMMON && defined(RADEON_COMMON_FOR_R600) + .CreateContext = r600CreateContext, + .DestroyContext = radeonDestroyContext, +#elif RADEON_COMMON && defined(RADEON_COMMON_FOR_R300) + .CreateContext = r300CreateContext, .DestroyContext = radeonDestroyContext, +#else + .CreateContext = r100CreateContext, + .DestroyContext = radeonDestroyContext, +#endif .CreateBuffer = radeonCreateBuffer, .DestroyBuffer = radeonDestroyBuffer, .SwapBuffers = radeonSwapBuffers, @@ -1280,23 +1796,7 @@ const struct __DriverAPIRec driDriverAPI = { .WaitForSBC = NULL, .SwapBuffersMSC = NULL, .CopySubBuffer = radeonCopySubBuffer, + /* DRI2 */ + .InitScreen2 = radeonInitScreen2, }; -#else -const struct __DriverAPIRec driDriverAPI = { - .InitScreen = radeonInitScreen, - .DestroyScreen = radeonDestroyScreen, - .CreateContext = r200CreateContext, - .DestroyContext = r200DestroyContext, - .CreateBuffer = radeonCreateBuffer, - .DestroyBuffer = radeonDestroyBuffer, - .SwapBuffers = r200SwapBuffers, - .MakeCurrent = r200MakeCurrent, - .UnbindContext = r200UnbindContext, - .GetSwapInfo = getSwapInfo, - .GetDrawableMSC = driDrawableGetMSC32, - .WaitForMSC = driWaitForMSC32, - .WaitForSBC = NULL, - .SwapBuffersMSC = NULL, - .CopySubBuffer = r200CopySubBuffer, -}; -#endif + diff --git a/src/mesa/drivers/dri/radeon/radeon_screen.h b/src/mesa/drivers/dri/radeon/radeon_screen.h index b84c70bfae9..15744e88284 100644 --- a/src/mesa/drivers/dri/radeon/radeon_screen.h +++ b/src/mesa/drivers/dri/radeon/radeon_screen.h @@ -54,11 +54,12 @@ typedef struct { drmAddress map; /* Mapping of the DRM region */ } radeonRegionRec, *radeonRegionPtr; -typedef struct { +typedef struct radeon_screen { int chip_family; int chip_flags; int cpp; int card_type; + int device_id; /* PCI ID */ int AGPMode; unsigned int irq; /* IRQ number (0 means none) */ @@ -98,14 +99,19 @@ typedef struct { GLboolean drmSupportsPointSprites; /* need radeon kernel module >= 1.13 */ GLboolean drmSupportsCubeMapsR100; /* need radeon kernel module >= 1.15 */ GLboolean drmSupportsVertexProgram; /* need radeon kernel module >= 1.25 */ + GLboolean drmSupportsOcclusionQueries; /* need radeon kernel module >= 1.30 */ GLboolean depthHasSurface; /* Configuration cache with default values for all contexts */ driOptionCache optionCache; - const __DRIextension *extensions[8]; + const __DRIextension *extensions[16]; int num_gb_pipes; + int num_z_pipes; + int kernel_mm; + drm_radeon_sarea_t *sarea; /* Private SAREA data */ + struct radeon_bo_manager *bom; } radeonScreenRec, *radeonScreenPtr; #define IS_R100_CLASS(screen) \ @@ -114,5 +120,8 @@ typedef struct { ((screen->chip_flags & RADEON_CLASS_MASK) == RADEON_CLASS_R200) #define IS_R300_CLASS(screen) \ ((screen->chip_flags & RADEON_CLASS_MASK) == RADEON_CLASS_R300) +#define IS_R600_CLASS(screen) \ + ((screen->chip_flags & RADEON_CLASS_MASK) == RADEON_CLASS_R600) +extern void radeonDestroyBuffer(__DRIdrawablePrivate *driDrawPriv); #endif /* __RADEON_SCREEN_H__ */ diff --git a/src/mesa/drivers/dri/radeon/radeon_span.c b/src/mesa/drivers/dri/radeon/radeon_span.c index 12051ff1c81..4e100d854ed 100644 --- a/src/mesa/drivers/dri/radeon/radeon_span.c +++ b/src/mesa/drivers/dri/radeon/radeon_span.c @@ -43,46 +43,222 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "main/glheader.h" #include "swrast/swrast.h" -#include "radeon_context.h" -#include "radeon_ioctl.h" -#include "radeon_state.h" +#include "radeon_common.h" +#include "radeon_lock.h" #include "radeon_span.h" -#include "radeon_tex.h" - -#include "drirenderbuffer.h" #define DBG 0 +static void radeonSetSpanFunctions(struct radeon_renderbuffer *rrb); + + +/* r200 depth buffer is always tiled - this is the formula + according to the docs unless I typo'ed in it +*/ +#if defined(RADEON_COMMON_FOR_R200) +static GLubyte *r200_depth_2byte(const struct radeon_renderbuffer * rrb, + GLint x, GLint y) +{ + GLubyte *ptr = rrb->bo->ptr; + GLint offset; + if (rrb->has_surface) { + offset = x * rrb->cpp + y * rrb->pitch; + } else { + GLuint b; + offset = 0; + b = (((y >> 4) * (rrb->pitch >> 8) + (x >> 6))); + offset += (b >> 1) << 12; + offset += (((rrb->pitch >> 8) & 0x1) ? (b & 0x1) : ((b & 0x1) ^ ((y >> 4) & 0x1))) << 11; + offset += ((y >> 2) & 0x3) << 9; + offset += ((x >> 3) & 0x1) << 8; + offset += ((x >> 4) & 0x3) << 6; + offset += ((x >> 2) & 0x1) << 5; + offset += ((y >> 1) & 0x1) << 4; + offset += ((x >> 1) & 0x1) << 3; + offset += (y & 0x1) << 2; + offset += (x & 0x1) << 1; + } + return &ptr[offset]; +} + +static GLubyte *r200_depth_4byte(const struct radeon_renderbuffer * rrb, + GLint x, GLint y) +{ + GLubyte *ptr = rrb->bo->ptr; + GLint offset; + if (rrb->has_surface) { + offset = x * rrb->cpp + y * rrb->pitch; + } else { + GLuint b; + offset = 0; + b = (((y & 0x7ff) >> 4) * (rrb->pitch >> 7) + (x >> 5)); + offset += (b >> 1) << 12; + offset += (((rrb->pitch >> 7) & 0x1) ? (b & 0x1) : ((b & 0x1) ^ ((y >> 4) & 0x1))) << 11; + offset += ((y >> 2) & 0x3) << 9; + offset += ((x >> 2) & 0x1) << 8; + offset += ((x >> 3) & 0x3) << 6; + offset += ((y >> 1) & 0x1) << 5; + offset += ((x >> 1) & 0x1) << 4; + offset += (y & 0x1) << 3; + offset += (x & 0x1) << 2; + } + return &ptr[offset]; +} +#endif + +/* radeon tiling on r300-r500 has 4 states, + macro-linear/micro-linear + macro-linear/micro-tiled + macro-tiled /micro-linear + macro-tiled /micro-tiled + 1 byte surface + 2 byte surface - two types - we only provide 8x2 microtiling + 4 byte surface + 8/16 byte (unused) +*/ +static GLubyte *radeon_ptr_4byte(const struct radeon_renderbuffer * rrb, + GLint x, GLint y) +{ + GLubyte *ptr = rrb->bo->ptr; + uint32_t mask = RADEON_BO_FLAGS_MACRO_TILE | RADEON_BO_FLAGS_MICRO_TILE; + GLint offset; + + if (rrb->has_surface || !(rrb->bo->flags & mask)) { + offset = x * rrb->cpp + y * rrb->pitch; + } else { + offset = 0; + if (rrb->bo->flags & RADEON_BO_FLAGS_MACRO_TILE) { + if (rrb->bo->flags & RADEON_BO_FLAGS_MICRO_TILE) { + offset = ((y >> 4) * (rrb->pitch >> 7) + (x >> 5)) << 11; + offset += (((y >> 3) ^ (x >> 5)) & 0x1) << 10; + offset += (((y >> 4) ^ (x >> 4)) & 0x1) << 9; + offset += (((y >> 2) ^ (x >> 4)) & 0x1) << 8; + offset += (((y >> 3) ^ (x >> 3)) & 0x1) << 7; + offset += ((y >> 1) & 0x1) << 6; + offset += ((x >> 2) & 0x1) << 5; + offset += (y & 1) << 4; + offset += (x & 3) << 2; + } else { + offset = ((y >> 3) * (rrb->pitch >> 8) + (x >> 6)) << 11; + offset += (((y >> 2) ^ (x >> 6)) & 0x1) << 10; + offset += (((y >> 3) ^ (x >> 5)) & 0x1) << 9; + offset += (((y >> 1) ^ (x >> 5)) & 0x1) << 8; + offset += (((y >> 2) ^ (x >> 4)) & 0x1) << 7; + offset += (y & 1) << 6; + offset += (x & 15) << 2; + } + } else { + offset = ((y >> 1) * (rrb->pitch >> 4) + (x >> 2)) << 5; + offset += (y & 1) << 4; + offset += (x & 3) << 2; + } + } + return &ptr[offset]; +} + +static GLubyte *radeon_ptr_2byte_8x2(const struct radeon_renderbuffer * rrb, + GLint x, GLint y) +{ + GLubyte *ptr = rrb->bo->ptr; + uint32_t mask = RADEON_BO_FLAGS_MACRO_TILE | RADEON_BO_FLAGS_MICRO_TILE; + GLint offset; + + if (rrb->has_surface || !(rrb->bo->flags & mask)) { + offset = x * rrb->cpp + y * rrb->pitch; + } else { + offset = 0; + if (rrb->bo->flags & RADEON_BO_FLAGS_MACRO_TILE) { + if (rrb->bo->flags & RADEON_BO_FLAGS_MICRO_TILE) { + offset = ((y >> 4) * (rrb->pitch >> 7) + (x >> 6)) << 11; + offset += (((y >> 3) ^ (x >> 6)) & 0x1) << 10; + offset += (((y >> 4) ^ (x >> 5)) & 0x1) << 9; + offset += (((y >> 2) ^ (x >> 5)) & 0x1) << 8; + offset += (((y >> 3) ^ (x >> 4)) & 0x1) << 7; + offset += ((y >> 1) & 0x1) << 6; + offset += ((x >> 3) & 0x1) << 5; + offset += (y & 1) << 4; + offset += (x & 3) << 2; + } else { + offset = ((y >> 3) * (rrb->pitch >> 8) + (x >> 7)) << 11; + offset += (((y >> 2) ^ (x >> 7)) & 0x1) << 10; + offset += (((y >> 3) ^ (x >> 6)) & 0x1) << 9; + offset += (((y >> 1) ^ (x >> 6)) & 0x1) << 8; + offset += (((y >> 2) ^ (x >> 5)) & 0x1) << 7; + offset += (y & 1) << 6; + offset += ((x >> 4) & 0x1) << 5; + offset += (x & 15) << 2; + } + } else { + offset = ((y >> 1) * (rrb->pitch >> 4) + (x >> 3)) << 5; + offset += (y & 0x1) << 4; + offset += (x & 0x7) << 1; + } + } + return &ptr[offset]; +} + +#ifndef COMPILE_R300 +static uint32_t +z24s8_to_s8z24(uint32_t val) +{ + return (val << 24) | (val >> 8); +} + +static uint32_t +s8z24_to_z24s8(uint32_t val) +{ + return (val >> 24) | (val << 8); +} +#endif + /* * Note that all information needed to access pixels in a renderbuffer * should be obtained through the gl_renderbuffer parameter, not per-context * information. */ #define LOCAL_VARS \ - driRenderbuffer *drb = (driRenderbuffer *) rb; \ - const __DRIdrawablePrivate *dPriv = drb->dPriv; \ - const GLuint bottom = dPriv->h - 1; \ - GLubyte *buf = (GLubyte *) drb->flippedData \ - + (dPriv->y * drb->flippedPitch + dPriv->x) * drb->cpp; \ - GLuint p; \ - (void) p; + struct radeon_context *radeon = RADEON_CONTEXT(ctx); \ + struct radeon_renderbuffer *rrb = (void *) rb; \ + const GLint yScale = ctx->DrawBuffer->Name ? 1 : -1; \ + const GLint yBias = ctx->DrawBuffer->Name ? 0 : rrb->base.Height - 1;\ + unsigned int num_cliprects; \ + struct drm_clip_rect *cliprects; \ + int x_off, y_off; \ + GLuint p; \ + (void)p; \ + radeon_get_cliprects(radeon, &cliprects, &num_cliprects, &x_off, &y_off); #define LOCAL_DEPTH_VARS \ - driRenderbuffer *drb = (driRenderbuffer *) rb; \ - const __DRIdrawablePrivate *dPriv = drb->dPriv; \ - const GLuint bottom = dPriv->h - 1; \ - GLuint xo = dPriv->x; \ - GLuint yo = dPriv->y; \ - GLubyte *buf = (GLubyte *) drb->Base.Data; + struct radeon_context *radeon = RADEON_CONTEXT(ctx); \ + struct radeon_renderbuffer *rrb = (void *) rb; \ + const GLint yScale = ctx->DrawBuffer->Name ? 1 : -1; \ + const GLint yBias = ctx->DrawBuffer->Name ? 0 : rrb->base.Height - 1;\ + unsigned int num_cliprects; \ + struct drm_clip_rect *cliprects; \ + int x_off, y_off; \ + radeon_get_cliprects(radeon, &cliprects, &num_cliprects, &x_off, &y_off); #define LOCAL_STENCIL_VARS LOCAL_DEPTH_VARS -#define Y_FLIP(Y) (bottom - (Y)) +#define Y_FLIP(_y) ((_y) * yScale + yBias) #define HW_LOCK() #define HW_UNLOCK() +/* XXX FBO: this is identical to the macro in spantmp2.h except we get + * the cliprect info from the context, not the driDrawable. + * Move this into spantmp2.h someday. + */ +#define HW_CLIPLOOP() \ + do { \ + int _nc = num_cliprects; \ + while ( _nc-- ) { \ + int minx = cliprects[_nc].x1 - x_off; \ + int miny = cliprects[_nc].y1 - y_off; \ + int maxx = cliprects[_nc].x2 - x_off; \ + int maxy = cliprects[_nc].y2 - y_off; + /* ================================================================ * Color buffer */ @@ -94,7 +270,41 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #define TAG(x) radeon##x##_RGB565 #define TAG2(x,y) radeon##x##_RGB565##y -#define GET_PTR(X,Y) (buf + ((Y) * drb->flippedPitch + (X)) * 2) +#define GET_PTR(X,Y) radeon_ptr_2byte_8x2(rrb, (X) + x_off, (Y) + y_off) +#include "spantmp2.h" + +/* 16 bit, ARGB1555 color spanline and pixel functions + */ +#define SPANTMP_PIXEL_FMT GL_BGRA +#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_SHORT_1_5_5_5_REV + +#define TAG(x) radeon##x##_ARGB1555 +#define TAG2(x,y) radeon##x##_ARGB1555##y +#define GET_PTR(X,Y) radeon_ptr_2byte_8x2(rrb, (X) + x_off, (Y) + y_off) +#include "spantmp2.h" + +/* 16 bit, RGBA4 color spanline and pixel functions + */ +#define SPANTMP_PIXEL_FMT GL_BGRA +#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_SHORT_4_4_4_4_REV + +#define TAG(x) radeon##x##_ARGB4444 +#define TAG2(x,y) radeon##x##_ARGB4444##y +#define GET_PTR(X,Y) radeon_ptr_2byte_8x2(rrb, (X) + x_off, (Y) + y_off) +#include "spantmp2.h" + +/* 32 bit, xRGB8888 color spanline and pixel functions + */ +#define SPANTMP_PIXEL_FMT GL_BGRA +#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_INT_8_8_8_8_REV + +#define TAG(x) radeon##x##_xRGB8888 +#define TAG2(x,y) radeon##x##_xRGB8888##y +#define GET_VALUE(_x, _y) ((*(GLuint*)(radeon_ptr_4byte(rrb, _x + x_off, _y + y_off)) | 0xff000000)) +#define PUT_VALUE(_x, _y, d) { \ + GLuint *_ptr = (GLuint*)radeon_ptr_4byte( rrb, _x + x_off, _y + y_off ); \ + *_ptr = d; \ +} while (0) #include "spantmp2.h" /* 32 bit, ARGB8888 color spanline and pixel functions @@ -104,7 +314,11 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #define TAG(x) radeon##x##_ARGB8888 #define TAG2(x,y) radeon##x##_ARGB8888##y -#define GET_PTR(X,Y) (buf + ((Y) * drb->flippedPitch + (X)) * 4) +#define GET_VALUE(_x, _y) (*(GLuint*)(radeon_ptr_4byte(rrb, _x + x_off, _y + y_off))) +#define PUT_VALUE(_x, _y, d) { \ + GLuint *_ptr = (GLuint*)radeon_ptr_4byte( rrb, _x + x_off, _y + y_off ); \ + *_ptr = d; \ +} while (0) #include "spantmp2.h" /* ================================================================ @@ -121,106 +335,127 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. * too... */ -static GLuint radeon_mba_z32(const driRenderbuffer * drb, GLint x, GLint y) -{ - GLuint pitch = drb->pitch; - if (drb->depthHasSurface) { - return 4 * (x + y * pitch); - } else { - GLuint ba, address = 0; /* a[0..1] = 0 */ - -#ifdef COMPILE_R300 - ba = (y / 8) * (pitch / 8) + (x / 8); -#else - ba = (y / 16) * (pitch / 16) + (x / 16); -#endif - - address |= (x & 0x7) << 2; /* a[2..4] = x[0..2] */ - address |= (y & 0x3) << 5; /* a[5..6] = y[0..1] */ - address |= (((x & 0x10) >> 2) ^ (y & 0x4)) << 5; /* a[7] = x[4] ^ y[2] */ - address |= (ba & 0x3) << 8; /* a[8..9] = ba[0..1] */ - - address |= (y & 0x8) << 7; /* a[10] = y[3] */ - address |= (((x & 0x8) << 1) ^ (y & 0x10)) << 7; /* a[11] = x[3] ^ y[4] */ - address |= (ba & ~0x3) << 10; /* a[12..] = ba[2..] */ - - return address; - } -} - -static INLINE GLuint -radeon_mba_z16(const driRenderbuffer * drb, GLint x, GLint y) -{ - GLuint pitch = drb->pitch; - if (drb->depthHasSurface) { - return 2 * (x + y * pitch); - } else { - GLuint ba, address = 0; /* a[0] = 0 */ - - ba = (y / 16) * (pitch / 32) + (x / 32); - - address |= (x & 0x7) << 1; /* a[1..3] = x[0..2] */ - address |= (y & 0x7) << 4; /* a[4..6] = y[0..2] */ - address |= (x & 0x8) << 4; /* a[7] = x[3] */ - address |= (ba & 0x3) << 8; /* a[8..9] = ba[0..1] */ - address |= (y & 0x8) << 7; /* a[10] = y[3] */ - address |= ((x & 0x10) ^ (y & 0x10)) << 7; /* a[11] = x[4] ^ y[4] */ - address |= (ba & ~0x3) << 10; /* a[12..] = ba[2..] */ - - return address; - } -} - /* 16-bit depth buffer functions */ #define VALUE_TYPE GLushort +#if defined(RADEON_COMMON_FOR_R200) +#define WRITE_DEPTH( _x, _y, d ) \ + *(GLushort *)r200_depth_2byte(rrb, _x + x_off, _y + y_off) = d +#else #define WRITE_DEPTH( _x, _y, d ) \ - *(GLushort *)(buf + radeon_mba_z16( drb, _x + xo, _y + yo )) = d; + *(GLushort *)radeon_ptr_2byte_8x2(rrb, _x + x_off, _y + y_off) = d +#endif +#if defined(RADEON_COMMON_FOR_R200) #define READ_DEPTH( d, _x, _y ) \ - d = *(GLushort *)(buf + radeon_mba_z16( drb, _x + xo, _y + yo )); + d = *(GLushort *)r200_depth_2byte(rrb, _x + x_off, _y + y_off) +#else +#define READ_DEPTH( d, _x, _y ) \ + d = *(GLushort *)radeon_ptr_2byte_8x2(rrb, _x + x_off, _y + y_off) +#endif #define TAG(x) radeon##x##_z16 #include "depthtmp.h" -/* 24 bit depth, 8 bit stencil depthbuffer functions +/* 24 bit depth * * Careful: It looks like the R300 uses ZZZS byte order while the R200 * uses SZZZ for 24 bit depth, 8 bit stencil mode. */ #define VALUE_TYPE GLuint -#ifdef COMPILE_R300 +#if defined(COMPILE_R300) #define WRITE_DEPTH( _x, _y, d ) \ do { \ - GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo ); \ - GLuint tmp = *(GLuint *)(buf + offset); \ + GLuint *_ptr = (GLuint*)radeon_ptr_4byte( rrb, _x + x_off, _y + y_off ); \ + GLuint tmp = *_ptr; \ tmp &= 0x000000ff; \ tmp |= ((d << 8) & 0xffffff00); \ - *(GLuint *)(buf + offset) = tmp; \ + *_ptr = tmp; \ +} while (0) +#elif defined(RADEON_COMMON_FOR_R200) +#define WRITE_DEPTH( _x, _y, d ) \ +do { \ + GLuint *_ptr = (GLuint*)r200_depth_4byte( rrb, _x + x_off, _y + y_off ); \ + GLuint tmp = *_ptr; \ + tmp &= 0xff000000; \ + tmp |= ((d) & 0x00ffffff); \ + *_ptr = tmp; \ } while (0) #else #define WRITE_DEPTH( _x, _y, d ) \ do { \ - GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo ); \ - GLuint tmp = *(GLuint *)(buf + offset); \ + GLuint *_ptr = (GLuint*)radeon_ptr_4byte( rrb, _x + x_off, _y + y_off ); \ + GLuint tmp = *_ptr; \ tmp &= 0xff000000; \ tmp |= ((d) & 0x00ffffff); \ - *(GLuint *)(buf + offset) = tmp; \ + *_ptr = tmp; \ } while (0) #endif -#ifdef COMPILE_R300 +#if defined(COMPILE_R300) #define READ_DEPTH( d, _x, _y ) \ - do { \ - d = (*(GLuint *)(buf + radeon_mba_z32( drb, _x + xo, \ - _y + yo )) & 0xffffff00) >> 8; \ + do { \ + d = (*(GLuint*)(radeon_ptr_4byte(rrb, _x + x_off, _y + y_off)) & 0xffffff00) >> 8; \ + }while(0) +#elif defined(RADEON_COMMON_FOR_R200) +#define READ_DEPTH( d, _x, _y ) \ + do { \ + d = *(GLuint*)(r200_depth_4byte(rrb, _x + x_off, _y + y_off)) & 0x00ffffff; \ }while(0) #else +#define READ_DEPTH( d, _x, _y ) \ + d = *(GLuint*)(radeon_ptr_4byte(rrb, _x + x_off, _y + y_off)) & 0x00ffffff; +#endif + +#define TAG(x) radeon##x##_z24 +#include "depthtmp.h" + +/* 24 bit depth, 8 bit stencil depthbuffer functions + * EXT_depth_stencil + * + * Careful: It looks like the R300 uses ZZZS byte order while the R200 + * uses SZZZ for 24 bit depth, 8 bit stencil mode. + */ +#define VALUE_TYPE GLuint + +#if defined(COMPILE_R300) +#define WRITE_DEPTH( _x, _y, d ) \ +do { \ + GLuint *_ptr = (GLuint*)radeon_ptr_4byte( rrb, _x + x_off, _y + y_off ); \ + *_ptr = d; \ +} while (0) +#elif defined(RADEON_COMMON_FOR_R200) +#define WRITE_DEPTH( _x, _y, d ) \ +do { \ + GLuint *_ptr = (GLuint*)r200_depth_4byte( rrb, _x + x_off, _y + y_off ); \ + GLuint tmp = z24s8_to_s8z24(d); \ + *_ptr = tmp; \ +} while (0) +#else +#define WRITE_DEPTH( _x, _y, d ) \ +do { \ + GLuint *_ptr = (GLuint*)radeon_ptr_4byte( rrb, _x + x_off, _y + y_off ); \ + GLuint tmp = z24s8_to_s8z24(d); \ + *_ptr = tmp; \ +} while (0) +#endif + +#if defined(COMPILE_R300) +#define READ_DEPTH( d, _x, _y ) \ + do { \ + d = (*(GLuint*)(radeon_ptr_4byte(rrb, _x + x_off, _y + y_off))); \ + }while(0) +#elif defined(RADEON_COMMON_FOR_R200) #define READ_DEPTH( d, _x, _y ) \ - d = *(GLuint *)(buf + radeon_mba_z32( drb, _x + xo, \ - _y + yo )) & 0x00ffffff; + do { \ + d = s8z24_to_z24s8(*(GLuint*)(r200_depth_4byte(rrb, _x + x_off, _y + y_off))); \ + }while(0) +#else +#define READ_DEPTH( d, _x, _y ) do { \ + d = s8z24_to_z24s8(*(GLuint*)(radeon_ptr_4byte(rrb, _x + x_off, _y + y_off ))); \ + } while (0) #endif #define TAG(x) radeon##x##_z24_s8 @@ -235,35 +470,51 @@ do { \ #ifdef COMPILE_R300 #define WRITE_STENCIL( _x, _y, d ) \ do { \ - GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo ); \ - GLuint tmp = *(GLuint *)(buf + offset); \ + GLuint *_ptr = (GLuint*)radeon_ptr_4byte(rrb, _x + x_off, _y + y_off); \ + GLuint tmp = *_ptr; \ tmp &= 0xffffff00; \ tmp |= (d) & 0xff; \ - *(GLuint *)(buf + offset) = tmp; \ + *_ptr = tmp; \ +} while (0) +#elif defined(RADEON_COMMON_FOR_R200) +#define WRITE_STENCIL( _x, _y, d ) \ +do { \ + GLuint *_ptr = (GLuint*)r200_depth_4byte(rrb, _x + x_off, _y + y_off); \ + GLuint tmp = *_ptr; \ + tmp &= 0x00ffffff; \ + tmp |= (((d) & 0xff) << 24); \ + *_ptr = tmp; \ } while (0) #else #define WRITE_STENCIL( _x, _y, d ) \ do { \ - GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo ); \ - GLuint tmp = *(GLuint *)(buf + offset); \ + GLuint *_ptr = (GLuint*)radeon_ptr_4byte(rrb, _x + x_off, _y + y_off); \ + GLuint tmp = *_ptr; \ tmp &= 0x00ffffff; \ tmp |= (((d) & 0xff) << 24); \ - *(GLuint *)(buf + offset) = tmp; \ + *_ptr = tmp; \ } while (0) #endif #ifdef COMPILE_R300 #define READ_STENCIL( d, _x, _y ) \ do { \ - GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo ); \ - GLuint tmp = *(GLuint *)(buf + offset); \ + GLuint *_ptr = (GLuint*)radeon_ptr_4byte( rrb, _x + x_off, _y + y_off ); \ + GLuint tmp = *_ptr; \ d = tmp & 0x000000ff; \ } while (0) +#elif defined(RADEON_COMMON_FOR_R200) +#define READ_STENCIL( d, _x, _y ) \ +do { \ + GLuint *_ptr = (GLuint*)r200_depth_4byte( rrb, _x + x_off, _y + y_off ); \ + GLuint tmp = *_ptr; \ + d = (tmp & 0xff000000) >> 24; \ +} while (0) #else #define READ_STENCIL( d, _x, _y ) \ do { \ - GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo ); \ - GLuint tmp = *(GLuint *)(buf + offset); \ + GLuint *_ptr = (GLuint*)radeon_ptr_4byte( rrb, _x + x_off, _y + y_off ); \ + GLuint tmp = *_ptr; \ d = (tmp & 0xff000000) >> 24; \ } while (0) #endif @@ -271,29 +522,110 @@ do { \ #define TAG(x) radeon##x##_z24_s8 #include "stenciltmp.h" -/* Move locking out to get reasonable span performance (10x better - * than doing this in HW_LOCK above). WaitForIdle() is the main - * culprit. - */ + +static void map_unmap_rb(struct gl_renderbuffer *rb, int flag) +{ + struct radeon_renderbuffer *rrb = radeon_renderbuffer(rb); + int r; + + if (rrb == NULL || !rrb->bo) + return; + + if (flag) { + if (rrb->bo->bom->funcs->bo_wait) + radeon_bo_wait(rrb->bo); + r = radeon_bo_map(rrb->bo, 1); + if (r) { + fprintf(stderr, "(%s) error(%d) mapping buffer.\n", + __FUNCTION__, r); + } + + radeonSetSpanFunctions(rrb); + } else { + radeon_bo_unmap(rrb->bo); + rb->GetRow = NULL; + rb->PutRow = NULL; + } +} + +static void +radeon_map_unmap_buffers(GLcontext *ctx, GLboolean map) +{ + GLuint i, j; + + /* color draw buffers */ + for (j = 0; j < ctx->DrawBuffer->_NumColorDrawBuffers; j++) + map_unmap_rb(ctx->DrawBuffer->_ColorDrawBuffers[j], map); + + /* check for render to textures */ + for (i = 0; i < BUFFER_COUNT; i++) { + struct gl_renderbuffer_attachment *att = + ctx->DrawBuffer->Attachment + i; + struct gl_texture_object *tex = att->Texture; + if (tex) { + /* Render to texture. Note that a mipmapped texture need not + * be complete for render to texture, so we must restrict to + * mapping only the attached image. + */ + radeon_texture_image *image = get_radeon_texture_image(tex->Image[att->CubeMapFace][att->TextureLevel]); + ASSERT(att->Renderbuffer); + + if (map) + radeon_teximage_map(image, GL_TRUE); + else + radeon_teximage_unmap(image); + } + } + + map_unmap_rb(ctx->ReadBuffer->_ColorReadBuffer, map); + + /* depth buffer (Note wrapper!) */ + if (ctx->DrawBuffer->_DepthBuffer) + map_unmap_rb(ctx->DrawBuffer->_DepthBuffer->Wrapped, map); + + if (ctx->DrawBuffer->_StencilBuffer) + map_unmap_rb(ctx->DrawBuffer->_StencilBuffer->Wrapped, map); +} static void radeonSpanRenderStart(GLcontext * ctx) { radeonContextPtr rmesa = RADEON_CONTEXT(ctx); -#ifdef COMPILE_R300 - r300ContextPtr r300 = (r300ContextPtr) rmesa; - R300_FIREVERTICES(r300); -#else - RADEON_FIREVERTICES(rmesa); -#endif - LOCK_HARDWARE(rmesa); - radeonWaitForIdleLocked(rmesa); + int i; + + radeon_firevertices(rmesa); + + /* The locking and wait for idle should really only be needed in classic mode. + * In a future memory manager based implementation, this should become + * unnecessary due to the fact that mapping our buffers, textures, etc. + * should implicitly wait for any previous rendering commands that must + * be waited on. */ + if (!rmesa->radeonScreen->driScreen->dri2.enabled) { + LOCK_HARDWARE(rmesa); + radeonWaitForIdleLocked(rmesa); + } + + for (i = 0; i < ctx->Const.MaxTextureImageUnits; i++) { + if (ctx->Texture.Unit[i]._ReallyEnabled) + ctx->Driver.MapTexture(ctx, ctx->Texture.Unit[i]._Current); + } + + radeon_map_unmap_buffers(ctx, 1); } static void radeonSpanRenderFinish(GLcontext * ctx) { radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + int i; _swrast_flush(ctx); - UNLOCK_HARDWARE(rmesa); + if (!rmesa->radeonScreen->driScreen->dri2.enabled) { + UNLOCK_HARDWARE(rmesa); + } + for (i = 0; i < ctx->Const.MaxTextureImageUnits; i++) { + if (ctx->Texture.Unit[i]._ReallyEnabled) + ctx->Driver.UnmapTexture(ctx, ctx->Texture.Unit[i]._Current); + } + + radeon_map_unmap_buffers(ctx, 0); } void radeonInitSpanFuncs(GLcontext * ctx) @@ -307,20 +639,27 @@ void radeonInitSpanFuncs(GLcontext * ctx) /** * Plug in the Get/Put routines for the given driRenderbuffer. */ -void radeonSetSpanFunctions(driRenderbuffer * drb, const GLvisual * vis) +static void radeonSetSpanFunctions(struct radeon_renderbuffer *rrb) { - if (drb->Base.InternalFormat == GL_RGBA) { - if (vis->redBits == 5 && vis->greenBits == 6 - && vis->blueBits == 5) { - radeonInitPointers_RGB565(&drb->Base); - } else { - radeonInitPointers_ARGB8888(&drb->Base); - } - } else if (drb->Base.InternalFormat == GL_DEPTH_COMPONENT16) { - radeonInitDepthPointers_z16(&drb->Base); - } else if (drb->Base.InternalFormat == GL_DEPTH_COMPONENT24) { - radeonInitDepthPointers_z24_s8(&drb->Base); - } else if (drb->Base.InternalFormat == GL_STENCIL_INDEX8_EXT) { - radeonInitStencilPointers_z24_s8(&drb->Base); + if (rrb->base._ActualFormat == GL_RGB5) { + radeonInitPointers_RGB565(&rrb->base); + } else if (rrb->base._ActualFormat == GL_RGB8) { + radeonInitPointers_xRGB8888(&rrb->base); + } else if (rrb->base._ActualFormat == GL_RGBA8) { + radeonInitPointers_ARGB8888(&rrb->base); + } else if (rrb->base._ActualFormat == GL_RGBA4) { + radeonInitPointers_ARGB4444(&rrb->base); + } else if (rrb->base._ActualFormat == GL_RGB5_A1) { + radeonInitPointers_ARGB1555(&rrb->base); + } else if (rrb->base._ActualFormat == GL_DEPTH_COMPONENT16) { + radeonInitDepthPointers_z16(&rrb->base); + } else if (rrb->base._ActualFormat == GL_DEPTH_COMPONENT24) { + radeonInitDepthPointers_z24(&rrb->base); + } else if (rrb->base._ActualFormat == GL_DEPTH24_STENCIL8_EXT) { + radeonInitDepthPointers_z24_s8(&rrb->base); + } else if (rrb->base._ActualFormat == GL_STENCIL_INDEX8_EXT) { + radeonInitStencilPointers_z24_s8(&rrb->base); + } else { + fprintf(stderr, "radeonSetSpanFunctions: bad actual format: 0x%04X\n", rrb->base._ActualFormat); } } diff --git a/src/mesa/drivers/dri/radeon/radeon_span.h b/src/mesa/drivers/dri/radeon/radeon_span.h index 9abe0864b17..ea6a2e7fb4e 100644 --- a/src/mesa/drivers/dri/radeon/radeon_span.h +++ b/src/mesa/drivers/dri/radeon/radeon_span.h @@ -42,9 +42,6 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #ifndef __RADEON_SPAN_H__ #define __RADEON_SPAN_H__ -#include "drirenderbuffer.h" - extern void radeonInitSpanFuncs(GLcontext * ctx); -extern void radeonSetSpanFunctions(driRenderbuffer * rb, const GLvisual * vis); #endif diff --git a/src/mesa/drivers/dri/radeon/radeon_state.c b/src/mesa/drivers/dri/radeon/radeon_state.c index b6561001e76..4d0d35ee0cd 100644 --- a/src/mesa/drivers/dri/radeon/radeon_state.c +++ b/src/mesa/drivers/dri/radeon/radeon_state.c @@ -40,6 +40,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "main/state.h" #include "main/context.h" #include "main/framebuffer.h" +#include "main/simple_list.h" #include "vbo/vbo.h" #include "tnl/tnl.h" @@ -47,6 +48,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "swrast_setup/swrast_setup.h" #include "radeon_context.h" +#include "radeon_mipmap_tree.h" #include "radeon_ioctl.h" #include "radeon_state.h" #include "radeon_tcl.h" @@ -62,7 +64,7 @@ static void radeonUpdateSpecular( GLcontext *ctx ); static void radeonAlphaFunc( GLcontext *ctx, GLenum func, GLfloat ref ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); int pp_misc = rmesa->hw.ctx.cmd[CTX_PP_MISC]; GLubyte refByte; @@ -106,7 +108,7 @@ static void radeonAlphaFunc( GLcontext *ctx, GLenum func, GLfloat ref ) static void radeonBlendEquationSeparate( GLcontext *ctx, GLenum modeRGB, GLenum modeA ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); GLuint b = rmesa->hw.ctx.cmd[CTX_RB3D_BLENDCNTL] & ~RADEON_COMB_FCN_MASK; GLboolean fallback = GL_FALSE; @@ -147,8 +149,8 @@ static void radeonBlendFuncSeparate( GLcontext *ctx, GLenum sfactorRGB, GLenum dfactorRGB, GLenum sfactorA, GLenum dfactorA ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - GLuint b = rmesa->hw.ctx.cmd[CTX_RB3D_BLENDCNTL] & + r100ContextPtr rmesa = R100_CONTEXT(ctx); + GLuint b = rmesa->hw.ctx.cmd[CTX_RB3D_BLENDCNTL] & ~(RADEON_SRC_BLEND_MASK | RADEON_DST_BLEND_MASK); GLboolean fallback = GL_FALSE; @@ -257,7 +259,7 @@ static void radeonBlendFuncSeparate( GLcontext *ctx, static void radeonDepthFunc( GLcontext *ctx, GLenum func ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); RADEON_STATECHANGE( rmesa, ctx ); rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] &= ~RADEON_Z_TEST_MASK; @@ -293,7 +295,7 @@ static void radeonDepthFunc( GLcontext *ctx, GLenum func ) static void radeonDepthMask( GLcontext *ctx, GLboolean flag ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); RADEON_STATECHANGE( rmesa, ctx ); if ( ctx->Depth.Mask ) { @@ -305,16 +307,16 @@ static void radeonDepthMask( GLcontext *ctx, GLboolean flag ) static void radeonClearDepth( GLcontext *ctx, GLclampd d ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); GLuint format = (rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] & RADEON_DEPTH_FORMAT_MASK); switch ( format ) { case RADEON_DEPTH_FORMAT_16BIT_INT_Z: - rmesa->state.depth.clear = d * 0x0000ffff; + rmesa->radeon.state.depth.clear = d * 0x0000ffff; break; case RADEON_DEPTH_FORMAT_24BIT_INT_Z: - rmesa->state.depth.clear = d * 0x00ffffff; + rmesa->radeon.state.depth.clear = d * 0x00ffffff; break; } } @@ -327,7 +329,7 @@ static void radeonClearDepth( GLcontext *ctx, GLclampd d ) static void radeonFogfv( GLcontext *ctx, GLenum pname, const GLfloat *param ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); union { int i; float f; } c, d; GLchan col[4]; @@ -391,7 +393,7 @@ static void radeonFogfv( GLcontext *ctx, GLenum pname, const GLfloat *param ) rmesa->hw.fog.cmd[FOG_D] = d.i; } break; - case GL_FOG_COLOR: + case GL_FOG_COLOR: RADEON_STATECHANGE( rmesa, ctx ); UNCLAMPED_FLOAT_TO_RGB_CHAN( col, ctx->Fog.Color ); rmesa->hw.ctx.cmd[CTX_PP_FOG_COLOR] &= ~RADEON_FOG_COLOR_MASK; @@ -406,109 +408,13 @@ static void radeonFogfv( GLcontext *ctx, GLenum pname, const GLfloat *param ) } } - -/* ============================================================= - * Scissoring - */ - - -static GLboolean intersect_rect( drm_clip_rect_t *out, - drm_clip_rect_t *a, - drm_clip_rect_t *b ) -{ - *out = *a; - if ( b->x1 > out->x1 ) out->x1 = b->x1; - if ( b->y1 > out->y1 ) out->y1 = b->y1; - if ( b->x2 < out->x2 ) out->x2 = b->x2; - if ( b->y2 < out->y2 ) out->y2 = b->y2; - if ( out->x1 >= out->x2 ) return GL_FALSE; - if ( out->y1 >= out->y2 ) return GL_FALSE; - return GL_TRUE; -} - - -void radeonRecalcScissorRects( radeonContextPtr rmesa ) -{ - drm_clip_rect_t *out; - int i; - - /* Grow cliprect store? - */ - if (rmesa->state.scissor.numAllocedClipRects < rmesa->numClipRects) { - while (rmesa->state.scissor.numAllocedClipRects < rmesa->numClipRects) { - rmesa->state.scissor.numAllocedClipRects += 1; /* zero case */ - rmesa->state.scissor.numAllocedClipRects *= 2; - } - - if (rmesa->state.scissor.pClipRects) - FREE(rmesa->state.scissor.pClipRects); - - rmesa->state.scissor.pClipRects = - MALLOC( rmesa->state.scissor.numAllocedClipRects * - sizeof(drm_clip_rect_t) ); - - if ( rmesa->state.scissor.pClipRects == NULL ) { - rmesa->state.scissor.numAllocedClipRects = 0; - return; - } - } - - out = rmesa->state.scissor.pClipRects; - rmesa->state.scissor.numClipRects = 0; - - for ( i = 0 ; i < rmesa->numClipRects ; i++ ) { - if ( intersect_rect( out, - &rmesa->pClipRects[i], - &rmesa->state.scissor.rect ) ) { - rmesa->state.scissor.numClipRects++; - out++; - } - } -} - - -static void radeonUpdateScissor( GLcontext *ctx ) -{ - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - - if ( rmesa->dri.drawable ) { - __DRIdrawablePrivate *dPriv = rmesa->dri.drawable; - - int x = ctx->Scissor.X; - int y = dPriv->h - ctx->Scissor.Y - ctx->Scissor.Height; - int w = ctx->Scissor.X + ctx->Scissor.Width - 1; - int h = dPriv->h - ctx->Scissor.Y - 1; - - rmesa->state.scissor.rect.x1 = x + dPriv->x; - rmesa->state.scissor.rect.y1 = y + dPriv->y; - rmesa->state.scissor.rect.x2 = w + dPriv->x + 1; - rmesa->state.scissor.rect.y2 = h + dPriv->y + 1; - - radeonRecalcScissorRects( rmesa ); - } -} - - -static void radeonScissor( GLcontext *ctx, - GLint x, GLint y, GLsizei w, GLsizei h ) -{ - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - - if ( ctx->Scissor.Enabled ) { - RADEON_FIREVERTICES( rmesa ); /* don't pipeline cliprect changes */ - radeonUpdateScissor( ctx ); - } - -} - - /* ============================================================= * Culling */ static void radeonCullFace( GLcontext *ctx, GLenum unused ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); GLuint s = rmesa->hw.set.cmd[SET_SE_CNTL]; GLuint t = rmesa->hw.tcl.cmd[TCL_UCP_VERT_BLEND_CTL]; @@ -545,7 +451,7 @@ static void radeonCullFace( GLcontext *ctx, GLenum unused ) static void radeonFrontFace( GLcontext *ctx, GLenum mode ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); RADEON_STATECHANGE( rmesa, set ); rmesa->hw.set.cmd[SET_SE_CNTL] &= ~RADEON_FFACE_CULL_DIR_MASK; @@ -553,6 +459,10 @@ static void radeonFrontFace( GLcontext *ctx, GLenum mode ) RADEON_STATECHANGE( rmesa, tcl ); rmesa->hw.tcl.cmd[TCL_UCP_VERT_BLEND_CTL] &= ~RADEON_CULL_FRONT_IS_CCW; + /* Winding is inverted when rendering to FBO */ + if (ctx->DrawBuffer && ctx->DrawBuffer->Name) + mode = (mode == GL_CW) ? GL_CCW : GL_CW; + switch ( mode ) { case GL_CW: rmesa->hw.set.cmd[SET_SE_CNTL] |= RADEON_FFACE_CULL_CW; @@ -570,7 +480,7 @@ static void radeonFrontFace( GLcontext *ctx, GLenum mode ) */ static void radeonLineWidth( GLcontext *ctx, GLfloat widthf ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); RADEON_STATECHANGE( rmesa, lin ); RADEON_STATECHANGE( rmesa, set ); @@ -587,10 +497,10 @@ static void radeonLineWidth( GLcontext *ctx, GLfloat widthf ) static void radeonLineStipple( GLcontext *ctx, GLint factor, GLushort pattern ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); RADEON_STATECHANGE( rmesa, lin ); - rmesa->hw.lin.cmd[LIN_RE_LINE_PATTERN] = + rmesa->hw.lin.cmd[LIN_RE_LINE_PATTERN] = ((((GLuint)factor & 0xff) << 16) | ((GLuint)pattern)); } @@ -602,12 +512,19 @@ static void radeonColorMask( GLcontext *ctx, GLboolean r, GLboolean g, GLboolean b, GLboolean a ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - GLuint mask = radeonPackColor( rmesa->radeonScreen->cpp, - ctx->Color.ColorMask[RCOMP], - ctx->Color.ColorMask[GCOMP], - ctx->Color.ColorMask[BCOMP], - ctx->Color.ColorMask[ACOMP] ); + r100ContextPtr rmesa = R100_CONTEXT(ctx); + struct radeon_renderbuffer *rrb; + GLuint mask; + + rrb = radeon_get_colorbuffer(&rmesa->radeon); + if (!rrb) + return; + + mask = radeonPackColor( rrb->cpp, + ctx->Color.ColorMask[RCOMP], + ctx->Color.ColorMask[GCOMP], + ctx->Color.ColorMask[BCOMP], + ctx->Color.ColorMask[ACOMP] ); if ( rmesa->hw.msk.cmd[MSK_RB3D_PLANEMASK] != mask ) { RADEON_STATECHANGE( rmesa, msk ); @@ -623,8 +540,9 @@ static void radeonColorMask( GLcontext *ctx, static void radeonPolygonOffset( GLcontext *ctx, GLfloat factor, GLfloat units ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - float_ui32_type constant = { units * rmesa->state.depth.scale }; + r100ContextPtr rmesa = R100_CONTEXT(ctx); + const GLfloat depthScale = 1.0F / ctx->DrawBuffer->_DepthMaxF; + float_ui32_type constant = { units * depthScale }; float_ui32_type factoru = { factor }; RADEON_STATECHANGE( rmesa, zbs ); @@ -632,41 +550,16 @@ static void radeonPolygonOffset( GLcontext *ctx, rmesa->hw.zbs.cmd[ZBS_SE_ZBIAS_CONSTANT] = constant.ui32; } -static void radeonPolygonStipple( GLcontext *ctx, const GLubyte *mask ) -{ - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - GLuint i; - drm_radeon_stipple_t stipple; - - /* Must flip pattern upside down. - */ - for ( i = 0 ; i < 32 ; i++ ) { - rmesa->state.stipple.mask[31 - i] = ((GLuint *) mask)[i]; - } - - /* TODO: push this into cmd mechanism - */ - RADEON_FIREVERTICES( rmesa ); - LOCK_HARDWARE( rmesa ); - - /* FIXME: Use window x,y offsets into stipple RAM. - */ - stipple.mask = rmesa->state.stipple.mask; - drmCommandWrite( rmesa->dri.fd, DRM_RADEON_STIPPLE, - &stipple, sizeof(drm_radeon_stipple_t) ); - UNLOCK_HARDWARE( rmesa ); -} - static void radeonPolygonMode( GLcontext *ctx, GLenum face, GLenum mode ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); GLboolean flag = (ctx->_TriangleCaps & DD_TRI_UNFILLED) != 0; /* Can't generally do unfilled via tcl, but some good special - * cases work. + * cases work. */ TCL_FALLBACK( ctx, RADEON_TCL_FALLBACK_UNFILLED, flag); - if (rmesa->TclFallback) { + if (rmesa->radeon.TclFallback) { radeonChooseRenderState( ctx ); radeonChooseVertexState( ctx ); } @@ -686,7 +579,7 @@ static void radeonPolygonMode( GLcontext *ctx, GLenum face, GLenum mode ) */ static void radeonUpdateSpecular( GLcontext *ctx ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); uint32_t p = rmesa->hw.ctx.cmd[CTX_PP_CNTL]; GLuint flag = 0; @@ -711,7 +604,7 @@ static void radeonUpdateSpecular( GLcontext *ctx ) rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] |= RADEON_TCL_VTX_PK_DIFFUSE; rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] |= RADEON_LIGHTING_ENABLE; p |= RADEON_SPECULAR_ENABLE; - rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] &= + rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] &= ~RADEON_DIFFUSE_SPECULAR_COMBINE; } else if (ctx->Light.Enabled) { @@ -741,7 +634,7 @@ static void radeonUpdateSpecular( GLcontext *ctx ) RADEON_TCL_COMPUTE_SPECULAR) != 0; } } - + TCL_FALLBACK( ctx, RADEON_TCL_FALLBACK_FOGCOORDSPEC, flag); if (NEED_SECONDARY_COLOR(ctx)) { @@ -757,7 +650,7 @@ static void radeonUpdateSpecular( GLcontext *ctx ) /* Update vertex/render formats */ - if (rmesa->TclFallback) { + if (rmesa->radeon.TclFallback) { radeonChooseRenderState( ctx ); radeonChooseVertexState( ctx ); } @@ -769,12 +662,12 @@ static void radeonUpdateSpecular( GLcontext *ctx ) */ -/* Update on colormaterial, material emmissive/ambient, +/* Update on colormaterial, material emmissive/ambient, * lightmodel.globalambient */ static void update_global_ambient( GLcontext *ctx ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); float *fcmd = (float *)RADEON_DB_STATE( glt ); /* Need to do more if both emmissive & ambient are PREMULT: @@ -782,23 +675,23 @@ static void update_global_ambient( GLcontext *ctx ) */ if ((rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] & ((3 << RADEON_EMISSIVE_SOURCE_SHIFT) | - (3 << RADEON_AMBIENT_SOURCE_SHIFT))) == 0) + (3 << RADEON_AMBIENT_SOURCE_SHIFT))) == 0) { - COPY_3V( &fcmd[GLT_RED], + COPY_3V( &fcmd[GLT_RED], ctx->Light.Material.Attrib[MAT_ATTRIB_FRONT_EMISSION]); ACC_SCALE_3V( &fcmd[GLT_RED], ctx->Light.Model.Ambient, ctx->Light.Material.Attrib[MAT_ATTRIB_FRONT_AMBIENT]); - } + } else { COPY_3V( &fcmd[GLT_RED], ctx->Light.Model.Ambient ); } - + RADEON_DB_STATECHANGE(rmesa, &rmesa->hw.glt); } -/* Update on change to +/* Update on change to * - light[p].colors * - light[p].enabled */ @@ -809,13 +702,13 @@ static void update_light_colors( GLcontext *ctx, GLuint p ) /* fprintf(stderr, "%s\n", __FUNCTION__); */ if (l->Enabled) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); float *fcmd = (float *)RADEON_DB_STATE( lit[p] ); - COPY_4V( &fcmd[LIT_AMBIENT_RED], l->Ambient ); + COPY_4V( &fcmd[LIT_AMBIENT_RED], l->Ambient ); COPY_4V( &fcmd[LIT_DIFFUSE_RED], l->Diffuse ); COPY_4V( &fcmd[LIT_SPECULAR_RED], l->Specular ); - + RADEON_DB_STATECHANGE( rmesa, &rmesa->hw.lit[p] ); } } @@ -829,7 +722,7 @@ static void check_twoside_fallback( GLcontext *ctx ) if (ctx->Light.Enabled && ctx->Light.Model.TwoSide) { if (ctx->Light.ColorMaterialEnabled && - (ctx->Light.ColorMaterialBitmask & BACK_MATERIAL_BITS) != + (ctx->Light.ColorMaterialBitmask & BACK_MATERIAL_BITS) != ((ctx->Light.ColorMaterialBitmask & FRONT_MATERIAL_BITS)<<1)) fallback = GL_TRUE; else { @@ -837,7 +730,7 @@ static void check_twoside_fallback( GLcontext *ctx ) if (memcmp( ctx->Light.Material.Attrib[i], ctx->Light.Material.Attrib[i+1], sizeof(GLfloat)*4) != 0) { - fallback = GL_TRUE; + fallback = GL_TRUE; break; } } @@ -849,14 +742,14 @@ static void check_twoside_fallback( GLcontext *ctx ) static void radeonColorMaterial( GLcontext *ctx, GLenum face, GLenum mode ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); GLuint light_model_ctl1 = rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL]; light_model_ctl1 &= ~((3 << RADEON_EMISSIVE_SOURCE_SHIFT) | (3 << RADEON_AMBIENT_SOURCE_SHIFT) | (3 << RADEON_DIFFUSE_SOURCE_SHIFT) | - (3 << RADEON_SPECULAR_SOURCE_SHIFT)); - + (3 << RADEON_SPECULAR_SOURCE_SHIFT)); + if (ctx->Light.ColorMaterialEnabled) { GLuint mask = ctx->Light.ColorMaterialBitmask; @@ -877,7 +770,7 @@ static void radeonColorMaterial( GLcontext *ctx, GLenum face, GLenum mode ) light_model_ctl1 |= (RADEON_LM_SOURCE_STATE_MULT << RADEON_AMBIENT_SOURCE_SHIFT); } - + if (mask & MAT_BIT_FRONT_DIFFUSE) { light_model_ctl1 |= (RADEON_LM_SOURCE_VERTEX_DIFFUSE << RADEON_DIFFUSE_SOURCE_SHIFT); @@ -886,7 +779,7 @@ static void radeonColorMaterial( GLcontext *ctx, GLenum face, GLenum mode ) light_model_ctl1 |= (RADEON_LM_SOURCE_STATE_MULT << RADEON_DIFFUSE_SOURCE_SHIFT); } - + if (mask & MAT_BIT_FRONT_SPECULAR) { light_model_ctl1 |= (RADEON_LM_SOURCE_VERTEX_DIFFUSE << RADEON_SPECULAR_SOURCE_SHIFT); @@ -904,27 +797,27 @@ static void radeonColorMaterial( GLcontext *ctx, GLenum face, GLenum mode ) (RADEON_LM_SOURCE_STATE_MULT << RADEON_DIFFUSE_SOURCE_SHIFT) | (RADEON_LM_SOURCE_STATE_MULT << RADEON_SPECULAR_SOURCE_SHIFT); } - + if (light_model_ctl1 != rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL]) { RADEON_STATECHANGE( rmesa, tcl ); - rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] = light_model_ctl1; + rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] = light_model_ctl1; } } void radeonUpdateMaterial( GLcontext *ctx ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); GLfloat (*mat)[4] = ctx->Light.Material.Attrib; GLfloat *fcmd = (GLfloat *)RADEON_DB_STATE( mtl ); GLuint mask = ~0; - + if (ctx->Light.ColorMaterialEnabled) mask &= ~ctx->Light.ColorMaterialBitmask; - if (RADEON_DEBUG & DEBUG_STATE) + if (RADEON_DEBUG & RADEON_STATE) fprintf(stderr, "%s\n", __FUNCTION__); - + if (mask & MAT_BIT_FRONT_EMISSION) { fcmd[MTL_EMMISSIVE_RED] = mat[MAT_ATTRIB_FRONT_EMISSION][0]; fcmd[MTL_EMMISSIVE_GREEN] = mat[MAT_ATTRIB_FRONT_EMISSION][1]; @@ -974,11 +867,11 @@ void radeonUpdateMaterial( GLcontext *ctx ) * * which are calculated in light.c and are correct for the current * lighting space (model or eye), hence dependencies on _NEW_MODELVIEW - * and _MESA_NEW_NEED_EYE_COORDS. + * and _MESA_NEW_NEED_EYE_COORDS. */ static void update_light( GLcontext *ctx ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); /* Have to check these, or have an automatic shortcircuit mechanism * to remove noop statechanges. (Or just do a better job on the @@ -991,12 +884,12 @@ static void update_light( GLcontext *ctx ) tmp &= ~RADEON_LIGHT_IN_MODELSPACE; else tmp |= RADEON_LIGHT_IN_MODELSPACE; - + /* Leave this test disabled: (unexplained q3 lockup) (even with new packets) */ - if (tmp != rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL]) + if (tmp != rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL]) { RADEON_STATECHANGE( rmesa, tcl ); rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] = tmp; @@ -1020,10 +913,10 @@ static void update_light( GLcontext *ctx ) if (ctx->Light.Light[p].Enabled) { struct gl_light *l = &ctx->Light.Light[p]; GLfloat *fcmd = (GLfloat *)RADEON_DB_STATE( lit[p] ); - + if (l->EyePosition[3] == 0.0) { - COPY_3FV( &fcmd[LIT_POSITION_X], l->_VP_inf_norm ); - COPY_3FV( &fcmd[LIT_DIRECTION_X], l->_h_inf_norm ); + COPY_3FV( &fcmd[LIT_POSITION_X], l->_VP_inf_norm ); + COPY_3FV( &fcmd[LIT_DIRECTION_X], l->_h_inf_norm ); fcmd[LIT_POSITION_W] = 0; fcmd[LIT_DIRECTION_W] = 0; } else { @@ -1043,30 +936,30 @@ static void update_light( GLcontext *ctx ) static void radeonLightfv( GLcontext *ctx, GLenum light, GLenum pname, const GLfloat *params ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); GLint p = light - GL_LIGHT0; struct gl_light *l = &ctx->Light.Light[p]; GLfloat *fcmd = (GLfloat *)rmesa->hw.lit[p].cmd; - + switch (pname) { - case GL_AMBIENT: + case GL_AMBIENT: case GL_DIFFUSE: case GL_SPECULAR: update_light_colors( ctx, p ); break; - case GL_SPOT_DIRECTION: - /* picked up in update_light */ + case GL_SPOT_DIRECTION: + /* picked up in update_light */ break; case GL_POSITION: { - /* positions picked up in update_light, but can do flag here */ + /* positions picked up in update_light, but can do flag here */ GLuint flag; GLuint idx = TCL_PER_LIGHT_CTL_0 + p/2; /* FIXME: Set RANGE_ATTEN only when needed */ - if (p&1) + if (p&1) flag = RADEON_LIGHT_1_IS_LOCAL; else flag = RADEON_LIGHT_0_IS_LOCAL; @@ -1158,16 +1051,16 @@ static void radeonLightfv( GLcontext *ctx, GLenum light, } } - + static void radeonLightModelfv( GLcontext *ctx, GLenum pname, const GLfloat *param ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); switch (pname) { - case GL_LIGHT_MODEL_AMBIENT: + case GL_LIGHT_MODEL_AMBIENT: update_global_ambient( ctx ); break; @@ -1188,7 +1081,7 @@ static void radeonLightModelfv( GLcontext *ctx, GLenum pname, check_twoside_fallback( ctx ); - if (rmesa->TclFallback) { + if (rmesa->radeon.TclFallback) { radeonChooseRenderState( ctx ); radeonChooseVertexState( ctx ); } @@ -1205,7 +1098,7 @@ static void radeonLightModelfv( GLcontext *ctx, GLenum pname, static void radeonShadeModel( GLcontext *ctx, GLenum mode ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); GLuint s = rmesa->hw.set.cmd[SET_SE_CNTL]; s &= ~(RADEON_DIFFUSE_SHADE_MASK | @@ -1244,7 +1137,7 @@ static void radeonShadeModel( GLcontext *ctx, GLenum mode ) static void radeonClipPlane( GLcontext *ctx, GLenum plane, const GLfloat *eq ) { GLint p = (GLint) plane - (GLint) GL_CLIP_PLANE0; - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); GLint *ip = (GLint *)ctx->Transform._ClipUserPlane[p]; RADEON_STATECHANGE( rmesa, ucp[p] ); @@ -1256,7 +1149,7 @@ static void radeonClipPlane( GLcontext *ctx, GLenum plane, const GLfloat *eq ) static void radeonUpdateClipPlanes( GLcontext *ctx ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); GLuint p; for (p = 0; p < ctx->Const.MaxClipPlanes; p++) { @@ -1281,7 +1174,7 @@ static void radeonStencilFuncSeparate( GLcontext *ctx, GLenum face, GLenum func, GLint ref, GLuint mask ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); GLuint refmask = (((ctx->Stencil.Ref[0] & 0xff) << RADEON_STENCIL_REF_SHIFT) | ((ctx->Stencil.ValueMask[0] & 0xff) << RADEON_STENCIL_MASK_SHIFT)); @@ -1325,7 +1218,7 @@ radeonStencilFuncSeparate( GLcontext *ctx, GLenum face, GLenum func, static void radeonStencilMaskSeparate( GLcontext *ctx, GLenum face, GLuint mask ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); RADEON_STATECHANGE( rmesa, msk ); rmesa->hw.msk.cmd[MSK_RB3D_STENCILREFMASK] &= ~RADEON_STENCIL_WRITE_MASK; @@ -1336,20 +1229,20 @@ radeonStencilMaskSeparate( GLcontext *ctx, GLenum face, GLuint mask ) static void radeonStencilOpSeparate( GLcontext *ctx, GLenum face, GLenum fail, GLenum zfail, GLenum zpass ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); /* radeon 7200 have stencil bug, DEC and INC_WRAP will actually both do DEC_WRAP, and DEC_WRAP (and INVERT) will do INVERT. No way to get correct INC_WRAP and DEC, but DEC_WRAP can be fixed by using DEC and INC_WRAP at least use INC. */ - + GLuint tempRADEON_STENCIL_FAIL_DEC_WRAP; GLuint tempRADEON_STENCIL_FAIL_INC_WRAP; GLuint tempRADEON_STENCIL_ZFAIL_DEC_WRAP; GLuint tempRADEON_STENCIL_ZFAIL_INC_WRAP; GLuint tempRADEON_STENCIL_ZPASS_DEC_WRAP; GLuint tempRADEON_STENCIL_ZPASS_INC_WRAP; - - if (rmesa->radeonScreen->chip_flags & RADEON_CHIPSET_BROKEN_STENCIL) { + + if (rmesa->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_BROKEN_STENCIL) { tempRADEON_STENCIL_FAIL_DEC_WRAP = RADEON_STENCIL_FAIL_DEC; tempRADEON_STENCIL_FAIL_INC_WRAP = RADEON_STENCIL_FAIL_INC; tempRADEON_STENCIL_ZFAIL_DEC_WRAP = RADEON_STENCIL_ZFAIL_DEC; @@ -1365,7 +1258,7 @@ static void radeonStencilOpSeparate( GLcontext *ctx, GLenum face, GLenum fail, tempRADEON_STENCIL_ZPASS_DEC_WRAP = RADEON_STENCIL_ZPASS_DEC_WRAP; tempRADEON_STENCIL_ZPASS_INC_WRAP = RADEON_STENCIL_ZPASS_INC_WRAP; } - + RADEON_STATECHANGE( rmesa, ctx ); rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] &= ~(RADEON_STENCIL_FAIL_MASK | RADEON_STENCIL_ZFAIL_MASK | @@ -1455,9 +1348,9 @@ static void radeonStencilOpSeparate( GLcontext *ctx, GLenum face, GLenum fail, static void radeonClearStencil( GLcontext *ctx, GLint s ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); - rmesa->state.stencil.clear = + rmesa->radeon.state.stencil.clear = ((GLuint) (ctx->Stencil.Clear & 0xff) | (0xff << RADEON_STENCIL_MASK_SHIFT) | ((ctx->Stencil.WriteMask[0] & 0xff) << RADEON_STENCIL_WRITEMASK_SHIFT)); @@ -1481,20 +1374,30 @@ static void radeonClearStencil( GLcontext *ctx, GLint s ) */ void radeonUpdateWindow( GLcontext *ctx ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - __DRIdrawablePrivate *dPriv = rmesa->dri.drawable; - GLfloat xoffset = (GLfloat)dPriv->x; - GLfloat yoffset = (GLfloat)dPriv->y + dPriv->h; + r100ContextPtr rmesa = R100_CONTEXT(ctx); + __DRIdrawablePrivate *dPriv = radeon_get_drawable(&rmesa->radeon); + GLfloat xoffset = dPriv ? (GLfloat) dPriv->x : 0; + GLfloat yoffset = dPriv ? (GLfloat) dPriv->y + dPriv->h : 0; const GLfloat *v = ctx->Viewport._WindowMap.m; + const GLboolean render_to_fbo = (ctx->DrawBuffer ? (ctx->DrawBuffer->Name != 0) : 0); + const GLfloat depthScale = 1.0F / ctx->DrawBuffer->_DepthMaxF; + GLfloat y_scale, y_bias; + + if (render_to_fbo) { + y_scale = 1.0; + y_bias = 0; + } else { + y_scale = -1.0; + y_bias = yoffset; + } float_ui32_type sx = { v[MAT_SX] }; float_ui32_type tx = { v[MAT_TX] + xoffset + SUBPIXEL_X }; - float_ui32_type sy = { - v[MAT_SY] }; - float_ui32_type ty = { (- v[MAT_TY]) + yoffset + SUBPIXEL_Y }; - float_ui32_type sz = { v[MAT_SZ] * rmesa->state.depth.scale }; - float_ui32_type tz = { v[MAT_TZ] * rmesa->state.depth.scale }; + float_ui32_type sy = { v[MAT_SY] * y_scale }; + float_ui32_type ty = { (v[MAT_TY] * y_scale) + y_bias + SUBPIXEL_Y }; + float_ui32_type sz = { v[MAT_SZ] * depthScale }; + float_ui32_type tz = { v[MAT_TZ] * depthScale }; - RADEON_FIREVERTICES( rmesa ); RADEON_STATECHANGE( rmesa, vpt ); rmesa->hw.vpt.cmd[VPT_SE_VPORT_XSCALE] = sx.ui32; @@ -1514,6 +1417,8 @@ static void radeonViewport( GLcontext *ctx, GLint x, GLint y, * values, or keep the originals hanging around. */ radeonUpdateWindow( ctx ); + + radeon_viewport(ctx, x, y, width, height); } static void radeonDepthRange( GLcontext *ctx, GLclampd nearval, @@ -1524,8 +1429,8 @@ static void radeonDepthRange( GLcontext *ctx, GLclampd nearval, void radeonUpdateViewportOffset( GLcontext *ctx ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - __DRIdrawablePrivate *dPriv = rmesa->dri.drawable; + r100ContextPtr rmesa = R100_CONTEXT(ctx); + __DRIdrawablePrivate *dPriv = radeon_get_drawable(&rmesa->radeon); GLfloat xoffset = (GLfloat)dPriv->x; GLfloat yoffset = (GLfloat)dPriv->y + dPriv->h; const GLfloat *v = ctx->Viewport._WindowMap.m; @@ -1555,8 +1460,8 @@ void radeonUpdateViewportOffset( GLcontext *ctx ) RADEON_STIPPLE_Y_OFFSET_MASK); /* add magic offsets, then invert */ - stx = 31 - ((rmesa->dri.drawable->x - 1) & RADEON_STIPPLE_COORD_MASK); - sty = 31 - ((rmesa->dri.drawable->y + rmesa->dri.drawable->h - 1) + stx = 31 - ((dPriv->x - 1) & RADEON_STIPPLE_COORD_MASK); + sty = 31 - ((dPriv->y + dPriv->h - 1) & RADEON_STIPPLE_COORD_MASK); m |= ((stx << RADEON_STIPPLE_X_OFFSET_SHIFT) | @@ -1580,20 +1485,26 @@ void radeonUpdateViewportOffset( GLcontext *ctx ) static void radeonClearColor( GLcontext *ctx, const GLfloat color[4] ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); GLubyte c[4]; + struct radeon_renderbuffer *rrb; + + rrb = radeon_get_colorbuffer(&rmesa->radeon); + if (!rrb) + return; + CLAMPED_FLOAT_TO_UBYTE(c[0], color[0]); CLAMPED_FLOAT_TO_UBYTE(c[1], color[1]); CLAMPED_FLOAT_TO_UBYTE(c[2], color[2]); CLAMPED_FLOAT_TO_UBYTE(c[3], color[3]); - rmesa->state.color.clear = radeonPackColor( rmesa->radeonScreen->cpp, + rmesa->radeon.state.color.clear = radeonPackColor( rrb->cpp, c[0], c[1], c[2], c[3] ); } static void radeonRenderMode( GLcontext *ctx, GLenum mode ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); FALLBACK( rmesa, RADEON_FALLBACK_RENDER_MODE, (mode != GL_RENDER) ); } @@ -1619,7 +1530,7 @@ static GLuint radeon_rop_tab[] = { static void radeonLogicOpCode( GLcontext *ctx, GLenum opcode ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); GLuint rop = (GLuint)opcode - GL_CLEAR; ASSERT( rop < 16 ); @@ -1628,108 +1539,16 @@ static void radeonLogicOpCode( GLcontext *ctx, GLenum opcode ) rmesa->hw.msk.cmd[MSK_RB3D_ROPCNTL] = radeon_rop_tab[rop]; } - -/** - * Set up the cliprects for either front or back-buffer drawing. - */ -void radeonSetCliprects( radeonContextPtr rmesa ) -{ - __DRIdrawablePrivate *const drawable = rmesa->dri.drawable; - __DRIdrawablePrivate *const readable = rmesa->dri.readable; - GLframebuffer *const draw_fb = (GLframebuffer*) drawable->driverPrivate; - GLframebuffer *const read_fb = (GLframebuffer*) readable->driverPrivate; - - if (draw_fb->_ColorDrawBufferIndexes[0] == BUFFER_BACK_LEFT) { - /* Can't ignore 2d windows if we are page flipping. - */ - if ( drawable->numBackClipRects == 0 || rmesa->doPageFlip ) { - rmesa->numClipRects = drawable->numClipRects; - rmesa->pClipRects = drawable->pClipRects; - } - else { - rmesa->numClipRects = drawable->numBackClipRects; - rmesa->pClipRects = drawable->pBackClipRects; - } - } - else { - /* front buffer (or none, or multiple buffers */ - rmesa->numClipRects = drawable->numClipRects; - rmesa->pClipRects = drawable->pClipRects; - } - - if ((draw_fb->Width != drawable->w) || (draw_fb->Height != drawable->h)) { - _mesa_resize_framebuffer(rmesa->glCtx, draw_fb, - drawable->w, drawable->h); - draw_fb->Initialized = GL_TRUE; - } - - if (drawable != readable) { - if ((read_fb->Width != readable->w) || (read_fb->Height != readable->h)) { - _mesa_resize_framebuffer(rmesa->glCtx, read_fb, - readable->w, readable->h); - read_fb->Initialized = GL_TRUE; - } - } - - if (rmesa->state.scissor.enabled) - radeonRecalcScissorRects( rmesa ); - - rmesa->lastStamp = drawable->lastStamp; -} - - -/** - * Called via glDrawBuffer. - */ -static void radeonDrawBuffer( GLcontext *ctx, GLenum mode ) -{ - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - - if (RADEON_DEBUG & DEBUG_DRI) - fprintf(stderr, "%s %s\n", __FUNCTION__, - _mesa_lookup_enum_by_nr( mode )); - - RADEON_FIREVERTICES(rmesa); /* don't pipeline cliprect changes */ - - if (ctx->DrawBuffer->_NumColorDrawBuffers != 1) { - /* 0 (GL_NONE) buffers or multiple color drawing buffers */ - FALLBACK( rmesa, RADEON_FALLBACK_DRAW_BUFFER, GL_TRUE ); - return; - } - - switch ( ctx->DrawBuffer->_ColorDrawBufferIndexes[0] ) { - case BUFFER_FRONT_LEFT: - case BUFFER_BACK_LEFT: - FALLBACK( rmesa, RADEON_FALLBACK_DRAW_BUFFER, GL_FALSE ); - break; - default: - FALLBACK( rmesa, RADEON_FALLBACK_DRAW_BUFFER, GL_TRUE ); - return; - } - - radeonSetCliprects( rmesa ); - - /* We'll set the drawing engine's offset/pitch parameters later - * when we update other state. - */ -} - -static void radeonReadBuffer( GLcontext *ctx, GLenum mode ) -{ - /* nothing, until we implement h/w glRead/CopyPixels or CopyTexImage */ -} - - /* ============================================================= * State enable/disable */ static void radeonEnable( GLcontext *ctx, GLenum cap, GLboolean state ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); GLuint p, flag; - if ( RADEON_DEBUG & DEBUG_STATE ) + if ( RADEON_DEBUG & RADEON_STATE ) fprintf( stderr, "%s( %s = %s )\n", __FUNCTION__, _mesa_lookup_enum_by_nr( cap ), state ? "GL_TRUE" : "GL_FALSE" ); @@ -1787,7 +1606,7 @@ static void radeonEnable( GLcontext *ctx, GLenum cap, GLboolean state ) case GL_CLIP_PLANE2: case GL_CLIP_PLANE3: case GL_CLIP_PLANE4: - case GL_CLIP_PLANE5: + case GL_CLIP_PLANE5: p = cap-GL_CLIP_PLANE0; RADEON_STATECHANGE( rmesa, tcl ); if (state) { @@ -1821,10 +1640,10 @@ static void radeonEnable( GLcontext *ctx, GLenum cap, GLboolean state ) RADEON_STATECHANGE(rmesa, ctx ); if ( state ) { rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |= RADEON_DITHER_ENABLE; - rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] &= ~rmesa->state.color.roundEnable; + rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] &= ~rmesa->radeon.state.color.roundEnable; } else { rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] &= ~RADEON_DITHER_ENABLE; - rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |= rmesa->state.color.roundEnable; + rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |= rmesa->radeon.state.color.roundEnable; } break; @@ -1852,13 +1671,13 @@ static void radeonEnable( GLcontext *ctx, GLenum cap, GLboolean state ) case GL_LIGHT7: RADEON_STATECHANGE(rmesa, tcl); p = cap - GL_LIGHT0; - if (p&1) + if (p&1) flag = (RADEON_LIGHT_1_ENABLE | - RADEON_LIGHT_1_ENABLE_AMBIENT | + RADEON_LIGHT_1_ENABLE_AMBIENT | RADEON_LIGHT_1_ENABLE_SPECULAR); else flag = (RADEON_LIGHT_0_ENABLE | - RADEON_LIGHT_0_ENABLE_AMBIENT | + RADEON_LIGHT_0_ENABLE_AMBIENT | RADEON_LIGHT_0_ENABLE_SPECULAR); if (state) @@ -1866,7 +1685,7 @@ static void radeonEnable( GLcontext *ctx, GLenum cap, GLboolean state ) else rmesa->hw.tcl.cmd[p/2 + TCL_PER_LIGHT_CTL_0] &= ~flag; - /* + /* */ update_light_colors( ctx, p ); break; @@ -1904,7 +1723,7 @@ static void radeonEnable( GLcontext *ctx, GLenum cap, GLboolean state ) rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] &= ~RADEON_ROP_ENABLE; } break; - + case GL_NORMALIZE: RADEON_STATECHANGE( rmesa, tcl ); if ( state ) { @@ -1971,21 +1790,30 @@ static void radeonEnable( GLcontext *ctx, GLenum cap, GLboolean state ) } case GL_SCISSOR_TEST: - RADEON_FIREVERTICES( rmesa ); - rmesa->state.scissor.enabled = state; + radeon_firevertices(&rmesa->radeon); + rmesa->radeon.state.scissor.enabled = state; radeonUpdateScissor( ctx ); break; case GL_STENCIL_TEST: - if ( rmesa->state.stencil.hwBuffer ) { - RADEON_STATECHANGE( rmesa, ctx ); - if ( state ) { - rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |= RADEON_STENCIL_ENABLE; + { + GLboolean hw_stencil = GL_FALSE; + if (ctx->DrawBuffer) { + struct radeon_renderbuffer *rrbStencil + = radeon_get_renderbuffer(ctx->DrawBuffer, BUFFER_STENCIL); + hw_stencil = (rrbStencil && rrbStencil->bo); + } + + if (hw_stencil) { + RADEON_STATECHANGE( rmesa, ctx ); + if ( state ) { + rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |= RADEON_STENCIL_ENABLE; + } else { + rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] &= ~RADEON_STENCIL_ENABLE; + } } else { - rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] &= ~RADEON_STENCIL_ENABLE; + FALLBACK( rmesa, RADEON_FALLBACK_STENCIL, state ); } - } else { - FALLBACK( rmesa, RADEON_FALLBACK_STENCIL, state ); } break; @@ -1995,7 +1823,7 @@ static void radeonEnable( GLcontext *ctx, GLenum cap, GLboolean state ) case GL_TEXTURE_GEN_T: /* Picked up in radeonUpdateTextureState. */ - rmesa->recheck_texgen[ctx->Texture.CurrentUnit] = GL_TRUE; + rmesa->recheck_texgen[ctx->Texture.CurrentUnit] = GL_TRUE; break; case GL_COLOR_SUM_EXT: @@ -2010,11 +1838,11 @@ static void radeonEnable( GLcontext *ctx, GLenum cap, GLboolean state ) static void radeonLightingSpaceChange( GLcontext *ctx ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); GLboolean tmp; RADEON_STATECHANGE( rmesa, tcl ); - if (RADEON_DEBUG & DEBUG_STATE) + if (RADEON_DEBUG & RADEON_STATE) fprintf(stderr, "%s %d BEFORE %x\n", __FUNCTION__, ctx->_NeedEyeCoords, rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL]); @@ -2029,7 +1857,7 @@ static void radeonLightingSpaceChange( GLcontext *ctx ) rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] &= ~RADEON_RESCALE_NORMALS; } - if (RADEON_DEBUG & DEBUG_STATE) + if (RADEON_DEBUG & RADEON_STATE) fprintf(stderr, "%s %d AFTER %x\n", __FUNCTION__, ctx->_NeedEyeCoords, rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL]); } @@ -2039,7 +1867,7 @@ static void radeonLightingSpaceChange( GLcontext *ctx ) */ -void radeonUploadTexMatrix( radeonContextPtr rmesa, +void radeonUploadTexMatrix( r100ContextPtr rmesa, int unit, GLboolean swapcols ) { /* Here's how this works: on r100, only 3 tex coords can be submitted, so the @@ -2065,7 +1893,7 @@ void radeonUploadTexMatrix( radeonContextPtr rmesa, int idx = TEXMAT_0 + unit; float *dest = ((float *)RADEON_DB_STATE( mat[idx] )) + MAT_ELT_0; int i; - struct gl_texture_unit tUnit = rmesa->glCtx->Texture.Unit[unit]; + struct gl_texture_unit tUnit = rmesa->radeon.glCtx->Texture.Unit[unit]; GLfloat *src = rmesa->tmpmat[unit].m; rmesa->TexMatColSwap &= ~(1 << unit); @@ -2119,7 +1947,7 @@ void radeonUploadTexMatrix( radeonContextPtr rmesa, } -static void upload_matrix( radeonContextPtr rmesa, GLfloat *src, int idx ) +static void upload_matrix( r100ContextPtr rmesa, GLfloat *src, int idx ) { float *dest = ((float *)RADEON_DB_STATE( mat[idx] ))+MAT_ELT_0; int i; @@ -2135,7 +1963,7 @@ static void upload_matrix( radeonContextPtr rmesa, GLfloat *src, int idx ) RADEON_DB_STATECHANGE( rmesa, &rmesa->hw.mat[idx] ); } -static void upload_matrix_t( radeonContextPtr rmesa, GLfloat *src, int idx ) +static void upload_matrix_t( r100ContextPtr rmesa, GLfloat *src, int idx ) { float *dest = ((float *)RADEON_DB_STATE( mat[idx] ))+MAT_ELT_0; memcpy(dest, src, 16*sizeof(float)); @@ -2145,7 +1973,7 @@ static void upload_matrix_t( radeonContextPtr rmesa, GLfloat *src, int idx ) static void update_texturematrix( GLcontext *ctx ) { - radeonContextPtr rmesa = RADEON_CONTEXT( ctx ); + r100ContextPtr rmesa = R100_CONTEXT( ctx ); GLuint tpc = rmesa->hw.tcl.cmd[TCL_TEXTURE_PROC_CTL]; GLuint vs = rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXSEL]; int unit; @@ -2209,64 +2037,74 @@ static void update_texturematrix( GLcontext *ctx ) } } - -/** - * Tell the card where to render (offset, pitch). - * Effected by glDrawBuffer, etc - */ -void -radeonUpdateDrawBuffer(GLcontext *ctx) +static GLboolean r100ValidateBuffers(GLcontext *ctx) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - struct