Merge branch 'develop' into develop

CMakeLists.txt

@@ -2,7 +2,7 @@ cmake_minimum_required(VERSION 3.0)
 project(paddle-mobile)
 
 option(DEBUGING "enable debug mode" ON)
-option(USE_OPENMP "openmp support" OFF)
+option(USE_OPENMP "openmp support" ON)
 option(USE_EXCEPTION "use std exception" ON)
 option(LOG_PROFILE "log profile" ON)
 # select the platform to build

src/io/executor.cpp

@@ -13,6 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "io/executor.h"
+#include <operators/math/gemm.h>
 #include <algorithm>
 #include <vector>
 #include "common/enforce.h"
@@ -25,6 +26,9 @@ limitations under the License. */
 #include "framework/program/var_desc.h"
 #include "framework/scope.h"
 #include "framework/tensor.h"
+#ifdef _OPENMP
+#include <omp.h>
+#endif  // _OPENMP
 #ifdef PADDLE_EXECUTOR_MULTITHREAD
 #include <queue>
 #include <utility>
@@ -403,6 +407,17 @@ std::vector<typename Executor<Dtype, P>::Ptype> Executor<Dtype, P>::Predict(
   return result_vector;
 }
 
+template <typename Dtype, Precision P>
+void Executor<Dtype, P>::SetThreadNum(int num) {
+  for (int k = 0; k < std::max(num, 3); ++k) {
+    operators::math::Gemmer::gemmers.push_back(new operators::math::Gemmer());
+  }
+#ifdef _OPENMP
+  //  omp_set_dynamic(0);
+  omp_set_num_threads(num);
+#endif
+}
+
 template class Executor<CPU, Precision::FP32>;
 template class Executor<FPGA, Precision::FP32>;
 template class Executor<GPU_MALI, Precision::FP32>;

src/io/executor.h

@@ -58,6 +58,8 @@ class Executor {
   std::vector<Ptype> Predict(const std::vector<Ptype> &input,
                              const std::vector<int64_t> &dims);
 
+  void SetThreadNum(int num);
+
  protected:
   Executor() = default;
   void InitMemory();

src/operators/kernel/central-arm-func/conv_add_arm_func.h

@@ -14,10 +14,14 @@ limitations under the License. */
 
 #ifdef FUSION_CONVADD_OP
 #pragma once
+#if _OPENMP
+#include <omp.h>
+#endif
 
 #include <vector>
 #include "operators/math/conv_func.h"
 #include "operators/math/depthwise_conv_3x3.h"
+#include "operators/math/gemm.h"
 #include "operators/math/im2col.h"
 #include "operators/math/math_function.h"
 #include "operators/math/vol2col.h"
@@ -106,9 +110,33 @@ void ConvAddBasic(const FusionConvAddParam &param) {
       // gemm
       Tensor out_slice = out_batch.Slice(g * out_step, (g + 1) * out_step);
       Tensor filter_slice = filter.Slice(g * out_step, (g + 1) * out_step);
-      math::matmul<float>(filter_slice, false, col_matrix, false,
-                          static_cast<float>(1), &out_slice,
-                          static_cast<float>(1));
+
+      auto dim_a = filter_slice.dims();
+      auto dim_b = col_matrix.dims();
+
+      auto dim_out = out_slice.dims();
+
+      int m = dim_out[0];
+      int n = dim_out[1];
+      int k = dim_a[1];
+
+      float *output_data = out_slice.data<float>();
+      int thread_num = 4;
+      int m1 = m / thread_num;
+      int m2 = m % thread_num;
+#pragma omp parallel for
+      for (int j = 0; j < thread_num; ++j) {
+        int row_count = m1;
+        if (j == thread_num - 1) {
+          row_count = m1 + m2;
+        }
+        math::Gemmer::gemmers[j]->Sgemm(
+            row_count, n, k, 1, filter_slice.data<float>() + j * m1 * k, k,
+            col_matrix.data<float>(), n, 1, output_data + j * m1 * n, n, false);
+      }
+      //        math::matmul<float>(filter_slice, false, col_matrix, false,
+      //                            static_cast<float>(1), &out_slice,
+      //                            static_cast<float>(1));
     }
   }
 }

src/operators/kernel/lrn_kernel.h

@@ -13,7 +13,9 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #ifdef LRN_OP
-
+#ifdef _OPENMP
+#include <omp.h>
+#endif
 #include "framework/operator.h"
 #include "operators/op_param.h"
 
@@ -47,6 +49,7 @@ struct LRNFunctor {
     std::fill(sqr_buffer_ptr, sqr_buffer_ptr + sqr_buffer.numel(), 0.0);
 
     for (int a = 0; a < N; a++) {
+#pragma parallel for
       for (int b = 0; b < C; b++) {
         for (int index = start; index < end; index++) {
           int channel = b + index;

src/operators/math/gemm.cpp

@@ -22,17 +22,11 @@ limitations under the License. */
 namespace paddle_mobile {
 namespace operators {
 namespace math {
-int MC = 0;
-int KC = 0;
-int NC = 0;
-
-float *packedA;
-float *packedB;
-float *packedC;
-float *zero;
+
+std::vector<Gemmer *> Gemmer::gemmers;
 // 将A矩阵分块复制到连续内存(ColMajor)
-void PackMatrixA(int m, int k, int m_tail, const float *A, int lda,
-                 float *buffer) {
+void Gemmer::PackMatrixA(int m, int k, int m_tail, const float *A, int lda,
+                         float *buffer) {
   int i, j;
   const float *Aij;
   for (i = 0; i < m - m_tail; i += MR) {
@@ -58,8 +52,8 @@ void PackMatrixA(int m, int k, int m_tail, const float *A, int lda,
 }
 
 // 将A矩阵分块复制到连续内存(RowMajor)
-void PackMatrixA_(int m, int k, int m_tail, const float *A, int lda,
-                  float *buffer) {
+void Gemmer::PackMatrixA_(int m, int k, int m_tail, const float *A, int lda,
+                          float *buffer) {
   const float *a0, *a1, *a2, *a3;
   for (int i = 0; i < m - m_tail; i += MR) {
     a0 = A + i * lda;
@@ -98,8 +92,8 @@ void PackMatrixA_(int m, int k, int m_tail, const float *A, int lda,
 }
 
 // 将B矩阵分块复制到连续内存(ColMajor)
-void PackMatrixB(int k, int n, int n_tail, const float *B, int ldb,
-                 float *buffer) {
+void Gemmer::PackMatrixB(int k, int n, int n_tail, const float *B, int ldb,
+                         float *buffer) {
   int i, j;
   const float *Bj, *Bj1, *Bj2, *Bj3;
   for (j = 0; j < n - n_tail; j += NR) {
@@ -127,8 +121,8 @@ void PackMatrixB(int k, int n, int n_tail, const float *B, int ldb,
 }
 
 // 将B矩阵分块复制到连续内存(RowMajor)
-void PackMatrixB_(int k, int n, int n_tail, const float *B, int ldb,
-                  float *buffer) {
+void Gemmer::PackMatrixB_(int k, int n, int n_tail, const float *B, int ldb,
+                          float *buffer) {
   const float *b0;
   for (int j = 0; j < n - n_tail; j += NR) {
     for (int i = 0; i < k; ++i) {
@@ -156,8 +150,9 @@ void PackMatrixB_(int k, int n, int n_tail, const float *B, int ldb,
 }
 
 // 分块矩阵乘法
-void InnerKernel(int mc, int nc, float alpha, const float *a, const float *b,
-                 float beta, float *c, float *C, int ldc, bool relu) {
+void Gemmer::InnerKernel(int mc, int nc, float alpha, const float *a,
+                         const float *b, float beta, float *c, float *C,
+                         int ldc, bool relu) {
   for (int j = 0; j < nc; j += NR) {
     for (int i = 0; i < mc; i += MR) {
       // AddDot4x4(KC, a + i * KC, b + j * KC, c + i * NC + j, NC);
@@ -184,9 +179,10 @@ void InnerKernel(int mc, int nc, float alpha, const float *a, const float *b,
 }
 
 // 分块矩阵乘法
-void InnerKernelWithBn(int mc, int nc, float alpha, const float *a,
-                       const float *b, float beta, float *c, float *C, int ldc,
-                       bool relu, float *new_scale, float *new_bias) {
+void Gemmer::InnerKernelWithBn(int mc, int nc, float alpha, const float *a,
+                               const float *b, float beta, float *c, float *C,
+                               int ldc, bool relu, float *new_scale,
+                               float *new_bias) {
   for (int j = 0; j < nc; j += NR) {
     for (int i = 0; i < mc; i += MR) {
       // AddDot4x4(KC, a + i * KC, b + j * KC, c + i * NC + j, NC);
@@ -202,7 +198,8 @@ void InnerKernelWithBn(int mc, int nc, float alpha, const float *a,
 }
 
 #if defined(IOS)
-void AddDot4x4(int k, const float *a, const float *b, float *C, int ldc) {
+void Gemmer::AddDot4x4(int k, const float *a, const float *b, float *C,
+                       int ldc) {
   // init C
   float32x4_t cv0 = vdupq_n_f32(0.0);
   float32x4_t cv1 = vdupq_n_f32(0.0);
@@ -253,7 +250,8 @@ void AddDot4x4(int k, const float *a, const float *b, float *C, int ldc) {
 }  // namespace math
 
 #elif defined(ARMV7)
-void AddDot4x4(int k, const float *a, const float *b, float *c, int ldc) {
+void Gemmer::AddDot4x4(int k, const float *a, const float *b, float *c,
+                       int ldc) {
   const float *a_ptr, *b_ptr;
   a_ptr = a;
   b_ptr = b;
@@ -324,7 +322,8 @@ void AddDot4x4(int k, const float *a, const float *b, float *c, int ldc) {
 }
 
 #else
-void AddDot4x4(int k, const float *a, const float *b, float *c, int ldc) {
+void Gemmer::AddDot4x4(int k, const float *a, const float *b, float *c,
+                       int ldc) {
   float *c0, *c1, *c2, *c3;
   c0 = c;
   c1 = c + ldc;
@@ -363,8 +362,9 @@ void AddDot4x4(int k, const float *a, const float *b, float *c, int ldc) {
 #endif
 
 // 32位 float 矩阵乘法
-void Sgemm(int m, int n, int k, float alpha, const float *A, int lda,
-           const float *B, int ldb, float beta, float *C, int ldc, bool relu) {
+void Gemmer::Sgemm(int m, int n, int k, float alpha, const float *A, int lda,
+                   const float *B, int ldb, float beta, float *C, int ldc,
+                   bool relu) {
   // L1 data cache is 32 kib (Per Contex-A57, Contex-A72, Contex-A73)
   // L2 cache is 0.5~4 Mib (Contex-A72 cluster)
   int L1 = 30 * 1024;
@@ -415,9 +415,10 @@ void Sgemm(int m, int n, int k, float alpha, const float *A, int lda,
   paddle_mobile::memory::Free(zero);
 }
 
-void SgemmWithBn(int m, int n, int k, float alpha, const float *A, int lda,
-                 const float *B, int ldb, float beta, float *C, int ldc,
-                 bool relu, float *new_scale, float *new_bias) {
+void Gemmer::SgemmWithBn(int m, int n, int k, float alpha, const float *A,
+                         int lda, const float *B, int ldb, float beta, float *C,
+                         int ldc, bool relu, float *new_scale,
+                         float *new_bias) {
   // L1 data cache is 32 kib (Per Contex-A57, Contex-A72, Contex-A73)
   // L2 cache is 0.5~4 Mib (Contex-A72 cluster)
   int L1 = 30 * 1024;
@@ -468,9 +469,9 @@ void SgemmWithBn(int m, int n, int k, float alpha, const float *A, int lda,
   paddle_mobile::memory::Free(zero);
 }
 
-void VectorKernel(int m, int n, int k, float alpha, const float *A, int lda,
-                  const float *B, int ldb, float beta, float *C, int ldc,
-                  bool relu) {
+void Gemmer::VectorKernel(int m, int n, int k, float alpha, const float *A,
+                          int lda, const float *B, int ldb, float beta,
+                          float *C, int ldc, bool relu) {
   float *bufferC = static_cast<float *>(memory::Alloc(sizeof(float) * n));
 
   const float *a0, *b0, *b1, *b2, *b3;
@@ -690,9 +691,10 @@ void VectorKernel(int m, int n, int k, float alpha, const float *A, int lda,
   }
 }
 
-void VectorKernelWithBn(int m, int n, int k, float alpha, const float *A,
-                        int lda, const float *B, int ldb, float beta, float *C,
-                        int ldc, bool relu, float *new_scale, float *new_bias) {
+void Gemmer::VectorKernelWithBn(int m, int n, int k, float alpha,
+                                const float *A, int lda, const float *B,
+                                int ldb, float beta, float *C, int ldc,
+                                bool relu, float *new_scale, float *new_bias) {
   float *bufferC = static_cast<float *>(memory::Alloc(sizeof(float) * n));
 
   const float *a0, *b0, *b1, *b2, *b3;
@@ -901,7 +903,8 @@ void VectorKernelWithBn(int m, int n, int k, float alpha, const float *A,
   }
 }
 
-void AddDot4x8(int k, const float *a, const float *b, float *c, int ldc) {
+void Gemmer::AddDot4x8(int k, const float *a, const float *b, float *c,
+                       int ldc) {
   const float *a_ptr, *b_ptr;
   a_ptr = a;
   b_ptr = b;
@@ -1009,7 +1012,7 @@ void AddDot4x8(int k, const float *a, const float *b, float *c, int ldc) {
 }
 
 // C = A * B
-void WriteBasic(int mc, int nc, float *c, float *C, int ldc) {
+void Gemmer::WriteBasic(int mc, int nc, float *c, float *C, int ldc) {
   int nc1 = nc / 16;
   int _nc1 = nc % 16;
   int step = 4 * ldc;
@@ -1066,10 +1069,10 @@ void WriteBasic(int mc, int nc, float *c, float *C, int ldc) {
 }
 
 // C = alpha * A * B + beta * C
-void WriteWithAlphaBeta(int mc, int nc, float *c, float *C, int ldc) {}
+void Gemmer::WriteWithAlphaBeta(int mc, int nc, float *c, float *C, int ldc) {}
 
 // C = A * B + C
-void WriteWithAdd(int mc, int nc, float *c, float *C, int ldc) {
+void Gemmer::WriteWithAdd(int mc, int nc, float *c, float *C, int ldc) {
   int nc1 = nc / 16;
   int _nc1 = nc % 16;
   int step = 4 * ldc;
@@ -1133,7 +1136,7 @@ void WriteWithAdd(int mc, int nc, float *c, float *C, int ldc) {
 }
 
