Fix 1x1 depthwise conv

4406f902 · hjchen2 · c02076f0 · 4406f902
隐藏空白更改
内联并排

Showing with 4 addition and 27 deletion

src/operators/math/math_function.cpp src/operators/math/math_function.cpp +4 -27

未找到文件。
--- a/src/operators/math/math_function.cpp
+++ b/src/operators/math/math_function.cpp
@@ -71,34 +71,11 @@ void MatMul<float, float>(const framework::Tensor &matrix_a, bool trans_a,
        a[index++] = tmp[i * n + j];
      }
    }
-    if (M == 1) {
+    cblas_sgemm(false, false, M, N, K, alpha, a, K, matrix_b.data<float>(), N,
-#ifdef _OPENMP
+                beta, matrix_out->data<float>(), N);
-      gemm.Sgemm_omp(M, N, K, alpha, a, K, matrix_b.data<float>(), N, beta,
-                     matrix_out->data<float>(), N, relu, bias);
-#else
-      gemm.Sgemm(M, N, K, alpha, a, K, matrix_b.data<float>(), N, beta,
-                 matrix_out->data<float>(), N, relu, bias);
-#endif
-    } else {
-      cblas_sgemm(false, false, M, N, K, alpha, a, K, matrix_b.data<float>(), N,
-                  beta, matrix_out->data<float>(), N);
-    }
  } else {
-    if (M == 1) {
+    cblas_sgemm(false, false, M, N, K, alpha, matrix_a.data<float>(), K,
-#ifdef _OPENMP
+                matrix_b.data<float>(), N, beta, matrix_out->data<float>(), N);
-      gemm.Sgemm_omp(M, N, K, alpha, matrix_a.data<float>(), K,
-                     matrix_b.data<float>(), N, beta, matrix_out->data<float>(),
-                     N, relu, bias);
-#else
-      gemm.Sgemm(M, N, K, alpha, matrix_a.data<float>(), K,
-                 matrix_b.data<float>(), N, beta, matrix_out->data<float>(), N,
-                 relu, bias);
-#endif
-    } else {
-      cblas_sgemm(false, false, M, N, K, alpha, matrix_a.data<float>(), K,
-                  matrix_b.data<float>(), N, beta, matrix_out->data<float>(),
-                  N);
-    }
  }
 }