gl_framebuffer *fb = ctx->DrawBuffer; - driRenderbuffer *drb; + r100ContextPtr rmesa = R100_CONTEXT(ctx); + struct radeon_renderbuffer *rrb; + int i, ret; - if (fb->_ColorDrawBufferIndexes[0] == BUFFER_FRONT_LEFT) { - /* draw to front */ - drb = (driRenderbuffer *) fb->Attachment[BUFFER_FRONT_LEFT].Renderbuffer; - } - else if (fb->_ColorDrawBufferIndexes[0] == BUFFER_BACK_LEFT) { - /* draw to back */ - drb = (driRenderbuffer *) fb->Attachment[BUFFER_BACK_LEFT].Renderbuffer; + radeon_cs_space_reset_bos(rmesa->radeon.cmdbuf.cs); + + rrb = radeon_get_colorbuffer(&rmesa->radeon); + /* color buffer */ + if (rrb && rrb->bo) { + radeon_cs_space_add_persistent_bo(rmesa->radeon.cmdbuf.cs, rrb->bo, + 0, RADEON_GEM_DOMAIN_VRAM); } - else { - /* drawing to multiple buffers, or none */ - return; + + /* depth buffer */ + rrb = radeon_get_depthbuffer(&rmesa->radeon); + /* color buffer */ + if (rrb && rrb->bo) { + radeon_cs_space_add_persistent_bo(rmesa->radeon.cmdbuf.cs, rrb->bo, + 0, RADEON_GEM_DOMAIN_VRAM); } - assert(drb); - assert(drb->flippedPitch); + for (i = 0; i < ctx->Const.MaxTextureImageUnits; ++i) { + radeonTexObj *t; - RADEON_STATECHANGE( rmesa, ctx ); + if (!ctx->Texture.Unit[i]._ReallyEnabled) + continue; - /* Note: we used the (possibly) page-flipped values */ - rmesa->hw.ctx.cmd[CTX_RB3D_COLOROFFSET] - = ((drb->flippedOffset + rmesa->radeonScreen->fbLocation) - & RADEON_COLOROFFSET_MASK); - rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] = drb->flippedPitch; - if (rmesa->sarea->tiling_enabled) { - rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] |= RADEON_COLOR_TILE_ENABLE; + t = rmesa->state.texture.unit[i].texobj; + if (t->image_override && t->bo) + radeon_cs_space_add_persistent_bo(rmesa->radeon.cmdbuf.cs, t->bo, + RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0); + else if (t->mt->bo) + radeon_cs_space_add_persistent_bo(rmesa->radeon.cmdbuf.cs, t->mt->bo, + RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0); } -} + ret = radeon_cs_space_check_with_bo(rmesa->radeon.cmdbuf.cs, first_elem(&rmesa->radeon.dma.reserved)->bo, RADEON_GEM_DOMAIN_GTT, 0); + if (ret) + return GL_FALSE; + return GL_TRUE; +} -void radeonValidateState( GLcontext *ctx ) +GLboolean radeonValidateState( GLcontext *ctx ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - GLuint new_state = rmesa->NewGLState; + r100ContextPtr rmesa = R100_CONTEXT(ctx); + GLuint new_state = rmesa->radeon.NewGLState; - if (new_state & (_NEW_BUFFERS | _NEW_COLOR | _NEW_PIXEL)) { - radeonUpdateDrawBuffer(ctx); + if (new_state & _NEW_BUFFERS) { + _mesa_update_framebuffer(ctx); + /* this updates the DrawBuffer's Width/Height if it's a FBO */ + _mesa_update_draw_buffer_bounds(ctx); + RADEON_STATECHANGE(rmesa, ctx); } if (new_state & _NEW_TEXTURE) { radeonUpdateTextureState( ctx ); - new_state |= rmesa->NewGLState; /* may add TEXTURE_MATRIX */ + new_state |= rmesa->radeon.NewGLState; /* may add TEXTURE_MATRIX */ } + /* we need to do a space check here */ + if (!r100ValidateBuffers(ctx)) + return GL_FALSE; + /* Need an event driven matrix update? */ - if (new_state & (_NEW_MODELVIEW|_NEW_PROJECTION)) + if (new_state & (_NEW_MODELVIEW|_NEW_PROJECTION)) upload_matrix( rmesa, ctx->_ModelProjectMatrix.m, MODEL_PROJ ); /* Need these for lighting (shouldn't upload otherwise) @@ -2290,12 +2128,14 @@ void radeonValidateState( GLcontext *ctx ) /* emit all active clip planes if projection matrix changes. */ if (new_state & (_NEW_PROJECTION)) { - if (ctx->Transform.ClipPlanesEnabled) + if (ctx->Transform.ClipPlanesEnabled) radeonUpdateClipPlanes( ctx ); } - rmesa->NewGLState = 0; + rmesa->radeon.NewGLState = 0; + + return GL_TRUE; } @@ -2306,7 +2146,7 @@ static void radeonInvalidateState( GLcontext *ctx, GLuint new_state ) _vbo_InvalidateState( ctx, new_state ); _tnl_InvalidateState( ctx, new_state ); _ae_invalidate_state( ctx, new_state ); - RADEON_CONTEXT(ctx)->NewGLState |= new_state; + R100_CONTEXT(ctx)->radeon.NewGLState |= new_state; } @@ -2317,8 +2157,8 @@ static GLboolean check_material( GLcontext *ctx ) TNLcontext *tnl = TNL_CONTEXT(ctx); GLint i; - for (i = _TNL_ATTRIB_MAT_FRONT_AMBIENT; - i < _TNL_ATTRIB_MAT_BACK_INDEXES; + for (i = _TNL_ATTRIB_MAT_FRONT_AMBIENT; + i < _TNL_ATTRIB_MAT_BACK_INDEXES; i++) if (tnl->vb.AttribPtr[i] && tnl->vb.AttribPtr[i]->stride) @@ -2326,20 +2166,21 @@ static GLboolean check_material( GLcontext *ctx ) return GL_FALSE; } - + static void radeonWrapRunPipeline( GLcontext *ctx ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); GLboolean has_material; if (0) - fprintf(stderr, "%s, newstate: %x\n", __FUNCTION__, rmesa->NewGLState); + fprintf(stderr, "%s, newstate: %x\n", __FUNCTION__, rmesa->radeon.NewGLState); /* Validate state: */ - if (rmesa->NewGLState) - radeonValidateState( ctx ); + if (rmesa->radeon.NewGLState) + if (!radeonValidateState( ctx )) + FALLBACK(rmesa, RADEON_FALLBACK_TEXTURE, GL_TRUE); has_material = (ctx->Light.Enabled && check_material( ctx )); @@ -2348,7 +2189,7 @@ static void radeonWrapRunPipeline( GLcontext *ctx ) } /* Run the pipeline. - */ + */ _tnl_run_pipeline( ctx ); if (has_material) { @@ -2356,12 +2197,28 @@ static void radeonWrapRunPipeline( GLcontext *ctx ) } } +static void radeonPolygonStipple( GLcontext *ctx, const GLubyte *mask ) +{ + r100ContextPtr r100 = R100_CONTEXT(ctx); + GLint i; + + radeon_firevertices(&r100->radeon); + + RADEON_STATECHANGE(r100, stp); + + /* Must flip pattern upside down. + */ + for ( i = 31 ; i >= 0; i--) { + r100->hw.stp.cmd[3 + i] = ((GLuint *) mask)[i]; + } +} + /* Initialize the driver's state functions. * Many of the ctx->Driver functions might have been initialized to * software defaults in the earlier _mesa_init_driver_functions() call. */ -void radeonInitStateFuncs( GLcontext *ctx ) +void radeonInitStateFuncs( GLcontext *ctx , GLboolean dri2 ) { ctx->Driver.UpdateState = radeonInvalidateState; ctx->Driver.LightingSpaceChange = radeonLightingSpaceChange; @@ -2394,7 +2251,10 @@ void radeonInitStateFuncs( GLcontext *ctx ) ctx->Driver.LogicOpcode = radeonLogicOpCode; ctx->Driver.PolygonMode = radeonPolygonMode; ctx->Driver.PolygonOffset = radeonPolygonOffset; - ctx->Driver.PolygonStipple = radeonPolygonStipple; + if (dri2) + ctx->Driver.PolygonStipple = radeonPolygonStipple; + else + ctx->Driver.PolygonStipple = radeonPolygonStipplePreKMS; ctx->Driver.RenderMode = radeonRenderMode; ctx->Driver.Scissor = radeonScissor; ctx->Driver.ShadeModel = radeonShadeModel; diff --git a/src/mesa/drivers/dri/radeon/radeon_state.h b/src/mesa/drivers/dri/radeon/radeon_state.h index 2171879f759..c780cff0cfb 100644 --- a/src/mesa/drivers/dri/radeon/radeon_state.h +++ b/src/mesa/drivers/dri/radeon/radeon_state.h @@ -39,30 +39,25 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "radeon_context.h" -extern void radeonInitState( radeonContextPtr rmesa ); -extern void radeonInitStateFuncs( GLcontext *ctx ); +extern void radeonInitState( r100ContextPtr rmesa ); +extern void radeonInitStateFuncs( GLcontext *ctx , GLboolean dri2); extern void radeonUpdateMaterial( GLcontext *ctx ); -extern void radeonSetCliprects( radeonContextPtr rmesa ); -extern void radeonRecalcScissorRects( radeonContextPtr rmesa ); extern void radeonUpdateViewportOffset( GLcontext *ctx ); extern void radeonUpdateWindow( GLcontext *ctx ); extern void radeonUpdateDrawBuffer( GLcontext *ctx ); -extern void radeonUploadTexMatrix( radeonContextPtr rmesa, +extern void radeonUploadTexMatrix( r100ContextPtr rmesa, int unit, GLboolean swapcols ); -extern void radeonValidateState( GLcontext *ctx ); - -extern void radeonPrintDirty( radeonContextPtr rmesa, - const char *msg ); +extern GLboolean radeonValidateState( GLcontext *ctx ); extern void radeonFallback( GLcontext *ctx, GLuint bit, GLboolean mode ); #define FALLBACK( rmesa, bit, mode ) do { \ if ( 0 ) fprintf( stderr, "FALLBACK in %s: #%d=%d\n", \ __FUNCTION__, bit, mode ); \ - radeonFallback( rmesa->glCtx, bit, mode ); \ + radeonFallback( rmesa->radeon.glCtx, bit, mode ); \ } while (0) diff --git a/src/mesa/drivers/dri/radeon/radeon_state_init.c b/src/mesa/drivers/dri/radeon/radeon_state_init.c index 57dc3800501..f3ad0dd17af 100644 --- a/src/mesa/drivers/dri/radeon/radeon_state_init.c +++ b/src/mesa/drivers/dri/radeon/radeon_state_init.c @@ -38,39 +38,141 @@ #include "swrast_setup/swrast_setup.h" #include "radeon_context.h" +#include "radeon_mipmap_tree.h" #include "radeon_ioctl.h" #include "radeon_state.h" #include "radeon_tcl.h" #include "radeon_tex.h" #include "radeon_swtcl.h" +#include "radeon_queryobj.h" + +#include "../r200/r200_reg.h" #include "xmlpool.h" +/* New (1.3) state mechanism. 3 commands (packet, scalar, vector) in + * 1.3 cmdbuffers allow all previous state to be updated as well as + * the tcl scalar and vector areas. + */ +static struct { + int start; + int len; + const char *name; +} packet[RADEON_MAX_STATE_PACKETS] = { + {RADEON_PP_MISC, 7, "RADEON_PP_MISC"}, + {RADEON_PP_CNTL, 3, "RADEON_PP_CNTL"}, + {RADEON_RB3D_COLORPITCH, 1, "RADEON_RB3D_COLORPITCH"}, + {RADEON_RE_LINE_PATTERN, 2, "RADEON_RE_LINE_PATTERN"}, + {RADEON_SE_LINE_WIDTH, 1, "RADEON_SE_LINE_WIDTH"}, + {RADEON_PP_LUM_MATRIX, 1, "RADEON_PP_LUM_MATRIX"}, + {RADEON_PP_ROT_MATRIX_0, 2, "RADEON_PP_ROT_MATRIX_0"}, + {RADEON_RB3D_STENCILREFMASK, 3, "RADEON_RB3D_STENCILREFMASK"}, + {RADEON_SE_VPORT_XSCALE, 6, "RADEON_SE_VPORT_XSCALE"}, + {RADEON_SE_CNTL, 2, "RADEON_SE_CNTL"}, + {RADEON_SE_CNTL_STATUS, 1, "RADEON_SE_CNTL_STATUS"}, + {RADEON_RE_MISC, 1, "RADEON_RE_MISC"}, + {RADEON_PP_TXFILTER_0, 6, "RADEON_PP_TXFILTER_0"}, + {RADEON_PP_BORDER_COLOR_0, 1, "RADEON_PP_BORDER_COLOR_0"}, + {RADEON_PP_TXFILTER_1, 6, "RADEON_PP_TXFILTER_1"}, + {RADEON_PP_BORDER_COLOR_1, 1, "RADEON_PP_BORDER_COLOR_1"}, + {RADEON_PP_TXFILTER_2, 6, "RADEON_PP_TXFILTER_2"}, + {RADEON_PP_BORDER_COLOR_2, 1, "RADEON_PP_BORDER_COLOR_2"}, + {RADEON_SE_ZBIAS_FACTOR, 2, "RADEON_SE_ZBIAS_FACTOR"}, + {RADEON_SE_TCL_OUTPUT_VTX_FMT, 11, "RADEON_SE_TCL_OUTPUT_VTX_FMT"}, + {RADEON_SE_TCL_MATERIAL_EMMISSIVE_RED, 17, + "RADEON_SE_TCL_MATERIAL_EMMISSIVE_RED"}, + {R200_PP_TXCBLEND_0, 4, "R200_PP_TXCBLEND_0"}, + {R200_PP_TXCBLEND_1, 4, "R200_PP_TXCBLEND_1"}, + {R200_PP_TXCBLEND_2, 4, "R200_PP_TXCBLEND_2"}, + {R200_PP_TXCBLEND_3, 4, "R200_PP_TXCBLEND_3"}, + {R200_PP_TXCBLEND_4, 4, "R200_PP_TXCBLEND_4"}, + {R200_PP_TXCBLEND_5, 4, "R200_PP_TXCBLEND_5"}, + {R200_PP_TXCBLEND_6, 4, "R200_PP_TXCBLEND_6"}, + {R200_PP_TXCBLEND_7, 4, "R200_PP_TXCBLEND_7"}, + {R200_SE_TCL_LIGHT_MODEL_CTL_0, 6, "R200_SE_TCL_LIGHT_MODEL_CTL_0"}, + {R200_PP_TFACTOR_0, 6, "R200_PP_TFACTOR_0"}, + {R200_SE_VTX_FMT_0, 4, "R200_SE_VTX_FMT_0"}, + {R200_SE_VAP_CNTL, 1, "R200_SE_VAP_CNTL"}, + {R200_SE_TCL_MATRIX_SEL_0, 5, "R200_SE_TCL_MATRIX_SEL_0"}, + {R200_SE_TCL_TEX_PROC_CTL_2, 5, "R200_SE_TCL_TEX_PROC_CTL_2"}, + {R200_SE_TCL_UCP_VERT_BLEND_CTL, 1, "R200_SE_TCL_UCP_VERT_BLEND_CTL"}, + {R200_PP_TXFILTER_0, 6, "R200_PP_TXFILTER_0"}, + {R200_PP_TXFILTER_1, 6, "R200_PP_TXFILTER_1"}, + {R200_PP_TXFILTER_2, 6, "R200_PP_TXFILTER_2"}, + {R200_PP_TXFILTER_3, 6, "R200_PP_TXFILTER_3"}, + {R200_PP_TXFILTER_4, 6, "R200_PP_TXFILTER_4"}, + {R200_PP_TXFILTER_5, 6, "R200_PP_TXFILTER_5"}, + {R200_PP_TXOFFSET_0, 1, "R200_PP_TXOFFSET_0"}, + {R200_PP_TXOFFSET_1, 1, "R200_PP_TXOFFSET_1"}, + {R200_PP_TXOFFSET_2, 1, "R200_PP_TXOFFSET_2"}, + {R200_PP_TXOFFSET_3, 1, "R200_PP_TXOFFSET_3"}, + {R200_PP_TXOFFSET_4, 1, "R200_PP_TXOFFSET_4"}, + {R200_PP_TXOFFSET_5, 1, "R200_PP_TXOFFSET_5"}, + {R200_SE_VTE_CNTL, 1, "R200_SE_VTE_CNTL"}, + {R200_SE_TCL_OUTPUT_VTX_COMP_SEL, 1, + "R200_SE_TCL_OUTPUT_VTX_COMP_SEL"}, + {R200_PP_TAM_DEBUG3, 1, "R200_PP_TAM_DEBUG3"}, + {R200_PP_CNTL_X, 1, "R200_PP_CNTL_X"}, + {R200_RB3D_DEPTHXY_OFFSET, 1, "R200_RB3D_DEPTHXY_OFFSET"}, + {R200_RE_AUX_SCISSOR_CNTL, 1, "R200_RE_AUX_SCISSOR_CNTL"}, + {R200_RE_SCISSOR_TL_0, 2, "R200_RE_SCISSOR_TL_0"}, + {R200_RE_SCISSOR_TL_1, 2, "R200_RE_SCISSOR_TL_1"}, + {R200_RE_SCISSOR_TL_2, 2, "R200_RE_SCISSOR_TL_2"}, + {R200_SE_VAP_CNTL_STATUS, 1, "R200_SE_VAP_CNTL_STATUS"}, + {R200_SE_VTX_STATE_CNTL, 1, "R200_SE_VTX_STATE_CNTL"}, + {R200_RE_POINTSIZE, 1, "R200_RE_POINTSIZE"}, + {R200_SE_TCL_INPUT_VTX_VECTOR_ADDR_0, 4, + "R200_SE_TCL_INPUT_VTX_VECTOR_ADDR_0"}, + {R200_PP_CUBIC_FACES_0, 1, "R200_PP_CUBIC_FACES_0"}, /* 61 */ + {R200_PP_CUBIC_OFFSET_F1_0, 5, "R200_PP_CUBIC_OFFSET_F1_0"}, /* 62 */ + {R200_PP_CUBIC_FACES_1, 1, "R200_PP_CUBIC_FACES_1"}, + {R200_PP_CUBIC_OFFSET_F1_1, 5, "R200_PP_CUBIC_OFFSET_F1_1"}, + {R200_PP_CUBIC_FACES_2, 1, "R200_PP_CUBIC_FACES_2"}, + {R200_PP_CUBIC_OFFSET_F1_2, 5, "R200_PP_CUBIC_OFFSET_F1_2"}, + {R200_PP_CUBIC_FACES_3, 1, "R200_PP_CUBIC_FACES_3"}, + {R200_PP_CUBIC_OFFSET_F1_3, 5, "R200_PP_CUBIC_OFFSET_F1_3"}, + {R200_PP_CUBIC_FACES_4, 1, "R200_PP_CUBIC_FACES_4"}, + {R200_PP_CUBIC_OFFSET_F1_4, 5, "R200_PP_CUBIC_OFFSET_F1_4"}, + {R200_PP_CUBIC_FACES_5, 1, "R200_PP_CUBIC_FACES_5"}, + {R200_PP_CUBIC_OFFSET_F1_5, 5, "R200_PP_CUBIC_OFFSET_F1_5"}, + {RADEON_PP_TEX_SIZE_0, 2, "RADEON_PP_TEX_SIZE_0"}, + {RADEON_PP_TEX_SIZE_1, 2, "RADEON_PP_TEX_SIZE_1"}, + {RADEON_PP_TEX_SIZE_2, 2, "RADEON_PP_TEX_SIZE_2"}, + {R200_RB3D_BLENDCOLOR, 3, "R200_RB3D_BLENDCOLOR"}, + {R200_SE_TCL_POINT_SPRITE_CNTL, 1, "R200_SE_TCL_POINT_SPRITE_CNTL"}, + {RADEON_PP_CUBIC_FACES_0, 1, "RADEON_PP_CUBIC_FACES_0"}, + {RADEON_PP_CUBIC_OFFSET_T0_0, 5, "RADEON_PP_CUBIC_OFFSET_T0_0"}, + {RADEON_PP_CUBIC_FACES_1, 1, "RADEON_PP_CUBIC_FACES_1"}, + {RADEON_PP_CUBIC_OFFSET_T1_0, 5, "RADEON_PP_CUBIC_OFFSET_T1_0"}, + {RADEON_PP_CUBIC_FACES_2, 1, "RADEON_PP_CUBIC_FACES_2"}, + {RADEON_PP_CUBIC_OFFSET_T2_0, 5, "RADEON_PP_CUBIC_OFFSET_T2_0"}, + {R200_PP_TRI_PERF, 2, "R200_PP_TRI_PERF"}, + {R200_PP_TXCBLEND_8, 32, "R200_PP_AFS_0"}, /* 85 */ + {R200_PP_TXCBLEND_0, 32, "R200_PP_AFS_1"}, + {R200_PP_TFACTOR_0, 8, "R200_ATF_TFACTOR"}, + {R200_PP_TXFILTER_0, 8, "R200_PP_TXCTLALL_0"}, + {R200_PP_TXFILTER_1, 8, "R200_PP_TXCTLALL_1"}, + {R200_PP_TXFILTER_2, 8, "R200_PP_TXCTLALL_2"}, + {R200_PP_TXFILTER_3, 8, "R200_PP_TXCTLALL_3"}, + {R200_PP_TXFILTER_4, 8, "R200_PP_TXCTLALL_4"}, + {R200_PP_TXFILTER_5, 8, "R200_PP_TXCTLALL_5"}, + {R200_VAP_PVS_CNTL_1, 2, "R200_VAP_PVS_CNTL"}, +}; + /* ============================================================= * State initialization */ - -void radeonPrintDirty( radeonContextPtr rmesa, const char *msg ) +static int cmdpkt( r100ContextPtr rmesa, int id ) { - struct radeon_state_atom *l; - - fprintf(stderr, msg); - fprintf(stderr, ": "); + drm_radeon_cmd_header_t h; - foreach(l, &rmesa->hw.atomlist) { - if (l->dirty || rmesa->hw.all_dirty) - fprintf(stderr, "%s, ", l->name); + if (rmesa->radeon.radeonScreen->kernel_mm) { + return CP_PACKET0(packet[id].start, packet[id].len - 1); + } else { + h.i = 0; + h.packet.cmd_type = RADEON_CMD_PACKET; + h.packet.packet_id = id; } - - fprintf(stderr, "\n"); -} - -static int cmdpkt( int id ) -{ - drm_radeon_cmd_header_t h; - h.i = 0; - h.packet.cmd_type = RADEON_CMD_PACKET; - h.packet.packet_id = id; return h.i; } @@ -96,131 +198,523 @@ static int cmdscl( int offset, int stride, int count ) return h.i; } -#define CHECK( NM, FLAG ) \ -static GLboolean check_##NM( GLcontext *ctx ) \ -{ \ - return FLAG; \ +#define CHECK( NM, FLAG, ADD ) \ +static int check_##NM( GLcontext *ctx, struct radeon_state_atom *atom ) \ +{ \ + return FLAG ? atom->cmd_size + (ADD) : 0; \ } -#define TCL_CHECK( NM, FLAG ) \ -static GLboolean check_##NM( GLcontext *ctx ) \ +#define TCL_CHECK( NM, FLAG, ADD ) \ +static int check_##NM( GLcontext *ctx, struct radeon_state_atom *atom ) \ { \ - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); \ - return !rmesa->TclFallback && (FLAG); \ + r100ContextPtr rmesa = R100_CONTEXT(ctx); \ + return (!rmesa->radeon.TclFallback && (FLAG)) ? atom->cmd_size + (ADD) : 0; \ } -CHECK( always, GL_TRUE ) -CHECK( never, GL_FALSE ) -CHECK( tex0, ctx->Texture.Unit[0]._ReallyEnabled ) -CHECK( tex1, ctx->Texture.Unit[1]._ReallyEnabled ) +CHECK( always, GL_TRUE, 0 ) +CHECK( always_add2, GL_TRUE, 2 ) +CHECK( always_add4, GL_TRUE, 4 ) +CHECK( never, GL_FALSE, 0 ) +CHECK( tex0_mm, ctx->Texture.Unit[0]._ReallyEnabled, 3 ) +CHECK( tex1_mm, ctx->Texture.Unit[1]._ReallyEnabled, 3 ) /* need this for the cubic_map on disabled unit 2 bug, maybe r100 only? */ -CHECK( tex2, ctx->Texture._EnabledUnits ) -CHECK( cube0, (ctx->Texture.Unit[0]._ReallyEnabled & TEXTURE_CUBE_BIT)) -CHECK( cube1, (ctx->Texture.Unit[1]._ReallyEnabled & TEXTURE_CUBE_BIT)) -CHECK( cube2, (ctx->Texture.Unit[2]._ReallyEnabled & TEXTURE_CUBE_BIT)) -CHECK( fog, ctx->Fog.Enabled ) -TCL_CHECK( tcl, GL_TRUE ) -TCL_CHECK( tcl_tex0, ctx->Texture.Unit[0]._ReallyEnabled ) -TCL_CHECK( tcl_tex1, ctx->Texture.Unit[1]._ReallyEnabled ) -TCL_CHECK( tcl_tex2, ctx->Texture.Unit[2]._ReallyEnabled ) -TCL_CHECK( tcl_lighting, ctx->Light.Enabled ) -TCL_CHECK( tcl_eyespace_or_lighting, ctx->_NeedEyeCoords || ctx->Light.Enabled ) -TCL_CHECK( tcl_lit0, ctx->Light.Enabled && ctx->Light.Light[0].Enabled ) -TCL_CHECK( tcl_lit1, ctx->Light.Enabled && ctx->Light.Light[1].Enabled ) -TCL_CHECK( tcl_lit2, ctx->Light.Enabled && ctx->Light.Light[2].Enabled ) -TCL_CHECK( tcl_lit3, ctx->Light.Enabled && ctx->Light.Light[3].Enabled ) -TCL_CHECK( tcl_lit4, ctx->Light.Enabled && ctx->Light.Light[4].Enabled ) -TCL_CHECK( tcl_lit5, ctx->Light.Enabled && ctx->Light.Light[5].Enabled ) -TCL_CHECK( tcl_lit6, ctx->Light.Enabled && ctx->Light.Light[6].Enabled ) -TCL_CHECK( tcl_lit7, ctx->Light.Enabled && ctx->Light.Light[7].Enabled ) -TCL_CHECK( tcl_ucp0, (ctx->Transform.ClipPlanesEnabled & 0x1) ) -TCL_CHECK( tcl_ucp1, (ctx->Transform.ClipPlanesEnabled & 0x2) ) -TCL_CHECK( tcl_ucp2, (ctx->Transform.ClipPlanesEnabled & 0x4) ) -TCL_CHECK( tcl_ucp3, (ctx->Transform.ClipPlanesEnabled & 0x8) ) -TCL_CHECK( tcl_ucp4, (ctx->Transform.ClipPlanesEnabled & 0x10) ) -TCL_CHECK( tcl_ucp5, (ctx->Transform.ClipPlanesEnabled & 0x20) ) -TCL_CHECK( tcl_eyespace_or_fog, ctx->_NeedEyeCoords || ctx->Fog.Enabled ) - -CHECK( txr0, (ctx->Texture.Unit[0]._ReallyEnabled & TEXTURE_RECT_BIT)) -CHECK( txr1, (ctx->Texture.Unit[1]._ReallyEnabled & TEXTURE_RECT_BIT)) -CHECK( txr2, (ctx->Texture.Unit[2]._ReallyEnabled & TEXTURE_RECT_BIT)) +CHECK( tex2_mm, ctx->Texture._EnabledUnits, 3 ) +CHECK( tex0, ctx->Texture.Unit[0]._ReallyEnabled, 2 ) +CHECK( tex1, ctx->Texture.Unit[1]._ReallyEnabled, 2 ) +CHECK( tex2, ctx->Texture._EnabledUnits, 2 ) +CHECK( cube0, (ctx->Texture.Unit[0]._ReallyEnabled & TEXTURE_CUBE_BIT), 3 + 3*5 - CUBE_STATE_SIZE ) +CHECK( cube1, (ctx->Texture.Unit[1]._ReallyEnabled & TEXTURE_CUBE_BIT), 3 + 3*5 - CUBE_STATE_SIZE ) +CHECK( cube2, (ctx->Texture.Unit[2]._ReallyEnabled & TEXTURE_CUBE_BIT), 3 + 3*5 - CUBE_STATE_SIZE ) +CHECK( cube0_mm, (ctx->Texture.Unit[0]._ReallyEnabled & TEXTURE_CUBE_BIT), 2 + 4*5 - CUBE_STATE_SIZE ) +CHECK( cube1_mm, (ctx->Texture.Unit[1]._ReallyEnabled & TEXTURE_CUBE_BIT), 2 + 4*5 - CUBE_STATE_SIZE ) +CHECK( cube2_mm, (ctx->Texture.Unit[2]._ReallyEnabled & TEXTURE_CUBE_BIT), 2 + 4*5 - CUBE_STATE_SIZE ) +CHECK( fog, ctx->Fog.Enabled, 0 ) +CHECK( fog_add4, ctx->Fog.Enabled, 4 ) +TCL_CHECK( tcl, GL_TRUE, 0 ) +TCL_CHECK( tcl_add4, GL_TRUE, 4 ) +TCL_CHECK( tcl_tex0, ctx->Texture.Unit[0]._ReallyEnabled, 0 ) +TCL_CHECK( tcl_tex1, ctx->Texture.Unit[1]._ReallyEnabled, 0 ) +TCL_CHECK( tcl_tex2, ctx->Texture.Unit[2]._ReallyEnabled, 0 ) +TCL_CHECK( tcl_tex0_add4, ctx->Texture.Unit[0]._ReallyEnabled, 4 ) +TCL_CHECK( tcl_tex1_add4, ctx->Texture.Unit[1]._ReallyEnabled, 4 ) +TCL_CHECK( tcl_tex2_add4, ctx->Texture.Unit[2]._ReallyEnabled, 4 ) +TCL_CHECK( tcl_lighting, ctx->Light.Enabled, 0 ) +TCL_CHECK( tcl_lighting_add4, ctx->Light.Enabled, 4 ) +TCL_CHECK( tcl_eyespace_or_lighting, ctx->_NeedEyeCoords || ctx->Light.Enabled, 0 ) +TCL_CHECK( tcl_eyespace_or_lighting_add4, ctx->_NeedEyeCoords || ctx->Light.Enabled, 4 ) +TCL_CHECK( tcl_lit0, ctx->Light.Enabled && ctx->Light.Light[0].Enabled, 0 ) +TCL_CHECK( tcl_lit1, ctx->Light.Enabled && ctx->Light.Light[1].Enabled, 0 ) +TCL_CHECK( tcl_lit2, ctx->Light.Enabled && ctx->Light.Light[2].Enabled, 0 ) +TCL_CHECK( tcl_lit3, ctx->Light.Enabled && ctx->Light.Light[3].Enabled, 0 ) +TCL_CHECK( tcl_lit4, ctx->Light.Enabled && ctx->Light.Light[4].Enabled, 0 ) +TCL_CHECK( tcl_lit5, ctx->Light.Enabled && ctx->Light.Light[5].Enabled, 0 ) +TCL_CHECK( tcl_lit6, ctx->Light.Enabled && ctx->Light.Light[6].Enabled, 0 ) +TCL_CHECK( tcl_lit7, ctx->Light.Enabled && ctx->Light.Light[7].Enabled, 0 ) +TCL_CHECK( tcl_lit0_add6, ctx->Light.Enabled && ctx->Light.Light[0].Enabled, 6 ) +TCL_CHECK( tcl_lit1_add6, ctx->Light.Enabled && ctx->Light.Light[1].Enabled, 6 ) +TCL_CHECK( tcl_lit2_add6, ctx->Light.Enabled && ctx->Light.Light[2].Enabled, 6 ) +TCL_CHECK( tcl_lit3_add6, ctx->Light.Enabled && ctx->Light.Light[3].Enabled, 6 ) +TCL_CHECK( tcl_lit4_add6, ctx->Light.Enabled && ctx->Light.Light[4].Enabled, 6 ) +TCL_CHECK( tcl_lit5_add6, ctx->Light.Enabled && ctx->Light.Light[5].Enabled, 6 ) +TCL_CHECK( tcl_lit6_add6, ctx->Light.Enabled && ctx->Light.Light[6].Enabled, 6 ) +TCL_CHECK( tcl_lit7_add6, ctx->Light.Enabled && ctx->Light.Light[7].Enabled, 6 ) +TCL_CHECK( tcl_ucp0, (ctx->Transform.ClipPlanesEnabled & 0x1), 0 ) +TCL_CHECK( tcl_ucp1, (ctx->Transform.ClipPlanesEnabled & 0x2), 0 ) +TCL_CHECK( tcl_ucp2, (ctx->Transform.ClipPlanesEnabled & 0x4), 0 ) +TCL_CHECK( tcl_ucp3, (ctx->Transform.ClipPlanesEnabled & 0x8), 0 ) +TCL_CHECK( tcl_ucp4, (ctx->Transform.ClipPlanesEnabled & 0x10), 0 ) +TCL_CHECK( tcl_ucp5, (ctx->Transform.ClipPlanesEnabled & 0x20), 0 ) +TCL_CHECK( tcl_ucp0_add4, (ctx->Transform.ClipPlanesEnabled & 0x1), 4 ) +TCL_CHECK( tcl_ucp1_add4, (ctx->Transform.ClipPlanesEnabled & 0x2), 4 ) +TCL_CHECK( tcl_ucp2_add4, (ctx->Transform.ClipPlanesEnabled & 0x4), 4 ) +TCL_CHECK( tcl_ucp3_add4, (ctx->Transform.ClipPlanesEnabled & 0x8), 4 ) +TCL_CHECK( tcl_ucp4_add4, (ctx->Transform.ClipPlanesEnabled & 0x10), 4 ) +TCL_CHECK( tcl_ucp5_add4, (ctx->Transform.ClipPlanesEnabled & 0x20), 4 ) +TCL_CHECK( tcl_eyespace_or_fog, ctx->_NeedEyeCoords || ctx->Fog.Enabled, 0 ) +TCL_CHECK( tcl_eyespace_or_fog_add4, ctx->_NeedEyeCoords || ctx->Fog.Enabled, 4 ) + +CHECK( txr0, (ctx->Texture.Unit[0]._ReallyEnabled & TEXTURE_RECT_BIT), 0 ) +CHECK( txr1, (ctx->Texture.Unit[1]._ReallyEnabled & TEXTURE_RECT_BIT), 0 ) +CHECK( txr2, (ctx->Texture.Unit[2]._ReallyEnabled & TEXTURE_RECT_BIT), 0 ) + +#define OUT_VEC(hdr, data) do { \ + drm_radeon_cmd_header_t h; \ + h.i = hdr; \ + OUT_BATCH(CP_PACKET0(RADEON_SE_TCL_STATE_FLUSH, 0)); \ + OUT_BATCH(0); \ + OUT_BATCH(CP_PACKET0(R200_SE_TCL_VECTOR_INDX_REG, 0)); \ + OUT_BATCH(h.vectors.offset | (h.vectors.stride << RADEON_VEC_INDX_OCTWORD_STRIDE_SHIFT)); \ + OUT_BATCH(CP_PACKET0_ONE(R200_SE_TCL_VECTOR_DATA_REG, h.vectors.count - 1)); \ + OUT_BATCH_TABLE((data), h.vectors.count); \ + } while(0) + +#define OUT_SCL(hdr, data) do { \ + drm_radeon_cmd_header_t h; \ + h.i = hdr; \ + OUT_BATCH(CP_PACKET0(R200_SE_TCL_SCALAR_INDX_REG, 0)); \ + OUT_BATCH((h.scalars.offset) | (h.scalars.stride << RADEON_SCAL_INDX_DWORD_STRIDE_SHIFT)); \ + OUT_BATCH(CP_PACKET0_ONE(R200_SE_TCL_SCALAR_DATA_REG, h.scalars.count - 1)); \ + OUT_BATCH_TABLE((data), h.scalars.count); \ + } while(0) + +static void scl_emit(GLcontext *ctx, struct radeon_state_atom *atom) +{ + r100ContextPtr r100 = R100_CONTEXT(ctx); + BATCH_LOCALS(&r100->radeon); + uint32_t dwords = atom->check(ctx, atom); + + BEGIN_BATCH_NO_AUTOSTATE(dwords); + OUT_SCL(atom->cmd[0], atom->cmd+1); + END_BATCH(); +} +static void vec_emit(GLcontext *ctx, struct radeon_state_atom *atom) +{ + r100ContextPtr r100 = R100_CONTEXT(ctx); + BATCH_LOCALS(&r100->radeon); + uint32_t dwords = atom->check(ctx, atom); -/* Initialize the context's hardware state. - */ -void radeonInitState( radeonContextPtr rmesa ) + BEGIN_BATCH_NO_AUTOSTATE(dwords); + OUT_VEC(atom->cmd[0], atom->cmd+1); + END_BATCH(); +} + + +static void lit_emit(GLcontext *ctx, struct radeon_state_atom *atom) { - GLcontext *ctx = rmesa->glCtx; - GLuint color_fmt, depth_fmt, i; - GLint drawPitch, drawOffset; + r100ContextPtr r100 = R100_CONTEXT(ctx); + BATCH_LOCALS(&r100->radeon); + uint32_t dwords = atom->check(ctx, atom); + + BEGIN_BATCH_NO_AUTOSTATE(dwords); + OUT_VEC(atom->cmd[LIT_CMD_0], atom->cmd+1); + OUT_SCL(atom->cmd[LIT_CMD_1], atom->cmd+LIT_CMD_1+1); + END_BATCH(); +} - switch ( rmesa->radeonScreen->cpp ) { - case 2: - color_fmt = RADEON_COLOR_FORMAT_RGB565; - break; - case 4: - color_fmt = RADEON_COLOR_FORMAT_ARGB8888; - break; - default: - fprintf( stderr, "Error: Unsupported pixel depth... exiting\n" ); - exit( -1 ); +static void ctx_emit(GLcontext *ctx, struct radeon_state_atom *atom) +{ + r100ContextPtr r100 = R100_CONTEXT(ctx); + BATCH_LOCALS(&r100->radeon); + struct radeon_renderbuffer *rrb; + uint32_t cbpitch; + uint32_t zbpitch, depth_fmt; + uint32_t dwords = atom->check(ctx, atom); + + /* output the first 7 bytes of context */ + BEGIN_BATCH_NO_AUTOSTATE(dwords); + OUT_BATCH_TABLE(atom->cmd, 5); + + rrb = radeon_get_depthbuffer(&r100->radeon); + if (!rrb) { + OUT_BATCH(0); + OUT_BATCH(0); + } else { + zbpitch = (rrb->pitch / rrb->cpp); + if (r100->using_hyperz) + zbpitch |= RADEON_DEPTH_HYPERZ; + + OUT_BATCH_RELOC(0, rrb->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0); + OUT_BATCH(zbpitch); + if (rrb->cpp == 4) + depth_fmt = RADEON_DEPTH_FORMAT_24BIT_INT_Z; + else + depth_fmt = RADEON_DEPTH_FORMAT_16BIT_INT_Z; + atom->cmd[CTX_RB3D_ZSTENCILCNTL] &= ~RADEON_DEPTH_FORMAT_MASK; + atom->cmd[CTX_RB3D_ZSTENCILCNTL] |= depth_fmt; + } + + OUT_BATCH(atom->cmd[CTX_RB3D_ZSTENCILCNTL]); + OUT_BATCH(atom->cmd[CTX_CMD_1]); + OUT_BATCH(atom->cmd[CTX_PP_CNTL]); + + rrb = radeon_get_colorbuffer(&r100->radeon); + if (!rrb || !rrb->bo) { + OUT_BATCH(atom->cmd[CTX_RB3D_CNTL]); + OUT_BATCH(atom->cmd[CTX_RB3D_COLOROFFSET]); + } else { + atom->cmd[CTX_RB3D_CNTL] &= ~(0xf << 10); + if (rrb->cpp == 4) + atom->cmd[CTX_RB3D_CNTL] |= RADEON_COLOR_FORMAT_ARGB8888; + else + atom->cmd[CTX_RB3D_CNTL] |= RADEON_COLOR_FORMAT_RGB565; + + OUT_BATCH(atom->cmd[CTX_RB3D_CNTL]); + OUT_BATCH_RELOC(0, rrb->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0); } - rmesa->state.color.clear = 0x00000000; + OUT_BATCH(atom->cmd[CTX_CMD_2]); + + if (!rrb || !rrb->bo) { + OUT_BATCH(atom->cmd[CTX_RB3D_COLORPITCH]); + } else { + cbpitch = (rrb->pitch / rrb->cpp); + if (rrb->bo->flags & RADEON_BO_FLAGS_MACRO_TILE) + cbpitch |= RADEON_COLOR_TILE_ENABLE; + OUT_BATCH(cbpitch); + } + + END_BATCH(); +} + +static int check_always_ctx( GLcontext *ctx, struct radeon_state_atom *atom) +{ + r100ContextPtr r100 = R100_CONTEXT(ctx); + struct radeon_renderbuffer *rrb, *drb; + uint32_t dwords; + + rrb = radeon_get_colorbuffer(&r100->radeon); + if (!rrb || !rrb->bo) { + return 0; + } + + drb = radeon_get_depthbuffer(&r100->radeon); + + dwords = 10; + if (drb) + dwords += 6; + if (rrb) + dwords += 8; + + return dwords; +} + +static void ctx_emit_cs(GLcontext *ctx, struct radeon_state_atom *atom) +{ + r100ContextPtr r100 = R100_CONTEXT(ctx); + BATCH_LOCALS(&r100->radeon); + struct radeon_renderbuffer *rrb, *drb; + uint32_t cbpitch = 0; + uint32_t zbpitch = 0; + uint32_t dwords = atom->check(ctx, atom); + uint32_t depth_fmt; + + rrb = radeon_get_colorbuffer(&r100->radeon); + if (!rrb || !rrb->bo) { + fprintf(stderr, "no rrb\n"); + return; + } + + atom->cmd[CTX_RB3D_CNTL] &= ~(0xf << 10); + if (rrb->cpp == 4) + atom->cmd[CTX_RB3D_CNTL] |= RADEON_COLOR_FORMAT_ARGB8888; + else switch (rrb->base._ActualFormat) { + case GL_RGB5: + atom->cmd[CTX_RB3D_CNTL] |= RADEON_COLOR_FORMAT_RGB565; + break; + case GL_RGBA4: + atom->cmd[CTX_RB3D_CNTL] |= RADEON_COLOR_FORMAT_ARGB4444; + break; + case GL_RGB5_A1: + atom->cmd[CTX_RB3D_CNTL] |= RADEON_COLOR_FORMAT_ARGB1555; + break; + } + + cbpitch = (rrb->pitch / rrb->cpp); + if (rrb->bo->flags & RADEON_BO_FLAGS_MACRO_TILE) + cbpitch |= R200_COLOR_TILE_ENABLE; + + drb = radeon_get_depthbuffer(&r100->radeon); + if (drb) { + zbpitch = (drb->pitch / drb->cpp); + if (drb->cpp == 4) + depth_fmt = RADEON_DEPTH_FORMAT_24BIT_INT_Z; + else + depth_fmt = RADEON_DEPTH_FORMAT_16BIT_INT_Z; + atom->cmd[CTX_RB3D_ZSTENCILCNTL] &= ~RADEON_DEPTH_FORMAT_MASK; + atom->cmd[CTX_RB3D_ZSTENCILCNTL] |= depth_fmt; + + } + + BEGIN_BATCH_NO_AUTOSTATE(dwords); + + /* In the CS case we need to split this up */ + OUT_BATCH(CP_PACKET0(packet[0].start, 3)); + OUT_BATCH_TABLE((atom->cmd + 1), 4); + + if (drb) { + OUT_BATCH(CP_PACKET0(RADEON_RB3D_DEPTHOFFSET, 0)); + OUT_BATCH_RELOC(0, drb->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0); + + OUT_BATCH(CP_PACKET0(RADEON_RB3D_DEPTHPITCH, 0)); + OUT_BATCH(zbpitch); + } + + OUT_BATCH(CP_PACKET0(RADEON_RB3D_ZSTENCILCNTL, 0)); + OUT_BATCH(atom->cmd[CTX_RB3D_ZSTENCILCNTL]); + OUT_BATCH(CP_PACKET0(RADEON_PP_CNTL, 1)); + OUT_BATCH(atom->cmd[CTX_PP_CNTL]); + OUT_BATCH(atom->cmd[CTX_RB3D_CNTL]); + + if (rrb) { + OUT_BATCH(CP_PACKET0(RADEON_RB3D_COLOROFFSET, 0)); + OUT_BATCH_RELOC(0, rrb->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0); + + OUT_BATCH(CP_PACKET0(RADEON_RB3D_COLORPITCH, 0)); + OUT_BATCH_RELOC(cbpitch, rrb->bo, cbpitch, 0, RADEON_GEM_DOMAIN_VRAM, 0); + } + + // if (atom->cmd_size == CTX_STATE_SIZE_NEWDRM) { + // OUT_BATCH_TABLE((atom->cmd + 14), 4); + // } + + END_BATCH(); + BEGIN_BATCH_NO_AUTOSTATE(4); + OUT_BATCH(CP_PACKET0(RADEON_RE_TOP_LEFT, 0)); + OUT_BATCH(0); + OUT_BATCH(CP_PACKET0(RADEON_RE_WIDTH_HEIGHT, 0)); + if (rrb) { + OUT_BATCH(((rrb->base.Width - 1) << RADEON_RE_WIDTH_SHIFT) | + ((rrb->base.Height - 1) << RADEON_RE_HEIGHT_SHIFT)); + } else { + OUT_BATCH(0); + } + END_BATCH(); +} + +static void cube_emit(GLcontext *ctx, struct radeon_state_atom *atom) +{ + r100ContextPtr r100 = R100_CONTEXT(ctx); + BATCH_LOCALS(&r100->radeon); + uint32_t dwords = atom->check(ctx, atom); + int i = atom->idx, j; + radeonTexObj *t = r100->state.texture.unit[i].texobj; + radeon_mipmap_level *lvl; + + if (!(ctx->Texture.Unit[i]._ReallyEnabled & TEXTURE_CUBE_BIT)) + return; + + if (!t) + return; + + if (!t->mt) + return; + + BEGIN_BATCH_NO_AUTOSTATE(dwords); + OUT_BATCH_TABLE(atom->cmd, 3); + lvl = &t->mt->levels[0]; + for (j = 0; j < 5; j++) { + OUT_BATCH_RELOC(lvl->faces[j].offset, t->mt->bo, lvl->faces[j].offset, + RADEON_GEM_DOMAIN_VRAM, 0, 0); + } + END_BATCH(); +} + +static void cube_emit_cs(GLcontext *ctx, struct radeon_state_atom *atom) +{ + r100ContextPtr r100 = R100_CONTEXT(ctx); + BATCH_LOCALS(&r100->radeon); + uint32_t dwords = atom->check(ctx, atom); + int i = atom->idx, j; + radeonTexObj *t = r100->state.texture.unit[i].texobj; + radeon_mipmap_level *lvl; + uint32_t base_reg; + + if (!