blender
/
blender
134
398
Fork 579
Code Issues 6.3k Pull Requests 754 Projects 19 Wiki Activity

Vulkan: Texture Data Conversions #105762

Merged
Jeroen Bakker merged 23 commits from Jeroen-Bakker/blender:vulkan-textures into main 2023-03-24 08:09:33 +01:00
Conversation 0 Commits 23 Files Changed 9 +1686 -18
9 changed files with 1686 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
source/blender/gpu/CMakeLists.txt
View File

@ -52,6 +52,7 @@ endif()
set(INC_SYS
${Epoxy_INCLUDE_DIRS}
${IMATH_INCLUDE_DIR}
)
set(SRC
@ -203,6 +204,7 @@ set(VULKAN_SRC
vulkan/vk_command_buffer.cc
vulkan/vk_common.cc
vulkan/vk_context.cc
vulkan/vk_data_conversion.cc
vulkan/vk_debug.cc
vulkan/vk_descriptor_pools.cc
vulkan/vk_descriptor_set.cc
@ -232,6 +234,7 @@ set(VULKAN_SRC
vulkan/vk_command_buffer.hh
vulkan/vk_common.hh
vulkan/vk_context.hh
vulkan/vk_data_conversion.hh
vulkan/vk_descriptor_pools.hh
vulkan/vk_descriptor_set.hh
vulkan/vk_drawlist.hh

3
source/blender/gpu/intern/gpu_texture.cc
View File

@ -213,7 +213,8 @@ void Texture::detach_from(FrameBuffer *fb)
void Texture::update(eGPUDataFormat format, const void *data)
{
int mip = 0;
int extent[3], offset[3] = {0, 0, 0};
int extent[3] = {1, 1, 1};
fclem marked this conversation as resolved
Jeroen Bakker commented 2023-03-17 13:36:56 +01:00
Review
Copy Link

This was added as it is easier to have a good default, than afterwards finding out what should be set and what not. Without this change I have to set the 1 based on the actual texture type. what adds a minimum overhead.

I think this is a better solution as it can reduce code complexity in other backends as well.

This was added as it is easier to have a good default, than afterwards finding out what should be set and what not. Without this change I have to set the 1 based on the actual texture type. what adds a minimum overhead. I think this is a better solution as it can reduce code complexity in other backends as well.
int offset[3] = {0, 0, 0};
this->mip_size_get(mip, extent);
this->update_sub(mip, offset, extent, format, data);
}

3
source/blender/gpu/intern/gpu_texture_private.hh
View File

@ -759,7 +759,8 @@ inline size_t to_bytesize(eGPUTextureFormat tex_format, eGPUDataFormat data_form
}
/* Definitely not complete, edit according to the gl specification. */
inline bool validate_data_format(eGPUTextureFormat tex_format, eGPUDataFormat data_format)
constexpr inline bool validate_data_format(eGPUTextureFormat tex_format,
eGPUDataFormat data_format)
{
switch (tex_format) {
/* Formats texture & render-buffer */

