#define __gen_address_type uint64_t
#define __gen_user_data void
-static inline uint64_t
-__gen_combine_address(void *data, void *loc, uint64_t addr, uint32_t delta)
+static uint64_t
+__gen_combine_address(__attribute__((unused)) void *data,
+ __attribute__((unused)) void *loc, uint64_t addr,
+ uint32_t delta)
{
return addr + delta;
}
#include "isl_priv.h"
-#define __PASTE2(x, y) x ## y
-#define __PASTE(x, y) __PASTE2(x, y)
-#define isl_genX(x) __PASTE(isl_, genX(x))
-
#if GEN_GEN >= 8
static const uint8_t isl_to_gen_halign[] = {
[4] = HALIGN4,
[ISL_TILING_Y0] = YMAJOR,
[ISL_TILING_Yf] = YMAJOR,
[ISL_TILING_Ys] = YMAJOR,
+#if GEN_GEN <= 11
[ISL_TILING_W] = WMAJOR,
+#endif
};
#endif
};
#endif
-#if GEN_GEN >= 9
+#if GEN_GEN >= 12
+static const uint32_t isl_to_gen_aux_mode[] = {
+ [ISL_AUX_USAGE_NONE] = AUX_NONE,
+ [ISL_AUX_USAGE_MCS] = AUX_CCS_E,
+ [ISL_AUX_USAGE_CCS_E] = AUX_CCS_E,
+ [ISL_AUX_USAGE_HIZ_CCS_WT] = AUX_CCS_E,
+ [ISL_AUX_USAGE_MCS_CCS] = AUX_MCS_LCE,
+ [ISL_AUX_USAGE_STC_CCS] = AUX_CCS_E,
+};
+#elif GEN_GEN >= 9
static const uint32_t isl_to_gen_aux_mode[] = {
[ISL_AUX_USAGE_NONE] = AUX_NONE,
[ISL_AUX_USAGE_HIZ] = AUX_HIZ,
assert(!(usage & ISL_SURF_USAGE_CUBE_BIT));
return SURFTYPE_1D;
case ISL_SURF_DIM_2D:
- if (usage & ISL_SURF_USAGE_STORAGE_BIT) {
- /* Storage images are always plain 2-D, not cube */
- return SURFTYPE_2D;
- } else if (usage & ISL_SURF_USAGE_CUBE_BIT) {
+ if ((usage & ISL_SURF_USAGE_CUBE_BIT) &&
+ (usage & ISL_SURF_USAGE_TEXTURE_BIT)) {
+ /* We need SURFTYPE_CUBE to make cube sampling work */
return SURFTYPE_CUBE;
} else {
+ /* Everything else (render and storage) treat cubes as plain
+ * 2D array textures
+ */
return SURFTYPE_2D;
}
case ISL_SURF_DIM_3D:
* hardware. Note that this does NOT give you the actual hardware enum values
* but an index into the isl_to_gen_[hv]align arrays above.
*/
-static inline struct isl_extent3d
+UNUSED static struct isl_extent3d
get_image_alignment(const struct isl_surf *surf)
{
if (GEN_GEN >= 9) {
if (info->surf->dim == ISL_SURF_DIM_1D)
assert(!isl_format_is_compressed(info->view->format));
+ if (isl_format_is_compressed(info->surf->format)) {
+ /* You're not allowed to make a view of a compressed format with any
+ * format other than the surface format. None of the userspace APIs
+ * allow for this directly and doing so would mess up a number of
+ * surface parameters such as Width, Height, and alignments. Ideally,
+ * we'd like to assert that the two formats match. However, we have an
+ * S3TC workaround that requires us to do reinterpretation. So assert
+ * that they're at least the same bpb and block size.
+ */
+ ASSERTED const struct isl_format_layout *surf_fmtl =
+ isl_format_get_layout(info->surf->format);
+ ASSERTED const struct isl_format_layout *view_fmtl =
+ isl_format_get_layout(info->surf->format);
+ assert(surf_fmtl->bpb == view_fmtl->bpb);
+ assert(surf_fmtl->bw == view_fmtl->bw);
+ assert(surf_fmtl->bh == view_fmtl->bh);
+ }
+
s.SurfaceFormat = info->view->format;
+#if GEN_GEN >= 12
+ s.DepthStencilResource =
+ isl_surf_usage_is_depth_or_stencil(info->surf->usage);
+#endif
+
+#if GEN_GEN <= 5
+ s.ColorBufferComponentWriteDisables = info->write_disables;
+#else
+ assert(info->write_disables == 0);
+#endif
+
#if GEN_IS_HASWELL
- s.IntegerSurfaceFormat = isl_format_has_int_channel(s.SurfaceFormat);
+ s.IntegerSurfaceFormat =
+ isl_format_has_int_channel((enum isl_format) s.SurfaceFormat);
#endif
+ assert(info->surf->logical_level0_px.width > 0 &&
+ info->surf->logical_level0_px.height > 0);
+
s.Width = info->surf->logical_level0_px.width - 1;
s.Height = info->surf->logical_level0_px.height - 1;
unreachable("bad SurfaceType");
}
-#if GEN_GEN >= 7
+#if GEN_GEN >= 12
+ /* GEN:BUG:1806565034: Only set SurfaceArray if arrayed surface is > 1. */
+ s.SurfaceArray = info->surf->dim != ISL_SURF_DIM_3D &&
+ info->view->array_len > 1;
+#elif GEN_GEN >= 7
s.SurfaceArray = info->surf->dim != ISL_SURF_DIM_3D;
#endif
/* For gen9 1-D textures, surface pitch is ignored */
s.SurfacePitch = 0;
} else {
- s.SurfacePitch = info->surf->row_pitch - 1;
+ s.SurfacePitch = info->surf->row_pitch_B - 1;
}
#if GEN_GEN >= 8
#endif
#if GEN_GEN >= 8
+ assert(GEN_GEN < 12 || info->surf->tiling != ISL_TILING_W);
s.TileMode = isl_to_gen_tiling[info->surf->tiling];
#else
s.TiledSurface = info->surf->tiling != ISL_TILING_LINEAR,
s.RenderCacheReadWriteMode = 0;
#endif
- if (info->view->usage & ISL_SURF_USAGE_CUBE_BIT) {
-#if GEN_GEN >= 8
- s.CubeFaceEnablePositiveZ = 1;
- s.CubeFaceEnableNegativeZ = 1;
- s.CubeFaceEnablePositiveY = 1;
- s.CubeFaceEnableNegativeY = 1;
- s.CubeFaceEnablePositiveX = 1;
- s.CubeFaceEnableNegativeX = 1;
-#else
- s.CubeFaceEnables = 0x3f;
+#if GEN_GEN >= 11
+ /* We've seen dEQP failures when enabling this bit with UINT formats,
+ * which particularly affects blorp_copy() operations. It shouldn't
+ * have any effect on UINT textures anyway, so disable it for them.
