[Mlir-commits] [mlir] [mlir][linalg] Support subtracting accumulation in partial reduction … (PR #214033)

[llvmlistbot at llvm.org]

pstarkcdpr wrote:

Thanks @banach-space. I have also never seen overflow flags in the wild, so I'm personally not affected by this. I'm reasonably convinced that maintaining overflow flags in this case would be wrong, so I'm thinking that we should err on the side of caution and bail until such time that a larger discussion thinks that it's safe. The original issue is actively causing errors on a current model for me (in IREE), so I'd rather not delay. I figure that it's better to fix the active issue and be conservative for now. Let me know if you disagree and I can change the code to propagate the flags for consistency with the other reducers.

I'll try find some time to address your previous review comments in the next day or two.

Unfortunately I can't start a thread on Discord since Discord is disallowed due to company security policies. If there's another avenue for discussion, I can summarize my findings there.

https://github.com/llvm/llvm-project/pull/214033