ilo: make ilo_cp based on ilo_builder

This makes ilo_cp use the builder to manage batch buffers, and use
ilo_builder_decode() to replace ilo_3d_pipeline_dump().

16 files changed, 114 insertions, 979 deletions

diff --git a/src/gallium/drivers/ilo/Makefile.sources b/src/gallium/drivers/ilo/Makefile.sources
index 7f0cf75938d..dbcbfb9fee7 100644
--- a/src/gallium/drivers/ilo/Makefile.sources
+++ b/src/gallium/drivers/ilo/Makefile.sources
@@ -2,7 +2,6 @@ C_SOURCES := \
 	ilo_3d.c \
 	ilo_3d.h \
 	ilo_3d_pipeline.c \
-	ilo_3d_pipeline_dump.c \
 	ilo_3d_pipeline_gen6.c \
 	ilo_3d_pipeline_gen6.h \
 	ilo_3d_pipeline_gen7.c \
diff --git a/src/gallium/drivers/ilo/ilo_3d.c b/src/gallium/drivers/ilo/ilo_3d.c
index 48984ba006a..e6c72d35484 100644
--- a/src/gallium/drivers/ilo/ilo_3d.c
+++ b/src/gallium/drivers/ilo/ilo_3d.c
@@ -396,7 +396,7 @@ void
 ilo_3d_cp_flushed(struct ilo_3d *hw3d)
 {
    if (ilo_debug & ILO_DEBUG_3D)
-      ilo_3d_pipeline_dump(hw3d->pipeline);
+      ilo_builder_decode(&hw3d->cp->builder);
 
    /* invalidate the pipeline */
    ilo_3d_pipeline_invalidate(hw3d->pipeline,
diff --git a/src/gallium/drivers/ilo/ilo_3d_pipeline.c b/src/gallium/drivers/ilo/ilo_3d_pipeline.c
index 1152a7ffc3e..bea5406404f 100644
--- a/src/gallium/drivers/ilo/ilo_3d_pipeline.c
+++ b/src/gallium/drivers/ilo/ilo_3d_pipeline.c
@@ -172,10 +172,10 @@ ilo_3d_pipeline_emit_draw(struct ilo_3d_pipeline *p,
 
 
    while (true) {
-      struct ilo_cp_jmp_buf jmp;
+      struct ilo_builder_snapshot snapshot;
 
       /* we will rewind if aperture check below fails */
-      ilo_cp_setjmp(p->cp, &jmp);
+      ilo_builder_batch_snapshot(&p->cp->builder, &snapshot);
 
       handle_invalid_batch_bo(p, false);
 
@@ -184,13 +184,13 @@ ilo_3d_pipeline_emit_draw(struct ilo_3d_pipeline *p,
       p->emit_draw(p, ilo);
       ilo_cp_assert_no_implicit_flush(p->cp, false);
 
-      if (intel_winsys_can_submit_bo(ilo->winsys, &p->cp->bo, 1)) {
+      if (ilo_builder_validate(&ilo->cp->builder, 0, NULL)) {
          success = true;
          break;
       }
 
       /* rewind */
-      ilo_cp_longjmp(p->cp, &jmp);
+      ilo_builder_batch_restore(&p->cp->builder, &snapshot);
 
       if (ilo_cp_empty(p->cp)) {
          success = false;
@@ -279,10 +279,10 @@ ilo_3d_pipeline_emit_rectlist(struct ilo_3d_pipeline *p,
       ilo_cp_flush(p->cp, "out of space");
 
    while (true) {
-      struct ilo_cp_jmp_buf jmp;
+      struct ilo_builder_snapshot snapshot;
 
       /* we will rewind if aperture check below fails */
-      ilo_cp_setjmp(p->cp, &jmp);
+      ilo_builder_batch_snapshot(&p->cp->builder, &snapshot);
 
       handle_invalid_batch_bo(p, false);
 
@@ -290,9 +290,9 @@ ilo_3d_pipeline_emit_rectlist(struct ilo_3d_pipeline *p,
       p->emit_rectlist(p, blitter);
       ilo_cp_assert_no_implicit_flush(p->cp, false);
 
-      if (!intel_winsys_can_submit_bo(blitter->ilo->winsys, &p->cp->bo, 1)) {
+      if (!ilo_builder_validate(&p->cp->builder, 0, NULL)) {
          /* rewind */
-         ilo_cp_longjmp(p->cp, &jmp);
+         ilo_builder_batch_restore(&p->cp->builder, &snapshot);
 
          /* flush and try again */
          if (!ilo_cp_empty(p->cp)) {
diff --git a/src/gallium/drivers/ilo/ilo_3d_pipeline.h b/src/gallium/drivers/ilo/ilo_3d_pipeline.h
index 90c626e5274..d5771c49641 100644
--- a/src/gallium/drivers/ilo/ilo_3d_pipeline.h
+++ b/src/gallium/drivers/ilo/ilo_3d_pipeline.h
@@ -194,7 +194,4 @@ ilo_3d_pipeline_get_sample_position(struct ilo_3d_pipeline *p,
                                     unsigned sample_index,
                                     float *x, float *y);
 
