search

intel / torch-xpu-ops

Apache License 2.0
30 stars 21 forks source link

Avoid early exit for bach norm CNL backward kernel to resolve hang issue #940

Closed xytintel closed 1 month ago

xytintel commented 1 month ago

Resolve issue: https://github.com/pytorch/pytorch/issues/136007

xytintel commented 1 month ago

@dvrogozh Double checked, out of the above 3 kernels, only BatchNormBackwardReduceChannelsLastKernelFunctor need barrier for reduction.