+ */
+ s.EnableUnormPathInColorPipe =
+ !isl_format_has_int_channel(info->view->format);
#endif
- }
+
+ s.CubeFaceEnablePositiveZ = 1;
+ s.CubeFaceEnableNegativeZ = 1;
+ s.CubeFaceEnablePositiveY = 1;
+ s.CubeFaceEnableNegativeY = 1;
+ s.CubeFaceEnablePositiveX = 1;
+ s.CubeFaceEnableNegativeX = 1;
#if GEN_GEN >= 6
s.NumberofMultisamples = ffs(info->surf->samples) - 1;
#endif
#if (GEN_GEN >= 8 || GEN_IS_HASWELL)
- s.ShaderChannelSelectRed = info->view->swizzle.r;
- s.ShaderChannelSelectGreen = info->view->swizzle.g;
- s.ShaderChannelSelectBlue = info->view->swizzle.b;
- s.ShaderChannelSelectAlpha = info->view->swizzle.a;
+ if (info->view->usage & ISL_SURF_USAGE_RENDER_TARGET_BIT)
+ assert(isl_swizzle_supports_rendering(dev->info, info->view->swizzle));
+
+ s.ShaderChannelSelectRed = (enum GENX(ShaderChannelSelect)) info->view->swizzle.r;
+ s.ShaderChannelSelectGreen = (enum GENX(ShaderChannelSelect)) info->view->swizzle.g;
+ s.ShaderChannelSelectBlue = (enum GENX(ShaderChannelSelect)) info->view->swizzle.b;
+ s.ShaderChannelSelectAlpha = (enum GENX(ShaderChannelSelect)) info->view->swizzle.a;
+#else
+ assert(isl_swizzle_is_identity(info->view->swizzle));
#endif
s.SurfaceBaseAddress = info->address;
#endif
#if GEN_GEN >= 7
- if (info->aux_surf && info->aux_usage != ISL_AUX_USAGE_NONE) {
+ if (info->aux_usage != ISL_AUX_USAGE_NONE) {
+ /* Check valid aux usages per-gen */
+ if (GEN_GEN >= 12) {
+ assert(info->aux_usage == ISL_AUX_USAGE_MCS ||
+ info->aux_usage == ISL_AUX_USAGE_CCS_E ||
+ info->aux_usage == ISL_AUX_USAGE_HIZ_CCS_WT ||
+ info->aux_usage == ISL_AUX_USAGE_MCS_CCS ||
+ info->aux_usage == ISL_AUX_USAGE_STC_CCS);
+ } else if (GEN_GEN >= 9) {
+ assert(info->aux_usage == ISL_AUX_USAGE_HIZ ||
+ info->aux_usage == ISL_AUX_USAGE_MCS ||
+ info->aux_usage == ISL_AUX_USAGE_CCS_D ||
+ info->aux_usage == ISL_AUX_USAGE_CCS_E);
+ } else if (GEN_GEN >= 8) {
+ assert(info->aux_usage == ISL_AUX_USAGE_HIZ ||
+ info->aux_usage == ISL_AUX_USAGE_MCS ||
+ info->aux_usage == ISL_AUX_USAGE_CCS_D);
+ } else if (GEN_GEN >= 7) {
+ assert(info->aux_usage == ISL_AUX_USAGE_MCS ||
+ info->aux_usage == ISL_AUX_USAGE_CCS_D);
+ }
+
+ /* The docs don't appear to say anything whatsoever about compression
+ * and the data port. Testing seems to indicate that the data port
+ * completely ignores the AuxiliarySurfaceMode field.
+ */
+ assert(!(info->view->usage & ISL_SURF_USAGE_STORAGE_BIT));
+
+ if (isl_surf_usage_is_depth(info->surf->usage))
+ assert(isl_aux_usage_has_hiz(info->aux_usage));
+
+ if (isl_surf_usage_is_stencil(info->surf->usage))
+ assert(info->aux_usage == ISL_AUX_USAGE_STC_CCS);
+
+ if (isl_aux_usage_has_hiz(info->aux_usage)) {
+ /* For Gen8-10, there are some restrictions around sampling from HiZ.
+ * The Skylake PRM docs for RENDER_SURFACE_STATE::AuxiliarySurfaceMode
+ * say:
+ *
+ * "If this field is set to AUX_HIZ, Number of Multisamples must
+ * be MULTISAMPLECOUNT_1, and Surface Type cannot be SURFTYPE_3D."
+ *
+ * On Gen12, the docs are a bit less obvious but the restriction is
+ * the same. The limitation isn't called out explicitly but the docs
+ * for the CCS_E value of RENDER_SURFACE_STATE::AuxiliarySurfaceMode
+ * say:
+ *
+ * "If Number of multisamples > 1, programming this value means
+ * MSAA compression is enabled for that surface. Auxillary surface
+ * is MSC with tile y."
