Issue |
91467
|
Summary |
[SLP] Vectorizer runs forever after PR #88504
|
Labels |
new issue
|
Assignees |
|
Reporter |
chsigg
|
After 37ae4ad0eef338776c7e2cffb3896153d43dcd90, the `vectorize-slp` pass does not complete on [this IR](https://github.com/llvm/llvm-project/files/15248904/repro.ll.txt) produced by Triton with
```
LLVM_IR_ENABLE_DUMP=1 pytest --verbose --capture=no \
python/test/unit/hopper/test_persistent_warp_specialized_gemm.py \
-k '128-128-128-4-2-360-4096-512-False-True-none-float32-True-3'
```
This blocks upgrading Triton to the latest LLVM version (https://github.com/openai/triton/pull/3818).
I'm not familiar with the vectorizer code and it's not obvious to me what's going wrong. I would therefore highly appreciate any help looking into the issue.
_______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs