summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorChia-I Wu <olvaffe@gmail.com>2014-08-08 15:36:36 +0800
committerChia-I Wu <olvaffe@gmail.com>2014-08-19 19:53:37 +0800
commitfb3d506431871fdb04fc84bbcc916d8f9d7c9954 (patch)
tree3f23bd8055812091cd48e89f67362d4511743cfd
parent925359bc784399c1a65a46e2b136da3c4d30388d (diff)
ilo: migrate to ilo_layout
Embed an ilo_layout in ilo_texture, and remove now duplicated members.
-rw-r--r--src/gallium/drivers/ilo/ilo_blitter_blt.c44
-rw-r--r--src/gallium/drivers/ilo/ilo_blitter_rectlist.c12
-rw-r--r--src/gallium/drivers/ilo/ilo_gpe_gen6.c47
-rw-r--r--src/gallium/drivers/ilo/ilo_gpe_gen7.c27
-rw-r--r--src/gallium/drivers/ilo/ilo_resource.c1381
-rw-r--r--src/gallium/drivers/ilo/ilo_resource.h54
-rw-r--r--src/gallium/drivers/ilo/ilo_state.c2
-rw-r--r--src/gallium/drivers/ilo/ilo_transfer.c100
8 files changed, 213 insertions, 1454 deletions
diff --git a/src/gallium/drivers/ilo/ilo_blitter_blt.c b/src/gallium/drivers/ilo/ilo_blitter_blt.c
index ac34383f5bd..1dd00af6c9c 100644
--- a/src/gallium/drivers/ilo/ilo_blitter_blt.c
+++ b/src/gallium/drivers/ilo/ilo_blitter_blt.c
@@ -521,19 +521,20 @@ tex_clear_region(struct ilo_blitter *blitter,
if (dst->separate_s8)
return false;
- if (dst->bo_stride > max_extent)
+ if (dst->layout.bo_stride > max_extent)
return false;
swctrl = ilo_blitter_blt_begin(blitter, dst_box->depth * 6,
- dst->bo, dst->tiling, NULL, INTEL_TILING_NONE);
+ dst->bo, dst->layout.tiling, NULL, INTEL_TILING_NONE);
for (slice = 0; slice < dst_box->depth; slice++) {
- const struct ilo_texture_slice *dst_slice =
- ilo_texture_get_slice(dst, dst_level, dst_box->z + slice);
unsigned x1, y1, x2, y2;
- x1 = dst_slice->x + dst_box->x;
- y1 = dst_slice->y + dst_box->y;
+ ilo_layout_get_slice_pos(&dst->layout,
+ dst_level, dst_box->z + slice, &x1, &y1);
+
+ x1 += dst_box->x;
+ y1 += dst_box->y;
x2 = x1 + dst_box->width;
y2 = y1 + dst_box->height;
@@ -542,7 +543,7 @@ tex_clear_region(struct ilo_blitter *blitter,
break;
gen6_emit_XY_COLOR_BLT(ilo->dev,
- dst->bo, dst->tiling, dst->bo_stride, 0,
+ dst->bo, dst->layout.tiling, dst->layout.bo_stride, 0,
x1, y1, x2, y2, val, rop, value_mask, write_mask,
ilo->cp);
}
@@ -562,7 +563,7 @@ tex_copy_region(struct ilo_blitter *blitter,
const struct pipe_box *src_box)
{
const struct util_format_description *desc =
- util_format_description(dst->bo_format);
+ util_format_description(dst->layout.format);
const unsigned max_extent = 32767; /* INT16_MAX */
const uint8_t rop = 0xcc; /* SRCCOPY */
struct ilo_context *ilo = blitter->ilo;
@@ -574,7 +575,8 @@ tex_copy_region(struct ilo_blitter *blitter,
if (dst->separate_s8 || src->separate_s8)
return false;
- if (dst->bo_stride > max_extent || src->bo_stride > max_extent)
+ if (dst->layout.bo_stride > max_extent ||
+ src->layout.bo_stride > max_extent)
return false;
cpp = desc->block.bits / 8;
@@ -605,21 +607,23 @@ tex_copy_region(struct ilo_blitter *blitter,
}
swctrl = ilo_blitter_blt_begin(blitter, src_box->depth * 8,
- dst->bo, dst->tiling, src->bo, src->tiling);
+ dst->bo, dst->layout.tiling, src->bo, src->layout.tiling);
for (slice = 0; slice < src_box->depth; slice++) {
- const struct ilo_texture_slice *dst_slice =
- ilo_texture_get_slice(dst, dst_level, dst_z + slice);
- const struct ilo_texture_slice *src_slice =
- ilo_texture_get_slice(src, src_level, src_box->z + slice);
unsigned x1, y1, x2, y2, src_x, src_y;
- x1 = (dst_slice->x + dst_x) * xscale;
- y1 = dst_slice->y + dst_y;
+ ilo_layout_get_slice_pos(&dst->layout,
+ dst_level, dst_z + slice, &x1, &y1);
+ x1 = (x1 + dst_x) * xscale;
+ y1 = y1 + dst_y;
x2 = (x1 + src_box->width) * xscale;
y2 = y1 + src_box->height;
- src_x = (src_slice->x + src_box->x) * xscale;
- src_y = src_slice->y + src_box->y;
+
+ ilo_layout_get_slice_pos(&src->layout,
+ src_level, src_box->z + slice, &src_x, &src_y);
+
+ src_x = (src_x + src_box->x) * xscale;
+ src_y += src_box->y;
/* in blocks */
x1 /= desc->block.width;
@@ -635,9 +639,9 @@ tex_copy_region(struct ilo_blitter *blitter,
break;
gen6_emit_XY_SRC_COPY_BLT(ilo->dev,
- dst->bo, dst->tiling, dst->bo_stride, 0,
+ dst->bo, dst->layout.tiling, dst->layout.bo_stride, 0,
x1, y1, x2, y2,
- src->bo, src->tiling, src->bo_stride, 0,
+ src->bo, src->layout.tiling, src->layout.bo_stride, 0,
src_x, src_y, rop, mask, mask,
ilo->cp);
}
diff --git a/src/gallium/drivers/ilo/ilo_blitter_rectlist.c b/src/gallium/drivers/ilo/ilo_blitter_rectlist.c
index f80dc5ddd2d..a2c3695cc4f 100644
--- a/src/gallium/drivers/ilo/ilo_blitter_rectlist.c
+++ b/src/gallium/drivers/ilo/ilo_blitter_rectlist.c
@@ -173,11 +173,13 @@ ilo_blitter_set_dsa(struct ilo_blitter *blitter,
static void
ilo_blitter_set_fb(struct ilo_blitter *blitter,
- const struct pipe_resource *res, unsigned level,
+ struct pipe_resource *res, unsigned level,
const struct ilo_surface_cso *cso)
{
- blitter->fb.width = u_minify(res->width0, level);
- blitter->fb.height = u_minify(res->height0, level);
+ struct ilo_texture *tex = ilo_texture(res);
+
+ blitter->fb.width = u_minify(tex->layout.width0, level);
+ blitter->fb.height = u_minify(tex->layout.height0, level);
blitter->fb.num_samples = res->nr_samples;
if (!blitter->fb.num_samples)
@@ -375,7 +377,7 @@ hiz_can_clear_zs(const struct ilo_blitter *blitter,
* The truth is when HiZ is enabled, separate stencil is also enabled on
* all GENs. The depth buffer format cannot be combined depth/stencil.
*/
- switch (tex->bo_format) {
+ switch (tex->layout.format) {
case PIPE_FORMAT_Z16_UNORM:
if (blitter->ilo->dev->gen == ILO_GEN(6) && tex->base.width0 % 16)
return false;
@@ -410,7 +412,7 @@ ilo_blitter_rectlist_clear_zs(struct ilo_blitter *blitter,
if (!hiz_can_clear_zs(blitter, tex))
return false;
- clear_value = util_pack_z(tex->bo_format, depth);
+ clear_value = util_pack_z(tex->layout.format, depth);
ilo_blit_resolve_surface(blitter->ilo, zs,
ILO_TEXTURE_RENDER_WRITE | ILO_TEXTURE_CLEAR);
diff --git a/src/gallium/drivers/ilo/ilo_gpe_gen6.c b/src/gallium/drivers/ilo/ilo_gpe_gen6.c
index 0a9fd98152d..bbc269faf47 100644
--- a/src/gallium/drivers/ilo/ilo_gpe_gen6.c
+++ b/src/gallium/drivers/ilo/ilo_gpe_gen6.c
@@ -1075,11 +1075,11 @@ zs_init_info(const struct ilo_dev_info *dev,
if (format != PIPE_FORMAT_S8_UINT) {
info->zs.bo = tex->bo;
- info->zs.stride = tex->bo_stride;
- info->zs.tiling = tex->tiling;
+ info->zs.stride = tex->layout.bo_stride;
+ info->zs.tiling = tex->layout.tiling;
if (offset_to_layer) {
- info->zs.offset = ilo_texture_get_slice_offset(tex,
+ info->zs.offset = ilo_layout_get_slice_tile_offset(&tex->layout,
level, first_layer, &x_offset[0], &y_offset[0]);
}
}
@@ -1096,22 +1096,23 @@ zs_init_info(const struct ilo_dev_info *dev,
* "The pitch must be set to 2x the value computed based on width,
* as the stencil buffer is stored with two rows interleaved."
*
- * According to the classic driver, we need to do the same for GEN7+
- * even though the Ivy Bridge PRM does not say anything about it.
+ * For GEN7, we still dobule the stride because we did not double the
+ * slice widths when initializing the layout.
*/
- info->stencil.stride = s8_tex->bo_stride * 2;
+ info->stencil.stride = s8_tex->layout.bo_stride * 2;
- info->stencil.tiling = s8_tex->tiling;
+ info->stencil.tiling = s8_tex->layout.tiling;
if (offset_to_layer) {
- info->stencil.offset = ilo_texture_get_slice_offset(s8_tex,
- level, first_layer, &x_offset[1], &y_offset[1]);
+ info->stencil.offset =
+ ilo_layout_get_slice_tile_offset(&s8_tex->layout,
+ level, first_layer, &x_offset[1], &y_offset[1]);
}
}
if (ilo_texture_can_enable_hiz(tex, level, first_layer, num_layers)) {
- info->hiz.bo = tex->hiz.bo;
- info->hiz.stride = tex->hiz.bo_stride;
+ info->hiz.bo = tex->aux_bo;
+ info->hiz.stride = tex->layout.aux_stride;
info->hiz.tiling = INTEL_TILING_Y;
/*
@@ -1128,8 +1129,8 @@ zs_init_info(const struct ilo_dev_info *dev,
}
}
- info->width = tex->base.width0;
- info->height = tex->base.height0;
+ info->width = tex->layout.width0;
+ info->height = tex->layout.height0;
info->depth = (tex->base.target == PIPE_TEXTURE_3D) ?
tex->base.depth0 : num_layers;
@@ -1934,11 +1935,11 @@ ilo_gpe_init_view_surface_for_texture_gen6(const struct ilo_dev_info *dev,
surface_format = ilo_translate_texture_format(dev, format);
assert(surface_format >= 0);
- width = tex->base.width0;
- height = tex->base.height0;
+ width = tex->layout.width0;
+ height = tex->layout.height0;
depth = (tex->base.target == PIPE_TEXTURE_3D) ?
tex->base.depth0 : num_layers;
- pitch = tex->bo_stride;
+ pitch = tex->layout.bo_stride;
if (surface_type == GEN6_SURFTYPE_CUBE) {
/*
@@ -1992,10 +1993,10 @@ ilo_gpe_init_view_surface_for_texture_gen6(const struct ilo_dev_info *dev,
}
/* non-full array spacing is supported only on GEN7+ */
- assert(tex->array_spacing_full);
+ assert(tex->layout.walk != ILO_LAYOUT_WALK_LOD);
/* non-interleaved samples are supported only on GEN7+ */
if (tex->base.nr_samples > 1)
- assert(tex->interleaved);
+ assert(tex->layout.interleaved_samples);
if (is_rt) {
assert(num_levels == 1);
@@ -2014,7 +2015,7 @@ ilo_gpe_init_view_surface_for_texture_gen6(const struct ilo_dev_info *dev,
/* we lose the capability for layered rendering */
assert(is_rt && num_layers == 1);
- layer_offset = ilo_texture_get_slice_offset(tex,
+ layer_offset = ilo_layout_get_slice_tile_offset(&tex->layout,
first_level, first_layer, &x_offset, &y_offset);
assert(x_offset % 4 == 0);
@@ -2056,7 +2057,7 @@ ilo_gpe_init_view_surface_for_texture_gen6(const struct ilo_dev_info *dev,
*
* "For linear surfaces, this field (X Offset) must be zero"
*/
- if (tex->tiling == INTEL_TILING_NONE) {
+ if (tex->layout.tiling == INTEL_TILING_NONE) {
if (is_rt) {
const int elem_size = util_format_get_blocksize(format);
assert(layer_offset % elem_size == 0);
@@ -2089,7 +2090,7 @@ ilo_gpe_init_view_surface_for_texture_gen6(const struct ilo_dev_info *dev,
dw[3] = (depth - 1) << GEN6_SURFACE_DW3_DEPTH__SHIFT |
(pitch - 1) << GEN6_SURFACE_DW3_PITCH__SHIFT |
- ilo_gpe_gen6_translate_winsys_tiling(tex->tiling);
+ ilo_gpe_gen6_translate_winsys_tiling(tex->layout.tiling);
dw[4] = first_level << GEN6_SURFACE_DW4_MIN_LOD__SHIFT |
first_layer << 17 |
@@ -2099,7 +2100,9 @@ ilo_gpe_init_view_surface_for_texture_gen6(const struct ilo_dev_info *dev,
dw[5] = x_offset << GEN6_SURFACE_DW5_X_OFFSET__SHIFT |
y_offset << GEN6_SURFACE_DW5_Y_OFFSET__SHIFT;
- if (tex->valign_4)
+
+ assert(tex->layout.align_j == 2 || tex->layout.align_j == 4);
+ if (tex->layout.align_j == 4)
dw[5] |= GEN6_SURFACE_DW5_VALIGN_4;
/* do not increment reference count */
diff --git a/src/gallium/drivers/ilo/ilo_gpe_gen7.c b/src/gallium/drivers/ilo/ilo_gpe_gen7.c
index cc666c80953..170ec0319f7 100644
--- a/src/gallium/drivers/ilo/ilo_gpe_gen7.c
+++ b/src/gallium/drivers/ilo/ilo_gpe_gen7.c
@@ -452,11 +452,11 @@ ilo_gpe_init_view_surface_for_texture_gen7(const struct ilo_dev_info *dev,
surface_format = ilo_translate_texture_format(dev, format);
assert(surface_format >= 0);
- width = tex->base.width0;
- height = tex->base.height0;
+ width = tex->layout.width0;
+ height = tex->layout.height0;
depth = (tex->base.target == PIPE_TEXTURE_3D) ?
tex->base.depth0 : num_layers;
- pitch = tex->bo_stride;
+ pitch = tex->layout.bo_stride;
if (surface_type == GEN6_SURFTYPE_CUBE) {
/*
@@ -522,7 +522,7 @@ ilo_gpe_init_view_surface_for_texture_gen7(const struct ilo_dev_info *dev,
/* we lose the capability for layered rendering */
assert(is_rt && num_layers == 1);
- layer_offset = ilo_texture_get_slice_offset(tex,
+ layer_offset = ilo_layout_get_slice_tile_offset(&tex->layout,
first_level, first_layer, &x_offset, &y_offset);
assert(x_offset % 4 == 0);
@@ -569,7 +569,7 @@ ilo_gpe_init_view_surface_for_texture_gen7(const struct ilo_dev_info *dev,
*
* "For linear surfaces, this field (X Offset) must be zero."
*/
- if (tex->tiling == INTEL_TILING_NONE) {
+ if (tex->layout.tiling == INTEL_TILING_NONE) {
if (is_rt) {
const int elem_size = util_format_get_blocksize(format);
assert(layer_offset % elem_size == 0);
@@ -584,7 +584,7 @@ ilo_gpe_init_view_surface_for_texture_gen7(const struct ilo_dev_info *dev,
dw[0] = surface_type << GEN7_SURFACE_DW0_TYPE__SHIFT |
surface_format << GEN7_SURFACE_DW0_FORMAT__SHIFT |
- ilo_gpe_gen6_translate_winsys_tiling(tex->tiling) << 13;
+ ilo_gpe_gen6_translate_winsys_tiling(tex->layout.tiling) << 13;
/*
* From the Ivy Bridge PRM, volume 4 part 1, page 63:
@@ -604,16 +604,19 @@ ilo_gpe_init_view_surface_for_texture_gen7(const struct ilo_dev_info *dev,
assert(depth == 1);
}
- if (tex->valign_4)
+ assert(tex->layout.align_i == 4 || tex->layout.align_i == 8);
+ assert(tex->layout.align_j == 2 || tex->layout.align_j == 4);
+
+ if (tex->layout.align_j == 4)
dw[0] |= GEN7_SURFACE_DW0_VALIGN_4;
- if (tex->halign_8)
+ if (tex->layout.align_i == 8)
dw[0] |= GEN7_SURFACE_DW0_HALIGN_8;
- if (tex->array_spacing_full)
- dw[0] |= GEN7_SURFACE_DW0_ARYSPC_FULL;
- else
+ if (tex->layout.walk == ILO_LAYOUT_WALK_LOD)
dw[0] |= GEN7_SURFACE_DW0_ARYSPC_LOD0;
+ else
+ dw[0] |= GEN7_SURFACE_DW0_ARYSPC_FULL;
if (is_rt)
dw[0] |= GEN7_SURFACE_DW0_RENDER_CACHE_RW;
@@ -637,7 +640,7 @@ ilo_gpe_init_view_surface_for_texture_gen7(const struct ilo_dev_info *dev,
* means the samples are interleaved. The layouts are the same when the
* number of samples is 1.
*/
- if (tex->interleaved && tex->base.nr_samples > 1) {
+ if (tex->layout.interleaved_samples && tex->base.nr_samples > 1) {
assert(!is_rt);
dw[4] |= GEN7_SURFACE_DW4_MSFMT_DEPTH_STENCIL;
}
diff --git a/src/gallium/drivers/ilo/ilo_resource.c b/src/gallium/drivers/ilo/ilo_resource.c
index 812ccafa93b..54d6c368ae9 100644
--- a/src/gallium/drivers/ilo/ilo_resource.c
+++ b/src/gallium/drivers/ilo/ilo_resource.c
@@ -25,43 +25,10 @@
* Chia-I Wu <olv@lunarg.com>
*/
+#include "ilo_layout.h"
#include "ilo_screen.h"
#include "ilo_resource.h"
-/* use PIPE_BIND_CUSTOM to indicate MCS */
-#define ILO_BIND_MCS PIPE_BIND_CUSTOM
-
-struct tex_layout {
- const struct ilo_dev_info *dev;
- const struct pipe_resource *templ;
-
- bool has_depth, has_stencil;
- bool hiz, separate_stencil;
-
- enum pipe_format format;
- unsigned block_width, block_height, block_size;
- bool compressed;
-
- enum intel_tiling_mode tiling;
- unsigned valid_tilings; /* bitmask of valid tiling modes */
-
- bool array_spacing_full;
- bool interleaved;
-
- struct {
- int w, h, d;
- struct ilo_texture_slice *slices;
- } levels[PIPE_MAX_TEXTURE_LEVELS];
-
- int align_i, align_j;
- int qpitch;
-
- int width, height;
-
- int bo_stride, bo_height;
- int hiz_stride, hiz_height;
-};
-
/*
* From the Ivy Bridge PRM, volume 1 part 1, page 105:
*
@@ -69,7 +36,7 @@ struct tex_layout {
* surfaces are also restricted to a maximum size in bytes. This
* maximum is 2 GB for all products and all surface types."
*/
-static const size_t max_resource_size = 1u << 31;
+static const size_t ilo_max_resource_size = 1u << 31;
static const char *
resource_get_bo_name(const struct pipe_resource *templ)
@@ -119,966 +86,6 @@ resource_get_bo_initial_domain(const struct pipe_resource *templ)
}
static void
-tex_layout_init_qpitch(struct tex_layout *layout)
-{
- const struct pipe_resource *templ = layout->templ;
- int h0, h1;
-
- if (templ->array_size <= 1)
- return;
-
- h0 = align(layout->levels[0].h, layout->align_j);
-
- if (!layout->array_spacing_full) {
- layout->qpitch = h0;
- return;
- }
-
- h1 = align(layout->levels[1].h, layout->align_j);
-
- /*
- * From the Sandy Bridge PRM, volume 1 part 1, page 115:
- *
- * "The following equation is used for surface formats other than
- * compressed textures:
- *
- * QPitch = (h0 + h1 + 11j)"
- *
- * "The equation for compressed textures (BC* and FXT1 surface formats)
- * follows:
- *
- * QPitch = (h0 + h1 + 11j) / 4"
- *
- * "[DevSNB] Errata: Sampler MSAA Qpitch will be 4 greater than the
- * value calculated in the equation above, for every other odd Surface
- * Height starting from 1 i.e. 1,5,9,13"
- *
- * From the Ivy Bridge PRM, volume 1 part 1, page 111-112:
- *
- * "If Surface Array Spacing is set to ARYSPC_FULL (note that the depth
- * buffer and stencil buffer have an implied value of ARYSPC_FULL):
- *
- * QPitch = (h0 + h1 + 12j)
- * QPitch = (h0 + h1 + 12j) / 4 (compressed)
- *
- * (There are many typos or missing words here...)"
