[all-commits] [llvm/llvm-project] 9a2d9c: [SCEV] Try to fold ZExt/SExt of AddRec before exis...

[Florian Hahn via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 9a2d9c7fef47c0af321150f75d8f21db06cccb43
      https://github.com/llvm/llvm-project/commit/9a2d9c7fef47c0af321150f75d8f21db06cccb43
  Author: Florian Hahn <flo at fhahn.com>
  Date:   2026-07-17 (Fri, 17 Jul 2026)

  Changed paths:
    M llvm/lib/Analysis/ScalarEvolution.cpp
    M llvm/test/Analysis/ScalarEvolution/exit-count-non-strict.ll
    M llvm/test/Analysis/ScalarEvolution/incorrect-exit-count.ll
    M llvm/test/CodeGen/AMDGPU/GlobalISel/divergence-divergent-i1-used-outside-loop.ll
    M llvm/test/CodeGen/AMDGPU/GlobalISel/divergence-structurizer.ll
    M llvm/test/CodeGen/AMDGPU/memset-pattern.ll
    M llvm/test/CodeGen/ARM/select-imm.ll
    A llvm/test/Transforms/IndVarSimplify/iv-widen-addrec-flag-inferred-late.ll

  Log Message:
  -----------
  [SCEV] Try to fold ZExt/SExt of AddRec before existing lookup. (#208805)

Move AddRec fold for ZExt & SExt before existing SCEV lookup. This can
improve results, when we constructed SCEV for a ZExt/SExt during earlier
analysis and flags on the operand AddRec have not been
refined/strengthend yet.

This improves results in some cases and fixes a regression here:
https://github.com/dtcxzyw/llvm-opt-benchmark-nightly/pull/654#discussion_r3554885279.

On a large C/C++ corpus with 32k bitcode files, ~2% more IVs get
widened, removing a large number of ZExt instructions.

Compile-time impact in the noise

https://llvm-compile-time-tracker.com/compare.php?from=167d65906cc3c636221158abe8289ab4a03dfc7d&to=381abca0d5ead4443d86287d754f47c21396a9be&stat=instructions:u



To unsubscribe from these emails, change your notification settings at https://github.com/llvm/llvm-project/settings/notifications