#include "isl_gen7.h"
#include "isl_gen8.h"
#include "isl_gen9.h"
+#include "isl_gen12.h"
#include "isl_priv.h"
+void
+isl_memcpy_linear_to_tiled(uint32_t xt1, uint32_t xt2,
+ uint32_t yt1, uint32_t yt2,
+ char *dst, const char *src,
+ uint32_t dst_pitch, int32_t src_pitch,
+ bool has_swizzling,
+ enum isl_tiling tiling,
+ isl_memcpy_type copy_type)
+{
+#ifdef USE_SSE41
+ if (copy_type == ISL_MEMCPY_STREAMING_LOAD) {
+ _isl_memcpy_linear_to_tiled_sse41(
+ xt1, xt2, yt1, yt2, dst, src, dst_pitch, src_pitch, has_swizzling,
+ tiling, copy_type);
+ return;
+ }
+#endif
+
+ _isl_memcpy_linear_to_tiled(
+ xt1, xt2, yt1, yt2, dst, src, dst_pitch, src_pitch, has_swizzling,
+ tiling, copy_type);
+}
+
+void
+isl_memcpy_tiled_to_linear(uint32_t xt1, uint32_t xt2,
+ uint32_t yt1, uint32_t yt2,
+ char *dst, const char *src,
+ int32_t dst_pitch, uint32_t src_pitch,
+ bool has_swizzling,
+ enum isl_tiling tiling,
+ isl_memcpy_type copy_type)
+{
+#ifdef USE_SSE41
+ if (copy_type == ISL_MEMCPY_STREAMING_LOAD) {
+ _isl_memcpy_tiled_to_linear_sse41(
+ xt1, xt2, yt1, yt2, dst, src, dst_pitch, src_pitch, has_swizzling,
+ tiling, copy_type);
+ return;
+ }
+#endif
+
+ _isl_memcpy_tiled_to_linear(
+ xt1, xt2, yt1, yt2, dst, src, dst_pitch, src_pitch, has_swizzling,
+ tiling, copy_type);
+}
+
void PRINTFLIKE(3, 4) UNUSED
__isl_finishme(const char *file, int line, const char *fmt, ...)
{
fprintf(stderr, "%s:%d: FINISHME: %s\n", file, line, buf);
}
+static void
+isl_device_setup_mocs(struct isl_device *dev)
+{
+ if (dev->info->gen >= 12) {
+ if (dev->info->is_dg1) {
+ /* L3CC=WB */
+ dev->mocs.internal = 5 << 1;
+ /* Displayables on DG1 are free to cache in L3 since L3 is transient
+ * and flushed at bottom of each submission.
+ */
+ dev->mocs.external = 5 << 1;
+ } else {
+ /* TODO: Set PTE to MOCS 61 when the kernel is ready */
+ /* TC=1/LLC Only, LeCC=1/Uncacheable, LRUM=0, L3CC=1/Uncacheable */
+ dev->mocs.external = 3 << 1;
+ /* TC=LLC/eLLC, LeCC=WB, LRUM=3, L3CC=WB */
+ dev->mocs.internal = 2 << 1;
+ }
+ } else if (dev->info->gen >= 9) {
+ /* TC=LLC/eLLC, LeCC=PTE, LRUM=3, L3CC=WB */
+ dev->mocs.external = 1 << 1;
+ /* TC=LLC/eLLC, LeCC=WB, LRUM=3, L3CC=WB */
+ dev->mocs.internal = 2 << 1;
+ } else if (dev->info->gen >= 8) {
+ /* MEMORY_OBJECT_CONTROL_STATE:
+ * .MemoryTypeLLCeLLCCacheabilityControl = UCwithFenceifcoherentcycle,
+ * .TargetCache = L3DefertoPATforLLCeLLCselection,
+ * .AgeforQUADLRU = 0
+ */
+ dev->mocs.external = 0x18;
+ /* MEMORY_OBJECT_CONTROL_STATE:
+ * .MemoryTypeLLCeLLCCacheabilityControl = WB,
+ * .TargetCache = L3DefertoPATforLLCeLLCselection,
+ * .AgeforQUADLRU = 0
+ */
+ dev->mocs.internal = 0x78;
+ } else if (dev->info->gen >= 7) {
+ if (dev->info->is_haswell) {
+ /* MEMORY_OBJECT_CONTROL_STATE:
+ * .LLCeLLCCacheabilityControlLLCCC = 0,
+ * .L3CacheabilityControlL3CC = 1,
+ */
+ dev->mocs.internal = 1;
+ dev->mocs.external = 1;
+ } else {
+ /* MEMORY_OBJECT_CONTROL_STATE:
+ * .GraphicsDataTypeGFDT = 0,
+ * .LLCCacheabilityControlLLCCC = 0,
+ * .L3CacheabilityControlL3CC = 1,
+ */
+ dev->mocs.internal = 1;
+ dev->mocs.external = 1;
+ }
+ } else {
+ dev->mocs.internal = 0;
+ dev->mocs.external = 0;
+ }
+}
+
void
isl_device_init(struct isl_device *dev,
const struct gen_device_info *info,
bool has_bit6_swizzling)
{
+ /* Gen8+ don't have bit6 swizzling, ensure callsite is not confused. */
+ assert(!(has_bit6_swizzling && info->gen >= 8));
+
dev->info = info;
dev->use_separate_stencil = ISL_DEV_GEN(dev) >= 6;
dev->has_bit6_swizzling = has_bit6_swizzling;
dev->ss.size = RENDER_SURFACE_STATE_length(info) * 4;
dev->ss.align = isl_align(dev->ss.size, 32);
+ dev->ss.clear_color_state_size =
+ isl_align(CLEAR_COLOR_length(info) * 4, 64);
+ dev->ss.clear_color_state_offset =
+ RENDER_SURFACE_STATE_ClearValueAddress_start(info) / 32 * 4;
+
+ dev->ss.clear_value_size =
+ isl_align(RENDER_SURFACE_STATE_RedClearColor_bits(info) +
+ RENDER_SURFACE_STATE_GreenClearColor_bits(info) +
+ RENDER_SURFACE_STATE_BlueClearColor_bits(info) +
+ RENDER_SURFACE_STATE_AlphaClearColor_bits(info), 32) / 8;
+
+ dev->ss.clear_value_offset =
+ RENDER_SURFACE_STATE_RedClearColor_start(info) / 32 * 4;
+
assert(RENDER_SURFACE_STATE_SurfaceBaseAddress_start(info) % 8 == 0);
dev->ss.addr_offset =
RENDER_SURFACE_STATE_SurfaceBaseAddress_start(info) / 8;
dev->ds.stencil_offset = 0;
dev->ds.hiz_offset = 0;
}
+
+ if (ISL_DEV_GEN(dev) >= 12) {
+ dev->ds.size += GEN12_MI_LOAD_REGISTER_IMM_length * 4 * 2;
+ }
+
+ isl_device_setup_mocs(dev);
}
/**
phys_B = isl_extent2d(128, 32);
break;
+ case ISL_TILING_GEN12_CCS:
+ /* From the Bspec, Gen Graphics > Gen12 > Memory Data Formats > Memory
+ * Compression > Memory Compression - Gen12:
+ *
+ * 4 bits of auxiliary plane data are required for 2 cachelines of
+ * main surface data. This results in a single cacheline of auxiliary
+ * plane data mapping to 4 4K pages of main surface data for the 4K
+ * pages (tile Y ) and 1 64K Tile Ys page.
+ *
+ * The Y-tiled pairing bit of 9 shown in the table below that Bspec
+ * section expresses that the 2 cachelines of main surface data are
+ * horizontally adjacent.
+ *
+ * TODO: Handle Ys, Yf and their pairing bits.
+ *
+ * Therefore, each CCS cacheline represents a 512Bx32 row area and each
+ * element represents a 32Bx4 row area.
+ */
+ assert(format_bpb == 4);
+ logical_el = isl_extent2d(16, 8);
+ phys_B = isl_extent2d(64, 1);
+ break;
+
default:
unreachable("not reached");
} /* end switch */
};
}
+bool
+isl_color_value_is_zero(union isl_color_value value,
+ enum isl_format format)
+{
+ const struct isl_format_layout *fmtl = isl_format_get_layout(format);
+
+#define RETURN_FALSE_IF_NOT_0(c, i) \
+ if (fmtl->channels.c.bits && value.u32[i] != 0) \
+ return false
+
+ RETURN_FALSE_IF_NOT_0(r, 0);
+ RETURN_FALSE_IF_NOT_0(g, 1);
+ RETURN_FALSE_IF_NOT_0(b, 2);
+ RETURN_FALSE_IF_NOT_0(a, 3);
+
+#undef RETURN_FALSE_IF_NOT_0
+
+ return true;
+}
+
+bool
+isl_color_value_is_zero_one(union isl_color_value value,
+ enum isl_format format)
+{
+ const struct isl_format_layout *fmtl = isl_format_get_layout(format);
+
+#define RETURN_FALSE_IF_NOT_0_1(c, i, field) \
+ if (fmtl->channels.c.bits && value.field[i] != 0 && value.field[i] != 1) \
+ return false
+
+ if (isl_format_has_int_channel(format)) {
+ RETURN_FALSE_IF_NOT_0_1(r, 0, u32);
+ RETURN_FALSE_IF_NOT_0_1(g, 1, u32);
+ RETURN_FALSE_IF_NOT_0_1(b, 2, u32);
+ RETURN_FALSE_IF_NOT_0_1(a, 3, u32);
+ } else {
+ RETURN_FALSE_IF_NOT_0_1(r, 0, f32);
+ RETURN_FALSE_IF_NOT_0_1(g, 1, f32);
+ RETURN_FALSE_IF_NOT_0_1(b, 2, f32);
+ RETURN_FALSE_IF_NOT_0_1(a, 3, f32);
+ }
+
+#undef RETURN_FALSE_IF_NOT_0_1
+
+ return true;
+}
+
/**
* @param[out] tiling is set only on success
*/
if (info->usage & ISL_SURF_USAGE_HIZ_BIT) {
assert(info->format == ISL_FORMAT_HIZ);
assert(tiling_flags == ISL_TILING_HIZ_BIT);
- *tiling = ISL_TILING_HIZ;
+ *tiling = isl_tiling_flag_to_enum(tiling_flags);
return true;
}
/* CCS surfaces always use the CCS tiling */
if (info->usage & ISL_SURF_USAGE_CCS_BIT) {
assert(isl_format_get_layout(info->format)->txc == ISL_TXC_CCS);
- assert(tiling_flags == ISL_TILING_CCS_BIT);
- *tiling = ISL_TILING_CCS;
+ UNUSED bool ivb_ccs = ISL_DEV_GEN(dev) < 12 &&
+ tiling_flags == ISL_TILING_CCS_BIT;
+ UNUSED bool tgl_ccs = ISL_DEV_GEN(dev) >= 12 &&
+ tiling_flags == ISL_TILING_GEN12_CCS_BIT;
+ assert(ivb_ccs != tgl_ccs);
+ *tiling = isl_tiling_flag_to_enum(tiling_flags);
return true;
}
* the storage for LODs other than LOD 0 is not needed.
*/
assert(info->levels == 1);
- assert(phys_level0_sa->array_len == 1);
return ISL_ARRAY_PITCH_SPAN_COMPACT;
} else {
if ((ISL_DEV_GEN(dev) == 5 || ISL_DEV_GEN(dev) == 6) &&
* compact QPitch possible in order to conserve memory.
*/
return ISL_ARRAY_PITCH_SPAN_COMPACT;
+
+ case ISL_DIM_LAYOUT_GEN6_STENCIL_HIZ:
+ /* Each array image in the gen6 stencil of HiZ surface is compact in the
+ * sense that every LOD is a compact array of the same size as LOD0.
+ */
+ return ISL_ARRAY_PITCH_SPAN_COMPACT;
}
unreachable("bad isl_dim_layout");
return;
} else if (info->format == ISL_FORMAT_HIZ) {
assert(ISL_DEV_GEN(dev) >= 6);
- /* HiZ surfaces are always aligned to 16x8 pixels in the primary surface
- * which works out to 2x2 HiZ elments.
- */
- *image_align_el = isl_extent3d(2, 2, 1);
+ if (ISL_DEV_GEN(dev) == 6) {
+ /* HiZ surfaces on Sandy Bridge are packed tightly. */
+ *image_align_el = isl_extent3d(1, 1, 1);
+ } else if (ISL_DEV_GEN(dev) < 12) {
+ /* On gen7+, HiZ surfaces are always aligned to 16x8 pixels in the
+ * primary surface which works out to 2x2 HiZ elments.
