junrushao commented on code in PR #11639:
URL: https://github.com/apache/tvm/pull/11639#discussion_r970271772
##########
src/meta_schedule/schedule_rule/cross_thread_reduction.cc:
##########
@@ -184,6 +184,13 @@ class CrossThreadReductionNode : public ScheduleRuleNode {
*/
std::tuple<bool, tir::LoopRV, tir::BlockRV, tir::LoopRV>
GetComputeTargetLoopAndBlock(
const tir::Schedule& sch, const tir::BlockRV& block_rv) {
+ // Step 0. Due to technical reason of some primitives (e.g., compute-at),
if the block is doing
+ // a tuple reduction, fusion is temporarily not supported.
Review Comment:
nit: let's improve the doc a bit by being more specific, like, "fusion with
the subsequent consumer is not allowed"
--
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.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
