Add PADDLE_USE_EIGEN_FOR_BLAS macro.

CMakeLists.txt

@@ -55,6 +55,7 @@ option(WITH_C_API       "Compile PaddlePaddle with C-API(Prediction)"   OFF)
 option(WITH_GOLANG      "Compile PaddlePaddle with GOLANG"              OFF)
 option(GLIDE_INSTALL    "Download and install go dependencies "         ON)
 option(USE_NNPACK       "Compile PaddlePaddle with NNPACK library"      OFF)
+option(USE_EIGEN_FOR_BLAS   "Use matrix multiplication in Eigen"        OFF)
 
 # CMAKE_BUILD_TYPE
 if(NOT CMAKE_BUILD_TYPE)

cmake/configure.cmake

@@ -28,6 +28,10 @@ if(NOT WITH_TIMER)
     add_definitions(-DPADDLE_DISABLE_TIMER)
 endif(NOT WITH_TIMER)
 
+if(USE_EIGEN_FOR_BLAS)
+    add_definitions(-DPADDLE_USE_EIGEN_FOR_BLAS)
+endif(USE_EIGEN_FOR_BLAS)
+
 if(NOT WITH_PROFILER)
     add_definitions(-DPADDLE_DISABLE_PROFILER)
 endif(NOT WITH_PROFILER)

paddle/function/GemmFunctor.cpp

@@ -32,6 +32,10 @@ struct BlasGemm<DEVICE_TYPE_CPU, T> {
                       const T beta,
                       T* C,
                       const int ldc) {
+#ifdef PADDLE_USE_EIGEN_FOR_BLAS
+    EigenBlasGemm<T>::compute(
+        transA, transB, M, N, K, alpha, A, lda, B, ldb, beta, C, ldc);
+#else
     gemm<T>(transA == false ? CblasNoTrans : CblasTrans,
             transB == false ? CblasNoTrans : CblasTrans,
             M,
@@ -45,6 +49,7 @@ struct BlasGemm<DEVICE_TYPE_CPU, T> {
             beta,
             C,
             ldc);
+#endif
   }
 };