718
source/blender/gpu/tests/texture_test.cc
View File

@ -5,10 +5,23 @@
#include "MEM_guardedalloc.h"
#include "BLI_math_vector.hh"
#include "BLI_vector.hh"
#include "GPU_context.h"
#include "GPU_texture.h"
#include "gpu_texture_private.hh"
/* Not all texture types are supported by all platforms. This define safe guards them until we have
* a working workaround or decided to remove support for those texture types. */
#define RUN_UNSUPPORTED false
/* Skip tests that haven't been developed yet due to non standard data types or it needs an
* framebuffer to create the texture. */
#define RUN_SRGB_UNIMPLEMENTED false
#define RUN_NON_STANDARD_UNIMPLEMENTED false
#define RUN_COMPONENT_UNIMPLEMENTED false
namespace blender::gpu::tests {
static void test_texture_read()
@ -48,4 +61,709 @@ static void test_texture_read()
}
GPU_TEST(texture_read)
template<typename DataType> static DataType *generate_test_data(size_t data_len)
{
DataType *data = static_cast<DataType *>(MEM_mallocN(data_len * sizeof(DataType), __func__));
for (int i : IndexRange(data_len)) {
if (std::is_same<DataType, float>()) {
data[i] = (DataType)(i % 8) / 8.0f;
}
else {
data[i] = (DataType)(i % 8);
}
}
return data;
}
template<eGPUTextureFormat DeviceFormat,
eGPUDataFormat HostFormat,
typename DataType,
int Size = 16>
static void texture_create_upload_read()
{
static_assert(!std::is_same<DataType, float>());
static_assert(validate_data_format(DeviceFormat, HostFormat));
eGPUTextureUsage usage = GPU_TEXTURE_USAGE_ATTACHMENT | GPU_TEXTURE_USAGE_HOST_READ;
GPUTexture *texture = GPU_texture_create_2d(
"texture", Size, Size, 1, DeviceFormat, usage, nullptr);
if (texture == nullptr) {
GTEST_SKIP() << "Platform doesn't support texture format [" << STRINGIFY(DeviceFormat) << "]";
}
size_t data_len = Size * Size * to_component_len(DeviceFormat);
DataType *data = static_cast<DataType *>(generate_test_data<DataType>(data_len));
GPU_texture_update(texture, HostFormat, data);
DataType *read_data = static_cast<DataType *>(GPU_texture_read(texture, HostFormat, 0));
bool failed = false;
for (int i : IndexRange(data_len)) {
bool ok = (read_data[i] - data[i]) == 0;
failed |= !ok;
}
EXPECT_FALSE(failed);
MEM_freeN(read_data);
MEM_freeN(data);
GPU_texture_free(texture);
}
template<eGPUTextureFormat DeviceFormat, eGPUDataFormat HostFormat, int Size = 16>
static void texture_create_upload_read_with_bias(float max_allowed_bias)
{
static_assert(validate_data_format(DeviceFormat, HostFormat));
eGPUTextureUsage usage = GPU_TEXTURE_USAGE_ATTACHMENT | GPU_TEXTURE_USAGE_HOST_READ;
GPUTexture *texture = GPU_texture_create_2d(
"texture", Size, Size, 1, DeviceFormat, usage, nullptr);
if (texture == nullptr) {
GTEST_SKIP() << "Platform doesn't support texture format [" << STRINGIFY(DeviceFormat) << "]";
}
size_t data_len = Size * Size * to_component_len(DeviceFormat);
float *data = static_cast<float *>(generate_test_data<float>(data_len));
GPU_texture_update(texture, HostFormat, data);
float *read_data = static_cast<float *>(GPU_texture_read(texture, HostFormat, 0));
float max_used_bias = 0.0f;
for (int i : IndexRange(data_len)) {
float bias = abs(read_data[i] - data[i]);
max_used_bias = max_ff(max_used_bias, bias);
}
EXPECT_LE(max_used_bias, max_allowed_bias);
MEM_freeN(read_data);
MEM_freeN(data);
GPU_texture_free(texture);
}
/* Derivative of texture_create_upload_read_pixels that doesn't test each component, but a pixel at
* a time. This is needed to check the R11G11B10 and similar types. */
template<eGPUTextureFormat DeviceFormat, eGPUDataFormat HostFormat, int Size = 16>
static void texture_create_upload_read_pixel()
{
using DataType = uint32_t;
static_assert(validate_data_format(DeviceFormat, HostFormat));
eGPUTextureUsage usage = GPU_TEXTURE_USAGE_ATTACHMENT | GPU_TEXTURE_USAGE_HOST_READ;
GPUTexture *texture = GPU_texture_create_2d(
"texture", Size, Size, 1, DeviceFormat, usage, nullptr);
ASSERT_NE(texture, nullptr);
size_t data_len = Size * Size;
DataType *data = static_cast<DataType *>(generate_test_data<DataType>(data_len));
GPU_texture_update(texture, HostFormat, data);
DataType *read_data = static_cast<DataType *>(GPU_texture_read(texture, HostFormat, 0));
bool failed = false;
for (int i : IndexRange(data_len)) {
bool ok = (read_data[i] - data[i]) == 0;
failed |= !ok;
}
EXPECT_FALSE(failed);
MEM_freeN(read_data);
MEM_freeN(data);
GPU_texture_free(texture);
}
/* -------------------------------------------------------------------- */
/** \name Roundtrip testing GPU_DATA_FLOAT
* \{ */
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_RGBA8()
{
texture_create_upload_read_with_bias<GPU_RGBA8, GPU_DATA_FLOAT>(0.004f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_RGBA8);
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_RGBA16F()
{
texture_create_upload_read_with_bias<GPU_RGBA16F, GPU_DATA_FLOAT>(0.9f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_RGBA16F);
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_RGBA16()
{
texture_create_upload_read_with_bias<GPU_RGBA16, GPU_DATA_FLOAT>(0.00002f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_RGBA16);
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_RGBA32F()
{
texture_create_upload_read_with_bias<GPU_RGBA32F, GPU_DATA_FLOAT>(0.0f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_RGBA32F);
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_RG8()
{
texture_create_upload_read_with_bias<GPU_RG8, GPU_DATA_FLOAT>(0.004f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_RG8);
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_RG16F()
{
texture_create_upload_read_with_bias<GPU_RG16F, GPU_DATA_FLOAT>(0.9f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_RG16F);
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_RG16()
{
texture_create_upload_read_with_bias<GPU_RG16, GPU_DATA_FLOAT>(0.00002f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_RG16);
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_RG32F()
{
texture_create_upload_read_with_bias<GPU_RG32F, GPU_DATA_FLOAT>(0.0f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_RG32F);
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_R8()
{
texture_create_upload_read_with_bias<GPU_R8, GPU_DATA_FLOAT>(0.004f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_R8);
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_R16F()
{
texture_create_upload_read_with_bias<GPU_R16F, GPU_DATA_FLOAT>(0.9f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_R16F);
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_R16()
{
texture_create_upload_read_with_bias<GPU_R16, GPU_DATA_FLOAT>(0.00002f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_R16);
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_R32F()
{
texture_create_upload_read_with_bias<GPU_R32F, GPU_DATA_FLOAT>(0.0f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_R32F);
#if RUN_NON_STANDARD_UNIMPLEMENTED
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_RGB10_A2()
{
texture_create_upload_read_with_bias<GPU_RGB10_A2, GPU_DATA_FLOAT>(0.0f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_RGB10_A2);
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_RGB10_A2UI()
{
texture_create_upload_read_with_bias<GPU_RGB10_A2UI, GPU_DATA_FLOAT>(0.0f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_RGB10_A2UI);
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_R11F_G11F_B10F()
{
texture_create_upload_read_with_bias<GPU_R11F_G11F_B10F, GPU_DATA_FLOAT>(0.0f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_R11F_G11F_B10F);
#endif
#if RUN_SRGB_UNIMPLEMENTED
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_SRGB8_A8()
{
texture_create_upload_read_with_bias<GPU_SRGB8_A8, GPU_DATA_FLOAT>(0.0f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_SRGB8_A8);
#endif
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_RGBA8_SNORM()
{
texture_create_upload_read_with_bias<GPU_RGBA8_SNORM, GPU_DATA_FLOAT>(0.004f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_RGBA8_SNORM);
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_RGBA16_SNORM()
{
texture_create_upload_read_with_bias<GPU_RGBA16_SNORM, GPU_DATA_FLOAT>(0.00002f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_RGBA16_SNORM);
#if RUN_UNSUPPORTED
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_RGB8()
{
texture_create_upload_read_with_bias<GPU_RGB8, GPU_DATA_FLOAT>(0.0f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_RGB8);
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_RGB8_SNORM()
{
texture_create_upload_read_with_bias<GPU_RGB8_SNORM, GPU_DATA_FLOAT>(0.0f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_RGB8_SNORM);
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_RGB16F()
{
texture_create_upload_read_with_bias<GPU_RGB16F, GPU_DATA_FLOAT>(0.0f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_RGB16F);
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_RGB16()
{
texture_create_upload_read_with_bias<GPU_RGB16, GPU_DATA_FLOAT>(0.0f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_RGB16);
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_RGB16_SNORM()
{
texture_create_upload_read_with_bias<GPU_RGB16_SNORM, GPU_DATA_FLOAT>(0.0f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_RGB16_SNORM);
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_RGB32F()
{
texture_create_upload_read_with_bias<GPU_RGB32F, GPU_DATA_FLOAT>(0.0f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_RGB32F);
#endif
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_RG8_SNORM()
{
texture_create_upload_read_with_bias<GPU_RG8_SNORM, GPU_DATA_FLOAT>(0.004f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_RG8_SNORM);
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_RG16_SNORM()
{
texture_create_upload_read_with_bias<GPU_RG16_SNORM, GPU_DATA_FLOAT>(0.00002f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_RG16_SNORM);
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_R8_SNORM()
{
texture_create_upload_read_with_bias<GPU_R8_SNORM, GPU_DATA_FLOAT>(0.004f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_R8_SNORM);
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_R16_SNORM()
{
texture_create_upload_read_with_bias<GPU_R16_SNORM, GPU_DATA_FLOAT>(0.00002f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_R16_SNORM);
#if RUN_NON_STANDARD_UNIMPLEMENTED
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_SRGB8_A8_DXT1()
{
texture_create_upload_read_with_bias<GPU_SRGB8_A8_DXT1, GPU_DATA_FLOAT>(0.0f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_SRGB8_A8_DXT1);
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_SRGB8_A8_DXT3()
{
texture_create_upload_read_with_bias<GPU_SRGB8_A8_DXT3, GPU_DATA_FLOAT>(0.0f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_SRGB8_A8_DXT3);
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_SRGB8_A8_DXT5()
{
texture_create_upload_read_with_bias<GPU_SRGB8_A8_DXT5, GPU_DATA_FLOAT>(0.0f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_SRGB8_A8_DXT5);
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_RGBA8_DXT1()
{
texture_create_upload_read_with_bias<GPU_RGBA8_DXT1, GPU_DATA_FLOAT>(0.0f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_RGBA8_DXT1);
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_RGBA8_DXT3()
{
texture_create_upload_read_with_bias<GPU_RGBA8_DXT3, GPU_DATA_FLOAT>(0.0f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_RGBA8_DXT3);
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_RGBA8_DXT5()
{
texture_create_upload_read_with_bias<GPU_RGBA8_DXT5, GPU_DATA_FLOAT>(0.0f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_RGBA8_DXT5);
#endif
#if RUN_SRGB_UNIMPLEMENTED
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_SRGB8()
{
texture_create_upload_read_with_bias<GPU_SRGB8, GPU_DATA_FLOAT>(0.0f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_SRGB8);
#endif
#if RUN_NON_STANDARD_UNIMPLEMENTED
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_RGB9_E5()
{
texture_create_upload_read_with_bias<GPU_RGB9_E5, GPU_DATA_FLOAT>(0.0f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_RGB9_E5);
#endif
#if RUN_UNSUPPORTED
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_DEPTH_COMPONENT32F()
{
texture_create_upload_read_with_bias<GPU_DEPTH_COMPONENT32F, GPU_DATA_FLOAT>(0.0f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_DEPTH_COMPONENT32F);
#endif
#if RUN_COMPONENT_UNIMPLEMENTED
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_DEPTH_COMPONENT24()
{
texture_create_upload_read_with_bias<GPU_DEPTH_COMPONENT24, GPU_DATA_FLOAT>(0.0f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_DEPTH_COMPONENT24);
static void test_texture_roundtrip__GPU_DATA_FLOAT__GPU_DEPTH_COMPONENT16()
{
texture_create_upload_read_with_bias<GPU_DEPTH_COMPONENT16, GPU_DATA_FLOAT>(0.0f);
}
GPU_TEST(texture_roundtrip__GPU_DATA_FLOAT__GPU_DEPTH_COMPONENT16);
#endif
/* \} */
/* -------------------------------------------------------------------- */
/** \name Roundtrip testing GPU_DATA_HALF_FLOAT
* \{ */
static void test_texture_roundtrip__GPU_DATA_HALF_FLOAT__GPU_RGBA16F()
{
texture_create_upload_read<GPU_RGBA16F, GPU_DATA_HALF_FLOAT, uint16_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_HALF_FLOAT__GPU_RGBA16F);
static void test_texture_roundtrip__GPU_DATA_HALF_FLOAT__GPU_RG16F()
{
texture_create_upload_read<GPU_RG16F, GPU_DATA_HALF_FLOAT, uint16_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_HALF_FLOAT__GPU_RG16F);
static void test_texture_roundtrip__GPU_DATA_HALF_FLOAT__GPU_R16F()
{
texture_create_upload_read<GPU_R16F, GPU_DATA_HALF_FLOAT, uint16_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_HALF_FLOAT__GPU_R16F);
#if RUN_UNSUPPORTED
static void test_texture_roundtrip__GPU_DATA_HALF_FLOAT__GPU_RGB16F()
{
texture_create_upload_read<GPU_RGB16F, GPU_DATA_HALF_FLOAT, uint16_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_HALF_FLOAT__GPU_RGB16F);
#endif
/* \} */
/* -------------------------------------------------------------------- */
/** \name Roundtrip testing GPU_DATA_INT
* \{ */
static void test_texture_roundtrip__GPU_DATA_INT__GPU_RGBA8I()
{
texture_create_upload_read<GPU_RGBA8I, GPU_DATA_INT, int32_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_INT__GPU_RGBA8I);
static void test_texture_roundtrip__GPU_DATA_INT__GPU_RGBA16I()
{
texture_create_upload_read<GPU_RGBA16I, GPU_DATA_INT, int32_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_INT__GPU_RGBA16I);
static void test_texture_roundtrip__GPU_DATA_INT__GPU_RGBA32I()
{
texture_create_upload_read<GPU_RGBA32I, GPU_DATA_INT, int32_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_INT__GPU_RGBA32I);
static void test_texture_roundtrip__GPU_DATA_INT__GPU_RG8I()
{
texture_create_upload_read<GPU_RG8I, GPU_DATA_INT, int32_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_INT__GPU_RG8I);
static void test_texture_roundtrip__GPU_DATA_INT__GPU_RG16I()
{
texture_create_upload_read<GPU_RG16I, GPU_DATA_INT, int32_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_INT__GPU_RG16I);
static void test_texture_roundtrip__GPU_DATA_INT__GPU_RG32I()
{
texture_create_upload_read<GPU_RG32I, GPU_DATA_INT, int32_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_INT__GPU_RG32I);
static void test_texture_roundtrip__GPU_DATA_INT__GPU_R8I()
{
texture_create_upload_read<GPU_R8I, GPU_DATA_INT, int32_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_INT__GPU_R8I);
static void test_texture_roundtrip__GPU_DATA_INT__GPU_R16I()
{
texture_create_upload_read<GPU_R16I, GPU_DATA_INT, int32_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_INT__GPU_R16I);
static void test_texture_roundtrip__GPU_DATA_INT__GPU_R32I()
{
texture_create_upload_read<GPU_R32I, GPU_DATA_INT, int32_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_INT__GPU_R32I);
#if RUN_UNSUPPORTED
static void test_texture_roundtrip__GPU_DATA_INT__GPU_RGB8I()
{
texture_create_upload_read<GPU_RGB8I, GPU_DATA_INT, int32_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_INT__GPU_RGB8I);
static void test_texture_roundtrip__GPU_DATA_INT__GPU_RGB16I()
{
