Mesa (master): turnip: implement VK_EXT_custom_border_color
GitLab Mirror
gitlab-mirror at kemper.freedesktop.org
Thu Aug 13 17:11:10 UTC 2020
Module: Mesa
Branch: master
Commit: 0c97e601a422532ea673e842683982b1b4012e6d
URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=0c97e601a422532ea673e842683982b1b4012e6d
Author: Jonathan Marek <jonathan at marek.ca>
Date: Mon Jul 27 13:20:04 2020 -0400
turnip: implement VK_EXT_custom_border_color
Signed-off-by: Jonathan Marek <jonathan at marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6093>
---
src/freedreno/registers/adreno/a6xx.xml | 2 +-
src/freedreno/vulkan/tu_cmd_buffer.c | 8 +-
src/freedreno/vulkan/tu_device.c | 142 ++++++++++++++------------
src/freedreno/vulkan/tu_extensions.py | 1 +
src/freedreno/vulkan/tu_private.h | 12 ++-
src/freedreno/vulkan/tu_util.h | 46 +++++++++
src/gallium/drivers/freedreno/a6xx/fd6_emit.c | 2 +-
7 files changed, 137 insertions(+), 76 deletions(-)
diff --git a/src/freedreno/registers/adreno/a6xx.xml b/src/freedreno/registers/adreno/a6xx.xml
index 15314fbb736..8c5d61c5487 100644
--- a/src/freedreno/registers/adreno/a6xx.xml
+++ b/src/freedreno/registers/adreno/a6xx.xml
@@ -3689,7 +3689,7 @@ to upconvert to 32b float internally?
<reg32 offset="2" name="2">
<bitfield name="REDUCTION_MODE" low="0" high="1" type="a6xx_reduction_mode"/>
<bitfield name="CHROMA_LINEAR" pos="5" type="boolean"/>
- <bitfield name="BCOLOR_OFFSET" low="7" high="31" shr="7"/>
+ <bitfield name="BCOLOR" low="7" high="31"/>
</reg32>
<reg32 offset="3" name="3"/>
</domain>
diff --git a/src/freedreno/vulkan/tu_cmd_buffer.c b/src/freedreno/vulkan/tu_cmd_buffer.c
index 36d0e462edc..d647cb13f34 100644
--- a/src/freedreno/vulkan/tu_cmd_buffer.c
+++ b/src/freedreno/vulkan/tu_cmd_buffer.c
@@ -902,10 +902,10 @@ tu6_init_hw(struct tu_cmd_buffer *cmd, struct tu_cs *cs)
tu_cs_emit_regs(cs,
A6XX_SP_TP_BORDER_COLOR_BASE_ADDR(.bo = &dev->global_bo,
- .bo_offset = gb_offset(border_color)));
+ .bo_offset = gb_offset(bcolor_builtin)));
tu_cs_emit_regs(cs,
A6XX_SP_PS_TP_BORDER_COLOR_BASE_ADDR(.bo = &dev->global_bo,
- .bo_offset = gb_offset(border_color)));
+ .bo_offset = gb_offset(bcolor_builtin)));
/* VSC buffers:
* use vsc pitches from the largest values used so far with this device
@@ -914,7 +914,7 @@ tu6_init_hw(struct tu_cmd_buffer *cmd, struct tu_cs *cs)
*
* if overflow is detected, the stream size is increased by 2x
*/
- mtx_lock(&dev->vsc_pitch_mtx);
+ mtx_lock(&dev->mutex);
struct tu6_global *global = dev->global_bo.map;
@@ -930,7 +930,7 @@ tu6_init_hw(struct tu_cmd_buffer *cmd, struct tu_cs *cs)
cmd->vsc_prim_strm_pitch = dev->vsc_prim_strm_pitch;
cmd->vsc_draw_strm_pitch = dev->vsc_draw_strm_pitch;
- mtx_unlock(&dev->vsc_pitch_mtx);
+ mtx_unlock(&dev->mutex);
struct tu_bo *vsc_bo;
uint32_t size0 = cmd->vsc_prim_strm_pitch * MAX_VSC_PIPES +
diff --git a/src/freedreno/vulkan/tu_device.c b/src/freedreno/vulkan/tu_device.c
index 9dd95ce62bb..0ab94e2bf6e 100644
--- a/src/freedreno/vulkan/tu_device.c
+++ b/src/freedreno/vulkan/tu_device.c
@@ -603,6 +603,12 @@ tu_GetPhysicalDeviceFeatures2(VkPhysicalDevice physicalDevice,
