masahi commented on a change in pull request #9737:
URL: https://github.com/apache/tvm/pull/9737#discussion_r768960030
##########
File path: python/tvm/relay/op/strategy/cuda.py
##########
@@ -324,7 +324,10 @@ def conv2d_strategy_cuda(attrs, inputs, out_type, target):
plevel=25,
)
- elif is_depthwise_conv2d(data.shape, layout, kernel.shape, kernel_layout,
groups):
+ elif (
+ is_depthwise_conv2d(data.shape, layout, kernel.shape, kernel_layout,
groups)
+ and "cudnn" not in target.libs
Review comment:
cuDNN requires a different kernel layout than AutoTVM when the input
layout is NHWC, in which case two implementations are not compatible.
But there is no problem with the NCHW layout, so I've refined the condition
to
```
elif is_depthwise_conv2d(data.shape, layout, kernel.shape,
kernel_layout, groups) and (
layout == "NCHW" or "cudnn" not in target.libs):
```
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
