Support npu kernel for eye op #34543
Conversation
|
Thanks for your contribution! |
| @@ -0,0 +1,105 @@ | |||
| # Copyright (c) 2019 PaddlePaddle Authors. All Rights Reserved. | |||
paddle/fluid/operators/eye_op_npu.cc
Outdated
| See the License for the specific language governing permissions and | ||
| limitations under the License. */ | ||
|
|
||
| #include "paddle/fluid/operators/crop_op.h" |
paddle/fluid/operators/eye_op_npu.cc
Outdated
| PADDLE_ENFORCE_EQ( | ||
| num_rows >= 0, true, | ||
| platform::errors::InvalidArgument( | ||
| "The value of Input(num_rows) should be non-negative int.")); |
There was a problem hiding this comment.
这个不需要检查，在EyeOp::InferShape的时候已经检查过了。
paddle/fluid/operators/eye_op_npu.cc
Outdated
| num_columns >= 0, true, | ||
| platform::errors::InvalidArgument( | ||
| "The value of Input(num_columns) should be non-negative int.")); | ||
|
|
paddle/fluid/operators/eye_op_npu.cc
Outdated
|
|
||
| if (ctx.HasAttr("num_columns")) { | ||
| num_columns = ctx.Attr<int64_t>("num_columns"); | ||
| } |
There was a problem hiding this comment.
这个最好参考eye_op.h通过判断num_columns是否等于-1，避免单测或者python API把这个num_columns设置为-1的情况。
| self.num_rows = 100 | ||
| self.num_columns = 100 | ||
|
|
||
|
|
There was a problem hiding this comment.
少了一个dtype是fp32的单测，另外参考 test_eye_op.py 的 class API_TestTensorEye(unittest.TestCase)，增加关于静态图API和动态图API的测试。
| expected_result = np.eye(10, dtype="int32") | ||
| self.assertEqual((result == expected_result).all(), True) | ||
|
|
||
| paddle.disable_static() |
There was a problem hiding this comment.
这里可以写成paddle.disable_static(paddle.NPUPlace(0)) 就可以把动态图的api跑到NPU上了
| paddle.enable_static() | ||
| self.assertEqual((out.numpy() == expected_result).all(), True) | ||
|
|
||
| paddle.disable_static() |
| True) | ||
| self.assertEqual((out.numpy() == expected_result).all(), True) | ||
|
|
||
| paddle.disable_static() |
| if (ctx.HasAttr("num_columns")) { | ||
| num_columns = ctx.Attr<int64_t>("num_columns"); | ||
| } | ||
| if (num_columns == -1) num_columns = num_rows; |
There was a problem hiding this comment.
33-37行合并为以下2行就可以，因为在python端的def eye中输入的num_columns不是None，因此ctx.HasAttr("num_columns")在API调用端永远是true.
auto num_columns = ctx.Attr<int64_t>("num_columns");
if (num_columns == -1) num_columns = num_rows;
3 participants
PR types
New features
PR changes
OPs
Describe
add support npu kernel for eye op