features->formatA4B4G4R4 = true;
break;
}
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CUSTOM_BORDER_COLOR_FEATURES_EXT: {
+ VkPhysicalDeviceCustomBorderColorFeaturesEXT *features = (void *) ext;
+ features->customBorderColors = true;
+ features->customBorderColorWithoutFormat = true;
+ break;
+ }
default:
break;
}
@@ -851,6 +857,11 @@ tu_GetPhysicalDeviceProperties2(VkPhysicalDevice physicalDevice,
props->maxVertexAttribDivisor = UINT32_MAX;
break;
}
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CUSTOM_BORDER_COLOR_PROPERTIES_EXT: {
+ VkPhysicalDeviceCustomBorderColorPropertiesEXT *props = (void *)ext;
+ props->maxCustomBorderColorSamplers = TU_BORDER_COLOR_COUNT;
+ break;
+ }
default:
break;
}
@@ -976,60 +987,6 @@ tu_get_device_extension_index(const char *name)
return -1;
}
-struct PACKED bcolor_entry {
- uint32_t fp32[4];
- uint16_t ui16[4];
- int16_t si16[4];
- uint16_t fp16[4];
- uint16_t rgb565;
- uint16_t rgb5a1;
- uint16_t rgba4;
- uint8_t __pad0[2];
- uint8_t ui8[4];
- int8_t si8[4];
- uint32_t rgb10a2;
- uint32_t z24; /* also s8? */
- uint16_t srgb[4]; /* appears to duplicate fp16[], but clamped, used for srgb */
- uint8_t __pad1[56];
-} border_color[] = {
- [VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK] = {},
- [VK_BORDER_COLOR_INT_TRANSPARENT_BLACK] = {},
- [VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK] = {
- .fp32[3] = 0x3f800000,
- .ui16[3] = 0xffff,
- .si16[3] = 0x7fff,
- .fp16[3] = 0x3c00,
- .rgb5a1 = 0x8000,
- .rgba4 = 0xf000,
- .ui8[3] = 0xff,
- .si8[3] = 0x7f,
- .rgb10a2 = 0xc0000000,
- .srgb[3] = 0x3c00,
- },
- [VK_BORDER_COLOR_INT_OPAQUE_BLACK] = {
- .fp32[3] = 1,
- .fp16[3] = 1,
- },
- [VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE] = {
- .fp32[0 ... 3] = 0x3f800000,
- .ui16[0 ... 3] = 0xffff,
- .si16[0 ... 3] = 0x7fff,
- .fp16[0 ... 3] = 0x3c00,
- .rgb565 = 0xffff,
- .rgb5a1 = 0xffff,
- .rgba4 = 0xffff,
- .ui8[0 ... 3] = 0xff,
- .si8[0 ... 3] = 0x7f,
- .rgb10a2 = 0xffffffff,
- .z24 = 0xffffff,
- .srgb[0 ... 3] = 0x3c00,
- },
- [VK_BORDER_COLOR_INT_OPAQUE_WHITE] = {
- .fp32[0 ... 3] = 1,
- .fp16[0 ... 3] = 1,
- },
-};
-
VkResult
tu_CreateDevice(VkPhysicalDevice physicalDevice,
const VkDeviceCreateInfo *pCreateInfo,
@@ -1039,6 +996,7 @@ tu_CreateDevice(VkPhysicalDevice physicalDevice,
TU_FROM_HANDLE(tu_physical_device, physical_device, physicalDevice);
VkResult result;
struct tu_device *device;
+ bool custom_border_colors = false;
/* Check enabled features */
if (pCreateInfo->pEnabledFeatures) {
@@ -1055,6 +1013,18 @@ tu_CreateDevice(VkPhysicalDevice physicalDevice,
}
}
+ vk_foreach_struct_const(ext, pCreateInfo->pNext) {
+ switch (ext->sType) {
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CUSTOM_BORDER_COLOR_FEATURES_EXT: {
+ const VkPhysicalDeviceCustomBorderColorFeaturesEXT *border_color_features = (const void *)ext;
+ custom_border_colors = border_color_features->customBorderColors;
+ break;
+ }
+ default:
+ break;
+ }
+ }
+
device = vk_zalloc2(&physical_device->instance->alloc, pAllocator,
sizeof(*device), 8, VK_SYSTEM_ALLOCATION_SCOPE_DEVICE);
if (!device)
@@ -1113,8 +1083,11 @@ tu_CreateDevice(VkPhysicalDevice physicalDevice,
device->vsc_draw_strm_pitch = 0x1000 + VSC_PAD;
device->vsc_prim_strm_pitch = 0x4000 + VSC_PAD;
- STATIC_ASSERT(sizeof(border_color) == sizeof(((struct tu6_global*) 0)->border_color));
- result = tu_bo_init_new(device, &device->global_bo, sizeof(struct tu6_global));
+ uint32_t global_size = sizeof(struct tu6_global);
+ if (custom_border_colors)
+ global_size += TU_BORDER_COLOR_COUNT * sizeof(struct bcolor_entry);
+
+ result = tu_bo_init_new(device, &device->global_bo, global_size);
if (result != VK_SUCCESS)
goto fail_global_bo;
@@ -1123,9 +1096,23 @@ tu_CreateDevice(VkPhysicalDevice physicalDevice,
goto fail_global_bo_map;
struct tu6_global *global = device->global_bo.map;
- memcpy(global->border_color, border_color, sizeof(border_color));
+ tu_init_clear_blit_shaders(device->global_bo.map);
global->predicate = 0;
- tu_init_clear_blit_shaders(global);
+ tu6_pack_border_color(&global->bcolor_builtin[VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK],
+ &(VkClearColorValue) {}, false);
+ tu6_pack_border_color(&global->bcolor_builtin[VK_BORDER_COLOR_INT_TRANSPARENT_BLACK],
+ &(VkClearColorValue) {}, true);
+ tu6_pack_border_color(&global->bcolor_builtin[VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK],
+ &(VkClearColorValue) { .float32[3] = 1.0f }, false);
+ tu6_pack_border_color(&global->bcolor_builtin[VK_BORDER_COLOR_INT_OPAQUE_BLACK],
+ &(VkClearColorValue) { .int32[3] = 1 }, true);
+ tu6_pack_border_color(&global->bcolor_builtin[VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE],
+ &(VkClearColorValue) { .float32[0 ... 3] = 1.0f }, false);
+ tu6_pack_border_color(&global->bcolor_builtin[VK_BORDER_COLOR_INT_OPAQUE_WHITE],
+ &(VkClearColorValue) { .int32[0 ... 3] = 1 }, true);
+
+ /* initialize to ones so ffs can be used to find unused slots */
+ BITSET_ONES(device->custom_border_color);
VkPipelineCacheCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_PIPELINE_CACHE_CREATE_INFO;
@@ -1144,7 +1131,7 @@ tu_CreateDevice(VkPhysicalDevice physicalDevice,
for (unsigned i = 0; i < ARRAY_SIZE(device->scratch_bos); i++)
mtx_init(&device->scratch_bos[i].construct_mtx, mtx_plain);
- mtx_init(&device->vsc_pitch_mtx, mtx_plain);
+ mtx_init(&device->mutex, mtx_plain);
*pDevice = tu_device_to_handle(device);
return VK_SUCCESS;
@@ -1901,6 +1888,24 @@ tu_init_sampler(struct tu_device *device,
vk_find_struct_const(pCreateInfo->pNext, SAMPLER_REDUCTION_MODE_CREATE_INFO);
const struct VkSamplerYcbcrConversionInfo *ycbcr_conversion =
vk_find_struct_const(pCreateInfo->pNext, SAMPLER_YCBCR_CONVERSION_INFO);
+ const VkSamplerCustomBorderColorCreateInfoEXT *custom_border_color =
+ vk_find_struct_const(pCreateInfo->pNext, SAMPLER_CUSTOM_BORDER_COLOR_CREATE_INFO_EXT);
+ /* for non-custom border colors, the VK enum is translated directly to an offset in
+ * the border color buffer. custom border colors are located immediately after the
+ * builtin colors, and thus an offset of TU_BORDER_COLOR_BUILTIN is added.