-void
-ilo_3d_pipeline_dump(struct ilo_3d_pipeline *p);
-
 #endif /* ILO_3D_PIPELINE_H */
diff --git a/src/gallium/drivers/ilo/ilo_3d_pipeline_dump.c b/src/gallium/drivers/ilo/ilo_3d_pipeline_dump.c
deleted file mode 100644
index de47e8d4558..00000000000
--- a/src/gallium/drivers/ilo/ilo_3d_pipeline_dump.c
+++ /dev/null
@@ -1,648 +0,0 @@
-/*
- * Copyright © 2007 Intel Corporation
- * Copyright (C) 2012-2013 LunarG, Inc.
- *
- * 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:
- *    Eric Anholt <eric@anholt.net>
- *    Chia-I Wu <olv@lunarg.com>
- */
-
-#include "intel_winsys.h"
-
-#include "ilo_cp.h"
-#include "ilo_3d_pipeline.h"
-
-#define PRINTFLIKE(f, a) _util_printf_format(f, a)
-typedef short GLshort;
-typedef int GLint;
-typedef unsigned char GLubyte;
-typedef unsigned int GLuint;
-typedef float GLfloat;
-#include <stdint.h>
-#include <stdarg.h>
-#include <stdio.h>
-#include "brw_structs.h"
-#include "brw_defines.h"
-
-struct intel_context {
-   int gen;
-
-   struct {
-      struct {
-         void *virtual;
-      } *bo, bo_dst;
-   } batch;
-};
-
-struct brw_context {
-   struct intel_context intel;
-};
-
-static void
-batch_out(struct brw_context *brw, const char *name, uint32_t offset,
-	  int index, char *fmt, ...) PRINTFLIKE(5, 6);
-
-static void
-batch_out(struct brw_context *brw, const char *name, uint32_t offset,
-	  int index, char *fmt, ...)
-{
-   struct intel_context *intel = &brw->intel;
-   uint32_t *data = intel->batch.bo->virtual + offset;
-   va_list va;
-
-   fprintf(stderr, "0x%08x:      0x%08x: %8s: ",
-	   offset + index * 4, data[index], name);
-   va_start(va, fmt);
-   vfprintf(stderr, fmt, va);
-   va_end(va);
-}
-
-static const char *
-get_965_surfacetype(unsigned int surfacetype)
-{
-    switch (surfacetype) {
-    case 0: return "1D";
-    case 1: return "2D";
-    case 2: return "3D";
-    case 3: return "CUBE";
-    case 4: return "BUFFER";
-    case 7: return "NULL";
-    default: return "unknown";
-    }
-}
-
-static const char *
-get_965_surface_format(unsigned int surface_format)
-{
-    switch (surface_format) {
-    case 0x000: return "r32g32b32a32_float";
-    case 0x0c1: return "b8g8r8a8_unorm";
-    case 0x100: return "b5g6r5_unorm";
-    case 0x102: return "b5g5r5a1_unorm";
-    case 0x104: return "b4g4r4a4_unorm";
-    default: return "unknown";
-    }
-}
-
-static void dump_vs_state(struct brw_context *brw, uint32_t offset)
-{
-   struct intel_context *intel = &brw->intel;
-   const char *name = "VS_STATE";
-   struct brw_vs_unit_state *vs = intel->batch.bo->virtual + offset;
-
-   batch_out(brw, name, offset, 0, "thread0\n");
-   batch_out(brw, name, offset, 1, "thread1\n");
-   batch_out(brw, name, offset, 2, "thread2\n");
-   batch_out(brw, name, offset, 3, "thread3\n");
-   batch_out(brw, name, offset, 4, "thread4: %d threads\n",
-	     vs->thread4.max_threads + 1);
-   batch_out(brw, name, offset, 5, "vs5\n");
-   batch_out(brw, name, offset, 6, "vs6\n");
-}
-
-static void dump_gs_state(struct brw_context *brw, uint32_t offset)
-{
-   struct intel_context *intel = &brw->intel;
-   const char *name = "GS_STATE";
-   struct brw_gs_unit_state *gs = intel->batch.bo->virtual + offset;
-
-   batch_out(brw, name, offset, 0, "thread0\n");
-   batch_out(brw, name, offset, 1, "thread1\n");
-   batch_out(brw, name, offset, 2, "thread2\n");
-   batch_out(brw, name, offset, 3, "thread3\n");
-   batch_out(brw, name, offset, 4, "thread4: %d threads\n",
-	     gs->thread4.max_threads + 1);
-   batch_out(brw, name, offset, 5, "vs5\n");
-   batch_out(brw, name, offset, 6, "vs6\n");
-}
-
-static void dump_clip_state(struct brw_context *brw, uint32_t offset)
-{
-   struct intel_context *intel = &brw->intel;
-   const char *name = "CLIP_STATE";
-   struct brw_clip_unit_state *clip = intel->batch.bo->virtual + offset;
-
-   batch_out(brw, name, offset, 0, "thread0\n");
-   batch_out(brw, name, offset, 1, "thread1\n");
-   batch_out(brw, name, offset, 2, "thread2\n");
-   batch_out(brw, name, offset, 3, "thread3\n");
-   batch_out(brw, name, offset, 4, "thread4: %d threads\n",
-	     clip->thread4.max_threads + 1);
-   batch_out(brw, name, offset, 5, "clip5\n");
-   batch_out(brw, name, offset, 6, "clip6\n");
-   batch_out(brw, name, offset, 7, "vp xmin %f\n", clip->viewport_xmin);
-   batch_out(brw, name, offset, 8, "vp xmax %f\n", clip->viewport_xmax);
-   batch_out(brw, name, offset, 9, "vp ymin %f\n", clip->viewport_ymin);
-   batch_out(brw, name, offset, 10, "vp ymax %f\n", clip->viewport_ymax);
-}
-
-static void dump_sf_state(struct brw_context *brw, uint32_t offset)
-{
-   struct intel_context *intel = &brw->intel;
-   const char *name = "SF_STATE";
-   struct brw_sf_unit_state *sf = intel->batch.bo->virtual + offset;
-
-   batch_out(brw, name, offset, 0, "thread0\n");
-   batch_out(brw, name, offset, 1, "thread1\n");
-   batch_out(brw, name, offset, 2, "thread2\n");
-   batch_out(brw, name, offset, 3, "thread3\n");
-   batch_out(brw, name, offset, 4, "thread4: %d threads\n",
-	     sf->thread4.max_threads + 1);
-   batch_out(brw, name, offset, 5, "sf5: viewport offset\n");
-   batch_out(brw, name, offset, 6, "sf6\n");
-   batch_out(brw, name, offset, 7, "sf7\n");
-}
-
-static void dump_wm_state(struct brw_context *brw, uint32_t offset)
-{
-   struct intel_context *intel = &brw->intel;
-   const char *name = "WM_STATE";
-   struct brw_wm_unit_state *wm = intel->batch.bo->virtual + offset;
-
-   batch_out(brw, name, offset, 0, "thread0\n");
-   batch_out(brw, name, offset, 1, "thread1\n");
-   batch_out(brw, name, offset, 2, "thread2\n");
-   batch_out(brw, name, offset, 3, "thread3\n");
-   batch_out(brw, name, offset, 4, "wm4\n");
-   batch_out(brw, name, offset, 5, "wm5: %s%s%s%s%s%s, %d threads\n",
-	     wm->wm5.enable_8_pix ? "8pix" : "",
-	     wm->wm5.enable_16_pix ? "16pix" : "",
-	     wm->wm5.program_uses_depth ? ", uses depth" : "",
-	     wm->wm5.program_computes_depth ? ", computes depth" : "",
-	     wm->wm5.program_uses_killpixel ? ", kills" : "",
-	     wm->wm5.thread_dispatch_enable ? "" : ", no dispatch",
-	     wm->wm5.max_threads + 1);
-   batch_out(brw, name, offset, 6, "depth offset constant %f\n",
-	     wm->global_depth_offset_constant);
-   batch_out(brw, name, offset, 7, "depth offset scale %f\n",
-	     wm->global_depth_offset_scale);
-   batch_out(brw, name, offset, 8, "wm8: kernel 1 (gen5+)\n");
-   batch_out(brw, name, offset, 9, "wm9: kernel 2 (gen5+)\n");
-   batch_out(brw, name, offset, 10, "wm10: kernel 3 (gen5+)\n");
-}
-
-static void dump_surface_state(struct brw_context *brw, uint32_t offset)
-{
-   const char *name = "SURF";
-   uint32_t *surf = brw->intel.batch.bo->virtual + offset;
-
-   batch_out(brw, name, offset, 0, "%s %s\n",
-	     get_965_surfacetype(GET_FIELD(surf[0], BRW_SURFACE_TYPE)),
-	     get_965_surface_format(GET_FIELD(surf[0], BRW_SURFACE_FORMAT)));
-   batch_out(brw, name, offset, 1, "offset\n");
-   batch_out(brw, name, offset, 2, "%dx%d size, %d mips\n",
-	     GET_FIELD(surf[2], BRW_SURFACE_WIDTH) + 1,
-	     GET_FIELD(surf[2], BRW_SURFACE_HEIGHT) + 1,
-	     GET_FIELD(surf[2], BRW_SURFACE_LOD));
-   batch_out(brw, name, offset, 3, "pitch %d, %s tiled\n",
-	     GET_FIELD(surf[3], BRW_SURFACE_PITCH) + 1,
-	     (surf[3] & BRW_SURFACE_TILED) ?
-	     ((surf[3] & BRW_SURFACE_TILED_Y) ? "Y" : "X") : "not");
-   batch_out(brw, name, offset, 4, "mip base %d\n",
-	     GET_FIELD(surf[4], BRW_SURFACE_MIN_LOD));
-   batch_out(brw, name, offset, 5, "x,y offset: %d,%d\n",
-	     GET_FIELD(surf[5], BRW_SURFACE_X_OFFSET),
-	     GET_FIELD(surf[5], BRW_SURFACE_Y_OFFSET));
-}
-
-static void dump_gen7_surface_state(struct brw_context *brw, uint32_t offset)
-{
