Merge pull request #11989 from tensor-tang/feature/libxsmm

introduce libxsmm

Author: tensor-tang

CMakeLists.txt

@@ -146,6 +146,7 @@ endif()
 ########################################################################################
 
 include(external/mklml)     # download mklml package
+include(external/libxsmm)   # download, build, install libxsmm
 include(external/zlib)      # download, build, install zlib
 include(external/gflags)    # download, build, install gflags
 include(external/glog)      # download, build, install glog
@@ -232,6 +233,10 @@ if(WITH_MKLML)
     list(APPEND EXTERNAL_LIBS ${MKLML_IOMP_LIB})
 endif()
 
+if(WITH_LIBXSMM)
+    list(APPEND EXTERNAL_LIBS ${LIBXSMM_LIBS})
+endif()
+
 if(WITH_MKLDNN)
     list(APPEND EXTERNAL_LIBS ${MKLDNN_LIB})
 endif()

cmake/external/libxsmm.cmake

@@ -0,0 +1,57 @@
+# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+
+OPTION(WITH_LIBXSMM "Compile with libxsmm" OFF)
+
+IF(NOT WITH_LIBXSMM)
+    return()
+ENDIF()
+
+IF(WIN32 OR APPLE OR ANDROID OR IOS)
+    MESSAGE(WARNING "Windows, Mac or Mobile are not supported with libxsmm in Paddle yet.")
+    SET(WITH_LIBXSMM OFF CACHE STRING "Disable LIBXSMM" FORCE)
+    return()
+ENDIF()
+
+INCLUDE (ExternalProject)
+
+SET(LIBXSMM_SOURCES_DIR ${THIRD_PARTY_PATH}/libxsmm)
+SET(LIBXSMM_INSTALL_DIR ${THIRD_PARTY_PATH}/install/libxsmm)
+SET(LIBXSMM_INCLUDE_DIR "${LIBXSMM_INSTALL_DIR}/include" CACHE PATH "LIBXSMM include directory." FORCE)
+SET(LIBXSMM_LIBRARY_DIR "${LIBXSMM_INSTALL_DIR}/lib" CACHE PATH "LIBXSMM library directory." FORCE)
+SET(LIBXSMM_LIBS        "${LIBXSMM_LIBRARY_DIR}/libxsmm.a"
+                        "${LIBXSMM_LIBRARY_DIR}/libxsmmnoblas.a")
+
+ExternalProject_Add(
+    extern_libxsmm
+    GIT_REPOSITORY  "https://github.com/hfp/libxsmm.git"
+    GIT_TAG         "7cc03b5b342fdbc6b6d990b190671c5dbb8489a2"
+    PREFIX          ${LIBXSMM_SOURCES_DIR}
+    UPDATE_COMMAND  ""
+    CONFIGURE_COMMAND ""
+    BUILD_IN_SOURCE 1
+    BUILD_COMMAND   $(MAKE) --silent PREFIX=${LIBXSMM_INSTALL_DIR} CXX=g++ CC=gcc WARP=0 install
+    INSTALL_COMMAND ""
+)
+ADD_LIBRARY(libxsmm STATIC IMPORTED GLOBAL)
+SET_PROPERTY(TARGET libxsmm PROPERTY IMPORTED_LOCATION "${LIBXSMM_LIBRARY_DIR}/libxsmm.a")
+SET_PROPERTY(TARGET libxsmm PROPERTY IMPORTED_LOCATION "${LIBXSMM_LIBRARY_DIR}/libxsmmnoblas.a")
+
+MESSAGE(STATUS "Libxsmm library: ${LIBXSMM_LIBS}")
+include_directories(${LIBXSMM_INCLUDE_DIR})
+ADD_DEFINITIONS(-DPADDLE_WITH_LIBXSMM)
+ADD_DEPENDENCIES(libxsmm extern_libxsmm)
+LIST(APPEND external_project_dependencies libxsmm)
+

cmake/external/openblas.cmake

@@ -121,6 +121,11 @@ ELSE()
   TARGET_LINK_LIBRARIES(cblas ${CBLAS_LIBRARIES})
 ENDIF("${CBLAS_PROVIDER}" STREQUAL "MKLML")
 
+IF(WITH_LIBXSMM)
+  TARGET_LINK_LIBRARIES(cblas ${LIBXSMM_LIBS})
+  ADD_DEPENDENCIES(cblas extern_libxsmm)
+ENDIF()
+
 IF(NOT ${CBLAS_FOUND})
     ADD_DEPENDENCIES(cblas extern_openblas)
     LIST(APPEND external_project_dependencies cblas)

paddle/fluid/operators/math/blas.h

@@ -21,6 +21,10 @@
 #include "paddle/fluid/platform/dynload/mklml.h"
 #endif
 
+#ifdef PADDLE_WITH_LIBXSMM
+#include <libxsmm.h>
+#endif
+
 #ifdef PADDLE_USE_OPENBLAS
 #include <cblas.h>
 #endif

paddle/fluid/operators/math/blas_impl.h

@@ -12,6 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 #pragma once
+#include <limits>
 #include <vector>
 #include "paddle/fluid/operators/math/math_function.h"
 
@@ -30,6 +31,12 @@ struct CBlas<float> {
     platform::dynload::cblas_sgemm(args...);
   }
 
+#ifdef PADDLE_WITH_LIBXSMM
+  template <typename... ARGS>
+  static void SMM_GEMM(ARGS... args) {
+    libxsmm_sgemm(args...);
+  }
+#endif
   template <typename... ARGS>
   static void AXPY(ARGS... args) {
     platform::dynload::cblas_saxpy(args...);
@@ -63,6 +70,12 @@ struct CBlas<double> {
     platform::dynload::cblas_dgemm(args...);
   }
 
+#ifdef PADDLE_WITH_LIBXSMM
+  template <typename... ARGS>
+  static void SMM_GEMM(ARGS... args) {
+    libxsmm_dgemm(args...);
+  }
+#endif
   template <typename... ARGS>
   static void AXPY(ARGS... args) {
     platform::dynload::cblas_daxpy(args...);
@@ -140,13 +153,43 @@ struct CBlas<double> {
 template <>
 struct CBlas<platform::float16> {
   static void GEMM(...) { PADDLE_THROW("float16 GEMM not supported on CPU"); }
+  static void SMM_GEMM(...) {
+    PADDLE_THROW("float16 SMM_GEMM not supported on CPU");
+  }
 #ifdef PADDLE_WITH_MKLML
   static void GEMM_BATCH(...) {
     PADDLE_THROW("float16 GEMM_BATCH not supported on CPU");
   }
 #endif
 };
 
+template <typename T>
+inline bool UseXSMM(const int &m, const int &n, const int &k, bool transa,
+                    bool transb, const T &alpha, const T &beta) {
+#ifdef PADDLE_WITH_LIBXSMM
+  // Refer to https://github.com/hfp/libxsmm/blob/master/README.md
+  // But the threshold is custom
+  constexpr int LIBXSMM_THRESHOLD = 20 * 20 * 20;
+  if (m * n * k > LIBXSMM_THRESHOLD || transa || transb ||
+      std::abs<T>(alpha - static_cast<T>(1) >
+                  std::numeric_limits<T>::epsilon()) ||
+      std::abs<T>(beta) > std::numeric_limits<T>::epsilon()) {
+    return false;
+  } else {
+    return true;
+  }
+#endif
+  return false;
+}
+
+template <>
+inline bool UseXSMM<platform::float16>(const int &m, const int &n, const int &k,
+                                       bool transa, bool transb,
+                                       const platform::float16 &alpha,
+                                       const platform::float16 &beta) {
+  return false;
+}
+
 template <>
 template <typename T>
 void Blas<platform::CPUDeviceContext>::GEMM(CBLAS_TRANSPOSE transA,
@@ -156,8 +199,21 @@ void Blas<platform::CPUDeviceContext>::GEMM(CBLAS_TRANSPOSE transA,
   int lda = (transA == CblasNoTrans) ? K : M;
   int ldb = (transB == CblasNoTrans) ? N : K;
   int ldc = N;
-  CBlas<T>::GEMM(CblasRowMajor, transA, transB, M, N, K, alpha, A, lda, B, ldb,
-                 beta, C, ldc);
+#ifdef PADDLE_WITH_LIBXSMM
+  if (UseXSMM(M, N, K, transA != CblasNoTrans, transB != CblasNoTrans, alpha,
+              beta)) {
+    // Note: SMM use ColMajor
+    const char transa = 'N';
+    const char transb = 'N';
+    CBlas<T>::SMM_GEMM(&transa, &transb, &N, &M, &K, &alpha, B, &ldb, A, &lda,
+                       &beta, C, &ldc);
+  } else {
+#endif
+    CBlas<T>::GEMM(CblasRowMajor, transA, transB, M, N, K, alpha, A, lda, B,
+                   ldb, beta, C, ldc);
+#ifdef PADDLE_WITH_LIBXSMM
+  }
+#endif
 }
 