+ */
+ uint32_t border_color = (unsigned) pCreateInfo->borderColor;
+ if (pCreateInfo->borderColor == VK_BORDER_COLOR_FLOAT_CUSTOM_EXT ||
+ pCreateInfo->borderColor == VK_BORDER_COLOR_INT_CUSTOM_EXT) {
+ mtx_lock(&device->mutex);
+ border_color = BITSET_FFS(device->custom_border_color);
+ BITSET_CLEAR(device->custom_border_color, border_color);
+ mtx_unlock(&device->mutex);
+ tu6_pack_border_color(device->global_bo.map + gb_offset(bcolor[border_color]),
+ &custom_border_color->customBorderColor,
+ pCreateInfo->borderColor == VK_BORDER_COLOR_INT_CUSTOM_EXT);
+ border_color += TU_BORDER_COLOR_BUILTIN;
+ }
unsigned aniso = pCreateInfo->anisotropyEnable ?
util_last_bit(MIN2((uint32_t)pCreateInfo->maxAnisotropy >> 1, 8)) : 0;
@@ -1924,13 +1929,7 @@ tu_init_sampler(struct tu_device *device,
A6XX_TEX_SAMP_1_MAX_LOD(max_lod) |
COND(pCreateInfo->compareEnable,
A6XX_TEX_SAMP_1_COMPARE_FUNC(tu6_compare_func(pCreateInfo->compareOp)));
- /* This is an offset into the border_color BO, which we fill with all the
- * possible Vulkan border colors in the correct order, so we can just use
- * the Vulkan enum with no translation necessary.
- */
- sampler->descriptor[2] =
- A6XX_TEX_SAMP_2_BCOLOR_OFFSET((unsigned) pCreateInfo->borderColor *
- sizeof(struct bcolor_entry));
+ sampler->descriptor[2] = A6XX_TEX_SAMP_2_BCOLOR(border_color);
sampler->descriptor[3] = 0;
if (reduction) {
@@ -1980,10 +1979,21 @@ tu_DestroySampler(VkDevice _device,
{
TU_FROM_HANDLE(tu_device, device, _device);
TU_FROM_HANDLE(tu_sampler, sampler, _sampler);
+ uint32_t border_color;
if (!sampler)
return;
+ border_color = (sampler->descriptor[2] & A6XX_TEX_SAMP_2_BCOLOR__MASK) >> A6XX_TEX_SAMP_2_BCOLOR__SHIFT;
+ if (border_color >= TU_BORDER_COLOR_BUILTIN) {
+ border_color -= TU_BORDER_COLOR_BUILTIN;
+ /* if the sampler had a custom border color, free it. TODO: no lock */
+ mtx_lock(&device->mutex);
+ assert(!BITSET_TEST(device->custom_border_color, border_color));
+ BITSET_SET(device->custom_border_color, border_color);
+ mtx_unlock(&device->mutex);
+ }
+
vk_object_free(&device->vk, pAllocator, sampler);
}
diff --git a/src/freedreno/vulkan/tu_extensions.py b/src/freedreno/vulkan/tu_extensions.py
index 50afba18920..4a2d37f94ed 100644
--- a/src/freedreno/vulkan/tu_extensions.py
+++ b/src/freedreno/vulkan/tu_extensions.py
@@ -91,6 +91,7 @@ EXTENSIONS = [
Extension('VK_KHR_draw_indirect_count', 1, True),
Extension('VK_EXT_4444_formats', 1, True),
Extension('VK_EXT_conditional_rendering', 1, True),
+ Extension('VK_EXT_custom_border_color', 12, True),
]
MAX_API_VERSION = VkVersion(MAX_API_VERSION)
diff --git a/src/freedreno/vulkan/tu_private.h b/src/freedreno/vulkan/tu_private.h
index 3d1ed0fe5f4..28ba3ec84c1 100644
--- a/src/freedreno/vulkan/tu_private.h
+++ b/src/freedreno/vulkan/tu_private.h
@@ -353,12 +353,12 @@ enum global_shader {
GLOBAL_SH_COUNT,
};
+#define TU_BORDER_COLOR_COUNT 4096
+#define TU_BORDER_COLOR_BUILTIN 6
+
/* This struct defines the layout of the global_bo */
struct tu6_global
{
- /* 6 bcolor_entry entries, one for each VK_BORDER_COLOR */
- uint8_t border_color[128 * 6];
-
/* clear/blit shaders, all <= 16 instrs (16 instr = 1 instrlen unit) */