-   const char *name = "SURF";
-   uint32_t *surf = brw->intel.batch.bo->virtual + offset;
-
-   batch_out(brw, name, offset, 0, "%s %s\n",
-             get_965_surfacetype(GET_FIELD(surf[0], BRW_SURFACE_TYPE)),
-             get_965_surface_format(GET_FIELD(surf[0], BRW_SURFACE_FORMAT)));
-   batch_out(brw, name, offset, 1, "offset\n");
-   batch_out(brw, name, offset, 2, "%dx%d size, %d mips\n",
-             GET_FIELD(surf[2], GEN7_SURFACE_WIDTH) + 1,
-             GET_FIELD(surf[2], GEN7_SURFACE_HEIGHT) + 1,
-             surf[5] & INTEL_MASK(3, 0));
-   batch_out(brw, name, offset, 3, "pitch %d, %stiled\n",
-	     (surf[3] & INTEL_MASK(17, 0)) + 1,
-             (surf[0] & (1 << 14)) ? "" : "not ");
-   batch_out(brw, name, offset, 4, "mip base %d\n",
-             GET_FIELD(surf[5], GEN7_SURFACE_MIN_LOD));
-   batch_out(brw, name, offset, 5, "x,y offset: %d,%d\n",
-             GET_FIELD(surf[5], BRW_SURFACE_X_OFFSET),
-             GET_FIELD(surf[5], BRW_SURFACE_Y_OFFSET));
-}
-
-static void
-dump_sdc(struct brw_context *brw, uint32_t offset)
-{
-   const char *name = "SDC";
-   struct intel_context *intel = &brw->intel;
-
-   if (intel->gen >= 5 && intel->gen <= 6) {
-      struct gen5_sampler_default_color *sdc = (intel->batch.bo->virtual +
-						offset);
-      batch_out(brw, name, offset, 0, "unorm rgba\n");
-      batch_out(brw, name, offset, 1, "r %f\n", sdc->f[0]);
-      batch_out(brw, name, offset, 2, "b %f\n", sdc->f[1]);
-      batch_out(brw, name, offset, 3, "g %f\n", sdc->f[2]);
-      batch_out(brw, name, offset, 4, "a %f\n", sdc->f[3]);
-      batch_out(brw, name, offset, 5, "half float rg\n");
-      batch_out(brw, name, offset, 6, "half float ba\n");
-      batch_out(brw, name, offset, 7, "u16 rg\n");
-      batch_out(brw, name, offset, 8, "u16 ba\n");
-      batch_out(brw, name, offset, 9, "s16 rg\n");
-      batch_out(brw, name, offset, 10, "s16 ba\n");
-      batch_out(brw, name, offset, 11, "s8 rgba\n");
-   } else {
-      struct brw_sampler_default_color *sdc = (intel->batch.bo->virtual +
-					       offset);
-      batch_out(brw, name, offset, 0, "r %f\n", sdc->color[0]);
-      batch_out(brw, name, offset, 1, "g %f\n", sdc->color[1]);
-      batch_out(brw, name, offset, 2, "b %f\n", sdc->color[2]);
-      batch_out(brw, name, offset, 3, "a %f\n", sdc->color[3]);
-   }
-}
-
-static void dump_sampler_state(struct brw_context *brw,
-			       uint32_t offset, uint32_t size)
-{
-   struct intel_context *intel = &brw->intel;
-   int i;
-   struct brw_sampler_state *samp = intel->batch.bo->virtual + offset;
-
-   assert(intel->gen < 7);
-
-   for (i = 0; i < size / sizeof(*samp); i++) {
-      char name[20];
-
-      sprintf(name, "WM SAMP%d", i);
-      batch_out(brw, name, offset, 0, "filtering\n");
-      batch_out(brw, name, offset, 1, "wrapping, lod\n");
-      batch_out(brw, name, offset, 2, "default color pointer\n");
-      batch_out(brw, name, offset, 3, "chroma key, aniso\n");
-
-      samp++;
-      offset += sizeof(*samp);
-   }
-}
-
-static void dump_gen7_sampler_state(struct brw_context *brw,
-				    uint32_t offset, uint32_t size)
-{
-   struct intel_context *intel = &brw->intel;
-   struct gen7_sampler_state *samp = intel->batch.bo->virtual + offset;
-   int i;
-
-   assert(intel->gen >= 7);
-
-   for (i = 0; i < size / sizeof(*samp); i++) {
-      char name[20];
-
-      sprintf(name, "WM SAMP%d", i);
-      batch_out(brw, name, offset, 0, "filtering\n");
-      batch_out(brw, name, offset, 1, "wrapping, lod\n");
-      batch_out(brw, name, offset, 2, "default color pointer\n");
-      batch_out(brw, name, offset, 3, "chroma key, aniso\n");
-
-      samp++;
-      offset += sizeof(*samp);
-   }
-}
-
-
-static void dump_sf_viewport_state(struct brw_context *brw,
-				   uint32_t offset)
-{
-   struct intel_context *intel = &brw->intel;
-   const char *name = "SF VP";
-   struct brw_sf_viewport *vp = intel->batch.bo->virtual + offset;
-
-   assert(intel->gen < 7);
-
-   batch_out(brw, name, offset, 0, "m00 = %f\n", vp->viewport.m00);
-   batch_out(brw, name, offset, 1, "m11 = %f\n", vp->viewport.m11);
-   batch_out(brw, name, offset, 2, "m22 = %f\n", vp->viewport.m22);
-   batch_out(brw, name, offset, 3, "m30 = %f\n", vp->viewport.m30);
-   batch_out(brw, name, offset, 4, "m31 = %f\n", vp->viewport.m31);
-   batch_out(brw, name, offset, 5, "m32 = %f\n", vp->viewport.m32);
-
-   batch_out(brw, name, offset, 6, "top left = %d,%d\n",
-	     vp->scissor.xmin, vp->scissor.ymin);
-   batch_out(brw, name, offset, 7, "bottom right = %d,%d\n",
-	     vp->scissor.xmax, vp->scissor.ymax);
-}
-
-static void dump_clip_viewport_state(struct brw_context *brw,
-				     uint32_t offset)
-{
-   struct intel_context *intel = &brw->intel;
-   const char *name = "CLIP VP";
-   struct brw_clipper_viewport *vp = intel->batch.bo->virtual + offset;
-
-   assert(intel->gen < 7);
-
-   batch_out(brw, name, offset, 0, "xmin = %f\n", vp->xmin);
-   batch_out(brw, name, offset, 1, "xmax = %f\n", vp->xmax);
-   batch_out(brw, name, offset, 2, "ymin = %f\n", vp->ymin);
-   batch_out(brw, name, offset, 3, "ymax = %f\n", vp->ymax);
-}
-
-static void dump_sf_clip_viewport_state(struct brw_context *brw,
-					uint32_t offset)
-{
-   struct intel_context *intel = &brw->intel;
-   const char *name = "SF_CLIP VP";
-   struct gen7_sf_clip_viewport *vp = intel->batch.bo->virtual + offset;
-
-   assert(intel->gen >= 7);
-
-   batch_out(brw, name, offset, 0, "m00 = %f\n", vp->viewport.m00);
-   batch_out(brw, name, offset, 1, "m11 = %f\n", vp->viewport.m11);
-   batch_out(brw, name, offset, 2, "m22 = %f\n", vp->viewport.m22);
-   batch_out(brw, name, offset, 3, "m30 = %f\n", vp->viewport.m30);
-   batch_out(brw, name, offset, 4, "m31 = %f\n", vp->viewport.m31);
-   batch_out(brw, name, offset, 5, "m32 = %f\n", vp->viewport.m32);
-   batch_out(brw, name, offset, 6, "guardband xmin = %f\n", vp->guardband.xmin);
-   batch_out(brw, name, offset, 7, "guardband xmax = %f\n", vp->guardband.xmax);
-   batch_out(brw, name, offset, 8, "guardband ymin = %f\n", vp->guardband.ymin);
-   batch_out(brw, name, offset, 9, "guardband ymax = %f\n", vp->guardband.ymax);
-}
-
-
-static void dump_cc_viewport_state(struct brw_context *brw, uint32_t offset)
-{
-   const char *name = "CC VP";
-   struct brw_cc_viewport *vp = brw->intel.batch.bo->virtual + offset;
-
-   batch_out(brw, name, offset, 0, "min_depth = %f\n", vp->min_depth);
-   batch_out(brw, name, offset, 1, "max_depth = %f\n", vp->max_depth);
-}
-
-static void dump_depth_stencil_state(struct brw_context *brw, uint32_t offset)
-{
-   const char *name = "D_S";
-   struct gen6_depth_stencil_state *ds = brw->intel.batch.bo->virtual + offset;
-
-   batch_out(brw, name, offset, 0,
-	     "stencil %sable, func %d, write %sable\n",
-	     ds->ds0.stencil_enable ? "en" : "dis",
-	     ds->ds0.stencil_func,
-	     ds->ds0.stencil_write_enable ? "en" : "dis");
-   batch_out(brw, name, offset, 1,
-	     "stencil test mask 0x%x, write mask 0x%x\n",
-	     ds->ds1.stencil_test_mask, ds->ds1.stencil_write_mask);
-   batch_out(brw, name, offset, 2,
-	     "depth test %sable, func %d, write %sable\n",
-	     ds->ds2.depth_test_enable ? "en" : "dis",
-	     ds->ds2.depth_test_func,
-	     ds->ds2.depth_write_enable ? "en" : "dis");
-}
-
-static void dump_cc_state_gen4(struct brw_context *brw, uint32_t offset)
-{
-   const char *name = "CC";
-
-   batch_out(brw, name, offset, 0, "cc0\n");
-   batch_out(brw, name, offset, 1, "cc1\n");
-   batch_out(brw, name, offset, 2, "cc2\n");
-   batch_out(brw, name, offset, 3, "cc3\n");
-   batch_out(brw, name, offset, 4, "cc4: viewport offset\n");
-   batch_out(brw, name, offset, 5, "cc5\n");
