[llvm] [InstCombine] Match swapped form of truncating saturation clamp (PR #226614)

[via llvm-commits]

llvmorg-github-actions[bot] wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-llvm-transforms

Author: Craig Topper (topperc)

<details>
<summary>Changes</summary>

Extend the fold added in #<!-- -->189703 to also handle the inverted select:
trunc (select (icmp ugt A, DestTy_umax), sext(icmp sgt A, 0), A) -->
trunc (smin (smax (0, A), DestTy_umax))

InstCombine canonicalizes (A & NegPow2) != 0 into the ult form with
swapped select operands, but if SCCP first rewrites the compare as
icmp uge A, C, InstCombine only turns it into icmp ugt and never swaps
the select, so the original fold is missed.

While here, match the compare constant with m_APInt instead of
m_Constant + getUniqueInteger, and build TruncatedMax directly with
APInt::getLowBitsSet. Comparing the constant against TruncatedMax + 1
or TruncatedMax makes the separate zero check unnecessary.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@<!-- -->anthropic.com>

---
Full diff: https://github.com/llvm/llvm-project/pull/226614.diff


2 Files Affected:

- (modified) llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp (+19-12) 
- (modified) llvm/test/Transforms/InstCombine/truncating-saturate.ll (+95) 


``````````diff
The server is unavailable at this time. Please wait a few minutes before you try again.
``````````

</details>


https://github.com/llvm/llvm-project/pull/226614