Merge pull request #585 from Eclipsess/develop

fix #584 add basic with group

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

@@ -104,7 +104,7 @@ void ConvAddBNReluBasic(const FusionConvAddBNReluParam &param) {
 
       math::matmulWithBn<float>(
           filter_slice, false, col_matrix, false, static_cast<float>(1),
-          &out_slice, static_cast<float>(0), true, &new_scale, &new_bias);
+          &out_slice, static_cast<float>(0), true, &new_scale, &new_bias, g);
     }
   }
 }

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

@@ -101,10 +101,9 @@ void DWConvBNReluBasic(const FusionDWConvBNReluParam &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);
-      std::cout << "***************" << std::endl;
       math::matmulWithBn<float>(
           filter_slice, false, col_matrix, false, static_cast<float>(1),
-          &out_slice, static_cast<float>(0), false, &new_scale, &new_bias);
+          &out_slice, static_cast<float>(0), true, &new_scale, &new_bias, g);
     }
   }
 }

src/operators/math/math_function.cpp

@@ -50,7 +50,7 @@ void matmulWithBn<float>(const framework::Tensor &matrix_a, bool trans_a,
                          const framework::Tensor &matrix_b, bool trans_b,
                          float alpha, framework::Tensor *matrix_out, float beta,
                          bool relu, framework::Tensor *new_scale,
-                         framework::Tensor *new_bias) {
+                         framework::Tensor *new_bias, int group) {
   auto dim_a = matrix_a.dims();
   auto dim_b = matrix_b.dims();
   auto dim_out = matrix_out->dims();
@@ -71,7 +71,8 @@ void matmulWithBn<float>(const framework::Tensor &matrix_a, bool trans_a,
 
   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>());
+              new_scale->data<float>() + group,
+              new_bias->data<float>() + group);
 }
 
 }  // namespace math

src/operators/math/math_function.h

@@ -31,7 +31,8 @@ template <typename T>
 void matmulWithBn(const framework::Tensor &matrix_a, bool trans_a,
                   const framework::Tensor &matrix_b, bool trans_b, T alpha,
                   framework::Tensor *matrix_out, T beta, bool relu,
-                  framework::Tensor *new_scale, framework::Tensor *new_bias);
+                  framework::Tensor *new_scale, framework::Tensor *new_bias,
+                  int group);
 }  // namespace math
 }  // namespace operators
 }  // namespace paddle_mobile