-   batch_out(brw, name, offset, 6, "cc6\n");
-   batch_out(brw, name, offset, 7, "cc7\n");
-}
-
-static void dump_cc_state_gen6(struct brw_context *brw, uint32_t offset)
-{
-   const char *name = "CC";
-   struct gen6_color_calc_state *cc = brw->intel.batch.bo->virtual + offset;
-
-   batch_out(brw, name, offset, 0,
-	     "alpha test format %s, round disable %d, stencil ref %d, "
-	     "bf stencil ref %d\n",
-	     cc->cc0.alpha_test_format ? "FLOAT32" : "UNORM8",
-	     cc->cc0.round_disable,
-	     cc->cc0.stencil_ref,
-	     cc->cc0.bf_stencil_ref);
-   batch_out(brw, name, offset, 1, "\n");
-   batch_out(brw, name, offset, 2, "constant red %f\n", cc->constant_r);
-   batch_out(brw, name, offset, 3, "constant green %f\n", cc->constant_g);
-   batch_out(brw, name, offset, 4, "constant blue %f\n", cc->constant_b);
-   batch_out(brw, name, offset, 5, "constant alpha %f\n", cc->constant_a);
-}
-
-static void dump_blend_state(struct brw_context *brw, uint32_t offset)
-{
-   const char *name = "BLEND";
-
-   batch_out(brw, name, offset, 0, "\n");
-   batch_out(brw, name, offset, 1, "\n");
-}
-
-static void
-dump_scissor(struct brw_context *brw, uint32_t offset)
-{
-   const char *name = "SCISSOR";
-   struct intel_context *intel = &brw->intel;
-   struct gen6_scissor_rect *scissor = intel->batch.bo->virtual + offset;
-
-   batch_out(brw, name, offset, 0, "xmin %d, ymin %d\n",
-	     scissor->xmin, scissor->ymin);
-   batch_out(brw, name, offset, 1, "xmax %d, ymax %d\n",
-	     scissor->xmax, scissor->ymax);
-}
-
-static void
-dump_vs_constants(struct brw_context *brw, uint32_t offset, uint32_t size)
-{
-   const char *name = "VS_CONST";
-   struct intel_context *intel = &brw->intel;
-   uint32_t *as_uint = intel->batch.bo->virtual + offset;
-   float *as_float = intel->batch.bo->virtual + offset;
-   int i;
-
-   for (i = 0; i < size / 4; i += 4) {
-      batch_out(brw, name, offset, i, "%3d: (% f % f % f % f) (0x%08x 0x%08x 0x%08x 0x%08x)\n",
-		i / 4,
-		as_float[i], as_float[i + 1], as_float[i + 2], as_float[i + 3],
-		as_uint[i], as_uint[i + 1], as_uint[i + 2], as_uint[i + 3]);
-   }
-}
-
-static void
-dump_wm_constants(struct brw_context *brw, uint32_t offset, uint32_t size)
-{
-   const char *name = "WM_CONST";
-   struct intel_context *intel = &brw->intel;
-   uint32_t *as_uint = intel->batch.bo->virtual + offset;
-   float *as_float = intel->batch.bo->virtual + offset;
-   int i;
-
-   for (i = 0; i < size / 4; i += 4) {
-      batch_out(brw, name, offset, i, "%3d: (% f % f % f % f) (0x%08x 0x%08x 0x%08x 0x%08x)\n",
-		i / 4,
-		as_float[i], as_float[i + 1], as_float[i + 2], as_float[i + 3],
-		as_uint[i], as_uint[i + 1], as_uint[i + 2], as_uint[i + 3]);
-   }
-}
-
-static void dump_binding_table(struct brw_context *brw, uint32_t offset,
-			       uint32_t size)
-{
-   char name[20];
-   int i;
-   uint32_t *data = brw->intel.batch.bo->virtual + offset;
-
-   for (i = 0; i < size / 4; i++) {
-      if (data[i] == 0)
-	 continue;
-
-      sprintf(name, "BIND%d", i);
-      batch_out(brw, name, offset, i, "surface state address\n");
-   }
-}
-
-static bool
-init_brw(struct brw_context *brw, struct ilo_3d_pipeline *p)
-{
-   brw->intel.gen = ILO_GEN_GET_MAJOR(p->dev->gen);
-   brw->intel.batch.bo = &brw->intel.batch.bo_dst;
-
-   brw->intel.batch.bo_dst.virtual = intel_bo_map(p->cp->bo, false);
-   if (!brw->intel.batch.bo_dst.virtual)
-      return false;
-
-   return true;
-}
-
-static void
-dump_3d_state(struct ilo_3d_pipeline *p)
-{
-   struct brw_context brw;
-   int num_states, i;
-
-   if (!init_brw(&brw, p))
-      return;
-
-   if (brw.intel.gen >= 7) {
-      dump_cc_viewport_state(&brw, p->state.CC_VIEWPORT);
-      dump_sf_clip_viewport_state(&brw, p->state.SF_CLIP_VIEWPORT);
-   }
-   else {
-      dump_clip_viewport_state(&brw, p->state.CLIP_VIEWPORT);
-      dump_sf_viewport_state(&brw, p->state.SF_VIEWPORT);
-      dump_cc_viewport_state(&brw, p->state.CC_VIEWPORT);
-   }
-
-   dump_blend_state(&brw, p->state.BLEND_STATE);
-   dump_cc_state_gen6(&brw, p->state.COLOR_CALC_STATE);
-   dump_depth_stencil_state(&brw, p->state.DEPTH_STENCIL_STATE);
-
-   /* VS */
-   num_states = p->state.vs.BINDING_TABLE_STATE_size;
-   for (i = 0; i < num_states; i++) {
-      if (brw.intel.gen < 7)
-         dump_surface_state(&brw, p->state.vs.SURFACE_STATE[i]);
-      else
-         dump_gen7_surface_state(&brw, p->state.vs.SURFACE_STATE[i]);
-   }
-   dump_binding_table(&brw, p->state.vs.BINDING_TABLE_STATE, num_states * 4);
-
-   num_states = 0;
-   for (i = 0; i < Elements(p->state.vs.SAMPLER_BORDER_COLOR_STATE); i++) {
-      if (!p->state.vs.SAMPLER_BORDER_COLOR_STATE[i])
-         continue;
-
-      dump_sdc(&brw, p->state.vs.SAMPLER_BORDER_COLOR_STATE[i]);
-      num_states++;
-   }
-   if (brw.intel.gen < 7)
-      dump_sampler_state(&brw, p->state.vs.SAMPLER_STATE, num_states * 16);
-   else
-      dump_gen7_sampler_state(&brw, p->state.vs.SAMPLER_STATE, num_states * 16);
-
-   if (p->state.vs.PUSH_CONSTANT_BUFFER_size) {
-      dump_vs_constants(&brw, p->state.vs.PUSH_CONSTANT_BUFFER,
-            p->state.vs.PUSH_CONSTANT_BUFFER_size);
-   }
-
-   /* GS */
-   num_states = p->state.gs.BINDING_TABLE_STATE_size;
-   for (i = 0; i < num_states; i++) {
-      if (!p->state.gs.SURFACE_STATE[i])
-         continue;
-
-      if (brw.intel.gen < 7)
-         dump_surface_state(&brw, p->state.gs.SURFACE_STATE[i]);
-      else
-         dump_gen7_surface_state(&brw, p->state.gs.SURFACE_STATE[i]);
-   }
-   dump_binding_table(&brw, p->state.gs.BINDING_TABLE_STATE, num_states * 4);
-
-   /* WM */
-   num_states = p->state.wm.BINDING_TABLE_STATE_size;
-   for (i = 0; i < num_states; i++) {
-      if (!p->state.wm.SURFACE_STATE[i])
-         continue;
-
-      if (brw.intel.gen < 7)
-         dump_surface_state(&brw, p->state.wm.SURFACE_STATE[i]);
-      else
-         dump_gen7_surface_state(&brw, p->state.wm.SURFACE_STATE[i]);
-   }
-   dump_binding_table(&brw, p->state.wm.BINDING_TABLE_STATE, num_states * 4);
-
-   num_states = 0;
-   for (i = 0; i < Elements(p->state.wm.SAMPLER_BORDER_COLOR_STATE); i++) {
-      if (!p->state.wm.SAMPLER_BORDER_COLOR_STATE[i])
-         continue;
-
-      dump_sdc(&brw, p->state.wm.SAMPLER_BORDER_COLOR_STATE[i]);
-      num_states++;
-   }
-   if (brw.intel.gen < 7)
-      dump_sampler_state(&brw, p->state.wm.SAMPLER_STATE, num_states * 16);
-   else
-      dump_gen7_sampler_state(&brw, p->state.wm.SAMPLER_STATE, num_states * 16);
-
-   if (p->state.wm.PUSH_CONSTANT_BUFFER_size) {
-      dump_wm_constants(&brw, p->state.wm.PUSH_CONSTANT_BUFFER,
-            p->state.wm.PUSH_CONSTANT_BUFFER_size);
-   }
-
-   dump_scissor(&brw, p->state.SCISSOR_RECT);
-
-   (void) dump_vs_state;
-   (void) dump_gs_state;
-   (void) dump_clip_state;
-   (void) dump_sf_state;
-   (void) dump_wm_state;
-   (void) dump_cc_state_gen4;
-
-   intel_bo_unmap(p->cp->bo);
-}
-
-/**
- * Dump the pipeline.
- */
-void
-ilo_3d_pipeline_dump(struct ilo_3d_pipeline *p)
-{
-   ilo_cp_dump(p->cp);
-   dump_3d_state(p);
-}
diff --git a/src/gallium/drivers/ilo/ilo_3d_pipeline_gen6.c b/src/gallium/drivers/ilo/ilo_3d_pipeline_gen6.c
index 227492a41bf..3bff8961ce8 100644
--- a/src/gallium/drivers/ilo/ilo_3d_pipeline_gen6.c
+++ b/src/gallium/drivers/ilo/ilo_3d_pipeline_gen6.c
@@ -215,11 +215,14 @@ gen6_pipeline_common_base_address(struct ilo_3d_pipeline *p,
    /* STATE_BASE_ADDRESS */
    if (session->state_bo_changed || session->kernel_bo_changed ||
        session->batch_bo_changed) {
+      const struct ilo_builder_writer *bat =
+         &p->cp->builder.writers[ILO_BUILDER_WRITER_BATCH];
+
       if (p->dev->gen == ILO_GEN(6))
          gen6_wa_pipe_control_post_sync(p, false);
 