(ctx->Texture.Unit[i]._ReallyEnabled & TEXTURE_CUBE_BIT)) + return; + + if (!t) + return; + + if (!t->mt) + return; + + switch(i) { + case 1: base_reg = RADEON_PP_CUBIC_OFFSET_T1_0; break; + case 2: base_reg = RADEON_PP_CUBIC_OFFSET_T2_0; break; + default: + case 0: base_reg = RADEON_PP_CUBIC_OFFSET_T0_0; break; + }; + BEGIN_BATCH_NO_AUTOSTATE(dwords); + OUT_BATCH_TABLE(atom->cmd, 2); + lvl = &t->mt->levels[0]; + for (j = 0; j < 5; j++) { + OUT_BATCH(CP_PACKET0(base_reg + (4 * j), 0)); + OUT_BATCH_RELOC(lvl->faces[j].offset, t->mt->bo, lvl->faces[j].offset, + RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0); + } + END_BATCH(); +} + +static void tex_emit(GLcontext *ctx, struct radeon_state_atom *atom) +{ + r100ContextPtr r100 = R100_CONTEXT(ctx); + BATCH_LOCALS(&r100->radeon); + uint32_t dwords = atom->cmd_size; + int i = atom->idx; + radeonTexObj *t = r100->state.texture.unit[i].texobj; + radeon_mipmap_level *lvl; + + if (t && t->mt && !t->image_override) + dwords += 2; + BEGIN_BATCH_NO_AUTOSTATE(dwords); + + OUT_BATCH_TABLE(atom->cmd, 3); + if (t && t->mt && !t->image_override) { + if ((ctx->Texture.Unit[i]._ReallyEnabled & TEXTURE_CUBE_BIT)) { + lvl = &t->mt->levels[0]; + OUT_BATCH_RELOC(lvl->faces[5].offset, t->mt->bo, lvl->faces[5].offset, + RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0); + } else { + OUT_BATCH_RELOC(t->tile_bits, t->mt->bo, 0, + RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0); + } + } else if (!t) { + /* workaround for old CS mechanism */ + OUT_BATCH(r100->radeon.radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP]); + // OUT_BATCH(r100->radeon.radeonScreen); + } else { + OUT_BATCH(t->override_offset); + } + + OUT_BATCH_TABLE((atom->cmd+4), 5); + END_BATCH(); +} + +static void tex_emit_cs(GLcontext *ctx, struct radeon_state_atom *atom) +{ + r100ContextPtr r100 = R100_CONTEXT(ctx); + BATCH_LOCALS(&r100->radeon); + uint32_t dwords = atom->cmd_size; + int i = atom->idx; + radeonTexObj *t = r100->state.texture.unit[i].texobj; + radeon_mipmap_level *lvl; + int hastexture = 1; + + if (!t) + hastexture = 0; + else { + if (!t->mt && !t->bo) + hastexture = 0; + } + dwords += 1; + if (hastexture) + dwords += 2; + else + dwords -= 2; + BEGIN_BATCH_NO_AUTOSTATE(dwords); + + OUT_BATCH(CP_PACKET0(RADEON_PP_TXFILTER_0 + (24 * i), 1)); + OUT_BATCH_TABLE((atom->cmd + 1), 2); + + if (hastexture) { + OUT_BATCH(CP_PACKET0(RADEON_PP_TXOFFSET_0 + (24 * i), 0)); + if (t->mt && !t->image_override) { + if ((ctx->Texture.Unit[i]._ReallyEnabled & TEXTURE_CUBE_BIT)) { + lvl = &t->mt->levels[0]; + OUT_BATCH_RELOC(lvl->faces[5].offset, t->mt->bo, lvl->faces[5].offset, + RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0); + } else { + OUT_BATCH_RELOC(t->tile_bits, t->mt->bo, 0, + RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0); + } + } else { + if (t->bo) + OUT_BATCH_RELOC(t->tile_bits, t->bo, 0, + RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0); + } + } + + OUT_BATCH(CP_PACKET0(RADEON_PP_TXCBLEND_0 + (i * 24), 1)); + OUT_BATCH_TABLE((atom->cmd+4), 2); + OUT_BATCH(CP_PACKET0(RADEON_PP_BORDER_COLOR_0 + (i * 4), 0)); + OUT_BATCH((atom->cmd[TEX_PP_BORDER_COLOR])); + END_BATCH(); +} + +/* Initialize the context's hardware state. + */ +void radeonInitState( r100ContextPtr rmesa ) +{ + GLcontext *ctx = rmesa->radeon.glCtx; + GLuint i; + + rmesa->radeon.state.color.clear = 0x00000000; switch ( ctx->Visual.depthBits ) { case 16: - rmesa->state.depth.clear = 0x0000ffff; - rmesa->state.depth.scale = 1.0 / (GLfloat)0xffff; - depth_fmt = RADEON_DEPTH_FORMAT_16BIT_INT_Z; - rmesa->state.stencil.clear = 0x00000000; + rmesa->radeon.state.depth.clear = 0x0000ffff; + rmesa->radeon.state.stencil.clear = 0x00000000; break; case 24: - rmesa->state.depth.clear = 0x00ffffff; - rmesa->state.depth.scale = 1.0 / (GLfloat)0xffffff; - depth_fmt = RADEON_DEPTH_FORMAT_24BIT_INT_Z; - rmesa->state.stencil.clear = 0xffff0000; + rmesa->radeon.state.depth.clear = 0x00ffffff; + rmesa->radeon.state.stencil.clear = 0xffff0000; break; default: - fprintf( stderr, "Error: Unsupported depth %d... exiting\n", - ctx->Visual.depthBits ); - exit( -1 ); + break; } - /* Only have hw stencil when depth buffer is 24 bits deep */ - rmesa->state.stencil.hwBuffer = ( ctx->Visual.stencilBits > 0 && - ctx->Visual.depthBits == 24 ); + rmesa->radeon.Fallback = 0; - rmesa->Fallback = 0; - if ( ctx->Visual.doubleBufferMode && rmesa->sarea->pfCurrentPage == 0 ) { - drawOffset = rmesa->radeonScreen->backOffset; - drawPitch = rmesa->radeonScreen->backPitch; - } else { - drawOffset = rmesa->radeonScreen->frontOffset; - drawPitch = rmesa->radeonScreen->frontPitch; - } + rmesa->radeon.hw.max_state_size = 0; - rmesa->hw.max_state_size = 0; - -#define ALLOC_STATE( ATOM, CHK, SZ, NM, FLAG ) \ +#define ALLOC_STATE_IDX( ATOM, CHK, SZ, NM, FLAG, IDX ) \ do { \ rmesa->hw.ATOM.cmd_size = SZ; \ - rmesa->hw.ATOM.cmd = (int *)CALLOC(SZ * sizeof(int)); \ - rmesa->hw.ATOM.lastcmd = (int *)CALLOC(SZ * sizeof(int)); \ - rmesa->hw.ATOM.name = NM; \ + rmesa->hw.ATOM.cmd = (GLuint *)CALLOC(SZ * sizeof(int)); \ + rmesa->hw.ATOM.lastcmd = (GLuint *)CALLOC(SZ * sizeof(int)); \ + rmesa->hw.ATOM.name = NM; \ rmesa->hw.ATOM.is_tcl = FLAG; \ rmesa->hw.ATOM.check = check_##CHK; \ - rmesa->hw.ATOM.dirty = GL_TRUE; \ - rmesa->hw.max_state_size += SZ * sizeof(int); \ + rmesa->hw.ATOM.dirty = GL_TRUE; \ + rmesa->hw.ATOM.idx = IDX; \ + rmesa->radeon.hw.max_state_size += SZ * sizeof(int); \ } while (0) - - + +#define ALLOC_STATE( ATOM, CHK, SZ, NM, FLAG ) \ + ALLOC_STATE_IDX(ATOM, CHK, SZ, NM, FLAG, 0) + /* Allocate state buffers: */ - ALLOC_STATE( ctx, always, CTX_STATE_SIZE, "CTX/context", 0 ); + ALLOC_STATE( ctx, always_add4, CTX_STATE_SIZE, "CTX/context", 0 ); + if (rmesa->radeon.radeonScreen->kernel_mm) { + rmesa->hw.ctx.emit = ctx_emit_cs; + rmesa->hw.ctx.check = check_always_ctx; + } else + rmesa->hw.ctx.emit = ctx_emit; ALLOC_STATE( lin, always, LIN_STATE_SIZE, "LIN/line", 0 ); ALLOC_STATE( msk, always, MSK_STATE_SIZE, "MSK/mask", 0 ); ALLOC_STATE( vpt, always, VPT_STATE_SIZE, "VPT/viewport", 0 ); @@ -229,82 +723,133 @@ void radeonInitState( radeonContextPtr rmesa ) ALLOC_STATE( zbs, always, ZBS_STATE_SIZE, "ZBS/zbias", 0 ); ALLOC_STATE( tcl, always, TCL_STATE_SIZE, "TCL/tcl", 1 ); ALLOC_STATE( mtl, tcl_lighting, MTL_STATE_SIZE, "MTL/material", 1 ); - ALLOC_STATE( grd, always, GRD_STATE_SIZE, "GRD/guard-band", 1 ); - ALLOC_STATE( fog, fog, FOG_STATE_SIZE, "FOG/fog", 1 ); - ALLOC_STATE( glt, tcl_lighting, GLT_STATE_SIZE, "GLT/light-global", 1 ); - ALLOC_STATE( eye, tcl_lighting, EYE_STATE_SIZE, "EYE/eye-vector", 1 ); - ALLOC_STATE( tex[0], tex0, TEX_STATE_SIZE, "TEX/tex-0", 0 ); - ALLOC_STATE( tex[1], tex1, TEX_STATE_SIZE, "TEX/tex-1", 0 ); - ALLOC_STATE( tex[2], tex2, TEX_STATE_SIZE, "TEX/tex-2", 0 ); - if (rmesa->radeonScreen->drmSupportsCubeMapsR100) + if (rmesa->radeon.radeonScreen->kernel_mm) { + ALLOC_STATE( grd, always_add2, GRD_STATE_SIZE, "GRD/guard-band", 1 ); + ALLOC_STATE( fog, fog_add4, FOG_STATE_SIZE, "FOG/fog", 1 ); + ALLOC_STATE( glt, tcl_lighting_add4, GLT_STATE_SIZE, "GLT/light-global", 1 ); + ALLOC_STATE( eye, tcl_lighting_add4, EYE_STATE_SIZE, "EYE/eye-vector", 1 ); + ALLOC_STATE_IDX( tex[0], tex0_mm, TEX_STATE_SIZE, "TEX/tex-0", 0, 0); + ALLOC_STATE_IDX( tex[1], tex1_mm, TEX_STATE_SIZE, "TEX/tex-1", 0, 1); + ALLOC_STATE_IDX( tex[2], tex2_mm, TEX_STATE_SIZE, "TEX/tex-2", 0, 2); + ALLOC_STATE( mat[0], tcl_add4, MAT_STATE_SIZE, "MAT/modelproject", 1 ); + ALLOC_STATE( mat[1], tcl_eyespace_or_fog_add4, MAT_STATE_SIZE, "MAT/modelview", 1 ); + ALLOC_STATE( mat[2], tcl_eyespace_or_lighting_add4, MAT_STATE_SIZE, "MAT/it-modelview", 1 ); + ALLOC_STATE( mat[3], tcl_tex0_add4, MAT_STATE_SIZE, "MAT/texmat0", 1 ); + ALLOC_STATE( mat[4], tcl_tex1_add4, MAT_STATE_SIZE, "MAT/texmat1", 1 ); + ALLOC_STATE( mat[5], tcl_tex2_add4, MAT_STATE_SIZE, "MAT/texmat2", 1 ); + ALLOC_STATE( lit[0], tcl_lit0_add6, LIT_STATE_SIZE, "LIT/light-0", 1 ); + ALLOC_STATE( lit[1], tcl_lit1_add6, LIT_STATE_SIZE, "LIT/light-1", 1 ); + ALLOC_STATE( lit[2], tcl_lit2_add6, LIT_STATE_SIZE, "LIT/light-2", 1 ); + ALLOC_STATE( lit[3], tcl_lit3_add6, LIT_STATE_SIZE, "LIT/light-3", 1 ); + ALLOC_STATE( lit[4], tcl_lit4_add6, LIT_STATE_SIZE, "LIT/light-4", 1 ); + ALLOC_STATE( lit[5], tcl_lit5_add6, LIT_STATE_SIZE, "LIT/light-5", 1 ); + ALLOC_STATE( lit[6], tcl_lit6_add6, LIT_STATE_SIZE, "LIT/light-6", 1 ); + ALLOC_STATE( lit[7], tcl_lit7_add6, LIT_STATE_SIZE, "LIT/light-7", 1 ); + ALLOC_STATE( ucp[0], tcl_ucp0_add4, UCP_STATE_SIZE, "UCP/userclip-0", 1 ); + ALLOC_STATE( ucp[1], tcl_ucp1_add4, UCP_STATE_SIZE, "UCP/userclip-1", 1 ); + ALLOC_STATE( ucp[2], tcl_ucp2_add4, UCP_STATE_SIZE, "UCP/userclip-2", 1 ); + ALLOC_STATE( ucp[3], tcl_ucp3_add4, UCP_STATE_SIZE, "UCP/userclip-3", 1 ); + ALLOC_STATE( ucp[4], tcl_ucp4_add4, UCP_STATE_SIZE, "UCP/userclip-4", 1 ); + ALLOC_STATE( ucp[5], tcl_ucp5_add4, UCP_STATE_SIZE, "UCP/userclip-5", 1 ); + } else { + ALLOC_STATE( grd, always, GRD_STATE_SIZE, "GRD/guard-band", 1 ); + ALLOC_STATE( fog, fog, FOG_STATE_SIZE, "FOG/fog", 1 ); + ALLOC_STATE( glt, tcl_lighting, GLT_STATE_SIZE, "GLT/light-global", 1 ); + ALLOC_STATE( eye, tcl_lighting, EYE_STATE_SIZE, "EYE/eye-vector", 1 ); + ALLOC_STATE_IDX( tex[0], tex0, TEX_STATE_SIZE, "TEX/tex-0", 0, 0); + ALLOC_STATE_IDX( tex[1], tex1, TEX_STATE_SIZE, "TEX/tex-1", 0, 1); + ALLOC_STATE_IDX( tex[2], tex2, TEX_STATE_SIZE, "TEX/tex-2", 0, 2); + ALLOC_STATE( mat[0], tcl, MAT_STATE_SIZE, "MAT/modelproject", 1 ); + ALLOC_STATE( mat[1], tcl_eyespace_or_fog, MAT_STATE_SIZE, "MAT/modelview", 1 ); + ALLOC_STATE( mat[2], tcl_eyespace_or_lighting, MAT_STATE_SIZE, "MAT/it-modelview", 1 ); + ALLOC_STATE( mat[3], tcl_tex0, MAT_STATE_SIZE, "MAT/texmat0", 1 ); + ALLOC_STATE( mat[4], tcl_tex1, MAT_STATE_SIZE, "MAT/texmat1", 1 ); + ALLOC_STATE( mat[5], tcl_tex2, MAT_STATE_SIZE, "MAT/texmat2", 1 ); + ALLOC_STATE( lit[0], tcl_lit0, LIT_STATE_SIZE, "LIT/light-0", 1 ); + ALLOC_STATE( lit[1], tcl_lit1, LIT_STATE_SIZE, "LIT/light-1", 1 ); + ALLOC_STATE( lit[2], tcl_lit2, LIT_STATE_SIZE, "LIT/light-2", 1 ); + ALLOC_STATE( lit[3], tcl_lit3, LIT_STATE_SIZE, "LIT/light-3", 1 ); + ALLOC_STATE( lit[4], tcl_lit4, LIT_STATE_SIZE, "LIT/light-4", 1 ); + ALLOC_STATE( lit[5], tcl_lit5, LIT_STATE_SIZE, "LIT/light-5", 1 ); + ALLOC_STATE( lit[6], tcl_lit6, LIT_STATE_SIZE, "LIT/light-6", 1 ); + ALLOC_STATE( lit[7], tcl_lit7, LIT_STATE_SIZE, "LIT/light-7", 1 ); + ALLOC_STATE( ucp[0], tcl_ucp0, UCP_STATE_SIZE, "UCP/userclip-0", 1 ); + ALLOC_STATE( ucp[1], tcl_ucp1, UCP_STATE_SIZE, "UCP/userclip-1", 1 ); + ALLOC_STATE( ucp[2], tcl_ucp2, UCP_STATE_SIZE, "UCP/userclip-2", 1 ); + ALLOC_STATE( ucp[3], tcl_ucp3, UCP_STATE_SIZE, "UCP/userclip-3", 1 ); + ALLOC_STATE( ucp[4], tcl_ucp4, UCP_STATE_SIZE, "UCP/userclip-4", 1 ); + ALLOC_STATE( ucp[5], tcl_ucp5, UCP_STATE_SIZE, "UCP/userclip-5", 1 ); + } + + if (rmesa->radeon.radeonScreen->kernel_mm) { + ALLOC_STATE( stp, always, STP_STATE_SIZE, "STP/stp", 0 ); + } + + for (i = 0; i < 3; i++) { + if (rmesa->radeon.radeonScreen->kernel_mm) + rmesa->hw.tex[i].emit = tex_emit_cs; + else + rmesa->hw.tex[i].emit = tex_emit; + } + if (rmesa->radeon.radeonScreen->drmSupportsCubeMapsR100) { - ALLOC_STATE( cube[0], cube0, CUBE_STATE_SIZE, "CUBE/cube-0", 0 ); - ALLOC_STATE( cube[1], cube1, CUBE_STATE_SIZE, "CUBE/cube-1", 0 ); - ALLOC_STATE( cube[2], cube2, CUBE_STATE_SIZE, "CUBE/cube-2", 0 ); + if (rmesa->radeon.radeonScreen->kernel_mm) { + ALLOC_STATE_IDX( cube[0], cube0_mm, CUBE_STATE_SIZE, "CUBE/cube-0", 0, 0 ); + ALLOC_STATE_IDX( cube[1], cube1_mm, CUBE_STATE_SIZE, "CUBE/cube-1", 0, 1 ); + ALLOC_STATE_IDX( cube[2], cube2_mm, CUBE_STATE_SIZE, "CUBE/cube-2", 0, 2 ); + for (i = 0; i < 3; i++) + rmesa->hw.cube[i].emit = cube_emit_cs; + } else { + ALLOC_STATE_IDX( cube[0], cube0, CUBE_STATE_SIZE, "CUBE/cube-0", 0, 0 ); + ALLOC_STATE_IDX( cube[1], cube1, CUBE_STATE_SIZE, "CUBE/cube-1", 0, 1 ); + ALLOC_STATE_IDX( cube[2], cube2, CUBE_STATE_SIZE, "CUBE/cube-2", 0, 2 ); + for (i = 0; i < 3; i++) + rmesa->hw.cube[i].emit = cube_emit; + } } else { - ALLOC_STATE( cube[0], never, CUBE_STATE_SIZE, "CUBE/cube-0", 0 ); - ALLOC_STATE( cube[1], never, CUBE_STATE_SIZE, "CUBE/cube-1", 0 ); - ALLOC_STATE( cube[2], never, CUBE_STATE_SIZE, "CUBE/cube-2", 0 ); - } - ALLOC_STATE( mat[0], tcl, MAT_STATE_SIZE, "MAT/modelproject", 1 ); - ALLOC_STATE( mat[1], tcl_eyespace_or_fog, MAT_STATE_SIZE, "MAT/modelview", 1 ); - ALLOC_STATE( mat[2], tcl_eyespace_or_lighting, MAT_STATE_SIZE, "MAT/it-modelview", 1 ); - ALLOC_STATE( mat[3], tcl_tex0, MAT_STATE_SIZE, "MAT/texmat0", 1 ); - ALLOC_STATE( mat[4], tcl_tex1, MAT_STATE_SIZE, "MAT/texmat1", 1 ); - ALLOC_STATE( mat[5], tcl_tex2, MAT_STATE_SIZE, "MAT/texmat2", 1 ); - ALLOC_STATE( ucp[0], tcl_ucp0, UCP_STATE_SIZE, "UCP/userclip-0", 1 ); - ALLOC_STATE( ucp[1], tcl_ucp1, UCP_STATE_SIZE, "UCP/userclip-1", 1 ); - ALLOC_STATE( ucp[2], tcl_ucp2, UCP_STATE_SIZE, "UCP/userclip-2", 1 ); - ALLOC_STATE( ucp[3], tcl_ucp3, UCP_STATE_SIZE, "UCP/userclip-3", 1 ); - ALLOC_STATE( ucp[4], tcl_ucp4, UCP_STATE_SIZE, "UCP/userclip-4", 1 ); - ALLOC_STATE( ucp[5], tcl_ucp5, UCP_STATE_SIZE, "UCP/userclip-5", 1 ); - ALLOC_STATE( lit[0], tcl_lit0, LIT_STATE_SIZE, "LIT/light-0", 1 ); - ALLOC_STATE( lit[1], tcl_lit1, LIT_STATE_SIZE, "LIT/light-1", 1 ); - ALLOC_STATE( lit[2], tcl_lit2, LIT_STATE_SIZE, "LIT/light-2", 1 ); - ALLOC_STATE( lit[3], tcl_lit3, LIT_STATE_SIZE, "LIT/light-3", 1 ); - ALLOC_STATE( lit[4], tcl_lit4, LIT_STATE_SIZE, "LIT/light-4", 1 ); - ALLOC_STATE( lit[5], tcl_lit5, LIT_STATE_SIZE, "LIT/light-5", 1 ); - ALLOC_STATE( lit[6], tcl_lit6, LIT_STATE_SIZE, "LIT/light-6", 1 ); - ALLOC_STATE( lit[7], tcl_lit7, LIT_STATE_SIZE, "LIT/light-7", 1 ); - ALLOC_STATE( txr[0], txr0, TXR_STATE_SIZE, "TXR/txr-0", 0 ); - ALLOC_STATE( txr[1], txr1, TXR_STATE_SIZE, "TXR/txr-1", 0 ); - ALLOC_STATE( txr[2], txr2, TXR_STATE_SIZE, "TXR/txr-2", 0 ); + ALLOC_STATE_IDX( cube[0], never, CUBE_STATE_SIZE, "CUBE/cube-0", 0, 0 ); + ALLOC_STATE_IDX( cube[1], never, CUBE_STATE_SIZE, "CUBE/cube-1", 0, 1 ); + ALLOC_STATE_IDX( cube[2], never, CUBE_STATE_SIZE, "CUBE/cube-2", 0, 2 ); + } + ALLOC_STATE_IDX( txr[0], txr0, TXR_STATE_SIZE, "TXR/txr-0", 0, 0 ); + ALLOC_STATE_IDX( txr[1], txr1, TXR_STATE_SIZE, "TXR/txr-1", 0, 1 ); + ALLOC_STATE_IDX( txr[2], txr2, TXR_STATE_SIZE, "TXR/txr-2", 0, 2 ); radeonSetUpAtomList( rmesa ); /* Fill in the packet headers: */ - rmesa->hw.ctx.cmd[CTX_CMD_0] = cmdpkt(RADEON_EMIT_PP_MISC); - rmesa->hw.ctx.cmd[CTX_CMD_1] = cmdpkt(RADEON_EMIT_PP_CNTL); - rmesa->hw.ctx.cmd[CTX_CMD_2] = cmdpkt(RADEON_EMIT_RB3D_COLORPITCH); - rmesa->hw.lin.cmd[LIN_CMD_0] = cmdpkt(RADEON_EMIT_RE_LINE_PATTERN); - rmesa->hw.lin.cmd[LIN_CMD_1] = cmdpkt(RADEON_EMIT_SE_LINE_WIDTH); - rmesa->hw.msk.cmd[MSK_CMD_0] = cmdpkt(RADEON_EMIT_RB3D_STENCILREFMASK); - rmesa->hw.vpt.cmd[VPT_CMD_0] = cmdpkt(RADEON_EMIT_SE_VPORT_XSCALE); - rmesa->hw.set.cmd[SET_CMD_0] = cmdpkt(RADEON_EMIT_SE_CNTL); - rmesa->hw.set.cmd[SET_CMD_1] = cmdpkt(RADEON_EMIT_SE_CNTL_STATUS); - rmesa->hw.msc.cmd[MSC_CMD_0] = cmdpkt(RADEON_EMIT_RE_MISC); - rmesa->hw.tex[0].cmd[TEX_CMD_0] = cmdpkt(RADEON_EMIT_PP_TXFILTER_0); - rmesa->hw.tex[0].cmd[TEX_CMD_1] = cmdpkt(RADEON_EMIT_PP_BORDER_COLOR_0); - rmesa->hw.tex[1].cmd[TEX_CMD_0] = cmdpkt(RADEON_EMIT_PP_TXFILTER_1); - rmesa->hw.tex[1].cmd[TEX_CMD_1] = cmdpkt(RADEON_EMIT_PP_BORDER_COLOR_1); - rmesa->hw.tex[2].cmd[TEX_CMD_0] = cmdpkt(RADEON_EMIT_PP_TXFILTER_2); - rmesa->hw.tex[2].cmd[TEX_CMD_1] = cmdpkt(RADEON_EMIT_PP_BORDER_COLOR_2); - rmesa->hw.cube[0].cmd[CUBE_CMD_0] = cmdpkt(RADEON_EMIT_PP_CUBIC_FACES_0); - rmesa->hw.cube[0].cmd[CUBE_CMD_1] = cmdpkt(RADEON_EMIT_PP_CUBIC_OFFSETS_T0); - rmesa->hw.cube[1].cmd[CUBE_CMD_0] = cmdpkt(RADEON_EMIT_PP_CUBIC_FACES_1); - rmesa->hw.cube[1].cmd[CUBE_CMD_1] = cmdpkt(RADEON_EMIT_PP_CUBIC_OFFSETS_T1); - rmesa->hw.cube[2].cmd[CUBE_CMD_0] = cmdpkt(RADEON_EMIT_PP_CUBIC_FACES_2); - rmesa->hw.cube[2].cmd[CUBE_CMD_1] = cmdpkt(RADEON_EMIT_PP_CUBIC_OFFSETS_T2); - rmesa->hw.zbs.cmd[ZBS_CMD_0] = cmdpkt(RADEON_EMIT_SE_ZBIAS_FACTOR); - rmesa->hw.tcl.cmd[TCL_CMD_0] = cmdpkt(RADEON_EMIT_SE_TCL_OUTPUT_VTX_FMT); + rmesa->hw.ctx.cmd[CTX_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_PP_MISC); + rmesa->hw.ctx.cmd[CTX_CMD_1] = cmdpkt(rmesa, RADEON_EMIT_PP_CNTL); + rmesa->hw.ctx.cmd[CTX_CMD_2] = cmdpkt(rmesa, RADEON_EMIT_RB3D_COLORPITCH); + rmesa->hw.lin.cmd[LIN_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_RE_LINE_PATTERN); + rmesa->hw.lin.cmd[LIN_CMD_1] = cmdpkt(rmesa, RADEON_EMIT_SE_LINE_WIDTH); + rmesa->hw.msk.cmd[MSK_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_RB3D_STENCILREFMASK); + rmesa->hw.vpt.cmd[VPT_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_SE_VPORT_XSCALE); + rmesa->hw.set.cmd[SET_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_SE_CNTL); + rmesa->hw.set.cmd[SET_CMD_1] = cmdpkt(rmesa, RADEON_EMIT_SE_CNTL_STATUS); + rmesa->hw.msc.cmd[MSC_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_RE_MISC); + rmesa->hw.tex[0].cmd[TEX_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_PP_TXFILTER_0); + rmesa->hw.tex[0].cmd[TEX_CMD_1] = cmdpkt(rmesa, RADEON_EMIT_PP_BORDER_COLOR_0); + rmesa->hw.tex[1].cmd[TEX_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_PP_TXFILTER_1); + rmesa->hw.tex[1].cmd[TEX_CMD_1] = cmdpkt(rmesa, RADEON_EMIT_PP_BORDER_COLOR_1); + rmesa->hw.tex[2].cmd[TEX_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_PP_TXFILTER_2); + rmesa->hw.tex[2].cmd[TEX_CMD_1] = cmdpkt(rmesa, RADEON_EMIT_PP_BORDER_COLOR_2); + rmesa->hw.cube[0].cmd[CUBE_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_PP_CUBIC_FACES_0); + rmesa->hw.cube[0].cmd[CUBE_CMD_1] = cmdpkt(rmesa, RADEON_EMIT_PP_CUBIC_OFFSETS_T0); + rmesa->hw.cube[1].cmd[CUBE_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_PP_CUBIC_FACES_1); + rmesa->hw.cube[1].cmd[CUBE_CMD_1] = cmdpkt(rmesa, RADEON_EMIT_PP_CUBIC_OFFSETS_T1); + rmesa->hw.cube[2].cmd[CUBE_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_PP_CUBIC_FACES_2); + rmesa->hw.cube[2].cmd[CUBE_CMD_1] = cmdpkt(rmesa, RADEON_EMIT_PP_CUBIC_OFFSETS_T2); + rmesa->hw.zbs.cmd[ZBS_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_SE_ZBIAS_FACTOR); + rmesa->hw.tcl.cmd[TCL_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_SE_TCL_OUTPUT_VTX_FMT); rmesa->hw.mtl.cmd[MTL_CMD_0] = - cmdpkt(RADEON_EMIT_SE_TCL_MATERIAL_EMMISSIVE_RED); - rmesa->hw.txr[0].cmd[TXR_CMD_0] = cmdpkt(RADEON_EMIT_PP_TEX_SIZE_0); - rmesa->hw.txr[1].cmd[TXR_CMD_0] = cmdpkt(RADEON_EMIT_PP_TEX_SIZE_1); - rmesa->hw.txr[2].cmd[TXR_CMD_0] = cmdpkt(RADEON_EMIT_PP_TEX_SIZE_2); + cmdpkt(rmesa, RADEON_EMIT_SE_TCL_MATERIAL_EMMISSIVE_RED); + rmesa->hw.txr[0].cmd[TXR_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_PP_TEX_SIZE_0); + rmesa->hw.txr[1].cmd[TXR_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_PP_TEX_SIZE_1); + rmesa->hw.txr[2].cmd[TXR_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_PP_TEX_SIZE_2); rmesa->hw.grd.cmd[GRD_CMD_0] = cmdscl( RADEON_SS_VERT_GUARD_CLIP_ADJ_ADDR, 1, 4 ); rmesa->hw.fog.cmd[FOG_CMD_0] = @@ -331,6 +876,26 @@ void radeonInitState( radeonContextPtr rmesa ) cmdvec( RADEON_VS_UCP_ADDR + i, 1, 4 ); } + if (rmesa->radeon.radeonScreen->kernel_mm) { + rmesa->hw.stp.cmd[STP_CMD_0] = CP_PACKET0(RADEON_RE_STIPPLE_ADDR, 0); + rmesa->hw.stp.cmd[STP_DATA_0] = 0; + rmesa->hw.stp.cmd[STP_CMD_1] = CP_PACKET0_ONE(RADEON_RE_STIPPLE_DATA, 31); + + rmesa->hw.grd.emit = scl_emit; + rmesa->hw.fog.emit = vec_emit; + rmesa->hw.glt.emit = vec_emit; + rmesa->hw.eye.emit = vec_emit; + + for (i = 0; i < 6; i++) + rmesa->hw.mat[i].emit = vec_emit; + + for (i = 0; i < 8; i++) + rmesa->hw.lit[i].emit = lit_emit; + + for (i = 0; i < 6; i++) + rmesa->hw.ucp[i].emit = vec_emit; + } + rmesa->last_ReallyEnabled = -1; /* Initial Harware state: @@ -352,19 +917,7 @@ void radeonInitState( radeonContextPtr rmesa ) RADEON_SRC_BLEND_GL_ONE | RADEON_DST_BLEND_GL_ZERO ); - rmesa->hw.ctx.cmd[CTX_RB3D_DEPTHOFFSET] = - rmesa->radeonScreen->depthOffset + rmesa->radeonScreen->fbLocation; - - rmesa->hw.ctx.cmd[CTX_RB3D_DEPTHPITCH] = - ((rmesa->radeonScreen->depthPitch & - RADEON_DEPTHPITCH_MASK) | - RADEON_DEPTH_ENDIAN_NO_SWAP); - - if (rmesa->using_hyperz) - rmesa->hw.ctx.cmd[CTX_RB3D_DEPTHPITCH] |= RADEON_DEPTH_HYPERZ; - - rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] = (depth_fmt | - RADEON_Z_TEST_LESS | + rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] = (RADEON_Z_TEST_LESS | RADEON_STENCIL_TEST_ALWAYS | RADEON_STENCIL_FAIL_KEEP | RADEON_STENCIL_ZPASS_KEEP | @@ -374,7 +927,7 @@ void radeonInitState( radeonContextPtr rmesa ) if (rmesa->using_hyperz) { rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_Z_COMPRESSION_ENABLE | RADEON_Z_DECOMPRESSION_ENABLE; - if (rmesa->radeonScreen->chip_flags & RADEON_CHIPSET_TCL) { + if (rmesa->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL) { /* works for q3, but slight rendering errors with glxgears ? */ /* rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_Z_HIERARCHY_ENABLE;*/ /* need this otherwise get lots of lockups with q3 ??? */ @@ -386,10 +939,9 @@ void radeonInitState( radeonContextPtr rmesa ) RADEON_ANTI_ALIAS_NONE); rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] = (RADEON_PLANE_MASK_ENABLE | - color_fmt | RADEON_ZBLOCK16); - switch ( driQueryOptioni( &rmesa->optionCache, "dither_mode" ) ) { + switch ( driQueryOptioni( &rmesa->radeon.optionCache, "dither_mode" ) ) { case DRI_CONF_DITHER_XERRORDIFFRESET: rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |= RADEON_DITHER_INIT; break; @@ -397,30 +949,17 @@ void radeonInitState( radeonContextPtr rmesa ) rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |= RADEON_SCALE_DITHER_ENABLE; break; } - if ( driQueryOptioni( &rmesa->optionCache, "round_mode" ) == + if ( driQueryOptioni( &rmesa->radeon.optionCache, "round_mode" ) == DRI_CONF_ROUND_ROUND ) - rmesa->state.color.roundEnable = RADEON_ROUND_ENABLE; + rmesa->radeon.state.color.roundEnable = RADEON_ROUND_ENABLE; else - rmesa->state.color.roundEnable = 0; - if ( driQueryOptioni (&rmesa->optionCache, "color_reduction" ) == + rmesa->radeon.state.color.roundEnable = 0; + if ( driQueryOptioni (&rmesa->radeon.optionCache, "color_reduction" ) == DRI_CONF_COLOR_REDUCTION_DITHER ) rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |= RADEON_DITHER_ENABLE; else - rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |= rmesa->state.color.roundEnable; - - rmesa->hw.ctx.cmd[CTX_RB3D_COLOROFFSET] = ((drawOffset + - rmesa->radeonScreen->fbLocation) - & RADEON_COLOROFFSET_MASK); + rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |= rmesa->radeon.state.color.roundEnable; - rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] = ((drawPitch & - RADEON_COLORPITCH_MASK) | - RADEON_COLOR_ENDIAN_NO_SWAP); - - - /* (fixed size) sarea is initialized to zero afaics so can omit version check. Phew! */ - if (rmesa->sarea->tiling_enabled) { - rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] |= RADEON_COLOR_TILE_ENABLE; - } rmesa->hw.set.cmd[SET_SE_CNTL] = (RADEON_FFACE_CULL_CCW | RADEON_BFACE_SOLID | @@ -444,7 +983,7 @@ void radeonInitState( radeonContextPtr rmesa ) RADEON_VC_NO_SWAP; #endif - if (!(rmesa->radeonScreen->chip_flags & RADEON_CHIPSET_TCL)) { + if (!(rmesa->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL)) { rmesa->hw.set.cmd[SET_SE_CNTL_STATUS] |= RADEON_TCL_BYPASS; } @@ -491,8 +1030,8 @@ void radeonInitState( radeonContextPtr rmesa ) (2 << RADEON_TXFORMAT_HEIGHT_SHIFT)); /* Initialize the texture offset to the start of the card texture heap */ - rmesa->hw.tex[i].cmd[TEX_PP_TXOFFSET] = - rmesa->radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP]; + // rmesa->hw.tex[i].cmd[TEX_PP_TXOFFSET] = + // rmesa->radeon.radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP]; rmesa->hw.tex[i].cmd[TEX_PP_BORDER_COLOR] = 0; rmesa->hw.tex[i].cmd[TEX_PP_TXCBLEND] = @@ -513,15 +1052,15 @@ void radeonInitState( radeonContextPtr rmesa ) rmesa->hw.cube[i].cmd[CUBE_PP_CUBIC_FACES] = 0; rmesa->hw.cube[i].cmd[CUBE_PP_CUBIC_OFFSET_0] = - rmesa->radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP]; + rmesa->radeon.radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP]; rmesa->hw.cube[i].cmd[CUBE_PP_CUBIC_OFFSET_1] = - rmesa->radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP]; + rmesa->radeon.radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP]; rmesa->hw.cube[i].cmd[CUBE_PP_CUBIC_OFFSET_2] = - rmesa->radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP]; + rmesa->radeon.radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP]; rmesa->hw.cube[i].cmd[CUBE_PP_CUBIC_OFFSET_3] = - rmesa->radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP]; + rmesa->radeon.radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP]; rmesa->hw.cube[i].cmd[CUBE_PP_CUBIC_OFFSET_4] = - rmesa->radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP]; + rmesa->radeon.radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP]; } /* Can only add ST1 at the time of doing some multitex but can keep @@ -612,6 +1151,14 @@ void radeonInitState( radeonContextPtr rmesa ) rmesa->hw.eye.cmd[EYE_Y] = 0; rmesa->hw.eye.cmd[EYE_Z] = IEEE_ONE; rmesa->hw.eye.cmd[EYE_RESCALE_FACTOR] = IEEE_ONE; - - rmesa->hw.all_dirty = GL_TRUE; + + if (rmesa->radeon.radeonScreen->kernel_mm) { + radeon_init_query_stateobj(&rmesa->radeon, R100_QUERYOBJ_CMDSIZE); + rmesa->radeon.query.queryobj.cmd[R100_QUERYOBJ_CMD_0] = CP_PACKET0(RADEON_RB3D_ZPASS_DATA, 0); + rmesa->radeon.query.queryobj.cmd[R100_QUERYOBJ_DATA_0] = 0; + } + + rmesa->radeon.hw.all_dirty = GL_TRUE; + + rcommonInitCmdBuf(&rmesa->radeon); } diff --git a/src/mesa/drivers/dri/radeon/radeon_swtcl.c b/src/mesa/drivers/dri/radeon/radeon_swtcl.c index ebea1fecdca..e61f59eaeaf 100644 --- a/src/mesa/drivers/dri/radeon/radeon_swtcl.c +++ b/src/mesa/drivers/dri/radeon/radeon_swtcl.c @@ -38,6 +38,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "main/enums.h" #include "main/imports.h" #include "main/macros.h" +#include "main/simple_list.h" #include "swrast_setup/swrast_setup.h" #include "math/m_translate.h" @@ -50,10 +51,9 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "radeon_state.h" #include "radeon_swtcl.h" #include "radeon_tcl.h" +#include "radeon_debug.h" -static void flush_last_swtcl_prim( radeonContextPtr rmesa ); - /* R100: xyzw, c0, c1/fog, stq[0..2] = 4+1+1+3*3 = 15 right? */ /* R200: xyzw, c0, c1/fog, strq[0..5] = 4+1+1+4*6 = 30 */ #define RADEON_MAX_TNL_VERTEX_SIZE (15 * sizeof(GLfloat)) /* for mesa _tnl stage */ @@ -64,18 +64,18 @@ static void flush_last_swtcl_prim( radeonContextPtr rmesa ); #define EMIT_ATTR( ATTR, STYLE, F0 ) \ do { \ - rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].attrib = (ATTR); \ - rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].format = (STYLE); \ - rmesa->swtcl.vertex_attr_count++; \ + rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].attrib = (ATTR); \ + rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].format = (STYLE); \ + rmesa->radeon.swtcl.vertex_attr_count++; \ fmt_0 |= F0; \ } while (0) #define EMIT_PAD( N ) \ do { \ - rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].attrib = 0; \ - rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].format = EMIT_PAD; \ - rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].offset = (N); \ - rmesa->swtcl.vertex_attr_count++; \ + rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].attrib = 0; \ + rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].format = EMIT_PAD; \ + rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].offset = (N); \ + rmesa->radeon.swtcl.vertex_attr_count++; \ } while (0) static GLuint radeon_cp_vc_frmts[3][2] = @@ -87,7 +87,7 @@ static GLuint radeon_cp_vc_frmts[3][2] = static void radeonSetVertexFormat( GLcontext *ctx ) { - radeonContextPtr rmesa = RADEON_CONTEXT( ctx ); + r100ContextPtr rmesa = R100_CONTEXT( ctx ); TNLcontext *tnl = TNL_CONTEXT(ctx); struct vertex_buffer *VB = &tnl->vb; DECLARE_RENDERINPUTS(index_bitset); @@ -106,7 +106,7 @@ static void radeonSetVertexFormat( GLcontext *ctx ) } assert( VB->AttribPtr[VERT_ATTRIB_POS] != NULL ); - rmesa->swtcl.vertex_attr_count = 0; + rmesa->radeon.swtcl.vertex_attr_count = 0; /* EMIT_ATTR's must be in order as they tell t_vertex.c how to * build up a hardware vertex. @@ -204,33 +204,52 @@ static void radeonSetVertexFormat( GLcontext *ctx ) } } - if (!RENDERINPUTS_EQUAL( rmesa->tnl_index_bitset, index_bitset ) || + if (!RENDERINPUTS_EQUAL( rmesa->radeon.tnl_index_bitset, index_bitset ) || fmt_0 != rmesa->swtcl.vertex_format) { RADEON_NEWPRIM(rmesa); rmesa->swtcl.vertex_format = fmt_0; - rmesa->swtcl.vertex_size = + rmesa->radeon.swtcl.vertex_size = _tnl_install_attrs( ctx, - rmesa->swtcl.vertex_attrs, - rmesa->swtcl.vertex_attr_count, + rmesa->radeon.swtcl.vertex_attrs, + rmesa->radeon.swtcl.vertex_attr_count, NULL, 0 ); - rmesa->swtcl.vertex_size /= 4; - RENDERINPUTS_COPY( rmesa->tnl_index_bitset, index_bitset ); - if (RADEON_DEBUG & DEBUG_VERTS) - fprintf( stderr, "%s: vertex_size= %d floats\n", - __FUNCTION__, rmesa->swtcl.vertex_size); + rmesa->radeon.swtcl.vertex_size /= 4; + RENDERINPUTS_COPY( rmesa->radeon.tnl_index_bitset, index_bitset ); + radeon_print(RADEON_SWRENDER, RADEON_VERBOSE, + "%s: vertex_size= %d floats\n", __FUNCTION__, rmesa->radeon.swtcl.vertex_size); } } +static void radeon_predict_emit_size( r100ContextPtr rmesa ) +{ + + if (!rmesa->radeon.swtcl.emit_prediction) { + const int state_size = radeonCountStateEmitSize( &rmesa->radeon ); + const int scissor_size = 8; + const int prims_size = 8; + const int vertex_size = 7; + + if (rcommonEnsureCmdBufSpace(&rmesa->radeon, + state_size + + (scissor_size + prims_size + vertex_size), + __FUNCTION__)) + rmesa->radeon.swtcl.emit_prediction = radeonCountStateEmitSize( &rmesa->radeon ); + else + rmesa->radeon.swtcl.emit_prediction = state_size; + rmesa->radeon.swtcl.emit_prediction += scissor_size + prims_size + vertex_size + + rmesa->radeon.cmdbuf.cs->cdw; + } +} static void radeonRenderStart( GLcontext *ctx ) { - radeonContextPtr rmesa = RADEON_CONTEXT( ctx ); + r100ContextPtr rmesa = R100_CONTEXT( ctx ); - radeonSetVertexFormat( ctx ); - - if (rmesa->dma.flush != 0 && - rmesa->dma.flush != flush_last_swtcl_prim) - rmesa->dma.flush( rmesa ); + radeonSetVertexFormat( ctx ); + + if (rmesa->radeon.dma.flush != 0 && + rmesa->radeon.dma.flush != rcommon_flush_last_swtcl_prim) + rmesa->radeon.dma.flush( ctx ); } @@ -241,7 +260,7 @@ static void radeonRenderStart( GLcontext *ctx ) */ void radeonChooseVertexState( GLcontext *ctx ) { - radeonContextPtr rmesa = RADEON_CONTEXT( ctx ); + r100ContextPtr rmesa = R100_CONTEXT( ctx ); TNLcontext *tnl = TNL_CONTEXT(ctx); GLuint se_coord_fmt = rmesa->hw.set.cmd[SET_SE_COORDFMT]; @@ -254,7 +273,7 @@ void radeonChooseVertexState( GLcontext *ctx ) * rasterization fallback. As this function will be called again when we * leave a rasterization fallback, we can just skip it for now. */ - if (rmesa->Fallback != 0) + if (rmesa->radeon.Fallback != 0) return; /* HW perspective divide is a win, but tiny vertex formats are a @@ -281,80 +300,33 @@ void radeonChooseVertexState( GLcontext *ctx ) } } - -/* Flush vertices in the current dma region. - */ -static void flush_last_swtcl_prim( radeonContextPtr rmesa ) +void r100_swtcl_flush(GLcontext *ctx, uint32_t current_offset) { - if (RADEON_DEBUG & DEBUG_IOCTL) - fprintf(stderr, "%s\n", __FUNCTION__); + r100ContextPtr rmesa = R100_CONTEXT(ctx); - rmesa->dma.flush = NULL; - if (rmesa->dma.current.buf) { - struct radeon_dma_region *current = &rmesa->dma.current; - GLuint current_offset = (rmesa->radeonScreen->gart_buffer_offset + - current->buf->buf->idx * RADEON_BUFFER_SIZE + - current->start); - assert (!