+ */
+ *image_align_el = isl_extent3d(2, 2, 1);
+ } else {
+ /* On gen12+, HiZ surfaces are always aligned to 16x16 pixels in the
+ * primary surface which works out to 2x4 HiZ elments.
+ * TODO: Verify
+ */
+ *image_align_el = isl_extent3d(2, 4, 1);
+ }
return;
}
- if (ISL_DEV_GEN(dev) >= 9) {
+ if (ISL_DEV_GEN(dev) >= 12) {
+ isl_gen12_choose_image_alignment_el(dev, info, tiling, dim_layout,
+ msaa_layout, image_align_el);
+ } else if (ISL_DEV_GEN(dev) >= 9) {
isl_gen9_choose_image_alignment_el(dev, info, tiling, dim_layout,
msaa_layout, image_align_el);
} else if (ISL_DEV_GEN(dev) >= 8) {
static enum isl_dim_layout
isl_surf_choose_dim_layout(const struct isl_device *dev,
enum isl_surf_dim logical_dim,
- enum isl_tiling tiling)
+ enum isl_tiling tiling,
+ isl_surf_usage_flags_t usage)
{
+ /* Sandy bridge needs a special layout for HiZ and stencil. */
+ if (ISL_DEV_GEN(dev) == 6 &&
+ (tiling == ISL_TILING_W || tiling == ISL_TILING_HIZ))
+ return ISL_DIM_LAYOUT_GEN6_STENCIL_HIZ;
+
if (ISL_DEV_GEN(dev) >= 9) {
switch (logical_dim) {
case ISL_SURF_DIM_1D:
switch (logical_dim) {
case ISL_SURF_DIM_1D:
case ISL_SURF_DIM_2D:
+ /* From the G45 PRM Vol. 1a, "6.17.4.1 Hardware Cube Map Layout":
+ *
+ * The cube face textures are stored in the same way as 3D surfaces
+ * are stored (see section 6.17.5 for details). For cube surfaces,
+ * however, the depth is equal to the number of faces (always 6) and
+ * is not reduced for each MIP.
+ */
+ if (ISL_DEV_GEN(dev) == 4 && (usage & ISL_SURF_USAGE_CUBE_BIT))
+ return ISL_DIM_LAYOUT_GEN4_3D;
+
return ISL_DIM_LAYOUT_GEN4_2D;
case ISL_SURF_DIM_3D:
return ISL_DIM_LAYOUT_GEN4_3D;
/**
* Calculate the physical extent of the surface's first level, in units of
- * surface samples. The result is aligned to the format's compression block.
+ * surface samples.
*/
static void
isl_calc_phys_level0_extent_sa(const struct isl_device *dev,
{
const struct isl_format_layout *fmtl = isl_format_get_layout(info->format);
- if (isl_format_is_yuv(info->format))
- isl_finishme("%s:%s: YUV format", __FILE__, __func__);
+ if (isl_format_is_planar(info->format))
+ unreachable("Planar formats unsupported");
switch (info->dim) {
case ISL_SURF_DIM_1D:
case ISL_DIM_LAYOUT_GEN9_1D:
case ISL_DIM_LAYOUT_GEN4_2D:
+ case ISL_DIM_LAYOUT_GEN6_STENCIL_HIZ:
*phys_level0_sa = (struct isl_extent4d) {
- .w = isl_align_npot(info->width, fmtl->bw),
- .h = fmtl->bh,
+ .w = info->width,
+ .h = 1,
.d = 1,
.a = info->array_len,
};
break;
case ISL_SURF_DIM_2D:
- assert(dim_layout == ISL_DIM_LAYOUT_GEN4_2D);
+ if (ISL_DEV_GEN(dev) == 4 && (info->usage & ISL_SURF_USAGE_CUBE_BIT))
+ assert(dim_layout == ISL_DIM_LAYOUT_GEN4_3D);
+ else
+ assert(dim_layout == ISL_DIM_LAYOUT_GEN4_2D ||
+ dim_layout == ISL_DIM_LAYOUT_GEN6_STENCIL_HIZ);
if (tiling == ISL_TILING_Ys && info->samples > 1)
isl_finishme("%s:%s: multisample TileYs layout", __FILE__, __func__);
assert(info->samples == 1);
*phys_level0_sa = (struct isl_extent4d) {
- .w = isl_align_npot(info->width, fmtl->bw),
- .h = isl_align_npot(info->height, fmtl->bh),
+ .w = info->width,
+ .h = info->height,
.d = 1,
.a = info->array_len,
};
isl_msaa_interleaved_scale_px_to_sa(info->samples,
&phys_level0_sa->w,
&phys_level0_sa->h);
-
- phys_level0_sa->w = isl_align(phys_level0_sa->w, fmtl->bw);
- phys_level0_sa->h = isl_align(phys_level0_sa->h, fmtl->bh);
break;
}
break;
switch (dim_layout) {
case ISL_DIM_LAYOUT_GEN9_1D:
+ case ISL_DIM_LAYOUT_GEN6_STENCIL_HIZ:
unreachable("bad isl_dim_layout");
case ISL_DIM_LAYOUT_GEN4_2D:
assert(ISL_DEV_GEN(dev) >= 9);
*phys_level0_sa = (struct isl_extent4d) {
- .w = isl_align_npot(info->width, fmtl->bw),
- .h = isl_align_npot(info->height, fmtl->bh),
+ .w = info->width,
+ .h = info->height,
.d = 1,
.a = info->depth,
};
case ISL_DIM_LAYOUT_GEN4_3D:
assert(ISL_DEV_GEN(dev) < 9);
*phys_level0_sa = (struct isl_extent4d) {
- .w = isl_align(info->width, fmtl->bw),
- .h = isl_align(info->height, fmtl->bh),
+ .w = info->width,
+ .h = info->height,
.d = info->depth,
.a = 1,
};
* surface elements.
*/
static uint32_t
-isl_calc_array_pitch_el_rows(const struct isl_device *dev,
- const struct isl_surf_init_info *restrict info,
- const struct isl_tile_info *tile_info,
- enum isl_dim_layout dim_layout,
- enum isl_array_pitch_span array_pitch_span,
- const struct isl_extent3d *image_align_sa,
- const struct isl_extent4d *phys_level0_sa,
- const struct isl_extent2d *phys_slice0_sa)
+isl_calc_array_pitch_el_rows_gen4_2d(
+ const struct isl_device *dev,
+ const struct isl_surf_init_info *restrict info,
+ const struct isl_tile_info *tile_info,
+ const struct isl_extent3d *image_align_sa,
+ const struct isl_extent4d *phys_level0_sa,
+ enum isl_array_pitch_span array_pitch_span,
+ const struct isl_extent2d *phys_slice0_sa)
{
const struct isl_format_layout *fmtl = isl_format_get_layout(info->format);
uint32_t pitch_sa_rows = 0;
- switch (dim_layout) {
- case ISL_DIM_LAYOUT_GEN9_1D:
- /* Each row is an array slice */
- pitch_sa_rows = 1;
+ switch (array_pitch_span) {
+ case ISL_ARRAY_PITCH_SPAN_COMPACT:
+ pitch_sa_rows = isl_align_npot(phys_slice0_sa->h, image_align_sa->h);
break;
- case ISL_DIM_LAYOUT_GEN4_2D:
- switch (array_pitch_span) {
- case ISL_ARRAY_PITCH_SPAN_COMPACT:
- pitch_sa_rows = isl_align_npot(phys_slice0_sa->h, image_align_sa->h);
- break;
- case ISL_ARRAY_PITCH_SPAN_FULL: {
- /* The QPitch equation is found in the Broadwell PRM >> Volume 5:
- * Memory Views >> Common Surface Formats >> Surface Layout >> 2D
- * Surfaces >> Surface Arrays.
- */
- uint32_t H0_sa = phys_level0_sa->h;
- uint32_t H1_sa = isl_minify(H0_sa, 1);
-
- uint32_t h0_sa = isl_align_npot(H0_sa, image_align_sa->h);
- uint32_t h1_sa = isl_align_npot(H1_sa, image_align_sa->h);
+ case ISL_ARRAY_PITCH_SPAN_FULL: {
+ /* The QPitch equation is found in the Broadwell PRM >> Volume 5:
+ * Memory Views >> Common Surface Formats >> Surface Layout >> 2D
+ * Surfaces >> Surface Arrays.
+ */
+ uint32_t H0_sa = phys_level0_sa->h;
+ uint32_t H1_sa = isl_minify(H0_sa, 1);
- uint32_t m;
- if (ISL_DEV_GEN(dev) >= 7) {
- /* The QPitch equation changed slightly in Ivybridge. */
- m = 12;
- } else {
- m = 11;
- }
+ uint32_t h0_sa = isl_align_npot(H0_sa, image_align_sa->h);
+ uint32_t h1_sa = isl_align_npot(H1_sa, image_align_sa->h);
- pitch_sa_rows = h0_sa + h1_sa + (m * image_align_sa->h);
+ uint32_t m;
+ if (ISL_DEV_GEN(dev) >= 7) {
+ /* The QPitch equation changed slightly in Ivybridge. */
+ m = 12;
+ } else {
+ m = 11;
+ }
- if (ISL_DEV_GEN(dev) == 6 && info->samples > 1 &&
- (info->height % 4 == 1)) {
- /* [SNB] Errata from the Sandy Bridge PRM >> Volume 4 Part 1:
- * Graphics Core >> Section 7.18.3.7: Surface Arrays:
- *
- * [SNB] 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.
- *
- * XXX(chadv): Is the errata natural corollary of the physical
- * layout of interleaved samples?
- */
- pitch_sa_rows += 4;
- }
+ pitch_sa_rows = h0_sa + h1_sa + (m * image_align_sa->h);
- pitch_sa_rows = isl_align_npot(pitch_sa_rows, fmtl->bh);
- } /* end case */
- break;
+ if (ISL_DEV_GEN(dev) == 6 && info->samples > 1 &&
+ (info->height % 4 == 1)) {
+ /* [SNB] Errata from the Sandy Bridge PRM >> Volume 4 Part 1:
+ * Graphics Core >> Section 7.18.3.7: Surface Arrays:
+ *
+ * [SNB] 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.
+ *
+ * XXX(chadv): Is the errata natural corollary of the physical
+ * layout of interleaved samples?
+ */
+ pitch_sa_rows += 4;
}
- break;
- case ISL_DIM_LAYOUT_GEN4_3D:
- assert(array_pitch_span == ISL_ARRAY_PITCH_SPAN_COMPACT);
- pitch_sa_rows = isl_align_npot(phys_slice0_sa->h, image_align_sa->h);
- break;
- default:
- unreachable("bad isl_dim_layout");
+
+ pitch_sa_rows = isl_align_npot(pitch_sa_rows, fmtl->bh);
+ } /* end case */
break;
}
assert(pitch_sa_rows % fmtl->bh == 0);
uint32_t pitch_el_rows = pitch_sa_rows / fmtl->bh;
- if (ISL_DEV_GEN(dev) >= 9 && fmtl->txc == ISL_TXC_CCS) {
+ if (ISL_DEV_GEN(dev) >= 9 && ISL_DEV_GEN(dev) <= 11 &&
+ fmtl->txc == ISL_TXC_CCS) {
/*
* From the Sky Lake PRM Vol 7, "MCS Buffer for Render Target(s)" (p. 632):
*
* The first restriction is already handled by isl_choose_image_alignment_el
* but the second restriction, which is an extension of the first, only
* applies to qpitch and must be applied here.
+ *
+ * The second restriction disappears on Gen12.
*/
assert(fmtl->bh == 4);
pitch_el_rows = isl_align(pitch_el_rows, 256 / 4);
const struct isl_extent4d *phys_level0_sa,
struct isl_extent2d *phys_slice0_sa)
{
- const struct isl_format_layout *fmtl = isl_format_get_layout(info->format);
-
assert(phys_level0_sa->depth == 1);
if (info->levels == 1) {
- /* Do not pad the surface to the image alignment. Instead, pad it only
- * to the pixel format's block alignment.
+ /* Do not pad the surface to the image alignment.
*
* For tiled surfaces, using a reduced alignment here avoids wasting CPU
* cycles on the below mipmap layout caluclations. Reducing the
* VkBufferImageCopy::bufferRowLength.
