Merge pull request #960 from yangfei963158659/develop

imp conv_transpose op

src/operators/math/math_function.cpp

@@ -36,6 +36,27 @@ void matmul<float>(const framework::Tensor &matrix_a, bool trans_a,
   int N = dim_out[1];
   int K = (!trans_a) ? dim_a[1] : dim_a[0];
 
+  if (trans_a) {
+    int numel = matrix_a.numel();
+    int m = matrix_a.dims()[0];
+    int n = matrix_a.dims()[1];
+    float *tmp = (float *)(matrix_a.data<float>());
+    float *a = static_cast<float *>(
+        paddle_mobile::memory::Alloc(sizeof(float) * numel));
+    int index = 0;
+    for (int j = 0; j < n; j++) {
+      for (int i = 0; i < m; i++) {
+        a[index++] = tmp[i * n + j];
+      }
+    }
+#ifdef _OPENMP
+    Sgemm_omp(M, N, K, alpha, a, K, matrix_b.data<float>(), N, beta,
+              matrix_out->data<float>(), N, relu, bias);
+#else
+    Sgemm(M, N, K, alpha, a, K, matrix_b.data<float>(), N, beta,
+          matrix_out->data<float>(), N, relu, bias);
+#endif
+  } else {
 #ifdef _OPENMP
     Sgemm_omp(M, N, K, alpha, matrix_a.data<float>(), K, matrix_b.data<float>(),
               N, beta, matrix_out->data<float>(), N, relu, bias);
@@ -43,6 +64,7 @@ void matmul<float>(const framework::Tensor &matrix_a, bool trans_a,
     Sgemm(M, N, K, alpha, matrix_a.data<float>(), K, matrix_b.data<float>(), N,
           beta, matrix_out->data<float>(), N, relu, bias);
 #endif
+  }
 }
 
 template <>