sxjscience commented on a change in pull request #15560: Add fp16 support for topk URL: https://github.com/apache/incubator-mxnet/pull/15560#discussion_r317255526
########## File path: src/operator/tensor/sort_op-inl.cuh ########## @@ -227,18 +251,18 @@ SortByKeyImpl(mshadow::Tensor<gpu, 1, KDType> keys, CHECK_EQ(values.CheckContiguous(), true); #if CUDA_VERSION >= 9000 cudaStream_t stream = mshadow::Stream<gpu>::GetStream(keys.stream_); - thrust::device_ptr<half> key_iter = thrust::device_pointer_cast( - reinterpret_cast<half*>(keys.dptr_)); - thrust::device_ptr<half> value_iter = thrust::device_pointer_cast( - reinterpret_cast<half*>(values.dptr_)); + thrust::device_ptr<__half> key_iter = thrust::device_pointer_cast( + reinterpret_cast<__half*>(keys.dptr_)); + thrust::device_ptr<__half> value_iter = thrust::device_pointer_cast( + reinterpret_cast<__half*>(values.dptr_)); if (is_ascend) { thrust::stable_sort_by_key( thrust::cuda::par.on(stream), - key_iter, key_iter + (keys.size(0)), value_iter, cuda::less_half<half>()); + key_iter, key_iter + (keys.size(0)), value_iter.get(), cuda::less_half<__half>()); Review comment: Why adding `.get()` to the `value_iter`? ---------------------------------------------------------------- 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