 // C = A * B + C, relu(C)
-void WriteWithAddRelu(int mc, int nc, float *c, float *C, int ldc) {
+void Gemmer::WriteWithAddRelu(int mc, int nc, float *c, float *C, int ldc) {
   int nc1 = nc / 16;
   int _nc1 = nc % 16;
   int step = 4 * ldc;
@@ -1207,8 +1210,8 @@ void WriteWithAddRelu(int mc, int nc, float *c, float *C, int ldc) {
 }
 
 // C = A * B, batchnorm(C)
-void WriteWithBn(int mc, int nc, float *c, float *C, int ldc, float *scale,
-                 float *bias) {
+void Gemmer::WriteWithBn(int mc, int nc, float *c, float *C, int ldc,
+                         float *scale, float *bias) {
   int nc1 = nc / 16;
   int _nc1 = nc % 16;
   int nc2 = _nc1 / 4;
@@ -1293,8 +1296,8 @@ void WriteWithBn(int mc, int nc, float *c, float *C, int ldc, float *scale,
 }
 
 // C = A * B, batchnorm(C), relu(C)
-void WriteWithBnRelu(int mc, int nc, float *c, float *C, int ldc, float *scale,
-                     float *bias) {
+void Gemmer::WriteWithBnRelu(int mc, int nc, float *c, float *C, int ldc,
+                             float *scale, float *bias) {
   int nc1 = nc / 16;
   int _nc1 = nc % 16;
   int nc2 = _nc1 / 4;
@@ -1386,7 +1389,7 @@ void WriteWithBnRelu(int mc, int nc, float *c, float *C, int ldc, float *scale,
 }
 
 // C = A * B
-void VecWriteBasic(int n, float *c, float *C, int ldc) {
+void Gemmer::VecWriteBasic(int n, float *c, float *C, int ldc) {
   int nc1 = n / 16;
   int _nc1 = n % 16;
   int nc2 = _nc1 / 4;
@@ -1432,10 +1435,10 @@ void VecWriteBasic(int n, float *c, float *C, int ldc) {
 }
 
 // C = alpha * A * B + beta * C
-void VecWriteWithAlphaBeta(int n, float *c, float *C, int ldc) {}
+void Gemmer::VecWriteWithAlphaBeta(int n, float *c, float *C, int ldc) {}
 
 // C = A * B + C
-void VecWriteWithAdd(int n, float *c, float *C, int ldc) {
+void Gemmer::VecWriteWithAdd(int n, float *c, float *C, int ldc) {
   int nc1 = n / 16;
   int _nc1 = n % 16;
 
@@ -1473,7 +1476,7 @@ void VecWriteWithAdd(int n, float *c, float *C, int ldc) {
 }
 
 // C = A * B + C, relu(C)
-void VecWriteWithAddRelu(int n, float *c, float *C, int ldc) {
+void Gemmer::VecWriteWithAddRelu(int n, float *c, float *C, int ldc) {
   int nc1 = n / 16;
   int _nc1 = n % 16;
 
@@ -1521,8 +1524,8 @@ void VecWriteWithAddRelu(int n, float *c, float *C, int ldc) {
 }
 
 // C = A * B, batchnorm(C)
-void VecWriteWithBn(int n, float *c, float *C, int ldc, float *scale,
-                    float *bias) {
+void Gemmer::VecWriteWithBn(int n, float *c, float *C, int ldc, float *scale,
+                            float *bias) {
   int nc1 = n / 16;
   int _nc1 = n % 16;
   int nc2 = _nc1 / 4;
@@ -1588,8 +1591,8 @@ void VecWriteWithBn(int n, float *c, float *C, int ldc, float *scale,
 }
 
 // C = A * B, batchnorm(C), relu(C)
-void VecWriteWithBnRelu(int n, float *c, float *C, int ldc, float *scale,
-                        float *bias) {
+void Gemmer::VecWriteWithBnRelu(int n, float *c, float *C, int ldc,
+                                float *scale, float *bias) {
   int nc1 = n / 16;
   int _nc1 = n % 16;
   int nc2 = _nc1 / 4;

src/operators/math/gemm.h

@@ -13,6 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #pragma once
+#include <vector>
 
 // 矩阵取值运算宏，假设矩阵按行存储
 #define A(i, j) A[(i)*lda + (j)]
@@ -27,88 +28,111 @@ limitations under the License. */
 namespace paddle_mobile {
 namespace operators {
 namespace math {
+struct Gemmer {
+  int MC = 0;
+  int KC = 0;
+  int NC = 0;
 