*/
*phys_slice0_sa = (struct isl_extent2d) {
- .w = isl_align_npot(phys_level0_sa->w, fmtl->bw),
- .h = isl_align_npot(phys_level0_sa->h, fmtl->bh),
+ .w = phys_level0_sa->w,
+ .h = phys_level0_sa->h,
};
return;
}
};
}
+static void
+isl_calc_phys_total_extent_el_gen4_2d(
+ const struct isl_device *dev,
+ const struct isl_surf_init_info *restrict info,
+ const struct isl_tile_info *tile_info,
+ enum isl_msaa_layout msaa_layout,
+ const struct isl_extent3d *image_align_sa,
+ const struct isl_extent4d *phys_level0_sa,
+ enum isl_array_pitch_span array_pitch_span,
+ uint32_t *array_pitch_el_rows,
+ struct isl_extent2d *total_extent_el)
+{
+ const struct isl_format_layout *fmtl = isl_format_get_layout(info->format);
+
+ struct isl_extent2d phys_slice0_sa;
+ isl_calc_phys_slice0_extent_sa_gen4_2d(dev, info, msaa_layout,
+ image_align_sa, phys_level0_sa,
+ &phys_slice0_sa);
+ *array_pitch_el_rows =
+ isl_calc_array_pitch_el_rows_gen4_2d(dev, info, tile_info,
+ image_align_sa, phys_level0_sa,
+ array_pitch_span,
+ &phys_slice0_sa);
+ *total_extent_el = (struct isl_extent2d) {
+ .w = isl_align_div_npot(phys_slice0_sa.w, fmtl->bw),
+ .h = *array_pitch_el_rows * (phys_level0_sa->array_len - 1) +
+ isl_align_div_npot(phys_slice0_sa.h, fmtl->bh),
+ };
+}
+
/**
* A variant of isl_calc_phys_slice0_extent_sa() specific to
* ISL_DIM_LAYOUT_GEN4_3D.
*/
static void
-isl_calc_phys_slice0_extent_sa_gen4_3d(
+isl_calc_phys_total_extent_el_gen4_3d(
const struct isl_device *dev,
const struct isl_surf_init_info *restrict info,
const struct isl_extent3d *image_align_sa,
const struct isl_extent4d *phys_level0_sa,
- struct isl_extent2d *phys_slice0_sa)
+ uint32_t *array_pitch_el_rows,
+ struct isl_extent2d *phys_total_el)
{
+ const struct isl_format_layout *fmtl = isl_format_get_layout(info->format);
+
assert(info->samples == 1);
- assert(phys_level0_sa->array_len == 1);
- uint32_t slice_w = 0;
- uint32_t slice_h = 0;
+ if (info->dim != ISL_SURF_DIM_3D) {
+ /* From the G45 PRM Vol. 1a, "6.17.4.1 Hardware Cube Map Layout":
+ *
+ * The cube face textures are stored in the same way as 3D surfaces
+ * are stored (see section 6.17.5 for details). For cube surfaces,
+ * however, the depth is equal to the number of faces (always 6) and
+ * is not reduced for each MIP.
+ */
+ assert(ISL_DEV_GEN(dev) == 4);
+ assert(info->usage & ISL_SURF_USAGE_CUBE_BIT);
+ assert(phys_level0_sa->array_len == 6);
+ } else {
+ assert(phys_level0_sa->array_len == 1);
+ }
+
+ uint32_t total_w = 0;
+ uint32_t total_h = 0;
uint32_t W0 = phys_level0_sa->w;
uint32_t H0 = phys_level0_sa->h;
uint32_t D0 = phys_level0_sa->d;
+ uint32_t A0 = phys_level0_sa->a;
for (uint32_t l = 0; l < info->levels; ++l) {
uint32_t level_w = isl_align_npot(isl_minify(W0, l), image_align_sa->w);
uint32_t level_h = isl_align_npot(isl_minify(H0, l), image_align_sa->h);
- uint32_t level_d = isl_align_npot(isl_minify(D0, l), image_align_sa->d);
+ uint32_t level_d = info->dim == ISL_SURF_DIM_3D ? isl_minify(D0, l) : A0;
uint32_t max_layers_horiz = MIN(level_d, 1u << l);
uint32_t max_layers_vert = isl_align(level_d, 1u << l) / (1u << l);
- slice_w = MAX(slice_w, level_w * max_layers_horiz);
- slice_h += level_h * max_layers_vert;
+ total_w = MAX(total_w, level_w * max_layers_horiz);
+ total_h += level_h * max_layers_vert;
}
- *phys_slice0_sa = (struct isl_extent2d) {
- .w = slice_w,
- .h = slice_h,
+ /* GEN4_3D layouts don't really have an array pitch since each LOD has a
+ * different number of horizontal and vertical layers. We have to set it
+ * to something, so at least make it true for LOD0.
+ */
+ *array_pitch_el_rows =
+ isl_align_npot(phys_level0_sa->h, image_align_sa->h) / fmtl->bw;
+ *phys_total_el = (struct isl_extent2d) {
+ .w = isl_assert_div(total_w, fmtl->bw),
+ .h = isl_assert_div(total_h, fmtl->bh),
+ };
+}
+
+/**
+ * A variant of isl_calc_phys_slice0_extent_sa() specific to
+ * ISL_DIM_LAYOUT_GEN6_STENCIL_HIZ.
+ */
+static void
+isl_calc_phys_total_extent_el_gen6_stencil_hiz(
+ const struct isl_device *dev,
+ const struct isl_surf_init_info *restrict info,
+ const struct isl_tile_info *tile_info,
+ const struct isl_extent3d *image_align_sa,
+ const struct isl_extent4d *phys_level0_sa,
+ uint32_t *array_pitch_el_rows,
+ struct isl_extent2d *phys_total_el)
+{
+ const struct isl_format_layout *fmtl = isl_format_get_layout(info->format);
+
+ const struct isl_extent2d tile_extent_sa = {
+ .w = tile_info->logical_extent_el.w * fmtl->bw,
+ .h = tile_info->logical_extent_el.h * fmtl->bh,
+ };
+ /* Tile size is a multiple of image alignment */
+ assert(tile_extent_sa.w % image_align_sa->w == 0);
+ assert(tile_extent_sa.h % image_align_sa->h == 0);
+
+ const uint32_t W0 = phys_level0_sa->w;
+ const uint32_t H0 = phys_level0_sa->h;
+
+ /* Each image has the same height as LOD0 because the hardware thinks
+ * everything is LOD0
+ */
+ const uint32_t H = isl_align(H0, image_align_sa->h) * phys_level0_sa->a;
+
+ uint32_t total_top_w = 0;
+ uint32_t total_bottom_w = 0;
+ uint32_t total_h = 0;
+
+ for (uint32_t l = 0; l < info->levels; ++l) {
+ const uint32_t W = isl_minify(W0, l);
+
+ const uint32_t w = isl_align(W, tile_extent_sa.w);
+ const uint32_t h = isl_align(H, tile_extent_sa.h);
+
+ if (l == 0) {
+ total_top_w = w;
+ total_h = h;
+ } else if (l == 1) {
+ total_bottom_w = w;
+ total_h += h;
+ } else {
+ total_bottom_w += w;
+ }
+ }
+
+ *array_pitch_el_rows =
+ isl_assert_div(isl_align(H0, image_align_sa->h), fmtl->bh);
+ *phys_total_el = (struct isl_extent2d) {
+ .w = isl_assert_div(MAX(total_top_w, total_bottom_w), fmtl->bw),
+ .h = isl_assert_div(total_h, fmtl->bh),
};
}
* ISL_DIM_LAYOUT_GEN9_1D.
*/
static void
-isl_calc_phys_slice0_extent_sa_gen9_1d(
+isl_calc_phys_total_extent_el_gen9_1d(
const struct isl_device *dev,
const struct isl_surf_init_info *restrict info,
const struct isl_extent3d *image_align_sa,
const struct isl_extent4d *phys_level0_sa,
- struct isl_extent2d *phys_slice0_sa)
+ uint32_t *array_pitch_el_rows,
+ struct isl_extent2d *phys_total_el)
{
- MAYBE_UNUSED const struct isl_format_layout *fmtl = isl_format_get_layout(info->format);
+ const struct isl_format_layout *fmtl = isl_format_get_layout(info->format);
assert(phys_level0_sa->height == 1);
assert(phys_level0_sa->depth == 1);
slice_w += w;
}
- *phys_slice0_sa = isl_extent2d(slice_w, 1);
+ *array_pitch_el_rows = 1;
+ *phys_total_el = (struct isl_extent2d) {
+ .w = isl_assert_div(slice_w, fmtl->bw),
+ .h = phys_level0_sa->array_len,
+ };
}
/**
- * Calculate the physical extent of the surface's first array slice, in units
- * of surface samples. If the surface is multi-leveled, then the result will
- * be aligned to \a image_align_sa.
+ * Calculate the two-dimensional total physical extent of the surface, in
+ * units of surface elements.
*/
static void
-isl_calc_phys_slice0_extent_sa(const struct isl_device *dev,
- const struct isl_surf_init_info *restrict info,
- enum isl_dim_layout dim_layout,
- enum isl_msaa_layout msaa_layout,
- const struct isl_extent3d *image_align_sa,
- const struct isl_extent4d *phys_level0_sa,
- struct isl_extent2d *phys_slice0_sa)
+isl_calc_phys_total_extent_el(const struct isl_device *dev,
+ const struct isl_surf_init_info *restrict info,
+ const struct isl_tile_info *tile_info,
+ enum isl_dim_layout dim_layout,
+ enum isl_msaa_layout msaa_layout,
+ const struct isl_extent3d *image_align_sa,
+ const struct isl_extent4d *phys_level0_sa,
+ enum isl_array_pitch_span array_pitch_span,
+ uint32_t *array_pitch_el_rows,
+ struct isl_extent2d *total_extent_el)
{
switch (dim_layout) {
case ISL_DIM_LAYOUT_GEN9_1D:
- isl_calc_phys_slice0_extent_sa_gen9_1d(dev, info,
- image_align_sa, phys_level0_sa,
- phys_slice0_sa);
+ assert(array_pitch_span == ISL_ARRAY_PITCH_SPAN_COMPACT);
+ isl_calc_phys_total_extent_el_gen9_1d(dev, info,
+ image_align_sa, phys_level0_sa,
+ array_pitch_el_rows,
+ total_extent_el);
return;
case ISL_DIM_LAYOUT_GEN4_2D:
- isl_calc_phys_slice0_extent_sa_gen4_2d(dev, info, msaa_layout,
- image_align_sa, phys_level0_sa,
- phys_slice0_sa);
+ isl_calc_phys_total_extent_el_gen4_2d(dev, info, tile_info, msaa_layout,
+ image_align_sa, phys_level0_sa,
+ array_pitch_span,
+ array_pitch_el_rows,
+ total_extent_el);
+ return;
+ case ISL_DIM_LAYOUT_GEN6_STENCIL_HIZ:
+ assert(array_pitch_span == ISL_ARRAY_PITCH_SPAN_COMPACT);
+ isl_calc_phys_total_extent_el_gen6_stencil_hiz(dev, info, tile_info,
+ image_align_sa,
+ phys_level0_sa,
+ array_pitch_el_rows,
+ total_extent_el);
return;
case ISL_DIM_LAYOUT_GEN4_3D:
- isl_calc_phys_slice0_extent_sa_gen4_3d(dev, info, image_align_sa,
- phys_level0_sa, phys_slice0_sa);
+ assert(array_pitch_span == ISL_ARRAY_PITCH_SPAN_COMPACT);
+ isl_calc_phys_total_extent_el_gen4_3d(dev, info,
+ image_align_sa, phys_level0_sa,
+ array_pitch_el_rows,
+ total_extent_el);
return;
}
+
+ unreachable("invalid value for dim_layout");
}
static uint32_t
-isl_calc_row_pitch_alignment(const struct isl_surf_init_info *surf_info,
+isl_calc_row_pitch_alignment(const struct isl_device *dev,
+ const struct isl_surf_init_info *surf_info,
const struct isl_tile_info *tile_info)
{
- if (tile_info->tiling != ISL_TILING_LINEAR)
+ if (tile_info->tiling != ISL_TILING_LINEAR) {
+ /* According to BSpec: 44930, Gen12's CCS-compressed surface pitches must
+ * be 512B-aligned. CCS is only support on Y tilings.
+ *
+ * Only consider 512B alignment when :
+ * - AUX is not explicitly disabled
+ * - the caller has specified no pitch
+ *
+ * isl_surf_get_ccs_surf() will check that the main surface alignment
+ * matches CCS expectations.