- *
- * To access the N-th slice, an offset of (Stride * QPitch * N) is added to
- * the base address. The PRM divides QPitch by 4 for compressed formats
- * because the block height for those formats are 4, and it wants QPitch to
- * mean the number of memory rows, as opposed to texel rows, between
- * slices. Since we use texel rows in tex->slice_offsets, we do not need
- * to divide QPitch by 4.
- */
- layout->qpitch = h0 + h1 +
- ((layout->dev->gen >= ILO_GEN(7)) ? 12 : 11) * layout->align_j;
-
- if (layout->dev->gen == ILO_GEN(6) && templ->nr_samples > 1 &&
- templ->height0 % 4 == 1)
- layout->qpitch += 4;
-}
-
-static void
-tex_layout_init_alignments(struct tex_layout *layout)
-{
- const struct pipe_resource *templ = layout->templ;
-
- /*
- * From the Sandy Bridge PRM, volume 1 part 1, page 113:
- *
- * "surface format align_i align_j
- * YUV 4:2:2 formats 4 *see below
- * BC1-5 4 4
- * FXT1 8 4
- * all other formats 4 *see below"
- *
- * "- align_j = 4 for any depth buffer
- * - align_j = 2 for separate stencil buffer
- * - align_j = 4 for any render target surface is multisampled (4x)
- * - align_j = 4 for any render target surface with Surface Vertical
- * Alignment = VALIGN_4
- * - align_j = 2 for any render target surface with Surface Vertical
- * Alignment = VALIGN_2
- * - align_j = 2 for all other render target surface
- * - align_j = 2 for any sampling engine surface with Surface Vertical
- * Alignment = VALIGN_2
- * - align_j = 4 for any sampling engine surface with Surface Vertical
- * Alignment = VALIGN_4"
- *
- * From the Sandy Bridge PRM, volume 4 part 1, page 86:
- *
- * "This field (Surface Vertical Alignment) must be set to VALIGN_2 if
- * the Surface Format is 96 bits per element (BPE)."
- *
- * They can be rephrased as
- *
- * align_i align_j
- * compressed formats block width block height
- * PIPE_FORMAT_S8_UINT 4 2
- * other depth/stencil formats 4 4
- * 4x multisampled 4 4
- * bpp 96 4 2
- * others 4 2 or 4
- */
-
- /*
- * From the Ivy Bridge PRM, volume 1 part 1, page 110:
- *
- * "surface defined by surface format align_i align_j
- * 3DSTATE_DEPTH_BUFFER D16_UNORM 8 4
- * not D16_UNORM 4 4
- * 3DSTATE_STENCIL_BUFFER N/A 8 8
- * SURFACE_STATE BC*, ETC*, EAC* 4 4
- * FXT1 8 4
- * all others (set by SURFACE_STATE)"
- *
- * From the Ivy Bridge PRM, volume 4 part 1, page 63:
- *
- * "- This field (Surface Vertical Aligment) is intended to be set to
- * VALIGN_4 if the surface was rendered as a depth buffer, for a
- * multisampled (4x) render target, or for a multisampled (8x)
- * render target, since these surfaces support only alignment of 4.
- * - Use of VALIGN_4 for other surfaces is supported, but uses more
- * memory.
- * - This field must be set to VALIGN_4 for all tiled Y Render Target
- * surfaces.
- * - Value of 1 is not supported for format YCRCB_NORMAL (0x182),
- * YCRCB_SWAPUVY (0x183), YCRCB_SWAPUV (0x18f), YCRCB_SWAPY (0x190)
- * - If Number of Multisamples is not MULTISAMPLECOUNT_1, this field
- * must be set to VALIGN_4."
- * - VALIGN_4 is not supported for surface format R32G32B32_FLOAT."
- *
- * "- This field (Surface Horizontal Aligment) is intended to be set to
- * HALIGN_8 only if the surface was rendered as a depth buffer with
- * Z16 format or a stencil buffer, since these surfaces support only
- * alignment of 8.
- * - Use of HALIGN_8 for other surfaces is supported, but uses more
- * memory.
- * - This field must be set to HALIGN_4 if the Surface Format is BC*.
- * - This field must be set to HALIGN_8 if the Surface Format is
- * FXT1."
- *
- * They can be rephrased as
- *
- * align_i align_j
- * compressed formats block width block height
- * PIPE_FORMAT_Z16_UNORM 8 4
- * PIPE_FORMAT_S8_UINT 8 8
- * other depth/stencil formats 4 or 8 4
- * 2x or 4x multisampled 4 or 8 4
- * tiled Y 4 or 8 4 (if rt)
- * PIPE_FORMAT_R32G32B32_FLOAT 4 or 8 2
- * others 4 or 8 2 or 4
- */
-
- if (layout->compressed) {
- /* this happens to be the case */
- layout->align_i = layout->block_width;
- layout->align_j = layout->block_height;
- }
- else if (layout->has_depth || layout->has_stencil) {
- if (layout->dev->gen >= ILO_GEN(7)) {
- switch (layout->format) {
- case PIPE_FORMAT_Z16_UNORM:
- layout->align_i = 8;
- layout->align_j = 4;
- break;
- case PIPE_FORMAT_S8_UINT:
- layout->align_i = 8;
- layout->align_j = 8;
- break;
- default:
- layout->align_i = 4;
- layout->align_j = 4;
- break;
- }
- }
- else {
- switch (layout->format) {
- case PIPE_FORMAT_S8_UINT:
- layout->align_i = 4;
- layout->align_j = 2;
- break;
- default:
- layout->align_i = 4;
- layout->align_j = 4;
- break;
- }
- }
- }
- else {
- const bool valign_4 = (templ->nr_samples > 1) ||
- (layout->dev->gen >= ILO_GEN(7) &&
- layout->tiling == INTEL_TILING_Y &&
- (templ->bind & PIPE_BIND_RENDER_TARGET));
-
- if (valign_4)
- assert(layout->block_size != 12);
-
- layout->align_i = 4;
- layout->align_j = (valign_4) ? 4 : 2;
- }
-
- /*
- * the fact that align i and j are multiples of block width and height
- * respectively is what makes the size of the bo a multiple of the block
- * size, slices start at block boundaries, and many of the computations
- * work.
- */
- assert(layout->align_i % layout->block_width == 0);
- assert(layout->align_j % layout->block_height == 0);
-
- /* make sure align() works */
- assert(util_is_power_of_two(layout->align_i) &&
- util_is_power_of_two(layout->align_j));
- assert(util_is_power_of_two(layout->block_width) &&
- util_is_power_of_two(layout->block_height));
-}
-
-static void
-tex_layout_init_levels(struct tex_layout *layout)
-{
- const struct pipe_resource *templ = layout->templ;
- int last_level, lv;
-
- last_level = templ->last_level;
-
- /* need at least 2 levels to compute full qpitch */
- if (last_level == 0 && templ->array_size > 1 && layout->array_spacing_full)
- last_level++;
-
- /* compute mip level sizes */
- for (lv = 0; lv <= last_level; lv++) {
- int w, h, d;
-
- w = u_minify(templ->width0, lv);
- h = u_minify(templ->height0, lv);
- d = u_minify(templ->depth0, lv);
-
- /*
- * From the Sandy Bridge PRM, volume 1 part 1, page 114:
- *
- * "The dimensions of the mip maps are first determined by applying
- * the sizing algorithm presented in Non-Power-of-Two Mipmaps
- * above. Then, if necessary, they are padded out to compression
- * block boundaries."
- */
- w = align(w, layout->block_width);
- h = align(h, layout->block_height);
-
- /*
- * From the Sandy Bridge PRM, volume 1 part 1, page 111:
- *
- * "If the surface is multisampled (4x), these values must be
- * adjusted as follows before proceeding:
- *
- * W_L = ceiling(W_L / 2) * 4
- * H_L = ceiling(H_L / 2) * 4"
- *
- * From the Ivy Bridge PRM, volume 1 part 1, page 108:
- *
- * "If the surface is multisampled and it is a depth or stencil
- * surface or Multisampled Surface StorageFormat in SURFACE_STATE
- * is MSFMT_DEPTH_STENCIL, W_L and H_L must be adjusted as follows
- * before proceeding:
- *
- * #samples W_L = H_L =
- * 2 ceiling(W_L / 2) * 4 HL [no adjustment]
- * 4 ceiling(W_L / 2) * 4 ceiling(H_L / 2) * 4
- * 8 ceiling(W_L / 2) * 8 ceiling(H_L / 2) * 4
- * 16 ceiling(W_L / 2) * 8 ceiling(H_L / 2) * 8"
- *
- * For interleaved samples (4x), where pixels
- *
- * (x, y ) (x+1, y )
- * (x, y+1) (x+1, y+1)
- *
- * would be is occupied by
- *
- * (x, y , si0) (x+1, y , si0) (x, y , si1) (x+1, y , si1)
- * (x, y+1, si0) (x+1, y+1, si0) (x, y+1, si1) (x+1, y+1, si1)
- * (x, y , si2) (x+1, y , si2) (x, y , si3) (x+1, y , si3)
- * (x, y+1, si2) (x+1, y+1, si2) (x, y+1, si3) (x+1, y+1, si3)
- *
- * Thus the need to
- *
- * w = align(w, 2) * 2;
- * y = align(y, 2) * 2;
- */
- if (layout->interleaved) {
- switch (templ->nr_samples) {
- case 0:
- case 1:
- break;
- case 2:
- w = align(w, 2) * 2;
- break;
- case 4:
- w = align(w, 2) * 2;
- h = align(h, 2) * 2;
- break;
- case 8:
- w = align(w, 2) * 4;
- h = align(h, 2) * 2;
- break;
- case 16:
- w = align(w, 2) * 4;
- h = align(h, 2) * 4;
- break;
- default:
- assert(!"unsupported sample count");
- break;
- }
- }
-
- layout->levels[lv].w = w;
- layout->levels[lv].h = h;
- layout->levels[lv].d = d;
- }
-}
-
-static void
-tex_layout_init_spacing(struct tex_layout *layout)
-{
- const struct pipe_resource *templ = layout->templ;
-
- if (layout->dev->gen >= ILO_GEN(7)) {
- /*
- * It is not explicitly states, but render targets are expected to be
- * UMS/CMS (samples non-interleaved) and depth/stencil buffers are
- * expected to be IMS (samples interleaved).
- *
- * See "Multisampled Surface Storage Format" field of SURFACE_STATE.
- */
- if (layout->has_depth || layout->has_stencil) {
- layout->interleaved = true;
-
- /*
- * From the Ivy Bridge PRM, volume 1 part 1, page 111:
- *
- * "note that the depth buffer and stencil buffer have an implied
- * value of ARYSPC_FULL"
- */
- layout->array_spacing_full = true;
- }
- else {
- layout->interleaved = false;
-
- /*
- * From the Ivy Bridge PRM, volume 4 part 1, page 66:
- *
- * "If Multisampled Surface Storage Format is MSFMT_MSS and
- * Number of Multisamples is not MULTISAMPLECOUNT_1, this field
- * (Surface Array Spacing) must be set to ARYSPC_LOD0."