       gen6_emit_STATE_BASE_ADDRESS(p->dev,
-            NULL, p->cp->bo, p->cp->bo, NULL, ilo->hw3d->kernel.bo,
+            NULL, bat->bo, bat->bo, NULL, ilo->hw3d->kernel.bo,
             0, 0, 0, 0, p->cp);
 
       /*
@@ -1633,8 +1636,8 @@ gen6_rectlist_commands(struct ilo_3d_pipeline *p,
 
    gen6_emit_STATE_BASE_ADDRESS(p->dev,
          NULL,                /* General State Base */
-         p->cp->bo,           /* Surface State Base */
-         p->cp->bo,           /* Dynamic State Base */
+         p->cp->builder.writers[0].bo,           /* Surface State Base */
+         p->cp->builder.writers[0].bo,           /* Dynamic State Base */
          NULL,                /* Indirect Object Base */
          NULL,                /* Instruction Base */
          0, 0, 0, 0, p->cp);
diff --git a/src/gallium/drivers/ilo/ilo_3d_pipeline_gen7.c b/src/gallium/drivers/ilo/ilo_3d_pipeline_gen7.c
index 557e5a8a271..f8c335aac78 100644
--- a/src/gallium/drivers/ilo/ilo_3d_pipeline_gen7.c
+++ b/src/gallium/drivers/ilo/ilo_3d_pipeline_gen7.c
@@ -802,8 +802,8 @@ gen7_rectlist_commands(struct ilo_3d_pipeline *p,
 
    gen6_emit_STATE_BASE_ADDRESS(p->dev,
          NULL,                /* General State Base */
-         p->cp->bo,           /* Surface State Base */
-         p->cp->bo,           /* Dynamic State Base */
+         p->cp->builder.writers[0].bo,           /* Surface State Base */
+         p->cp->builder.writers[0].bo,           /* Dynamic State Base */
          NULL,                /* Indirect Object Base */
          NULL,                /* Instruction Base */
          0, 0, 0, 0, p->cp);
diff --git a/src/gallium/drivers/ilo/ilo_blitter_blt.c b/src/gallium/drivers/ilo/ilo_blitter_blt.c
index b9156412aee..703aef236b0 100644
--- a/src/gallium/drivers/ilo/ilo_blitter_blt.c
+++ b/src/gallium/drivers/ilo/ilo_blitter_blt.c
@@ -321,7 +321,7 @@ ilo_blitter_blt_begin(struct ilo_blitter *blitter, int max_cmd_size,
                       struct intel_bo *src, enum intel_tiling_mode src_tiling)
 {
    struct ilo_context *ilo = blitter->ilo;
-   struct intel_bo *aper_check[3];
+   struct intel_bo *aper_check[2];
    int count;
    uint32_t swctrl;
 
@@ -330,16 +330,15 @@ ilo_blitter_blt_begin(struct ilo_blitter *blitter, int max_cmd_size,
    ilo_cp_set_owner(ilo->cp, NULL, 0);
 
    /* check aperture space */
-   aper_check[0] = ilo->cp->bo;
-   aper_check[1] = dst;
-   count = 2;
+   aper_check[0] = dst;
+   count = 1;
 
    if (src) {
-      aper_check[2] = src;
+      aper_check[1] = src;
       count++;
    }
 
-   if (!intel_winsys_can_submit_bo(ilo->winsys, aper_check, count))
+   if (!ilo_builder_validate(&ilo->cp->builder, count, aper_check))
       ilo_cp_flush(ilo->cp, "out of aperture");
 
    /* set BCS_SWCTRL */
diff --git a/src/gallium/drivers/ilo/ilo_context.c b/src/gallium/drivers/ilo/ilo_context.c
index 04df389c6c2..119d940e3f9 100644
--- a/src/gallium/drivers/ilo/ilo_context.c
+++ b/src/gallium/drivers/ilo/ilo_context.c
@@ -46,13 +46,6 @@ ilo_context_cp_flushed(struct ilo_cp *cp, void *data)
 {
    struct ilo_context *ilo = ilo_context(data);
 
-   if (ilo->last_cp_bo)
-      intel_bo_unreference(ilo->last_cp_bo);
-
-   /* remember the just flushed bo, on which fences could wait */
-   ilo->last_cp_bo = cp->bo;
-   intel_bo_reference(ilo->last_cp_bo);
-
    ilo_3d_cp_flushed(ilo->hw3d);
 }
 
@@ -68,7 +61,7 @@ ilo_flush(struct pipe_context *pipe,
 
    if (f) {
       *f = (struct pipe_fence_handle *)
-         ilo_fence_create(pipe->screen, ilo->last_cp_bo);
+         ilo_fence_create(pipe->screen, ilo->cp->last_submitted_bo);
    }
 }
 
@@ -79,9 +72,6 @@ ilo_context_destroy(struct pipe_context *pipe)
 
    ilo_cleanup_states(ilo);
 
-   if (ilo->last_cp_bo)
-      intel_bo_unreference(ilo->last_cp_bo);
-
    if (ilo->uploader)
       u_upload_destroy(ilo->uploader);
 
@@ -119,8 +109,7 @@ ilo_context_create(struct pipe_screen *screen, void *priv)
    util_slab_create(&ilo->transfer_mempool,
          sizeof(struct ilo_transfer), 64, UTIL_SLAB_SINGLETHREADED);
 
-   /* 8192 DWords */
-   ilo->cp = ilo_cp_create(ilo->winsys, 8192, is->dev.has_llc);
+   ilo->cp = ilo_cp_create(ilo->dev, ilo->winsys);
    ilo->shader_cache = ilo_shader_cache_create();
    if (ilo->cp)
       ilo->hw3d = ilo_3d_create(ilo->cp, ilo->dev);
diff --git a/src/gallium/drivers/ilo/ilo_context.h b/src/gallium/drivers/ilo/ilo_context.h
index a5c1ce0755a..d3c74a4db1b 100644
--- a/src/gallium/drivers/ilo/ilo_context.h
+++ b/src/gallium/drivers/ilo/ilo_context.h
@@ -53,7 +53,6 @@ struct ilo_context {
    struct util_slab_mempool transfer_mempool;
 
    struct ilo_cp *cp;
-   struct intel_bo *last_cp_bo;
 
    struct ilo_shader_cache *shader_cache;
    struct ilo_3d *hw3d;
diff --git a/src/gallium/drivers/ilo/ilo_cp.c b/src/gallium/drivers/ilo/ilo_cp.c
index 3ddec6c7bd9..4f65d443f49 100644
--- a/src/gallium/drivers/ilo/ilo_cp.c
+++ b/src/gallium/drivers/ilo/ilo_cp.c
@@ -25,168 +25,46 @@
  *    Chia-I Wu <olv@lunarg.com>
  */
 
-#include "genhw/genhw.h"
 #include "intel_winsys.h"
 
 #include "ilo_cp.h"
 