+ */
+ if (ISL_DEV_GEN(dev) >= 12 &&
+ isl_format_supports_ccs_e(dev->info, surf_info->format) &&
+ tile_info->tiling != ISL_TILING_X &&
+ !(surf_info->usage & ISL_SURF_USAGE_DISABLE_AUX_BIT) &&
+ surf_info->row_pitch_B == 0) {
+ return isl_align(tile_info->phys_extent_B.width, 512);
+ }
+
return tile_info->phys_extent_B.width;
+ }
/* From the Broadwel PRM >> Volume 2d: Command Reference: Structures >>
* RENDER_SURFACE_STATE Surface Pitch (p349):
*/
const struct isl_format_layout *fmtl = isl_format_get_layout(surf_info->format);
const uint32_t bs = fmtl->bpb / 8;
+ uint32_t alignment;
if (surf_info->usage & ISL_SURF_USAGE_RENDER_TARGET_BIT) {
if (isl_format_is_yuv(surf_info->format)) {
- return 2 * bs;
+ alignment = 2 * bs;
} else {
- return bs;
+ alignment = bs;
}
+ } else {
+ alignment = 1;
}
- return 1;
+ /* From the Broadwell PRM >> Volume 2c: Command Reference: Registers >>
+ * PRI_STRIDE Stride (p1254):
+ *
+ * "When using linear memory, this must be at least 64 byte aligned."
+ */
+ if (surf_info->usage & ISL_SURF_USAGE_DISPLAY_BIT)
+ alignment = isl_align(alignment, 64);
+
+ return alignment;
}
static uint32_t
isl_calc_linear_min_row_pitch(const struct isl_device *dev,
const struct isl_surf_init_info *info,
- const struct isl_extent2d *phys_slice0_sa,
- uint32_t alignment)
+ const struct isl_extent2d *phys_total_el,
+ uint32_t alignment_B)
{
const struct isl_format_layout *fmtl = isl_format_get_layout(info->format);
const uint32_t bs = fmtl->bpb / 8;
- assert(phys_slice0_sa->w % fmtl->bw == 0);
-
- return isl_align_npot(bs * (phys_slice0_sa->w / fmtl->bw), alignment);
+ return isl_align_npot(bs * phys_total_el->w, alignment_B);
}
static uint32_t
isl_calc_tiled_min_row_pitch(const struct isl_device *dev,
const struct isl_surf_init_info *surf_info,
const struct isl_tile_info *tile_info,
- const struct isl_extent2d *phys_slice0_sa,
- uint32_t alignment)
+ const struct isl_extent2d *phys_total_el,
+ uint32_t alignment_B)
{
const struct isl_format_layout *fmtl = isl_format_get_layout(surf_info->format);
assert(fmtl->bpb % tile_info->format_bpb == 0);
- assert(phys_slice0_sa->w % fmtl->bw == 0);
const uint32_t tile_el_scale = fmtl->bpb / tile_info->format_bpb;
- const uint32_t total_w_el = phys_slice0_sa->width / fmtl->bw;
const uint32_t total_w_tl =
- isl_align_div(total_w_el * tile_el_scale,
+ isl_align_div(phys_total_el->w * tile_el_scale,
tile_info->logical_extent_el.width);
- assert(alignment == tile_info->phys_extent_B.width);
- return total_w_tl * tile_info->phys_extent_B.width;
+ /* In some cases the alignment of the pitch might be > to the tile size
+ * (for example Gen12 CCS requires 512B alignment while the tile's width
+ * can be 128B), so align the row pitch to the alignment.
+ */
+ assert(alignment_B >= tile_info->phys_extent_B.width);
+ return isl_align(total_w_tl * tile_info->phys_extent_B.width, alignment_B);
}
static uint32_t
isl_calc_min_row_pitch(const struct isl_device *dev,
const struct isl_surf_init_info *surf_info,
const struct isl_tile_info *tile_info,
- const struct isl_extent2d *phys_slice0_sa,
- uint32_t alignment)
+ const struct isl_extent2d *phys_total_el,
+ uint32_t alignment_B)
{
if (tile_info->tiling == ISL_TILING_LINEAR) {
- return isl_calc_linear_min_row_pitch(dev, surf_info, phys_slice0_sa,
- alignment);
+ return isl_calc_linear_min_row_pitch(dev, surf_info, phys_total_el,
+ alignment_B);
} else {
return isl_calc_tiled_min_row_pitch(dev, surf_info, tile_info,
- phys_slice0_sa, alignment);
+ phys_total_el, alignment_B);
}
}
const struct isl_surf_init_info *surf_info,
const struct isl_tile_info *tile_info,
enum isl_dim_layout dim_layout,
- const struct isl_extent2d *phys_slice0_sa,
- uint32_t *out_row_pitch)
+ const struct isl_extent2d *phys_total_el,
+ uint32_t *out_row_pitch_B)
{
- const uint32_t alignment =
- isl_calc_row_pitch_alignment(surf_info, tile_info);
-
- const uint32_t min_row_pitch =
- isl_calc_min_row_pitch(dev, surf_info, tile_info, phys_slice0_sa,
- alignment);
+ uint32_t alignment_B =
+ isl_calc_row_pitch_alignment(dev, surf_info, tile_info);
- uint32_t row_pitch = min_row_pitch;
+ const uint32_t min_row_pitch_B =
+ isl_calc_min_row_pitch(dev, surf_info, tile_info, phys_total_el,
+ alignment_B);
- if (surf_info->row_pitch != 0) {
- row_pitch = surf_info->row_pitch;
-
- if (row_pitch < min_row_pitch)
+ if (surf_info->row_pitch_B != 0) {
+ if (surf_info->row_pitch_B < min_row_pitch_B)
return false;
- if (row_pitch % alignment != 0)
+ if (surf_info->row_pitch_B % alignment_B != 0)
return false;
}
- const uint32_t row_pitch_tiles = row_pitch / tile_info->phys_extent_B.width;
+ const uint32_t row_pitch_B =
+ surf_info->row_pitch_B != 0 ? surf_info->row_pitch_B : min_row_pitch_B;
+
+ const uint32_t row_pitch_tl = row_pitch_B / tile_info->phys_extent_B.width;
- if (row_pitch == 0)
+ if (row_pitch_B == 0)
return false;
if (dim_layout == ISL_DIM_LAYOUT_GEN9_1D) {
if ((surf_info->usage & (ISL_SURF_USAGE_RENDER_TARGET_BIT |
ISL_SURF_USAGE_TEXTURE_BIT |
ISL_SURF_USAGE_STORAGE_BIT)) &&
- !pitch_in_range(row_pitch, RENDER_SURFACE_STATE_SurfacePitch_bits(dev->info)))
+ !pitch_in_range(row_pitch_B, RENDER_SURFACE_STATE_SurfacePitch_bits(dev->info)))
return false;
if ((surf_info->usage & (ISL_SURF_USAGE_CCS_BIT |
ISL_SURF_USAGE_MCS_BIT)) &&
- !pitch_in_range(row_pitch_tiles, RENDER_SURFACE_STATE_AuxiliarySurfacePitch_bits(dev->info)))
+ !pitch_in_range(row_pitch_tl, RENDER_SURFACE_STATE_AuxiliarySurfacePitch_bits(dev->info)))
return false;
if ((surf_info->usage & ISL_SURF_USAGE_DEPTH_BIT) &&
- !pitch_in_range(row_pitch, _3DSTATE_DEPTH_BUFFER_SurfacePitch_bits(dev->info)))
+ !pitch_in_range(row_pitch_B, _3DSTATE_DEPTH_BUFFER_SurfacePitch_bits(dev->info)))
return false;
if ((surf_info->usage & ISL_SURF_USAGE_HIZ_BIT) &&
- !pitch_in_range(row_pitch, _3DSTATE_HIER_DEPTH_BUFFER_SurfacePitch_bits(dev->info)))
+ !pitch_in_range(row_pitch_B, _3DSTATE_HIER_DEPTH_BUFFER_SurfacePitch_bits(dev->info)))
return false;
- if (surf_info->usage & ISL_SURF_USAGE_STENCIL_BIT)
- isl_finishme("validate row pitch of stencil surfaces");
+ const uint32_t stencil_pitch_bits = dev->use_separate_stencil ?
+ _3DSTATE_STENCIL_BUFFER_SurfacePitch_bits(dev->info) :
+ _3DSTATE_DEPTH_BUFFER_SurfacePitch_bits(dev->info);
+
+ if ((surf_info->usage & ISL_SURF_USAGE_STENCIL_BIT) &&
+ !pitch_in_range(row_pitch_B, stencil_pitch_bits))
+ return false;
done:
- *out_row_pitch = row_pitch;
+ *out_row_pitch_B = row_pitch_B;
return true;
}
-/**
- * Calculate and apply any padding required for the surface.
- *
- * @param[inout] total_h_el is updated with the new height
- * @param[out] pad_bytes is overwritten with additional padding requirements.
- */
-static void
-isl_apply_surface_padding(const struct isl_device *dev,
- const struct isl_surf_init_info *restrict info,
- const struct isl_tile_info *tile_info,
- uint32_t *total_h_el,
- uint32_t *pad_bytes)
-{
- const struct isl_format_layout *fmtl = isl_format_get_layout(info->format);
-
- *pad_bytes = 0;
-
- /* From the Broadwell PRM >> Volume 5: Memory Views >> Common Surface
- * Formats >> Surface Padding Requirements >> Render Target and Media
- * Surfaces:
- *
- * The data port accesses data (pixels) outside of the surface if they
- * are contained in the same cache request as pixels that are within the
- * surface. These pixels will not be returned by the requesting message,
- * however if these pixels lie outside of defined pages in the GTT,
- * a GTT error will result when the cache request is processed. In
- * order to avoid these GTT errors, “padding” at the bottom of the
- * surface is sometimes necessary.
- *
- * From the Broadwell PRM >> Volume 5: Memory Views >> Common Surface
- * Formats >> Surface Padding Requirements >> Sampling Engine Surfaces:
- *
- * ... Lots of padding requirements, all listed separately below.
- */
-
- /* We can safely ignore the first padding requirement, quoted below,
- * because isl doesn't do buffers.
- *
- * - [pre-BDW] For buffers, which have no inherent “height,” padding
- * requirements are different. A buffer must be padded to the next
- * multiple of 256 array elements, with an additional 16 bytes added
- * beyond that to account for the L1 cache line.
- */
-
- /*
- * - For compressed textures [...], padding at the bottom of the surface
- * is to an even compressed row.
- */
- if (isl_format_is_compressed(info->format))
- *total_h_el = isl_align(*total_h_el, 2);
-
- /*
- * - For cube surfaces, an additional two rows of padding are required
- * at the bottom of the surface.
- */
- if (info->usage & ISL_SURF_USAGE_CUBE_BIT)
- *total_h_el += 2;
-
- /*
- * - For packed YUV, 96 bpt, 48 bpt, and 24 bpt surface formats,
- * additional padding is required. These surfaces require an extra row
- * plus 16 bytes of padding at the bottom in addition to the general
- * padding requirements.
- */
- if (isl_format_is_yuv(info->format) &&
- (fmtl->bpb == 96 || fmtl->bpb == 48|| fmtl->bpb == 24)) {
- *total_h_el += 1;
- *pad_bytes += 16;
- }
-
- /*
- * - 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 (tile_info->tiling == ISL_TILING_LINEAR)
- *pad_bytes += 64;
-
- /* The below text weakens, not strengthens, the padding requirements for
- * linear surfaces. Therefore we can safely ignore it.
- *
- * - [BDW+] For SURFTYPE_BUFFER, SURFTYPE_1D, and SURFTYPE_2D non-array,
- * non-MSAA, non-mip-mapped surfaces in linear memory, the only
- * padding requirement is to the next aligned 64-byte boundary beyond
- * the end of the surface. The rest of the padding requirements
- * documented above do not apply to these surfaces.
- */
-
- /*
- * - [SKL+] For SURFTYPE_2D and SURFTYPE_3D with linear mode and
- * height % 4 != 0, the surface must be padded with
- * 4-(height % 4)*Surface Pitch # of bytes.