texture_create_upload_read<GPU_RGB16I, GPU_DATA_INT, int32_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_INT__GPU_RGB16I);
static void test_texture_roundtrip__GPU_DATA_INT__GPU_RGB32I()
{
texture_create_upload_read<GPU_RGB32I, GPU_DATA_INT, int32_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_INT__GPU_RGB32I);
#endif
/* \} */
/* -------------------------------------------------------------------- */
/** \name Roundtrip testing GPU_DATA_UINT
* \{ */
static void test_texture_roundtrip__GPU_DATA_UINT__GPU_RGBA8UI()
{
texture_create_upload_read<GPU_RGBA8UI, GPU_DATA_UINT, uint32_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_UINT__GPU_RGBA8UI);
static void test_texture_roundtrip__GPU_DATA_UINT__GPU_RGBA16UI()
{
texture_create_upload_read<GPU_RGBA16UI, GPU_DATA_UINT, uint32_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_UINT__GPU_RGBA16UI);
static void test_texture_roundtrip__GPU_DATA_UINT__GPU_RGBA32UI()
{
texture_create_upload_read<GPU_RGBA32UI, GPU_DATA_UINT, uint32_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_UINT__GPU_RGBA32UI);
static void test_texture_roundtrip__GPU_DATA_UINT__GPU_RG8UI()
{
texture_create_upload_read<GPU_RG8UI, GPU_DATA_UINT, uint32_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_UINT__GPU_RG8UI);
static void test_texture_roundtrip__GPU_DATA_UINT__GPU_RG16UI()
{
texture_create_upload_read<GPU_RG16UI, GPU_DATA_UINT, uint32_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_UINT__GPU_RG16UI);
static void test_texture_roundtrip__GPU_DATA_UINT__GPU_RG32UI()
{
texture_create_upload_read<GPU_RG32UI, GPU_DATA_UINT, uint32_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_UINT__GPU_RG32UI);
static void test_texture_roundtrip__GPU_DATA_UINT__GPU_R8UI()
{
texture_create_upload_read<GPU_R8UI, GPU_DATA_UINT, uint32_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_UINT__GPU_R8UI);
static void test_texture_roundtrip__GPU_DATA_UINT__GPU_R16UI()
{
texture_create_upload_read<GPU_R16UI, GPU_DATA_UINT, uint32_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_UINT__GPU_R16UI);
static void test_texture_roundtrip__GPU_DATA_UINT__GPU_R32UI()
{
texture_create_upload_read<GPU_R32UI, GPU_DATA_UINT, uint32_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_UINT__GPU_R32UI);
#if RUN_UNSUPPORTED
static void test_texture_roundtrip__GPU_DATA_UINT__GPU_DEPTH32F_STENCIL8()
{
texture_create_upload_read<GPU_DEPTH32F_STENCIL8, GPU_DATA_UINT, uint32_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_UINT__GPU_DEPTH32F_STENCIL8);
static void test_texture_roundtrip__GPU_DATA_UINT__GPU_DEPTH24_STENCIL8()
{
texture_create_upload_read<GPU_DEPTH24_STENCIL8, GPU_DATA_UINT, uint32_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_UINT__GPU_DEPTH24_STENCIL8);
#endif
#if RUN_UNSUPPORTED
static void test_texture_roundtrip__GPU_DATA_UINT__GPU_RGB8UI()
{
texture_create_upload_read<GPU_RGB8UI, GPU_DATA_UINT, uint32_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_UINT__GPU_RGB8UI);
static void test_texture_roundtrip__GPU_DATA_UINT__GPU_RGB16UI()
{
texture_create_upload_read<GPU_RGB16UI, GPU_DATA_UINT, uint32_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_UINT__GPU_RGB16UI);
static void test_texture_roundtrip__GPU_DATA_UINT__GPU_RGB32UI()
{
texture_create_upload_read<GPU_RGB32UI, GPU_DATA_UINT, uint32_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_UINT__GPU_RGB32UI);
#endif
#if RUN_COMPONENT_UNIMPLEMENTED
static void test_texture_roundtrip__GPU_DATA_UINT__GPU_DEPTH_COMPONENT32F()
{
texture_create_upload_read<GPU_DEPTH_COMPONENT32F, GPU_DATA_UINT, uint32_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_UINT__GPU_DEPTH_COMPONENT32F);
static void test_texture_roundtrip__GPU_DATA_UINT__GPU_DEPTH_COMPONENT24()
{
texture_create_upload_read<GPU_DEPTH_COMPONENT24, GPU_DATA_UINT, uint32_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_UINT__GPU_DEPTH_COMPONENT24);
static void test_texture_roundtrip__GPU_DATA_UINT__GPU_DEPTH_COMPONENT16()
{
texture_create_upload_read<GPU_DEPTH_COMPONENT16, GPU_DATA_UINT, uint32_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_UINT__GPU_DEPTH_COMPONENT16);
#endif
/* \} */
/* -------------------------------------------------------------------- */
/** \name Roundtrip testing GPU_DATA_UBYTE
* \{ */
static void test_texture_roundtrip__GPU_DATA_UBYTE__GPU_RGBA8UI()
{
texture_create_upload_read<GPU_RGBA8UI, GPU_DATA_UBYTE, uint8_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_UBYTE__GPU_RGBA8UI);
static void test_texture_roundtrip__GPU_DATA_UBYTE__GPU_RGBA8()
{
texture_create_upload_read<GPU_RGBA8, GPU_DATA_UBYTE, uint8_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_UBYTE__GPU_RGBA8);
static void test_texture_roundtrip__GPU_DATA_UBYTE__GPU_RG8UI()
{
texture_create_upload_read<GPU_RG8UI, GPU_DATA_UBYTE, uint8_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_UBYTE__GPU_RG8UI);
static void test_texture_roundtrip__GPU_DATA_UBYTE__GPU_RG8()
{
texture_create_upload_read<GPU_RG8, GPU_DATA_UBYTE, uint8_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_UBYTE__GPU_RG8);
static void test_texture_roundtrip__GPU_DATA_UBYTE__GPU_R8UI()
{
texture_create_upload_read<GPU_R8UI, GPU_DATA_UBYTE, uint8_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_UBYTE__GPU_R8UI);
static void test_texture_roundtrip__GPU_DATA_UBYTE__GPU_R8()
{
texture_create_upload_read<GPU_R8, GPU_DATA_UBYTE, uint8_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_UBYTE__GPU_R8);
#if RUN_SRGB_UNIMPLEMENTED
static void test_texture_roundtrip__GPU_DATA_UBYTE__GPU_SRGB8_A8()
{
texture_create_upload_read<GPU_SRGB8_A8, GPU_DATA_UBYTE, uint8_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_UBYTE__GPU_SRGB8_A8);
#endif
#if RUN_UNSUPPORTED
static void test_texture_roundtrip__GPU_DATA_UBYTE__GPU_RGB8I()
{
texture_create_upload_read<GPU_RGB8I, GPU_DATA_UBYTE, uint8_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_UBYTE__GPU_RGB8I);
static void test_texture_roundtrip__GPU_DATA_UBYTE__GPU_RGB8()
{
texture_create_upload_read<GPU_RGB8, GPU_DATA_UBYTE, uint8_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_UBYTE__GPU_RGB8);
static void test_texture_roundtrip__GPU_DATA_UBYTE__GPU_SRGB8()
{
texture_create_upload_read<GPU_SRGB8, GPU_DATA_UBYTE, uint8_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_UBYTE__GPU_SRGB8);
#endif
/* \} */
/* -------------------------------------------------------------------- */
/** \name Roundtrip testing GPU_DATA_UINT_24_8
* \{ */
#if RUN_UNSUPPORTED
static void test_texture_roundtrip__GPU_DATA_UINT_24_8__GPU_DEPTH32F_STENCIL8()
{
texture_create_upload_read<GPU_DEPTH32F_STENCIL8, GPU_DATA_UINT_24_8, void>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_UINT_24_8__GPU_DEPTH32F_STENCIL8);
static void test_texture_roundtrip__GPU_DATA_UINT_24_8__GPU_DEPTH24_STENCIL8()
{
texture_create_upload_read<GPU_DEPTH24_STENCIL8, GPU_DATA_UINT_24_8, void>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_UINT_24_8__GPU_DEPTH24_STENCIL8);
#endif
/* \} */
/* -------------------------------------------------------------------- */
/** \name Roundtrip testing GPU_DATA_10_11_11_REV
* \{ */
static void test_texture_roundtrip__GPU_DATA_10_11_11_REV__GPU_R11F_G11F_B10F()
{
texture_create_upload_read<GPU_R11F_G11F_B10F, GPU_DATA_10_11_11_REV, uint8_t>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_10_11_11_REV__GPU_R11F_G11F_B10F);
/* \} */
/* -------------------------------------------------------------------- */
/** \name Roundtrip testing GPU_DATA_2_10_10_10_REV
* \{ */
static void test_texture_roundtrip__GPU_DATA_2_10_10_10_REV__GPU_RGB10_A2()
{
texture_create_upload_read_pixel<GPU_RGB10_A2, GPU_DATA_2_10_10_10_REV>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_2_10_10_10_REV__GPU_RGB10_A2);
static void test_texture_roundtrip__GPU_DATA_2_10_10_10_REV__GPU_RGB10_A2UI()
{
texture_create_upload_read_pixel<GPU_RGB10_A2UI, GPU_DATA_2_10_10_10_REV>();
}
GPU_TEST(texture_roundtrip__GPU_DATA_2_10_10_10_REV__GPU_RGB10_A2UI);
/* \} */
} // namespace blender::gpu::tests