-/* the size of the private space */
-static const int ilo_cp_private = 2;
-
-/**
- * Dump the contents of the parser bo.  This can only be called in the flush
- * callback.
- */
-void
-ilo_cp_dump(struct ilo_cp *cp)
-{
-   ilo_printf("dumping %d bytes\n", cp->used * 4);
-   if (cp->used)
-      intel_winsys_decode_bo(cp->winsys, cp->bo, cp->used * 4);
-}
-
-/**
- * Save the command parser state for rewind.
- *
- * Note that this cannot rewind a flush, and the caller must make sure
- * that does not happend.
- */
-void
-ilo_cp_setjmp(struct ilo_cp *cp, struct ilo_cp_jmp_buf *jmp)
+static struct intel_bo *
+ilo_cp_end_batch(struct ilo_cp *cp, unsigned *used)
 {
-   jmp->id = pointer_to_intptr(cp->bo);
-
-   jmp->size = cp->size;
-   jmp->used = cp->used;
-   jmp->stolen = cp->stolen;
-   /* save reloc count to rewind ilo_cp_write_bo() */
-   jmp->reloc_count = intel_bo_get_reloc_count(cp->bo);
-}
-
-/**
- * Rewind to the saved state.
- */
-void
-ilo_cp_longjmp(struct ilo_cp *cp, const struct ilo_cp_jmp_buf *jmp)
-{
-   if (jmp->id != pointer_to_intptr(cp->bo)) {
-      assert(!"invalid use of CP longjmp");
-      return;
-   }
-
-   cp->size = jmp->size;
-   cp->used = jmp->used;
-   cp->stolen = jmp->stolen;
-   intel_bo_truncate_relocs(cp->bo, jmp->reloc_count);
-}
-
-/**
- * Clear the parser buffer.
- */
-static void
-ilo_cp_clear_buffer(struct ilo_cp *cp)
-{
-   cp->cmd_cur = 0;
-   cp->cmd_end = 0;
-
-   cp->used = 0;
-   cp->stolen = 0;
-
-   /*
-    * Recalculate cp->size.  This is needed not only because cp->stolen is
-    * reset above, but also that ilo_cp_private are added to cp->size in
-    * ilo_cp_end_buffer().
-    */
-   cp->size = cp->bo_size - ilo_cp_private;
-}
-
-/**
- * Add MI_BATCH_BUFFER_END to the private space of the parser buffer.
- */
-static void
-ilo_cp_end_buffer(struct ilo_cp *cp)
-{
-   /* make the private space available */
-   cp->size += ilo_cp_private;
-
-   assert(cp->used + 2 <= cp->size);
-
-   cp->ptr[cp->used++] = GEN6_MI_CMD(MI_BATCH_BUFFER_END);
-
-   /*
-    * From the Sandy Bridge PRM, volume 1 part 1, page 107:
-    *
-    *     "The batch buffer must be QWord aligned and a multiple of QWords in
-    *      length."
-    */
-   if (cp->used & 1)
-      cp->ptr[cp->used++] = GEN6_MI_CMD(MI_NOOP);
-}
-
-/**
- * Upload the parser buffer to the bo.
- */
-static int
-ilo_cp_upload_buffer(struct ilo_cp *cp)
-{
-   int err;
-
-   if (!cp->sys) {
-      intel_bo_unmap(cp->bo);
-      return 0;
-   }
-
-   err = intel_bo_pwrite(cp->bo, 0, cp->used * 4, cp->ptr);
-   if (likely(!err && cp->stolen)) {
-      const int offset = cp->bo_size - cp->stolen;
-
-      err = intel_bo_pwrite(cp->bo, offset * 4,
-            cp->stolen * 4, &cp->ptr[offset]);
-   }
-
-   return err;
-}
-
-/**
- * Reallocate the parser bo.
- */
-static void
-ilo_cp_realloc_bo(struct ilo_cp *cp)
-{
-   struct intel_bo *bo;
+   ilo_cp_set_owner(cp, NULL, 0);
 
-   /*
-    * allocate the new bo before unreferencing the old one so that they
-    * won't point at the same address, which is needed for jmpbuf
-    */
-   bo = intel_winsys_alloc_buffer(cp->winsys,
-         "batch buffer", cp->bo_size * 4, true);
-   if (unlikely(!bo)) {
-      /* reuse the old one */
-      bo = cp->bo;
-      intel_bo_reference(bo);
+   if (!ilo_builder_batch_used(&cp->builder)) {
+      ilo_builder_batch_discard(&cp->builder);
+      return NULL;
    }
 
-   if (cp->bo)
-      intel_bo_unreference(cp->bo);
-   cp->bo = bo;
+   /* see ilo_cp_space() */
+   assert(ilo_builder_batch_space(&cp->builder) >= 2);
+   ilo_builder_batch_mi_batch_buffer_end(&cp->builder);
 
-   if (!cp->sys)
-      cp->ptr = intel_bo_map(cp->bo, true);
+   return ilo_builder_end(&cp->builder, used);
 }
 
 /**
- * Execute the parser bo.
+ * Flush the command parser and execute the commands.  When the parser buffer
+ * is empty, the callback is not invoked.
  */
-static int
-ilo_cp_exec_bo(struct ilo_cp *cp)
+void
+ilo_cp_flush_internal(struct ilo_cp *cp)
 {
    const bool do_exec = !(ilo_debug & ILO_DEBUG_NOHW);
+   struct intel_bo *bo;
+   unsigned used;
    int err;
 
+   bo = ilo_cp_end_batch(cp, &used);
+   if (!bo)
+      return;
+
    if (likely(do_exec)) {
       err = intel_winsys_submit_bo(cp->winsys, cp->ring,
-            cp->bo, cp->used * 4, cp->render_ctx, cp->one_off_flags);
+            bo, used, cp->render_ctx, cp->one_off_flags);
    }
    else {
       err = 0;
@@ -194,42 +72,17 @@ ilo_cp_exec_bo(struct ilo_cp *cp)
 
    cp->one_off_flags = 0;
 
-   return err;
-}
-
-/**
- * Flush the command parser and execute the commands.  When the parser buffer
- * is empty, the callback is not invoked.
- */
-void
-ilo_cp_flush_internal(struct ilo_cp *cp)
-{
-   int err;
-
-   ilo_cp_set_owner(cp, NULL, 0);
-
-   /* sanity check */
-   assert(cp->bo_size == cp->size + cp->stolen + ilo_cp_private);
+   if (!err) {
+      if (cp->last_submitted_bo)
+         intel_bo_unreference(cp->last_submitted_bo);
+      cp->last_submitted_bo = bo;
+      intel_bo_reference(cp->last_submitted_bo);
 
-   if (!cp->used) {
-      /* return the space stolen and etc. */
-      ilo_cp_clear_buffer(cp);
-
-      return;
+      if (cp->flush_callback)
+         cp->flush_callback(cp, cp->flush_callback_data);
    }
 
-   ilo_cp_end_buffer(cp);
-
-   /* upload and execute */
-   err = ilo_cp_upload_buffer(cp);
-   if (likely(!err))
-      err = ilo_cp_exec_bo(cp);
-
-   if (likely(!err && cp->flush_callback))
-      cp->flush_callback(cp, cp->flush_callback_data);
-
-   ilo_cp_clear_buffer(cp);
-   ilo_cp_realloc_bo(cp);
+   ilo_builder_begin(&cp->builder);
 }
 
 /**
@@ -238,16 +91,9 @@ ilo_cp_flush_internal(struct ilo_cp *cp)
 void
 ilo_cp_destroy(struct ilo_cp *cp)
 {
-   if (cp->bo) {
-      if (!cp->sys)
-         intel_bo_unmap(cp->bo);
-
-      intel_bo_unreference(cp->bo);
-   }
+   ilo_builder_reset(&cp->builder);
 
    intel_winsys_destroy_context(cp->winsys, cp->render_ctx);
-
-   FREE(cp->sys);
    FREE(cp);
 }
 
@@ -255,7 +101,7 @@ ilo_cp_destroy(struct ilo_cp *cp)
  * Create a command parser.
  */
 struct ilo_cp *
-ilo_cp_create(struct intel_winsys *winsys, int size, bool direct_map)
+ilo_cp_create(const struct ilo_dev_info *dev, struct intel_winsys *winsys)
 {
    struct ilo_cp *cp;
 
@@ -273,26 +119,12 @@ ilo_cp_create(struct intel_winsys *winsys, int size, bool direct_map)
    cp->ring = INTEL_RING_RENDER;
    cp->no_implicit_flush = false;
 
-   cp->bo_size = size;
+   ilo_builder_init(&cp->builder, dev, winsys);
 
-   if (!direct_map) {
-      cp->sys = MALLOC(cp->bo_size * 4);
-      if (!cp->sys) {
-         FREE(cp);
-         return NULL;
-      }
-
-      cp->ptr = cp->sys;
-   }
-
-   ilo_cp_realloc_bo(cp);
-   if (!cp->bo) {
-      FREE(cp->sys);
-      FREE(cp);
+   if (!ilo_builder_begin(&cp->builder)) {
+      ilo_cp_destroy(cp);
       return NULL;
    }
 
-   ilo_cp_clear_buffer(cp);
-
    return cp;
 }
diff --git a/src/gallium/drivers/ilo/ilo_cp.h b/src/gallium/drivers/ilo/ilo_cp.h
index b4963022b00..bdc75c63d87 100644
--- a/src/gallium/drivers/ilo/ilo_cp.h
+++ b/src/gallium/drivers/ilo/ilo_cp.h
@@ -30,6 +30,7 @@
 
 #include "intel_winsys.h"
 
+#include "ilo_builder.h"
 #include "ilo_common.h"
 
 struct ilo_cp;
@@ -58,27 +59,16 @@ struct ilo_cp {
    bool no_implicit_flush;
    unsigned one_off_flags;
 
-   int bo_size;
-   struct intel_bo *bo;
-   uint32_t *sys;
+   struct ilo_builder builder;
+   struct intel_bo *last_submitted_bo;
 
+   unsigned pos;
    uint32_t *ptr;
-   int size, used, stolen;
-
    int cmd_cur, cmd_end;
 };
 