(rmesa->swtcl.hw_primitive & RADEON_CP_VC_CNTL_PRIM_WALK_IND)); + radeonEmitState(&rmesa->radeon); + radeonEmitVertexAOS( rmesa, + rmesa->radeon.swtcl.vertex_size, + first_elem(&rmesa->radeon.dma.reserved)->bo, + current_offset); - assert (current->start + - rmesa->swtcl.numverts * rmesa->swtcl.vertex_size * 4 == - current->ptr); + + radeonEmitVbufPrim( rmesa, + rmesa->swtcl.vertex_format, + rmesa->radeon.swtcl.hw_primitive, + rmesa->radeon.swtcl.numverts); + if ( rmesa->radeon.swtcl.emit_prediction < rmesa->radeon.cmdbuf.cs->cdw ) + WARN_ONCE("Rendering was %d commands larger than predicted size." + " We might overflow command buffer.\n", + rmesa->radeon.cmdbuf.cs->cdw - rmesa->radeon.swtcl.emit_prediction ); - if (rmesa->dma.current.start != rmesa->dma.current.ptr) { - radeonEnsureCmdBufSpace( rmesa, VERT_AOS_BUFSZ + - rmesa->hw.max_state_size + VBUF_BUFSZ ); - radeonEmitVertexAOS( rmesa, - rmesa->swtcl.vertex_size, - current_offset); + rmesa->radeon.swtcl.emit_prediction = 0; - radeonEmitVbufPrim( rmesa, - rmesa->swtcl.vertex_format, - rmesa->swtcl.hw_primitive, - rmesa->swtcl.numverts); - } - - rmesa->swtcl.numverts = 0; - current->start = current->ptr; - } } - -/* Alloc space in the current dma region. - */ -static INLINE void * -radeonAllocDmaLowVerts( radeonContextPtr rmesa, int nverts, int vsize ) -{ - GLuint bytes = vsize * nverts; - - if ( rmesa->dma.current.ptr + bytes > rmesa->dma.current.end ) - radeonRefillCurrentDmaRegion( rmesa ); - - if (!rmesa->dma.flush) { - rmesa->glCtx->Driver.NeedFlush |= FLUSH_STORED_VERTICES; - rmesa->dma.flush = flush_last_swtcl_prim; - } - - assert( vsize == rmesa->swtcl.vertex_size * 4 ); - assert( rmesa->dma.flush == flush_last_swtcl_prim ); - assert (rmesa->dma.current.start + - rmesa->swtcl.numverts * rmesa->swtcl.vertex_size * 4 == - rmesa->dma.current.ptr); - - - { - GLubyte *head = (GLubyte *)(rmesa->dma.current.address + rmesa->dma.current.ptr); - rmesa->dma.current.ptr += bytes; - rmesa->swtcl.numverts += nverts; - return head; - } - -} - - /* * Render unclipped vertex buffers by emitting vertices directly to * dma buffers. Use strip/fan hardware primitives where possible. @@ -387,22 +359,31 @@ static const GLuint hw_prim[GL_POLYGON+1] = { }; static INLINE void -radeonDmaPrimitive( radeonContextPtr rmesa, GLenum prim ) +radeonDmaPrimitive( r100ContextPtr rmesa, GLenum prim ) { RADEON_NEWPRIM( rmesa ); - rmesa->swtcl.hw_primitive = hw_prim[prim]; - assert(rmesa->dma.current.ptr == rmesa->dma.current.start); + rmesa->radeon.swtcl.hw_primitive = hw_prim[prim]; + // assert(rmesa->radeon.dma.current.ptr == rmesa->radeon.dma.current.start); +} + +static void* radeon_alloc_verts( r100ContextPtr rmesa , GLuint nr, GLuint size ) +{ + void *rv; + do { + radeon_predict_emit_size( rmesa ); + rv = rcommonAllocDmaLowVerts( &rmesa->radeon, nr, size ); + } while (!rv); + return rv; } -#define LOCAL_VARS radeonContextPtr rmesa = RADEON_CONTEXT(ctx) +#define LOCAL_VARS r100ContextPtr rmesa = R100_CONTEXT(ctx) #define INIT( prim ) radeonDmaPrimitive( rmesa, prim ) #define FLUSH() RADEON_NEWPRIM( rmesa ) -#define GET_CURRENT_VB_MAX_VERTS() \ - (((int)rmesa->dma.current.end - (int)rmesa->dma.current.ptr) / (rmesa->swtcl.vertex_size*4)) +#define GET_CURRENT_VB_MAX_VERTS() 10\ +// (((int)rmesa->radeon.dma.current.end - (int)rmesa->radeon.dma.current.ptr) / (rmesa->radeon.swtcl.vertex_size*4)) #define GET_SUBSEQUENT_VB_MAX_VERTS() \ - ((RADEON_BUFFER_SIZE) / (rmesa->swtcl.vertex_size*4)) -#define ALLOC_VERTS( nr ) \ - radeonAllocDmaLowVerts( rmesa, nr, rmesa->swtcl.vertex_size * 4 ) + ((RADEON_BUFFER_SIZE) / (rmesa->radeon.swtcl.vertex_size*4)) +#define ALLOC_VERTS( nr ) radeon_alloc_verts( rmesa, nr, rmesa->radeon.swtcl.vertex_size * 4 ) #define EMIT_VERTS( ctx, j, nr, buf ) \ _tnl_emit_vertices_to_buffer(ctx, j, (j)+(nr), buf) @@ -418,16 +399,13 @@ radeonDmaPrimitive( radeonContextPtr rmesa, GLenum prim ) static GLboolean radeon_run_render( GLcontext *ctx, struct tnl_pipeline_stage *stage ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); TNLcontext *tnl = TNL_CONTEXT(ctx); struct vertex_buffer *VB = &tnl->vb; tnl_render_func *tab = TAG(render_tab_verts); GLuint i; - if (rmesa->swtcl.indexed_verts.buf) - RELEASE_ELT_VERTS(); - - if (rmesa->swtcl.RenderIndex != 0 || + if (rmesa->radeon.swtcl.RenderIndex != 0 || !radeon_dma_validate_render( ctx, VB )) return GL_TRUE; @@ -442,8 +420,8 @@ static GLboolean radeon_run_render( GLcontext *ctx, if (!length) continue; - if (RADEON_DEBUG & DEBUG_PRIMS) - fprintf(stderr, "radeon_render.c: prim %s %d..%d\n", + radeon_print(RADEON_SWRENDER, RADEON_NORMAL, + "radeon_render.c: prim %s %d..%d\n", _mesa_lookup_enum_by_nr(prim & PRIM_MODE_MASK), start, start+length); @@ -496,13 +474,13 @@ static void radeonResetLineStipple( GLcontext *ctx ); #undef LOCAL_VARS #undef ALLOC_VERTS -#define CTX_ARG radeonContextPtr rmesa -#define GET_VERTEX_DWORDS() rmesa->swtcl.vertex_size -#define ALLOC_VERTS( n, size ) radeonAllocDmaLowVerts( rmesa, n, (size) * 4 ) +#define CTX_ARG r100ContextPtr rmesa +#define GET_VERTEX_DWORDS() rmesa->radeon.swtcl.vertex_size +#define ALLOC_VERTS( n, size ) radeon_alloc_verts( rmesa, n, (size) * 4 ) #undef LOCAL_VARS #define LOCAL_VARS \ - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); \ - const char *radeonverts = (char *)rmesa->swtcl.verts; + r100ContextPtr rmesa = R100_CONTEXT(ctx); \ + const char *radeonverts = (char *)rmesa->radeon.swtcl.verts; #define VERT(x) (radeonVertex *)(radeonverts + ((x) * (vertsize) * sizeof(int))) #define VERTEX radeonVertex #undef TAG @@ -560,7 +538,7 @@ static struct { #define VERT_Y(_v) _v->v.y #define VERT_Z(_v) _v->v.z #define AREA_IS_CCW( a ) (a < 0) -#define GET_VERTEX(e) (rmesa->swtcl.verts + ((e) * rmesa->swtcl.vertex_size * sizeof(int))) +#define GET_VERTEX(e) (rmesa->radeon.swtcl.verts + ((e) * rmesa->radeon.swtcl.vertex_size * sizeof(int))) #define VERT_SET_RGBA( v, c ) \ do { \ @@ -606,8 +584,8 @@ do { \ #undef INIT #define LOCAL_VARS(n) \ - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); \ - GLuint color[n], spec[n]; \ + r100ContextPtr rmesa = R100_CONTEXT(ctx); \ + GLuint color[n] = {0}, spec[n] = {0}; \ GLuint coloroffset = rmesa->swtcl.coloroffset; \ GLuint specoffset = rmesa->swtcl.specoffset; \ (void) color; (void) spec; (void) coloroffset; (void) specoffset; @@ -617,7 +595,7 @@ do { \ ***********************************************************************/ #define RASTERIZE(x) radeonRasterPrimitive( ctx, reduced_hw_prim[x] ) -#define RENDER_PRIMITIVE rmesa->swtcl.render_primitive +#define RENDER_PRIMITIVE rmesa->radeon.swtcl.render_primitive #undef TAG #define TAG(x) x #include "tnl_dd/t_dd_unfilled.h" @@ -673,9 +651,9 @@ static void init_rast_tab( void ) } while (0) #undef LOCAL_VARS #define LOCAL_VARS \ - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); \ - const GLuint vertsize = rmesa->swtcl.vertex_size; \ - const char *radeonverts = (char *)rmesa->swtcl.verts; \ + r100ContextPtr rmesa = R100_CONTEXT(ctx); \ + const GLuint vertsize = rmesa->radeon.swtcl.vertex_size; \ + const char *radeonverts = (char *)rmesa->radeon.swtcl.verts; \ const GLuint * const elt = TNL_CONTEXT(ctx)->vb.Elts; \ const GLboolean stipple = ctx->Line.StippleFlag; \ (void) elt; (void) stipple; @@ -700,17 +678,17 @@ static void init_rast_tab( void ) void radeonChooseRenderState( GLcontext *ctx ) { TNLcontext *tnl = TNL_CONTEXT(ctx); - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); GLuint index = 0; GLuint flags = ctx->_TriangleCaps; - if (!rmesa->TclFallback || rmesa->Fallback) + if (!rmesa->radeon.TclFallback || rmesa->radeon.Fallback) return; if (flags & DD_TRI_LIGHT_TWOSIDE) index |= RADEON_TWOSIDE_BIT; if (flags & DD_TRI_UNFILLED) index |= RADEON_UNFILLED_BIT; - if (index != rmesa->swtcl.RenderIndex) { + if (index != rmesa->radeon.swtcl.RenderIndex) { tnl->Driver.Render.Points = rast_tab[index].points; tnl->Driver.Render.Line = rast_tab[index].line; tnl->Driver.Render.ClippedLine = rast_tab[index].line; @@ -727,7 +705,7 @@ void radeonChooseRenderState( GLcontext *ctx ) tnl->Driver.Render.ClippedPolygon = _tnl_RenderClippedPolygon; } - rmesa->swtcl.RenderIndex = index; + rmesa->radeon.swtcl.RenderIndex = index; } } @@ -739,18 +717,18 @@ void radeonChooseRenderState( GLcontext *ctx ) static void radeonRasterPrimitive( GLcontext *ctx, GLuint hwprim ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); - if (rmesa->swtcl.hw_primitive != hwprim) { + if (rmesa->radeon.swtcl.hw_primitive != hwprim) { RADEON_NEWPRIM( rmesa ); - rmesa->swtcl.hw_primitive = hwprim; + rmesa->radeon.swtcl.hw_primitive = hwprim; } } static void radeonRenderPrimitive( GLcontext *ctx, GLenum prim ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - rmesa->swtcl.render_primitive = prim; + r100ContextPtr rmesa = R100_CONTEXT(ctx); + rmesa->radeon.swtcl.render_primitive = prim; if (prim < GL_TRIANGLES || !(ctx->_TriangleCaps & DD_TRI_UNFILLED)) radeonRasterPrimitive( ctx, reduced_hw_prim[prim] ); } @@ -761,7 +739,7 @@ static void radeonRenderFinish( GLcontext *ctx ) static void radeonResetLineStipple( GLcontext *ctx ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); RADEON_STATECHANGE( rmesa, lin ); } @@ -795,25 +773,25 @@ static const char *getFallbackString(GLuint bit) void radeonFallback( GLcontext *ctx, GLuint bit, GLboolean mode ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); TNLcontext *tnl = TNL_CONTEXT(ctx); - GLuint oldfallback = rmesa->Fallback; + GLuint oldfallback = rmesa->radeon.Fallback; if (mode) { - rmesa->Fallback |= bit; + rmesa->radeon.Fallback |= bit; if (oldfallback == 0) { - RADEON_FIREVERTICES( rmesa ); + radeon_firevertices(&rmesa->radeon); TCL_FALLBACK( ctx, RADEON_TCL_FALLBACK_RASTER, GL_TRUE ); _swsetup_Wakeup( ctx ); - rmesa->swtcl.RenderIndex = ~0; - if (RADEON_DEBUG & DEBUG_FALLBACKS) { + rmesa->radeon.swtcl.RenderIndex = ~0; + if (RADEON_DEBUG & RADEON_FALLBACKS) { fprintf(stderr, "Radeon begin rasterization fallback: 0x%x %s\n", bit, getFallbackString(bit)); } } } else { - rmesa->Fallback &= ~bit; + rmesa->radeon.Fallback &= ~bit; if (oldfallback == bit) { _swrast_flush( ctx ); tnl->Driver.Render.Start = radeonRenderStart; @@ -826,18 +804,18 @@ void radeonFallback( GLcontext *ctx, GLuint bit, GLboolean mode ) tnl->Driver.Render.ResetLineStipple = radeonResetLineStipple; TCL_FALLBACK( ctx, RADEON_TCL_FALLBACK_RASTER, GL_FALSE ); - if (rmesa->TclFallback) { - /* These are already done if rmesa->TclFallback goes to + if (rmesa->radeon.TclFallback) { + /* These are already done if rmesa->radeon.TclFallback goes to * zero above. But not if it doesn't (RADEON_NO_TCL for * example?) */ _tnl_invalidate_vertex_state( ctx, ~0 ); _tnl_invalidate_vertices( ctx, ~0 ); - RENDERINPUTS_ZERO( rmesa->tnl_index_bitset ); + RENDERINPUTS_ZERO( rmesa->radeon.tnl_index_bitset ); radeonChooseVertexState( ctx ); radeonChooseRenderState( ctx ); } - if (RADEON_DEBUG & DEBUG_FALLBACKS) { + if (RADEON_DEBUG & RADEON_FALLBACKS) { fprintf(stderr, "Radeon end rasterization fallback: 0x%x %s\n", bit, getFallbackString(bit)); } @@ -853,13 +831,14 @@ void radeonFallback( GLcontext *ctx, GLuint bit, GLboolean mode ) void radeonInitSwtcl( GLcontext *ctx ) { TNLcontext *tnl = TNL_CONTEXT(ctx); - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); static int firsttime = 1; if (firsttime) { init_rast_tab(); firsttime = 0; } + rmesa->radeon.swtcl.emit_prediction = 0; tnl->Driver.Render.Start = radeonRenderStart; tnl->Driver.Render.Finish = radeonRenderFinish; @@ -872,18 +851,9 @@ void radeonInitSwtcl( GLcontext *ctx ) _tnl_init_vertices( ctx, ctx->Const.MaxArrayLockSize + 12, RADEON_MAX_TNL_VERTEX_SIZE); - rmesa->swtcl.verts = (GLubyte *)tnl->clipspace.vertex_buf; - rmesa->swtcl.RenderIndex = ~0; - rmesa->swtcl.render_primitive = GL_TRIANGLES; - rmesa->swtcl.hw_primitive = 0; + rmesa->radeon.swtcl.verts = (GLubyte *)tnl->clipspace.vertex_buf; + rmesa->radeon.swtcl.RenderIndex = ~0; + rmesa->radeon.swtcl.render_primitive = GL_TRIANGLES; + rmesa->radeon.swtcl.hw_primitive = 0; } - -void radeonDestroySwtcl( GLcontext *ctx ) -{ - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - - if (rmesa->swtcl.indexed_verts.buf) - radeonReleaseDmaRegion( rmesa, &rmesa->swtcl.indexed_verts, - __FUNCTION__ ); -} diff --git a/src/mesa/drivers/dri/radeon/radeon_swtcl.h b/src/mesa/drivers/dri/radeon/radeon_swtcl.h index e485052ad77..da89158eeb9 100644 --- a/src/mesa/drivers/dri/radeon/radeon_swtcl.h +++ b/src/mesa/drivers/dri/radeon/radeon_swtcl.h @@ -40,7 +40,6 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. #include "radeon_context.h" extern void radeonInitSwtcl( GLcontext *ctx ); -extern void radeonDestroySwtcl( GLcontext *ctx ); extern void radeonChooseRenderState( GLcontext *ctx ); extern void radeonChooseVertexState( GLcontext *ctx ); @@ -63,5 +62,5 @@ extern void radeon_translate_vertex( GLcontext *ctx, extern void radeon_print_vertex( GLcontext *ctx, const radeonVertex *v ); - +extern void r100_swtcl_flush(GLcontext *ctx, uint32_t current_offset); #endif diff --git a/src/mesa/drivers/dri/radeon/radeon_tcl.c b/src/mesa/drivers/dri/radeon/radeon_tcl.c index 779e9ae5df0..b334ea05e5b 100644 --- a/src/mesa/drivers/dri/radeon/radeon_tcl.c +++ b/src/mesa/drivers/dri/radeon/radeon_tcl.c @@ -42,6 +42,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "tnl/tnl.h" #include "tnl/t_pipeline.h" +#include "radeon_common.h" #include "radeon_context.h" #include "radeon_state.h" #include "radeon_ioctl.h" @@ -49,6 +50,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "radeon_tcl.h" #include "radeon_swtcl.h" #include "radeon_maos.h" +#include "radeon_common_context.h" @@ -104,7 +106,7 @@ static GLboolean discrete_prim[0x10] = { }; -#define LOCAL_VARS radeonContextPtr rmesa = RADEON_CONTEXT(ctx) +#define LOCAL_VARS r100ContextPtr rmesa = R100_CONTEXT(ctx) #define ELT_TYPE GLushort #define ELT_INIT(prim, hw_prim) \ @@ -125,7 +127,7 @@ static GLboolean discrete_prim[0x10] = { #define RESET_STIPPLE() do { \ RADEON_STATECHANGE( rmesa, lin ); \ - radeonEmitState( rmesa ); \ + radeonEmitState(&rmesa->radeon); \ } while (0) #define AUTO_STIPPLE( mode ) do { \ @@ -136,31 +138,26 @@ static GLboolean discrete_prim[0x10] = { else \ rmesa->hw.lin.cmd[LIN_RE_LINE_PATTERN] &= \ ~RADEON_LINE_PATTERN_AUTO_RESET; \ - radeonEmitState( rmesa ); \ + radeonEmitState(&rmesa->radeon); \ } while (0) #define ALLOC_ELTS(nr) radeonAllocElts( rmesa, nr ) -static GLushort *radeonAllocElts( radeonContextPtr rmesa, GLuint nr ) +static GLushort *radeonAllocElts( r100ContextPtr rmesa, GLuint nr ) { - if (rmesa->dma.flush) - rmesa->dma.flush( rmesa ); + if (rmesa->radeon.dma.flush) + rmesa->radeon.dma.flush( rmesa->radeon.glCtx ); - radeonEnsureCmdBufSpace(rmesa, AOS_BUFSZ(rmesa->tcl.nr_aos_components) + - rmesa->hw.max_state_size + ELTS_BUFSZ(nr)); + radeonEmitAOS( rmesa, + rmesa->radeon.tcl.aos_count, 0 ); - radeonEmitAOS( rmesa, - rmesa->tcl.aos_components, - rmesa->tcl.nr_aos_components, 0 ); - - return radeonAllocEltsOpenEnded( rmesa, - rmesa->tcl.vertex_format, - rmesa->tcl.hw_primitive, nr ); + return radeonAllocEltsOpenEnded( rmesa, rmesa->tcl.vertex_format, + rmesa->tcl.hw_primitive, nr ); } -#define CLOSE_ELTS() RADEON_NEWPRIM( rmesa ) +#define CLOSE_ELTS() if (0) RADEON_NEWPRIM( rmesa ) @@ -174,15 +171,11 @@ static void radeonEmitPrim( GLcontext *ctx, GLuint start, GLuint count) { - radeonContextPtr rmesa = RADEON_CONTEXT( ctx ); + r100ContextPtr rmesa = R100_CONTEXT( ctx ); radeonTclPrimitive( ctx, prim, hwprim ); - radeonEnsureCmdBufSpace( rmesa, AOS_BUFSZ(rmesa->tcl.nr_aos_components) + - rmesa->hw.max_state_size + VBUF_BUFSZ ); - radeonEmitAOS( rmesa, - rmesa->tcl.aos_components, - rmesa->tcl.nr_aos_components, + rmesa->radeon.tcl.aos_count, start ); /* Why couldn't this packet have taken an offset param? @@ -197,6 +190,8 @@ static void radeonEmitPrim( GLcontext *ctx, radeonEmitPrim( ctx, prim, hwprim, start, count ); \ (void) rmesa; } while (0) +#define MAX_CONVERSION_SIZE 40 + /* Try & join small primitives */ #if 0 @@ -254,7 +249,7 @@ void radeonTclPrimitive( GLcontext *ctx, GLenum prim, int hw_prim ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); GLuint se_cntl; GLuint newprim = hw_prim | RADEON_CP_VC_CNTL_TCL_ENABLE; @@ -361,6 +356,73 @@ radeonComputeFogBlendFactor( GLcontext *ctx, GLfloat fogcoord ) } } +/** + * Predict total emit size for next rendering operation so there is no flush in middle of rendering + * Prediction has to aim towards the best possible value that is worse than worst case scenario + */ +static GLuint radeonEnsureEmitSize( GLcontext * ctx , GLuint inputs ) +{ + r100ContextPtr rmesa = R100_CONTEXT(ctx); + TNLcontext *tnl = TNL_CONTEXT(ctx); + struct vertex_buffer *VB = &tnl->vb; + GLuint space_required; + GLuint state_size; + GLuint nr_aos = 1; /* radeonEmitArrays does always emit one */ + int i; + /* list of flags that are allocating aos object */ + const GLuint flags_to_check[] = { + VERT_BIT_NORMAL, + VERT_BIT_COLOR0, + VERT_BIT_COLOR1, + VERT_BIT_FOG + }; + /* predict number of aos to emit */ + for (i=0; i < sizeof(flags_to_check)/sizeof(flags_to_check[0]); ++i) + { + if (inputs & flags_to_check[i]) + ++nr_aos; + } + for (i = 0; i < ctx->Const.MaxTextureUnits; ++i) + { + if (inputs & VERT_BIT_TEX(i)) + ++nr_aos; + } + + { + /* count the prediction for state size */ + space_required = 0; + state_size = radeonCountStateEmitSize( &rmesa->radeon ); + /* tcl may be changed in radeonEmitArrays so account for it if not dirty */ + if (!rmesa->hw.tcl.dirty) + state_size += rmesa->hw.tcl.check( rmesa->radeon.glCtx, &rmesa->hw.tcl ); + /* predict size for elements */ + for (i = 0; i < VB->PrimitiveCount; ++i) + { + if (!VB->Primitive[i].count) + continue; + /* If primitive.count is less than MAX_CONVERSION_SIZE + rendering code may decide convert to elts. + In that case we have to make pessimistic prediction. + and use larger of 2 paths. */ + const GLuint elts = ELTS_BUFSZ(nr_aos); + const GLuint index = INDEX_BUFSZ; + const GLuint vbuf = VBUF_BUFSZ; + if ( (!VB->Elts && VB->Primitive[i].count >= MAX_CONVERSION_SIZE) + || vbuf > index + elts) + space_required += vbuf; + else + space_required += index + elts; + space_required += AOS_BUFSZ(nr_aos); + } + space_required += SCISSOR_BUFSZ; + } + /* flush the buffer in case we need more than is left. */ + if (rcommonEnsureCmdBufSpace(&rmesa->radeon, space_required, __FUNCTION__)) + return space_required + radeonCountStateEmitSize( &rmesa->radeon ); + else + return space_required + state_size; +} + /**********************************************************************/ /* Render pipeline stage */ /**********************************************************************/ @@ -371,7 +433,7 @@ radeonComputeFogBlendFactor( GLcontext *ctx, GLfloat fogcoord ) static GLboolean radeon_run_tcl_render( GLcontext *ctx, struct tnl_pipeline_stage *stage ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); TNLcontext *tnl = TNL_CONTEXT(ctx); struct vertex_buffer *VB = &tnl->vb; GLuint inputs = VERT_BIT_POS | VERT_BIT_COLOR0; @@ -379,7 +441,7 @@ static GLboolean radeon_run_tcl_render( GLcontext *ctx, /* TODO: separate this from the swtnl pipeline */ - if (rmesa->TclFallback) + if (rmesa->radeon.TclFallback) return GL_TRUE; /* fallback to software t&l */ if (VB->Count == 0) @@ -411,6 +473,8 @@ static GLboolean radeon_run_tcl_render( GLcontext *ctx, } radeonReleaseArrays( ctx, ~0 ); + GLuint emit_end = radeonEnsureEmitSize( ctx, inputs ) + + rmesa->radeon.cmdbuf.cs->cdw; radeonEmitArrays( ctx, inputs ); rmesa->tcl.Elts = VB->Elts; @@ -430,6 +494,10 @@ static GLboolean radeon_run_tcl_render( GLcontext *ctx, radeonEmitPrimitive( ctx, start, start+length, prim ); } + if (emit_end < rmesa->radeon.cmdbuf.cs->cdw) + WARN_ONCE("Rendering was %d commands larger than predicted size." + " We might overflow command buffer.\n", rmesa->radeon.cmdbuf.cs->cdw - emit_end); + return GL_FALSE; /* finished the pipe */ } @@ -461,7 +529,7 @@ const struct tnl_pipeline_stage _radeon_tcl_stage = static void transition_to_swtnl( GLcontext *ctx ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); TNLcontext *tnl = TNL_CONTEXT(ctx); GLuint se_cntl; @@ -490,7 +558,7 @@ static void transition_to_swtnl( GLcontext *ctx ) static void transition_to_hwtnl( GLcontext *ctx ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); TNLcontext *tnl = TNL_CONTEXT(ctx); GLuint se_coord_fmt = rmesa->hw.set.cmd[SET_SE_COORDFMT]; @@ -509,17 +577,17 @@ static void transition_to_hwtnl( GLcontext *ctx ) tnl->Driver.NotifyMaterialChange = radeonUpdateMaterial; - if ( rmesa->dma.flush ) - rmesa->dma.flush( rmesa ); + if ( rmesa->radeon.dma.flush ) + rmesa->radeon.dma.flush( rmesa->radeon.glCtx ); - rmesa->dma.flush = NULL; + rmesa->radeon.dma.flush = NULL; rmesa->swtcl.vertex_format = 0; - if (rmesa->swtcl.indexed_verts.buf) - radeonReleaseDmaRegion( rmesa, &rmesa->swtcl.indexed_verts, - __FUNCTION__ ); + // if (rmesa->swtcl.indexed_verts.buf) + // radeonReleaseDmaRegion( rmesa, &rmesa->swtcl.indexed_verts, + // __FUNCTION__ ); - if (RADEON_DEBUG & DEBUG_FALLBACKS) + if (RADEON_DEBUG & RADEON_FALLBACKS) fprintf(stderr, "Radeon end tcl fallback\n"); } @@ -550,22 +618,22 @@ static char *getFallbackString(GLuint bit) void radeonTclFallback( GLcontext *ctx, GLuint bit, GLboolean mode ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - GLuint oldfallback = rmesa->TclFallback; + r100ContextPtr rmesa = R100_CONTEXT(ctx); + GLuint oldfallback = rmesa->radeon.TclFallback; if (mode) { - rmesa->TclFallback |= bit; + rmesa->radeon.TclFallback |= bit; if (oldfallback == 0) { - if (RADEON_DEBUG & DEBUG_FALLBACKS) + if (RADEON_DEBUG & RADEON_FALLBACKS) fprintf(stderr, "Radeon begin tcl fallback %s\n", getFallbackString( bit )); transition_to_swtnl( ctx ); } } else { - rmesa->TclFallback &= ~bit; + rmesa->radeon.TclFallback &= ~bit; if (oldfallback == bit) { - if (RADEON_DEBUG & DEBUG_FALLBACKS) + if (RADEON_DEBUG & RADEON_FALLBACKS) fprintf(stderr, "Radeon end tcl fallback %s\n", getFallbackString( bit )); transition_to_hwtnl( ctx ); diff --git a/src/mesa/drivers/dri/radeon/radeon_tex.c b/src/mesa/drivers/dri/radeon/radeon_tex.c index f2b6deb9c04..99865fff27b 100644 --- a/src/mesa/drivers/dri/radeon/radeon_tex.c +++ b/src/mesa/drivers/dri/radeon/radeon_tex.c @@ -44,6 +44,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "main/texobj.h" #include "radeon_context.h" +#include "radeon_mipmap_tree.h" #include "radeon_state.h" #include "radeon_ioctl.h" #include "radeon_swtcl.h" @@ -170,10 +171,13 @@ static void radeonSetTexFilter( radeonTexObjPtr t, GLenum minf, GLenum magf ) { GLuint anisotropy = (t->pp_txfilter & RADEON_MAX_ANISO_MASK); + /* Force revalidation to account for switches from/to mipmapping. */ + t->validated = GL_FALSE; + t->pp_txfilter &= ~(RADEON_MIN_FILTER_MASK | RADEON_MAG_FILTER_MASK); /* r100 chips can't handle mipmaps/aniso for cubemap/volume textures */ - if ( t->base.tObj->Target == GL_TEXTURE_CUBE_MAP ) { + if ( t->base.Target == GL_TEXTURE_CUBE_MAP ) { switch ( minf ) { case GL_NEAREST: case GL_NEAREST_MIPMAP_NEAREST: @@ -249,437 +253,17 @@ static void radeonSetTexBorderColor( radeonTexObjPtr t, const GLfloat color[4] ) t->pp_border_color = radeonPackColor( 4, c[0], c[1], c[2], c[3] ); } - -/** - * Allocate space for and load the mesa images into the texture memory block. - * This will happen before drawing with a new texture, or drawing with a - * texture after it was swapped out or teximaged again. - */ - -static radeonTexObjPtr radeonAllocTexObj( struct gl_texture_object *texObj ) -{ - radeonTexObjPtr t; - - t = CALLOC_STRUCT( radeon_tex_obj ); - texObj->DriverData = t; - if ( t != NULL ) { - if ( RADEON_DEBUG & DEBUG_TEXTURE ) { - fprintf( stderr, "%s( %p, %p )\n", __FUNCTION__, (void *)texObj, (void *)t ); - } - - /* Initialize non-image-dependent parts of the state: - */ - t->base.tObj = texObj; - t->border_fallback = GL_FALSE; - - t->pp_txfilter = RADEON_BORDER_MODE_OGL; - t->pp_txformat = (RADEON_TXFORMAT_ENDIAN_NO_SWAP | - RADEON_TXFORMAT_PERSPECTIVE_ENABLE); - - make_empty_list( & t->base ); - - radeonSetTexWrap( t, texObj->WrapS, texObj->WrapT ); - radeonSetTexMaxAnisotropy( t, texObj->MaxAnisotropy ); - radeonSetTexFilter( t, texObj->MinFilter, texObj->MagFilter ); - radeonSetTexBorderColor( t, texObj->BorderColor ); - } - - return t; -} - - -static const struct gl_texture_format * -radeonChooseTextureFormat( GLcontext *ctx, GLint internalFormat, - GLenum format, GLenum type ) -{ - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - const GLboolean do32bpt = - ( rmesa->texture_depth == DRI_CONF_TEXTURE_DEPTH_32 ); - const GLboolean force16bpt = - ( rmesa->texture_depth == DRI_CONF_TEXTURE_DEPTH_FORCE_16 ); - (void) format; - - switch ( internalFormat ) { - case 4: - case GL_RGBA: - case GL_COMPRESSED_RGBA: - switch ( type ) { - case GL_UNSIGNED_INT_10_10_10_2: - case GL_UNSIGNED_INT_2_10_10_10_REV: - return do32bpt ? _dri_texformat_argb8888 : _dri_texformat_argb1555; - case GL_UNSIGNED_SHORT_4_4_4_4: - case GL_UNSIGNED_SHORT_4_4_4_4_REV: - return _dri_texformat_argb4444; - case GL_UNSIGNED_SHORT_5_5_5_1: - case GL_UNSIGNED_SHORT_1_5_5_5_REV: - return _dri_texformat_argb1555; - default: - return do32bpt ? _dri_texformat_argb8888 : _dri_texformat_argb4444; - } - - case 3: - case GL_RGB: - case GL_COMPRESSED_RGB: - switch ( type ) { - case GL_UNSIGNED_SHORT_4_4_4_4: - case GL_UNSIGNED_SHORT_4_4_4_4_REV: - return _dri_texformat_argb4444; - case GL_UNSIGNED_SHORT_5_5_5_1: - case GL_UNSIGNED_SHORT_1_5_5_5_REV: - return _dri_texformat_argb1555; - case GL_UNSIGNED_SHORT_5_6_5: - case GL_UNSIGNED_SHORT_5_6_5_REV: - return _dri_texformat_rgb565; - default: - return do32bpt ? _dri_texformat_argb8888 : _dri_texformat_rgb565; - } - - case GL_RGBA8: - case GL_RGB10_A2: - case GL_RGBA12: - case GL_RGBA16: - return !force16bpt ? - _dri_texformat_argb8888 : _dri_texformat_argb4444; - - case GL_RGBA4: - case GL_RGBA2: - return _dri_texformat_argb4444; - - case GL_RGB5_A1: - return _dri_texformat_argb1555; - - case GL_RGB8: - case GL_RGB10: - case GL_RGB12: - case GL_RGB16: - return !force16bpt ? _dri_texformat_argb8888 : _dri_texformat_rgb565; - - case GL_RGB5: - case GL_RGB4: - case GL_R3_G3_B2: - return _dri_texformat_rgb565; - - case GL_ALPHA: - case GL_ALPHA4: - case GL_ALPHA8: - case GL_ALPHA12: - case GL_ALPHA16: - case GL_COMPRESSED_ALPHA: - return _dri_texformat_a8; - - case 1: - case GL_LUMINANCE: - case GL_LUMINANCE4: - case GL_LUMINANCE8: - case GL_LUMINANCE12: - case GL_LUMINANCE16: - case GL_COMPRESSED_LUMINANCE: - return _dri_texformat_l8; - - case 2: - case GL_LUMINANCE_ALPHA: - case GL_LUMINANCE4_ALPHA4: - case GL_LUMINANCE6_ALPHA2: - case GL_LUMINANCE8_ALPHA8: - case GL_LUMINANCE12_ALPHA4: - case GL_LUMINANCE12_ALPHA12: - case GL_LUMINANCE16_ALPHA16: - case GL_COMPRESSED_LUMINANCE_ALPHA: - return _dri_texformat_al88; - - case GL_INTENSITY: - case GL_INTENSITY4: - case GL_INTENSITY8: - case GL_INTENSITY12: - case GL_INTENSITY16: - case GL_COMPRESSED_INTENSITY: - return _dri_texformat_i8; - - case GL_YCBCR_MESA: - if (type == GL_UNSIGNED_SHORT_8_8_APPLE || - type == GL_UNSIGNED_BYTE) - return &_mesa_texformat_ycbcr; - else - return &_mesa_texformat_ycbcr_rev; - - case GL_RGB_S3TC: - case GL_RGB4_S3TC: - case GL_COMPRESSED_RGB_S3TC_DXT1_EXT: - return &_mesa_texformat_rgb_dxt1; - - case GL_COMPRESSED_RGBA_S3TC_DXT1_EXT: - return &_mesa_texformat_rgba_dxt1; - - case GL_RGBA_S3TC: - case GL_RGBA4_S3TC: - case GL_COMPRESSED_RGBA_S3TC_DXT3_EXT: - return &_mesa_texformat_rgba_dxt3; - - case GL_COMPRESSED_RGBA_S3TC_DXT5_EXT: - return &_mesa_texformat_rgba_dxt5; - - default: - _mesa_problem(ctx, "unexpected texture format in %s", __FUNCTION__); - return NULL; - } - - return NULL; /* never get here */ -} - - -static void radeonTexImage1D( GLcontext *ctx, GLenum target, GLint level, - GLint internalFormat, - GLint width, GLint border, - GLenum format, GLenum type, const GLvoid *pixels, - const struct gl_pixelstore_attrib *packing, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage ) -{ - driTextureObject * t = (driTextureObject *) texObj->DriverData; - - if ( t ) { - driSwapOutTextureObject( t ); - } - else { - t = (driTextureObject *) radeonAllocTexObj( texObj ); - if (!t) { - _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage1D"); - return; - } - } - - /* Note, this will call ChooseTextureFormat */ - _mesa_store_teximage1d(ctx, target, level, internalFormat, - width, border, format, type, pixels, - &ctx->Unpack, texObj, texImage); - - t->dirty_images[0] |= (1 << level); -} - - -static void radeonTexSubImage1D( GLcontext *ctx, GLenum target, GLint level, - GLint xoffset, - GLsizei width, - GLenum format, GLenum type, - const GLvoid *pixels, - const struct gl_pixelstore_attrib *packing, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage ) -{ - driTextureObject * t = (driTextureObject *) texObj->DriverData; - - assert( t ); /* this _should_ be true */ - if ( t ) { - driSwapOutTextureObject( t ); - } - else { - t = (driTextureObject *) radeonAllocTexObj( texObj ); - if (!t) { - _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage1D"); - return; - } - } - - _mesa_store_texsubimage1d(ctx, target, level, xoffset, width, - format, type, pixels, packing, texObj, - texImage); - - t->dirty_images[0] |= (1 << level); -} - - -static void radeonTexImage2D( GLcontext *ctx, GLenum target, GLint level, - GLint internalFormat, - GLint width, GLint height, GLint border, - GLenum format, GLenum type, const GLvoid *pixels, - const struct gl_pixelstore_attrib *packing, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage ) -{ - driTextureObject * t = (driTextureObject *) texObj->DriverData; - GLuint face; - - /* which cube face or ordinary 2D image */ - switch (target) { - case GL_TEXTURE_CUBE_MAP_POSITIVE_X: - case GL_TEXTURE_CUBE_MAP_NEGATIVE_X: - case GL_TEXTURE_CUBE_MAP_POSITIVE_Y: - case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y: - case GL_TEXTURE_CUBE_MAP_POSITIVE_Z: - case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z: - face = (GLuint) target - (GLuint) GL_TEXTURE_CUBE_MAP_POSITIVE_X; - ASSERT(face < 6); - break; - default: - face = 0; - } - - if ( t != NULL ) { - driSwapOutTextureObject( t ); - } - else { - t = (driTextureObject *) radeonAllocTexObj( texObj ); - if (!t) { - _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage2D"); - return; - } - } - - /* Note, this will call ChooseTextureFormat */ - _mesa_store_teximage2d(ctx, target, level, internalFormat, - width, height, border, format, type, pixels, - &ctx->Unpack, texObj, texImage); - - t->dirty_images[face] |= (1 << level); -} - - -static void radeonTexSubImage2D( GLcontext *ctx, GLenum target, GLint level, - GLint xoffset, GLint yoffset, - GLsizei width, GLsizei height, - GLenum format, GLenum type, - const GLvoid *pixels, - const struct gl_pixelstore_attrib *packing, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage ) -{ - driTextureObject * t = (driTextureObject *) texObj->DriverData; - GLuint face; - - /* which cube face or ordinary 2D image */ - switch (target) { - case GL_TEXTURE_CUBE_MAP_POSITIVE_X: - case GL_TEXTURE_CUBE_MAP_NEGATIVE_X: - case GL_TEXTURE_CUBE_MAP_POSITIVE_Y: - case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y: - case GL_TEXTURE_CUBE_MAP_POSITIVE_Z: - case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z: - face = (GLuint) target - (GLuint) GL_TEXTURE_CUBE_MAP_POSITIVE_X; - ASSERT(face < 6); - break; - default: - face = 0; - } - - assert( t ); /* this _should_ be true */ - if ( t ) { - driSwapOutTextureObject( t ); - } - else { - t = (driTextureObject *) radeonAllocTexObj( texObj ); - if (!t) { - _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage2D"); - return; - } - } - - _mesa_store_texsubimage2d(ctx, target, level, xoffset, yoffset, width, - height, format, type, pixels, packing, texObj, - texImage); - - t->dirty_images[face] |= (1 << level); -} - -static void radeonCompressedTexImage2D( GLcontext *ctx, GLenum target, GLint level, - GLint internalFormat, - GLint width, GLint height, GLint border, - GLsizei imageSize, const GLvoid *data, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage ) -{ - driTextureObject * t = (driTextureObject *) texObj->DriverData; - GLuint face; - - /* which cube face or ordinary 2D image */ - switch (target) { - case GL_TEXTURE_CUBE_MAP_POSITIVE_X: - case GL_TEXTURE_CUBE_MAP_NEGATIVE_X: - case GL_TEXTURE_CUBE_MAP_POSITIVE_Y: - case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y: - case GL_TEXTURE_CUBE_MAP_POSITIVE_Z: - case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z: - face = (GLuint) target - (GLuint) GL_TEXTURE_CUBE_MAP_POSITIVE_X; - ASSERT(face < 6); - break; - default: - face = 0; - } - - if ( t != NULL ) { - driSwapOutTextureObject( t ); - } - else { - t = (driTextureObject *) radeonAllocTexObj( texObj ); - if (!t) { - _mesa_error(ctx, GL_OUT_OF_MEMORY, "glCompressedTexImage2D"); - return; - } - } - - /* Note, this will call ChooseTextureFormat */ - _mesa_store_compressed_teximage2d(ctx, target, level, internalFormat, width, - height, border, imageSize, data, texObj, texImage); - - t->dirty_images[face] |= (1 << level); -} - - -static void radeonCompressedTexSubImage2D( GLcontext *ctx, GLenum target, GLint level, - GLint xoffset, GLint yoffset, - GLsizei width, GLsizei height, - GLenum format, - GLsizei imageSize, const GLvoid *data, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage ) -{ - driTextureObject * t = (driTextureObject *) texObj->DriverData; - GLuint face; - - - /* which cube face or ordinary 2D image */ - switch (target) { - case GL_TEXTURE_CUBE_MAP_POSITIVE_X: - case GL_TEXTURE_CUBE_MAP_NEGATIVE_X: - case GL_TEXTURE_CUBE_MAP_POSITIVE_Y: - case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y: - case GL_TEXTURE_CUBE_MAP_POSITIVE_Z: - case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z: - face = (GLuint) target - (GLuint) GL_TEXTURE_CUBE_MAP_POSITIVE_X; - ASSERT(face < 6); - break; - default: - face = 0; - } - - assert( t ); /* this _should_ be true */ - if ( t ) { - driSwapOutTextureObject( t ); - } - else { - t = (driTextureObject *) radeonAllocTexObj( texObj ); - if (!t) { - _mesa_error(ctx, GL_OUT_OF_MEMORY, "glCompressedTexSubImage2D"); - return; - } - } - - _mesa_store_compressed_texsubimage2d(ctx, target, level, xoffset, yoffset, width, - height, format, imageSize, data, texObj, texImage); - - t->dirty_images[face] |= (1 << level); -} - #define SCALED_FLOAT_TO_BYTE( x, scale ) \ (((GLuint)((255.0F / scale) * (x))) / 2) static void radeonTexEnv( GLcontext *ctx, GLenum target, GLenum pname, const GLfloat *param ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); GLuint unit = ctx->Texture.CurrentUnit; struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit]; - if ( RADEON_DEBUG & DEBUG_STATE ) { + if ( RADEON_DEBUG & RADEON_STATE ) { fprintf( stderr, "%s( %s )\n", __FUNCTION__, _mesa_lookup_enum_by_nr( pname ) ); } @@ -706,7 +290,7 @@ static void radeonTexEnv( GLcontext *ctx, GLenum target, * functions, one mapping [-1.0,0.0] to [-128,0] and one mapping * [0.0,4.0] to [0,127]. */ - min = driQueryOptionb (&rmesa->optionCache, "no_neg_lod_bias") ? + min = driQueryOptionb (&rmesa->radeon.optionCache, "no_neg_lod_bias") ? 0.0 : -1.0; bias = CLAMP( *param, min, 4.0 ); if ( bias == 0 ) { @@ -739,12 +323,10 @@ static void radeonTexParameter( GLcontext *ctx, GLenum target, struct gl_texture_object *texObj, GLenum pname, const GLfloat *params ) { - radeonTexObjPtr t = (radeonTexObjPtr) texObj->DriverData; + radeonTexObj* t = radeon_tex_obj(texObj); - if ( RADEON_DEBUG & (DEBUG_STATE|DEBUG_TEXTURE) ) { - fprintf( stderr, "%s( %s )\n", __FUNCTION__, + radeon_print(RADEON_TEXTURE, RADEON_VERBOSE, "%s( %s )\n", __FUNCTION__, _mesa_lookup_enum_by_nr( pname ) ); - } switch ( pname ) { case GL_TEXTURE_MIN_FILTER: @@ -767,57 +349,50 @@ static void radeonTexParameter( GLcontext *ctx, GLenum target, case GL_TEXTURE_MAX_LEVEL: case GL_TEXTURE_MIN_LOD: case GL_TEXTURE_MAX_LOD: + /* This isn't the most efficient solution but there doesn't appear to * be a nice alternative. Since there's no LOD clamping, * we just have to rely on loading the right subset of mipmap levels * to simulate a clamped LOD. */ - driSwapOutTextureObject( (driTextureObject *) t ); + if (t->mt) { + radeon_miptree_unreference(t->mt); + t->mt = 0; + t->validated = GL_FALSE; + } break; default: return; } - - /* Mark this texobj as dirty (one bit per tex unit) - */ - t->dirty_state = TEX_ALL; -} - - -static void radeonBindTexture( GLcontext *ctx, GLenum target, - struct gl_texture_object *texObj ) -{ - if ( RADEON_DEBUG & (DEBUG_STATE|DEBUG_TEXTURE) ) { - fprintf( stderr, "%s( %p ) unit=%d\n", __FUNCTION__, (void *)texObj, - ctx->Texture.CurrentUnit ); - } - - assert( (target != GL_TEXTURE_1D && target != GL_TEXTURE_2D && - target != GL_TEXTURE_RECTANGLE_NV && target != GL_TEXTURE_CUBE_MAP) || - (texObj->DriverData != NULL) ); } - static void radeonDeleteTexture( GLcontext *ctx, struct gl_texture_object *texObj ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - driTextureObject * t = (driTextureObject *) texObj->DriverData; + r100ContextPtr rmesa = R100_CONTEXT(ctx); + radeonTexObj* t = radeon_tex_obj(texObj); + int i; - if ( RADEON_DEBUG & (DEBUG_STATE|DEBUG_TEXTURE) ) { - fprintf( stderr, "%s( %p (target = %s) )\n", __FUNCTION__, (void *)texObj, + radeon_print(RADEON_TEXTURE, RADEON_NORMAL, + "%s( %p (target = %s) )\n", __FUNCTION__, (void *)texObj, _mesa_lookup_enum_by_nr( texObj->Target ) ); - } - - if ( t != NULL ) { - if ( rmesa ) { - RADEON_FIREVERTICES( rmesa ); - } - driDestroyTextureObject( t ); + if ( rmesa ) { + radeon_firevertices(&rmesa->radeon); + for ( i = 0 ; i < rmesa->radeon.glCtx->Const.MaxTextureUnits ; i++ ) { + if ( t == rmesa->state.texture.unit[i].texobj ) { + rmesa->state.texture.unit[i].texobj = NULL; + rmesa->hw.tex[i].dirty = GL_FALSE; + rmesa->hw.cube[i].dirty = GL_FALSE; + } + } } + if (t->mt) { + radeon_miptree_unreference(t->mt); + t->mt = 0; + } /* Free mipmap images and the texture object itself */ _mesa_delete_texture_object(ctx, texObj); } @@ -837,7 +412,7 @@ static void radeonTexGen( GLcontext *ctx, GLenum pname, const GLfloat *params ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); GLuint unit = ctx->Texture.CurrentUnit; rmesa->recheck_texgen[unit] = GL_TRUE; } @@ -851,29 +426,40 @@ static void radeonTexGen( GLcontext *ctx, static struct gl_texture_object * radeonNewTextureObject( GLcontext *ctx, GLuint name, GLenum target ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - struct gl_texture_object *obj; - obj = _mesa_new_texture_object(ctx, name, target); - if (!obj) - return NULL; - obj->MaxAnisotropy = rmesa->initialMaxAnisotropy; - radeonAllocTexObj( obj ); - return obj; + r100ContextPtr rmesa = R100_CONTEXT(ctx); + radeonTexObj* t = CALLOC_STRUCT(radeon_tex_obj); + + _mesa_initialize_texture_object(&t->base, name, target); + t->base.MaxAnisotropy = rmesa->radeon.initialMaxAnisotropy; + + t->border_fallback = GL_FALSE; + + t->pp_txfilter = RADEON_BORDER_MODE_OGL; + t->pp_txformat = (RADEON_TXFORMAT_ENDIAN_NO_SWAP | + RADEON_TXFORMAT_PERSPECTIVE_ENABLE); + + radeonSetTexWrap( t, t->base.WrapS, t->base.WrapT ); + radeonSetTexMaxAnisotropy( t, t->base.MaxAnisotropy ); + radeonSetTexFilter( t, t->base.MinFilter, t->base.MagFilter ); + radeonSetTexBorderColor( t, t->base.BorderColor ); + return &t->base; } + void radeonInitTextureFuncs( struct dd_function_table *functions ) { - functions->ChooseTextureFormat = radeonChooseTextureFormat; + functions->ChooseTextureFormat = radeonChooseTextureFormat_mesa; functions->TexImage1D = radeonTexImage1D; functions->TexImage2D = radeonTexImage2D; functions->TexSubImage1D = radeonTexSubImage1D; functions->TexSubImage2D = radeonTexSubImage2D; + functions->GetTexImage = radeonGetTexImage; + functions->GetCompressedTexImage = radeonGetCompressedTexImage; functions->NewTextureObject = radeonNewTextureObject; - functions->BindTexture = radeonBindTexture; + // functions->BindTexture = radeonBindTexture; functions->DeleteTexture = radeonDeleteTexture; - functions->IsTextureResident = driIsTextureResident; functions->TexEnv = radeonTexEnv; functions->TexParameter = radeonTexParameter; @@ -882,5 +468,12 @@ void radeonInitTextureFuncs( struct dd_function_table *functions ) functions->CompressedTexImage2D = radeonCompressedTexImage2D; functions->CompressedTexSubImage2D = radeonCompressedTexSubImage2D; + functions->GenerateMipmap = radeonGenerateMipmap; + + functions->NewTextureImage = radeonNewTextureImage; + functions->FreeTexImageData = radeonFreeTexImageData; + functions->MapTexture = radeonMapTexture; + functions->UnmapTexture = radeonUnmapTexture; + driInitTextureFormats(); } diff --git a/src/mesa/drivers/dri/radeon/radeon_tex.h b/src/mesa/drivers/dri/radeon/radeon_tex.h index 80008808289..a4aaddc74fa 100644 --- a/src/mesa/drivers/dri/radeon/radeon_tex.h +++ b/src/mesa/drivers/dri/radeon/radeon_tex.h @@ -41,12 +41,16 @@ extern void radeonSetTexOffset(__DRIcontext *pDRICtx, GLint texname, unsigned long long offset, GLint depth, GLuint pitch); +extern void radeonSetTexBuffer(__DRIcontext *pDRICtx, GLint target, __DRIdrawable *dPriv); +extern void radeonSetTexBuffer2(__DRIcontext *pDRICtx, GLint target, GLint glx_texture_format, + __DRIdrawable *dPriv); + extern void radeonUpdateTextureState( GLcontext *ctx ); -extern int radeonUploadTexImages( radeonContextPtr rmesa, radeonTexObjPtr t, +extern int radeonUploadTexImages( r100ContextPtr rmesa, radeonTexObjPtr t, GLuint face ); -extern void radeonDestroyTexObj( radeonContextPtr rmesa, radeonTexObjPtr t ); +extern void radeonDestroyTexObj( r100ContextPtr rmesa, radeonTexObjPtr t ); extern void radeonInitTextureFuncs( struct dd_function_table *functions ); diff --git a/src/mesa/drivers/dri/radeon/radeon_texmem.c b/src/mesa/drivers/dri/radeon/radeon_texmem.c deleted file mode 100644 index 5f7bbe6a4cb..00000000000 --- a/src/mesa/drivers/dri/radeon/radeon_texmem.c +++ /dev/null @@ -1,404 +0,0 @@ -/************************************************************************** - -Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and - VA Linux Systems Inc., Fremont, California. - -All Rights Reserved. - -Permission is hereby granted, free of charge, to any person obtaining -a copy of this software and associated documentation files (the -"Software"), to deal in the Software without restriction, including -without limitation on the rights to use, copy, modify, merge, publish, -distribute, sub license, and/or sell copies of the Software, and to -permit persons to whom the Software is furnished to do so, subject to -the following conditions: - -The above copyright notice and this permission notice (including the -next paragraph) shall be included in all copies or substantial -portions of the Software. - -THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, -EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF -MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND -NON-INFRINGEMENT. IN NO EVENT SHALL ATI, VA LINUX SYSTEMS AND/OR THEIR -SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER -IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR -IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE -SOFTWARE. - -**************************************************************************/ - -/* - * Authors: - * Kevin E. Martin <martin@valinux.com> - * Gareth Hughes <gareth@valinux.com> - * - */ -#include <errno.h> - -#include "main/glheader.h" -#include "main/imports.h" -#include "main/context.h" -#include "main/macros.h" - -#include "radeon_context.h" -#include "radeon_ioctl.h" -#include "radeon_tex.h" - -#include <unistd.h> /* for usleep() */ - - -/** - * Destroy any device-dependent state associated with the texture. This may - * include NULLing out hardware state that points to the texture. - */ -void -radeonDestroyTexObj( radeonContextPtr rmesa, radeonTexObjPtr t ) -{ - if ( RADEON_DEBUG & DEBUG_TEXTURE ) { - fprintf( stderr, "%s( %p, %p )\n", __FUNCTION__, (void *)t, (void *)t->base.tObj ); - } - - if ( rmesa != NULL ) { - unsigned i; - - - for ( i = 0 ; i < rmesa->glCtx->Const.MaxTextureUnits ; i++ ) { - if ( t == rmesa->state.texture.unit[i].texobj ) { - rmesa->state.texture.unit[i].texobj = NULL; - } - } - } -} - - -/* ------------------------------------------------------------ - * Texture image conversions - */ - - -static void radeonUploadRectSubImage( radeonContextPtr rmesa, - radeonTexObjPtr t, - struct gl_texture_image *texImage, - GLint x, GLint y, - GLint width, GLint height ) -{ - const struct gl_texture_format *texFormat = texImage->TexFormat; - int blit_format, dstPitch, done; - - switch ( texFormat->TexelBytes ) { - case 1: - blit_format = RADEON_GMC_DST_8BPP_CI; - break; - case 2: - blit_format = RADEON_GMC_DST_16BPP; - break; - case 4: - blit_format = RADEON_GMC_DST_32BPP; - break; - default: - fprintf( stderr, "radeonUploadRectSubImage: unknown blit_format (texelbytes=%d)\n", - texFormat->TexelBytes); - return; - } - - t->image[0][0].data = texImage->Data; - - /* Currently don't need to cope with small pitches. - */ - width = texImage->Width; - height = texImage->Height; - dstPitch = t->pp_txpitch + 32; - - { /* FIXME: prefer GART-texturing if possible */ - /* Data not in GART memory, or bad pitch. - */ - for (done = 0; done < height ; ) { - struct radeon_dma_region region; - int lines = MIN2( height - done, RADEON_BUFFER_SIZE / dstPitch ); - int src_pitch; - char *tex; - - src_pitch = texImage->RowStride * texFormat->TexelBytes; - - tex = (char *)texImage->Data + done * src_pitch; - - memset(®ion, 0, sizeof(region)); - radeonAllocDmaRegion( rmesa, ®ion, lines * dstPitch, 1024 ); - - /* Copy texdata to dma: - */ - if (0) - fprintf(stderr, "%s: src_pitch %d dst_pitch %d\n", - __FUNCTION__, src_pitch, dstPitch); - - if (src_pitch == dstPitch) { - memcpy( region.address + region.start, tex, lines * src_pitch ); - } - else { - char *buf = region.address + region.start; - int i; - for (i = 0 ; i < lines ; i++) { - memcpy( buf, tex, src_pitch ); - buf += dstPitch; - tex += src_pitch; - } - } - - radeonEmitWait( rmesa, RADEON_WAIT_3D ); - - - - /* Blit to framebuffer - */ - radeonEmitBlit( rmesa, - blit_format, - dstPitch, GET_START( ®ion ), - dstPitch, t->bufAddr, - 0, 0, - 0, done, - width, lines ); - - radeonEmitWait( rmesa, RADEON_WAIT_2D ); - - radeonReleaseDmaRegion( rmesa, ®ion, __FUNCTION__ ); - done += lines; - } - } -} - - -/** - * Upload the texture image associated with texture \a t at the specified - * level at the address relative to \a start. - */ -static void uploadSubImage( radeonContextPtr rmesa, radeonTexObjPtr t, - GLint hwlevel, - GLint x, GLint y, GLint width, GLint height, - GLuint face ) -{ - struct gl_texture_image *texImage = NULL; - GLuint offset; - GLint imageWidth, imageHeight; - GLint ret; - drm_radeon_texture_t tex; - drm_radeon_tex_image_t tmp; - const int level = hwlevel + t->base.firstLevel; - - if ( RADEON_DEBUG & DEBUG_TEXTURE ) { - fprintf( stderr, "%s( %p, %p ) level/width/height/face = %d/%d/%d/%u\n", - __FUNCTION__, (void *)t, (void *)t->base.tObj, level, width, height, face ); - } - - ASSERT(face < 6); - - /* Ensure we have a valid texture to upload */ - if ( ( hwlevel < 0 ) || ( hwlevel >= RADEON_MAX_TEXTURE_LEVELS ) ) { - _mesa_problem(NULL, "bad texture level in %s", __FUNCTION__); - return; - } - - texImage = t->base.tObj->Image[face][level]; - - if ( !texImage ) { - if ( RADEON_DEBUG & DEBUG_TEXTURE ) - fprintf( stderr, "%s: texImage %d is NULL!\n", __FUNCTION__, level ); - return; - } - if ( !texImage->Data ) { - if ( RADEON_DEBUG & DEBUG_TEXTURE ) - fprintf( stderr, "%s: image data is NULL!\n", __FUNCTION__ ); - return; - } - - - if (t->base.tObj->Target == GL_TEXTURE_RECTANGLE_NV) { - assert(level == 0); - assert(hwlevel == 0); - if ( RADEON_DEBUG & DEBUG_TEXTURE ) - fprintf( stderr, "%s: image data is rectangular\n", __FUNCTION__); - radeonUploadRectSubImage( rmesa, t, texImage, x, y, width, height ); - return; - } - - imageWidth = texImage->Width; - imageHeight = texImage->Height; - - offset = t->bufAddr + t->base.totalSize * face / 6; - - if ( RADEON_DEBUG & (DEBUG_TEXTURE|DEBUG_IOCTL) ) { - GLint imageX = 0; - GLint imageY = 0; - GLint blitX = t->image[face][hwlevel].x; - GLint blitY = t->image[face][hwlevel].y; - GLint blitWidth = t->image[face][hwlevel].width; - GLint blitHeight = t->image[face][hwlevel].height; - fprintf( stderr, " upload image: %d,%d at %d,%d\n", - imageWidth, imageHeight, imageX, imageY ); - fprintf( stderr, " upload blit: %d,%d at %d,%d\n", - blitWidth, blitHeight, blitX, blitY ); - fprintf( stderr, " blit ofs: 0x%07x level: %d/%d\n", - (GLuint)offset, hwlevel, level ); - } - - t->image[face][hwlevel].data = texImage->Data; - - /* Init the DRM_RADEON_TEXTURE command / drm_radeon_texture_t struct. - * NOTE: we're always use a 1KB-wide blit and I8 texture format. - * We used to use 1, 2 and 4-byte texels and used to use the texture - * width to dictate the blit width - but that won't work for compressed - * textures. (Brian) - * NOTE: can't do that with texture tiling. (sroland) - */ - tex.offset = offset; - tex.image = &tmp; - /* copy (x,y,width,height,data) */ - memcpy( &tmp, &t->image[face][hwlevel], sizeof(drm_radeon_tex_image_t) ); - - if (texImage->TexFormat->TexelBytes) { - /* use multi-byte upload scheme */ - tex.height = imageHeight; - tex.width = imageWidth; - tex.format = t->pp_txformat & RADEON_TXFORMAT_FORMAT_MASK; - tex.pitch = MAX2((texImage->Width * texImage->TexFormat->TexelBytes) / 64, 1); - tex.offset += tmp.x & ~1023; - tmp.x = tmp.x % 1024; - if (t->tile_bits & RADEON_TXO_MICRO_TILE_X2) { - /* need something like "tiled coordinates" ? */ - tmp.y = tmp.x / (tex.pitch * 128) * 2; - tmp.x = tmp.x % (tex.pitch * 128) / 2 / texImage->TexFormat->TexelBytes; - tex.pitch |= RADEON_DST_TILE_MICRO >> 22; - } - else { - tmp.x = tmp.x >> (texImage->TexFormat->TexelBytes >> 1); - } - if ((t->tile_bits & RADEON_TXO_MACRO_TILE) && - (texImage->Width * texImage->TexFormat->TexelBytes >= 256)) { - /* radeon switches off macro tiling for small textures/mipmaps it seems */ - tex.pitch |= RADEON_DST_TILE_MACRO >> 22; - } - } - else { - /* In case of for instance 8x8 texture (2x2 dxt blocks), padding after the first two blocks is - needed (only with dxt1 since 2 dxt3/dxt5 blocks already use 32 Byte). */ - /* set tex.height to 1/4 since 1 "macropixel" (dxt-block) has 4 real pixels. Needed - so the kernel module reads the right amount of data. */ - tex.format = RADEON_TXFORMAT_I8; /* any 1-byte texel format */ - tex.pitch = (BLIT_WIDTH_BYTES / 64); - tex.height = (imageHeight + 3) / 4; - tex.width = (imageWidth + 3) / 4; - switch (t->pp_txformat & RADEON_TXFORMAT_FORMAT_MASK) { - case RADEON_TXFORMAT_DXT1: - tex.width *= 8; - break; - case RADEON_TXFORMAT_DXT23: - case RADEON_TXFORMAT_DXT45: - tex.width *= 16; - break; - } - } - - LOCK_HARDWARE( rmesa ); - do { - ret = drmCommandWriteRead( rmesa->dri.fd, DRM_RADEON_TEXTURE, - &tex, sizeof(drm_radeon_texture_t) ); - } while ( ret == -EAGAIN ); - - UNLOCK_HARDWARE( rmesa ); - - if ( ret ) { - fprintf( stderr, "DRM_RADEON_TEXTURE: return = %d\n", ret ); - fprintf( stderr, " offset=0x%08x\n", - offset ); - fprintf( stderr, " image width=%d height=%d\n", - imageWidth, imageHeight ); - fprintf( stderr, " blit width=%d height=%d data=%p\n", - t->image[face][hwlevel].width, t->image[face][hwlevel].height, - t->image[face][hwlevel].data ); - exit( 1 ); - } -} - - -/** - * Upload the texture images associated with texture \a t. This might - * require the allocation of texture memory. - * - * \param rmesa Context pointer - * \param t Texture to be uploaded - * \param face Cube map face to be uploaded. Zero for non-cube maps. - */ - -int radeonUploadTexImages( radeonContextPtr rmesa, radeonTexObjPtr t, GLuint face ) -{ - int numLevels; - - if ( !t || t->base.totalSize == 0 || t->image_override ) - return 0; - - if ( RADEON_DEBUG & (DEBUG_TEXTURE|DEBUG_IOCTL) ) { - fprintf( stderr, "%s( %p, %p ) sz=%d lvls=%d-%d\n", __FUNCTION__, - (void *)rmesa->glCtx, (void *)t->base.tObj, t->base.totalSize, - t->base.firstLevel, t->base.lastLevel ); - } - - numLevels = t->base.lastLevel - t->base.firstLevel + 1; - - if (RADEON_DEBUG & DEBUG_SYNC) { - fprintf(stderr, "%s: Syncing\n", __FUNCTION__ ); - radeonFinish( rmesa->glCtx ); - } - - LOCK_HARDWARE( rmesa ); - - if ( t->base.memBlock == NULL ) { - int heap; - - heap = driAllocateTexture( rmesa->texture_heaps, rmesa->nr_heaps, - (driTextureObject *) t ); - if ( heap == -1 ) { - UNLOCK_HARDWARE( rmesa ); - return -1; - } - - /* Set the base offset of the texture image */ - t->bufAddr = rmesa->radeonScreen->texOffset[heap] - + t->base.memBlock->ofs; - t->pp_txoffset = t->bufAddr; - - if (!(t->base.tObj->Image[0][0]->IsClientData)) { - /* hope it's safe to add that here... */ - t->pp_txoffset |= t->tile_bits; - } - - /* Mark this texobj as dirty on all units: - */ - t->dirty_state = TEX_ALL; - } - - - /* Let the world know we've used this memory recently. - */ - driUpdateTextureLRU( (driTextureObject *) t ); - UNLOCK_HARDWARE( rmesa ); - - - /* Upload any images that are new */ - if (t->base.dirty_images[face]) { - int i; - for ( i = 0 ; i < numLevels ; i++ ) { - if ( (t->base.dirty_images[face] & (1 << (i+t->base.firstLevel))) != 0 ) { - uploadSubImage( rmesa, t, i, 0, 0, t->image[face][i].width, - t->image[face][i].height, face ); - } - } - t->base.dirty_images[face] = 0; - } - - if (RADEON_DEBUG & DEBUG_SYNC) { - fprintf(stderr, "%s: Syncing\n", __FUNCTION__ ); - radeonFinish( rmesa->glCtx ); - } - - return 0; -} diff --git a/src/mesa/drivers/dri/radeon/radeon_texstate.c b/src/mesa/drivers/dri/radeon/radeon_texstate.c index b165205c093..9d252aa74c5 100644 --- a/src/mesa/drivers/dri/radeon/radeon_texstate.c +++ b/src/mesa/drivers/dri/radeon/radeon_texstate.c @@ -39,10 +39,12 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "main/context.h" #include "main/macros.h" #include "main/texformat.h" +#include "main/teximage.h" #include "main/texobj.h" #include "main/enums.h" #include "radeon_context.h" +#include "radeon_mipmap_tree.h" #include "radeon_state.h" #include "radeon_ioctl.h" #include "radeon_swtcl.h" @@ -75,10 +77,11 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #define VALID_FORMAT(f) ( ((f) <= MESA_FORMAT_RGBA_DXT5) \ && (tx_table[f].format != 0xffffffff) ) -static const struct { +struct tx_table { GLuint format, filter; -} -tx_table[] = +}; + +static const struct tx_table tx_table[] = { _ALPHA(RGBA8888), _ALPHA_REV(RGBA8888), @@ -111,252 +114,6 @@ tx_table[] = #undef _ALPHA #undef _INVALID -/** - * This function computes the number of bytes of storage needed for - * the given texture object (all mipmap levels, all cube faces). - * The \c image[face][level].x/y/width/height parameters for upload/blitting - * are computed here. \c pp_txfilter, \c pp_txformat, etc. will be set here - * too. - * - * \param rmesa Context pointer - * \param tObj GL texture object whose images are to be posted to - * hardware state. - */ -static void radeonSetTexImages( radeonContextPtr rmesa, - struct gl_texture_object *tObj ) -{ - radeonTexObjPtr t = (radeonTexObjPtr)tObj->DriverData; - const struct gl_texture_image *baseImage = tObj->Image[0][tObj->BaseLevel]; - GLint curOffset, blitWidth; - GLint i, texelBytes; - GLint numLevels; - GLint log2Width, log2Height, log2Depth; - - /* Set the hardware texture format - */ - if ( !t->image_override ) { - t->pp_txformat &= ~(RADEON_TXFORMAT_FORMAT_MASK | - RADEON_TXFORMAT_ALPHA_IN_MAP); - t->pp_txfilter &= ~RADEON_YUV_TO_RGB; - - if ( VALID_FORMAT( baseImage->TexFormat->MesaFormat ) ) { - t->pp_txformat |= tx_table[ baseImage->TexFormat->MesaFormat ].format; - t->pp_txfilter |= tx_table[ baseImage->TexFormat->MesaFormat ].filter; - } - else { - _mesa_problem(NULL, "unexpected texture format in %s", __FUNCTION__); - return; - } - } - - texelBytes = baseImage->TexFormat->TexelBytes; - - /* Compute which mipmap levels we really want to send to the hardware. - */ - - if (tObj->Target != GL_TEXTURE_CUBE_MAP) - driCalculateTextureFirstLastLevel( (driTextureObject *) t ); - else { - /* r100 can't handle mipmaps for cube/3d textures, so don't waste - memory for them */ - t->base.firstLevel = t->base.lastLevel = tObj->BaseLevel; - } - log2Width = tObj->Image[0][t->base.firstLevel]->WidthLog2; - log2Height = tObj->Image[0][t->base.firstLevel]->HeightLog2; - log2Depth = tObj->Image[0][t->base.firstLevel]->DepthLog2; - - numLevels = t->base.lastLevel - t->base.firstLevel + 1; - - assert(numLevels <= RADEON_MAX_TEXTURE_LEVELS); - - /* Calculate mipmap offsets and dimensions for blitting (uploading) - * The idea is that we lay out the mipmap levels within a block of - * memory organized as a rectangle of width BLIT_WIDTH_BYTES. - */ - curOffset = 0; - blitWidth = BLIT_WIDTH_BYTES; - t->tile_bits = 0; - - /* figure out if this texture is suitable for tiling. */ - if (texelBytes && (tObj->Target != GL_TEXTURE_RECTANGLE_NV)) { - if (rmesa->texmicrotile && (baseImage->Height > 1)) { - /* allow 32 (bytes) x 1 mip (which will use two times the space - the non-tiled version would use) max if base texture is large enough */ - if ((numLevels == 1) || - (((baseImage->Width * texelBytes / baseImage->Height) <= 32) && - (baseImage->Width * texelBytes > 64)) || - ((baseImage->Width * texelBytes / baseImage->Height) <= 16)) { - /* R100 has two microtile bits (only the txoffset reg, not the blitter) - weird: X2 + OPT: 32bit correct, 16bit completely hosed - X2: 32bit correct, 16bit correct - OPT: 32bit large mips correct, small mips hosed, 16bit completely hosed */ - t->tile_bits |= RADEON_TXO_MICRO_TILE_X2 /*| RADEON_TXO_MICRO_TILE_OPT*/; - } - } - if ((baseImage->Width * texelBytes >= 256) && (baseImage->Height >= 16)) { - /* R100 disables macro tiling only if mip width is smaller than 256 bytes, and not - in the case if height is smaller than 16 (not 100% sure), as does the r200, - so need to disable macro tiling in that case */ - if ((numLevels == 1) || ((baseImage->Width * texelBytes / baseImage->Height) <= 4)) { - t->tile_bits |= RADEON_TXO_MACRO_TILE; - } - } - } - - for (i = 0; i < numLevels; i++) { - const struct gl_texture_image *texImage; - GLuint size; - - texImage = tObj->Image[0][i + t->base.firstLevel]; - if ( !texImage ) - break; - - /* find image size in bytes */ - if (texImage->IsCompressed) { - /* need to calculate the size AFTER padding even though the texture is - submitted without padding. - Only handle pot textures currently - don't know if npot is even possible, - size calculation would certainly need (trivial) adjustments. - Align (and later pad) to 32byte, not sure what that 64byte blit width is - good for? */ - if ((t->pp_txformat & RADEON_TXFORMAT_FORMAT_MASK) == RADEON_TXFORMAT_DXT1) { - /* RGB_DXT1/RGBA_DXT1, 8 bytes per block */ - if ((texImage->Width + 3) < 8) /* width one block */ - size = texImage->CompressedSize * 4; - else if ((texImage->Width + 3) < 16) - size = texImage->CompressedSize * 2; - else size = texImage->CompressedSize; - } - else /* DXT3/5, 16 bytes per block */ - if ((texImage->Width + 3) < 8) - size = texImage->CompressedSize * 2; - else size = texImage->CompressedSize; - } - else if (tObj->Target == GL_TEXTURE_RECTANGLE_NV) { - size = ((texImage->Width * texelBytes + 63) & ~63) * texImage->Height; - } - else if (t->tile_bits & RADEON_TXO_MICRO_TILE_X2) { - /* tile pattern is 16 bytes x2. mipmaps stay 32 byte aligned, - though the actual offset may be different (if texture is less than - 32 bytes width) to the untiled case */ - int w = (texImage->Width * texelBytes * 2 + 31) & ~31; - size = (w * ((texImage->Height + 1) / 2)) * texImage->Depth; - blitWidth = MAX2(texImage->Width, 64 / texelBytes); - } - else { - int w = (texImage->Width * texelBytes + 31) & ~31; - size = w * texImage->Height * texImage->Depth; - blitWidth = MAX2(texImage->Width, 64 / texelBytes); - } - assert(size > 0); - - /* Align to 32-byte offset. It is faster to do this unconditionally - * (no branch penalty). - */ - - curOffset = (curOffset + 0x1f) & ~0x1f; - - if (texelBytes) { - t->image[0][i].x = curOffset; /* fix x and y coords up later together with offset */ - t->image[0][i].y = 0; - t->image[0][i].width = MIN2(size / texelBytes, blitWidth); - t->image[0][i].height = (size / texelBytes) / t->image[0][i].width; - } - else { - t->image[0][i].x = curOffset % BLIT_WIDTH_BYTES; - t->image[0][i].y = curOffset / BLIT_WIDTH_BYTES; - t->image[0][i].width = MIN2(size, BLIT_WIDTH_BYTES); - t->image[0][i].height = size / t->image[0][i].width; - } - -#if 0 - /* for debugging only and only applicable to non-rectangle targets */ - assert(size % t->image[0][i].width == 0); - assert(t->image[0][i].x == 0 - || (size < BLIT_WIDTH_BYTES && t->image[0][i].height == 1)); -#endif - - if (0) - fprintf(stderr, - "level %d: %dx%d x=%d y=%d w=%d h=%d size=%d at %d\n", - i, texImage->Width, texImage->Height, - t->image[0][i].x, t->image[0][i].y, - t->image[0][i].width, t->image[0][i].height, size, curOffset); - - curOffset += size; - - } - - /* Align the total size of texture memory block. - */ - t->base.totalSize = (curOffset + RADEON_OFFSET_MASK) & ~RADEON_OFFSET_MASK; - - /* Setup remaining cube face blits, if needed */ - if (tObj->Target == GL_TEXTURE_CUBE_MAP) { - const GLuint faceSize = t->base.totalSize; - GLuint face; - /* reuse face 0 x/y/width/height - just update the offset when uploading */ - for (face = 1; face < 6; face++) { - for (i = 0; i < numLevels; i++) { - t->image[face][i].x = t->image[0][i].x; - t->image[face][i].y = t->image[0][i].y; - t->image[face][i].width = t->image[0][i].width; - t->image[face][i].height = t->image[0][i].height; - } - } - t->base.totalSize = 6 * faceSize; /* total texmem needed */ - } - - /* Hardware state: - */ - t->pp_txfilter &= ~RADEON_MAX_MIP_LEVEL_MASK; - t->pp_txfilter |= (numLevels - 1) << RADEON_MAX_MIP_LEVEL_SHIFT; - - t->pp_txformat &= ~(RADEON_TXFORMAT_WIDTH_MASK | - RADEON_TXFORMAT_HEIGHT_MASK | - RADEON_TXFORMAT_CUBIC_MAP_ENABLE | - RADEON_TXFORMAT_F5_WIDTH_MASK | - RADEON_TXFORMAT_F5_HEIGHT_MASK); - t->pp_txformat |= ((log2Width << RADEON_TXFORMAT_WIDTH_SHIFT) | - (log2Height << RADEON_TXFORMAT_HEIGHT_SHIFT)); - - if (tObj->Target == GL_TEXTURE_CUBE_MAP) { - assert(log2Width == log2Height); - t->pp_txformat |= ((log2Width << RADEON_TXFORMAT_F5_WIDTH_SHIFT) | - (log2Height << RADEON_TXFORMAT_F5_HEIGHT_SHIFT) | - (RADEON_TXFORMAT_CUBIC_MAP_ENABLE)); - t->pp_cubic_faces = ((log2Width << RADEON_FACE_WIDTH_1_SHIFT) | - (log2Height << RADEON_FACE_HEIGHT_1_SHIFT) | - (log2Width << RADEON_FACE_WIDTH_2_SHIFT) | - (log2Height << RADEON_FACE_HEIGHT_2_SHIFT) | - (log2Width << RADEON_FACE_WIDTH_3_SHIFT) | - (log2Height << RADEON_FACE_HEIGHT_3_SHIFT) | - (log2Width << RADEON_FACE_WIDTH_4_SHIFT) | - (log2Height << RADEON_FACE_HEIGHT_4_SHIFT)); - } - - t->pp_txsize = (((tObj->Image[0][t->base.firstLevel]->Width - 1) << 0) | - ((tObj->Image[0][t->base.firstLevel]->Height - 1) << 16)); - - /* Only need to round to nearest 32 for textures, but the blitter - * requires 64-byte aligned pitches, and we may/may not need the - * blitter. NPOT only! - */ - if ( !t->image_override ) { - if (baseImage->IsCompressed) - t->pp_txpitch = (tObj->Image[0][t->base.firstLevel]->Width + 63) & ~(63); - else - t->pp_txpitch = ((tObj->Image[0][t->base.firstLevel]->Width * texelBytes) + 63) & ~(63); - t->pp_txpitch -= 32; - } - - t->dirty_state = TEX_ALL; - - /* FYI: radeonUploadTexImages( rmesa, t ); used to be called here */ -} - - - /* ================================================================ * Texture combine functions */ @@ -503,7 +260,7 @@ do { \ static GLboolean radeonUpdateTextureEnv( GLcontext *ctx, int unit ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); const struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit]; GLuint color_combine, alpha_combine; const GLuint color_combine0 = RADEON_COLOR_ARG_A_ZERO | RADEON_COLOR_ARG_B_ZERO @@ -520,7 +277,7 @@ static GLboolean radeonUpdateTextureEnv( GLcontext *ctx, int unit ) assert( (texUnit->_ReallyEnabled == 0) || (texUnit->_Current != NULL) ); - if ( RADEON_DEBUG & DEBUG_TEXTURE ) { + if ( RADEON_DEBUG & RADEON_TEXTURE ) { fprintf( stderr, "%s( %p, %d )\n", __FUNCTION__, (void *)ctx, unit ); } @@ -846,22 +603,21 @@ static GLboolean radeonUpdateTextureEnv( GLcontext *ctx, int unit ) void radeonSetTexOffset(__DRIcontext * pDRICtx, GLint texname, unsigned long long offset, GLint depth, GLuint pitch) { - radeonContextPtr rmesa = pDRICtx->driverPrivate; + r100ContextPtr rmesa = pDRICtx->driverPrivate; struct gl_texture_object *tObj = - _mesa_lookup_texture(rmesa->glCtx, texname); - radeonTexObjPtr t; + _mesa_lookup_texture(rmesa->radeon.glCtx, texname); + radeonTexObjPtr t = radeon_tex_obj(tObj); if (tObj == NULL) return; - t = (radeonTexObjPtr) tObj->DriverData; - t->image_override = GL_TRUE; if (!offset) return; - - t->pp_txoffset = offset; + + t->bo = NULL; + t->override_offset = offset; t->pp_txpitch = pitch - 32; switch (depth) { @@ -881,6 +637,125 @@ void radeonSetTexOffset(__DRIcontext * pDRICtx, GLint texname, } } +void radeonSetTexBuffer2(__DRIcontext *pDRICtx, GLint target, GLint glx_texture_format, + __DRIdrawable *dPriv) +{ + struct gl_texture_unit *texUnit; + struct gl_texture_object *texObj; + struct gl_texture_image *texImage; + struct radeon_renderbuffer *rb; + radeon_texture_image *rImage; + radeonContextPtr radeon; + r100ContextPtr rmesa; + struct radeon_framebuffer *rfb; + radeonTexObjPtr t; + uint32_t pitch_val; + uint32_t internalFormat, type, format; + + type = GL_BGRA; + format = GL_UNSIGNED_BYTE; + internalFormat = (glx_texture_format == GLX_TEXTURE_FORMAT_RGB_EXT ? 