24
source/blender/gpu/vulkan/vk_command_buffer.cc
View File

@ -103,6 +103,30 @@ void VKCommandBuffer::copy(VKBuffer &dst_buffer,
regions.size(),
regions.data());
}
void VKCommandBuffer::copy(VKTexture &dst_texture,
VKBuffer &src_buffer,
Span<VkBufferImageCopy> regions)
{
vkCmdCopyBufferToImage(vk_command_buffer_,
src_buffer.vk_handle(),
dst_texture.vk_image_handle(),
VK_IMAGE_LAYOUT_GENERAL,
regions.size(),
regions.data());
}
void VKCommandBuffer::clear(VkImage vk_image,
VkImageLayout vk_image_layout,
const VkClearColorValue &vk_clear_color,
Span<VkImageSubresourceRange> ranges)
{
vkCmdClearColorImage(vk_command_buffer_,
vk_image,
vk_image_layout,
&vk_clear_color,
ranges.size(),
ranges.data());
}
void VKCommandBuffer::pipeline_barrier(VkPipelineStageFlags source_stages,
VkPipelineStageFlags destination_stages)

8
source/blender/gpu/vulkan/vk_command_buffer.hh
View File

@ -50,9 +50,17 @@ class VKCommandBuffer : NonCopyable, NonMovable {
void dispatch(int groups_x_len, int groups_y_len, int groups_z_len);
/** Copy the contents of a texture MIP level to the dst buffer. */
void copy(VKBuffer &dst_buffer, VKTexture &src_texture, Span<VkBufferImageCopy> regions);
void copy(VKTexture &dst_texture, VKBuffer &src_buffer, Span<VkBufferImageCopy> regions);
void pipeline_barrier(VkPipelineStageFlags source_stages,
VkPipelineStageFlags destination_stages);
void pipeline_barrier(Span<VkImageMemoryBarrier> image_memory_barriers);
/**
* Clear color image resource.
*/
void clear(VkImage vk_image,
VkImageLayout vk_image_layout,
const VkClearColorValue &vk_clear_color,
Span<VkImageSubresourceRange> ranges);
void fill(VKBuffer &buffer, uint32_t data);
/**

780
source/blender/gpu/vulkan/vk_data_conversion.cc Normal file
View File

@ -0,0 +1,780 @@
/* SPDX-License-Identifier: GPL-2.0-or-later
* Copyright 2023 Blender Foundation. All rights reserved. */
/** \file
* \ingroup gpu
*/
#include "vk_data_conversion.hh"
#include "Imath/half.h"
namespace blender::gpu {
/* -------------------------------------------------------------------- */
/** \name Conversion types
* \{ */
enum class ConversionType {
/** No conversion needed, result can be directly read back to host memory. */
PASS_THROUGH,
FLOAT_TO_UNORM8,
UNORM8_TO_FLOAT,
FLOAT_TO_SNORM8,
SNORM8_TO_FLOAT,
FLOAT_TO_UNORM16,
UNORM16_TO_FLOAT,
FLOAT_TO_SNORM16,
SNORM16_TO_FLOAT,
UI32_TO_UI16,
UI16_TO_UI32,
UI32_TO_UI8,
UI8_TO_UI32,
I32_TO_I16,
I16_TO_I32,
I32_TO_I8,
I8_TO_I32,
/** Convert device 16F to floats. */
HALF_TO_FLOAT,
FLOAT_TO_HALF,
/**
* The requested conversion isn't supported.
*/
UNSUPPORTED,
};
Jeroen Bakker commented 2023-03-17 13:03:33 +01:00
Outdated
Review
Copy Link

Perhaps we can still use texture data format to find out which conversion should be used to remove these large switch statements.

