- User Since
- Sep 26 2016, 7:58 AM (295 w, 4 d)
Fri, May 13
Thu, May 12
Mon, May 9
Drive-by-nit: Just happened to notice that the commit msg says "RagAlloc...".
Tue, May 3
This seem to help out avoiding complaints from clangsa that the OpTy passed on to getCastInstrCost could be a nullptr (implementations of getCastInstrCost doesn't seem to check that the Src/Dst Type pointers are non-null. One could perhaps question the API (if those pointers must not be null then I guess one could have passed the Type arguments by reference instead of passing the pointer). This patch in itself seem good anyway.
Thu, Apr 28
Calculate the ExtendCode outside the loop (also no longer needing the extra bool
to determine how we extend inside the loop).
Shorten a variable name according to review feedback.
Apr 26 2022
Apr 17 2022
Thanks for the patch. I've verified that it solves the problem I found.
Apr 16 2022
Seen assertion failures downstream with this patch. Here is a reproducer:
Apr 4 2022
Have had problems with failing stage 2 builds since this patch:
Mar 23 2022
Looked around a bit.
Hello. We've got some problem in our downstream tests after this patch and I'm trying to figure out how things are supposed to work. Maybe someone being part of this review knows.
Mar 22 2022
Mar 14 2022
Mar 10 2022
Sorry, I don't know much about bugpoint to understand if this is feasible or not. I doubt that the CFG simplification done by bugpoint really need to take target specific costs etc into account? We just want to simplify the flow, but not optimize it for a given target, right?
I don't find the motivation that strong and there are no test cases etc included that show any gains.
Just a few nits. But otherwise this LGTM.
Mar 9 2022
Feb 17 2022
Feb 16 2022
Fixed clang-format failures.
(let me know if I should try to find alternative reviewers)
Feb 7 2022
I'm far from an expert at this (I don't know that much about TokenFactors and the DAGCombiner worklist).
But one thing that caught my attention was the code that https://github.com/llvm/llvm-project/issues/52858 is talking about in visitTokenFactor:
Rename the new test case.
Feb 6 2022
Updated after review feedback:
- Now using APInt for comparisons.
- For USHLSAT we now compare minimum number of known leading zeroes to the shift amount when checking if all shifted out bits are zero. That is hopefully easier to read/understand compared to using MaskedValueIsZero.
Feb 4 2022
Feb 3 2022
Fix typo (there was an extra "Scale =" that wasn't supposed to be there).
Feb 2 2022
Added check that it is "legal" to use SHL.
Feb 1 2022
Jan 31 2022
Added test cases for undef/poison/out-of-bounds-shamt:
(shlsat undef/poison, x) -> 0 (shlsat x, undef/poison) -> undef (shlsat x, too_large_shamt) -> undef
Jan 28 2022
Jan 27 2022
Abandoned in favor of D118369 which adds the passes to the registry instead. That seems like a nicer solution.
Updated based on the suggestion to drop -aa-pipeline=basic-aa in test cases for which the output would be the same when using the default aa-pipeline (that was the case for all but one file, so that one was left out from this patch).
Updated based on review feedback.