mbrookhart commented on a change in pull request #6839:
URL: https://github.com/apache/tvm/pull/6839#discussion_r540658345
##########
File path: python/tvm/topi/cuda/sort.py
##########
@@ -103,9 +103,9 @@ def sort_ir(data, values_out, axis, is_ascend,
indices_out=None):
nthread_bx = shape[axis] // max_threads + 1
tx = te.thread_axis("threadIdx.x")
- bx = te.thread_axis("vthread")
+ bx = te.thread_axis("blockIdx.x")
ib.scope_attr(tx, "thread_extent", nthread_tx)
- ib.scope_attr(bx, "virtual_thread", nthread_bx)
+ ib.scope_attr(bx, "thread_extent", nthread_bx)
Review comment:
I'm working on a fuller rewrite of the sort kernel for stabilty in
another branch, I understand why this isn't fast or stable, and I can do better
for small inputs, but I need to rework things more to support full sized inputs.
----------------------------------------------------------------
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:
[email protected]
