Fix conv_fill_spec_pad.mlir #706
Conversation
Well, I still see some failures for this test https://github.com/nod-ai/iree-amd-aie/actions/runs/10567875494/job/29277762727?pr=706#step:9:11037. But then it passed after repeating. So I guess now it doesn't consistently fail but only flaky. |
|
Ah ok I thought this one failed consistently on OSX and didn't have 5 chances to pass |
| // %conv_pre_contract = | ||
| // transform.structured.match ops{["linalg.conv_1d_ncw_fcw"]} | ||
| // in %fused_for_all : (!any) -> !any | ||
|
|
||
| transform.structured.vectorize %conv_pre_contract : !any | ||
| // transform.structured.vectorize %conv_pre_contract : !any |
There was a problem hiding this comment.
Any reason to save this as a comment rather than just remove?
There was a problem hiding this comment.
Uh, I was just trying to see if this part of the codes caused flaky crash, but seems not. Without these, it still have some failures. So I'm going to keep blocking this test for MacOS Ci until figuring out the reason.
Is there a MacOS machine that I can use to debug the CI failure?
|
Q: why not just get rid of it? Personally I'm not a fan of transform schedules but that's just me 🤷 |
This change will get rid of the warning
conv_fill_spec_pad.mlir:57:3: warning: op argument #0 is not consumed in the body but is marked as consumed.Also attempt to fix #704.