root
/
MNN
mirror of https://github.com/alibaba/MNN.git
1
0
Fork 0
Code Issues Actions 1 Packages Projects Releases Wiki Activity

Vulkan:Bugfix: Fix bug for windows compile error

This commit is contained in:
xiaying 2025-02-12 11:56:52 +08:00
parent aee17faedf
commit ee9fff2d6c
1 changed files with 3 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
source/backend/vulkan/image/backend/VulkanBackend.cpp
View File

@ -541,6 +541,9 @@ float VulkanBackend::getPipelineTime(const VulkanPipeline* pipeline, std::shared
return time; return time;
} }
static bool checkInvalid(uint32_t x, uint32_t y, uint32_t z, uint32_t subgroupSize) {
return x * y * z > 4 * subgroupSize || x > 128 || y > 128 || z > 128;
}
std::vector<uint32_t> VulkanBackend::autoTunePipeline(SharedPtr<VulkanPipeline> pipeline, std::shared_ptr<VulkanLayout::DescriptorSet> des, const std::vector<uint32_t> gws, const uint32_t tuneDimension, std::vector<uint32_t> defaultLws,float * const minCostPtr) { std::vector<uint32_t> VulkanBackend::autoTunePipeline(SharedPtr<VulkanPipeline> pipeline, std::shared_ptr<VulkanLayout::DescriptorSet> des, const std::vector<uint32_t> gws, const uint32_t tuneDimension, std::vector<uint32_t> defaultLws,float * const minCostPtr) {
bool isPrivate = !(pipeline->mTuneName.empty()); bool isPrivate = !(pipeline->mTuneName.empty());
@ -585,10 +588,6 @@ std::vector<uint32_t> VulkanBackend::autoTunePipeline(SharedPtr<VulkanPipeline>
bool tuneNormalFlag = (mRuntime->mGpuMode & MNNGpuMode::MNN_GPU_TUNING_WIDE); bool tuneNormalFlag = (mRuntime->mGpuMode & MNNGpuMode::MNN_GPU_TUNING_WIDE);
auto checkInvalid = tuneNormalFlag
? [](uint32_t x, uint32_t y, uint32_t z, uint32_t subgroupSize) -> bool { return x * y * z > 4 * subgroupSize || x > 128 || y > 128 || z > 128 ; } // MNN_GPU_TUNING_WIDE
: [](uint32_t x, uint32_t y, uint32_t z, uint32_t subgroupSize) -> bool { return x * y * z > 16 * subgroupSize; }; // MNN_GPU_TUNING_HEAVY
for (uint32_t z = localSizeRangeZ.first; z <= localSizeRangeZ.second; z = z << 1) { for (uint32_t z = localSizeRangeZ.first; z <= localSizeRangeZ.second; z = z << 1) {
for (uint32_t y = localSizeRangeY.first; y <= localSizeRangeY.second; y = y << 1) { for (uint32_t y = localSizeRangeY.first; y <= localSizeRangeY.second; y = y << 1) {
for (uint32_t x = localSizeRangeX.first; x <= localSizeRangeX.second; x = x << 1) { for (uint32_t x = localSizeRangeX.first; x <= localSizeRangeX.second; x = x << 1) {