 template <>

paddle/fluid/operators/math/math_function_test.cc

@@ -54,8 +54,64 @@ TEST(math_function, gemm_notrans_cblas) {
   EXPECT_EQ(input3_ptr[6], 86);
   EXPECT_EQ(input3_ptr[7], 99);
 }
+#ifdef PADDLE_WITH_LIBXSMM
+template <typename T>
+void MklSmmCompare(int m, int n, int k) {
+  paddle::framework::Tensor mat_a;
+  paddle::framework::Tensor mat_b;
+  paddle::framework::Tensor mat_c_smm;
+  paddle::framework::Tensor mat_c_mkl;
+  auto* cpu_place = new paddle::platform::CPUPlace();
+
+  T* A = mat_a.mutable_data<T>({m, k}, *cpu_place);
+  T* B = mat_b.mutable_data<T>({k, n}, *cpu_place);
+  T* CSMM = mat_c_smm.mutable_data<T>({m, n}, *cpu_place);
+  T* CMKL = mat_c_mkl.mutable_data<T>({m, n}, *cpu_place);
+  T alpha = static_cast<T>(1);
+  T beta = static_cast<T>(0);
+  for (int i = 0; i < mat_a.numel(); ++i) {
+    A[i] = static_cast<T>(i);
+  }
+  for (int i = 0; i < mat_b.numel(); ++i) {
+    B[i] = static_cast<T>(i);
+  }
+  // lda,ldb,ldc follow RowMajor
+  int lda = k;
+  int ldb = n;
+  int ldc = n;
+
+  auto smm = [&, m, n, k, lda, ldb, ldc, alpha, beta]() {
+    const char transa = 'N';
+    const char transb = 'N';
+    paddle::operators::math::CBlas<T>::SMM_GEMM(&transa, &transb, &n, &m, &k,
+                                                &alpha, B, &ldb, A, &lda, &beta,
+                                                CSMM, &ldc);
+  };
+
+  auto mkl = [&, m, n, k, lda, ldb, ldc, alpha, beta]() {
+    paddle::operators::math::CBlas<T>::GEMM(CblasRowMajor, CblasNoTrans,
+                                            CblasNoTrans, m, n, k, alpha, A,
+                                            lda, B, ldb, beta, CMKL, ldc);
+  };
+
+  smm();
+  mkl();
+  ASSERT_EQ(mat_c_mkl.numel(), mat_c_smm.numel());
+  for (int i = 0; i < mat_c_mkl.numel(); ++i) {
+    EXPECT_FLOAT_EQ(CSMM[i], CMKL[i]);
+  }
+}
+TEST(math_function, gemm_mkl_vs_smm) {
+  MklSmmCompare<float>(1, 2, 3);
+  MklSmmCompare<double>(1, 2, 3);
+  MklSmmCompare<float>(3, 2, 1);
+  MklSmmCompare<double>(3, 2, 1);
+  MklSmmCompare<float>(3, 8, 5);
+  MklSmmCompare<double>(3, 8, 5);
+}
+#endif
 
-TEST(math_function, gemm_trans_clbas) {
+TEST(math_function, gemm_trans_cblas) {
   paddle::framework::Tensor input1;
   paddle::framework::Tensor input2;
   paddle::framework::Tensor input3;