- */
- if (ISL_DEV_GEN(dev) >= 9 &&
- tile_info->tiling == ISL_TILING_LINEAR &&
- (info->dim == ISL_SURF_DIM_2D || info->dim == ISL_SURF_DIM_3D)) {
- *total_h_el = isl_align(*total_h_el, 4);
- }
-
- /*
- * - [SKL+] For SURFTYPE_1D with linear mode, the surface must be padded
- * to 4 times the Surface Pitch # of bytes
- */
- if (ISL_DEV_GEN(dev) >= 9 &&
- tile_info->tiling == ISL_TILING_LINEAR &&
- info->dim == ISL_SURF_DIM_1D) {
- *total_h_el += 4;
- }
-}
-
bool
isl_surf_init_s(const struct isl_device *dev,
struct isl_surf *surf,
isl_tiling_get_info(tiling, fmtl->bpb, &tile_info);
const enum isl_dim_layout dim_layout =
- isl_surf_choose_dim_layout(dev, info->dim, tiling);
+ isl_surf_choose_dim_layout(dev, info->dim, tiling, info->usage);
enum isl_msaa_layout msaa_layout;
if (!isl_choose_msaa_layout(dev, info, tiling, &msaa_layout))
struct isl_extent4d phys_level0_sa;
isl_calc_phys_level0_extent_sa(dev, info, dim_layout, tiling, msaa_layout,
&phys_level0_sa);
- assert(phys_level0_sa.w % fmtl->bw == 0);
- assert(phys_level0_sa.h % fmtl->bh == 0);
enum isl_array_pitch_span array_pitch_span =
isl_choose_array_pitch_span(dev, info, dim_layout, &phys_level0_sa);
- struct isl_extent2d phys_slice0_sa;
- isl_calc_phys_slice0_extent_sa(dev, info, dim_layout, msaa_layout,
- &image_align_sa, &phys_level0_sa,
- &phys_slice0_sa);
- assert(phys_slice0_sa.w % fmtl->bw == 0);
- assert(phys_slice0_sa.h % fmtl->bh == 0);
-
- const uint32_t array_pitch_el_rows =
- isl_calc_array_pitch_el_rows(dev, info, &tile_info, dim_layout,
- array_pitch_span, &image_align_sa,
- &phys_level0_sa, &phys_slice0_sa);
+ uint32_t array_pitch_el_rows;
+ struct isl_extent2d phys_total_el;
+ isl_calc_phys_total_extent_el(dev, info, &tile_info,
+ dim_layout, msaa_layout,
+ &image_align_sa, &phys_level0_sa,
+ array_pitch_span, &array_pitch_el_rows,
+ &phys_total_el);
- uint32_t total_h_el = phys_level0_sa.array_len * array_pitch_el_rows;
-
- uint32_t pad_bytes;
- isl_apply_surface_padding(dev, info, &tile_info, &total_h_el, &pad_bytes);
-
- uint32_t row_pitch;
+ uint32_t row_pitch_B;
if (!isl_calc_row_pitch(dev, info, &tile_info, dim_layout,
- &phys_slice0_sa, &row_pitch))
+ &phys_total_el, &row_pitch_B))
return false;
- uint32_t size, base_alignment;
+ uint32_t base_alignment_B;
+ uint64_t size_B;
if (tiling == ISL_TILING_LINEAR) {
- size = row_pitch * total_h_el + pad_bytes;
+ size_B = (uint64_t) row_pitch_B * phys_total_el.h;
/* From the Broadwell PRM Vol 2d, RENDER_SURFACE_STATE::SurfaceBaseAddress:
*
* surfaces have no alignment requirements (byte alignment is
* sufficient.)"
*/
- base_alignment = MAX(1, info->min_alignment);
+ base_alignment_B = MAX(1, info->min_alignment_B);
if (info->usage & ISL_SURF_USAGE_RENDER_TARGET_BIT) {
if (isl_format_is_yuv(info->format)) {
- base_alignment = MAX(base_alignment, fmtl->bpb / 4);
+ base_alignment_B = MAX(base_alignment_B, fmtl->bpb / 4);
} else {
- base_alignment = MAX(base_alignment, fmtl->bpb / 8);
+ base_alignment_B = MAX(base_alignment_B, fmtl->bpb / 8);
}
}
- base_alignment = isl_round_up_to_power_of_two(base_alignment);
+ base_alignment_B = isl_round_up_to_power_of_two(base_alignment_B);
+
+ /* From the Skylake PRM Vol 2c, PLANE_STRIDE::Stride:
+ *
+ * "For Linear memory, this field specifies the stride in chunks of
+ * 64 bytes (1 cache line)."
+ */
+ if (isl_surf_usage_is_display(info->usage))
+ base_alignment_B = MAX(base_alignment_B, 64);
} else {
- total_h_el += isl_align_div_npot(pad_bytes, row_pitch);
const uint32_t total_h_tl =
- isl_align_div(total_h_el, tile_info.logical_extent_el.height);
+ isl_align_div(phys_total_el.h, tile_info.logical_extent_el.height);
- size = total_h_tl * tile_info.phys_extent_B.height * row_pitch;
+ size_B = (uint64_t) total_h_tl * tile_info.phys_extent_B.height * row_pitch_B;
- const uint32_t tile_size = tile_info.phys_extent_B.width *
- tile_info.phys_extent_B.height;
- assert(isl_is_pow2(info->min_alignment) && isl_is_pow2(tile_size));
- base_alignment = MAX(info->min_alignment, tile_size);
- }
+ const uint32_t tile_size_B = tile_info.phys_extent_B.width *
+ tile_info.phys_extent_B.height;
+ assert(isl_is_pow2(info->min_alignment_B) && isl_is_pow2(tile_size_B));
+ base_alignment_B = MAX(info->min_alignment_B, tile_size_B);
- *surf = (struct isl_surf) {
- .dim = info->dim,
+ /* The diagram in the Bspec section Memory Compression - Gen12, shows
+ * that the CCS is indexed in 256B chunks. However, the
+ * PLANE_AUX_DIST::Auxiliary Surface Distance field is in units of 4K
+ * pages. We currently don't assign the usage field like we do for main
+ * surfaces, so just use 4K for now.
+ */
+ if (tiling == ISL_TILING_GEN12_CCS)
+ base_alignment_B = MAX(base_alignment_B, 4096);
+
+ /* Gen12+ requires that images be 64K-aligned if they're going to used
+ * with CCS. This is because the Aux translation table maps main
+ * surface addresses to aux addresses at a 64K (in the main surface)
+ * granularity. Because we don't know for sure in ISL if a surface will
+ * use CCS, we have to guess based on the DISABLE_AUX usage bit. The
+ * one thing we do know is that we haven't enable CCS on linear images
+ * yet so we can avoid the extra alignment there.
+ */
+ if (ISL_DEV_GEN(dev) >= 12 &&
+ !(info->usage & ISL_SURF_USAGE_DISABLE_AUX_BIT)) {
+ base_alignment_B = MAX(base_alignment_B, 64 * 1024);
+ }
+ }
+
+ if (ISL_DEV_GEN(dev) < 9) {
+ /* From the Broadwell PRM Vol 5, Surface Layout:
+ *
+ * "In addition to restrictions on maximum height, width, and depth,
+ * surfaces are also restricted to a maximum size in bytes. This
+ * maximum is 2 GB for all products and all surface types."
+ *
+ * This comment is applicable to all Pre-gen9 platforms.
+ */
+ if (size_B > (uint64_t) 1 << 31)
+ return false;
+ } else if (ISL_DEV_GEN(dev) < 11) {
+ /* From the Skylake PRM Vol 5, Maximum Surface Size in Bytes:
+ * "In addition to restrictions on maximum height, width, and depth,
+ * surfaces are also restricted to a maximum size of 2^38 bytes.
+ * All pixels within the surface must be contained within 2^38 bytes
+ * of the base address."
+ */
+ if (size_B > (uint64_t) 1 << 38)
+ return false;
+ } else {
+ /* gen11+ platforms raised this limit to 2^44 bytes. */
+ if (size_B > (uint64_t) 1 << 44)
+ return false;
+ }
+
+ *surf = (struct isl_surf) {
+ .dim = info->dim,
.dim_layout = dim_layout,
.msaa_layout = msaa_layout,
.tiling = tiling,
.logical_level0_px = logical_level0_px,
.phys_level0_sa = phys_level0_sa,
- .size = size,
- .alignment = base_alignment,
- .row_pitch = row_pitch,
+ .size_B = size_B,
+ .alignment_B = base_alignment_B,
+ .row_pitch_B = row_pitch_B,
.array_pitch_el_rows = array_pitch_el_rows,
.array_pitch_span = array_pitch_span,
{
assert(ISL_DEV_GEN(dev) >= 5 && ISL_DEV_USE_SEPARATE_STENCIL(dev));
+ if (!isl_surf_usage_is_depth(surf->usage))
+ return false;
+
+ /* HiZ only works with Y-tiled depth buffers */
+ if (!isl_tiling_is_any_y(surf->tiling))
+ return false;
+
+ /* On SNB+, compressed depth buffers cannot be interleaved with stencil. */
+ switch (surf->format) {
+ case ISL_FORMAT_R24_UNORM_X8_TYPELESS:
+ if (isl_surf_usage_is_depth_and_stencil(surf->usage)) {
+ assert(ISL_DEV_GEN(dev) == 5);
+ unreachable("This should work, but is untested");
+ }
+ /* Fall through */
+ case ISL_FORMAT_R16_UNORM:
+ case ISL_FORMAT_R32_FLOAT:
+ break;
+ case ISL_FORMAT_R32_FLOAT_X8X24_TYPELESS:
+ if (ISL_DEV_GEN(dev) == 5) {
+ assert(isl_surf_usage_is_depth_and_stencil(surf->usage));
+ unreachable("This should work, but is untested");
+ }
+ /* Fall through */
+ default:
+ return false;
+ }
+
/* Multisampled depth is always interleaved */
assert(surf->msaa_layout == ISL_MSAA_LAYOUT_NONE ||
surf->msaa_layout == ISL_MSAA_LAYOUT_INTERLEAVED);
struct isl_surf *mcs_surf)
{
/* It must be multisampled with an array layout */
- assert(surf->samples > 1 && surf->msaa_layout == ISL_MSAA_LAYOUT_ARRAY);
+ if (surf->msaa_layout != ISL_MSAA_LAYOUT_ARRAY)
+ return false;
+
+ if (mcs_surf->size_B > 0)
+ return false;
/* The following are true of all multisampled surfaces */
+ assert(surf->samples > 1);
assert(surf->dim == ISL_SURF_DIM_2D);
assert(surf->levels == 1);
assert(surf->logical_level0_px.depth == 1);
+ /* From the Ivy Bridge PRM, Vol4 Part1 p77 ("MCS Enable"):
+ *
+ * This field must be set to 0 for all SINT MSRTs when all RT channels
+ * are not written
+ *
+ * In practice this means that we have to disable MCS for all signed
+ * integer MSAA buffers. The alternative, to disable MCS only when one
+ * of the render target channels is disabled, is impractical because it
+ * would require converting between CMS and UMS MSAA layouts on the fly,
+ * which is expensive.
+ */
+ if (ISL_DEV_GEN(dev) == 7 && isl_format_has_sint_channel(surf->format))
+ return false;
+
/* The "Auxiliary Surface Pitch" field in RENDER_SURFACE_STATE is only 9
* bits which means the maximum pitch of a compression surface is 512
* tiles or 64KB (since MCS is always Y-tiled). Since a 16x MCS buffer is
}
bool
-isl_surf_get_ccs_surf(const struct isl_device *dev,
- const struct isl_surf *surf,
- struct isl_surf *ccs_surf)
+isl_surf_supports_ccs(const struct isl_device *dev,
+ const struct isl_surf *surf)
{
- assert(surf->samples == 1 && surf->msaa_layout == ISL_MSAA_LAYOUT_NONE);
- assert(ISL_DEV_GEN(dev) >= 7);
+ /* CCS support does not exist prior to Gen7 */
+ if (ISL_DEV_GEN(dev) <= 6)
+ return false;
if (surf->usage & ISL_SURF_USAGE_DISABLE_AUX_BIT)
return false;
- if (ISL_DEV_GEN(dev) <= 8 && surf->dim != ISL_SURF_DIM_2D)
+ if (isl_format_is_compressed(surf->format))
return false;
- if (isl_format_is_compressed(surf->format))
+ if (!isl_is_pow2(isl_format_get_layout(surf->format)->bpb))
return false;
- /* TODO: More conditions where it can fail. */
+ /* From the Ivy Bridge PRM, Vol2 Part1 11.7 "MCS Buffer for Render
+ * Target(s)", beneath the "Fast Color Clear" bullet (p326):
+ *
+ * - Support is limited to tiled render targets.
+ *
+ * From the Skylake documentation, it is made clear that X-tiling is no
+ * longer supported:
+ *
+ * - MCS and Lossless compression is supported for
+ * TiledY/TileYs/TileYf non-MSRTs only.
