CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-31 08:51:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

vulkan: compile a test shader in cmake to check for coopmat2 support (#10713)

Browse Source
This commit is contained in:
Jeff Bolz
2024-12-08 02:05:55 -06:00
committed by GitHub
parent 62e84d9848
commit ecc93d0558
4 changed files with 36 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
ggml/src/ggml-vulkan/CMakeLists.txt
View File

@@ -8,6 +8,20 @@ if (Vulkan_FOUND)
../../include/ggml-vulkan.h
)
# Compile a test shader to determine whether GL_NV_cooperative_matrix2 is supported.
# If it's not, there will be an error to stderr.
# If it's supported, set a define to indicate that we should compile those shaders
execute_process(COMMAND ${Vulkan_GLSLC_EXECUTABLE} -o - -fshader-stage=compute --target-env=vulkan1.3 "${CMAKE_CURRENT_SOURCE_DIR}/vulkan-shaders/test_coopmat2_support.comp"
OUTPUT_VARIABLE glslc_output
ERROR_VARIABLE glslc_error)
if (${glslc_error} MATCHES ".*extension not supported: GL_NV_cooperative_matrix2.*")
message(STATUS "GL_NV_cooperative_matrix2 not supported by glslc")
else()
message(STATUS "GL_NV_cooperative_matrix2 supported by glslc")
add_compile_definitions(GGML_VULKAN_COOPMAT2_GLSLC_SUPPORT)
endif()
target_link_libraries(ggml-vulkan PRIVATE Vulkan::Vulkan)
target_include_directories(ggml-vulkan PRIVATE ${CMAKE_CURRENT_BINARY_DIR})

19
ggml/src/ggml-vulkan/ggml-vulkan.cpp
View File

@@ -1513,7 +1513,7 @@ static void ggml_vk_load_shaders(vk_device& device) {
compiles.push_back(std::async(ggml_vk_create_pipeline_func, std::ref(device), std::ref(pipeline), name, spv_size, spv_data, entrypoint, parameter_count, push_constant_size, wg_denoms, specialization_constants, align, disable_robustness));
};
#if defined(VK_NV_cooperative_matrix2)
#if defined(VK_NV_cooperative_matrix2) && defined(GGML_VULKAN_COOPMAT2_GLSLC_SUPPORT)
if (device->coopmat2) {
auto const &fa_wg_denoms = [&](uint32_t D, uint32_t clamp, ggml_type type, bool small_rows) -> std::array<uint32_t, 3> {
@@ -1611,7 +1611,7 @@ static void ggml_vk_load_shaders(vk_device& device) {
#undef CREATE_MM
#undef CREATE_MM2
} else
#endif // defined(VK_NV_cooperative_matrix2)
#endif // defined(VK_NV_cooperative_matrix2) && defined(GGML_VULKAN_COOPMAT2_GLSLC_SUPPORT)
if (device->coopmat_support) {
// Create 6 variants, {s,m,l}x{unaligned,aligned}
#define CREATE_MM(PIPELINE_NAME, NAMELC, F16ACC, WG_DENOMS, WARPTILE, PUSHCONST, PARAMCOUNT, ID) \
@@ -2153,7 +2153,7 @@ static vk_device ggml_vk_get_device(size_t idx) {
device->coopmat_support = device->coopmat_support && coopmat_features.cooperativeMatrix;
if (coopmat2_support) {
#if defined(VK_NV_cooperative_matrix2)
#if defined(VK_NV_cooperative_matrix2) && defined(GGML_VULKAN_COOPMAT2_GLSLC_SUPPORT)
if (coopmat2_features.cooperativeMatrixWorkgroupScope &&
coopmat2_features.cooperativeMatrixFlexibleDimensions &&
coopmat2_features.cooperativeMatrixReductions &&
@@ -2414,14 +2414,19 @@ static void ggml_vk_print_gpu_info(size_t idx) {
bool fp16_storage = false;
bool fp16_compute = false;
bool coopmat_support = false;
bool coopmat2_support = false;
for (auto properties : ext_props) {
if (strcmp("VK_KHR_16bit_storage", properties.extensionName) == 0) {
fp16_storage = true;
} else if (strcmp("VK_KHR_shader_float16_int8", properties.extensionName) == 0) {
fp16_compute = true;
} else if (strcmp("VK_KHR_cooperative_matrix", properties.extensionName) == 0) {
} else if (strcmp("VK_KHR_cooperative_matrix", properties.extensionName) == 0 &&
!getenv("GGML_VK_DISABLE_COOPMAT")) {
coopmat_support = true;
} else if (strcmp("VK_NV_cooperative_matrix2", properties.extensionName) == 0 &&
!getenv("GGML_VK_DISABLE_COOPMAT2")) {
coopmat2_support = true;
}
}
@@ -2472,9 +2477,11 @@ static void ggml_vk_print_gpu_info(size_t idx) {
coopmat_support = coopmat_support && coopmat_features.cooperativeMatrix;
std::string matrix_cores = coopmat2_support ? "NV_coopmat2" : coopmat_support ? "KHR_coopmat" : "none";
std::string device_name = props2.properties.deviceName.data();
GGML_LOG_DEBUG("ggml_vulkan: %zu = %s (%s) | uma: %d | fp16: %d | warp size: %zu | matrix cores: %d\n",
idx, device_name.c_str(), driver_props.driverName.data(), uma, fp16, subgroup_size, coopmat_support);
GGML_LOG_DEBUG("ggml_vulkan: %zu = %s (%s) | uma: %d | fp16: %d | warp size: %zu | matrix cores: %s\n",
idx, device_name.c_str(), driver_props.driverName.data(), uma, fp16, subgroup_size, matrix_cores.c_str());
if (props2.properties.deviceType == vk::PhysicalDeviceType::eCpu) {
GGML_LOG_DEBUG("ggml_vulkan: Warning: Device type is CPU. This is probably not the device you want.\n");

7
ggml/src/ggml-vulkan/vulkan-shaders/test_coopmat2_support.comp Normal file
View File

@@ -0,0 +1,7 @@
#version 460
#extension GL_NV_cooperative_matrix2 : require
void main()
{
}

4
ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp
View File

@@ -342,14 +342,14 @@ void process_shaders() {
matmul_shaders(true, matmul_id, true, false, false);
matmul_shaders(true, matmul_id, true, false, true);
#if defined(VK_NV_cooperative_matrix2)
#if defined(GGML_VULKAN_COOPMAT2_GLSLC_SUPPORT)
// Coopmat2, fp32acc and fp16acc
matmul_shaders(true, matmul_id, false, true, false);
matmul_shaders(true, matmul_id, false, true, true);
#endif
}
#if defined(VK_NV_cooperative_matrix2)
#if defined(GGML_VULKAN_COOPMAT2_GLSLC_SUPPORT)
// flash attention
for (const auto& f16acc : {false, true}) {
std::string acctype = f16acc ? "float16_t" : "float";