+ *
+ * Since this interpretation ignores whether the surface is
+ * depth/stencil or not and since multisampled depth buffers use
+ * ISL_MSAA_LAYOUT_INTERLEAVED which is incompatible with MCS
+ * compression, this means that we can't even specify MSAA depth CCS
+ * in RENDER_SURFACE_STATE::AuxiliarySurfaceMode.
+ */
+ assert(info->surf->samples == 1);
+
+ /* The dimension must not be 3D */
+ assert(info->surf->dim != ISL_SURF_DIM_3D);
+
+ /* The format must be one of the following: */
+ switch (info->view->format) {
+ case ISL_FORMAT_R32_FLOAT:
+ case ISL_FORMAT_R24_UNORM_X8_TYPELESS:
+ case ISL_FORMAT_R16_UNORM:
+ break;
+ default:
+ assert(!"Incompatible HiZ Sampling format");
+ break;
+ }
+ }
+
+#if GEN_GEN >= 8
+ s.AuxiliarySurfaceMode = isl_to_gen_aux_mode[info->aux_usage];
+#else
+ s.MCSEnable = true;
+#endif
+ }
+
+ /* The auxiliary buffer info is filled when it's useable by the HW.
+ *
+ * Starting with Gen12, the only form of compression that can be used
+ * with RENDER_SURFACE_STATE which requires an aux surface is MCS.
+ * HiZ still requires a surface but the HiZ surface can only be
+ * accessed through 3DSTATE_HIER_DEPTH_BUFFER.
+ *
+ * On all earlier hardware, an aux surface is required for all forms
+ * of compression.
+ */
+ if ((GEN_GEN < 12 && info->aux_usage != ISL_AUX_USAGE_NONE) ||
+ (GEN_GEN >= 12 && isl_aux_usage_has_mcs(info->aux_usage))) {
+
+ assert(info->aux_surf != NULL);
+
struct isl_tile_info tile_info;
- isl_surf_get_tile_info(dev, info->aux_surf, &tile_info);
+ isl_surf_get_tile_info(info->aux_surf, &tile_info);
uint32_t pitch_in_tiles =
- info->aux_surf->row_pitch / tile_info.phys_extent_B.width;
+ info->aux_surf->row_pitch_B / tile_info.phys_extent_B.width;
-#if GEN_GEN >= 8
- assert(GEN_GEN >= 9 || info->aux_usage != ISL_AUX_USAGE_CCS_E);
+ s.AuxiliarySurfaceBaseAddress = info->aux_address;
s.AuxiliarySurfacePitch = pitch_in_tiles - 1;
+
+#if GEN_GEN >= 8
/* Auxiliary surfaces in ISL have compressed formats but the hardware
* doesn't expect our definition of the compression, it expects qpitch
* in units of samples on the main surface.
*/
s.AuxiliarySurfaceQPitch =
isl_surf_get_array_pitch_sa_rows(info->aux_surf) >> 2;
- s.AuxiliarySurfaceBaseAddress = info->aux_address;
- s.AuxiliarySurfaceMode = isl_to_gen_aux_mode[info->aux_usage];
-#else
- assert(info->aux_usage == ISL_AUX_USAGE_MCS ||
- info->aux_usage == ISL_AUX_USAGE_CCS_D);
- s.MCSBaseAddress = info->aux_address,
- s.MCSSurfacePitch = pitch_in_tiles - 1;
- s.MCSEnable = true;
#endif
}
#endif
-#if GEN_GEN >= 8
+#if GEN_GEN >= 8 && GEN_GEN < 11
/* From the CHV PRM, Volume 2d, page 321 (RENDER_SURFACE_STATE dword 0
* bit 9 "Sampler L2 Bypass Mode Disable" Programming Notes):
*
s.SamplerL2BypassModeDisable = true;
break;
default:
+ /* From the SKL PRM, Programming Note under Sampler Output Channel
+ * Mapping:
+ *
+ * If a surface has an associated HiZ Auxilliary surface, the
+ * Sampler L2 Bypass Mode Disable field in the RENDER_SURFACE_STATE
+ * must be set.