-// 将 A 矩阵分块复制到连续内存(ColMajor)
-void PackMatrixA(int m, int k, int m_tail, const float *A, int lda,
-                 float *buffer);
-
-// 将 B 矩阵分块复制到连续内存(ColMajor)
-void PackMatrixB(int k, int n, int n_tail, const float *B, int ldb,
-                 float *buffer);
-
-// 将 A 矩阵分块复制到连续内存(RowMajor)
-void PackMatrixA_(int m, int k, int m_tail, const float *A, int lda,
-                  float *buffer);
-
-// 将 B 矩阵分块复制到连续内存(RowMajor)
-void PackMatrixB_(int k, int n, int n_tail, const float *B, int ldb,
-                  float *buffer);
-
-// 分块矩阵乘法
-void InnerKernel(int mc, int nc, float alpha, const float *a, const float *b,
-                 float beta, float *c, float *C, int ldc, bool relu);
-
-void InnerKernelWithBn(int mc, int nc, float alpha, const float *a,
-                       const float *b, float beta, float *c, float *C, int ldc,
-                       bool relu, float *new_scale, float *new_bias);
-
-// 向量矩阵乘法 (M = 1)
-void VectorKernel(int m, int n, int k, float alpha, const float *A, int lda,
-                  const float *B, int ldb, float beta, float *C, int ldc,
-                  bool relu);
-
-void VectorKernelWithBn(int m, int n, int k, float alpha, const float *A,
-                        int lda, const float *B, int ldb, float beta, float *C,
-                        int ldc, bool relu, float *new_scale, float *new_bias);
-
-// 计算一个更小的 C 矩阵分块
-void AddDot4x4(int k, const float *a, const float *b, float *c, int ldc);
-void AddDot4x8(int k, const float *a, const float *b, float *c, int ldc);
-
-// 分块矩阵乘法结果回写
-// C = A * B
-void WriteBasic(int mc, int nc, float *c, float *C, int ldc);
-// C = alpha * A * B + beta * C
-void WriteWithAlphaBeta(int mc, int nc, float *c, float *C, int ldc);
-// C = A * B + C
-void WriteWithAdd(int mc, int nc, float *c, float *C, int ldc);
-// C = A * B + C, relu(C)
-void WriteWithAddRelu(int mc, int nc, float *c, float *C, int ldc);
-// C = A * B, batchnorm(C)
-void WriteWithBn(int mc, int nc, float *c, float *C, int ldc, float *new_scale,
-                 float *new_bias);
-// C = A * B, batchnorm(C), relu(C)
-void WriteWithBnRelu(int mc, int nc, float *c, float *C, int ldc,
-                     float *new_scale, float *new_bias);
-
-// 向量矩阵乘法结果回写
-// C = A * B
-void VecWriteBasic(int n, float *c, float *C, int ldc);
-// C = alpha * A * B + beta * C
-void VecWriteWithAlphaBeta(int n, float *c, float *C, int ldc);
-// C = A * B + C
-void VecWriteWithAdd(int n, float *c, float *C, int ldc);
-// C = A * B + C, relu(C)
-void VecWriteWithAddRelu(int n, float *c, float *C, int ldc);
-// C = A * B, batchnorm(C)
-void VecWriteWithBn(int n, float *c, float *C, int ldc, float *new_scale,
-                    float *new_bias);
-// C = A * B, batchnorm(C), relu(C)
-void VecWriteWithBnRelu(int n, float *c, float *C, int ldc, float *new_scale,
-                        float *new_bias);
-
-// 32位 float 矩阵乘法
-void Sgemm(int m, int n, int k, float alpha, const float *A, int lda,
-           const float *B, int ldb, float beta, float *C, int ldc, bool relu);
-
-// 32位 float 矩阵乘法, 并对结果进行 batchnrom
-void SgemmWithBn(int m, int n, int k, float alpha, const float *A, int lda,
-                 const float *B, int ldb, float beta, float *C, int ldc,
-                 bool relu, float *new_scale, float *new_bias);
-
-// 64位 double 矩阵乘法
-void dgemm(int m, int n, int k, float alpha, const double *A, int lda,
-           const double *B, int ldb, float beta, double *C, int ldc);
+  float *packedA;
+  float *packedB;
+  float *packedC;
+  float *zero;
+  static std::vector<Gemmer *> gemmers;
+
+  // 将 A 矩阵分块复制到连续内存(ColMajor)
+  void PackMatrixA(int m, int k, int m_tail, const float *A, int lda,
+                   float *buffer);
+
+  // 将 B 矩阵分块复制到连续内存(ColMajor)
+  void PackMatrixB(int k, int n, int n_tail, const float *B, int ldb,
+                   float *buffer);
+
+  // 将 A 矩阵分块复制到连续内存(RowMajor)
+  void PackMatrixA_(int m, int k, int m_tail, const float *A, int lda,
+                    float *buffer);
+
+  // 将 B 矩阵分块复制到连续内存(RowMajor)
+  void PackMatrixB_(int k, int n, int n_tail, const float *B, int ldb,
+                    float *buffer);
+
+  // 分块矩阵乘法
+  void InnerKernel(int mc, int nc, float alpha, const float *a, const float *b,
+                   float beta, float *c, float *C, int ldc, bool relu);
+
+  void InnerKernelWithBn(int mc, int nc, float alpha, const float *a,
+                         const float *b, float beta, float *c, float *C,
+                         int ldc, bool relu, float *new_scale, float *new_bias);
+
+  // 向量矩阵乘法 (M = 1)
+  void VectorKernel(int m, int n, int k, float alpha, const float *A, int lda,
+                    const float *B, int ldb, float beta, float *C, int ldc,
+                    bool relu);
+
+  void VectorKernelWithBn(int m, int n, int k, float alpha, const float *A,
+                          int lda, const float *B, int ldb, float beta,
+                          float *C, int ldc, bool relu, float *new_scale,
+                          float *new_bias);
+
+  // 计算一个更小的 C 矩阵分块
+  void AddDot4x4(int k, const float *a, const float *b, float *c, int ldc);
+
+  void AddDot4x8(int k, const float *a, const float *b, float *c, int ldc);
+
+  // 分块矩阵乘法结果回写
+  // C = A * B
+  void WriteBasic(int mc, int nc, float *c, float *C, int ldc);
+
+  // C = alpha * A * B + beta * C
+  void WriteWithAlphaBeta(int mc, int nc, float *c, float *C, int ldc);
+
+  // C = A * B + C
+  void WriteWithAdd(int mc, int nc, float *c, float *C, int ldc);
+
+  // C = A * B + C, relu(C)
+  void WriteWithAddRelu(int mc, int nc, float *c, float *C, int ldc);
+
+  // C = A * B, batchnorm(C)
+  void WriteWithBn(int mc, int nc, float *c, float *C, int ldc,
+                   float *new_scale, float *new_bias);
+
+  // C = A * B, batchnorm(C), relu(C)
+  void WriteWithBnRelu(int mc, int nc, float *c, float *C, int ldc,
+                       float *new_scale, float *new_bias);
+
+  // 向量矩阵乘法结果回写
+  // C = A * B
+  void VecWriteBasic(int n, float *c, float *C, int ldc);
+
+  // C = alpha * A * B + beta * C
+  void VecWriteWithAlphaBeta(int n, float *c, float *C, int ldc);
+
+  // C = A * B + C
+  void VecWriteWithAdd(int n, float *c, float *C, int ldc);
+
+  // C = A * B + C, relu(C)
+  void VecWriteWithAddRelu(int n, float *c, float *C, int ldc);
+
+  // C = A * B, batchnorm(C)
+  void VecWriteWithBn(int n, float *c, float *C, int ldc, float *new_scale,
+                      float *new_bias);
+
+  // C = A * B, batchnorm(C), relu(C)
+  void VecWriteWithBnRelu(int n, float *c, float *C, int ldc, float *new_scale,
+                          float *new_bias);
+
+  // 32位 float 矩阵乘法
+  void Sgemm(int m, int n, int k, float alpha, const float *A, int lda,
+             const float *B, int ldb, float beta, float *C, int ldc, bool relu);
+
+  // 32位 float 矩阵乘法, 并对结果进行 batchnrom
+  void SgemmWithBn(int m, int n, int k, float alpha, const float *A, int lda,
+                   const float *B, int ldb, float beta, float *C, int ldc,
+                   bool relu, float *new_scale, float *new_bias);
+
+  // 64位 double 矩阵乘法
+  void dgemm(int m, int n, int k, float alpha, const double *A, int lda,
+             const double *B, int ldb, float beta, double *C, int ldc);
+};
 
 }  // namespace math
 }  // namespace operators

src/operators/math/math_function.cpp

@@ -26,23 +26,14 @@ void matmul<float>(const framework::Tensor &matrix_a, bool trans_a,
   auto dim_a = matrix_a.dims();
   auto dim_b = matrix_b.dims();
   auto dim_out = matrix_out->dims();
-  //  PADDLE_ENFORCE(dim_a.size() == 2 && dim_b.size() == 2 &&
-  //  dim_out.size() ==
-  //  2,
-  //                 "The input and output of matmul be matrix");
-  //
-  //  PADDLE_ENFORCE(platform::is_cpu_place(matrix_a.place()) &&
-  //                     platform::is_cpu_place(matrix_b.place())
-  //                     &&
-  //                     platform::is_cpu_place(matrix_out->place()),
-  //                 "Matrix must all be in CPUPlace");
 
   int M = dim_out[0];
   int N = dim_out[1];
   int K = (!trans_a) ? dim_a[1] : dim_a[0];
 
-  Sgemm(M, N, K, alpha, matrix_a.data<float>(), K, matrix_b.data<float>(), N,
-        beta, matrix_out->data<float>(), N, relu);
+  Gemmer::gemmers[0]->Sgemm(M, N, K, alpha, matrix_a.data<float>(), K,
+                            matrix_b.data<float>(), N, beta,
+                            matrix_out->data<float>(), N, relu);
 }
 
 template <>
@@ -54,24 +45,15 @@ void matmulWithBn<float>(const framework::Tensor &matrix_a, bool trans_a,
   auto dim_a = matrix_a.dims();
   auto dim_b = matrix_b.dims();
   auto dim_out = matrix_out->dims();
-  //  PADDLE_ENFORCE(dim_a.size() == 2 && dim_b.size() == 2 &&
-  //  dim_out.size() ==
-  //  2,
-  //                 "The input and output of matmul be matrix");
-  //
-  //  PADDLE_ENFORCE(platform::is_cpu_place(matrix_a.place()) &&
-  //                     platform::is_cpu_place(matrix_b.place())
-  //                     &&
-  //                     platform::is_cpu_place(matrix_out->place()),
-  //                 "Matrix must all be in CPUPlace");
 
   int M = dim_out[0];
   int N = dim_out[1];
   int K = (!trans_a) ? dim_a[1] : dim_a[0];
 