+ *
+ * From the BSpec (44930) for Gen12:
+ *
+ * Linear CCS is only allowed for Untyped Buffers but only via HDC
+ * Data-Port messages.
+ *
+ * We never use untyped messages on surfaces created by ISL on Gen9+ so
+ * this means linear is out on Gen12+ as well.
+ */
+ if (surf->tiling == ISL_TILING_LINEAR)
+ return false;
- enum isl_format ccs_format;
- if (ISL_DEV_GEN(dev) >= 9) {
- if (!isl_tiling_is_any_y(surf->tiling))
+ if (ISL_DEV_GEN(dev) >= 12) {
+ if (isl_surf_usage_is_stencil(surf->usage) && surf->samples > 1)
return false;
- switch (isl_format_get_layout(surf->format)->bpb) {
- case 32: ccs_format = ISL_FORMAT_GEN9_CCS_32BPP; break;
- case 64: ccs_format = ISL_FORMAT_GEN9_CCS_64BPP; break;
- case 128: ccs_format = ISL_FORMAT_GEN9_CCS_128BPP; break;
- default:
+ /* On Gen12, 8BPP surfaces cannot be compressed if any level is not
+ * 32Bx4row-aligned. For now, just reject the cases where alignment
+ * matters.
+ */
+ if (isl_format_get_layout(surf->format)->bpb == 8 && surf->levels >= 3) {
+ isl_finishme("%s:%s: CCS for 8BPP textures with 3+ miplevels is "
+ "disabled, but support for more levels is possible.",
+ __FILE__, __func__);
return false;
}
- } else if (surf->tiling == ISL_TILING_Y0) {
- switch (isl_format_get_layout(surf->format)->bpb) {
- case 32: ccs_format = ISL_FORMAT_GEN7_CCS_32BPP_Y; break;
- case 64: ccs_format = ISL_FORMAT_GEN7_CCS_64BPP_Y; break;
- case 128: ccs_format = ISL_FORMAT_GEN7_CCS_128BPP_Y; break;
- default:
+
+ /* On Gen12, all CCS-compressed surface pitches must be multiples of
+ * 512B.
+ */
+ if (surf->row_pitch_B % 512 != 0)
+ return false;
+
+ /* According to GEN:BUG:1406738321, 3D textures need a blit to a new
+ * surface in order to perform a resolve. For now, just disable CCS.
+ */
+ if (surf->dim == ISL_SURF_DIM_3D) {
+ isl_finishme("%s:%s: CCS for 3D textures is disabled, but a workaround"
+ " is available.", __FILE__, __func__);
return false;
}
- } else if (surf->tiling == ISL_TILING_X) {
+
+ /* GEN:BUG:1207137018
+ *
+ * TODO: implement following workaround currently covered by the
+ * restriction above. If following conditions are met:
+ *
+ * - RENDER_SURFACE_STATE.Surface Type == 3D
+ * - RENDER_SURFACE_STATE.Auxiliary Surface Mode != AUX_NONE
+ * - RENDER_SURFACE_STATE.Tiled ResourceMode is TYF or TYS
+ *
+ * Set the value of RENDER_SURFACE_STATE.Mip Tail Start LOD to a mip
+ * that larger than those present in the surface (i.e. 15)
+ */
+
+ /* TODO: Handle the other tiling formats */
+ if (surf->tiling != ISL_TILING_Y0)
+ return false;
+ } else {
+ /* ISL_DEV_GEN(dev) < 12 */
+ if (surf->samples > 1)
+ return false;
+
+ /* CCS is only for color images on Gen7-11 */
+ if (isl_surf_usage_is_depth_or_stencil(surf->usage))
+ return false;
+
+ /* The PRM doesn't say this explicitly, but fast-clears don't appear to
+ * work for 3D textures until gen9 where the layout of 3D textures
+ * changes to match 2D array textures.
+ */
+ if (ISL_DEV_GEN(dev) <= 8 && surf->dim != ISL_SURF_DIM_2D)
+ return false;
+
+ /* From the HSW PRM Volume 7: 3D-Media-GPGPU, page 652 (Color Clear of
+ * Non-MultiSampler Render Target Restrictions):
+ *
+ * "Support is for non-mip-mapped and non-array surface types only."
+ *
+ * This restriction is lifted on gen8+. Technically, it may be possible
+ * to create a CCS for an arrayed or mipmapped image and only enable
+ * CCS_D when rendering to the base slice. However, there is no
+ * documentation tell us what the hardware would do in that case or what
+ * it does if you walk off the bases slice. (Does it ignore CCS or does
+ * it start scribbling over random memory?) We play it safe and just
+ * follow the docs and don't allow CCS_D for arrayed or mip-mapped
+ * surfaces.
+ */
+ if (ISL_DEV_GEN(dev) <= 7 &&
+ (surf->levels > 1 || surf->logical_level0_px.array_len > 1))
+ return false;
+
+ /* From the Ivy Bridge PRM, Vol2 Part1 11.7 "MCS Buffer for Render
+ * Target(s)", beneath the "Fast Color Clear" bullet (p326):
+ *
+ * - MCS buffer for non-MSRT is supported only for RT formats 32bpp,
+ * 64bpp, and 128bpp.
+ */
+ if (isl_format_get_layout(surf->format)->bpb < 32)
+ return false;
+
+ /* From the Skylake documentation, it is made clear that X-tiling is no
+ * longer supported:
+ *
+ * - MCS and Lossless compression is supported for
+ * TiledY/TileYs/TileYf non-MSRTs only.
+ */
+ if (ISL_DEV_GEN(dev) >= 9 && !isl_tiling_is_any_y(surf->tiling))
+ return false;
+ }
+
+ return true;
+}
+
+bool
+isl_surf_get_ccs_surf(const struct isl_device *dev,
+ const struct isl_surf *surf,
+ struct isl_surf *aux_surf,
+ struct isl_surf *extra_aux_surf,
+ uint32_t row_pitch_B)
+{
+ assert(aux_surf);
+
+ /* An uninitialized surface is needed to get a CCS surface. */
+ if (aux_surf->size_B > 0 &&
+ (extra_aux_surf == NULL || extra_aux_surf->size_B > 0)) {
+ return false;
+ }
+
+ /* A surface can't have two CCS surfaces. */
+ if (aux_surf->usage & ISL_SURF_USAGE_CCS_BIT)
+ return false;
+
+ if (!isl_surf_supports_ccs(dev, surf))
+ return false;
+
+ if (ISL_DEV_GEN(dev) >= 12) {
+ enum isl_format ccs_format;
switch (isl_format_get_layout(surf->format)->bpb) {
- case 32: ccs_format = ISL_FORMAT_GEN7_CCS_32BPP_X; break;
- case 64: ccs_format = ISL_FORMAT_GEN7_CCS_64BPP_X; break;
- case 128: ccs_format = ISL_FORMAT_GEN7_CCS_128BPP_X; break;
+ case 8: ccs_format = ISL_FORMAT_GEN12_CCS_8BPP_Y0; break;
+ case 16: ccs_format = ISL_FORMAT_GEN12_CCS_16BPP_Y0; break;
+ case 32: ccs_format = ISL_FORMAT_GEN12_CCS_32BPP_Y0; break;
+ case 64: ccs_format = ISL_FORMAT_GEN12_CCS_64BPP_Y0; break;
+ case 128: ccs_format = ISL_FORMAT_GEN12_CCS_128BPP_Y0; break;
default:
return false;
}
+
+ /* On Gen12, the CCS is a scaled-down version of the main surface. We
+ * model this as the CCS compressing a 2D-view of the entire surface.
+ */
+ struct isl_surf *ccs_surf =
+ aux_surf->size_B > 0 ? extra_aux_surf : aux_surf;
+ const bool ok =
+ isl_surf_init(dev, ccs_surf,
+ .dim = ISL_SURF_DIM_2D,
+ .format = ccs_format,
+ .width = isl_surf_get_row_pitch_el(surf),
+ .height = surf->size_B / surf->row_pitch_B,
+ .depth = 1,
+ .levels = 1,
+ .array_len = 1,
+ .samples = 1,
+ .row_pitch_B = row_pitch_B,
+ .usage = ISL_SURF_USAGE_CCS_BIT,
+ .tiling_flags = ISL_TILING_GEN12_CCS_BIT);
+ assert(!ok || ccs_surf->size_B == surf->size_B / 256);
+ return ok;
} else {
- return false;
- }
+ enum isl_format ccs_format;
+ if (ISL_DEV_GEN(dev) >= 9) {
+ switch (isl_format_get_layout(surf->format)->bpb) {
+ case 32: ccs_format = ISL_FORMAT_GEN9_CCS_32BPP; break;
+ case 64: ccs_format = ISL_FORMAT_GEN9_CCS_64BPP; break;
+ case 128: ccs_format = ISL_FORMAT_GEN9_CCS_128BPP; break;
+ default: unreachable("Unsupported CCS format");
+ return false;
+ }
+ } else if (surf->tiling == ISL_TILING_Y0) {
+ switch (isl_format_get_layout(surf->format)->bpb) {
+ case 32: ccs_format = ISL_FORMAT_GEN7_CCS_32BPP_Y; break;
+ case 64: ccs_format = ISL_FORMAT_GEN7_CCS_64BPP_Y; break;
+ case 128: ccs_format = ISL_FORMAT_GEN7_CCS_128BPP_Y; break;
+ default: unreachable("Unsupported CCS format");
+ }
+ } else if (surf->tiling == ISL_TILING_X) {
+ switch (isl_format_get_layout(surf->format)->bpb) {
+ case 32: ccs_format = ISL_FORMAT_GEN7_CCS_32BPP_X; break;
+ case 64: ccs_format = ISL_FORMAT_GEN7_CCS_64BPP_X; break;
+ case 128: ccs_format = ISL_FORMAT_GEN7_CCS_128BPP_X; break;
+ default: unreachable("Unsupported CCS format");
+ }
+ } else {
+ unreachable("Invalid tiling format");
+ }
- return isl_surf_init(dev, ccs_surf,
- .dim = surf->dim,
- .format = ccs_format,
- .width = surf->logical_level0_px.width,
- .height = surf->logical_level0_px.height,
- .depth = surf->logical_level0_px.depth,
- .levels = surf->levels,
- .array_len = surf->logical_level0_px.array_len,
- .samples = 1,
- .usage = ISL_SURF_USAGE_CCS_BIT,
- .tiling_flags = ISL_TILING_CCS_BIT);
+ return isl_surf_init(dev, aux_surf,
+ .dim = surf->dim,
+ .format = ccs_format,
+ .width = surf->logical_level0_px.width,
+ .height = surf->logical_level0_px.height,
+ .depth = surf->logical_level0_px.depth,
+ .levels = surf->levels,
+ .array_len = surf->logical_level0_px.array_len,
+ .samples = 1,
+ .row_pitch_B = row_pitch_B,
+ .usage = ISL_SURF_USAGE_CCS_BIT,
+ .tiling_flags = ISL_TILING_CCS_BIT);
+ }
}
+#define isl_genX_call(dev, func, ...) \
+ switch (ISL_DEV_GEN(dev)) { \
+ case 4: \
+ /* G45 surface state is the same as gen5 */ \
+ if (ISL_DEV_IS_G4X(dev)) { \
+ isl_gen5_##func(__VA_ARGS__); \
+ } else { \
+ isl_gen4_##func(__VA_ARGS__); \
+ } \
+ break; \
+ case 5: \
+ isl_gen5_##func(__VA_ARGS__); \
+ break; \
+ case 6: \
+ isl_gen6_##func(__VA_ARGS__); \
+ break; \
+ case 7: \
+ if (ISL_DEV_IS_HASWELL(dev)) { \
+ isl_gen75_##func(__VA_ARGS__); \
+ } else { \
+ isl_gen7_##func(__VA_ARGS__); \
+ } \
+ break; \
+ case 8: \
+ isl_gen8_##func(__VA_ARGS__); \
+ break; \
+ case 9: \
+ isl_gen9_##func(__VA_ARGS__); \
+ break; \
+ case 10: \
+ isl_gen10_##func(__VA_ARGS__); \
+ break; \
+ case 11: \
+ isl_gen11_##func(__VA_ARGS__); \
+ break; \
+ case 12: \
+ isl_gen12_##func(__VA_ARGS__); \
+ break; \
+ default: \
+ assert(!"Unknown hardware generation"); \
+ }
+
void
isl_surf_fill_state_s(const struct isl_device *dev, void *state,
const struct isl_surf_fill_state_info *restrict info)
info->surf->logical_level0_px.array_len);
}
- switch (ISL_DEV_GEN(dev)) {
- case 4:
- if (ISL_DEV_IS_G4X(dev)) {
- /* G45 surface state is the same as gen5 */
- isl_gen5_surf_fill_state_s(dev, state, info);
- } else {
- isl_gen4_surf_fill_state_s(dev, state, info);
- }
- break;
- case 5:
- isl_gen5_surf_fill_state_s(dev, state, info);
- break;
- case 6:
- isl_gen6_surf_fill_state_s(dev, state, info);
- break;
- case 7:
- if (ISL_DEV_IS_HASWELL(dev)) {
- isl_gen75_surf_fill_state_s(dev, state, info);