- *
- * As multisampled resources are not mipmapped, we never use
- * ARYSPC_FULL for them.
- */
- if (templ->nr_samples > 1)
- assert(templ->last_level == 0);
- layout->array_spacing_full = (templ->last_level > 0);
- }
- }
- else {
- /* GEN6 supports only interleaved samples */
- layout->interleaved = true;
-
- /*
- * From the Sandy Bridge PRM, volume 1 part 1, page 115:
- *
- * "The separate stencil buffer does not support mip mapping, thus
- * the storage for LODs other than LOD 0 is not needed. The
- * following QPitch equation applies only to the separate stencil
- * buffer:
- *
- * QPitch = h_0"
- *
- * GEN6 does not support compact spacing otherwise.
- */
- layout->array_spacing_full = (layout->format != PIPE_FORMAT_S8_UINT);
- }
-}
-
-static void
-tex_layout_init_tiling(struct tex_layout *layout)
-{
- const struct pipe_resource *templ = layout->templ;
- const enum pipe_format format = layout->format;
- const unsigned tile_none = 1 << INTEL_TILING_NONE;
- const unsigned tile_x = 1 << INTEL_TILING_X;
- const unsigned tile_y = 1 << INTEL_TILING_Y;
- unsigned valid_tilings = tile_none | tile_x | tile_y;
-
- /*
- * From the Sandy Bridge PRM, volume 1 part 2, page 32:
- *
- * "Display/Overlay Y-Major not supported.
- * X-Major required for Async Flips"
- */
- if (unlikely(templ->bind & PIPE_BIND_SCANOUT))
- valid_tilings &= tile_x;
-
- /*
- * From the Sandy Bridge PRM, volume 3 part 2, page 158:
- *
- * "The cursor surface address must be 4K byte aligned. The cursor must
- * be in linear memory, it cannot be tiled."
- */
- if (unlikely(templ->bind & (PIPE_BIND_CURSOR | PIPE_BIND_LINEAR)))
- valid_tilings &= tile_none;
-
- /*
- * From the Ivy Bridge PRM, volume 4 part 1, page 76:
- *
- * "The MCS surface must be stored as Tile Y."
- */
- if (templ->bind & ILO_BIND_MCS)
- valid_tilings &= tile_y;
-
- /*
- * From the Sandy Bridge PRM, volume 2 part 1, page 318:
- *
- * "[DevSNB+]: This field (Tiled Surface) must be set to TRUE. Linear
- * Depth Buffer is not supported."
- *
- * "The Depth Buffer, if tiled, must use Y-Major tiling."
- *
- * From the Sandy Bridge PRM, volume 1 part 2, page 22:
- *
- * "W-Major Tile Format is used for separate stencil."
- *
- * Since the HW does not support W-tiled fencing, we have to do it in the
- * driver.
- */
- if (templ->bind & PIPE_BIND_DEPTH_STENCIL) {
- switch (format) {
- case PIPE_FORMAT_S8_UINT:
- valid_tilings &= tile_none;
- break;
- default:
- valid_tilings &= tile_y;
- break;
- }
- }
-
- if (templ->bind & PIPE_BIND_RENDER_TARGET) {
- /*
- * From the Sandy Bridge PRM, volume 1 part 2, page 32:
- *
- * "NOTE: 128BPE Format Color buffer ( render target ) MUST be
- * either TileX or Linear."
- */
- if (layout->block_size == 16)
- valid_tilings &= ~tile_y;
-
- /*
- * From the Ivy Bridge PRM, volume 4 part 1, page 63:
- *
- * "This field (Surface Vertical Aligment) must be set to VALIGN_4
- * for all tiled Y Render Target surfaces."
- *
- * "VALIGN_4 is not supported for surface format R32G32B32_FLOAT."
- */
- if (layout->dev->gen >= ILO_GEN(7) && layout->block_size == 12)
- valid_tilings &= ~tile_y;
- }
-
- /* no conflicting binding flags */
- assert(valid_tilings);
-
- layout->valid_tilings = valid_tilings;
-
- if (templ->bind & (PIPE_BIND_RENDER_TARGET | PIPE_BIND_SAMPLER_VIEW)) {
- /*
- * heuristically set a minimum width/height for enabling tiling
- */
- if (templ->width0 < 64 && (valid_tilings & ~tile_x))
- valid_tilings &= ~tile_x;
-
- if ((templ->width0 < 32 || templ->height0 < 16) &&
- (templ->width0 < 16 || templ->height0 < 32) &&
- (valid_tilings & ~tile_y))
- valid_tilings &= ~tile_y;
- }
- else {
- /* force linear if we are not sure where the texture is bound to */
- if (valid_tilings & tile_none)
- valid_tilings &= tile_none;
- }
-
- /* prefer tiled over linear */
- if (valid_tilings & tile_y)
- layout->tiling = INTEL_TILING_Y;
- else if (valid_tilings & tile_x)
- layout->tiling = INTEL_TILING_X;
- else
- layout->tiling = INTEL_TILING_NONE;
-}
-
-static void
-tex_layout_init_format(struct tex_layout *layout)
-{
- const struct pipe_resource *templ = layout->templ;
- enum pipe_format format;
-
- switch (templ->format) {
- case PIPE_FORMAT_ETC1_RGB8:
- format = PIPE_FORMAT_R8G8B8X8_UNORM;
- break;
- case PIPE_FORMAT_Z24_UNORM_S8_UINT:
- if (layout->separate_stencil)
- format = PIPE_FORMAT_Z24X8_UNORM;
- else
- format = templ->format;
- break;
- case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
- if (layout->separate_stencil)
- format = PIPE_FORMAT_Z32_FLOAT;
- else
- format = templ->format;
- break;
- default:
- format = templ->format;
- break;
- }
-
- layout->format = format;
-
- layout->block_width = util_format_get_blockwidth(format);
- layout->block_height = util_format_get_blockheight(format);
- layout->block_size = util_format_get_blocksize(format);
- layout->compressed = util_format_is_compressed(format);
-}
-
-static void
-tex_layout_init_hiz(struct tex_layout *layout)
-{
- const struct pipe_resource *templ = layout->templ;
- const struct util_format_description *desc;
-
- desc = util_format_description(templ->format);
- layout->has_depth = util_format_has_depth(desc);
- layout->has_stencil = util_format_has_stencil(desc);
-
- if (!layout->has_depth)
- return;
-
- layout->hiz = true;
-
- /* no point in having HiZ */
- if (templ->usage == PIPE_USAGE_STAGING)
- layout->hiz = false;
-
- if (layout->dev->gen == ILO_GEN(6)) {
- /*
- * From the Sandy Bridge PRM, volume 2 part 1, page 312:
- *
- * "The hierarchical depth buffer does not support the LOD field, it
- * is assumed by hardware to be zero. A separate hierarachical
- * depth buffer is required for each LOD used, and the
- * corresponding buffer's state delivered to hardware each time a
- * new depth buffer state with modified LOD is delivered."
- *
- * But we have a stronger requirement. Because of layer offsetting
- * (check out the callers of ilo_texture_get_slice_offset()), we already
- * have to require the texture to be non-mipmapped and non-array.
- */
- if (templ->last_level > 0 || templ->array_size > 1 || templ->depth0 > 1)
- layout->hiz = false;
- }
-
- if (ilo_debug & ILO_DEBUG_NOHIZ)
- layout->hiz = false;
-
- if (layout->has_stencil) {
- /*
- * From the Sandy Bridge PRM, volume 2 part 1, page 317:
- *
- * "This field (Separate Stencil Buffer Enable) must be set to the
- * same value (enabled or disabled) as Hierarchical Depth Buffer
- * Enable."
- *
- * GEN7+ requires separate stencil buffers.
- */
- if (layout->dev->gen >= ILO_GEN(7))
- layout->separate_stencil = true;
- else
- layout->separate_stencil = layout->hiz;
-
- if (layout->separate_stencil)
- layout->has_stencil = false;
- }
-}
-
-static bool
-tex_layout_init(struct tex_layout *layout,
- struct pipe_screen *screen,
- const struct pipe_resource *templ,
- struct ilo_texture_slice **slices)
-{
- struct ilo_screen *is = ilo_screen(screen);
-
- memset(layout, 0, sizeof(*layout));
-
- layout->dev = &is->dev;
- layout->templ = templ;
-
- /* note that there are dependencies between these functions */
- tex_layout_init_hiz(layout);
- tex_layout_init_format(layout);
- tex_layout_init_tiling(layout);
- tex_layout_init_spacing(layout);
- tex_layout_init_levels(layout);
- tex_layout_init_alignments(layout);
- tex_layout_init_qpitch(layout);
-
- if (templ->flags & PIPE_RESOURCE_FLAG_MAP_PERSISTENT) {
- /* require on-the-fly tiling/untiling or format conversion */
- if (layout->separate_stencil ||
- layout->format == PIPE_FORMAT_S8_UINT ||
- layout->format != templ->format)
- return false;
- }
-
- if (slices) {
- int lv;
-
- for (lv = 0; lv <= templ->last_level; lv++)
- layout->levels[lv].slices = slices[lv];
- }
-
- return true;
-}
-
-static void
-tex_layout_align(struct tex_layout *layout)
-{
- int align_w = 1, align_h = 1, pad_h = 0;
-
- /*
- * From the Sandy Bridge PRM, volume 1 part 1, page 118:
- *
- * "To determine the necessary padding on the bottom and right side of
- * the surface, refer to the table in Section 7.18.3.4 for the i and j
- * parameters for the surface format in use. The surface must then be
- * extended to the next multiple of the alignment unit size in each
- * dimension, and all texels contained in this extended surface must
- * have valid GTT entries."
- *
- * "For cube surfaces, an additional two rows of padding are required
- * at the bottom of the surface. This must be ensured regardless of
- * whether the surface is stored tiled or linear. This is due to the
- * potential rotation of cache line orientation from memory to cache."
- *
- * "For compressed textures (BC* and FXT1 surface formats), padding at
- * the bottom of the surface is to an even compressed row, which is
- * equal to a multiple of 8 uncompressed texel rows. Thus, for padding
- * purposes, these surfaces behave as if j = 8 only for surface
- * padding purposes. The value of 4 for j still applies for mip level
- * alignment and QPitch calculation."
- */
- if (layout->templ->bind & PIPE_BIND_SAMPLER_VIEW) {
- align_w = MAX2(align_w, layout->align_i);
- align_h = MAX2(align_h, layout->align_j);
-
- if (layout->templ->target == PIPE_TEXTURE_CUBE)
- pad_h += 2;
-
- if (layout->compressed)
- align_h = MAX2(align_h, layout->align_j * 2);
- }
-
- /*
- * From the Sandy Bridge PRM, volume 1 part 1, page 118:
- *
- * "If the surface contains an odd number of rows of data, a final row
- * below the surface must be allocated."
