From ec2ba242060fc10b2045533fdcb410cfbd473cec Mon Sep 17 00:00:00 2001
From: hedaoyuan <hedaoyuan@github.com>
Date: Thu, 17 Aug 2017 16:22:30 +0800
Subject: [PATCH] Fix GemmConvFunction.

---
 paddle/function/CMakeLists.txt |  2 +
 paddle/function/GemmConvOp.cpp | 82 ++++++++++++++++------------------
 2 files changed, 41 insertions(+), 43 deletions(-)

diff --git a/paddle/function/CMakeLists.txt b/paddle/function/CMakeLists.txt
index 7dfb6f61c50..9187294a49c 100644
--- a/paddle/function/CMakeLists.txt
+++ b/paddle/function/CMakeLists.txt
@@ -4,6 +4,8 @@ file(GLOB cpp_files . *Op.cpp)
 list(APPEND h_files Function.h)
 list(APPEND cpp_files Function.cpp)
 list(APPEND cpp_files BufferArg.cpp)
+list(APPEND cpp_files GemmFunctor.cpp)
+list(APPEND cpp_files EigenGemm.cpp)
 
 if(WITH_GPU)
     file(GLOB cu_files . *OpGpu.cu)
diff --git a/paddle/function/GemmConvOp.cpp b/paddle/function/GemmConvOp.cpp
index 0ada4d70a0c..f8cf4ebea8d 100644
--- a/paddle/function/GemmConvOp.cpp
+++ b/paddle/function/GemmConvOp.cpp
@@ -85,7 +85,6 @@ public:
     }
 
     Im2ColFunctor<kCFO, Device, real> im2col;
-    GemmFunctor<Device, real> gemm;
     size_t inputOffset = imShape.getElements();
     size_t outputOffset =
         (outputChannels / groups_) * outputHeight * outputWidth;
@@ -108,19 +107,19 @@ public:
         int M = outputChannels / groups_;
         int N = outputHeight * outputWidth;
         int K = inputChannels / groups_ * filterHeight * filterWidth;
-        gemm(CblasNoTrans,
-             CblasNoTrans,
-             M,
-             N,
-             K,
-             1.0f,
-             filterData + g * filterOffset,
-             K,
-             colData,
-             N,
-             beta,
-             outputData + g * outputOffset,
-             N);
+        BlasGemm<Device, real>::compute(false,
+                                        false,
+                                        M,
+                                        N,
+                                        K,
+                                        1.0f,
+                                        filterData + g * filterOffset,
+                                        K,
+                                        colData,
+                                        N,
+                                        beta,
+                                        outputData + g * outputOffset,
+                                        N);
       }
       inputData += inputChannels * inputHeight * inputWidth;
       outputData += outputChannels * outputHeight * outputWidth;
@@ -188,8 +187,6 @@ public:
     }
 
     Col2ImFunctor<kCFO, Device, real> col2im;
-    GemmFunctor<Device, real> gemm;
-
     size_t inputOffset = imShape.getElements();
     size_t outputOffset =
         (outputChannels / groups_) * outputHeight * outputWidth;
@@ -205,19 +202,19 @@ public:
           colData = inputGrad + g * inputOffset;
           scale = 1.0f;
         }
-        gemm(CblasTrans,
-             CblasNoTrans,
-             M,
-             N,
-             K,
-             1.0f,
-             filterData + g * filterOffset,
-             M,
-             outputGrad + g * outputOffset,
-             N,
-             scale,
-             colData,
-             N);
+        BlasGemm<Device, real>::compute(true,
+                                        false,
+                                        M,
+                                        N,
+                                        K,
+                                        1.0f,
+                                        filterData + g * filterOffset,
+                                        M,
+                                        outputGrad + g * outputOffset,
+                                        N,
+                                        scale,
+                                        colData,
+                                        N);
         if (needIm2col) {
           col2im(inputGrad + g * inputOffset,
                  imShape,
@@ -299,7 +296,6 @@ public:
     }
 
     Im2ColFunctor<kCFO, Device, real> im2col;
-    GemmFunctor<Device, real> gemm;
     size_t inputOffset = imShape.getElements();
     size_t outputOffset =
         (outputChannels / groups_) * outputHeight * outputWidth;
@@ -321,19 +317,19 @@ public:
         int M = outputChannels / groups_;
         int K = outputHeight * outputWidth;
         int N = inputChannels / groups_ * filterHeight * filterWidth;
-        gemm(CblasNoTrans,
-             CblasTrans,
-             M,
-             N,
-             K,
-             1.0f,
-             outputGrad + g * outputOffset,
-             K,
-             colData,
-             K,
-             i == 0 ? beta : 1.0f,
-             filterGrad + g * filterOffset,
-             N);
+        BlasGemm<Device, real>::compute(false,
+                                        true,
+                                        M,
+                                        N,
+                                        K,
+                                        1.0f,
+                                        outputGrad + g * outputOffset,
+                                        K,
+                                        colData,
+                                        K,
+                                        i == 0 ? beta : 1.0f,
+                                        filterGrad + g * filterOffset,
+                                        N);
       }
       inputData += inputChannels * inputHeight * inputWidth;
       outputGrad += outputChannels * outputHeight * outputWidth;
-- 
GitLab