-  SgemmWithBn(M, N, K, alpha, matrix_a.data<float>(), K, matrix_b.data<float>(),
-              N, beta, matrix_out->data<float>(), N, relu,
-              new_scale->data<float>(), new_bias->data<float>());
+  Gemmer::gemmers[0]->SgemmWithBn(
+      M, N, K, alpha, matrix_a.data<float>(), K, matrix_b.data<float>(), N,
+      beta, matrix_out->data<float>(), N, relu, new_scale->data<float>(),
+      new_bias->data<float>());
 }
 
 }  // namespace math

src/operators/math/pool_3x3.cpp

@@ -13,8 +13,12 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #ifdef POOL_OP
-#include "pool_3x3.h"
+#define __ARM_NEON true
+#ifdef _OPENMP
+#include <omp.h>
+#endif
 #include "framework/tensor.h"
+#include "pool_3x3.h"
 #if __ARM_NEON
 #include <arm_neon.h>
 #endif  // __ARM_NEON
@@ -40,46 +44,52 @@ void Pool3x3Avgs1p1(const Tensor *input, Tensor *output) {
   const int w_out = output->dims()[3];
   const int outputdata_channel_stride = h_out * w_out;
   const int inputdata_channel_stride = h_in * w_in;
+  const int input_batch_stride = output_channels * inputdata_channel_stride;
+  const int output_batch_stride = output_channels * outputdata_channel_stride;
   float *out_data = output->data<float>();
   const float *input_data = input->data<float>();
+
   const float coef = 1.0 / 9.0;
   for (int k = 0; k < batch_size; ++k) {
+#pragma omp parallel for
     for (int c = 0; c < output_channels; ++c) {
+      const float *input_seg = input_data + c * inputdata_channel_stride;
+      float *output_seg = out_data + c * outputdata_channel_stride;
       // four corner point
-      out_data[0] = (input_data[0] + input_data[1] + input_data[w_in] +
-                     input_data[w_in + 1]) *
-                    coef;
-      out_data[w_out - 1] =
-          (input_data[w_in - 2] + input_data[w_in - 1] +
-           input_data[w_in * 2 - 2] + input_data[2 * w_in - 1]) *
+      output_seg[0] = (input_seg[0] + input_seg[1] + input_seg[w_in] +
+                       input_seg[w_in + 1]) *
+                      coef;
+      output_seg[w_out - 1] =
+          (input_seg[w_in - 2] + input_seg[w_in - 1] + input_seg[w_in * 2 - 2] +
+           input_seg[2 * w_in - 1]) *
           coef;
-      out_data[(h_out - 1) * w_out] =
-          (input_data[(h_in - 2) * w_in] + input_data[(h_in - 2) * w_in + 1] +
-           input_data[(h_in - 1) * w_in] + input_data[(h_in - 1) * w_in + 1]) *
+      output_seg[(h_out - 1) * w_out] =
+          (input_seg[(h_in - 2) * w_in] + input_seg[(h_in - 2) * w_in + 1] +
+           input_seg[(h_in - 1) * w_in] + input_seg[(h_in - 1) * w_in + 1]) *
           coef;
-      out_data[h_out * w_out - 1] =
-          (input_data[h_in * w_in - 1] + input_data[h_in * w_in - 2] +
-           input_data[(h_in - 1) * w_in - 1] +
-           input_data[(h_in - 1) * w_in - 2]) *
+      output_seg[h_out * w_out - 1] =
+          (input_seg[h_in * w_in - 1] + input_seg[h_in * w_in - 2] +
+           input_seg[(h_in - 1) * w_in - 1] +
+           input_seg[(h_in - 1) * w_in - 2]) *
           coef;
       // left side & right side
       for (int i = 1; i < h_in - 1; ++i) {
-        out_data[i * w_out] =
-            (input_data[i * w_in - w_in] + input_data[i * w_in - w_in + 1] +
-             input_data[i * w_in] + input_data[i * w_in + 1] +
-             input_data[i * w_in + w_in] + input_data[i * w_in + w_in + 1]) *
+        output_seg[i * w_out] =
+            (input_seg[i * w_in - w_in] + input_seg[i * w_in - w_in + 1] +
+             input_seg[i * w_in] + input_seg[i * w_in + 1] +
+             input_seg[i * w_in + w_in] + input_seg[i * w_in + w_in + 1]) *
             coef;
-        out_data[i * w_out + w_out - 1] =
-            (input_data[i * w_in - w_in + w_in - 2] +
-             input_data[i * w_in - w_in + 1 + w_in - 2] +
-             input_data[i * w_in + w_in - 2] +
-             input_data[i * w_in + 1 + w_in - 2] +
-             input_data[i * w_in + w_in + w_in - 2] +
-             input_data[i * w_in + w_in + 1 + w_in - 2]) *
+        output_seg[i * w_out + w_out - 1] =
+            (input_seg[i * w_in - w_in + w_in - 2] +
+             input_seg[i * w_in - w_in + 1 + w_in - 2] +
+             input_seg[i * w_in + w_in - 2] +
+             input_seg[i * w_in + 1 + w_in - 2] +
+             input_seg[i * w_in + w_in + w_in - 2] +
+             input_seg[i * w_in + w_in + 1 + w_in - 2]) *
             coef;
       }
       // top 1 row & bottom 1 row
-      const float *input_tmp = input_data;
+      const float *input_tmp = input_seg;
 
       float32x4_t in0, in1, in2, in3, in4, in5, in6, in7, tmp0, tmp1, tmp2,
           tmp3, tmp4, tmp5, sum, out0;
@@ -90,7 +100,7 @@ void Pool3x3Avgs1p1(const Tensor *input, Tensor *output) {
       in4 = vld1q_f32(input_tmp_end);
       in6 = vld1q_f32(input_tmp_end + w_in);
       int c_mid = w_out - 2;
-      auto output_ptr = out_data + 1;
+      auto output_ptr = output_seg + 1;
       for (; c_mid > 3; c_mid -= 4) {
         in1 = vld1q_f32(input_tmp + 4);
         in3 = vld1q_f32(input_tmp + w_in + 4);
@@ -135,8 +145,8 @@ void Pool3x3Avgs1p1(const Tensor *input, Tensor *output) {
         in6 = in7;
       }
       // top right remain
-      float32x4_t pad0 = vdupq_n_f32(input_data[w_in - 1]);
-      float32x4_t pad1 = vdupq_n_f32(input_data[2 * w_in - 1]);
+      float32x4_t pad0 = vdupq_n_f32(input_seg[w_in - 1]);
+      float32x4_t pad1 = vdupq_n_f32(input_seg[2 * w_in - 1]);
 
       tmp0 = vextq_f32(in0, pad0, 1);
       tmp1 = vextq_f32(in0, pad0, 2);
@@ -163,8 +173,8 @@ void Pool3x3Avgs1p1(const Tensor *input, Tensor *output) {
       }
 
