ptrendx commented on a change in pull request #17462: Updated PartialSortSmallK for LT support URL: https://github.com/apache/incubator-mxnet/pull/17462#discussion_r381431132
########## File path: src/operator/tensor/ordering_op-inl.h ########## @@ -362,7 +362,7 @@ MSHADOW_FORCE_INLINE void TopKSort(const Tensor<gpu, 1, DType>& dat, } } else { const int nthreads(mshadow::cuda::kBaseThreadNum); - PartialSortSmallK<<<M, nthreads, nthreads*K*(sizeof(int)+sizeof(DType)), + PartialSortSmallK<<<M, nthreads, nthreads*K*(sizeof(IType)+sizeof(DType)), Review comment: Yeah, I don't believe the increase of GPU memory usage that you see comes from this particular change. Shared memory is part of L1 cache of the GPU, it is not counted towards the DRAM size. ---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services