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

Merge pull request #3770 from SixtyWang/kai/fix-bugs
android / android_build (push) Has been cancelled Details
ios / ios_build (push) Has been cancelled Details
linux / linux_buil_test (push) Has been cancelled Details
macos / macos_buil_test (push) Has been cancelled Details
windows / windows_build_test (push) Has been cancelled Details 

Fix bug in KleidiAI Code
This commit is contained in:
jxt1234 2025-08-01 11:15:24 +08:00 committed by GitHub
parent ba68171bbe 380370fcf8
commit 7a92132554
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
5 changed files with 12 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
source/backend/cpu/arm/CMakeLists.txt
View File

@ -103,7 +103,7 @@ if (MNN_KLEIDIAI)
${KLEIDIAI_SRC}/kai/ukernels/matmul/pack/kai_rhs_imatmul_pack_kxn_x16p2vlx2b_x16_x16_sme.c
)
set_source_files_properties(${MNN_SOURCES_KLEIDIAI} PROPERTIES COMPILE_OPTIONS -march=armv8.2-a+i8mm+dotprod+sve+sve2+fp16)
set_source_files_properties(${MNN_SOURCES_KLEIDIAI} PROPERTIES COMPILE_OPTIONS "-fno-tree-vectorize;-march=armv8.2-a+i8mm+dotprod+sve+sve2+fp16")
set_source_files_properties(${KLEIDIAI_FILES_SME2} PROPERTIES COMPILE_OPTIONS "-fno-tree-vectorize;-march=armv8.2-a+sve+sve2")
endif()

5
source/backend/cpu/arm/mnn_kleidiai.h
View File

@ -211,6 +211,11 @@ namespace MNN {
return false;
}
}
if(type == AccelType::QI4_SYM_CHNLQT_F32){
if(common->inputCount() % 2 != 0) {
return false;
}
}
if(common->kernelX() == 1 && common->kernelY() == 1
&& common->padX() == 0 && common->padY() == 0
&& common->strideX() == 1 && common->strideY() == 1

6
source/backend/cpu/compute/KleidiAIConvolution.cpp
View File

@ -175,8 +175,11 @@ ErrorCode KleidiAIConvolution::onResize(const std::vector<Tensor *> &inputs, con
if (outputOriginFmt != MNN_DATA_FORMAT_NHWC){
b->onReleaseBuffer(mOutputConvertBuffer.get(), Backend::DYNAMIC);
}
mPostParameters = getPostParameters();
return NO_ERROR;
}
ErrorCode KleidiAIConvolution::onExecute(const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs) {
auto input = inputs[0];
auto output = outputs[0];
@ -209,13 +212,12 @@ ErrorCode KleidiAIConvolution::onExecute(const std::vector<Tensor *> &inputs, co
}
auto outputDes = TensorUtils::getDescribe(outputs[0]);
auto postPtr = getPostParameters();
auto outputPtr = output->host<uint8_t>();
if(outputDes->dimensionFormat != MNN_DATA_FORMAT_NHWC){
outputPtr = mOutputConvertBuffer->host<uint8_t>();
}
kai.runMatmul(mAccelType, m, n, k, 0, lhsPacked, weightPtr, outputPtr, n * elementSize, elementSize, postPtr[3], postPtr[2]);
kai.runMatmul(mAccelType, m, n, k, 0, lhsPacked, weightPtr, outputPtr, n * elementSize, elementSize, mPostParameters[3], mPostParameters[2]);
if(outputDes->dimensionFormat != MNN_DATA_FORMAT_NHWC){
MNN_CONCURRENCY_BEGIN(tId, threadNum) {

2
source/backend/cpu/compute/KleidiAIConvolution.hpp
View File

@ -28,7 +28,7 @@ class KleidiAIConvolution : public CPUConvolution{
std::shared_ptr<Tensor> mOutputConvertBuffer;
std::shared_ptr<CPUConvolution::Resource> mResource;
KleidiAI::AccelType mAccelType = KleidiAI::AccelType::ACC_TYPE_NUMBER;
std::vector<float> mPostParameters;
};
#endif //MNN_KLEIDIAI_ENABLED

2
source/backend/cpu/compute/KleidiAIDenseConvolution.hpp
View File

@ -37,7 +37,7 @@ struct ConvParams {
int dilatedKernelWidth = kernelSizeWithDilated(kernelWidth, dilatedWidth);
int outputHeight = outputSize(inputHeight, padTop, padBottom, dilatedKernelHeight, strideHeight);
int outputWidth = outputSize(inputHeight, padLeft, padRight, dilatedKernelWidth, strideWidth);
int outputWidth = outputSize(inputWidth, padLeft, padRight, dilatedKernelWidth, strideWidth);
return {outputHeight, outputWidth};
}