       // bottom_right remain
-      float32x4_t pad2 = vdupq_n_f32(input_data[(h_in - 1) * w_in - 1]);
-      float32x4_t pad3 = vdupq_n_f32(input_data[h_in * w_in - 1]);
+      float32x4_t pad2 = vdupq_n_f32(input_seg[(h_in - 1) * w_in - 1]);
+      float32x4_t pad3 = vdupq_n_f32(input_seg[h_in * w_in - 1]);
 
       tmp0 = vextq_f32(in4, pad2, 1);
       tmp1 = vextq_f32(in4, pad2, 2);
@@ -191,8 +201,8 @@ void Pool3x3Avgs1p1(const Tensor *input, Tensor *output) {
       }
       // mid
       for (int j = 0; j < h_out - 2; ++j) {
-        output_ptr = out_data + w_out * (j + 1) + 1;
-        input_tmp = input_data + j * w_in;
+        output_ptr = output_seg + w_out * (j + 1) + 1;
+        input_tmp = input_seg + j * w_in;
 
         in0 = vld1q_f32(input_tmp);
         in2 = vld1q_f32(input_tmp + w_in);
@@ -228,9 +238,9 @@ void Pool3x3Avgs1p1(const Tensor *input, Tensor *output) {
           in4 = in5;
         }
         // mid remain
-        float32x4_t pad0 = vdupq_n_f32(input_data[(j + 1) * w_in - 1]);
-        float32x4_t pad1 = vdupq_n_f32(input_data[(j + 2) * w_in - 1]);
-        float32x4_t pad2 = vdupq_n_f32(input_data[(j + 2) * w_in - 1]);
+        float32x4_t pad0 = vdupq_n_f32(input_seg[(j + 1) * w_in - 1]);
+        float32x4_t pad1 = vdupq_n_f32(input_seg[(j + 2) * w_in - 1]);
+        float32x4_t pad2 = vdupq_n_f32(input_seg[(j + 2) * w_in - 1]);
 
         tmp0 = vextq_f32(in0, pad0, 1);
         tmp1 = vextq_f32(in0, pad0, 2);
@@ -261,9 +271,11 @@ void Pool3x3Avgs1p1(const Tensor *input, Tensor *output) {
           }
         }
       }
-      input_data += inputdata_channel_stride;
-      out_data += outputdata_channel_stride;
+      //      input_data += inputdata_channel_stride;
+      //      out_data += outputdata_channel_stride;
     }
+    input_data += input_batch_stride;
+    out_data += output_batch_stride;
   }
 #endif
 }
@@ -282,44 +294,50 @@ void Pool3x3Maxs1p1(const Tensor *input, Tensor *output) {
   const int w_out = output->dims()[3];
   const int outputdata_channel_stride = h_out * w_out;
   const int inputdata_channel_stride = h_in * w_in;
+  const int input_batch_stride = output_channels * inputdata_channel_stride;
+  const int output_batch_stride = output_channels * outputdata_channel_stride;
   float *out_data = output->data<float>();
   const float *input_data = input->data<float>();
   for (int k = 0; k < batch_size; ++k) {
+#pragma omp parallel for
     for (int c = 0; c < output_channels; ++c) {
+      const float *input_seg = input_data + c * inputdata_channel_stride;
+      float *output_seg = out_data + c * outputdata_channel_stride;
       // four corner point
-      out_data[0] = std::max(std::max(input_data[0], input_data[1]),
-                             std::max(input_data[w_in], input_data[w_in + 1]));
-      out_data[w_out - 1] = std::max(
-          std::max(input_data[w_in - 2], input_data[w_in - 1]),
-          std::max(input_data[w_in * 2 - 2], input_data[2 * w_in - 1]));
-      out_data[(h_out - 1) * w_out] =
-          std::max(std::max(input_data[(h_in - 2) * w_in],
-                            input_data[(h_in - 2) * w_in + 1]),
-                   std::max(input_data[(h_in - 1) * w_in],
-                            input_data[(h_in - 1) * w_in + 1]));
-      out_data[h_out * w_out - 1] = std::max(
-          std::max(input_data[(h_in - 1) * w_in - 1],
-                   input_data[(h_in - 1) * w_in - 2]),
-          std::max(input_data[h_in * w_in - 1], input_data[h_in * w_in - 2]));
+      output_seg[0] = std::max(std::max(input_seg[0], input_seg[1]),
+                               std::max(input_seg[w_in], input_seg[w_in + 1]));
+      output_seg[w_out - 1] =
+          std::max(std::max(input_seg[w_in - 2], input_seg[w_in - 1]),
+                   std::max(input_seg[w_in * 2 - 2], input_seg[2 * w_in - 1]));
+      output_seg[(h_out - 1) * w_out] =
+          std::max(std::max(input_seg[(h_in - 2) * w_in],
+                            input_seg[(h_in - 2) * w_in + 1]),
+                   std::max(input_seg[(h_in - 1) * w_in],
+                            input_seg[(h_in - 1) * w_in + 1]));
+      output_seg[h_out * w_out - 1] = std::max(
+          std::max(input_seg[(h_in - 1) * w_in - 1],
+                   input_seg[(h_in - 1) * w_in - 2]),
+          std::max(input_seg[h_in * w_in - 1], input_seg[h_in * w_in - 2]));
       // left side & right side
       for (int i = 1; i < h_in - 1; ++i) {
-        float max1 = std::max(input_data[i * w_in - w_in],
-                              input_data[i * w_in - w_in + 1]);
-        float max2 = std::max(input_data[i * w_in], input_data[i * w_in + 1]);
-        float max3 = std::max(input_data[i * w_in + w_in],
-                              input_data[i * w_in + w_in + 1]);
-        out_data[i * w_out] = std::max(std::max(max1, max2), max3);
-
-        max1 = std::max(input_data[i * w_in - w_in + w_in - 2],
-                        input_data[i * w_in - w_in + 1 + w_in - 2]);
-        max2 = std::max(input_data[i * w_in + w_in - 2],
-                        input_data[i * w_in + 1 + w_in - 2]);
-        max3 = std::max(input_data[i * w_in + w_in + w_in - 2],
-                        input_data[i * w_in + w_in + 1 + w_in - 2]);
-        out_data[i * w_out + w_out - 1] = std::max(std::max(max1, max2), max3);
+        float max1 = std::max(input_seg[i * w_in - w_in],
+                              input_seg[i * w_in - w_in + 1]);
+        float max2 = std::max(input_seg[i * w_in], input_seg[i * w_in + 1]);
+        float max3 = std::max(input_seg[i * w_in + w_in],
+                              input_seg[i * w_in + w_in + 1]);
+        output_seg[i * w_out] = std::max(std::max(max1, max2), max3);
+
+        max1 = std::max(input_seg[i * w_in - w_in + w_in - 2],
+                        input_seg[i * w_in - w_in + 1 + w_in - 2]);
+        max2 = std::max(input_seg[i * w_in + w_in - 2],
+                        input_seg[i * w_in + 1 + w_in - 2]);
+        max3 = std::max(input_seg[i * w_in + w_in + w_in - 2],
+                        input_seg[i * w_in + w_in + 1 + w_in - 2]);
+        output_seg[i * w_out + w_out - 1] =
+            std::max(std::max(max1, max2), max3);
       }
       // top 1 row & bottom 1 row
-      const float *input_tmp = input_data;
+      const float *input_tmp = input_seg;
 
