提交 b5996fa1 编写于 作者: Y Yu Yang

Fix unstable selected_rows_functor_test.cu

上级 d402234b
...@@ -107,7 +107,7 @@ struct SelectedRowsAddTensor<platform::CUDADeviceContext, T> { ...@@ -107,7 +107,7 @@ struct SelectedRowsAddTensor<platform::CUDADeviceContext, T> {
PADDLE_ENFORCE_EQ(in1_height, out_dims[0]); PADDLE_ENFORCE_EQ(in1_height, out_dims[0]);
auto& in1_value = input1.value(); auto& in1_value = input1.value();
framework::Vector<int64_t> in1_rows(input1.rows()); auto& in1_rows = input1.rows();
int64_t in1_row_numel = in1_value.numel() / in1_rows.size(); int64_t in1_row_numel = in1_value.numel() / in1_rows.size();
PADDLE_ENFORCE_EQ(in1_row_numel, input2.numel() / in1_height); PADDLE_ENFORCE_EQ(in1_row_numel, input2.numel() / in1_height);
...@@ -206,7 +206,7 @@ struct SelectedRowsAddToTensor<platform::CUDADeviceContext, T> { ...@@ -206,7 +206,7 @@ struct SelectedRowsAddToTensor<platform::CUDADeviceContext, T> {
PADDLE_ENFORCE_EQ(in1_height, in2_dims[0]); PADDLE_ENFORCE_EQ(in1_height, in2_dims[0]);
auto& in1_value = input1.value(); auto& in1_value = input1.value();
framework::Vector<int64_t> in1_rows(input1.rows()); auto& in1_rows = input1.rows();
int64_t in1_row_numel = in1_value.numel() / in1_rows.size(); int64_t in1_row_numel = in1_value.numel() / in1_rows.size();
PADDLE_ENFORCE_EQ(in1_row_numel, input2->numel() / in1_height); PADDLE_ENFORCE_EQ(in1_row_numel, input2->numel() / in1_height);
......
...@@ -20,7 +20,9 @@ limitations under the License. */ ...@@ -20,7 +20,9 @@ limitations under the License. */
TEST(selected_rows_functor, gpu_add) { TEST(selected_rows_functor, gpu_add) {
paddle::platform::CUDAPlace gpu_place(0); paddle::platform::CUDAPlace gpu_place(0);
paddle::platform::CPUPlace cpu_place; paddle::platform::CPUPlace cpu_place;
paddle::platform::CUDADeviceContext ctx(gpu_place); paddle::platform::CUDADeviceContext& ctx =
*reinterpret_cast<paddle::platform::CUDADeviceContext*>(
paddle::platform::DeviceContextPool::Instance().Get(gpu_place));
paddle::operators::math::SetConstant<paddle::platform::CUDADeviceContext, paddle::operators::math::SetConstant<paddle::platform::CUDADeviceContext,
float> float>
functor; functor;
...@@ -132,7 +134,9 @@ TEST(selected_rows_functor, gpu_add) { ...@@ -132,7 +134,9 @@ TEST(selected_rows_functor, gpu_add) {
TEST(selected_rows_functor, gpu_add_to) { TEST(selected_rows_functor, gpu_add_to) {
paddle::platform::CUDAPlace gpu_place(0); paddle::platform::CUDAPlace gpu_place(0);
paddle::platform::CPUPlace cpu_place; paddle::platform::CPUPlace cpu_place;
paddle::platform::CUDADeviceContext ctx(gpu_place); paddle::platform::CUDADeviceContext& ctx =
*reinterpret_cast<paddle::platform::CUDADeviceContext*>(
paddle::platform::DeviceContextPool::Instance().Get(gpu_place));
paddle::operators::math::SetConstant<paddle::platform::CUDADeviceContext, paddle::operators::math::SetConstant<paddle::platform::CUDADeviceContext,
float> float>
functor; functor;
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册