-/**
- * Jump buffer to save command parser state for rewind.
- */
-struct ilo_cp_jmp_buf {
-   intptr_t id;
-   int size, used, stolen;
-   int reloc_count;
-};
-
 struct ilo_cp *
-ilo_cp_create(struct intel_winsys *winsys, int size, bool direct_map);
+ilo_cp_create(const struct ilo_dev_info *dev, struct intel_winsys *winsys);
 
 void
 ilo_cp_destroy(struct ilo_cp *cp);
@@ -90,32 +80,21 @@ static inline void
 ilo_cp_flush(struct ilo_cp *cp, const char *reason)
 {
    if (ilo_debug & ILO_DEBUG_FLUSH) {
-      ilo_printf("cp flushed for %s with %d+%d DWords (%.1f%%) because of %s\n",
-            (cp->ring == INTEL_RING_RENDER) ? "render" : "other",
-             cp->used, cp->stolen,
-             (float) (100 * (cp->used + cp->stolen)) / cp->bo_size,
-             reason);
+      ilo_printf("cp flushed for %s because of %s: ",
+            (cp->ring == INTEL_RING_RENDER) ? "render" : "other", reason);
+      ilo_builder_batch_print_stats(&cp->builder);
    }
 
    ilo_cp_flush_internal(cp);
 }
 
-void
-ilo_cp_dump(struct ilo_cp *cp);
-
-void
-ilo_cp_setjmp(struct ilo_cp *cp, struct ilo_cp_jmp_buf *jmp);
-
-void
-ilo_cp_longjmp(struct ilo_cp *cp, const struct ilo_cp_jmp_buf *jmp);
-
 /**
  * Return true if the parser buffer is empty.
  */
 static inline bool
 ilo_cp_empty(struct ilo_cp *cp)
 {
-   return !cp->used;
+   return !ilo_builder_batch_used(&cp->builder);
 }
 
 /**
@@ -124,7 +103,12 @@ ilo_cp_empty(struct ilo_cp *cp)
 static inline int
 ilo_cp_space(struct ilo_cp *cp)
 {
-   return cp->size - cp->used;
+   const int space = ilo_builder_batch_space(&cp->builder);
+   const int mi_batch_buffer_end_space = 2;
+
+   assert(space >= cp->owner_reserve + mi_batch_buffer_end_space);
+
+   return space - cp->owner_reserve - mi_batch_buffer_end_space;
 }
 
 /**
@@ -135,8 +119,7 @@ ilo_cp_implicit_flush(struct ilo_cp *cp)
 {
    if (cp->no_implicit_flush) {
       assert(!"unexpected command parser flush");
-      /* discard the commands */
-      cp->used = 0;
+      ilo_builder_batch_discard(&cp->builder);
    }
 
    ilo_cp_flush(cp, "out of space (implicit)");
@@ -201,7 +184,6 @@ ilo_cp_set_owner(struct ilo_cp *cp, const struct ilo_cp_owner *owner,
       const bool no_implicit_flush = cp->no_implicit_flush;
 
       /* reclaim the reserved space */
-      cp->size += cp->owner_reserve;
       cp->owner_reserve = 0;
 
       /* invoke the release callback */
@@ -215,15 +197,13 @@ ilo_cp_set_owner(struct ilo_cp *cp, const struct ilo_cp_owner *owner,
    if (cp->owner_reserve != reserve) {
       const int extra = reserve - cp->owner_reserve;
 
-      if (cp->used > cp->size - extra) {
+      if (ilo_cp_space(cp) < extra) {
          ilo_cp_implicit_flush(cp);
-         assert(cp->used <= cp->size - reserve);
 
-         cp->size -= reserve;
+         assert(ilo_cp_space(cp) >= reserve);
          cp->owner_reserve = reserve;
       }
       else {
-         cp->size -= extra;
          cp->owner_reserve += extra;
       }
    }
@@ -240,57 +220,47 @@ ilo_cp_set_owner(struct ilo_cp *cp, const struct ilo_cp_owner *owner,
 static inline void
 ilo_cp_begin(struct ilo_cp *cp, int cmd_size)
 {
-   if (cp->used + cmd_size > cp->size) {
+   if (ilo_cp_space(cp) < cmd_size) {
       ilo_cp_implicit_flush(cp);
-      assert(cp->used + cmd_size <= cp->size);
+      assert(ilo_cp_space(cp) >= cmd_size);
    }
 
-   assert(cp->cmd_cur == cp->cmd_end);
-   cp->cmd_cur = cp->used;
-   cp->cmd_end = cp->cmd_cur + cmd_size;
-   cp->used = cp->cmd_end;
+   cp->pos = ilo_builder_batch_pointer(&cp->builder, cmd_size, &cp->ptr);
+
+   cp->cmd_cur = 0;
+   cp->cmd_end = cmd_size;
 }
 
 /**
  * Begin writing data to a space stolen from the top of the parser buffer.
  *
- * \param desc informative description of the data to be written
+ * \param item builder item type
  * \param data_size in dwords
  * \param align in dwords
  * \param bo_offset in bytes to the stolen space
  */
 static inline void
-ilo_cp_steal(struct ilo_cp *cp, const char *desc,
+ilo_cp_steal(struct ilo_cp *cp, enum ilo_builder_item_type item,
              int data_size, int align, uint32_t *bo_offset)
 {
-   int pad, steal;
-
    if (!align)
       align = 1;
 
-   pad = (cp->bo_size - cp->stolen - data_size) % align;
-   steal = data_size + pad;
-
    /* flush if there is not enough space after stealing */
-   if (cp->used > cp->size - steal) {
+   if (ilo_cp_space(cp) < data_size + align - 1) {
       ilo_cp_implicit_flush(cp);
-
-      pad = (cp->bo_size - cp->stolen - data_size) % align;
-      steal = data_size + steal;
-
-      assert(cp->used <= cp->size - steal);
+      assert(ilo_cp_space(cp) >= data_size + align - 1);
    }
 
-   cp->size -= steal;
-   cp->stolen += steal;
+   cp->pos = ilo_builder_state_pointer(&cp->builder,
+         item, align << 2, data_size, &cp->ptr) >> 2;
 
-   assert(cp->cmd_cur == cp->cmd_end);
-   cp->cmd_cur = cp->bo_size - cp->stolen;
-   cp->cmd_end = cp->cmd_cur + data_size;
+   cp->cmd_cur = 0;
+   cp->cmd_end = data_size;
 
    /* offset in cp->bo */
    if (bo_offset)
-      *bo_offset = cp->cmd_cur * 4;
+      *bo_offset = cp->pos << 2;
 }
 
 /**
@@ -323,20 +293,14 @@ static inline void
 ilo_cp_write_bo(struct ilo_cp *cp, uint32_t val,
                 struct intel_bo *bo, uint32_t flags)
 {
-   uint64_t presumed_offset;
-
    if (bo) {
-      intel_bo_add_reloc(cp->bo, cp->cmd_cur * 4, bo, val, flags,
-            &presumed_offset);
+      ilo_builder_batch_reloc(&cp->builder, cp->pos + cp->cmd_cur,
+            bo, val, flags);
+      cp->cmd_cur++;
    }
    else {
-      presumed_offset = 0;
+      ilo_cp_write(cp, 0);
    }
-
-   /* 32-bit addressing */
-   assert(presumed_offset == (uint64_t) ((uint32_t) presumed_offset));
-
-   ilo_cp_write(cp, (uint32_t) presumed_offset);
 }
 
 /**
@@ -357,12 +321,12 @@ ilo_cp_end(struct ilo_cp *cp)
  *             change is made to the parser.
  */
 static inline void *
-ilo_cp_steal_ptr(struct ilo_cp *cp, const char *desc,
+ilo_cp_steal_ptr(struct ilo_cp *cp, enum ilo_builder_item_type item,
                  int data_size, int align, uint32_t *bo_offset)
 {
    void *ptr;
 
-   ilo_cp_steal(cp, desc, data_size, align, bo_offset);
+   ilo_cp_steal(cp, item, data_size, align, bo_offset);
 
    ptr = &cp->ptr[cp->cmd_cur];
    cp->cmd_cur = cp->cmd_end;
diff --git a/src/gallium/drivers/ilo/ilo_gpe_gen6.h b/src/gallium/drivers/ilo/ilo_gpe_gen6.h
index 473337228af..24f7d156934 100644
--- a/src/gallium/drivers/ilo/ilo_gpe_gen6.h
+++ b/src/gallium/drivers/ilo/ilo_gpe_gen6.h
@@ -1946,7 +1946,7 @@ gen6_emit_INTERFACE_DESCRIPTOR_DATA(const struct ilo_dev_info *dev,
 
    ILO_GPE_VALID_GEN(dev, 6, 6);
 