- } else {
- isl_gen7_surf_fill_state_s(dev, state, info);
- }
- break;
- case 8:
- isl_gen8_surf_fill_state_s(dev, state, info);
- break;
- case 9:
- isl_gen9_surf_fill_state_s(dev, state, info);
- break;
- default:
- assert(!"Cannot fill surface state for this gen");
- }
+ isl_genX_call(dev, surf_fill_state_s, dev, state, info);
}
void
isl_buffer_fill_state_s(const struct isl_device *dev, void *state,
const struct isl_buffer_fill_state_info *restrict info)
{
- switch (ISL_DEV_GEN(dev)) {
- case 4:
- case 5:
- /* Gen 4-5 are all the same when it comes to buffer surfaces */
- isl_gen5_buffer_fill_state_s(state, info);
- break;
- case 6:
- isl_gen6_buffer_fill_state_s(state, info);
- break;
- case 7:
- if (ISL_DEV_IS_HASWELL(dev)) {
- isl_gen75_buffer_fill_state_s(state, info);
- } else {
- isl_gen7_buffer_fill_state_s(state, info);
- }
- break;
- case 8:
- isl_gen8_buffer_fill_state_s(state, info);
- break;
- case 9:
- isl_gen9_buffer_fill_state_s(state, info);
- break;
- default:
- assert(!"Cannot fill surface state for this gen");
- }
+ isl_genX_call(dev, buffer_fill_state_s, dev, state, info);
+}
+
+void
+isl_null_fill_state(const struct isl_device *dev, void *state,
+ struct isl_extent3d size)
+{
+ isl_genX_call(dev, null_fill_state, state, size);
}
void
}
}
- switch (ISL_DEV_GEN(dev)) {
- case 4:
- if (ISL_DEV_IS_G4X(dev)) {
- /* G45 surface state is the same as gen5 */
- isl_gen5_emit_depth_stencil_hiz_s(dev, batch, info);
- } else {
- isl_gen4_emit_depth_stencil_hiz_s(dev, batch, info);
- }
- break;
- case 5:
- isl_gen5_emit_depth_stencil_hiz_s(dev, batch, info);
- break;
- case 6:
- isl_gen6_emit_depth_stencil_hiz_s(dev, batch, info);
- break;
- case 7:
- if (ISL_DEV_IS_HASWELL(dev)) {
- isl_gen75_emit_depth_stencil_hiz_s(dev, batch, info);
- } else {
- isl_gen7_emit_depth_stencil_hiz_s(dev, batch, info);
- }
- break;
- case 8:
- isl_gen8_emit_depth_stencil_hiz_s(dev, batch, info);
- break;
- case 9:
- isl_gen9_emit_depth_stencil_hiz_s(dev, batch, info);
- break;
- default:
- assert(!"Cannot fill surface state for this gen");
- }
+ isl_genX_call(dev, emit_depth_stencil_hiz_s, dev, batch, info);
}
/**
*y_offset_sa = y;
}
+static void
+get_image_offset_sa_gen6_stencil_hiz(const struct isl_surf *surf,
+ uint32_t level,
+ uint32_t logical_array_layer,
+ uint32_t *x_offset_sa,
+ uint32_t *y_offset_sa)
+{
+ assert(level < surf->levels);
+ assert(surf->logical_level0_px.depth == 1);
+ assert(logical_array_layer < surf->logical_level0_px.array_len);
+
+ const struct isl_format_layout *fmtl = isl_format_get_layout(surf->format);
+
+ const struct isl_extent3d image_align_sa =
+ isl_surf_get_image_alignment_sa(surf);
+
+ struct isl_tile_info tile_info;
+ isl_tiling_get_info(surf->tiling, fmtl->bpb, &tile_info);
+ const struct isl_extent2d tile_extent_sa = {
+ .w = tile_info.logical_extent_el.w * fmtl->bw,
+ .h = tile_info.logical_extent_el.h * fmtl->bh,
+ };
+ /* Tile size is a multiple of image alignment */
+ assert(tile_extent_sa.w % image_align_sa.w == 0);
+ assert(tile_extent_sa.h % image_align_sa.h == 0);
+
+ const uint32_t W0 = surf->phys_level0_sa.w;
+ const uint32_t H0 = surf->phys_level0_sa.h;
+
+ /* Each image has the same height as LOD0 because the hardware thinks
+ * everything is LOD0
+ */
+ const uint32_t H = isl_align(H0, image_align_sa.h);
+
+ /* Quick sanity check for consistency */
+ if (surf->phys_level0_sa.array_len > 1)
+ assert(surf->array_pitch_el_rows == isl_assert_div(H, fmtl->bh));
+
+ uint32_t x = 0, y = 0;
+ for (uint32_t l = 0; l < level; ++l) {
+ const uint32_t W = isl_minify(W0, l);
+
+ const uint32_t w = isl_align(W, tile_extent_sa.w);
+ const uint32_t h = isl_align(H * surf->phys_level0_sa.a,
+ tile_extent_sa.h);
+
+ if (l == 0) {
+ y += h;
+ } else {
+ x += w;
+ }
+ }
+
+ y += H * logical_array_layer;
+
+ *x_offset_sa = x;
+ *y_offset_sa = y;
+}
+
/**
* A variant of isl_surf_get_image_offset_sa() specific to
* ISL_DIM_LAYOUT_GEN9_1D.
logical_z_offset_px,
x_offset_sa, y_offset_sa);
break;
+ case ISL_DIM_LAYOUT_GEN6_STENCIL_HIZ:
+ get_image_offset_sa_gen6_stencil_hiz(surf, level, logical_array_layer +
+ logical_z_offset_px,
+ x_offset_sa, y_offset_sa);
+ break;
default:
unreachable("not reached");
*y_offset_el = y_offset_sa / fmtl->bh;
}
+void
+isl_surf_get_image_offset_B_tile_sa(const struct isl_surf *surf,
+ uint32_t level,
+ uint32_t logical_array_layer,
+ uint32_t logical_z_offset_px,
+ uint32_t *offset_B,
+ uint32_t *x_offset_sa,
+ uint32_t *y_offset_sa)
+{
+ const struct isl_format_layout *fmtl = isl_format_get_layout(surf->format);
+
+ uint32_t total_x_offset_el, total_y_offset_el;
+ isl_surf_get_image_offset_el(surf, level, logical_array_layer,
+ logical_z_offset_px,
+ &total_x_offset_el,
+ &total_y_offset_el);
+
+ uint32_t x_offset_el, y_offset_el;
+ isl_tiling_get_intratile_offset_el(surf->tiling, fmtl->bpb,
+ surf->row_pitch_B,
+ total_x_offset_el,
+ total_y_offset_el,
+ offset_B,
+ &x_offset_el,
+ &y_offset_el);
+
+ if (x_offset_sa) {
+ *x_offset_sa = x_offset_el * fmtl->bw;
+ } else {
+ assert(x_offset_el == 0);
+ }
+
+ if (y_offset_sa) {
+ *y_offset_sa = y_offset_el * fmtl->bh;
+ } else {
+ assert(y_offset_el == 0);
+ }
+}
+
+void
+isl_surf_get_image_range_B_tile(const struct isl_surf *surf,
+ uint32_t level,
+ uint32_t logical_array_layer,
+ uint32_t logical_z_offset_px,
+ uint32_t *start_tile_B,
+ uint32_t *end_tile_B)
+{
+ uint32_t start_x_offset_el, start_y_offset_el;
+ isl_surf_get_image_offset_el(surf, level, logical_array_layer,
+ logical_z_offset_px,
+ &start_x_offset_el,
+ &start_y_offset_el);
+
+ /* Compute the size of the subimage in surface elements */
+ const uint32_t subimage_w_sa = isl_minify(surf->phys_level0_sa.w, level);
+ const uint32_t subimage_h_sa = isl_minify(surf->phys_level0_sa.h, level);
+ const struct isl_format_layout *fmtl = isl_format_get_layout(surf->format);
+ const uint32_t subimage_w_el = isl_align_div_npot(subimage_w_sa, fmtl->bw);
+ const uint32_t subimage_h_el = isl_align_div_npot(subimage_h_sa, fmtl->bh);
+
+ /* Find the last pixel */
+ uint32_t end_x_offset_el = start_x_offset_el + subimage_w_el - 1;
+ uint32_t end_y_offset_el = start_y_offset_el + subimage_h_el - 1;
+
+ UNUSED uint32_t x_offset_el, y_offset_el;
+ isl_tiling_get_intratile_offset_el(surf->tiling, fmtl->bpb,
+ surf->row_pitch_B,
+ start_x_offset_el,
+ start_y_offset_el,
+ start_tile_B,
+ &x_offset_el,
+ &y_offset_el);
+
+ isl_tiling_get_intratile_offset_el(surf->tiling, fmtl->bpb,
+ surf->row_pitch_B,
+ end_x_offset_el,
+ end_y_offset_el,
+ end_tile_B,
+ &x_offset_el,
+ &y_offset_el);
+
+ /* We want the range we return to be exclusive but the tile containing the
+ * last pixel (what we just calculated) is inclusive. Add one.
+ */
+ (*end_tile_B)++;
+
+ assert(*end_tile_B <= surf->size_B);
+}
+
+void
+isl_surf_get_image_surf(const struct isl_device *dev,
+ const struct isl_surf *surf,
+ uint32_t level,
+ uint32_t logical_array_layer,
+ uint32_t logical_z_offset_px,
+ struct isl_surf *image_surf,
+ uint32_t *offset_B,
+ uint32_t *x_offset_sa,
+ uint32_t *y_offset_sa)
+{
+ isl_surf_get_image_offset_B_tile_sa(surf,
+ level,
+ logical_array_layer,
+ logical_z_offset_px,
+ offset_B,
+ x_offset_sa,
+ y_offset_sa);
+
+ /* Even for cube maps there will be only single face, therefore drop the
+ * corresponding flag if present.
+ */
+ const isl_surf_usage_flags_t usage =
+ surf->usage & (~ISL_SURF_USAGE_CUBE_BIT);
+
+ bool ok UNUSED;
+ ok = isl_surf_init(dev, image_surf,
+ .dim = ISL_SURF_DIM_2D,
+ .format = surf->format,
+ .width = isl_minify(surf->logical_level0_px.w, level),
+ .height = isl_minify(surf->logical_level0_px.h, level),
+ .depth = 1,
+ .levels = 1,
+ .array_len = 1,
+ .samples = surf->samples,
+ .row_pitch_B = surf->row_pitch_B,
+ .usage = usage,
+ .tiling_flags = (1 << surf->tiling));
+ assert(ok);
+}
+
void
isl_tiling_get_intratile_offset_el(enum isl_tiling tiling,
- uint8_t bs,
- uint32_t row_pitch,
+ uint32_t bpb,
+ uint32_t row_pitch_B,
uint32_t total_x_offset_el,
uint32_t total_y_offset_el,
uint32_t *base_address_offset,
uint32_t *y_offset_el)
{
if (tiling == ISL_TILING_LINEAR) {
- *base_address_offset = total_y_offset_el * row_pitch +
- total_x_offset_el * bs;
+ assert(bpb % 8 == 0);
+ *base_address_offset = total_y_offset_el * row_pitch_B +
+ total_x_offset_el * (bpb / 8);
*x_offset_el = 0;
*y_offset_el = 0;
return;
}
- const uint32_t bpb = bs * 8;
-
struct isl_tile_info tile_info;
isl_tiling_get_info(tiling, bpb, &tile_info);
- assert(row_pitch % tile_info.phys_extent_B.width == 0);
+ assert(row_pitch_B % tile_info.phys_extent_B.width == 0);
/* For non-power-of-two formats, we need the address to be both tile and
* element-aligned. The easiest way to achieve this is to work with a tile
uint32_t y_offset_tl = total_y_offset_el / tile_info.logical_extent_el.h;
*base_address_offset =
- y_offset_tl * tile_info.phys_extent_B.h * row_pitch +
+ y_offset_tl * tile_info.phys_extent_B.h * row_pitch_B +
x_offset_tl * tile_info.phys_extent_B.h * tile_info.phys_extent_B.w;
}
return 5; /* D16_UNORM */
}
}
+
+bool
+isl_swizzle_supports_rendering(const struct gen_device_info *devinfo,
+ struct isl_swizzle swizzle)
+{
+ if (devinfo->is_haswell) {
+ /* From the Haswell PRM,
+ * RENDER_SURFACE_STATE::Shader Channel Select Red
+ *
+ * "The Shader channel selects also define which shader channels are
+ * written to which surface channel. If the Shader channel select is
+ * SCS_ZERO or SCS_ONE then it is not written to the surface. If the
+ * shader channel select is SCS_RED it is written to the surface red
+ * channel and so on. If more than one shader channel select is set
+ * to the same surface channel only the first shader channel in RGBA
+ * order will be written."