- */
- if (layout->templ->bind & PIPE_BIND_RENDER_TARGET)
- align_h = MAX2(align_h, 2);
-
- /*
- * Depth Buffer Clear/Resolve works in 8x4 sample blocks. In
- * ilo_texture_can_enable_hiz(), we always return true for the first slice.
- * To avoid out-of-bound access, we have to pad.
- */
- if (layout->hiz) {
- align_w = MAX2(align_w, 8);
- align_h = MAX2(align_h, 4);
- }
-
- layout->width = align(layout->width, align_w);
- layout->height = align(layout->height + pad_h, align_h);
-}
-
-/**
- * Layout a 2D texture.
- */
-static void
-tex_layout_2d(struct tex_layout *layout)
-{
- const struct pipe_resource *templ = layout->templ;
- unsigned int level_x, level_y, num_slices;
- int lv;
-
- level_x = 0;
- level_y = 0;
- for (lv = 0; lv <= templ->last_level; lv++) {
- const unsigned int level_w = layout->levels[lv].w;
- const unsigned int level_h = layout->levels[lv].h;
- int slice;
-
- /* set slice offsets */
- if (layout->levels[lv].slices) {
- for (slice = 0; slice < templ->array_size; slice++) {
- layout->levels[lv].slices[slice].x = level_x;
- /* slices are qpitch apart in Y-direction */
- layout->levels[lv].slices[slice].y =
- level_y + layout->qpitch * slice;
- }
- }
-
- /* extend the size of the monolithic bo to cover this mip level */
- if (layout->width < level_x + level_w)
- layout->width = level_x + level_w;
- if (layout->height < level_y + level_h)
- layout->height = level_y + level_h;
-
- /* MIPLAYOUT_BELOW */
- if (lv == 1)
- level_x += align(level_w, layout->align_i);
- else
- level_y += align(level_h, layout->align_j);
- }
-
- num_slices = templ->array_size;
- /* samples of the same index are stored in a slice */
- if (templ->nr_samples > 1 && !layout->interleaved)
- num_slices *= templ->nr_samples;
-
- /* we did not take slices into consideration in the computation above */
- layout->height += layout->qpitch * (num_slices - 1);
-
- tex_layout_align(layout);
-}
-
-/**
- * Layout a 3D texture.
- */
-static void
-tex_layout_3d(struct tex_layout *layout)
-{
- const struct pipe_resource *templ = layout->templ;
- unsigned int level_y;
- int lv;
-
- level_y = 0;
- for (lv = 0; lv <= templ->last_level; lv++) {
- const unsigned int level_w = layout->levels[lv].w;
- const unsigned int level_h = layout->levels[lv].h;
- const unsigned int level_d = layout->levels[lv].d;
- const unsigned int slice_pitch = align(level_w, layout->align_i);
- const unsigned int slice_qpitch = align(level_h, layout->align_j);
- const unsigned int num_slices_per_row = 1 << lv;
- int slice;
-
- for (slice = 0; slice < level_d; slice += num_slices_per_row) {
- int i;
-
- /* set slice offsets */
- if (layout->levels[lv].slices) {
- for (i = 0; i < num_slices_per_row && slice + i < level_d; i++) {
- layout->levels[lv].slices[slice + i].x = slice_pitch * i;
- layout->levels[lv].slices[slice + i].y = level_y;
- }
- }
-
- /* move on to the next slice row */
- level_y += slice_qpitch;
- }
-
- /* rightmost slice */
- slice = MIN2(num_slices_per_row, level_d) - 1;
-
- /* extend the size of the monolithic bo to cover this slice */
- if (layout->width < slice_pitch * slice + level_w)
- layout->width = slice_pitch * slice + level_w;
- if (lv == templ->last_level)
- layout->height = (level_y - slice_qpitch) + level_h;
- }
-
- tex_layout_align(layout);
-}
-
-/* note that this may force the texture to be linear */
-static bool
-tex_layout_calculate_bo_size(struct tex_layout *layout)
-{
- assert(layout->width % layout->block_width == 0);
- assert(layout->height % layout->block_height == 0);
- assert(layout->qpitch % layout->block_height == 0);
-
- layout->bo_stride =
- (layout->width / layout->block_width) * layout->block_size;
- layout->bo_height = layout->height / layout->block_height;
-
- while (true) {
- int w = layout->bo_stride, h = layout->bo_height;
- int align_w, align_h;
-
- /*
- * From the Haswell PRM, volume 5, page 163:
- *
- * "For linear surfaces, additional padding of 64 bytes is required
- * at the bottom of the surface. This is in addition to the padding
- * required above."
- */
- if (layout->dev->gen >= ILO_GEN(7.5) &&
- (layout->templ->bind & PIPE_BIND_SAMPLER_VIEW) &&
- layout->tiling == INTEL_TILING_NONE) {
- layout->bo_height +=
- (64 + layout->bo_stride - 1) / layout->bo_stride;
- }
-
- /*
- * From the Sandy Bridge PRM, volume 4 part 1, page 81:
- *
- * "- For linear render target surfaces, the pitch must be a
- * multiple of the element size for non-YUV surface formats.
- * Pitch must be a multiple of 2 * element size for YUV surface
- * formats.
- * - For other linear surfaces, the pitch can be any multiple of
- * bytes.
- * - For tiled surfaces, the pitch must be a multiple of the tile
- * width."
- *
- * Different requirements may exist when the bo is used in different
- * places, but our alignments here should be good enough that we do not
- * need to check layout->templ->bind.
- */
- switch (layout->tiling) {
- case INTEL_TILING_X:
- align_w = 512;
- align_h = 8;
- break;
- case INTEL_TILING_Y:
- align_w = 128;
- align_h = 32;
- break;
- default:
- if (layout->format == PIPE_FORMAT_S8_UINT) {
- /*
- * From the Sandy Bridge PRM, volume 1 part 2, page 22:
- *
- * "A 4KB tile is subdivided into 8-high by 8-wide array of
- * Blocks for W-Major Tiles (W Tiles). Each Block is 8 rows by 8
- * bytes."
- *
- * Since we asked for INTEL_TILING_NONE instead of the non-existent
- * INTEL_TILING_W, we want to align to W tiles here.
- */
- align_w = 64;
- align_h = 64;
- }
- else {
- /* some good enough values */
- align_w = 64;
- align_h = 2;
- }
- break;
- }
-
- w = align(w, align_w);
- h = align(h, align_h);
-
- /* make sure the bo is mappable */
- if (layout->tiling != INTEL_TILING_NONE) {
- /*
- * Usually only the first 256MB of the GTT is mappable.
- *
- * See also how intel_context::max_gtt_map_object_size is calculated.
- */
- const size_t mappable_gtt_size = 256 * 1024 * 1024;
-
- /*
- * Be conservative. We may be able to switch from VALIGN_4 to
- * VALIGN_2 if the layout was Y-tiled, but let's keep it simple.
- */
- if (mappable_gtt_size / w / 4 < h) {
- if (layout->valid_tilings & (1 << INTEL_TILING_NONE)) {
- layout->tiling = INTEL_TILING_NONE;
- continue;
- }
- else {
- ilo_warn("cannot force texture to be linear\n");
- }
- }
- }
-
- layout->bo_stride = w;
- layout->bo_height = h;
- break;
- }
-
- return (layout->bo_height <= max_resource_size / layout->bo_stride);
-}
-
-static void
-tex_layout_calculate_hiz_size(struct tex_layout *layout)
-{
- const struct pipe_resource *templ = layout->templ;
- const int hz_align_j = 8;
- int hz_width, hz_height;
-
- if (!layout->hiz)
- return;
-
- /*
- * See the Sandy Bridge PRM, volume 2 part 1, page 312, and the Ivy Bridge
- * PRM, volume 2 part 1, page 312-313.
- *
- * It seems HiZ buffer is aligned to 8x8, with every two rows packed into a
- * memory row.
- */
-
- hz_width = align(layout->levels[0].w, 16);
-
- if (templ->target == PIPE_TEXTURE_3D) {
- unsigned lv;
-
- hz_height = 0;
-
- for (lv = 0; lv <= templ->last_level; lv++) {
- const unsigned h = align(layout->levels[lv].h, hz_align_j);
- hz_height += h * layout->levels[lv].d;
- }
-
- hz_height /= 2;
- }
- else {
- const unsigned h0 = align(layout->levels[0].h, hz_align_j);
- unsigned hz_qpitch = h0;
-
- if (layout->array_spacing_full) {
- const unsigned h1 = align(layout->levels[1].h, hz_align_j);
- const unsigned htail =
- ((layout->dev->gen >= ILO_GEN(7)) ? 12 : 11) * hz_align_j;
-
- hz_qpitch += h1 + htail;
- }
-
- hz_height = hz_qpitch * templ->array_size / 2;
-
- if (layout->dev->gen >= ILO_GEN(7))
- hz_height = align(hz_height, 8);
- }
-
- /* align to Y-tile */
- layout->hiz_stride = align(hz_width, 128);
- layout->hiz_height = align(hz_height, 32);
-}
-
-static void
tex_free_slices(struct ilo_texture *tex)
{
FREE(tex->slices[0]);
@@ -1118,7 +125,6 @@ tex_alloc_slices(struct ilo_texture *tex)
static bool
tex_import_handle(struct ilo_texture *tex,
- const struct tex_layout *layout,
const struct winsys_handle *handle)
{
struct ilo_screen *is = ilo_screen(tex->base.screen);
@@ -1127,20 +133,17 @@ tex_import_handle(struct ilo_texture *tex,
unsigned long pitch;
tex->bo = intel_winsys_import_handle(is->winsys, name, handle,
- tex->bo_height, &tiling, &pitch);
+ tex->layout.bo_height, &tiling, &pitch);
if (!tex->bo)
return false;
- if (!(layout->valid_tilings & (1 << tiling))) {
- ilo_err("imported handle has incompatible tiling\n");
+ if (!ilo_layout_update_for_imported_bo(&tex->layout, tiling, pitch)) {
+ ilo_err("imported handle has incompatible tiling/pitch\n");
intel_bo_unreference(tex->bo);
tex->bo = NULL;
return false;
}
- tex->tiling = tiling;
- tex->bo_stride = pitch;
-
return true;
}
@@ -1152,8 +155,8 @@ tex_create_bo(struct ilo_texture *tex)
const enum intel_domain_flag initial_domain =
resource_get_bo_initial_domain(&tex->base);
- tex->bo = intel_winsys_alloc_bo(is->winsys, name, tex->tiling,
- tex->bo_stride, tex->bo_height, initial_domain);
+ tex->bo = intel_winsys_alloc_bo(is->winsys, name, tex->layout.tiling,
+ tex->layout.bo_stride, tex->layout.bo_height, initial_domain);
return (tex->bo != NULL);
}
@@ -1177,95 +180,34 @@ tex_create_separate_stencil(struct ilo_texture *tex)
tex->separate_s8 = ilo_texture(s8);
- assert(tex->separate_s8->bo_format == PIPE_FORMAT_S8_UINT);
+ assert(tex->separate_s8->layout.format == PIPE_FORMAT_S8_UINT);
return true;
}
static bool
-tex_create_hiz(struct ilo_texture *tex, const struct tex_layout *layout)
+tex_create_hiz(struct ilo_texture *tex)
{
+ const struct pipe_resource *templ = &tex->base;
struct ilo_screen *is = ilo_screen(tex->base.screen);
- const struct pipe_resource *templ = layout->templ;
unsigned lv;
- tex->hiz.bo = intel_winsys_alloc_bo(is->winsys, "hiz texture",
- INTEL_TILING_Y, layout->hiz_stride, layout->hiz_height,
+ tex->aux_bo = intel_winsys_alloc_bo(is->winsys, "hiz texture",
+ INTEL_TILING_Y, tex->layout.aux_stride, tex->layout.aux_height,
INTEL_DOMAIN_RENDER);
- if (!tex->hiz.bo)
+ if (!tex->aux_bo)
return false;
- tex->hiz.bo_stride = layout->hiz_stride;
-
- /*
- * From the Sandy Bridge PRM, volume 2 part 1, page 313-314:
- *
- * "A rectangle primitive representing the clear area is delivered. The
- * primitive must adhere to the following restrictions on size:
- *
- * - If Number of Multisamples is NUMSAMPLES_1, the rectangle must be
- * aligned to an 8x4 pixel block relative to the upper left corner
- * of the depth buffer, and contain an integer number of these pixel
- * blocks, and all 8x4 pixels must be lit.
- *
- * - If Number of Multisamples is NUMSAMPLES_4, the rectangle must be
- * aligned to a 4x2 pixel block (8x4 sample block) relative to the
- * upper left corner of the depth buffer, and contain an integer
- * number of these pixel blocks, and all samples of the 4x2 pixels
- * must be lit
- *
- * - If Number of Multisamples is NUMSAMPLES_8, the rectangle must be
- * aligned to a 2x2 pixel block (8x4 sample block) relative to the
- * upper left corner of the depth buffer, and contain an integer
- * number of these pixel blocks, and all samples of the 2x2 pixels
- * must be list."
- *
- * "The following is required when performing a depth buffer resolve:
- *
- * - A rectangle primitive of the same size as the previous depth
- * buffer clear operation must be delivered, and depth buffer state
- * cannot have changed since the previous depth buffer clear
- * operation."
- *
- * Experiments on Haswell show that depth buffer resolves have the same
- * alignment requirements, and aligning the RECTLIST primitive and
- * 3DSTATE_DRAWING_RECTANGLE alone are not enough. The mipmap size must be
- * aligned.
- */
for (lv = 0; lv <= templ->last_level; lv++) {
- unsigned align_w = 8, align_h = 4;
- unsigned flags = 0;
-
- switch (templ->nr_samples) {
- case 0:
- case 1:
- break;
- case 2:
- align_w /= 2;
- break;
- case 4:
- align_w /= 2;
- align_h /= 2;
- break;
- case 8:
- default:
- align_w /= 4;
- align_h /= 2;
- break;
- }
-
- if (u_minify(templ->width0, lv) % align_w == 0 &&
- u_minify(templ->height0, lv) % align_h == 0) {
- flags |= ILO_TEXTURE_HIZ;
+ if (tex->layout.aux_enables & (1 << lv)) {
+ const unsigned num_slices = (templ->target == PIPE_TEXTURE_3D) ?
+ u_minify(templ->depth0, lv) : templ->array_size;
+ unsigned flags = ILO_TEXTURE_HIZ;
/* this will trigger a HiZ resolve */
if (tex->imported)
flags |= ILO_TEXTURE_CPU_WRITE;
- }
- if (flags) {
- const unsigned num_slices = (templ->target == PIPE_TEXTURE_3D) ?
- u_minify(templ->depth0, lv) : templ->array_size;
ilo_texture_set_slice_flags(tex, lv, 0, num_slices, flags, flags);
}
}
@@ -1274,98 +216,98 @@ tex_create_hiz(struct ilo_texture *tex, const struct tex_layout *layout)
}
static bool
-tex_apply_layout(struct ilo_texture *tex,
- const struct tex_layout *layout,
- const struct winsys_handle *handle)
+tex_create_mcs(struct ilo_texture *tex)
{
- tex->bo_format = layout->format;
- tex->block_width = layout->block_width;
- tex->block_height = layout->block_height;
- tex->block_size = layout->block_size;
+ struct ilo_screen *is = ilo_screen(tex->base.screen);
- tex->tiling = layout->tiling;
- tex->bo_stride = layout->bo_stride;
- tex->bo_height = layout->bo_height;
+ assert(tex->layout.aux_enables == (1 << (tex->base.last_level + 1)) - 1);
- tex->halign_8 = (layout->align_i == 8);
- tex->valign_4 = (layout->align_j == 4);
- tex->array_spacing_full = layout->array_spacing_full;
- tex->interleaved = layout->interleaved;
+ tex->aux_bo = intel_winsys_alloc_bo(is->winsys, "mcs texture",
+ INTEL_TILING_Y, tex->layout.aux_stride, tex->layout.aux_height,
+ INTEL_DOMAIN_RENDER);
+ if (!tex->aux_bo)
+ return false;
+
+ return true;
+}
+
+static void
+tex_destroy(struct ilo_texture *tex)
+{
+ if (tex->aux_bo)
+ intel_bo_unreference(tex->aux_bo);
+
+ if (tex->separate_s8)
+ tex_destroy(tex->separate_s8);
+
+ if (tex->bo)
+ intel_bo_unreference(tex->bo);
+
+ tex_free_slices(tex);
+ FREE(tex);
+}
+
+static bool
+tex_alloc_bos(struct ilo_texture *tex,
+ const struct winsys_handle *handle)
+{
+ struct ilo_screen *is = ilo_screen(tex->base.screen);
if (handle) {
- if (!tex_import_handle(tex, layout, handle))
+ if (!tex_import_handle(tex, handle))
return false;
- }
- else {
+ } else {
if (!tex_create_bo(tex))
return false;
}
/* allocate separate stencil resource */
- if (layout->separate_stencil && !tex_create_separate_stencil(tex))
+ if (tex->layout.separate_stencil && !tex_create_separate_stencil(tex))
return false;
- if (layout->hiz && !tex_create_hiz(tex, layout)) {
- /* Separate Stencil Buffer requires HiZ to be enabled */
- if (layout->dev->gen == ILO_GEN(6) && layout->separate_stencil)
+ switch (tex->layout.aux) {
+ case ILO_LAYOUT_AUX_HIZ:
+ if (!tex_create_hiz(tex)) {
+ /* Separate Stencil Buffer requires HiZ to be enabled */
+ if (is->dev.gen == ILO_GEN(6) && tex->layout.separate_stencil)
+ return false;
+ }
+ break;
+ case ILO_LAYOUT_AUX_MCS:
+ if (!tex_create_mcs(tex))
return false;
+ break;
+ default:
+ break;
}
return true;
}
-/**
- * The texutre is for transfer only. We can define our own layout to save
- * space.
- */
static bool
-tex_apply_transfer_layout(struct ilo_texture *tex)
+tex_init_layout(struct ilo_texture *tex)
{
+ struct ilo_screen *is = ilo_screen(tex->base.screen);
const struct pipe_resource *templ = &tex->base;
- const unsigned num_slices = (templ->target == PIPE_TEXTURE_3D) ?
- templ->depth0 : templ->array_size;
- unsigned slice_width, slice_height, i;
-
- assert(templ->last_level == 0);
-
- tex->bo_format = templ->format;
- tex->block_width = util_format_get_blockwidth(templ->format);
- tex->block_height = util_format_get_blockheight(templ->format);
- tex->block_size = util_format_get_blocksize(templ->format);
-
- assert(util_is_power_of_two(tex->block_width) &&
- util_is_power_of_two(tex->block_height));
-
- /* use packed layout */
- slice_width = align(templ->width0, tex->block_width);
- slice_height = align(templ->height0, tex->block_height);
- for (i = 0; i < num_slices; i++) {
- tex->slices[0][i].x = 0;
- tex->slices[0][i].y = slice_height * i;
- }
+ struct ilo_layout *layout = &tex->layout;
- tex->tiling = INTEL_TILING_NONE;
- tex->bo_stride = (slice_width / tex->block_width) * tex->block_size;
- tex->bo_stride = align(tex->bo_stride, 64);
- tex->bo_height = (slice_height / tex->block_height) * num_slices;
+ ilo_layout_init(layout, &is->dev, templ);
- return tex_create_bo(tex);
-}
-
-static void
-tex_destroy(struct ilo_texture *tex)
-{
- if (tex->hiz.bo)
- intel_bo_unreference(tex->hiz.bo);
+ if (layout->bo_height > ilo_max_resource_size / layout->bo_stride)
+ return false;
- if (tex->separate_s8)
- tex_destroy(tex->separate_s8);
+ if (templ->flags & PIPE_RESOURCE_FLAG_MAP_PERSISTENT) {
+ /* require on-the-fly tiling/untiling or format conversion */
+ if (layout->separate_stencil ||
+ layout->format == PIPE_FORMAT_S8_UINT ||
+ layout->format != templ->format)
+ return false;
+ }
- if (tex->bo)
- intel_bo_unreference(tex->bo);
+ if (!tex_alloc_slices(tex))
+ return false;
- tex_free_slices(tex);
- FREE(tex);
+ return true;
}
static struct pipe_resource *
@@ -1373,9 +315,7 @@ tex_create(struct pipe_screen *screen,
const struct pipe_resource *templ,
const struct winsys_handle *handle)
{
- struct tex_layout layout;
struct ilo_texture *tex;
- bool transfer_only;
tex = CALLOC_STRUCT(ilo_texture);
if (!tex)
@@ -1385,56 +325,14 @@ tex_create(struct pipe_screen *screen,
tex->base.screen = screen;
pipe_reference_init(&tex->base.reference, 1);
- if (!tex_alloc_slices(tex)) {
- FREE(tex);
- return NULL;
- }
-
tex->imported = (handle != NULL);
- /* use transfer layout when the texture is never bound to GPU */
- transfer_only = !(templ->bind & ~(PIPE_BIND_TRANSFER_WRITE |
- PIPE_BIND_TRANSFER_READ));
- if (transfer_only && templ->last_level == 0) {
- if (!tex_apply_transfer_layout(tex)) {
- tex_destroy(tex);
- return NULL;
- }
-
- return &tex->base;
- }
-
- if (!tex_layout_init(&layout, screen, templ, tex->slices)) {
- tex_destroy(tex);
- return NULL;
- }
-
- switch (templ->target) {
- case PIPE_TEXTURE_1D:
- case PIPE_TEXTURE_2D:
- case PIPE_TEXTURE_CUBE:
- case PIPE_TEXTURE_RECT:
- case PIPE_TEXTURE_1D_ARRAY:
- case PIPE_TEXTURE_2D_ARRAY:
- case PIPE_TEXTURE_CUBE_ARRAY:
- tex_layout_2d(&layout);
- break;
- case PIPE_TEXTURE_3D:
- tex_layout_3d(&layout);
- break;
- default:
- assert(!"unknown resource target");
- break;
- }
-
- if (!tex_layout_calculate_bo_size(&layout)) {
- tex_destroy(tex);
+ if (!tex_init_layout(tex)) {
+ FREE(tex);
return NULL;
}
- tex_layout_calculate_hiz_size(&layout);
-
- if (!tex_apply_layout(tex, &layout, handle)) {
+ if (!tex_alloc_bos(tex, handle)) {
tex_destroy(tex);
return NULL;
}
@@ -1448,8 +346,8 @@ tex_get_handle(struct ilo_texture *tex, struct winsys_handle *handle)
struct ilo_screen *is = ilo_screen(tex->base.screen);
int err;
- err = intel_winsys_export_handle(is->winsys, tex->bo,
- tex->tiling, tex->bo_stride, tex->bo_height, handle);
+ err = intel_winsys_export_handle(is->winsys, tex->bo, tex->layout.tiling,
+ tex->layout.bo_stride, tex->layout.bo_height, handle);
return !err;
}
@@ -1521,7 +419,7 @@ buf_create(struct pipe_screen *screen, const struct pipe_resource *templ)
}
if (buf->bo_size < templ->width0 ||
- buf->bo_size > max_resource_size ||
+ buf->bo_size > ilo_max_resource_size ||
!buf_create_bo(buf)) {
FREE(buf);
return NULL;
@@ -1534,23 +432,15 @@ static boolean
ilo_can_create_resource(struct pipe_screen *screen,
const struct pipe_resource *templ)
{
- struct tex_layout layout;
+ struct ilo_layout layout;
if (templ->target == PIPE_BUFFER)
- return (templ->width0 <= max_resource_size);
+ return (templ->width0 <= ilo_max_resource_size);
- tex_layout_init(&layout, screen, templ, NULL);
-
- switch (templ->target) {
- case PIPE_TEXTURE_3D:
- tex_layout_3d(&layout);
- break;
- default:
- tex_layout_2d(&layout);
- break;
- }
+ memset(&layout, 0, sizeof(layout));
+ ilo_layout_init(&layout, &ilo_screen(screen)->dev, templ);
- return tex_layout_calculate_bo_size(&layout);
+ return (layout.bo_height <= ilo_max_resource_size / layout.bo_stride);
}
static struct pipe_resource *
@@ -1642,88 +532,3 @@ ilo_texture_rename_bo(struct ilo_texture *tex)
return false;
}
}
-
-/**
- * Return the offset (in bytes) to a slice within the bo.
- *
- * The returned offset is aligned to tile size. Since slices are not
- * guaranteed to start at tile boundaries, the X and Y offsets (in pixels)
- * from the tile origin to the slice are also returned. X offset is always a
- * multiple of 4 and Y offset is always a multiple of 2.
- */
-unsigned
-ilo_texture_get_slice_offset(const struct ilo_texture *tex,
- unsigned level, unsigned slice,
- unsigned *x_offset, unsigned *y_offset)
-{
- const struct ilo_texture_slice *s =
- ilo_texture_get_slice(tex, level, slice);
- unsigned tile_w, tile_h, tile_size, row_size;
- unsigned x, y, slice_offset;
-
- /* see the Sandy Bridge PRM, volume 1 part 2, page 24 */
-
- switch (tex->tiling) {
- case INTEL_TILING_NONE:
- /* W-tiled */
- if (tex->bo_format == PIPE_FORMAT_S8_UINT) {
- tile_w = 64;
- tile_h = 64;
- }
- else {
- tile_w = 1;
- tile_h = 1;
- }
- break;
- case INTEL_TILING_X:
- tile_w = 512;
- tile_h = 8;
- break;
- case INTEL_TILING_Y:
- tile_w = 128;
- tile_h = 32;
- break;
- default:
- assert(!"unknown tiling");
- tile_w = 1;
- tile_h = 1;
- break;
- }
-
- tile_size = tile_w * tile_h;
- row_size = tex->bo_stride * tile_h;
-
- /* in bytes */
- x = s->x / tex->block_width * tex->block_size;
- y = s->y / tex->block_height;
- slice_offset = row_size * (y / tile_h) + tile_size * (x / tile_w);
-
- /*
- * Since tex->bo_stride is a multiple of tile_w, slice_offset should be
- * aligned at this point.
- */
- assert(slice_offset % tile_size == 0);
-
- /*
- * because of the possible values of align_i and align_j in
- * tex_layout_init_alignments(), x_offset is guaranteed to be a multiple of
- * 4 and y_offset is guaranteed to be a multiple of 2.
- */
- if (x_offset) {
- /* in pixels */
- x = (x % tile_w) / tex->block_size * tex->block_width;
- assert(x % 4 == 0);
-
- *x_offset = x;
- }
-
- if (y_offset) {
- /* in pixels */
- y = (y % tile_h) * tex->block_height;
- assert(y % 2 == 0);
-
- *y_offset = y;
- }
-
- return slice_offset;
-}
diff --git a/src/gallium/drivers/ilo/ilo_resource.h b/src/gallium/drivers/ilo/ilo_resource.h
index b0eab5b9c0e..be42386b8e8 100644
--- a/src/gallium/drivers/ilo/ilo_resource.h
+++ b/src/gallium/drivers/ilo/ilo_resource.h
@@ -31,6 +31,7 @@
#include "intel_winsys.h"
#include "ilo_common.h"
+#include "ilo_layout.h"
#include "ilo_screen.h"
enum ilo_texture_flags {
@@ -87,8 +88,6 @@ struct ilo_buffer {
* A 3D image slice, cube face, or array layer.
*/
struct ilo_texture_slice {
- /* 2D offset to the slice */
- unsigned x, y;
unsigned flags;
/*
@@ -109,32 +108,15 @@ struct ilo_texture {
bool imported;
- enum pipe_format bo_format;
- struct intel_bo *bo;
-
- enum intel_tiling_mode tiling;
- unsigned long bo_stride; /* distance between two block rows in bytes */
- unsigned long bo_height;
-
- unsigned block_width;
- unsigned block_height;
- unsigned block_size;
-
- /* true if the mip level alignments are stricter */
- bool halign_8, valign_4;
- /* true if space is reserved between layers */
- bool array_spacing_full;
- /* true if samples are interleaved */
- bool interleaved;
+ struct ilo_layout layout;
+ /* XXX thread-safety */
+ struct intel_bo *bo;
struct ilo_texture_slice *slices[PIPE_MAX_TEXTURE_LEVELS];
- struct ilo_texture *separate_s8;
+ struct intel_bo *aux_bo;
- struct {
- struct intel_bo *bo;
- unsigned long bo_stride;
- } hiz;
+ struct ilo_texture *separate_s8;
};
static inline struct ilo_buffer *
@@ -181,11 +163,6 @@ ilo_texture_get_slice(const struct ilo_texture *tex,
return &tex->slices[level][slice];
}
-unsigned
-ilo_texture_get_slice_offset(const struct ilo_texture *tex,
- unsigned level, unsigned slice,
- unsigned *x_offset, unsigned *y_offset);
-
static inline void
ilo_texture_set_slice_flags(struct ilo_texture *tex, unsigned level,
unsigned first_slice, unsigned num_slices,
@@ -222,25 +199,14 @@ static inline bool
ilo_texture_can_enable_hiz(const struct ilo_texture *tex, unsigned level,
unsigned first_slice, unsigned num_slices)
{
- const struct ilo_screen *is = ilo_screen(tex->base.screen);
- const struct ilo_texture_slice *slice =
- ilo_texture_get_slice(tex, level, first_slice);
-
- if (!tex->hiz.bo)
- return false;
-
- /* we can adjust 3DSTATE_DEPTH_BUFFER for the first slice */
- if (level == 0 && first_slice == 0 && num_slices == 1)
- return true;
-
- /* HiZ is non-mipmapped and non-array on GEN6 */
- assert(is->dev.gen > ILO_GEN(6));
-
/*
* Either all or none of the slices in the same level have ILO_TEXTURE_HIZ
* set. It suffices to check only the first slice.
