public_git
/
whisper.cpp
mirror of https://github.com/ggml-org/whisper.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

vulkan: fix spv shadowing (llama/22760)

This commit is contained in:
miyan 2026-05-08 15:35:22 +08:00 committed by Georgi Gerganov
parent b1aaea6e5a
commit 0d871bd207
No known key found for this signature in database
GPG Key ID: 449E073F9DC10735
1 changed files with 10 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
ggml/src/ggml-vulkan/ggml-vulkan.cpp
View File

@ -2149,11 +2149,11 @@ static void ggml_vk_create_pipeline_func(vk_device& device, vk_pipeline& pipelin
// Patch SPIR-V to enable RTE rounding for FP16, avoiding the need for
// separate shader variants compiled with -DRTE16.
std::vector<uint32_t> spv;
std::vector<uint32_t> spirv;
if (device->float_controls_rte_fp16) {
const uint32_t* spv_words = reinterpret_cast<const uint32_t *>(spv_data);
size_t word_count = spv_size / sizeof(uint32_t);
spv.assign(spv_words, spv_words + word_count);
spirv.assign(spv_words, spv_words + word_count);
// Find insertion points respecting SPIR-V layout order:
// Header(5) -> OpCapability -> OpExtension -> ... -> OpEntryPoint -> OpExecutionMode -> ...
@ -2163,9 +2163,9 @@ static void ggml_vk_create_pipeline_func(vk_device& device, vk_pipeline& pipelin
size_t exec_insert_pos = pos;
uint32_t entry_point_id = 0;
while (pos < spv.size()) {
uint32_t opcode = spv[pos] & spv::OpCodeMask;
uint32_t len = spv[pos] >> spv::WordCountShift;
while (pos < spirv.size()) {
uint32_t opcode = spirv[pos] & spv::OpCodeMask;
uint32_t len = spirv[pos] >> spv::WordCountShift;
if (len == 0) break;
if (opcode == spv::OpCapability) {
@ -2174,7 +2174,7 @@ static void ggml_vk_create_pipeline_func(vk_device& device, vk_pipeline& pipelin
} else if (opcode == spv::OpExtension) {
ext_insert_pos = pos + len;
} else if (opcode == spv::OpEntryPoint) {
entry_point_id = spv[pos + 2];
entry_point_id = spirv[pos + 2];
exec_insert_pos = pos + len;
} else if (opcode == spv::OpExecutionMode || opcode == spv::OpExecutionModeId) {
exec_insert_pos = pos + len;
@ -2189,7 +2189,7 @@ static void ggml_vk_create_pipeline_func(vk_device& device, vk_pipeline& pipelin
// OpExecutionMode %entrypoint RoundingModeRTE 16
uint32_t exec_mode[] = { (4u << spv::WordCountShift) | spv::OpExecutionMode, entry_point_id, spv::ExecutionModeRoundingModeRTE, 16 };
spv.insert(spv.begin() + exec_insert_pos, std::begin(exec_mode), std::end(exec_mode));
spirv.insert(spirv.begin() + exec_insert_pos, std::begin(exec_mode), std::end(exec_mode));
// OpExtension "SPV_KHR_float_controls"
const char ext_str[] = "SPV_KHR_float_controls";
@ -2197,13 +2197,13 @@ static void ggml_vk_create_pipeline_func(vk_device& device, vk_pipeline& pipelin
std::vector<uint32_t> extension(1 + ext_str_words, 0);
extension[0] = (uint32_t)((1 + ext_str_words) << spv::WordCountShift) | spv::OpExtension;
memcpy(&extension[1], ext_str, sizeof(ext_str));
spv.insert(spv.begin() + ext_insert_pos, extension.begin(), extension.end());
spirv.insert(spirv.begin() + ext_insert_pos, extension.begin(), extension.end());
// OpCapability RoundingModeRTE
uint32_t capability[] = { (2u << spv::WordCountShift) | spv::OpCapability, spv::CapabilityRoundingModeRTE };
spv.insert(spv.begin() + cap_insert_pos, std::begin(capability), std::end(capability));
spirv.insert(spirv.begin() + cap_insert_pos, std::begin(capability), std::end(capability));
shader_module_create_info = vk::ShaderModuleCreateInfo({}, spv.size() * sizeof(uint32_t), spv.data());
shader_module_create_info = vk::ShaderModuleCreateInfo({}, spirv.size() * sizeof(uint32_t), spirv.data());
}
pipeline->shader_module = device->device.createShaderModule(shader_module_create_info);