       float32x4_t in0, in1, in2, in3, in4, in5, in6, in7, tmp0, tmp1, tmp2,
           tmp3, tmp4, tmp5, max;
@@ -329,7 +347,7 @@ void Pool3x3Maxs1p1(const Tensor *input, Tensor *output) {
       in4 = vld1q_f32(input_tmp_end);
       in6 = vld1q_f32(input_tmp_end + w_in);
       int c_mid = w_out - 2;
-      auto output_ptr = out_data + 1;
+      auto output_ptr = output_seg + 1;
       for (; c_mid > 3; c_mid -= 4) {
         in1 = vld1q_f32(input_tmp + 4);
         in3 = vld1q_f32(input_tmp + w_in + 4);
@@ -373,8 +391,8 @@ void Pool3x3Maxs1p1(const Tensor *input, Tensor *output) {
         in6 = in7;
       }
       // top right remain
-      float32x4_t pad0 = vdupq_n_f32(input_data[w_in - 1]);
-      float32x4_t pad1 = vdupq_n_f32(input_data[2 * w_in - 1]);
+      float32x4_t pad0 = vdupq_n_f32(input_seg[w_in - 1]);
+      float32x4_t pad1 = vdupq_n_f32(input_seg[2 * w_in - 1]);
 
       tmp0 = vextq_f32(in0, pad0, 1);
       tmp1 = vextq_f32(in0, pad0, 2);
@@ -400,8 +418,8 @@ void Pool3x3Maxs1p1(const Tensor *input, Tensor *output) {
       }
 
       // bottom_right remain
-      float32x4_t pad2 = vdupq_n_f32(input_data[(h_in - 1) * w_in - 1]);
-      float32x4_t pad3 = vdupq_n_f32(input_data[h_in * w_in - 1]);
+      float32x4_t pad2 = vdupq_n_f32(input_seg[(h_in - 1) * w_in - 1]);
+      float32x4_t pad3 = vdupq_n_f32(input_seg[h_in * w_in - 1]);
 
       tmp0 = vextq_f32(in4, pad2, 1);
       tmp1 = vextq_f32(in4, pad2, 2);
@@ -427,8 +445,8 @@ void Pool3x3Maxs1p1(const Tensor *input, Tensor *output) {
       }
       // mid
       for (int j = 0; j < h_out - 2; ++j) {
-        output_ptr = out_data + (j + 1) * w_out + 1;
-        input_tmp = input_data + j * w_in;
+        output_ptr = output_seg + (j + 1) * w_out + 1;
+        input_tmp = input_seg + j * w_in;
 
         in0 = vld1q_f32(input_tmp);
         in2 = vld1q_f32(input_tmp + w_in);
@@ -463,9 +481,9 @@ void Pool3x3Maxs1p1(const Tensor *input, Tensor *output) {
           in4 = in5;
         }
         // mid remain
-        float32x4_t pad0 = vdupq_n_f32(input_data[(j + 1) * w_in - 1]);
-        float32x4_t pad1 = vdupq_n_f32(input_data[(j + 2) * w_in - 1]);
-        float32x4_t pad2 = vdupq_n_f32(input_data[(j + 3) * w_in - 1]);
+        float32x4_t pad0 = vdupq_n_f32(input_seg[(j + 1) * w_in - 1]);
+        float32x4_t pad1 = vdupq_n_f32(input_seg[(j + 2) * w_in - 1]);
+        float32x4_t pad2 = vdupq_n_f32(input_seg[(j + 3) * w_in - 1]);
 
         tmp0 = vextq_f32(in0, pad0, 1);
         tmp1 = vextq_f32(in0, pad0, 2);
@@ -495,9 +513,11 @@ void Pool3x3Maxs1p1(const Tensor *input, Tensor *output) {
           }
         }
       }
-      input_data += inputdata_channel_stride;
-      out_data += outputdata_channel_stride;
+      //      input_data += inputdata_channel_stride;
+      //      out_data += outputdata_channel_stride;
     }
+    input_data += input_batch_stride;
+    out_data += output_batch_stride;
   }
 #endif
 }
@@ -515,11 +535,11 @@ void Pool3x3Max(vector<int> strides, vector<int> paddings, const Tensor *input,
 
   const int output_height = output->dims()[2];
   const int output_width = output->dims()[3];
-  const int _kernel_size = 3;
-  const int stride_height = strides[0];
-  const int stride_width = strides[1];
-  const int padding_height = paddings[0];
-  const int padding_width = paddings[1];
+  //  const int _kernel_size = 3;
+  const int stride = strides[0];
+  //  const int stride_width = strides[1];
+  const int padding = paddings[0];
+  //  const int padding_width = paddings[1];
   const float negative_max = -INT_MAX;
   const int input_channel_stride = input_height * input_width;
   const int output_channel_stride = output_height * output_width;
@@ -529,36 +549,39 @@ void Pool3x3Max(vector<int> strides, vector<int> paddings, const Tensor *input,
 
   const int input_batch_stride = output_channels * input_channel_stride;
   const int output_batch_stride = output_channels * output_channel_stride;
-  const float *pos1, *pos2, *pos3, *output_ptr;
+  const float *pos1, *output_ptr;
   int hstart, wstart, hend, wend;
   for (int i = 0; i < batch_size; ++i) {
+#pragma omp parallel for
     for (int c = 0; c < output_channels; ++c) {
+      const float *input_seg = input_data + c * input_channel_stride;
+      float *output_seg = output_data + c * output_channel_stride;
       for (int ph = 0; ph < output_height; ph++) {
         for (int pw = 0; pw < output_width; pw++) {
-          hstart = ph * stride_height - padding_height;
-          wstart = pw * stride_width - padding_width;
-          hend = min(hstart + _kernel_size, input_height + padding_height);
-          wend = min(wstart + _kernel_size, input_width + padding_width);
+          int hstart = ph * stride - padding;
+          int wstart = pw * stride - padding;
+          int hend = min(hstart + 3, input_height + padding);
+          int wend = min(wstart + 3, input_width + padding);
           hstart = max(hstart, 0);
           wstart = max(wstart, 0);
           hend = min(hend, input_height);
           wend = min(wend, input_width);
-          pos1 = input_data + hstart * input_width + wstart;
-          pos2 = input_data + (hstart + 1) * input_width + wstart;
-          pos3 = input_data + (hstart + 2) * input_width + wstart;
-          output_ptr = output_data + ph * output_width + pw;
+          const float *pos1 = input_seg + hstart * input_width + wstart;
+          const float *pos2 = input_seg + (hstart + 1) * input_width + wstart;
+          const float *pos3 = input_seg + (hstart + 2) * input_width + wstart;
+          output_ptr = output_seg + ph * output_width + pw;
 