-   dw = ilo_cp_steal_ptr(cp, "INTERFACE_DESCRIPTOR_DATA",
+   dw = ilo_cp_steal_ptr(cp, ILO_BUILDER_ITEM_BLOB,
          state_len, state_align, &state_offset);
 
    for (i = 0; i < num_ids; i++) {
@@ -1989,7 +1989,7 @@ gen6_emit_SF_VIEWPORT(const struct ilo_dev_info *dev,
     */
    assert(num_viewports && num_viewports <= 16);
 
-   dw = ilo_cp_steal_ptr(cp, "SF_VIEWPORT",
+   dw = ilo_cp_steal_ptr(cp, ILO_BUILDER_ITEM_SF_VIEWPORT,
          state_len, state_align, &state_offset);
 
    for (i = 0; i < num_viewports; i++) {
@@ -2031,7 +2031,7 @@ gen6_emit_CLIP_VIEWPORT(const struct ilo_dev_info *dev,
     */
    assert(num_viewports && num_viewports <= 16);
 
-   dw = ilo_cp_steal_ptr(cp, "CLIP_VIEWPORT",
+   dw = ilo_cp_steal_ptr(cp, ILO_BUILDER_ITEM_CLIP_VIEWPORT,
          state_len, state_align, &state_offset);
 
    for (i = 0; i < num_viewports; i++) {
@@ -2068,7 +2068,7 @@ gen6_emit_CC_VIEWPORT(const struct ilo_dev_info *dev,
     */
    assert(num_viewports && num_viewports <= 16);
 
-   dw = ilo_cp_steal_ptr(cp, "CC_VIEWPORT",
+   dw = ilo_cp_steal_ptr(cp, ILO_BUILDER_ITEM_CC_VIEWPORT,
          state_len, state_align, &state_offset);
 
    for (i = 0; i < num_viewports; i++) {
@@ -2096,7 +2096,7 @@ gen6_emit_COLOR_CALC_STATE(const struct ilo_dev_info *dev,
 
    ILO_GPE_VALID_GEN(dev, 6, 7.5);
 
-   dw = ilo_cp_steal_ptr(cp, "COLOR_CALC_STATE",
+   dw = ilo_cp_steal_ptr(cp, ILO_BUILDER_ITEM_COLOR_CALC,
          state_len, state_align, &state_offset);
 
    dw[0] = stencil_ref->ref_value[0] << 24 |
@@ -2142,7 +2142,7 @@ gen6_emit_BLEND_STATE(const struct ilo_dev_info *dev,
 
    state_len = 2 * num_targets;
 
-   dw = ilo_cp_steal_ptr(cp, "BLEND_STATE",
+   dw = ilo_cp_steal_ptr(cp, ILO_BUILDER_ITEM_BLEND,
          state_len, state_align, &state_offset);
 
    for (i = 0; i < num_targets; i++) {
@@ -2249,7 +2249,7 @@ gen6_emit_DEPTH_STENCIL_STATE(const struct ilo_dev_info *dev,
 
    ILO_GPE_VALID_GEN(dev, 6, 7.5);
 
-   dw = ilo_cp_steal_ptr(cp, "DEPTH_STENCIL_STATE",
+   dw = ilo_cp_steal_ptr(cp, ILO_BUILDER_ITEM_DEPTH_STENCIL,
          state_len, state_align, &state_offset);
 
    dw[0] = dsa->payload[0];
@@ -2279,7 +2279,7 @@ gen6_emit_SCISSOR_RECT(const struct ilo_dev_info *dev,
     */
    assert(num_viewports && num_viewports <= 16);
 
-   dw = ilo_cp_steal_ptr(cp, "SCISSOR_RECT",
+   dw = ilo_cp_steal_ptr(cp, ILO_BUILDER_ITEM_SCISSOR_RECT,
          state_len, state_align, &state_offset);
 
    memcpy(dw, scissor->payload, state_len * 4);
@@ -2309,7 +2309,7 @@ gen6_emit_BINDING_TABLE_STATE(const struct ilo_dev_info *dev,
    if (!num_surface_states)
       return 0;
 
-   dw = ilo_cp_steal_ptr(cp, "BINDING_TABLE_STATE",
+   dw = ilo_cp_steal_ptr(cp, ILO_BUILDER_ITEM_BINDING_TABLE,
          state_len, state_align, &state_offset);
    memcpy(dw, surface_states,
          num_surface_states * sizeof(surface_states[0]));
@@ -2329,7 +2329,8 @@ gen6_emit_SURFACE_STATE(const struct ilo_dev_info *dev,
 
    ILO_GPE_VALID_GEN(dev, 6, 7.5);
 
-   ilo_cp_steal(cp, "SURFACE_STATE", state_len, state_align, &state_offset);
+   ilo_cp_steal(cp, ILO_BUILDER_ITEM_SURFACE,
+         state_len, state_align, &state_offset);
 
    STATIC_ASSERT(Elements(surf->payload) >= 8);
 
@@ -2418,7 +2419,7 @@ gen6_emit_SAMPLER_STATE(const struct ilo_dev_info *dev,
    if (!num_samplers)
       return 0;
 
-   dw = ilo_cp_steal_ptr(cp, "SAMPLER_STATE",
+   dw = ilo_cp_steal_ptr(cp, ILO_BUILDER_ITEM_SAMPLER,
          state_len, state_align, &state_offset);
 
    for (i = 0; i < num_samplers; i++) {
@@ -2514,7 +2515,7 @@ gen6_emit_SAMPLER_BORDER_COLOR_STATE(const struct ilo_dev_info *dev,
 
    ILO_GPE_VALID_GEN(dev, 6, 7.5);
 
-   dw = ilo_cp_steal_ptr(cp, "SAMPLER_BORDER_COLOR_STATE",
+   dw = ilo_cp_steal_ptr(cp, ILO_BUILDER_ITEM_BLOB,
          state_len, state_align, &state_offset);
 
    /* see ilo_gpe_init_sampler_cso() */
@@ -2539,7 +2540,7 @@ gen6_emit_push_constant_buffer(const struct ilo_dev_info *dev,
 
    ILO_GPE_VALID_GEN(dev, 6, 7.5);
 
-   buf = ilo_cp_steal_ptr(cp, "PUSH_CONSTANT_BUFFER",
+   buf = ilo_cp_steal_ptr(cp, ILO_BUILDER_ITEM_BLOB,
          state_len, state_align, &state_offset);
 
    /* zero out the unused range */
diff --git a/src/gallium/drivers/ilo/ilo_gpe_gen7.h b/src/gallium/drivers/ilo/ilo_gpe_gen7.h
index 0fab6e3474d..f3059ae9fd9 100644
--- a/src/gallium/drivers/ilo/ilo_gpe_gen7.h
+++ b/src/gallium/drivers/ilo/ilo_gpe_gen7.h
@@ -1158,7 +1158,7 @@ gen7_emit_SF_CLIP_VIEWPORT(const struct ilo_dev_info *dev,
     */
    assert(num_viewports && num_viewports <= 16);
 
-   dw = ilo_cp_steal_ptr(cp, "SF_CLIP_VIEWPORT",
+   dw = ilo_cp_steal_ptr(cp, ILO_BUILDER_ITEM_SF_VIEWPORT,
          state_len, state_align, &state_offset);
 
    for (i = 0; i < num_viewports; i++) {
diff --git a/src/gallium/drivers/ilo/ilo_query.c b/src/gallium/drivers/ilo/ilo_query.c
index 72fc1c70a70..c789397d12c 100644
--- a/src/gallium/drivers/ilo/ilo_query.c
+++ b/src/gallium/drivers/ilo/ilo_query.c
@@ -185,7 +185,7 @@ ilo_get_query_result(struct pipe_context *pipe, struct pipe_query *query,
       return false;
 
    if (q->bo) {
-      if (intel_bo_has_reloc(ilo->cp->bo, q->bo))
+      if (ilo_builder_has_reloc(&ilo->cp->builder, q->bo))
          ilo_cp_flush(ilo->cp, "syncing for queries");
 
       if (!wait && intel_bo_is_busy(q->bo))
diff --git a/src/gallium/drivers/ilo/ilo_transfer.c b/src/gallium/drivers/ilo/ilo_transfer.c
index de6f4324b9a..50eda3da56a 100644
--- a/src/gallium/drivers/ilo/ilo_transfer.c
+++ b/src/gallium/drivers/ilo/ilo_transfer.c
@@ -1045,7 +1045,7 @@ copy_staging_resource(struct ilo_context *ilo,
 static bool
 is_bo_busy(struct ilo_context *ilo, struct intel_bo *bo, bool *need_flush)
 {
-   const bool referenced = intel_bo_has_reloc(ilo->cp->bo, bo);
+   const bool referenced = ilo_builder_has_reloc(&ilo->cp->builder, bo);
 
    if (need_flush)
       *need_flush = referenced;