3 : 4); + + radeon = pDRICtx->driverPrivate; + rmesa = pDRICtx->driverPrivate; + + rfb = dPriv->driverPrivate; + texUnit = &radeon->glCtx->Texture.Unit[radeon->glCtx->Texture.CurrentUnit]; + texObj = _mesa_select_tex_object(radeon->glCtx, texUnit, target); + texImage = _mesa_get_tex_image(radeon->glCtx, texObj, target, 0); + + rImage = get_radeon_texture_image(texImage); + t = radeon_tex_obj(texObj); + if (t == NULL) { + return; + } + + radeon_update_renderbuffers(pDRICtx, dPriv); + /* back & depth buffer are useless free them right away */ + rb = (void*)rfb->base.Attachment[BUFFER_DEPTH].Renderbuffer; + if (rb && rb->bo) { + radeon_bo_unref(rb->bo); + rb->bo = NULL; + } + rb = (void*)rfb->base.Attachment[BUFFER_BACK_LEFT].Renderbuffer; + if (rb && rb->bo) { + radeon_bo_unref(rb->bo); + rb->bo = NULL; + } + rb = rfb->color_rb[0]; + if (rb->bo == NULL) { + /* Failed to BO for the buffer */ + return; + } + + _mesa_lock_texture(radeon->glCtx, texObj); + if (t->bo) { + radeon_bo_unref(t->bo); + t->bo = NULL; + } + if (rImage->bo) { + radeon_bo_unref(rImage->bo); + rImage->bo = NULL; + } + if (t->mt) { + radeon_miptree_unreference(t->mt); + t->mt = NULL; + } + if (rImage->mt) { + radeon_miptree_unreference(rImage->mt); + rImage->mt = NULL; + } + _mesa_init_teximage_fields(radeon->glCtx, target, texImage, + rb->base.Width, rb->base.Height, 1, 0, rb->cpp); + texImage->RowStride = rb->pitch / rb->cpp; + texImage->TexFormat = radeonChooseTextureFormat(radeon->glCtx, + internalFormat, + type, format, 0); + rImage->bo = rb->bo; + radeon_bo_ref(rImage->bo); + t->bo = rb->bo; + radeon_bo_ref(t->bo); + t->tile_bits = 0; + t->image_override = GL_TRUE; + t->override_offset = 0; + t->pp_txpitch &= (1 << 13) -1; + pitch_val = rb->pitch; + switch (rb->cpp) { + case 4: + if (glx_texture_format == GLX_TEXTURE_FORMAT_RGB_EXT) + t->pp_txformat = tx_table[MESA_FORMAT_RGB888].format; + else + t->pp_txformat = tx_table[MESA_FORMAT_ARGB8888].format; + t->pp_txfilter |= tx_table[MESA_FORMAT_ARGB8888].filter; + break; + case 3: + default: + t->pp_txformat = tx_table[MESA_FORMAT_RGB888].format; + t->pp_txfilter |= tx_table[MESA_FORMAT_RGB888].filter; + break; + case 2: + t->pp_txformat = tx_table[MESA_FORMAT_RGB565].format; + t->pp_txfilter |= tx_table[MESA_FORMAT_RGB565].filter; + break; + } + t->pp_txsize = ((rb->base.Width - 1) << RADEON_TEX_USIZE_SHIFT) + | ((rb->base.Height - 1) << RADEON_TEX_VSIZE_SHIFT); + t->pp_txformat |= RADEON_TXFORMAT_NON_POWER2; + t->pp_txpitch = pitch_val; + t->pp_txpitch -= 32; + + t->validated = GL_TRUE; + _mesa_unlock_texture(radeon->glCtx, texObj); + return; +} + + +void radeonSetTexBuffer(__DRIcontext *pDRICtx, GLint target, __DRIdrawable *dPriv) +{ + radeonSetTexBuffer2(pDRICtx, target, GLX_TEXTURE_FORMAT_RGBA_EXT, dPriv); +} + + #define TEXOBJ_TXFILTER_MASK (RADEON_MAX_MIP_LEVEL_MASK | \ RADEON_MIN_FILTER_MASK | \ RADEON_MAG_FILTER_MASK | \ @@ -901,12 +776,53 @@ void radeonSetTexOffset(__DRIcontext * pDRICtx, GLint texname, RADEON_TXFORMAT_NON_POWER2) -static void import_tex_obj_state( radeonContextPtr rmesa, +static void disable_tex_obj_state( r100ContextPtr rmesa, + int unit ) +{ + RADEON_STATECHANGE( rmesa, tex[unit] ); + + RADEON_STATECHANGE( rmesa, tcl ); + rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] &= ~(RADEON_ST_BIT(unit) | + RADEON_Q_BIT(unit)); + + if (rmesa->radeon.TclFallback & (RADEON_TCL_FALLBACK_TEXGEN_0<<unit)) { + TCL_FALLBACK( rmesa->radeon.glCtx, (RADEON_TCL_FALLBACK_TEXGEN_0<<unit), GL_FALSE); + rmesa->recheck_texgen[unit] = GL_TRUE; + } + + if (rmesa->hw.tex[unit].cmd[TEX_PP_TXFORMAT] & RADEON_TXFORMAT_CUBIC_MAP_ENABLE) { + /* this seems to be a genuine (r100 only?) hw bug. Need to remove the + cubic_map bit on unit 2 when the unit is disabled, otherwise every + 2nd (2d) mipmap on unit 0 will be broken (may not be needed for other + units, better be safe than sorry though).*/ + RADEON_STATECHANGE( rmesa, tex[unit] ); + rmesa->hw.tex[unit].cmd[TEX_PP_TXFORMAT] &= ~RADEON_TXFORMAT_CUBIC_MAP_ENABLE; + } + + { + GLuint inputshift = RADEON_TEXGEN_0_INPUT_SHIFT + unit*4; + GLuint tmp = rmesa->TexGenEnabled; + + rmesa->TexGenEnabled &= ~(RADEON_TEXGEN_TEXMAT_0_ENABLE<<unit); + rmesa->TexGenEnabled &= ~(RADEON_TEXMAT_0_ENABLE<<unit); + rmesa->TexGenEnabled &= ~(RADEON_TEXGEN_INPUT_MASK<<inputshift); + rmesa->TexGenNeedNormals[unit] = 0; + rmesa->TexGenEnabled |= + (RADEON_TEXGEN_INPUT_TEXCOORD_0+unit) << inputshift; + + if (tmp != rmesa->TexGenEnabled) { + rmesa->recheck_texgen[unit] = GL_TRUE; + rmesa->radeon.NewGLState |= _NEW_TEXTURE_MATRIX; + } + } +} + +static void import_tex_obj_state( r100ContextPtr rmesa, int unit, radeonTexObjPtr texobj ) { /* do not use RADEON_DB_STATE to avoid stale texture caches */ - int *cmd = &rmesa->hw.tex[unit].cmd[TEX_CMD_0]; + uint32_t *cmd = &rmesa->hw.tex[unit].cmd[TEX_CMD_0]; GLuint se_coord_fmt = rmesa->hw.set.cmd[SET_SE_COORDFMT]; RADEON_STATECHANGE( rmesa, tex[unit] ); @@ -915,10 +831,9 @@ static void import_tex_obj_state( radeonContextPtr rmesa, cmd[TEX_PP_TXFILTER] |= texobj->pp_txfilter & TEXOBJ_TXFILTER_MASK; cmd[TEX_PP_TXFORMAT] &= ~TEXOBJ_TXFORMAT_MASK; cmd[TEX_PP_TXFORMAT] |= texobj->pp_txformat & TEXOBJ_TXFORMAT_MASK; - cmd[TEX_PP_TXOFFSET] = texobj->pp_txoffset; cmd[TEX_PP_BORDER_COLOR] = texobj->pp_border_color; - if (texobj->base.tObj->Target == GL_TEXTURE_RECTANGLE_NV) { + if (texobj->base.Target == GL_TEXTURE_RECTANGLE_NV) { GLuint *txr_cmd = RADEON_DB_STATE( txr[unit] ); txr_cmd[TXR_PP_TEX_SIZE] = texobj->pp_txsize; /* NPOT only! */ txr_cmd[TXR_PP_TEX_PITCH] = texobj->pp_txpitch; /* NPOT only! */ @@ -928,22 +843,12 @@ static void import_tex_obj_state( radeonContextPtr rmesa, else { se_coord_fmt &= ~(RADEON_VTX_ST0_NONPARAMETRIC << unit); - if (texobj->base.tObj->Target == GL_TEXTURE_CUBE_MAP) { - int *cube_cmd = &rmesa->hw.cube[unit].cmd[CUBE_CMD_0]; - GLuint bytesPerFace = texobj->base.totalSize / 6; - ASSERT(texobj->base.totalSize % 6 == 0); + if (texobj->base.Target == GL_TEXTURE_CUBE_MAP) { + uint32_t *cube_cmd = &rmesa->hw.cube[unit].cmd[CUBE_CMD_0]; RADEON_STATECHANGE( rmesa, cube[unit] ); cube_cmd[CUBE_PP_CUBIC_FACES] = texobj->pp_cubic_faces; - /* dont know if this setup conforms to OpenGL.. - * at least it matches the behavior of mesa software renderer - */ - cube_cmd[CUBE_PP_CUBIC_OFFSET_0] = texobj->pp_txoffset; /* right */ - cube_cmd[CUBE_PP_CUBIC_OFFSET_1] = texobj->pp_txoffset + 1 * bytesPerFace; /* left */ - cube_cmd[CUBE_PP_CUBIC_OFFSET_2] = texobj->pp_txoffset + 2 * bytesPerFace; /* top */ - cube_cmd[CUBE_PP_CUBIC_OFFSET_3] = texobj->pp_txoffset + 3 * bytesPerFace; /* bottom */ - cube_cmd[CUBE_PP_CUBIC_OFFSET_4] = texobj->pp_txoffset + 4 * bytesPerFace; /* front */ - cmd[TEX_PP_TXOFFSET] = texobj->pp_txoffset + 5 * bytesPerFace; /* back */ + /* state filled out in the cube_emit */ } } @@ -952,13 +857,11 @@ static void import_tex_obj_state( radeonContextPtr rmesa, rmesa->hw.set.cmd[SET_SE_COORDFMT] = se_coord_fmt; } - texobj->dirty_state &= ~(1<<unit); + rmesa->radeon.NewGLState |= _NEW_TEXTURE_MATRIX; } - - -static void set_texgen_matrix( radeonContextPtr rmesa, +static void set_texgen_matrix( r100ContextPtr rmesa, GLuint unit, const GLfloat *s_plane, const GLfloat *t_plane, @@ -986,14 +889,14 @@ static void set_texgen_matrix( radeonContextPtr rmesa, rmesa->TexGenMatrix[unit].m[15] = q_plane[3]; rmesa->TexGenEnabled |= RADEON_TEXMAT_0_ENABLE << unit; - rmesa->NewGLState |= _NEW_TEXTURE_MATRIX; + rmesa->radeon.NewGLState |= _NEW_TEXTURE_MATRIX; } /* Returns GL_FALSE if fallback required. */ static GLboolean radeon_validate_texgen( GLcontext *ctx, GLuint unit ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit]; GLuint inputshift = RADEON_TEXGEN_0_INPUT_SHIFT + unit*4; GLuint tmp = rmesa->TexGenEnabled; @@ -1030,7 +933,7 @@ static GLboolean radeon_validate_texgen( GLcontext *ctx, GLuint unit ) (texUnit->GenS.Mode != texUnit->GenQ.Mode)) ) { /* Mixed modes, fallback: */ - if (RADEON_DEBUG & DEBUG_FALLBACKS) + if (RADEON_DEBUG & RADEON_FALLBACKS) fprintf(stderr, "fallback mixed texgen\n"); return GL_FALSE; } @@ -1038,7 +941,7 @@ static GLboolean radeon_validate_texgen( GLcontext *ctx, GLuint unit ) } else { /* some texgen mode not including both S and T bits */ - if (RADEON_DEBUG & DEBUG_FALLBACKS) + if (RADEON_DEBUG & RADEON_FALLBACKS) fprintf(stderr, "fallback mixed texgen/nontexgen\n"); return GL_FALSE; } @@ -1088,289 +991,195 @@ static GLboolean radeon_validate_texgen( GLcontext *ctx, GLuint unit ) default: /* Unsupported mode, fallback: */ - if (RADEON_DEBUG & DEBUG_FALLBACKS) + if (RADEON_DEBUG & RADEON_FALLBACKS) fprintf(stderr, "fallback GL_SPHERE_MAP\n"); return GL_FALSE; } if (tmp != rmesa->TexGenEnabled) { - rmesa->NewGLState |= _NEW_TEXTURE_MATRIX; + rmesa->radeon.NewGLState |= _NEW_TEXTURE_MATRIX; } return GL_TRUE; } - -static void disable_tex( GLcontext *ctx, int unit ) -{ - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - - if (rmesa->hw.ctx.cmd[CTX_PP_CNTL] & (RADEON_TEX_0_ENABLE<<unit)) { - /* Texture unit disabled */ - if ( rmesa->state.texture.unit[unit].texobj != NULL ) { - /* The old texture is no longer bound to this texture unit. - * Mark it as such. - */ - - rmesa->state.texture.unit[unit].texobj->base.bound &= ~(1UL << unit); - rmesa->state.texture.unit[unit].texobj = NULL; - } - - RADEON_STATECHANGE( rmesa, ctx ); - rmesa->hw.ctx.cmd[CTX_PP_CNTL] &= - ~((RADEON_TEX_0_ENABLE | RADEON_TEX_BLEND_0_ENABLE) << unit); - - RADEON_STATECHANGE( rmesa, tcl ); - rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] &= ~(RADEON_ST_BIT(unit) | - RADEON_Q_BIT(unit)); - - if (rmesa->TclFallback & (RADEON_TCL_FALLBACK_TEXGEN_0<<unit)) { - TCL_FALLBACK( ctx, (RADEON_TCL_FALLBACK_TEXGEN_0<<unit), GL_FALSE); - rmesa->recheck_texgen[unit] = GL_TRUE; - } - - if (rmesa->hw.tex[unit].cmd[TEX_PP_TXFORMAT] & RADEON_TXFORMAT_CUBIC_MAP_ENABLE) { - /* this seems to be a genuine (r100 only?) hw bug. Need to remove the - cubic_map bit on unit 2 when the unit is disabled, otherwise every - 2nd (2d) mipmap on unit 0 will be broken (may not be needed for other - units, better be safe than sorry though).*/ - RADEON_STATECHANGE( rmesa, tex[unit] ); - rmesa->hw.tex[unit].cmd[TEX_PP_TXFORMAT] &= ~RADEON_TXFORMAT_CUBIC_MAP_ENABLE; - } - - { - GLuint inputshift = RADEON_TEXGEN_0_INPUT_SHIFT + unit*4; - GLuint tmp = rmesa->TexGenEnabled; - - rmesa->TexGenEnabled &= ~(RADEON_TEXGEN_TEXMAT_0_ENABLE<<unit); - rmesa->TexGenEnabled &= ~(RADEON_TEXMAT_0_ENABLE<<unit); - rmesa->TexGenEnabled &= ~(RADEON_TEXGEN_INPUT_MASK<<inputshift); - rmesa->TexGenNeedNormals[unit] = 0; - rmesa->TexGenEnabled |= - (RADEON_TEXGEN_INPUT_TEXCOORD_0+unit) << inputshift; - - if (tmp != rmesa->TexGenEnabled) { - rmesa->recheck_texgen[unit] = GL_TRUE; - rmesa->NewGLState |= _NEW_TEXTURE_MATRIX; - } - } - } -} - -static GLboolean enable_tex_2d( GLcontext *ctx, int unit ) -{ - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit]; - struct gl_texture_object *tObj = texUnit->_Current; - radeonTexObjPtr t = (radeonTexObjPtr) tObj->DriverData; - - /* Need to load the 2d images associated with this unit. - */ - if (t->pp_txformat & RADEON_TXFORMAT_NON_POWER2) { - t->pp_txformat &= ~RADEON_TXFORMAT_NON_POWER2; - t->base.dirty_images[0] = ~0; - } - - ASSERT(tObj->Target == GL_TEXTURE_2D || tObj->Target == GL_TEXTURE_1D); - - if ( t->base.dirty_images[0] ) { - RADEON_FIREVERTICES( rmesa ); - radeonSetTexImages( rmesa, tObj ); - radeonUploadTexImages( rmesa, (radeonTexObjPtr) tObj->DriverData, 0 ); - if ( !t->base.memBlock && !t->image_override ) - return GL_FALSE; - } - - return GL_TRUE; -} - -static GLboolean enable_tex_cube( GLcontext *ctx, int unit ) +/** + * Compute the cached hardware register values for the given texture object. + * + * \param rmesa Context pointer + * \param t the r300 texture object + */ +static GLboolean setup_hardware_state(r100ContextPtr rmesa, radeonTexObj *t, int unit) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit]; - struct gl_texture_object *tObj = texUnit->_Current; - radeonTexObjPtr t = (radeonTexObjPtr) tObj->DriverData; - GLuint face; + const struct gl_texture_image *firstImage; + GLint log2Width, log2Height, log2Depth, texelBytes; - /* Need to load the 2d images associated with this unit. - */ - if (t->pp_txformat & RADEON_TXFORMAT_NON_POWER2) { - t->pp_txformat &= ~RADEON_TXFORMAT_NON_POWER2; - for (face = 0; face < 6; face++) - t->base.dirty_images[face] = ~0; + if ( t->bo ) { + return GL_TRUE; } - ASSERT(tObj->Target == GL_TEXTURE_CUBE_MAP); + firstImage = t->base.Image[0][t->mt->firstLevel]; - if ( t->base.dirty_images[0] || t->base.dirty_images[1] || - t->base.dirty_images[2] || t->base.dirty_images[3] || - t->base.dirty_images[4] || t->base.dirty_images[5] ) { - /* flush */ - RADEON_FIREVERTICES( rmesa ); - /* layout memory space, once for all faces */ - radeonSetTexImages( rmesa, tObj ); + if (firstImage->Border > 0) { + fprintf(stderr, "%s: border\n", __FUNCTION__); + return GL_FALSE; } - /* upload (per face) */ - for (face = 0; face < 6; face++) { - if (t->base.dirty_images[face]) { - radeonUploadTexImages( rmesa, (radeonTexObjPtr) tObj->DriverData, face ); + log2Width = firstImage->WidthLog2; + log2Height = firstImage->HeightLog2; + log2Depth = firstImage->DepthLog2; + texelBytes = firstImage->TexFormat->TexelBytes; + + if (!t->image_override) { + if (VALID_FORMAT(firstImage->TexFormat->MesaFormat)) { + const struct tx_table *table = tx_table; + + t->pp_txformat &= ~(RADEON_TXFORMAT_FORMAT_MASK | + RADEON_TXFORMAT_ALPHA_IN_MAP); + t->pp_txfilter &= ~RADEON_YUV_TO_RGB; + + t->pp_txformat |= table[ firstImage->TexFormat->MesaFormat ].format; + t->pp_txfilter |= table[ firstImage->TexFormat->MesaFormat ].filter; + } else { + _mesa_problem(NULL, "unexpected texture format in %s", + __FUNCTION__); + return GL_FALSE; } } - - if ( !t->base.memBlock ) { - /* texmem alloc failed, use s/w fallback */ - return GL_FALSE; + + t->pp_txfilter &= ~RADEON_MAX_MIP_LEVEL_MASK; + t->pp_txfilter |= (t->mt->lastLevel - t->mt->firstLevel) << RADEON_MAX_MIP_LEVEL_SHIFT; + + t->pp_txformat &= ~(RADEON_TXFORMAT_WIDTH_MASK | + RADEON_TXFORMAT_HEIGHT_MASK | + RADEON_TXFORMAT_CUBIC_MAP_ENABLE | + RADEON_TXFORMAT_F5_WIDTH_MASK | + RADEON_TXFORMAT_F5_HEIGHT_MASK); + t->pp_txformat |= ((log2Width << RADEON_TXFORMAT_WIDTH_SHIFT) | + (log2Height << RADEON_TXFORMAT_HEIGHT_SHIFT)); + + t->tile_bits = 0; + + if (t->base.Target == GL_TEXTURE_CUBE_MAP) { + ASSERT(log2Width == log2Height); + t->pp_txformat |= ((log2Width << RADEON_TXFORMAT_F5_WIDTH_SHIFT) | + (log2Height << RADEON_TXFORMAT_F5_HEIGHT_SHIFT) | + /* don't think we need this bit, if it exists at all - fglrx does not set it */ + (RADEON_TXFORMAT_CUBIC_MAP_ENABLE)); + t->pp_cubic_faces = ((log2Width << RADEON_FACE_WIDTH_1_SHIFT) | + (log2Height << RADEON_FACE_HEIGHT_1_SHIFT) | + (log2Width << RADEON_FACE_WIDTH_2_SHIFT) | + (log2Height << RADEON_FACE_HEIGHT_2_SHIFT) | + (log2Width << RADEON_FACE_WIDTH_3_SHIFT) | + (log2Height << RADEON_FACE_HEIGHT_3_SHIFT) | + (log2Width << RADEON_FACE_WIDTH_4_SHIFT) | + (log2Height << RADEON_FACE_HEIGHT_4_SHIFT)); } - return GL_TRUE; -} - -static GLboolean enable_tex_rect( GLcontext *ctx, int unit ) -{ - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit]; - struct gl_texture_object *tObj = texUnit->_Current; - radeonTexObjPtr t = (radeonTexObjPtr) tObj->DriverData; + t->pp_txsize = (((firstImage->Width - 1) << RADEON_TEX_USIZE_SHIFT) + | ((firstImage->Height - 1) << RADEON_TEX_VSIZE_SHIFT)); - if (!(t->pp_txformat & RADEON_TXFORMAT_NON_POWER2)) { - t->pp_txformat |= RADEON_TXFORMAT_NON_POWER2; - t->base.dirty_images[0] = ~0; + if ( !t->image_override ) { + if (firstImage->IsCompressed) + t->pp_txpitch = (firstImage->Width + 63) & ~(63); + else + t->pp_txpitch = ((firstImage->Width * texelBytes) + 63) & ~(63); + t->pp_txpitch -= 32; } - ASSERT(tObj->Target == GL_TEXTURE_RECTANGLE_NV); - - if ( t->base.dirty_images[0] ) { - RADEON_FIREVERTICES( rmesa ); - radeonSetTexImages( rmesa, tObj ); - radeonUploadTexImages( rmesa, (radeonTexObjPtr) tObj->DriverData, 0 ); - if ( !t->base.memBlock && - !t->image_override /* && !rmesa->prefer_gart_client_texturing FIXME */ ) { - fprintf(stderr, "%s: upload failed\n", __FUNCTION__); - return GL_FALSE; - } + if (t->base.Target == GL_TEXTURE_RECTANGLE_NV) { + t->pp_txformat |= RADEON_TXFORMAT_NON_POWER2; } return GL_TRUE; } - -static GLboolean update_tex_common( GLcontext *ctx, int unit ) +static GLboolean radeon_validate_texture(GLcontext *ctx, struct gl_texture_object *texObj, int unit) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit]; - struct gl_texture_object *tObj = texUnit->_Current; - radeonTexObjPtr t = (radeonTexObjPtr) tObj->DriverData; - GLenum format; + r100ContextPtr rmesa = R100_CONTEXT(ctx); + radeonTexObj *t = radeon_tex_obj(texObj); + int ret; - /* Fallback if there's a texture border */ - if ( tObj->Image[0][tObj->BaseLevel]->Border > 0 ) { - fprintf(stderr, "%s: border\n", __FUNCTION__); + if (!radeon_validate_texture_miptree(ctx, texObj)) return GL_FALSE; - } + + ret = setup_hardware_state(rmesa, t, unit); + if (ret == GL_FALSE) + return GL_FALSE; + /* yuv conversion only works in first unit */ if (unit != 0 && (t->pp_txfilter & RADEON_YUV_TO_RGB)) return GL_FALSE; - /* Update state if this is a different texture object to last - * time. - */ - if ( rmesa->state.texture.unit[unit].texobj != t ) { - if ( rmesa->state.texture.unit[unit].texobj != NULL ) { - /* The old texture is no longer bound to this texture unit. - * Mark it as such. - */ - - rmesa->state.texture.unit[unit].texobj->base.bound &= - ~(1UL << unit); - } - - rmesa->state.texture.unit[unit].texobj = t; - t->base.bound |= (1UL << unit); - t->dirty_state |= 1<<unit; - driUpdateTextureLRU( (driTextureObject *) t ); /* XXX: should be locked! */ - } + RADEON_STATECHANGE( rmesa, ctx ); + rmesa->hw.ctx.cmd[CTX_PP_CNTL] |= + (RADEON_TEX_0_ENABLE | RADEON_TEX_BLEND_0_ENABLE) << unit; + RADEON_STATECHANGE( rmesa, tcl ); + rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] |= RADEON_ST_BIT(unit); - /* Newly enabled? - */ - if ( !(rmesa->hw.ctx.cmd[CTX_PP_CNTL] & (RADEON_TEX_0_ENABLE<<unit))) { - RADEON_STATECHANGE( rmesa, ctx ); - rmesa->hw.ctx.cmd[CTX_PP_CNTL] |= - (RADEON_TEX_0_ENABLE | RADEON_TEX_BLEND_0_ENABLE) << unit; + rmesa->recheck_texgen[unit] = GL_TRUE; - RADEON_STATECHANGE( rmesa, tcl ); - - rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] |= RADEON_ST_BIT(unit); - - rmesa->recheck_texgen[unit] = GL_TRUE; - } - - if (t->dirty_state & (1<<unit)) { - import_tex_obj_state( rmesa, unit, t ); - /* may need to update texture matrix (for texrect adjustments) */ - rmesa->NewGLState |= _NEW_TEXTURE_MATRIX; - } + import_tex_obj_state( rmesa, unit, t ); if (rmesa->recheck_texgen[unit]) { GLboolean fallback = !radeon_validate_texgen( ctx, unit ); TCL_FALLBACK( ctx, (RADEON_TCL_FALLBACK_TEXGEN_0<<unit), fallback); rmesa->recheck_texgen[unit] = 0; - rmesa->NewGLState |= _NEW_TEXTURE_MATRIX; + rmesa->radeon.NewGLState |= _NEW_TEXTURE_MATRIX; } - format = tObj->Image[0][tObj->BaseLevel]->_BaseFormat; - if ( rmesa->state.texture.unit[unit].format != format || - rmesa->state.texture.unit[unit].envMode != texUnit->EnvMode ) { - rmesa->state.texture.unit[unit].format = format; - rmesa->state.texture.unit[unit].envMode = texUnit->EnvMode; - if ( ! radeonUpdateTextureEnv( ctx, unit ) ) { - return GL_FALSE; - } + if ( ! radeonUpdateTextureEnv( ctx, unit ) ) { + return GL_FALSE; } - FALLBACK( rmesa, RADEON_FALLBACK_BORDER_MODE, t->border_fallback ); + + t->validated = GL_TRUE; return !t->border_fallback; } - - static GLboolean radeonUpdateTextureUnit( GLcontext *ctx, int unit ) { - struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit]; + r100ContextPtr rmesa = R100_CONTEXT(ctx); - if ( texUnit->_ReallyEnabled & (TEXTURE_RECT_BIT) ) { - return (enable_tex_rect( ctx, unit ) && - update_tex_common( ctx, unit )); - } - else if ( texUnit->_ReallyEnabled & (TEXTURE_1D_BIT | TEXTURE_2D_BIT) ) { - return (enable_tex_2d( ctx, unit ) && - update_tex_common( ctx, unit )); - } - else if ( texUnit->_ReallyEnabled & (TEXTURE_CUBE_BIT) ) { - return (enable_tex_cube( ctx, unit ) && - update_tex_common( ctx, unit )); + if (ctx->Texture.Unit[unit]._ReallyEnabled & TEXTURE_3D_BIT) { + rmesa->state.texture.unit[unit].texobj = NULL; + return GL_FALSE; } - else if ( texUnit->_ReallyEnabled ) { - return GL_FALSE; + + if (!ctx->Texture.Unit[unit]._ReallyEnabled) { + /* disable the unit */ + disable_tex_obj_state(rmesa, unit); + rmesa->state.texture.unit[unit].texobj = NULL; + return GL_TRUE; } - else { - disable_tex( ctx, unit ); - return GL_TRUE; + + if (!radeon_validate_texture(ctx, ctx->Texture.Unit[unit]._Current, unit)) { + _mesa_warning(ctx, + "failed to validate texture for unit %d.\n", + unit); + rmesa->state.texture.unit[unit].texobj = NULL; + return GL_FALSE; } + rmesa->state.texture.unit[unit].texobj = radeon_tex_obj(ctx->Texture.Unit[unit]._Current); + return GL_TRUE; } void radeonUpdateTextureState( GLcontext *ctx ) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + r100ContextPtr rmesa = R100_CONTEXT(ctx); GLboolean ok; + /* set the ctx all textures off */ + RADEON_STATECHANGE( rmesa, ctx ); + rmesa->hw.ctx.cmd[CTX_PP_CNTL] &= ~((RADEON_TEX_ENABLE_MASK) | (RADEON_TEX_BLEND_ENABLE_MASK)); + ok = (radeonUpdateTextureUnit( ctx, 0 ) && radeonUpdateTextureUnit( ctx, 1 ) && radeonUpdateTextureUnit( ctx, 2 )); FALLBACK( rmesa, RADEON_FALLBACK_TEXTURE, !ok ); - if (rmesa->TclFallback) + if (rmesa->radeon.TclFallback) radeonChooseVertexState( ctx ); } diff --git a/src/mesa/drivers/dri/radeon/radeon_texture.c b/src/mesa/drivers/dri/radeon/radeon_texture.c new file mode 100644 index 00000000000..fad3d1cedaf --- /dev/null +++ b/src/mesa/drivers/dri/radeon/radeon_texture.c @@ -0,0 +1,1052 @@ +/* + * Copyright (C) 2008 Nicolai Haehnle. + * Copyright (C) The Weather Channel, Inc. 2002. All Rights Reserved. + * + * The Weather Channel (TM) funded Tungsten Graphics to develop the + * initial release of the Radeon 8500 driver under the XFree86 license. + * This notice must be preserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sublicense, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial + * portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + */ + +#include "main/glheader.h" +#include "main/imports.h" +#include "main/context.h" +#include "main/convolve.h" +#include "main/mipmap.h" +#include "main/texcompress.h" +#include "main/texformat.h" +#include "main/texstore.h" +#include "main/teximage.h" +#include "main/texobj.h" +#include "main/texgetimage.h" + +#include "xmlpool.h" /* for symbolic values of enum-type options */ + +#include "radeon_common.h" + +#include "radeon_mipmap_tree.h" + + +static void copy_rows(void* dst, GLuint dststride, const void* src, GLuint srcstride, + GLuint numrows, GLuint rowsize) +{ + assert(rowsize <= dststride); + assert(rowsize <= srcstride); + + if (rowsize == srcstride && rowsize == dststride) { + memcpy(dst, src, numrows*rowsize); + } else { + GLuint i; + for(i = 0; i < numrows; ++i) { + memcpy(dst, src, rowsize); + dst += dststride; + src += srcstride; + } + } +} + +/* textures */ +/** + * Allocate an empty texture image object. + */ +struct gl_texture_image *radeonNewTextureImage(GLcontext *ctx) +{ + return CALLOC(sizeof(radeon_texture_image)); +} + +/** + * Free memory associated with this texture image. + */ +void radeonFreeTexImageData(GLcontext *ctx, struct gl_texture_image *timage) +{ + radeon_texture_image* image = get_radeon_texture_image(timage); + + if (image->mt) { + radeon_miptree_unreference(image->mt); + image->mt = 0; + assert(!image->base.Data); + } else { + _mesa_free_texture_image_data(ctx, timage); + } + if (image->bo) { + radeon_bo_unref(image->bo); + image->bo = NULL; + } + if (timage->Data) { + _mesa_free_texmemory(timage->Data); + timage->Data = NULL; + } +} + +/* Set Data pointer and additional data for mapped texture image */ +static void teximage_set_map_data(radeon_texture_image *image) +{ + radeon_mipmap_level *lvl = &image->mt->levels[image->mtlevel]; + + image->base.Data = image->mt->bo->ptr + lvl->faces[image->mtface].offset; + image->base.RowStride = lvl->rowstride / image->mt->bpp; +} + + +/** + * Map a single texture image for glTexImage and friends. + */ +void radeon_teximage_map(radeon_texture_image *image, GLboolean write_enable) +{ + if (image->mt) { + assert(!image->base.Data); + + radeon_bo_map(image->mt->bo, write_enable); + teximage_set_map_data(image); + } +} + + +void radeon_teximage_unmap(radeon_texture_image *image) +{ + if (image->mt) { + assert(image->base.Data); + + image->base.Data = 0; + radeon_bo_unmap(image->mt->bo); + } +} + +static void map_override(GLcontext *ctx, radeonTexObj *t) +{ + radeon_texture_image *img = get_radeon_texture_image(t->base.Image[0][0]); + + radeon_bo_map(t->bo, GL_FALSE); + + img->base.Data = t->bo->ptr; + _mesa_set_fetch_functions(&img->base, 2); +} + +static void unmap_override(GLcontext *ctx, radeonTexObj *t) +{ + radeon_texture_image *img = get_radeon_texture_image(t->base.Image[0][0]); + + radeon_bo_unmap(t->bo); + + img->base.Data = NULL; +} + +/** + * Map a validated texture for reading during software rendering. + */ +void radeonMapTexture(GLcontext *ctx, struct gl_texture_object *texObj) +{ + radeonTexObj* t = radeon_tex_obj(texObj); + int face, level; + + if (!radeon_validate_texture_miptree(ctx, texObj)) + return; + + /* for r100 3D sw fallbacks don't have mt */ + if (t->image_override && t->bo) + map_override(ctx, t); + + if (!t->mt) + return; + + radeon_bo_map(t->mt->bo, GL_FALSE); + for(face = 0; face < t->mt->faces; ++face) { + for(level = t->mt->firstLevel; level <= t->mt->lastLevel; ++level) + teximage_set_map_data(get_radeon_texture_image(texObj->Image[face][level])); + } +} + +void radeonUnmapTexture(GLcontext *ctx, struct gl_texture_object *texObj) +{ + radeonTexObj* t = radeon_tex_obj(texObj); + int face, level; + + if (t->image_override && t->bo) + unmap_override(ctx, t); + /* for r100 3D sw fallbacks don't have mt */ + if (!t->mt) + return; + + for(face = 0; face < t->mt->faces; ++face) { + for(level = t->mt->firstLevel; level <= t->mt->lastLevel; ++level) + texObj->Image[face][level]->Data = 0; + } + radeon_bo_unmap(t->mt->bo); +} + +GLuint radeon_face_for_target(GLenum target) +{ + switch (target) { + case GL_TEXTURE_CUBE_MAP_POSITIVE_X: + case GL_TEXTURE_CUBE_MAP_NEGATIVE_X: + case GL_TEXTURE_CUBE_MAP_POSITIVE_Y: + case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y: + case GL_TEXTURE_CUBE_MAP_POSITIVE_Z: + case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z: + return (GLuint) target - (GLuint) GL_TEXTURE_CUBE_MAP_POSITIVE_X; + default: + return 0; + } +} + +/** + * Wraps Mesa's implementation to ensure that the base level image is mapped. + * + * This relies on internal details of _mesa_generate_mipmap, in particular + * the fact that the memory for recreated texture images is always freed. + */ +static void radeon_generate_mipmap(GLcontext *ctx, GLenum target, + struct gl_texture_object *texObj) +{ + radeonTexObj* t = radeon_tex_obj(texObj); + GLuint nr_faces = (t->base.Target == GL_TEXTURE_CUBE_MAP) ? 