Perhaps we can still use texture data format to find out which conversion should be used to remove these large switch statements.
static ConversionType type_of_conversion_float(eGPUTextureFormat device_format)
{
switch (device_format) {
case GPU_RGBA32F:
case GPU_RG32F:
case GPU_R32F:
case GPU_DEPTH_COMPONENT32F:
return ConversionType::PASS_THROUGH;
case GPU_RGBA16F:
case GPU_RG16F:
case GPU_R16F:
case GPU_RGB16F:
return ConversionType::FLOAT_TO_HALF;
case GPU_RGBA8:
case GPU_RG8:
case GPU_R8:
return ConversionType::FLOAT_TO_UNORM8;
case GPU_RGBA8_SNORM:
case GPU_RGB8_SNORM:
case GPU_RG8_SNORM:
case GPU_R8_SNORM:
return ConversionType::FLOAT_TO_SNORM8;
case GPU_RGBA16:
case GPU_RG16:
case GPU_R16:
return ConversionType::FLOAT_TO_UNORM16;
case GPU_RGBA16_SNORM:
case GPU_RGB16_SNORM:
case GPU_RG16_SNORM:
case GPU_R16_SNORM:
return ConversionType::FLOAT_TO_SNORM16;
case GPU_RGB32F: /* GPU_RGB32F Not supported by vendors. */
case GPU_RGBA8UI:
case GPU_RGBA8I:
case GPU_RGBA16UI:
case GPU_RGBA16I:
case GPU_RGBA32UI:
case GPU_RGBA32I:
case GPU_RG8UI:
case GPU_RG8I:
case GPU_RG16UI:
case GPU_RG16I:
case GPU_RG32UI:
case GPU_RG32I:
case GPU_R8UI:
case GPU_R8I:
case GPU_R16UI:
case GPU_R16I:
case GPU_R32UI:
case GPU_R32I:
case GPU_RGB10_A2:
case GPU_RGB10_A2UI:
case GPU_R11F_G11F_B10F:
case GPU_DEPTH32F_STENCIL8:
case GPU_DEPTH24_STENCIL8:
case GPU_SRGB8_A8:
case GPU_RGB8UI:
case GPU_RGB8I:
case GPU_RGB8:
case GPU_RGB16UI:
case GPU_RGB16I:
case GPU_RGB16:
case GPU_RGB32UI:
case GPU_RGB32I:
case GPU_SRGB8_A8_DXT1:
case GPU_SRGB8_A8_DXT3:
case GPU_SRGB8_A8_DXT5:
case GPU_RGBA8_DXT1:
case GPU_RGBA8_DXT3:
case GPU_RGBA8_DXT5:
case GPU_SRGB8:
case GPU_RGB9_E5:
case GPU_DEPTH_COMPONENT24:
case GPU_DEPTH_COMPONENT16:
return ConversionType::UNSUPPORTED;
}
return ConversionType::UNSUPPORTED;
}
static ConversionType type_of_conversion_int(eGPUTextureFormat device_format)
{
switch (device_format) {
case GPU_RGBA32I:
case GPU_RG32I:
case GPU_R32I:
return ConversionType::PASS_THROUGH;
case GPU_RGBA16I:
case GPU_RG16I:
case GPU_R16I:
return ConversionType::I32_TO_I16;
case GPU_RGBA8I:
case GPU_RG8I:
case GPU_R8I:
return ConversionType::I32_TO_I8;
case GPU_RGBA8UI:
case GPU_RGBA8:
case GPU_RGBA16UI:
case GPU_RGBA16F:
case GPU_RGBA16:
case GPU_RGBA32UI:
case GPU_RGBA32F:
case GPU_RG8UI:
case GPU_RG8:
case GPU_RG16UI:
case GPU_RG16F:
case GPU_RG32UI:
case GPU_RG32F:
case GPU_RG16:
case GPU_R8UI:
case GPU_R8:
case GPU_R16UI:
case GPU_R16F:
case GPU_R16:
case GPU_R32UI:
case GPU_R32F:
case GPU_RGB10_A2:
case GPU_RGB10_A2UI:
case GPU_R11F_G11F_B10F:
case GPU_DEPTH32F_STENCIL8:
case GPU_DEPTH24_STENCIL8:
case GPU_SRGB8_A8:
case GPU_RGBA8_SNORM:
case GPU_RGBA16_SNORM:
case GPU_RGB8UI:
case GPU_RGB8I:
case GPU_RGB8:
case GPU_RGB8_SNORM:
case GPU_RGB16UI:
case GPU_RGB16I:
case GPU_RGB16F:
case GPU_RGB16:
case GPU_RGB16_SNORM:
case GPU_RGB32UI:
case GPU_RGB32I:
case GPU_RGB32F:
case GPU_RG8_SNORM:
case GPU_RG16_SNORM:
case GPU_R8_SNORM:
case GPU_R16_SNORM:
case GPU_SRGB8_A8_DXT1:
case GPU_SRGB8_A8_DXT3:
case GPU_SRGB8_A8_DXT5:
case GPU_RGBA8_DXT1:
case GPU_RGBA8_DXT3:
case GPU_RGBA8_DXT5:
case GPU_SRGB8:
case GPU_RGB9_E5:
case GPU_DEPTH_COMPONENT32F:
case GPU_DEPTH_COMPONENT24:
case GPU_DEPTH_COMPONENT16:
return ConversionType::UNSUPPORTED;
}
return ConversionType::UNSUPPORTED;
}
static ConversionType type_of_conversion_uint(eGPUTextureFormat device_format)
{
switch (device_format) {
case GPU_RGBA32UI:
case GPU_RG32UI:
case GPU_R32UI:
return ConversionType::PASS_THROUGH;
case GPU_RGBA16UI:
case GPU_RG16UI:
case GPU_R16UI:
case GPU_RGB16UI:
return ConversionType::UI32_TO_UI16;
case GPU_RGBA8UI:
case GPU_RG8UI:
case GPU_R8UI:
return ConversionType::UI32_TO_UI8;
case GPU_RGBA8I:
case GPU_RGBA8:
case GPU_RGBA16I:
case GPU_RGBA16F:
case GPU_RGBA16:
case GPU_RGBA32I:
case GPU_RGBA32F:
case GPU_RG8I:
case GPU_RG8:
case GPU_RG16I:
case GPU_RG16F:
case GPU_RG16:
case GPU_RG32I:
case GPU_RG32F:
case GPU_R8I:
case GPU_R8:
case GPU_R16I:
case GPU_R16F:
case GPU_R16:
case GPU_R32I:
case GPU_R32F:
case GPU_RGB10_A2:
case GPU_RGB10_A2UI:
case GPU_R11F_G11F_B10F:
case GPU_DEPTH32F_STENCIL8:
case GPU_DEPTH24_STENCIL8:
case GPU_SRGB8_A8:
case GPU_RGBA8_SNORM:
case GPU_RGBA16_SNORM:
case GPU_RGB8UI:
case GPU_RGB8I:
case GPU_RGB8:
case GPU_RGB8_SNORM:
case GPU_RGB16I:
case GPU_RGB16F:
case GPU_RGB16:
case GPU_RGB16_SNORM:
case GPU_RGB32UI:
case GPU_RGB32I:
case GPU_RGB32F:
case GPU_RG8_SNORM:
case GPU_RG16_SNORM:
case GPU_R8_SNORM:
case GPU_R16_SNORM:
case GPU_SRGB8_A8_DXT1:
case GPU_SRGB8_A8_DXT3:
case GPU_SRGB8_A8_DXT5:
case GPU_RGBA8_DXT1:
case GPU_RGBA8_DXT3:
case GPU_RGBA8_DXT5:
case GPU_SRGB8:
case GPU_RGB9_E5:
case GPU_DEPTH_COMPONENT32F:
case GPU_DEPTH_COMPONENT24:
case GPU_DEPTH_COMPONENT16:
return ConversionType::UNSUPPORTED;
}
return ConversionType::UNSUPPORTED;
}
static ConversionType type_of_conversion_half(eGPUTextureFormat device_format)
{
switch (device_format) {
case GPU_RGBA16F:
case GPU_RG16F:
case GPU_R16F:
return ConversionType::PASS_THROUGH;
case GPU_RGBA8UI:
case GPU_RGBA8I:
case GPU_RGBA8:
case GPU_RGBA16UI:
case GPU_RGBA16I:
case GPU_RGBA16:
case GPU_RGBA32UI:
case GPU_RGBA32I:
case GPU_RGBA32F:
case GPU_RG8UI:
case GPU_RG8I:
case GPU_RG8:
case GPU_RG16UI:
case GPU_RG16I:
case GPU_RG16:
case GPU_RG32UI:
case GPU_RG32I:
case GPU_RG32F:
case GPU_R8UI:
case GPU_R8I:
case GPU_R8:
case GPU_R16UI:
case GPU_R16I:
case GPU_R16:
case GPU_R32UI:
case GPU_R32I:
case GPU_R32F:
case GPU_RGB10_A2:
case GPU_RGB10_A2UI:
case GPU_R11F_G11F_B10F:
case GPU_DEPTH32F_STENCIL8:
case GPU_DEPTH24_STENCIL8:
case GPU_SRGB8_A8:
case GPU_RGBA8_SNORM:
case GPU_RGBA16_SNORM:
case GPU_RGB8UI:
case GPU_RGB8I:
case GPU_RGB8:
case GPU_RGB8_SNORM:
case GPU_RGB16UI:
case GPU_RGB16I:
case GPU_RGB16F:
case GPU_RGB16:
case GPU_RGB16_SNORM:
case GPU_RGB32UI:
case GPU_RGB32I:
case GPU_RGB32F:
case GPU_RG8_SNORM:
case GPU_RG16_SNORM:
case GPU_R8_SNORM:
case GPU_R16_SNORM:
case GPU_SRGB8_A8_DXT1:
case GPU_SRGB8_A8_DXT3:
case GPU_SRGB8_A8_DXT5:
case GPU_RGBA8_DXT1:
case GPU_RGBA8_DXT3:
case GPU_RGBA8_DXT5:
case GPU_SRGB8:
case GPU_RGB9_E5:
case GPU_DEPTH_COMPONENT32F:
case GPU_DEPTH_COMPONENT24:
case GPU_DEPTH_COMPONENT16:
return ConversionType::UNSUPPORTED;
}
return ConversionType::UNSUPPORTED;
}
static ConversionType type_of_conversion_ubyte(eGPUTextureFormat device_format)
{
switch (device_format) {
case GPU_RGBA8UI:
case GPU_RGBA8:
case GPU_RG8UI:
case GPU_RG8:
case GPU_R8UI:
case GPU_R8:
return ConversionType::PASS_THROUGH;
case GPU_RGBA8I:
case GPU_RGBA16UI:
case GPU_RGBA16I:
case GPU_RGBA16F:
case GPU_RGBA16:
case GPU_RGBA32UI:
case GPU_RGBA32I:
case GPU_RGBA32F:
case GPU_RG8I:
case GPU_RG16UI:
case GPU_RG16I:
case GPU_RG16F:
case GPU_RG16:
case GPU_RG32UI:
case GPU_RG32I:
case GPU_RG32F:
case GPU_R8I:
case GPU_R16UI:
case GPU_R16I:
case GPU_R16F:
case GPU_R16:
case GPU_R32UI:
case GPU_R32I:
case GPU_R32F:
case GPU_RGB10_A2:
case GPU_RGB10_A2UI:
case GPU_R11F_G11F_B10F:
case GPU_DEPTH32F_STENCIL8:
case GPU_DEPTH24_STENCIL8:
case GPU_SRGB8_A8:
case GPU_RGBA8_SNORM:
case GPU_RGBA16_SNORM:
case GPU_RGB8UI:
case GPU_RGB8I:
case GPU_RGB8:
case GPU_RGB8_SNORM:
case GPU_RGB16UI:
case GPU_RGB16I:
case GPU_RGB16F:
case GPU_RGB16:
case GPU_RGB16_SNORM:
case GPU_RGB32UI:
case GPU_RGB32I:
case GPU_RGB32F:
case GPU_RG8_SNORM:
case GPU_RG16_SNORM:
case GPU_R8_SNORM:
case GPU_R16_SNORM:
case GPU_SRGB8_A8_DXT1:
case GPU_SRGB8_A8_DXT3:
case GPU_SRGB8_A8_DXT5:
case GPU_RGBA8_DXT1:
case GPU_RGBA8_DXT3:
case GPU_RGBA8_DXT5:
case GPU_SRGB8:
case GPU_RGB9_E5:
case GPU_DEPTH_COMPONENT32F:
case GPU_DEPTH_COMPONENT24:
case GPU_DEPTH_COMPONENT16:
return ConversionType::UNSUPPORTED;
}
return ConversionType::UNSUPPORTED;
}
static ConversionType type_of_conversion_r11g11b10(eGPUTextureFormat device_format)
{
if (device_format == GPU_R11F_G11F_B10F) {
return ConversionType::PASS_THROUGH;
}
return ConversionType::UNSUPPORTED;
}
static ConversionType type_of_conversion_r10g10b10a2(eGPUTextureFormat device_format)
{
if (ELEM(device_format, GPU_RGB10_A2, GPU_RGB10_A2UI)) {
return ConversionType::PASS_THROUGH;
}
return ConversionType::UNSUPPORTED;
}
static ConversionType host_to_device(eGPUDataFormat host_format, eGPUTextureFormat device_format)
{
BLI_assert(validate_data_format(device_format, host_format));
switch (host_format) {
case GPU_DATA_FLOAT:
return type_of_conversion_float(device_format);
case GPU_DATA_UINT:
return type_of_conversion_uint(device_format);
case GPU_DATA_INT:
return type_of_conversion_int(device_format);
case GPU_DATA_HALF_FLOAT:
return type_of_conversion_half(device_format);
case GPU_DATA_UBYTE:
return type_of_conversion_ubyte(device_format);
case GPU_DATA_10_11_11_REV:
return type_of_conversion_r11g11b10(device_format);
case GPU_DATA_2_10_10_10_REV:
return type_of_conversion_r10g10b10a2(device_format);
case GPU_DATA_UINT_24_8:
return ConversionType::UNSUPPORTED;
}
return ConversionType::UNSUPPORTED;
}
static ConversionType reversed(ConversionType type)
{
#define CASE_SINGLE(a, b) \
case ConversionType::a##_TO_##b: \
return ConversionType::b##_TO_##a;
#define CASE_PAIR(a, b) \
CASE_SINGLE(a, b) \
CASE_SINGLE(b, a)
switch (type) {
case ConversionType::PASS_THROUGH:
return ConversionType::PASS_THROUGH;
CASE_PAIR(FLOAT, UNORM8)
CASE_PAIR(FLOAT, SNORM8)
CASE_PAIR(FLOAT, UNORM16)
CASE_PAIR(FLOAT, SNORM16)
CASE_PAIR(UI32, UI16)
CASE_PAIR(I32, I16)
CASE_PAIR(UI32, UI8)
CASE_PAIR(I32, I8)
CASE_PAIR(FLOAT, HALF)
case ConversionType::UNSUPPORTED:
return ConversionType::UNSUPPORTED;
}
#undef CASE_PAIR
#undef CASE_SINGLE
return ConversionType::UNSUPPORTED;
}
/* \} */
/* -------------------------------------------------------------------- */
/** \name Data Conversion
* \{ */
template<typename InnerType> struct SignedNormalized {
static_assert(std::is_same<InnerType, uint8_t>() || std::is_same<InnerType, uint16_t>());
InnerType value;
static constexpr int32_t scalar()
{
return (1 << (sizeof(InnerType) * 8 - 1));
}
static constexpr int32_t delta()
{
return (1 << (sizeof(InnerType) * 8 - 1)) - 1;
}
static constexpr int32_t max()
{
return ((1 << (sizeof(InnerType) * 8)) - 1);
}
};
template<typename InnerType> struct UnsignedNormalized {
static_assert(std::is_same<InnerType, uint8_t>() || std::is_same<InnerType, uint16_t>());
InnerType value;
static constexpr int32_t scalar()
{
return (1 << (sizeof(InnerType) * 8)) - 1;
}
static constexpr int32_t max()
{
return ((1 << (sizeof(InnerType) * 8)) - 1);
}
};
template<typename InnerType> struct ComponentValue {
InnerType value;
};
using UI8 = ComponentValue<uint8_t>;
using UI16 = ComponentValue<uint16_t>;
using UI32 = ComponentValue<uint32_t>;
using I8 = ComponentValue<int8_t>;
using I16 = ComponentValue<int16_t>;
using I32 = ComponentValue<int32_t>;
using F32 = ComponentValue<float>;
using F16 = ComponentValue<uint16_t>;
template<typename StorageType>
void convert_component(SignedNormalized<StorageType> &dst, const F32 &src)
{
static constexpr int32_t scalar = SignedNormalized<StorageType>::scalar();
static constexpr int32_t delta = SignedNormalized<StorageType>::delta();
static constexpr int32_t max = SignedNormalized<StorageType>::max();
dst.value = (clamp_i((src.value * scalar + delta), 0, max));
}
template<typename StorageType>
void convert_component(F32 &dst, const SignedNormalized<StorageType> &src)
{
static constexpr int32_t scalar = SignedNormalized<StorageType>::scalar();
static constexpr int32_t delta = SignedNormalized<StorageType>::delta();
dst.value = float(int32_t(src.value) - delta) / scalar;
}
template<typename StorageType>
void convert_component(UnsignedNormalized<StorageType> &dst, const F32 &src)
{
static constexpr int32_t scalar = UnsignedNormalized<StorageType>::scalar();
static constexpr int32_t max = scalar;
dst.value = (clamp_i((src.value * scalar), 0, max));
}
template<typename StorageType>
void convert_component(F32 &dst, const UnsignedNormalized<StorageType> &src)
{
static constexpr int32_t scalar = UnsignedNormalized<StorageType>::scalar();
dst.value = float(src.value) / scalar;
}
/* Copy the contents of src to dst with out performing any actual conversion.*/
template<typename DestinationType, typename SourceType>
void convert_component(DestinationType &dst, const SourceType &src)
{
static_assert(std::is_same<DestinationType, UI8>() || std::is_same<DestinationType, UI16>() ||
std::is_same<DestinationType, UI32>() || std::is_same<DestinationType, I8>() ||
std::is_same<DestinationType, I16>() || std::is_same<DestinationType, I32>());
static_assert(std::is_same<SourceType, UI8>() || std::is_same<SourceType, UI16>() ||
std::is_same<SourceType, UI32>() || std::is_same<SourceType, I8>() ||
std::is_same<SourceType, I16>() || std::is_same<SourceType, I32>());
static_assert(!std::is_same<DestinationType, SourceType>());
dst.value = src.value;
}
static void convert_component(F16 &dst, const F32 &src)
{
dst.value = imath_float_to_half(src.value);
}
static void convert_component(F32 &dst, const F16 &src)
{
dst.value = imath_half_to_float(src.value);
}
/* \} */
template<typename DestinationType, typename SourceType>
void convert_per_component(MutableSpan<DestinationType> dst, Span<SourceType> src)
{
BLI_assert(src.size() == dst.size());
for (int64_t index : IndexRange(src.size())) {
convert_component(dst[index], src[index]);
}
}
template<typename DestinationType, typename SourceType>
void convert_per_component(void *dst_memory,
const void *src_memory,
size_t buffer_size,
eGPUTextureFormat device_format)
{
size_t total_components = to_component_len(device_format) * buffer_size;
Span<SourceType> src = Span<SourceType>(static_cast<const SourceType *>(src_memory),
total_components);
MutableSpan<DestinationType> dst = MutableSpan<DestinationType>(
static_cast<DestinationType *>(dst_memory), total_components);
convert_per_component<DestinationType, SourceType>(dst, src);
}
static void convert_buffer(void *dst_memory,
const void *src_memory,
size_t buffer_size,
eGPUTextureFormat device_format,
ConversionType type)
{
switch (type) {
case ConversionType::UNSUPPORTED:
return;
case ConversionType::PASS_THROUGH:
memcpy(dst_memory, src_memory, buffer_size * to_bytesize(device_format));
return;
case ConversionType::UI32_TO_UI16:
convert_per_component<UI16, UI32>(dst_memory, src_memory, buffer_size, device_format);
break;
case ConversionType::UI16_TO_UI32:
convert_per_component<UI32, UI16>(dst_memory, src_memory, buffer_size, device_format);
break;
case ConversionType::UI32_TO_UI8:
convert_per_component<UI8, UI32>(dst_memory, src_memory, buffer_size, device_format);
break;
case ConversionType::UI8_TO_UI32:
convert_per_component<UI32, UI8>(dst_memory, src_memory, buffer_size, device_format);
break;
case ConversionType::I32_TO_I16:
convert_per_component<I16, I32>(dst_memory, src_memory, buffer_size, device_format);
break;
case ConversionType::I16_TO_I32:
convert_per_component<I32, I16>(dst_memory, src_memory, buffer_size, device_format);
break;
case ConversionType::I32_TO_I8:
convert_per_component<I8, I32>(dst_memory, src_memory, buffer_size, device_format);
break;
case ConversionType::I8_TO_I32:
convert_per_component<I32, I8>(dst_memory, src_memory, buffer_size, device_format);
break;
case ConversionType::FLOAT_TO_SNORM8:
convert_per_component<SignedNormalized<uint8_t>, F32>(
dst_memory, src_memory, buffer_size, device_format);
break;
case ConversionType::SNORM8_TO_FLOAT:
convert_per_component<F32, SignedNormalized<uint8_t>>(
dst_memory, src_memory, buffer_size, device_format);
break;
case ConversionType::FLOAT_TO_SNORM16:
convert_per_component<SignedNormalized<uint16_t>, F32>(
dst_memory, src_memory, buffer_size, device_format);
break;
case ConversionType::SNORM16_TO_FLOAT:
convert_per_component<F32, SignedNormalized<uint16_t>>(
dst_memory, src_memory, buffer_size, device_format);
break;
case ConversionType::FLOAT_TO_UNORM8:
convert_per_component<UnsignedNormalized<uint8_t>, F32>(
dst_memory, src_memory, buffer_size, device_format);
break;
case ConversionType::UNORM8_TO_FLOAT:
convert_per_component<F32, UnsignedNormalized<uint8_t>>(
dst_memory, src_memory, buffer_size, device_format);
break;
case ConversionType::FLOAT_TO_UNORM16:
convert_per_component<UnsignedNormalized<uint16_t>, F32>(
dst_memory, src_memory, buffer_size, device_format);
break;
case ConversionType::UNORM16_TO_FLOAT:
convert_per_component<F32, UnsignedNormalized<uint16_t>>(
dst_memory, src_memory, buffer_size, device_format);
break;
case ConversionType::FLOAT_TO_HALF:
convert_per_component<F16, F32>(dst_memory, src_memory, buffer_size, device_format);
break;
case ConversionType::HALF_TO_FLOAT:
convert_per_component<F32, F16>(dst_memory, src_memory, buffer_size, device_format);
break;
}
}
/* -------------------------------------------------------------------- */
/** \name API
* \{ */
void convert_host_to_device(void *dst_buffer,
const void *src_buffer,
size_t buffer_size,
eGPUDataFormat host_format,
eGPUTextureFormat device_format)
{
ConversionType conversion_type = host_to_device(host_format, device_format);
BLI_assert(conversion_type != ConversionType::UNSUPPORTED);
convert_buffer(dst_buffer, src_buffer, buffer_size, device_format, conversion_type);
}
void convert_device_to_host(void *dst_buffer,
const void *src_buffer,
size_t buffer_size,
eGPUDataFormat host_format,
eGPUTextureFormat device_format)
{
ConversionType conversion_type = reversed(host_to_device(host_format, device_format));
BLI_assert(conversion_type != ConversionType::UNSUPPORTED);
convert_buffer(dst_buffer, src_buffer, buffer_size, device_format, conversion_type);
}
/* \} */
} // namespace blender::gpu