+ */
+ return true;
+ } else if (devinfo->gen <= 7) {
+ /* Ivy Bridge and early doesn't have any swizzling */
+ return isl_swizzle_is_identity(swizzle);
+ } else {
+ /* From the Sky Lake PRM Vol. 2d,
+ * RENDER_SURFACE_STATE::Shader Channel Select Red
+ *
+ * "For Render Target, Red, Green and Blue Shader Channel Selects
+ * MUST be such that only valid components can be swapped i.e. only
+ * change the order of components in the pixel. Any other values for
+ * these Shader Channel Select fields are not valid for Render
+ * Targets. This also means that there MUST not be multiple shader
+ * channels mapped to the same RT channel."
+ *
+ * From the Sky Lake PRM Vol. 2d,
+ * RENDER_SURFACE_STATE::Shader Channel Select Alpha
+ *
+ * "For Render Target, this field MUST be programmed to
+ * value = SCS_ALPHA."
+ */
+ return (swizzle.r == ISL_CHANNEL_SELECT_RED ||
+ swizzle.r == ISL_CHANNEL_SELECT_GREEN ||
+ swizzle.r == ISL_CHANNEL_SELECT_BLUE) &&
+ (swizzle.g == ISL_CHANNEL_SELECT_RED ||
+ swizzle.g == ISL_CHANNEL_SELECT_GREEN ||
+ swizzle.g == ISL_CHANNEL_SELECT_BLUE) &&
+ (swizzle.b == ISL_CHANNEL_SELECT_RED ||
+ swizzle.b == ISL_CHANNEL_SELECT_GREEN ||
+ swizzle.b == ISL_CHANNEL_SELECT_BLUE) &&
+ swizzle.r != swizzle.g &&
+ swizzle.r != swizzle.b &&
+ swizzle.g != swizzle.b &&
+ swizzle.a == ISL_CHANNEL_SELECT_ALPHA;
+ }
+}
+
+static enum isl_channel_select
+swizzle_select(enum isl_channel_select chan, struct isl_swizzle swizzle)
+{
+ switch (chan) {
+ case ISL_CHANNEL_SELECT_ZERO:
+ case ISL_CHANNEL_SELECT_ONE:
+ return chan;
+ case ISL_CHANNEL_SELECT_RED:
+ return swizzle.r;
+ case ISL_CHANNEL_SELECT_GREEN:
+ return swizzle.g;
+ case ISL_CHANNEL_SELECT_BLUE:
+ return swizzle.b;
+ case ISL_CHANNEL_SELECT_ALPHA:
+ return swizzle.a;
+ default:
+ unreachable("Invalid swizzle component");
+ }
+}
+
+/**
+ * Returns the single swizzle that is equivalent to applying the two given
+ * swizzles in sequence.
+ */
+struct isl_swizzle
+isl_swizzle_compose(struct isl_swizzle first, struct isl_swizzle second)
+{
+ return (struct isl_swizzle) {
+ .r = swizzle_select(first.r, second),
+ .g = swizzle_select(first.g, second),
+ .b = swizzle_select(first.b, second),
+ .a = swizzle_select(first.a, second),
+ };
+}
+
+/**
+ * Returns a swizzle that is the pseudo-inverse of this swizzle.
+ */
+struct isl_swizzle
+isl_swizzle_invert(struct isl_swizzle swizzle)
+{
+ /* Default to zero for channels which do not show up in the swizzle */
+ enum isl_channel_select chans[4] = {
+ ISL_CHANNEL_SELECT_ZERO,
+ ISL_CHANNEL_SELECT_ZERO,
+ ISL_CHANNEL_SELECT_ZERO,
+ ISL_CHANNEL_SELECT_ZERO,
+ };
+
+ /* We go in ABGR order so that, if there are any duplicates, the first one
+ * is taken if you look at it in RGBA order. This is what Haswell hardware
+ * does for render target swizzles.
+ */
+ if ((unsigned)(swizzle.a - ISL_CHANNEL_SELECT_RED) < 4)
+ chans[swizzle.a - ISL_CHANNEL_SELECT_RED] = ISL_CHANNEL_SELECT_ALPHA;
+ if ((unsigned)(swizzle.b - ISL_CHANNEL_SELECT_RED) < 4)
+ chans[swizzle.b - ISL_CHANNEL_SELECT_RED] = ISL_CHANNEL_SELECT_BLUE;
+ if ((unsigned)(swizzle.g - ISL_CHANNEL_SELECT_RED) < 4)
+ chans[swizzle.g - ISL_CHANNEL_SELECT_RED] = ISL_CHANNEL_SELECT_GREEN;
+ if ((unsigned)(swizzle.r - ISL_CHANNEL_SELECT_RED) < 4)
+ chans[swizzle.r - ISL_CHANNEL_SELECT_RED] = ISL_CHANNEL_SELECT_RED;
+
+ return (struct isl_swizzle) { chans[0], chans[1], chans[2], chans[3] };
+}
+
+/** Applies an inverse swizzle to a color value */
+union isl_color_value
+isl_color_value_swizzle_inv(union isl_color_value src,
+ struct isl_swizzle swizzle)
+{
+ union isl_color_value dst = { .u32 = { 0, } };
+
+ /* We assign colors in ABGR order so that the first one will be taken in
+ * RGBA precedence order. According to the PRM docs for shader channel
+ * select, this matches Haswell hardware behavior.
+ */
+ if ((unsigned)(swizzle.a - ISL_CHANNEL_SELECT_RED) < 4)
+ dst.u32[swizzle.a - ISL_CHANNEL_SELECT_RED] = src.u32[3];
+ if ((unsigned)(swizzle.b - ISL_CHANNEL_SELECT_RED) < 4)
+ dst.u32[swizzle.b - ISL_CHANNEL_SELECT_RED] = src.u32[2];
+ if ((unsigned)(swizzle.g - ISL_CHANNEL_SELECT_RED) < 4)
+ dst.u32[swizzle.g - ISL_CHANNEL_SELECT_RED] = src.u32[1];
+ if ((unsigned)(swizzle.r - ISL_CHANNEL_SELECT_RED) < 4)
+ dst.u32[swizzle.r - ISL_CHANNEL_SELECT_RED] = src.u32[0];
+
+ return dst;
+}
+
+uint8_t
+isl_format_get_aux_map_encoding(enum isl_format format)
+{
+ switch(format) {
+ case ISL_FORMAT_R32G32B32A32_FLOAT: return 0x11;
+ case ISL_FORMAT_R32G32B32X32_FLOAT: return 0x11;
+ case ISL_FORMAT_R32G32B32A32_SINT: return 0x12;
+ case ISL_FORMAT_R32G32B32A32_UINT: return 0x13;
+ case ISL_FORMAT_R16G16B16A16_UNORM: return 0x14;
+ case ISL_FORMAT_R16G16B16A16_SNORM: return 0x15;
+ case ISL_FORMAT_R16G16B16A16_SINT: return 0x16;
+ case ISL_FORMAT_R16G16B16A16_UINT: return 0x17;
+ case ISL_FORMAT_R16G16B16A16_FLOAT: return 0x10;
+ case ISL_FORMAT_R16G16B16X16_FLOAT: return 0x10;
+ case ISL_FORMAT_R32G32_FLOAT: return 0x11;
+ case ISL_FORMAT_R32G32_SINT: return 0x12;
+ case ISL_FORMAT_R32G32_UINT: return 0x13;
+ case ISL_FORMAT_B8G8R8A8_UNORM: return 0xA;
+ case ISL_FORMAT_B8G8R8X8_UNORM: return 0xA;
+ case ISL_FORMAT_B8G8R8A8_UNORM_SRGB: return 0xA;
+ case ISL_FORMAT_B8G8R8X8_UNORM_SRGB: return 0xA;
+ case ISL_FORMAT_R10G10B10A2_UNORM: return 0x18;
+ case ISL_FORMAT_R10G10B10A2_UNORM_SRGB: return 0x18;
+ case ISL_FORMAT_R10G10B10_FLOAT_A2_UNORM: return 0x19;
+ case ISL_FORMAT_R10G10B10A2_UINT: return 0x1A;
+ case ISL_FORMAT_R8G8B8A8_UNORM: return 0xA;
+ case ISL_FORMAT_R8G8B8A8_UNORM_SRGB: return 0xA;
+ case ISL_FORMAT_R8G8B8A8_SNORM: return 0x1B;
+ case ISL_FORMAT_R8G8B8A8_SINT: return 0x1C;
+ case ISL_FORMAT_R8G8B8A8_UINT: return 0x1D;
+ case ISL_FORMAT_R16G16_UNORM: return 0x14;
+ case ISL_FORMAT_R16G16_SNORM: return 0x15;
+ case ISL_FORMAT_R16G16_SINT: return 0x16;
+ case ISL_FORMAT_R16G16_UINT: return 0x17;
+ case ISL_FORMAT_R16G16_FLOAT: return 0x10;
+ case ISL_FORMAT_B10G10R10A2_UNORM: return 0x18;
+ case ISL_FORMAT_B10G10R10A2_UNORM_SRGB: return 0x18;
+ case ISL_FORMAT_R11G11B10_FLOAT: return 0x1E;
+ case ISL_FORMAT_R32_SINT: return 0x12;
+ case ISL_FORMAT_R32_UINT: return 0x13;
+ case ISL_FORMAT_R32_FLOAT: return 0x11;
+ case ISL_FORMAT_R24_UNORM_X8_TYPELESS: return 0x11;
+ case ISL_FORMAT_B5G6R5_UNORM: return 0xA;
+ case ISL_FORMAT_B5G6R5_UNORM_SRGB: return 0xA;
+ case ISL_FORMAT_B5G5R5A1_UNORM: return 0xA;
+ case ISL_FORMAT_B5G5R5A1_UNORM_SRGB: return 0xA;
+ case ISL_FORMAT_B4G4R4A4_UNORM: return 0xA;
+ case ISL_FORMAT_B4G4R4A4_UNORM_SRGB: return 0xA;
+ case ISL_FORMAT_R8G8_UNORM: return 0xA;
+ case ISL_FORMAT_R8G8_SNORM: return 0x1B;
+ case ISL_FORMAT_R8G8_SINT: return 0x1C;
+ case ISL_FORMAT_R8G8_UINT: return 0x1D;
+ case ISL_FORMAT_R16_UNORM: return 0x14;
+ case ISL_FORMAT_R16_SNORM: return 0x15;
+ case ISL_FORMAT_R16_SINT: return 0x16;
+ case ISL_FORMAT_R16_UINT: return 0x17;
+ case ISL_FORMAT_R16_FLOAT: return 0x10;
+ case ISL_FORMAT_B5G5R5X1_UNORM: return 0xA;
+ case ISL_FORMAT_B5G5R5X1_UNORM_SRGB: return 0xA;
+ case ISL_FORMAT_A1B5G5R5_UNORM: return 0xA;
+ case ISL_FORMAT_A4B4G4R4_UNORM: return 0xA;
+ case ISL_FORMAT_R8_UNORM: return 0xA;
+ case ISL_FORMAT_R8_SNORM: return 0x1B;
+ case ISL_FORMAT_R8_SINT: return 0x1C;
+ case ISL_FORMAT_R8_UINT: return 0x1D;
+ case ISL_FORMAT_A8_UNORM: return 0xA;
+ default:
+ unreachable("Unsupported aux-map format!");
+ return 0;
+ }
+}
projects / mesa.git / blobdiff