[all-commits] [llvm/llvm-project] c9d9c3: [InstCombine] Fold `icmp pred X + K, Y -> icmp pre...

[Yingwei Zheng via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: c9d9c3e349848a63c67271bcef29cda218042bc0
      https://github.com/llvm/llvm-project/commit/c9d9c3e349848a63c67271bcef29cda218042bc0
  Author: Yingwei Zheng <dtcxzyw2333 at gmail.com>
  Date:   2025-07-05 (Sat, 05 Jul 2025)

  Changed paths:
    M llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
    M llvm/test/Transforms/InstCombine/icmp.ll

  Log Message:
  -----------
  [InstCombine] Fold `icmp pred X + K, Y -> icmp pred2 X, Y` if both X and Y is divisible by K (#147130)

This patch generalizes `icmp ule X +nuw 1, Y -> icmp ult X, Y`-like
optimizations to handle the case that the added RHS constant is a common
power-of-2 divisor of both X and Y. We can further generalize this
pattern to handle non-power-of-2 divisors as well.
Alive2: https://alive2.llvm.org/ce/z/QgpeM_

Compile-time improvement (Stage2-O3 -0.09%):
https://llvm-compile-time-tracker.com/compare.php?from=0ba59587fa98849ed5107fee4134e810e84b69a3&to=f80e5fe0bb2e63c05401bde7cd42899ea270909b&stat=instructions:u

The original case is from the comparison of expanded GEP offsets:
https://github.com/dtcxzyw/llvm-opt-benchmark/pull/2530/files#r2183005292



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