Skip to content

P
Paddle

PaddlePaddle / Paddle
1 年多 前同步成功

通知 2302
Star 20931
Fork 5422
P
Paddle
提交 ec2ba242 编写于 作者: H hedaoyuan
浏览文件
操作

Fix GemmConvFunction.

上级 53b0e427
隐藏空白更改
内联 并排
Showing with 41 addition and 43 deletion
paddle/function/CMakeLists.txt
浏览文件 @ ec2ba242
...@@ -4,6 +4,8 @@ file(GLOB cpp_files . *Op.cpp) ...@@ -4,6 +4,8 @@ file(GLOB cpp_files . *Op.cpp)
list(APPEND h_files Function.h) list(APPEND h_files Function.h)
list(APPEND cpp_files Function.cpp) list(APPEND cpp_files Function.cpp)
list(APPEND cpp_files BufferArg.cpp) list(APPEND cpp_files BufferArg.cpp)
list(APPEND cpp_files GemmFunctor.cpp)
list(APPEND cpp_files EigenGemm.cpp)
if(WITH_GPU) if(WITH_GPU)
file(GLOB cu_files . *OpGpu.cu) file(GLOB cu_files . *OpGpu.cu)
......
paddle/function/GemmConvOp.cpp
浏览文件 @ ec2ba242
...@@ -85,7 +85,6 @@ public: ...@@ -85,7 +85,6 @@ public:
} }
Im2ColFunctor<kCFO, Device, real> im2col; Im2ColFunctor<kCFO, Device, real> im2col;
GemmFunctor<Device, real> gemm;
size_t inputOffset = imShape.getElements(); size_t inputOffset = imShape.getElements();
size_t outputOffset = size_t outputOffset =
(outputChannels / groups_) * outputHeight * outputWidth; (outputChannels / groups_) * outputHeight * outputWidth;
...@@ -108,19 +107,19 @@ public: ...@@ -108,19 +107,19 @@ public:
int M = outputChannels / groups_; int M = outputChannels / groups_;
int N = outputHeight * outputWidth; int N = outputHeight * outputWidth;
int K = inputChannels / groups_ * filterHeight * filterWidth; int K = inputChannels / groups_ * filterHeight * filterWidth;
gemm(CblasNoTrans, BlasGemm<Device, real>::compute(false,
CblasNoTrans, false,
M, M,
N, N,
K, K,
1.0f, 1.0f,
filterData + g * filterOffset, filterData + g * filterOffset,
K, K,
colData, colData,
N, N,
beta, beta,
outputData + g * outputOffset, outputData + g * outputOffset,
N); N);
} }
inputData += inputChannels * inputHeight * inputWidth; inputData += inputChannels * inputHeight * inputWidth;
outputData += outputChannels * outputHeight * outputWidth; outputData += outputChannels * outputHeight * outputWidth;
...@@ -188,8 +187,6 @@ public: ...@@ -188,8 +187,6 @@ public:
} }
Col2ImFunctor<kCFO, Device, real> col2im; Col2ImFunctor<kCFO, Device, real> col2im;
GemmFunctor<Device, real> gemm;
size_t inputOffset = imShape.getElements(); size_t inputOffset = imShape.getElements();
size_t outputOffset = size_t outputOffset =
(outputChannels / groups_) * outputHeight * outputWidth; (outputChannels / groups_) * outputHeight * outputWidth;
...@@ -205,19 +202,19 @@ public: ...@@ -205,19 +202,19 @@ public:
colData = inputGrad + g * inputOffset; colData = inputGrad + g * inputOffset;
scale = 1.0f; scale = 1.0f;
} }
gemm(CblasTrans, BlasGemm<Device, real>::compute(true,
CblasNoTrans, false,
M, M,
N, N,
K, K,
1.0f, 1.0f,
filterData + g * filterOffset, filterData + g * filterOffset,
M, M,
outputGrad + g * outputOffset, outputGrad + g * outputOffset,
N, N,
scale, scale,
colData, colData,
N); N);
if (needIm2col) { if (needIm2col) {
col2im(inputGrad + g * inputOffset, col2im(inputGrad + g * inputOffset,
imShape, imShape,
...@@ -299,7 +296,6 @@ public: ...@@ -299,7 +296,6 @@ public:
} }
Im2ColFunctor<kCFO, Device, real> im2col; Im2ColFunctor<kCFO, Device, real> im2col;
GemmFunctor<Device, real> gemm;
size_t inputOffset = imShape.getElements(); size_t inputOffset = imShape.getElements();
size_t outputOffset = size_t outputOffset =
(outputChannels / groups_) * outputHeight * outputWidth; (outputChannels / groups_) * outputHeight * outputWidth;
...@@ -321,19 +317,19 @@ public: ...@@ -321,19 +317,19 @@ public:
int M = outputChannels / groups_; int M = outputChannels / groups_;
int K = outputHeight * outputWidth; int K = outputHeight * outputWidth;
int N = inputChannels / groups_ * filterHeight * filterWidth; int N = inputChannels / groups_ * filterHeight * filterWidth;
gemm(CblasNoTrans, BlasGemm<Device, real>::compute(false,
CblasTrans, true,
M, M,
N, N,
K, K,
1.0f, 1.0f,
outputGrad + g * outputOffset, outputGrad + g * outputOffset,
K, K,
colData, colData,
K, K,
i == 0 ? beta : 1.0f, i == 0 ? beta : 1.0f,
filterGrad + g * filterOffset, filterGrad + g * filterOffset,
N); N);
} }
inputData += inputChannels * inputHeight * inputWidth; inputData += inputChannels * inputHeight * inputWidth;
outputGrad += outputChannels * outputHeight * outputWidth; outputGrad += outputChannels * outputHeight * outputWidth;
......
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或