+ */
+ if (GEN_GEN >= 9 && info->aux_usage == ISL_AUX_USAGE_HIZ)
+ s.SamplerL2BypassModeDisable = true;
break;
}
}
#endif
if (info->aux_usage != ISL_AUX_USAGE_NONE) {
-#if GEN_GEN >= 9
- s.RedClearColor = info->clear_color.u32[0];
- s.GreenClearColor = info->clear_color.u32[1];
- s.BlueClearColor = info->clear_color.u32[2];
- s.AlphaClearColor = info->clear_color.u32[3];
+ if (info->use_clear_address) {
+#if GEN_GEN >= 10
+ s.ClearValueAddressEnable = true;
+ s.ClearValueAddress = info->clear_address;
+#else
+ unreachable("Gen9 and earlier do not support indirect clear colors");
+#endif
+ }
+
+#if GEN_GEN == 11
+ /*
+ * From BXML > GT > Shared Functions > vol5c Shared Functions >
+ * [Structure] RENDER_SURFACE_STATE [BDW+] > ClearColorConversionEnable:
+ *
+ * Project: Gen11
+ *
+ * "Enables Pixel backend hw to convert clear values into native format
+ * and write back to clear address, so that display and sampler can use
+ * the converted value for resolving fast cleared RTs."
+ *
+ * Summary:
+ * Clear color conversion must be enabled if the clear color is stored
+ * indirectly and fast color clears are enabled.
+ */
+ if (info->use_clear_address) {
+ s.ClearColorConversionEnable = true;
+ }
+#endif
+
+#if GEN_GEN >= 12
+ assert(info->use_clear_address);
+#elif GEN_GEN >= 9
+ if (!info->use_clear_address) {
+ s.RedClearColor = info->clear_color.u32[0];
+ s.GreenClearColor = info->clear_color.u32[1];
+ s.BlueClearColor = info->clear_color.u32[2];
+ s.AlphaClearColor = info->clear_color.u32[3];
+ }
#elif GEN_GEN >= 7
/* Prior to Sky Lake, we only have one bit for the clear color which
* gives us 0 or 1 in whatever the surface's format happens to be.
}
void
-isl_genX(buffer_fill_state_s)(void *state,
+isl_genX(buffer_fill_state_s)(const struct isl_device *dev, void *state,
const struct isl_buffer_fill_state_info *restrict info)
{
- uint32_t num_elements = info->size / info->stride;
+ uint64_t buffer_size = info->size_B;
+
+ /* Uniform and Storage buffers need to have surface size not less that the
+ * aligned 32-bit size of the buffer. To calculate the array lenght on
+ * unsized arrays in StorageBuffer the last 2 bits store the padding size
+ * added to the surface, so we can calculate latter the original buffer
+ * size to know the number of elements.
+ *
+ * surface_size = isl_align(buffer_size, 4) +
+ * (isl_align(buffer_size) - buffer_size)
+ *
+ * buffer_size = (surface_size & ~3) - (surface_size & 3)
+ */
+ if (info->format == ISL_FORMAT_RAW ||
+ info->stride_B < isl_format_get_layout(info->format)->bpb / 8) {
+ assert(info->stride_B == 1);
+ uint64_t aligned_size = isl_align(buffer_size, 4);
+ buffer_size = aligned_size + (aligned_size - buffer_size);
+ }
+
+ uint32_t num_elements = buffer_size / info->stride_B;
if (GEN_GEN >= 7) {
/* From the IVB PRM, SURFACE_STATE::Height,
*/
if (info->format == ISL_FORMAT_RAW) {
assert(num_elements <= (1ull << 30));
- assert((num_elements & 3) == 0);
+ assert(num_elements > 0);
} else {
assert(num_elements <= (1ull << 27));
}
s.Depth = ((num_elements - 1) >> 20) & 0x7f;
#endif
- s.SurfacePitch = info->stride - 1;
+ if (GEN_GEN == 12 && dev->info->revision == 0) {
+ /* TGL-LP A0 has a HW bug (fixed in later HW) which causes buffer
+ * textures with very close base addresses (delta < 64B) to corrupt each
+ * other. We can sort-of work around this by making small buffer
+ * textures 1D textures instead. This doesn't fix the problem for large
+ * buffer textures but the liklihood of large, overlapping, and very
+ * close buffer textures is fairly low and the point is to hack around
+ * the bug so we can run apps and tests.
