未验证 提交 15b05c75 编写于 作者: A Aurelius84 提交者: GitHub

[Custom Op]Remove PADDLE_WITH_MKLDNN in custom_op (#32903)

上级 c4787d76
...@@ -469,10 +469,6 @@ def normalize_extension_kwargs(kwargs, use_cuda=False): ...@@ -469,10 +469,6 @@ def normalize_extension_kwargs(kwargs, use_cuda=False):
########################### -- END -- ########################### ########################### -- END -- ###########################
add_compile_flag(extra_compile_args, ['-w']) # disable warning add_compile_flag(extra_compile_args, ['-w']) # disable warning
# Note(Aurelius84): This marco will impact memory layout of `Tensor`.
# We align it automatically with pre-installed Paddle.
if core.is_compiled_with_mkldnn():
add_compile_flag(extra_compile_args, ['-DPADDLE_WITH_MKLDNN'])
if use_cuda: if use_cuda:
extra_link_args.append('-lcudart') extra_link_args.append('-lcudart')
......
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