6 : 1; + int i, face; + + + _mesa_generate_mipmap(ctx, target, texObj); + + for (face = 0; face < nr_faces; face++) { + for (i = texObj->BaseLevel + 1; i < texObj->MaxLevel; i++) { + radeon_texture_image *image; + + image = get_radeon_texture_image(texObj->Image[face][i]); + + if (image == NULL) + break; + + image->mtlevel = i; + image->mtface = face; + + radeon_miptree_unreference(image->mt); + image->mt = NULL; + } + } + +} + +void radeonGenerateMipmap(GLcontext* ctx, GLenum target, struct gl_texture_object *texObj) +{ + GLuint face = radeon_face_for_target(target); + radeon_texture_image *baseimage = get_radeon_texture_image(texObj->Image[face][texObj->BaseLevel]); + + radeon_teximage_map(baseimage, GL_FALSE); + radeon_generate_mipmap(ctx, target, texObj); + radeon_teximage_unmap(baseimage); +} + + +/* try to find a format which will only need a memcopy */ +static const struct gl_texture_format *radeonChoose8888TexFormat(radeonContextPtr rmesa, + GLenum srcFormat, + GLenum srcType, GLboolean fbo) +{ + const GLuint ui = 1; + const GLubyte littleEndian = *((const GLubyte *)&ui); + + /* r100 can only do this */ + if (IS_R100_CLASS(rmesa->radeonScreen) || fbo) + return _dri_texformat_argb8888; + + if ((srcFormat == GL_RGBA && srcType == GL_UNSIGNED_INT_8_8_8_8) || + (srcFormat == GL_RGBA && srcType == GL_UNSIGNED_BYTE && !littleEndian) || + (srcFormat == GL_ABGR_EXT && srcType == GL_UNSIGNED_INT_8_8_8_8_REV) || + (srcFormat == GL_ABGR_EXT && srcType == GL_UNSIGNED_BYTE && littleEndian)) { + return &_mesa_texformat_rgba8888; + } else if ((srcFormat == GL_RGBA && srcType == GL_UNSIGNED_INT_8_8_8_8_REV) || + (srcFormat == GL_RGBA && srcType == GL_UNSIGNED_BYTE && littleEndian) || + (srcFormat == GL_ABGR_EXT && srcType == GL_UNSIGNED_INT_8_8_8_8) || + (srcFormat == GL_ABGR_EXT && srcType == GL_UNSIGNED_BYTE && !littleEndian)) { + return &_mesa_texformat_rgba8888_rev; + } else if (IS_R200_CLASS(rmesa->radeonScreen)) { + return _dri_texformat_argb8888; + } else if (srcFormat == GL_BGRA && ((srcType == GL_UNSIGNED_BYTE && !littleEndian) || + srcType == GL_UNSIGNED_INT_8_8_8_8)) { + return &_mesa_texformat_argb8888_rev; + } else if (srcFormat == GL_BGRA && ((srcType == GL_UNSIGNED_BYTE && littleEndian) || + srcType == GL_UNSIGNED_INT_8_8_8_8_REV)) { + return &_mesa_texformat_argb8888; + } else + return _dri_texformat_argb8888; +} + +const struct gl_texture_format *radeonChooseTextureFormat_mesa(GLcontext * ctx, + GLint internalFormat, + GLenum format, + GLenum type) +{ + return radeonChooseTextureFormat(ctx, internalFormat, format, + type, 0); +} + +const struct gl_texture_format *radeonChooseTextureFormat(GLcontext * ctx, + GLint internalFormat, + GLenum format, + GLenum type, GLboolean fbo) +{ + radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + const GLboolean do32bpt = + (rmesa->texture_depth == DRI_CONF_TEXTURE_DEPTH_32); + const GLboolean force16bpt = + (rmesa->texture_depth == DRI_CONF_TEXTURE_DEPTH_FORCE_16); + (void)format; + +#if 0 + fprintf(stderr, "InternalFormat=%s(%d) type=%s format=%s\n", + _mesa_lookup_enum_by_nr(internalFormat), internalFormat, + _mesa_lookup_enum_by_nr(type), _mesa_lookup_enum_by_nr(format)); + fprintf(stderr, "do32bpt=%d force16bpt=%d\n", do32bpt, force16bpt); +#endif + + switch (internalFormat) { + case 4: + case GL_RGBA: + case GL_COMPRESSED_RGBA: + switch (type) { + case GL_UNSIGNED_INT_10_10_10_2: + case GL_UNSIGNED_INT_2_10_10_10_REV: + return do32bpt ? _dri_texformat_argb8888 : + _dri_texformat_argb1555; + case GL_UNSIGNED_SHORT_4_4_4_4: + case GL_UNSIGNED_SHORT_4_4_4_4_REV: + return _dri_texformat_argb4444; + case GL_UNSIGNED_SHORT_5_5_5_1: + case GL_UNSIGNED_SHORT_1_5_5_5_REV: + return _dri_texformat_argb1555; + default: + return do32bpt ? radeonChoose8888TexFormat(rmesa, format, type, fbo) : + _dri_texformat_argb4444; + } + + case 3: + case GL_RGB: + case GL_COMPRESSED_RGB: + switch (type) { + case GL_UNSIGNED_SHORT_4_4_4_4: + case GL_UNSIGNED_SHORT_4_4_4_4_REV: + return _dri_texformat_argb4444; + case GL_UNSIGNED_SHORT_5_5_5_1: + case GL_UNSIGNED_SHORT_1_5_5_5_REV: + return _dri_texformat_argb1555; + case GL_UNSIGNED_SHORT_5_6_5: + case GL_UNSIGNED_SHORT_5_6_5_REV: + return _dri_texformat_rgb565; + default: + return do32bpt ? _dri_texformat_argb8888 : + _dri_texformat_rgb565; + } + + case GL_RGBA8: + case GL_RGB10_A2: + case GL_RGBA12: + case GL_RGBA16: + return !force16bpt ? + radeonChoose8888TexFormat(rmesa, format, type, fbo) : + _dri_texformat_argb4444; + + case GL_RGBA4: + case GL_RGBA2: + return _dri_texformat_argb4444; + + case GL_RGB5_A1: + return _dri_texformat_argb1555; + + case GL_RGB8: + case GL_RGB10: + case GL_RGB12: + case GL_RGB16: + return !force16bpt ? _dri_texformat_argb8888 : + _dri_texformat_rgb565; + + case GL_RGB5: + case GL_RGB4: + case GL_R3_G3_B2: + return _dri_texformat_rgb565; + + case GL_ALPHA: + case GL_ALPHA4: + case GL_ALPHA8: + case GL_ALPHA12: + case GL_ALPHA16: + case GL_COMPRESSED_ALPHA: + /* r200: can't use a8 format since interpreting hw I8 as a8 would result + in wrong rgb values (same as alpha value instead of 0). */ + if (IS_R200_CLASS(rmesa->radeonScreen)) + return _dri_texformat_al88; + else + return _dri_texformat_a8; + case 1: + case GL_LUMINANCE: + case GL_LUMINANCE4: + case GL_LUMINANCE8: + case GL_LUMINANCE12: + case GL_LUMINANCE16: + case GL_COMPRESSED_LUMINANCE: + return _dri_texformat_l8; + + case 2: + case GL_LUMINANCE_ALPHA: + case GL_LUMINANCE4_ALPHA4: + case GL_LUMINANCE6_ALPHA2: + case GL_LUMINANCE8_ALPHA8: + case GL_LUMINANCE12_ALPHA4: + case GL_LUMINANCE12_ALPHA12: + case GL_LUMINANCE16_ALPHA16: + case GL_COMPRESSED_LUMINANCE_ALPHA: + return _dri_texformat_al88; + + case GL_INTENSITY: + case GL_INTENSITY4: + case GL_INTENSITY8: + case GL_INTENSITY12: + case GL_INTENSITY16: + case GL_COMPRESSED_INTENSITY: + return _dri_texformat_i8; + + case GL_YCBCR_MESA: + if (type == GL_UNSIGNED_SHORT_8_8_APPLE || + type == GL_UNSIGNED_BYTE) + return &_mesa_texformat_ycbcr; + else + return &_mesa_texformat_ycbcr_rev; + + case GL_RGB_S3TC: + case GL_RGB4_S3TC: + case GL_COMPRESSED_RGB_S3TC_DXT1_EXT: + return &_mesa_texformat_rgb_dxt1; + + case GL_COMPRESSED_RGBA_S3TC_DXT1_EXT: + return &_mesa_texformat_rgba_dxt1; + + case GL_RGBA_S3TC: + case GL_RGBA4_S3TC: + case GL_COMPRESSED_RGBA_S3TC_DXT3_EXT: + return &_mesa_texformat_rgba_dxt3; + + case GL_COMPRESSED_RGBA_S3TC_DXT5_EXT: + return &_mesa_texformat_rgba_dxt5; + + case GL_ALPHA16F_ARB: + return &_mesa_texformat_alpha_float16; + case GL_ALPHA32F_ARB: + return &_mesa_texformat_alpha_float32; + case GL_LUMINANCE16F_ARB: + return &_mesa_texformat_luminance_float16; + case GL_LUMINANCE32F_ARB: + return &_mesa_texformat_luminance_float32; + case GL_LUMINANCE_ALPHA16F_ARB: + return &_mesa_texformat_luminance_alpha_float16; + case GL_LUMINANCE_ALPHA32F_ARB: + return &_mesa_texformat_luminance_alpha_float32; + case GL_INTENSITY16F_ARB: + return &_mesa_texformat_intensity_float16; + case GL_INTENSITY32F_ARB: + return &_mesa_texformat_intensity_float32; + case GL_RGB16F_ARB: + return &_mesa_texformat_rgba_float16; + case GL_RGB32F_ARB: + return &_mesa_texformat_rgba_float32; + case GL_RGBA16F_ARB: + return &_mesa_texformat_rgba_float16; + case GL_RGBA32F_ARB: + return &_mesa_texformat_rgba_float32; + + case GL_DEPTH_COMPONENT: + case GL_DEPTH_COMPONENT16: + case GL_DEPTH_COMPONENT24: + case GL_DEPTH_COMPONENT32: + case GL_DEPTH_STENCIL_EXT: + case GL_DEPTH24_STENCIL8_EXT: + return &_mesa_texformat_s8_z24; + + /* EXT_texture_sRGB */ + case GL_SRGB: + case GL_SRGB8: + case GL_SRGB_ALPHA: + case GL_SRGB8_ALPHA8: + case GL_COMPRESSED_SRGB: + case GL_COMPRESSED_SRGB_ALPHA: + return &_mesa_texformat_srgba8; + + case GL_SLUMINANCE: + case GL_SLUMINANCE8: + case GL_COMPRESSED_SLUMINANCE: + return &_mesa_texformat_sl8; + + case GL_SLUMINANCE_ALPHA: + case GL_SLUMINANCE8_ALPHA8: + case GL_COMPRESSED_SLUMINANCE_ALPHA: + return &_mesa_texformat_sla8; + + default: + _mesa_problem(ctx, + "unexpected internalFormat 0x%x in %s", + (int)internalFormat, __func__); + return NULL; + } + + return NULL; /* never get here */ +} + +/** + * All glTexImage calls go through this function. + */ +static void radeon_teximage( + GLcontext *ctx, int dims, + GLenum target, GLint level, + GLint internalFormat, + GLint width, GLint height, GLint depth, + GLsizei imageSize, + GLenum format, GLenum type, const GLvoid * pixels, + const struct gl_pixelstore_attrib *packing, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage, + int compressed) +{ + radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + radeonTexObj* t = radeon_tex_obj(texObj); + radeon_texture_image* image = get_radeon_texture_image(texImage); + GLuint dstRowStride; + GLint postConvWidth = width; + GLint postConvHeight = height; + GLuint texelBytes; + GLuint face = radeon_face_for_target(target); + + radeon_firevertices(rmesa); + + t->validated = GL_FALSE; + + if (ctx->_ImageTransferState & IMAGE_CONVOLUTION_BIT) { + _mesa_adjust_image_for_convolution(ctx, dims, &postConvWidth, + &postConvHeight); + } + + /* Choose and fill in the texture format for this image */ + texImage->TexFormat = radeonChooseTextureFormat(ctx, internalFormat, format, type, 0); + _mesa_set_fetch_functions(texImage, dims); + + if (texImage->TexFormat->TexelBytes == 0) { + texelBytes = 0; + texImage->IsCompressed = GL_TRUE; + texImage->CompressedSize = + ctx->Driver.CompressedTextureSize(ctx, texImage->Width, + texImage->Height, texImage->Depth, + texImage->TexFormat->MesaFormat); + } else { + texImage->IsCompressed = GL_FALSE; + texImage->CompressedSize = 0; + + texelBytes = texImage->TexFormat->TexelBytes; + /* Minimum pitch of 32 bytes */ + if (postConvWidth * texelBytes < 32) { + postConvWidth = 32 / texelBytes; + texImage->RowStride = postConvWidth; + } + if (!image->mt) { + assert(texImage->RowStride == postConvWidth); + } + } + + /* Allocate memory for image */ + radeonFreeTexImageData(ctx, texImage); /* Mesa core only clears texImage->Data but not image->mt */ + + if (t->mt && + t->mt->firstLevel == level && + t->mt->lastLevel == level && + t->mt->target != GL_TEXTURE_CUBE_MAP_ARB && + !radeon_miptree_matches_image(t->mt, texImage, face, level)) { + radeon_miptree_unreference(t->mt); + t->mt = NULL; + } + + if (!t->mt) + radeon_try_alloc_miptree(rmesa, t, image, face, level); + if (t->mt && radeon_miptree_matches_image(t->mt, texImage, face, level)) { + radeon_mipmap_level *lvl; + image->mt = t->mt; + image->mtlevel = level - t->mt->firstLevel; + image->mtface = face; + radeon_miptree_reference(t->mt); + lvl = &image->mt->levels[image->mtlevel]; + dstRowStride = lvl->rowstride; + } else { + int size; + if (texImage->IsCompressed) { + size = texImage->CompressedSize; + } else { + size = texImage->Width * texImage->Height * texImage->Depth * texImage->TexFormat->TexelBytes; + } + texImage->Data = _mesa_alloc_texmemory(size); + } + + /* Upload texture image; note that the spec allows pixels to be NULL */ + if (compressed) { + pixels = _mesa_validate_pbo_compressed_teximage( + ctx, imageSize, pixels, packing, "glCompressedTexImage"); + } else { + pixels = _mesa_validate_pbo_teximage( + ctx, dims, width, height, depth, + format, type, pixels, packing, "glTexImage"); + } + + if (pixels) { + radeon_teximage_map(image, GL_TRUE); + if (compressed) { + if (image->mt) { + uint32_t srcRowStride, bytesPerRow, rows; + srcRowStride = _mesa_compressed_row_stride(texImage->TexFormat->MesaFormat, width); + bytesPerRow = srcRowStride; + rows = (height + 3) / 4; + copy_rows(texImage->Data, image->mt->levels[level].rowstride, + pixels, srcRowStride, rows, bytesPerRow); + } else { + memcpy(texImage->Data, pixels, imageSize); + } + } else { + GLuint dstRowStride; + GLuint *dstImageOffsets; + + if (image->mt) { + radeon_mipmap_level *lvl = &image->mt->levels[image->mtlevel]; + dstRowStride = lvl->rowstride; + } else { + dstRowStride = texImage->Width * texImage->TexFormat->TexelBytes; + } + + if (dims == 3) { + int i; + + dstImageOffsets = _mesa_malloc(depth * sizeof(GLuint)) ; + if (!dstImageOffsets) + _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage"); + + for (i = 0; i < depth; ++i) { + dstImageOffsets[i] = dstRowStride/texImage->TexFormat->TexelBytes * height * i; + } + } else { + dstImageOffsets = texImage->ImageOffsets; + } + + if (!texImage->TexFormat->StoreImage(ctx, dims, + texImage->_BaseFormat, + texImage->TexFormat, + texImage->Data, 0, 0, 0, /* dstX/Y/Zoffset */ + dstRowStride, + dstImageOffsets, + width, height, depth, + format, type, pixels, packing)) + _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage"); + + if (dims == 3) + _mesa_free(dstImageOffsets); + } + + /* SGIS_generate_mipmap */ + if (level == texObj->BaseLevel && texObj->GenerateMipmap) { + radeon_generate_mipmap(ctx, target, texObj); + } + } + + _mesa_unmap_teximage_pbo(ctx, packing); + + if (pixels) + radeon_teximage_unmap(image); + + +} + +void radeonTexImage1D(GLcontext * ctx, GLenum target, GLint level, + GLint internalFormat, + GLint width, GLint border, + GLenum format, GLenum type, const GLvoid * pixels, + const struct gl_pixelstore_attrib *packing, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + radeon_teximage(ctx, 1, target, level, internalFormat, width, 1, 1, + 0, format, type, pixels, packing, texObj, texImage, 0); +} + +void radeonTexImage2D(GLcontext * ctx, GLenum target, GLint level, + GLint internalFormat, + GLint width, GLint height, GLint border, + GLenum format, GLenum type, const GLvoid * pixels, + const struct gl_pixelstore_attrib *packing, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) + +{ + radeon_teximage(ctx, 2, target, level, internalFormat, width, height, 1, + 0, format, type, pixels, packing, texObj, texImage, 0); +} + +void radeonCompressedTexImage2D(GLcontext * ctx, GLenum target, + GLint level, GLint internalFormat, + GLint width, GLint height, GLint border, + GLsizei imageSize, const GLvoid * data, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + radeon_teximage(ctx, 2, target, level, internalFormat, width, height, 1, + imageSize, 0, 0, data, &ctx->Unpack, texObj, texImage, 1); +} + +void radeonTexImage3D(GLcontext * ctx, GLenum target, GLint level, + GLint internalFormat, + GLint width, GLint height, GLint depth, + GLint border, + GLenum format, GLenum type, const GLvoid * pixels, + const struct gl_pixelstore_attrib *packing, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + radeon_teximage(ctx, 3, target, level, internalFormat, width, height, depth, + 0, format, type, pixels, packing, texObj, texImage, 0); +} + +/** + * Update a subregion of the given texture image. + */ +static void radeon_texsubimage(GLcontext* ctx, int dims, GLenum target, int level, + GLint xoffset, GLint yoffset, GLint zoffset, + GLsizei width, GLsizei height, GLsizei depth, + GLsizei imageSize, + GLenum format, GLenum type, + const GLvoid * pixels, + const struct gl_pixelstore_attrib *packing, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage, + int compressed) +{ + radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + radeonTexObj* t = radeon_tex_obj(texObj); + radeon_texture_image* image = get_radeon_texture_image(texImage); + + radeon_firevertices(rmesa); + + t->validated = GL_FALSE; + if (compressed) { + pixels = _mesa_validate_pbo_compressed_teximage( + ctx, imageSize, pixels, packing, "glCompressedTexImage"); + } else { + pixels = _mesa_validate_pbo_teximage(ctx, dims, + width, height, depth, format, type, pixels, packing, "glTexSubImage1D"); + } + + if (pixels) { + GLint dstRowStride; + radeon_teximage_map(image, GL_TRUE); + + if (image->mt) { + radeon_mipmap_level *lvl = &image->mt->levels[image->mtlevel]; + dstRowStride = lvl->rowstride; + } else { + dstRowStride = texImage->RowStride * texImage->TexFormat->TexelBytes; + } + + if (compressed) { + uint32_t srcRowStride, bytesPerRow, rows; + GLubyte *img_start; + if (!image->mt) { + dstRowStride = _mesa_compressed_row_stride(texImage->TexFormat->MesaFormat, texImage->Width); + img_start = _mesa_compressed_image_address(xoffset, yoffset, 0, + texImage->TexFormat->MesaFormat, + texImage->Width, texImage->Data); + } + else { + uint32_t blocks_x = dstRowStride / (image->mt->bpp * 4); + img_start = texImage->Data + image->mt->bpp * 4 * (blocks_x * (yoffset / 4) + xoffset / 4); + } + srcRowStride = _mesa_compressed_row_stride(texImage->TexFormat->MesaFormat, width); + bytesPerRow = srcRowStride; + rows = (height + 3) / 4; + + copy_rows(img_start, dstRowStride, pixels, srcRowStride, rows, bytesPerRow); + + } else { + if (!texImage->TexFormat->StoreImage(ctx, dims, texImage->_BaseFormat, + texImage->TexFormat, texImage->Data, + xoffset, yoffset, zoffset, + dstRowStride, + texImage->ImageOffsets, + width, height, depth, + format, type, pixels, packing)) + _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage"); + } + + /* GL_SGIS_generate_mipmap */ + if (level == texObj->BaseLevel && texObj->GenerateMipmap) { + radeon_generate_mipmap(ctx, target, texObj); + } + } + + radeon_teximage_unmap(image); + + _mesa_unmap_teximage_pbo(ctx, packing); + + +} + +void radeonTexSubImage1D(GLcontext * ctx, GLenum target, GLint level, + GLint xoffset, + GLsizei width, + GLenum format, GLenum type, + const GLvoid * pixels, + const struct gl_pixelstore_attrib *packing, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + radeon_texsubimage(ctx, 1, target, level, xoffset, 0, 0, width, 1, 1, 0, + format, type, pixels, packing, texObj, texImage, 0); +} + +void radeonTexSubImage2D(GLcontext * ctx, GLenum target, GLint level, + GLint xoffset, GLint yoffset, + GLsizei width, GLsizei height, + GLenum format, GLenum type, + const GLvoid * pixels, + const struct gl_pixelstore_attrib *packing, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + radeon_texsubimage(ctx, 2, target, level, xoffset, yoffset, 0, width, height, 1, + 0, format, type, pixels, packing, texObj, texImage, + 0); +} + +void radeonCompressedTexSubImage2D(GLcontext * ctx, GLenum target, + GLint level, GLint xoffset, + GLint yoffset, GLsizei width, + GLsizei height, GLenum format, + GLsizei imageSize, const GLvoid * data, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + radeon_texsubimage(ctx, 2, target, level, xoffset, yoffset, 0, width, height, 1, + imageSize, format, 0, data, &ctx->Unpack, texObj, texImage, 1); +} + + +void radeonTexSubImage3D(GLcontext * ctx, GLenum target, GLint level, + GLint xoffset, GLint yoffset, GLint zoffset, + GLsizei width, GLsizei height, GLsizei depth, + GLenum format, GLenum type, + const GLvoid * pixels, + const struct gl_pixelstore_attrib *packing, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + radeon_texsubimage(ctx, 3, target, level, xoffset, yoffset, zoffset, width, height, depth, 0, + format, type, pixels, packing, texObj, texImage, 0); +} + + + +/** + * Ensure that the given image is stored in the given miptree from now on. + */ +static void migrate_image_to_miptree(radeon_mipmap_tree *mt, radeon_texture_image *image, int face, int level) +{ + radeon_mipmap_level *dstlvl = &mt->levels[level - mt->firstLevel]; + unsigned char *dest; + + assert(image->mt != mt); + assert(dstlvl->width == image->base.Width); + assert(dstlvl->height == image->base.Height); + assert(dstlvl->depth == image->base.Depth); + + + radeon_bo_map(mt->bo, GL_TRUE); + dest = mt->bo->ptr + dstlvl->faces[face].offset; + + if (image->mt) { + /* Format etc. should match, so we really just need a memcpy(). + * In fact, that memcpy() could be done by the hardware in many + * cases, provided that we have a proper memory manager. + */ + radeon_mipmap_level *srclvl = &image->mt->levels[image->mtlevel-image->mt->firstLevel]; + + assert(srclvl->size == dstlvl->size); + assert(srclvl->rowstride == dstlvl->rowstride); + + radeon_bo_map(image->mt->bo, GL_FALSE); + + memcpy(dest, + image->mt->bo->ptr + srclvl->faces[face].offset, + dstlvl->size); + radeon_bo_unmap(image->mt->bo); + + radeon_miptree_unreference(image->mt); + } else { + uint32_t srcrowstride; + uint32_t height; + /* need to confirm this value is correct */ + if (mt->compressed) { + height = (image->base.Height + 3) / 4; + srcrowstride = _mesa_compressed_row_stride(image->base.TexFormat->MesaFormat, image->base.Width); + } else { + height = image->base.Height * image->base.Depth; + srcrowstride = image->base.Width * image->base.TexFormat->TexelBytes; + } + +// if (mt->tilebits) +// WARN_ONCE("%s: tiling not supported yet", __FUNCTION__); + + copy_rows(dest, dstlvl->rowstride, image->base.Data, srcrowstride, + height, srcrowstride); + + _mesa_free_texmemory(image->base.Data); + image->base.Data = 0; + } + + radeon_bo_unmap(mt->bo); + + image->mt = mt; + image->mtface = face; + image->mtlevel = level; + radeon_miptree_reference(image->mt); +} + +int radeon_validate_texture_miptree(GLcontext * ctx, struct gl_texture_object *texObj) +{ + radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + radeonTexObj *t = radeon_tex_obj(texObj); + radeon_texture_image *baseimage = get_radeon_texture_image(texObj->Image[0][texObj->BaseLevel]); + int face, level; + + if (t->validated || t->image_override) + return GL_TRUE; + + if (RADEON_DEBUG & RADEON_TEXTURE) + fprintf(stderr, "%s: Validating texture %p now\n", __FUNCTION__, texObj); + + if (baseimage->base.Border > 0) + return GL_FALSE; + + /* Ensure a matching miptree exists. + * + * Differing mipmap trees can result when the app uses TexImage to + * change texture dimensions. + * + * Prefer to use base image's miptree if it + * exists, since that most likely contains more valid data (remember + * that the base level is usually significantly larger than the rest + * of the miptree, so cubemaps are the only possible exception). + */ + if (baseimage->mt && + baseimage->mt != t->mt && + radeon_miptree_matches_texture(baseimage->mt, &t->base)) { + radeon_miptree_unreference(t->mt); + t->mt = baseimage->mt; + radeon_miptree_reference(t->mt); + } else if (t->mt && !radeon_miptree_matches_texture(t->mt, &t->base)) { + radeon_miptree_unreference(t->mt); + t->mt = 0; + } + + if (!t->mt) { + if (RADEON_DEBUG & RADEON_TEXTURE) + fprintf(stderr, " Allocate new miptree\n"); + radeon_try_alloc_miptree(rmesa, t, baseimage, 0, texObj->BaseLevel); + if (!t->mt) { + _mesa_problem(ctx, "radeon_validate_texture failed to alloc miptree"); + return GL_FALSE; + } + } + + /* Ensure all images are stored in the single main miptree */ + for(face = 0; face < t->mt->faces; ++face) { + for(level = t->mt->firstLevel; level <= t->mt->lastLevel; ++level) { + radeon_texture_image *image = get_radeon_texture_image(texObj->Image[face][level]); + if (RADEON_DEBUG & RADEON_TEXTURE) + fprintf(stderr, " face %i, level %i... %p vs %p ", face, level, t->mt, image->mt); + if (t->mt == image->mt) { + if (RADEON_DEBUG & RADEON_TEXTURE) + fprintf(stderr, "OK\n"); + + continue; + } + + if (RADEON_DEBUG & RADEON_TEXTURE) + fprintf(stderr, "migrating\n"); + migrate_image_to_miptree(t->mt, image, face, level); + } + } + + return GL_TRUE; +} + + +/** + * Need to map texture image into memory before copying image data, + * then unmap it. + */ +static void +radeon_get_tex_image(GLcontext * ctx, GLenum target, GLint level, + GLenum format, GLenum type, GLvoid * pixels, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage, int compressed) +{ + radeon_texture_image *image = get_radeon_texture_image(texImage); + + if (image->mt) { + /* Map the texture image read-only */ + radeon_teximage_map(image, GL_FALSE); + } else { + /* Image hasn't been uploaded to a miptree yet */ + assert(image->base.Data); + } + + if (compressed) { + /* FIXME: this can't work for small textures (mips) which + use different hw stride */ + _mesa_get_compressed_teximage(ctx, target, level, pixels, + texObj, texImage); + } else { + _mesa_get_teximage(ctx, target, level, format, type, pixels, + texObj, texImage); + } + + if (image->mt) { + radeon_teximage_unmap(image); + } +} + +void +radeonGetTexImage(GLcontext * ctx, GLenum target, GLint level, + GLenum format, GLenum type, GLvoid * pixels, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + radeon_get_tex_image(ctx, target, level, format, type, pixels, + texObj, texImage, 0); +} + +void +radeonGetCompressedTexImage(GLcontext *ctx, GLenum target, GLint level, + GLvoid *pixels, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + radeon_get_tex_image(ctx, target, level, 0, 0, pixels, + texObj, texImage, 1); +} diff --git a/src/mesa/drivers/dri/radeon/radeon_texture.h b/src/mesa/drivers/dri/radeon/radeon_texture.h new file mode 100644 index 00000000000..888a55ba911 --- /dev/null +++ b/src/mesa/drivers/dri/radeon/radeon_texture.h @@ -0,0 +1,122 @@ +/* + * Copyright (C) 2008 Nicolai Haehnle. + * Copyright (C) The Weather Channel, Inc. 2002. All Rights Reserved. + * + * The Weather Channel (TM) funded Tungsten Graphics to develop the + * initial release of the Radeon 8500 driver under the XFree86 license. + * This notice must be preserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sublicense, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial + * portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + */ + +#ifndef RADEON_TEXTURE_H +#define RADEON_TEXTURE_H +struct gl_texture_image *radeonNewTextureImage(GLcontext *ctx); +void radeonFreeTexImageData(GLcontext *ctx, struct gl_texture_image *timage); + +void radeon_teximage_map(radeon_texture_image *image, GLboolean write_enable); +void radeon_teximage_unmap(radeon_texture_image *image); +void radeonMapTexture(GLcontext *ctx, struct gl_texture_object *texObj); +void radeonUnmapTexture(GLcontext *ctx, struct gl_texture_object *texObj); +void radeonGenerateMipmap(GLcontext* ctx, GLenum target, struct gl_texture_object *texObj); +int radeon_validate_texture_miptree(GLcontext * ctx, struct gl_texture_object *texObj); +GLuint radeon_face_for_target(GLenum target); +const struct gl_texture_format *radeonChooseTextureFormat_mesa(GLcontext * ctx, + GLint internalFormat, + GLenum format, + GLenum type); +const struct gl_texture_format *radeonChooseTextureFormat(GLcontext * ctx, + GLint internalFormat, + GLenum format, + GLenum type, GLboolean fbo); + +void radeonTexImage1D(GLcontext * ctx, GLenum target, GLint level, + GLint internalFormat, + GLint width, GLint border, + GLenum format, GLenum type, const GLvoid * pixels, + const struct gl_pixelstore_attrib *packing, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage); +void radeonTexImage2D(GLcontext * ctx, GLenum target, GLint level, + GLint internalFormat, + GLint width, GLint height, GLint border, + GLenum format, GLenum type, const GLvoid * pixels, + const struct gl_pixelstore_attrib *packing, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage); +void radeonCompressedTexImage2D(GLcontext * ctx, GLenum target, + GLint level, GLint internalFormat, + GLint width, GLint height, GLint border, + GLsizei imageSize, const GLvoid * data, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage); +void radeonTexImage3D(GLcontext * ctx, GLenum target, GLint level, + GLint internalFormat, + GLint width, GLint height, GLint depth, + GLint border, + GLenum format, GLenum type, const GLvoid * pixels, + const struct gl_pixelstore_attrib *packing, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage); +void radeonTexSubImage1D(GLcontext * ctx, GLenum target, GLint level, + GLint xoffset, + GLsizei width, + GLenum format, GLenum type, + const GLvoid * pixels, + const struct gl_pixelstore_attrib *packing, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage); +void radeonTexSubImage2D(GLcontext * ctx, GLenum target, GLint level, + GLint xoffset, GLint yoffset, + GLsizei width, GLsizei height, + GLenum format, GLenum type, + const GLvoid * pixels, + const struct gl_pixelstore_attrib *packing, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage); +void radeonCompressedTexSubImage2D(GLcontext * ctx, GLenum target, + GLint level, GLint xoffset, + GLint yoffset, GLsizei width, + GLsizei height, GLenum format, + GLsizei imageSize, const GLvoid * data, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage); + +void radeonTexSubImage3D(GLcontext * ctx, GLenum target, GLint level, + GLint xoffset, GLint yoffset, GLint zoffset, + GLsizei width, GLsizei height, GLsizei depth, + GLenum format, GLenum type, + const GLvoid * pixels, + const struct gl_pixelstore_attrib *packing, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage); + +void radeonGetTexImage(GLcontext * ctx, GLenum target, GLint level, + GLenum format, GLenum type, GLvoid * pixels, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage); +void radeonGetCompressedTexImage(GLcontext *ctx, GLenum target, GLint level, + GLvoid *pixels, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage); + +#endif diff --git a/src/mesa/drivers/dri/radeon/server/radeon_reg.h b/src/mesa/drivers/dri/radeon/server/radeon_reg.h index ae2ccdf292a..e81d7fdcd0e 100644 --- a/src/mesa/drivers/dri/radeon/server/radeon_reg.h +++ b/src/mesa/drivers/dri/radeon/server/radeon_reg.h @@ -1601,6 +1601,8 @@ # define RADEON_STENCIL_VALUE_MASK (0xff << 16) # define RADEON_STENCIL_WRITEMASK_SHIFT 24 # define RADEON_STENCIL_WRITE_MASK (0xff << 24) +#define RADEON_RB3D_ZPASS_DATA 0x3290 +#define RADEON_RB3D_ZPASS_ADDR 0x3294 #define RADEON_RB3D_ZSTENCILCNTL 0x1c2c # define RADEON_DEPTH_FORMAT_MASK (0xf << 0) # define RADEON_DEPTH_FORMAT_16BIT_INT_Z (0 << 0) @@ -1661,6 +1663,9 @@ # define RADEON_FORCE_Z_DIRTY (1 << 29) # define RADEON_Z_WRITE_ENABLE (1 << 30) # define RADEON_Z_DECOMPRESSION_ENABLE (1 << 31) + +#define RADEON_RE_STIPPLE_ADDR 0x1cc8 +#define RADEON_RE_STIPPLE_DATA 0x1ccc #define RADEON_RE_LINE_PATTERN 0x1cd0 # define RADEON_LINE_PATTERN_MASK 0x0000ffff # define RADEON_LINE_REPEAT_COUNT_SHIFT 16 @@ -2031,6 +2036,9 @@ #define RADEON_CP_PACKET3_3D_DRAW_INDX 0xC0002A00 #define RADEON_CP_PACKET3_LOAD_PALETTE 0xC0002C00 #define RADEON_CP_PACKET3_3D_LOAD_VBPNTR 0xC0002F00 +#define R200_CP_CMD_3D_DRAW_VBUF_2 0xC0003400 +#define R200_CP_CMD_3D_DRAW_IMMD_2 0xC0003500 +#define R200_CP_CMD_3D_DRAW_INDX_2 0xC0003600 #define RADEON_CP_PACKET3_CNTL_PAINT 0xC0009100 #define RADEON_CP_PACKET3_CNTL_BITBLT 0xC0009200 #define RADEON_CP_PACKET3_CNTL_SMALLTEXT 0xC0009300 |