Do not apply these explicit unroll pragmas for ROCm. (#18204)
authorJohannes M Dieterich <johannes.dieterich@amd.com>
Wed, 20 Mar 2019 14:58:11 +0000 (07:58 -0700)
committerFacebook Github Bot <facebook-github-bot@users.noreply.github.com>
Wed, 20 Mar 2019 15:06:07 +0000 (08:06 -0700)
commit4741d613eed2588a13c0f7a331500716393de3de
tree62584ad1606fe9d72b69a4229cd8fa04a91a526c
parent8f1db1c6c177203ef73c1bd59116ed0c3c4720b0
Do not apply these explicit unroll pragmas for ROCm. (#18204)

Summary:
Loop analysis indicates that there is a runtime trip count and hence
unrolling cannot take place.

This will silence compile-time warnings we have been observing with recent ROCm releases.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18204

Differential Revision: D14539875

Pulled By: ezyang

fbshipit-source-id: a7ea7f2a95603754296b76a6b62a154f56f4ad4d
aten/src/ATen/native/cuda/WeightNorm.cu
aten/src/THC/THCSortUtils.cuh
aten/src/THCUNN/SpatialDepthwiseConvolution.cu
caffe2/operators/top_k_heap_selection.cuh