instr_t shaders[GLOBAL_SH_COUNT][16];
@@ -375,6 +375,9 @@ struct tu6_global
uint32_t offset;
uint32_t pad[7];
} flush_base[4];
+
+ /* note: larger global bo will be used for customBorderColors */
+ struct bcolor_entry bcolor_builtin[TU_BORDER_COLOR_BUILTIN], bcolor[];
};
#define gb_offset(member) offsetof(struct tu6_global, member)
#define global_iova(cmd, member) ((cmd)->device->global_bo.iova + gb_offset(member))
@@ -417,7 +420,8 @@ struct tu_device
uint32_t vsc_draw_strm_pitch;
uint32_t vsc_prim_strm_pitch;
- mtx_t vsc_pitch_mtx;
+ BITSET_DECLARE(custom_border_color, TU_BORDER_COLOR_COUNT);
+ mtx_t mutex;
};
VkResult _tu_device_set_lost(struct tu_device *device,
diff --git a/src/freedreno/vulkan/tu_util.h b/src/freedreno/vulkan/tu_util.h
index e4dd3094fd3..266dca24548 100644
--- a/src/freedreno/vulkan/tu_util.h
+++ b/src/freedreno/vulkan/tu_util.h
@@ -14,6 +14,8 @@
#include "util/macros.h"
#include "util/u_math.h"
+#include "util/format/u_format_pack.h"
+#include "util/format/u_format_zs.h"
#include "compiler/shader_enums.h"
#include "adreno_common.xml.h"
@@ -250,4 +252,48 @@ tu6_polygon_mode(VkPolygonMode mode)
}
}
+struct bcolor_entry {
+ uint32_t fp32[4];
+ uint64_t ui16;
+ uint64_t si16;
+ uint64_t fp16;
+ uint16_t rgb565;
+ uint16_t rgb5a1;
+ uint16_t rgba4;
+ uint8_t __pad0[2];
+ uint32_t ui8;
+ uint32_t si8;
+ uint32_t rgb10a2;
+ uint32_t z24; /* also s8? */
+ uint64_t srgb;
+ uint8_t __pad1[56];
+} __attribute__((aligned(128)));
+
+static inline void
+tu6_pack_border_color(struct bcolor_entry *bcolor, const VkClearColorValue *val, bool is_int)
+{
+ memcpy(bcolor->fp32, val, 4 * sizeof(float));
+ if (is_int) {
+ /* TODO: clamp? */
+ util_format_r16g16b16a16_uint_pack_unsigned((uint8_t*) &bcolor->fp16,
+ 0, val->uint32, 0, 1, 1);
+ return;
+ }
+#define PACK_F(x, type) util_format_##type##_pack_rgba_float \
+ ( (uint8_t*) (&bcolor->x), 0, val->float32, 0, 1, 1)
+ PACK_F(ui16, r16g16b16a16_unorm);
+ PACK_F(si16, r16g16b16a16_snorm);
+ PACK_F(fp16, r16g16b16a16_float);
+ PACK_F(rgb565, r5g6b5_unorm);
+ PACK_F(rgb5a1, r5g5b5a1_unorm);
+ PACK_F(rgba4, r4g4b4a4_unorm);
+ PACK_F(ui8, r8g8b8a8_unorm);
+ PACK_F(si8, r8g8b8a8_snorm);
+ PACK_F(rgb10a2, r10g10b10a2_unorm);
+ util_format_x8z24_unorm_pack_z_float((uint8_t*) &bcolor->z24,
+ 0, val->float32, 0, 1, 1);
+ PACK_F(srgb, r16g16b16a16_float); /* TODO: clamp? */
+#undef PACK_F
+}
+
#endif /* TU_UTIL_H */
diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_emit.c b/src/gallium/drivers/freedreno/a6xx/fd6_emit.c
index 67d9a8f13ae..c055ac0d2f4 100644
--- a/src/gallium/drivers/freedreno/a6xx/fd6_emit.c
+++ b/src/gallium/drivers/freedreno/a6xx/fd6_emit.c
@@ -346,7 +346,7 @@ fd6_emit_textures(struct fd_pipe *pipe, struct fd_ringbuffer *ring,
OUT_RING(state, sampler->texsamp0);
OUT_RING(state, sampler->texsamp1);
OUT_RING(state, sampler->texsamp2 |
- A6XX_TEX_SAMP_2_BCOLOR_OFFSET((i + bcolor_offset) * sizeof(struct bcolor_entry)));
+ A6XX_TEX_SAMP_2_BCOLOR(i + bcolor_offset));
OUT_RING(state, sampler->texsamp3);
needs_border |= sampler->needs_border;
}
More information about the mesa-commit
mailing list