*/
- return (slice->flags & ILO_TEXTURE_HIZ);
+ const struct ilo_texture_slice *slice =
+ ilo_texture_get_slice(tex, level, 0);
+
+ return (tex->aux_bo && (slice->flags & ILO_TEXTURE_HIZ));
}
#endif /* ILO_RESOURCE_H */
diff --git a/src/gallium/drivers/ilo/ilo_state.c b/src/gallium/drivers/ilo/ilo_state.c
index 0cc813c1fd7..229cc296fbd 100644
--- a/src/gallium/drivers/ilo/ilo_state.c
+++ b/src/gallium/drivers/ilo/ilo_state.c
@@ -924,7 +924,7 @@ ilo_create_sampler_view(struct pipe_context *pipe,
struct ilo_texture *tex = ilo_texture(res);
/* warn about degraded performance because of a missing binding flag */
- if (tex->tiling == INTEL_TILING_NONE &&
+ if (tex->layout.tiling == INTEL_TILING_NONE &&
!(tex->base.bind & PIPE_BIND_SAMPLER_VIEW)) {
ilo_warn("creating sampler view for a resource "
"not created for sampling\n");
diff --git a/src/gallium/drivers/ilo/ilo_transfer.c b/src/gallium/drivers/ilo/ilo_transfer.c
index 208e1b0327e..d40dce8c541 100644
--- a/src/gallium/drivers/ilo/ilo_transfer.c
+++ b/src/gallium/drivers/ilo/ilo_transfer.c
@@ -92,9 +92,9 @@ resource_get_transfer_method(struct pipe_resource *res, unsigned usage,
bool need_convert = true;
/* we may need to convert on the fly */
- if (tex->separate_s8 || tex->bo_format == PIPE_FORMAT_S8_UINT)
+ if (tex->separate_s8 || tex->layout.format == PIPE_FORMAT_S8_UINT)
m = ILO_TRANSFER_MAP_SW_ZS;
- else if (tex->bo_format != tex->base.format)
+ else if (tex->layout.format != tex->base.format)
m = ILO_TRANSFER_MAP_SW_CONVERT;
else
need_convert = false;
@@ -107,7 +107,7 @@ resource_get_transfer_method(struct pipe_resource *res, unsigned usage,
return true;
}
- tiled = (tex->tiling != INTEL_TILING_NONE);
+ tiled = (tex->layout.tiling != INTEL_TILING_NONE);
}
if (tiled)
@@ -194,8 +194,10 @@ xfer_alloc_staging_res(struct ilo_transfer *xfer)
xfer->staging.res = res->screen->resource_create(res->screen, &templ);
- if (xfer->staging.res && xfer->staging.res->target != PIPE_BUFFER)
- assert(ilo_texture(xfer->staging.res)->tiling == INTEL_TILING_NONE);
+ if (xfer->staging.res && xfer->staging.res->target != PIPE_BUFFER) {
+ assert(ilo_texture(xfer->staging.res)->layout.tiling ==
+ INTEL_TILING_NONE);
+ }
return (xfer->staging.res != NULL);
}
@@ -344,17 +346,13 @@ tex_get_box_origin(const struct ilo_texture *tex,
const struct pipe_box *box,
unsigned *mem_x, unsigned *mem_y)
{
- const struct ilo_texture_slice *s =
- ilo_texture_get_slice(tex, level, slice + box->z);
unsigned x, y;
- x = s->x + box->x;
- y = s->y + box->y;
-
- assert(x % tex->block_width == 0 && y % tex->block_height == 0);
+ ilo_layout_get_slice_pos(&tex->layout, level, box->z + slice, &x, &y);
+ x += box->x;
+ y += box->y;
- *mem_x = x / tex->block_width * tex->block_size;
- *mem_y = y / tex->block_height;
+ ilo_layout_pos_to_mem(&tex->layout, x, y, mem_x, mem_y);
}
static unsigned
@@ -365,36 +363,13 @@ tex_get_box_offset(const struct ilo_texture *tex, unsigned level,
tex_get_box_origin(tex, level, 0, box, &mem_x, &mem_y);
- return mem_y * tex->bo_stride + mem_x;
+ return ilo_layout_mem_to_linear(&tex->layout, mem_x, mem_y);
}
static unsigned
tex_get_slice_stride(const struct ilo_texture *tex, unsigned level)
{
- const struct ilo_texture_slice *s0, *s1;
- unsigned qpitch;
-
- /* there is no 3D array texture */
- assert(tex->base.array_size == 1 || tex->base.depth0 == 1);
-
- if (tex->base.array_size == 1) {
- /* non-array, non-3D */
- if (tex->base.depth0 == 1)
- return 0;
-
- /* only the first level has a fixed slice stride */
- if (level > 0) {
- assert(!"no slice stride for 3D texture with level > 0");
- return 0;
- }
- }
-
- s0 = ilo_texture_get_slice(tex, level, 0);
- s1 = ilo_texture_get_slice(tex, level, 1);
- qpitch = s1->y - s0->y;
- assert(qpitch % tex->block_height == 0);
-
- return (qpitch / tex->block_height) * tex->bo_stride;
+ return ilo_layout_get_slice_stride(&tex->layout, level);
}
static unsigned
@@ -542,22 +517,22 @@ static tex_tile_offset_func
tex_tile_choose_offset_func(const struct ilo_texture *tex,
unsigned *tiles_per_row)
{
- switch (tex->tiling) {
+ switch (tex->layout.tiling) {
case INTEL_TILING_X:
- *tiles_per_row = tex->bo_stride / 512;
+ *tiles_per_row = tex->layout.bo_stride / 512;
return tex_tile_x_offset;
case INTEL_TILING_Y:
- *tiles_per_row = tex->bo_stride / 128;
+ *tiles_per_row = tex->layout.bo_stride / 128;
return tex_tile_y_offset;
case INTEL_TILING_NONE:
default:
/* W-tiling */
- if (tex->bo_format == PIPE_FORMAT_S8_UINT) {
- *tiles_per_row = tex->bo_stride / 64;
+ if (tex->layout.format == PIPE_FORMAT_S8_UINT) {
+ *tiles_per_row = tex->layout.bo_stride / 64;
return tex_tile_w_offset;
}
else {
- *tiles_per_row = tex->bo_stride;
+ *tiles_per_row = tex->layout.bo_stride;
return tex_tile_none_offset;
}
}
@@ -572,7 +547,8 @@ tex_staging_sys_map_bo(struct ilo_texture *tex,
const bool prefer_cpu = (is->dev.has_llc || for_read_back);
void *ptr;
- if (prefer_cpu && (tex->tiling == INTEL_TILING_NONE || !linear_view))
+ if (prefer_cpu && (tex->layout.tiling == INTEL_TILING_NONE ||
+ !linear_view))
ptr = intel_bo_map(tex->bo, !for_read_back);
else
ptr = intel_bo_map_gtt(tex->bo);
@@ -604,7 +580,7 @@ tex_staging_sys_zs_read(struct ilo_texture *tex,
tile_offset = tex_tile_choose_offset_func(tex, &tiles_per_row);
- assert(tex->block_width == 1 && tex->block_height == 1);
+ assert(tex->layout.block_width == 1 && tex->layout.block_height == 1);
if (tex->separate_s8) {
struct ilo_texture *s8_tex = tex->separate_s8;
@@ -622,7 +598,7 @@ tex_staging_sys_zs_read(struct ilo_texture *tex,
s8_tile_offset = tex_tile_choose_offset_func(s8_tex, &s8_tiles_per_row);
if (tex->base.format == PIPE_FORMAT_Z24_UNORM_S8_UINT) {
- assert(tex->bo_format == PIPE_FORMAT_Z24X8_UNORM);
+ assert(tex->layout.format == PIPE_FORMAT_Z24X8_UNORM);
dst_cpp = 4;
dst_s8_pos = 3;
@@ -630,7 +606,7 @@ tex_staging_sys_zs_read(struct ilo_texture *tex,
}
else {
assert(tex->base.format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT);
- assert(tex->bo_format == PIPE_FORMAT_Z32_FLOAT);
+ assert(tex->layout.format == PIPE_FORMAT_Z32_FLOAT);
dst_cpp = 8;
dst_s8_pos = 4;
@@ -663,7 +639,7 @@ tex_staging_sys_zs_read(struct ilo_texture *tex,
d[dst_s8_pos] = s8_src[s8_offset];
d += dst_cpp;
- x += tex->block_size;
+ x += tex->layout.block_size;
s8_x++;
}
@@ -676,7 +652,7 @@ tex_staging_sys_zs_read(struct ilo_texture *tex,
tex_staging_sys_unmap_bo(s8_tex);
}
else {
- assert(tex->bo_format == PIPE_FORMAT_S8_UINT);
+ assert(tex->layout.format == PIPE_FORMAT_S8_UINT);
for (slice = 0; slice < box->depth; slice++) {
unsigned mem_x, mem_y;
@@ -731,7 +707,7 @@ tex_staging_sys_zs_write(struct ilo_texture *tex,
tile_offset = tex_tile_choose_offset_func(tex, &tiles_per_row);
- assert(tex->block_width == 1 && tex->block_height == 1);
+ assert(tex->layout.block_width == 1 && tex->layout.block_height == 1);
if (tex->separate_s8) {
struct ilo_texture *s8_tex = tex->separate_s8;
@@ -749,7 +725,7 @@ tex_staging_sys_zs_write(struct ilo_texture *tex,
s8_tile_offset = tex_tile_choose_offset_func(s8_tex, &s8_tiles_per_row);
if (tex->base.format == PIPE_FORMAT_Z24_UNORM_S8_UINT) {
- assert(tex->bo_format == PIPE_FORMAT_Z24X8_UNORM);
+ assert(tex->layout.format == PIPE_FORMAT_Z24X8_UNORM);
src_cpp = 4;
src_s8_pos = 3;
@@ -757,7 +733,7 @@ tex_staging_sys_zs_write(struct ilo_texture *tex,
}
else {
assert(tex->base.format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT);
- assert(tex->bo_format == PIPE_FORMAT_Z32_FLOAT);
+ assert(tex->layout.format == PIPE_FORMAT_Z32_FLOAT);
src_cpp = 8;
src_s8_pos = 4;
@@ -790,7 +766,7 @@ tex_staging_sys_zs_write(struct ilo_texture *tex,
s8_dst[s8_offset] = s[src_s8_pos];
s += src_cpp;
- x += tex->block_size;
+ x += tex->layout.block_size;
s8_x++;
}
@@ -803,7 +779,7 @@ tex_staging_sys_zs_write(struct ilo_texture *tex,
tex_staging_sys_unmap_bo(s8_tex);
}
else {
- assert(tex->bo_format == PIPE_FORMAT_S8_UINT);
+ assert(tex->layout.format == PIPE_FORMAT_S8_UINT);
for (slice = 0; slice < box->depth; slice++) {
unsigned mem_x, mem_y;
@@ -861,9 +837,9 @@ tex_staging_sys_convert_write(struct ilo_texture *tex,
else
dst_slice_stride = 0;
- if (unlikely(tex->bo_format == tex->base.format)) {
- util_copy_box(dst, tex->bo_format, tex->bo_stride, dst_slice_stride,
- 0, 0, 0, box->width, box->height, box->depth,
+ if (unlikely(tex->layout.format == tex->base.format)) {
+ util_copy_box(dst, tex->layout.format, tex->layout.bo_stride,
+ dst_slice_stride, 0, 0, 0, box->width, box->height, box->depth,
xfer->staging.sys, xfer->base.stride, xfer->base.layer_stride,
0, 0, 0);
@@ -874,14 +850,14 @@ tex_staging_sys_convert_write(struct ilo_texture *tex,
switch (tex->base.format) {
case PIPE_FORMAT_ETC1_RGB8:
- assert(tex->bo_format == PIPE_FORMAT_R8G8B8X8_UNORM);
+ assert(tex->layout.format == PIPE_FORMAT_R8G8B8X8_UNORM);
for (slice = 0; slice < box->depth; slice++) {
const void *src =
xfer->staging.sys + xfer->base.layer_stride * slice;
util_format_etc1_rgb8_unpack_rgba_8unorm(dst,
- tex->bo_stride, src, xfer->base.stride,
+ tex->layout.bo_stride, src, xfer->base.stride,
box->width, box->height);
dst += dst_slice_stride;
@@ -977,7 +953,7 @@ tex_map(struct ilo_transfer *xfer)
ptr += tex_get_box_offset(tex, xfer->base.level, &xfer->base.box);
/* stride is for a block row, not a texel row */
- xfer->base.stride = tex->bo_stride;
+ xfer->base.stride = tex->layout.bo_stride;
/* note that slice stride is not always available */
xfer->base.layer_stride = (xfer->base.box.depth > 1) ?
tex_get_slice_stride(tex, xfer->base.level) : 0;
@@ -987,7 +963,7 @@ tex_map(struct ilo_transfer *xfer)
ptr = xfer_map(xfer);
if (ptr) {
const struct ilo_texture *staging = ilo_texture(xfer->staging.res);
- xfer->base.stride = staging->bo_stride;
+ xfer->base.stride = staging->layout.bo_stride;
xfer->base.layer_stride = tex_get_slice_stride(staging, 0);
}
break;