53
source/blender/gpu/vulkan/vk_data_conversion.hh Normal file
View File

@ -0,0 +1,53 @@
/* SPDX-License-Identifier: GPL-2.0-or-later
* Copyright 2023 Blender Foundation. All rights reserved. */
/** \file
* \ingroup gpu
*/
#pragma once
#include "gpu_texture_private.hh"
namespace blender::gpu {
/**
* Convert host buffer to device buffer.
*
* \param dst_buffer: device buffer.
* \param src_buffer: host buffer.
* \param buffer_size: number of pixels to convert from the start of the given buffer.
* \param host_format: format of the host buffer
* \param device_format: format of the device buffer.
*
* \note Will assert when the host_format/device_format combination isn't valid
* (#validate_data_format) or supported. Some combinations aren't supported in Vulkan due to
* platform incompatibility.
*/
void convert_host_to_device(void *dst_buffer,
const void *src_buffer,
size_t buffer_size,
eGPUDataFormat host_format,
eGPUTextureFormat device_format);
/**
* Convert device buffer to host buffer.
*
* \param dst_buffer: host buffer
* \param src_buffer: device buffer.
* \param buffer_size: number of pixels to convert from the start of the given buffer.
* \param host_format: format of the host buffer
* \param device_format: format of the device buffer.
*
* \note Will assert when the host_format/device_format combination isn't valid
* (#validate_data_format) or supported. Some combinations aren't supported in Vulkan due to
* platform incompatibility.
*/
void convert_device_to_host(void *dst_buffer,
const void *src_buffer,
size_t buffer_size,
eGPUDataFormat host_format,
eGPUTextureFormat device_format);
}; // namespace blender::gpu

112
source/blender/gpu/vulkan/vk_texture.cc
View File

@ -9,10 +9,13 @@
#include "vk_buffer.hh"
#include "vk_context.hh"
#include "vk_data_conversion.hh"
#include "vk_memory.hh"
#include "vk_shader.hh"
#include "vk_shader_interface.hh"
#include "BLI_math_vector.hh"
#include "BKE_global.h"
namespace blender::gpu {
@ -34,8 +37,64 @@ void VKTexture::copy_to(Texture * /*tex*/)
{
}
void VKTexture::clear(eGPUDataFormat /*format*/, const void * /*data*/)
template<typename T> void copy_color(T dst[4], const T *src)
{
dst[0] = src[0];
dst[1] = src[1];
dst[2] = src[2];
dst[3] = src[3];
}
static VkClearColorValue to_vk_clear_color_value(eGPUDataFormat format, const void *data)
{
VkClearColorValue result = {0.0f};
switch (format) {
case GPU_DATA_FLOAT: {
const float *float_data = static_cast<const float *>(data);
copy_color<float>(result.float32, float_data);
break;
}
case GPU_DATA_INT: {
const int32_t *int_data = static_cast<const int32_t *>(data);
copy_color<int32_t>(result.int32, int_data);
break;
}
case GPU_DATA_UINT: {
const uint32_t *uint_data = static_cast<const uint32_t *>(data);
copy_color<uint32_t>(result.uint32, uint_data);
break;
}
case GPU_DATA_HALF_FLOAT:
case GPU_DATA_UBYTE:
case GPU_DATA_UINT_24_8:
case GPU_DATA_10_11_11_REV:
case GPU_DATA_2_10_10_10_REV: {
BLI_assert_unreachable();
break;
}
}
return result;
}
void VKTexture::clear(eGPUDataFormat format, const void *data)
{
if (!is_allocated()) {
allocate();
}
VKContext &context = *VKContext::get();
VKCommandBuffer &command_buffer = context.command_buffer_get();
VkClearColorValue clear_color = to_vk_clear_color_value(format, data);
VkImageSubresourceRange range = {0};
range.aspectMask = to_vk_image_aspect_flag_bits(format_);
range.levelCount = VK_REMAINING_MIP_LEVELS;
range.layerCount = VK_REMAINING_ARRAY_LAYERS;
command_buffer.clear(
vk_image_, VK_IMAGE_LAYOUT_GENERAL, clear_color, Span<VkImageSubresourceRange>(&range, 1));
}
void VKTexture::swizzle_set(const char /*swizzle_mask*/[4])
@ -79,22 +138,41 @@ void *VKTexture::read(int mip, eGPUDataFormat format)
command_buffer.submit();
void *data = MEM_mallocN(host_memory_size, __func__);
/* TODO: add conversion when data format is different. */
BLI_assert_msg(device_memory_size == host_memory_size,
"Memory data conversions not implemented yet");
staging_buffer.read(data);
convert_device_to_host(data, staging_buffer.mapped_memory_get(), sample_len, format, format_);
return data;
}
void VKTexture::update_sub(int /*mip*/,
int /*offset*/[3],
int /*extent*/[3],
eGPUDataFormat /*format*/,
const void * /*data*/)
void VKTexture::update_sub(
int mip, int offset[3], int extent[3], eGPUDataFormat format, const void *data)
{
if (!is_allocated()) {
allocate();
}
/* Vulkan images cannot be directly mapped to host memory and requires a staging buffer. */
VKContext &context = *VKContext::get();
VKBuffer staging_buffer;
size_t sample_len = extent[0] * extent[1] * extent[2];
size_t device_memory_size = sample_len * to_bytesize(format_);
staging_buffer.create(
context, device_memory_size, GPU_USAGE_DEVICE_ONLY, VK_BUFFER_USAGE_TRANSFER_SRC_BIT);
convert_host_to_device(staging_buffer.mapped_memory_get(), data, sample_len, format, format_);
VkBufferImageCopy region = {};
region.imageExtent.width = extent[0];
region.imageExtent.height = extent[1];
region.imageExtent.depth = extent[2];
region.imageOffset.x = offset[0];
region.imageOffset.y = offset[1];
region.imageOffset.z = offset[2];
region.imageSubresource.aspectMask = to_vk_image_aspect_flag_bits(format_);
region.imageSubresource.mipLevel = mip;
region.imageSubresource.layerCount = 1;
VKCommandBuffer &command_buffer = context.command_buffer_get();
command_buffer.copy(*this, staging_buffer, Span<VkBufferImageCopy>(&region, 1));
command_buffer.submit();
}
void VKTexture::update_sub(int /*offset*/[3],
@ -115,6 +193,8 @@ bool VKTexture::init_internal()
/* Initialization can only happen after the usage is known. By the current API this isn't set
* at this moment, so we cannot initialize here. The initialization is postponed until the
* allocation of the texture on the device. */
/* TODO: return false when texture format isn't supported. */
return true;
}
@ -152,12 +232,12 @@ bool VKTexture::allocate()
image_info.format = to_vk_format(format_);
image_info.tiling = VK_IMAGE_TILING_LINEAR;
image_info.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
image_info.usage = VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_SAMPLED_BIT |
VK_IMAGE_USAGE_STORAGE_BIT;
image_info.usage = VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_TRANSFER_DST_BIT |
VK_IMAGE_USAGE_SAMPLED_BIT | VK_IMAGE_USAGE_STORAGE_BIT;
image_info.samples = VK_SAMPLE_COUNT_1_BIT;
VkResult result;
if (G.debug &= G_DEBUG_GPU) {
if (G.debug & G_DEBUG_GPU) {
VkImageFormatProperties image_format = {};
result = vkGetPhysicalDeviceImageFormatProperties(context.physical_device_get(),
image_info.format,