+ */
+ if (info->format != ISL_FORMAT_RAW &&
+ info->stride_B == isl_format_get_layout(info->format)->bpb / 8 &&
+ num_elements <= (1 << 14)) {
+ s.SurfaceType = SURFTYPE_1D;
+ s.Width = num_elements - 1;
+ s.Height = 0;
+ s.Depth = 0;
+ }
+ }
+
+ s.SurfacePitch = info->stride_B - 1;
#if GEN_GEN >= 6
s.NumberofMultisamples = MULTISAMPLECOUNT_1;
#endif
#if (GEN_GEN >= 8 || GEN_IS_HASWELL)
- s.ShaderChannelSelectRed = SCS_RED;
- s.ShaderChannelSelectGreen = SCS_GREEN;
- s.ShaderChannelSelectBlue = SCS_BLUE;
- s.ShaderChannelSelectAlpha = SCS_ALPHA;
+ s.ShaderChannelSelectRed = (enum GENX(ShaderChannelSelect)) info->swizzle.r;
+ s.ShaderChannelSelectGreen = (enum GENX(ShaderChannelSelect)) info->swizzle.g;
+ s.ShaderChannelSelectBlue = (enum GENX(ShaderChannelSelect)) info->swizzle.b;
+ s.ShaderChannelSelectAlpha = (enum GENX(ShaderChannelSelect)) info->swizzle.a;
#endif
GENX(RENDER_SURFACE_STATE_pack)(NULL, state, &s);
}
+
+void
+isl_genX(null_fill_state)(void *state, struct isl_extent3d size)
+{
+ struct GENX(RENDER_SURFACE_STATE) s = {
+ .SurfaceType = SURFTYPE_NULL,
+ /* We previously had this format set to B8G8R8A8_UNORM but ran into
+ * hangs on IVB. R32_UINT seems to work for everybody.
+ *
+ * https://gitlab.freedesktop.org/mesa/mesa/issues/1872
+ */
+ .SurfaceFormat = ISL_FORMAT_R32_UINT,
+#if GEN_GEN >= 7
+ .SurfaceArray = size.depth > 1,
+#endif
+#if GEN_GEN >= 8
+ .TileMode = YMAJOR,
+#else
+ .TiledSurface = true,
+ .TileWalk = TILEWALK_YMAJOR,
+#endif
+#if GEN_GEN == 7
+ /* According to PRMs: "Volume 4 Part 1: Subsystem and Cores – Shared
+ * Functions"
+ *
+ * RENDER_SURFACE_STATE::Surface Vertical Alignment
+ *
+ * "This field must be set to VALIGN_4 for all tiled Y Render Target
+ * surfaces."
+ *
+ * Affect IVB, HSW.
+ */
+ .SurfaceVerticalAlignment = VALIGN_4,
+#endif
+ .Width = size.width - 1,
+ .Height = size.height - 1,
+ .Depth = size.depth - 1,
+ .RenderTargetViewExtent = size.depth - 1,
+#if GEN_GEN <= 5
+ .ColorBufferComponentWriteDisables = 0xf,
+#endif
+ };
+ GENX(RENDER_SURFACE_STATE_pack)(NULL, state, &s);
+}