           if (hend - hstart != 3 || wend - wstart != 3) {
             float max_value = -INT_MAX;
             for (int h = hstart; h < hend; h++) {
               for (int w = wstart; w < wend; w++) {
-                float value = input_data[h * input_width + w];
+                float value = input_seg[h * input_width + w];
                 if (value > max_value) {
                   max_value = value;
                 }
               }
             }
-            output_data[ph * output_width + pw] = max_value;
+            output_seg[ph * output_width + pw] = max_value;
           } else {
 #if defined(ARMV7)
             asm volatile(
@@ -572,27 +595,25 @@ void Pool3x3Max(vector<int> strides, vector<int> paddings, const Tensor *input,
                 "vpmax.f32  d7, d6, d6             \n\t"
                 "vst1.32 {d7[0]},[%[output_ptr]]    \n\t"
                 :
-                : [input_data] "r"(input_data), [pos1] "r"(pos1),
+                : [input_seg] "r"(input_seg), [pos1] "r"(pos1),
                   [pos2] "r"(pos2), [pos3] "r"(pos3),
                   [output_ptr] "r"(output_ptr), [negative_max] "r"(negative_max)
                 : "memory", "q1", "q2", "q3", "q4");
 #else
             const float32x4_t data1 = vld1q_f32(pos1);
-            const float32x4_t data2 = vld1q_f32(pos2);
-            const float32x4_t data3 = vld1q_f32(pos3);
+            const float32x4_t data2 = vld1q_f32(pos1 + input_width);
+            const float32x4_t data3 = vld1q_f32(pos1 + 2 * input_width);
             const float32x4_t max_data =
-                vmaxq_f32(vmaxq_f32(data1, data3), data2);
+                vmaxq_f32(vmaxq_f32(data1, data2), data3);
             float32x2_t res =
                 vpmax_f32(vget_high_f32(vsetq_lane_f32(-INT_MAX, max_data, 3)),
                           vget_low_f32(max_data));
             res = vpmax_f32(res, res);
-            output_data[ph * output_width + pw] = vget_lane_f32(res, 0);
+            output_seg[ph * output_width + pw] = vget_lane_f32(res, 0);
 #endif
           }
         }
       }
-      input_data += input_channel_stride;
-      output_data += output_channel_stride;
     }
     input_data += input_batch_stride;
     output_data += output_batch_stride;
@@ -613,11 +634,8 @@ void Pool3x3Avg(vector<int> strides, vector<int> paddings, const Tensor *input,
 
   const int output_height = output->dims()[2];
   const int output_width = output->dims()[3];
-  const int _kernel_size = 3;
-  const int stride_height = strides[0];
-  const int stride_width = strides[1];
-  const int padding_height = paddings[0];
-  const int padding_width = paddings[1];
+  const int stride = strides[0];
+  const int padding = paddings[0];
 
   const int input_channel_stride = input_height * input_width;
   const int output_channel_stride = output_height * output_width;
@@ -631,30 +649,33 @@ void Pool3x3Avg(vector<int> strides, vector<int> paddings, const Tensor *input,
   const int input_batch_stride = output_channels * input_channel_stride;
   const int output_batch_stride = output_channels * output_channel_stride;
   for (int i = 0; i < batch_size; ++i) {
+#pragma omp parallel for
     for (int c = 0; c < output_channels; ++c) {
+      const float *input_seg = input_data + c * input_channel_stride;
+      float *output_seg = output_data + c * output_channel_stride;
       for (int ph = 0; ph < output_height; ph++) {
         for (int pw = 0; pw < output_width; pw++) {
-          int hstart = ph * stride_height - padding_height;
-          int wstart = pw * stride_width - padding_width;
-          int hend = min(hstart + _kernel_size, input_height + padding_height);
-          int wend = min(wstart + _kernel_size, input_width + padding_width);
+          int hstart = ph * stride - padding;
+          int wstart = pw * stride - padding;
+          int hend = min(hstart + 3, input_height + padding);
+          int wend = min(wstart + 3, input_width + padding);
           hstart = max(hstart, 0);
           wstart = max(wstart, 0);
           hend = min(hend, input_height);
           wend = min(wend, input_width);
-          const float *pos1 = input_data + hstart * input_width + wstart;
-          const float *pos2 = input_data + (hstart + 1) * input_width + wstart;
-          const float *pos3 = input_data + (hstart + 2) * input_width + wstart;
-          const float *output_ptr = output_data + ph * output_width + pw;
+          const float *pos1 = input_seg + hstart * input_width + wstart;
+          const float *pos2 = input_seg + (hstart + 1) * input_width + wstart;
+          const float *pos3 = input_seg + (hstart + 2) * input_width + wstart;
+          float *output_ptr = output_seg + ph * output_width + pw;
 
           if (hend - hstart != 3 || wend - wstart != 3) {
             float sum = 0;
             for (int h = hstart; h < hend; h++) {
               for (int w = wstart; w < wend; w++) {
-                sum += input_data[h * input_width + w];
+                sum += input_seg[h * input_width + w];
               }
             }
-            output_data[ph * output_width + pw] = sum / 9.0;
+            output_seg[ph * output_width + pw] = sum / 9.0;
           } else {
 #if defined(ARMV7)
 
@@ -671,7 +692,7 @@ void Pool3x3Avg(vector<int> strides, vector<int> paddings, const Tensor *input,
                 "vmul.f32 d6,d7                     \n\t"
                 "vst1.32 {d6[0]},[%[output_ptr]]    \n\t"
                 :
-                : [input_data] "r"(input_data), [pos1] "r"(pos1),
+                : [input_seg] "r"(input_seg), [pos1] "r"(pos1),
                   [pos2] "r"(pos2), [pos3] "r"(pos3),
                   [output_ptr] "r"(output_ptr), [zero] "r"(zero),
                   [nine_ptr] "r"(nine_ptr)
@@ -686,13 +707,11 @@ void Pool3x3Avg(vector<int> strides, vector<int> paddings, const Tensor *input,
                 vpadd_f32(vget_high_f32(vsetq_lane_f32(0, sum_data, 3)),
                           vget_low_f32(sum_data));
             res = vpadd_f32(res, res);
-            output_data[ph * output_width + pw] = vget_lane_f32(res, 0) / 9.0;
+            output_seg[ph * output_width + pw] = vget_lane_f32(res, 0) / 9.0;
 #endif
           }
         }
       }
-      input_data += input_channel_stride;
-      output_data += output_channel_stride;
     }
     input_data += input_batch_stride;
     output_data += output_batch_stride;

src/operators/math/pool_3x3.h

@@ -15,6 +15,9 @@ limitations under the License. */
 #ifdef POOL_OP
 
 #pragma once
+#ifdef _OPENMP
+#include <omp.h>
+#endif
 #include <algorithm>
 #include <vector>
 #include "framework/tensor.h"

src/operators/math/pooling.cpp

@@ -16,6 +16,9 @@ limitations under the License. */
 
 #include "pooling.h"
 #include "common/types.h"
+#ifdef _OPENMP
+#include <omp.h>
+#endif
 
 namespace paddle_mobile {
 namespace operators {
@@ -57,8 +60,8 @@ class PoolFunctor<CPU, PoolProcess, T> {
     T *output_data = output->mutable_data<T>();
 
     for (int i = 0; i < batch_size; i++) {
-      //  #pragma omp parallel for
       for (int c = 0; c < output_channels; ++c) {
+#pragma omp parallel for
         for (int ph = 0; ph < output_height; ++ph) {
           int hstart = ph * stride_height - padding_height;
           int hend = std::min(hstart + ksize_height, input_height);

test/net/test_googlenet.cpp

@@ -26,16 +26,17 @@ int main() {
   auto time2 = time();
   DLOG << "load cost :" << time_diff(time1, time2) << "ms\n";
   paddle_mobile::Executor<paddle_mobile::CPU> executor(program, 1, optimize);
+  executor.SetThreadNum(4);
   std::vector<float> input;
   std::vector<int64_t> dims{1, 3, 224, 224};
   GetInput<float>(g_test_image_1x3x224x224, &input, dims);
   auto time3 = time();
-
-  for (int i = 0; i < 10; ++i) {
+  int count = 1;
+  for (int i = 0; i < count; ++i) {
     executor.Predict(input, dims);
   }
 
   auto time4 = time();
-  DLOG << "predict cost :" << time_diff(time3, time4) << "ms\n";
+  DLOG << "predict cost :" << time_diff(time3, time4) / count << "ms\n";
   return 0;
 }