pre-commit: PR147180

[zyw-bot]

Link: llvm/llvm-project#147180
Requested by: @dtcxzyw

[zyw-bot]

Diff mode

runner: ariselab-64c-docker
baseline: llvm/llvm-project@db4e927
patch: llvm/llvm-project#147180
sha256: ff2bc55c4a7bb17df56acb501e5f664f9c6725e197bbca3b4675b31306f88093
commit: 83544bc

819 files changed, 3051 insertions(+), 3051 deletions(-)

Improvements:
Regressions:
  instcombine.NumCombined 130200074 -> 130185765 -0.01%

1 1 bench/abc/optimized/abcNpn.ll
2 2 bench/actix-rs/optimized/19s5ghr2x6s6bmh4.ll
3 3 bench/bullet3/optimized/btSoftBody.ll
8 8 bench/git/optimized/gc.ll
5 5 bench/glslang/optimized/hlslParseHelper.ll
13 13 bench/grpc/optimized/promise_based_filter.ll
28 28 bench/hermes/optimized/Triple.ll
25 25 bench/image-rs/optimized/2ndzmzcdt55acj4k.ll
61 61 bench/llvm/optimized/AArch64ISelLowering.ll
202 202 bench/llvm/optimized/AttrImpl.ll
38 38 bench/llvm/optimized/Consumed.ll
11 11 bench/llvm/optimized/DeclSpec.ll
43 43 bench/llvm/optimized/LegalizerHelper.ll
20 20 bench/llvm/optimized/OpenMPKinds.ll
82 82 bench/llvm/optimized/SymbolGraphSerializer.ll
32 32 bench/llvm/optimized/Triple.ll
108 108 bench/llvm/optimized/TypeLoc.ll
31 31 bench/llvm/optimized/X86ISelLowering.ll
4 4 bench/miniaudio/optimized/unity.ll
15 15 bench/node/optimized/libnode.cares_wrap.ll
10 10 bench/openusd/optimized/listOp.ll
12 12 bench/php/optimized/simplexml.ll
22 22 bench/postgres/optimized/postmaster.ll
7 7 bench/pugixml/optimized/pugixml.ll
6 6 bench/ruby/optimized/prism.ll
9 9 bench/tls-rs/optimized/59h61akxu6z29dlt.ll
29 29 bench/typst-rs/optimized/1ru1rhojhbz2vfey.ll
26 26 bench/typst-rs/optimized/3rk2ctuzbghb17s4.ll
16 16 bench/typst-rs/optimized/49m3cs7hus53ztof.ll
17 17 bench/typst-rs/optimized/4qskctz4kwc33g7b.ll
14 14 bench/typst-rs/optimized/59tuvc5m3xlovl3o.ll
30 30 bench/typst-rs/optimized/d6l9ieo9tcw33dn.ll
87 87 bench/wasmtime-rs/optimized/2ly4gzztxx8hlwxv.ll
40 40 bench/wasmtime-rs/optimized/3r0osxvwe4cd326n.ll

[github-actions]

The patch applies the nneg (non-negative) attribute to multiple zext (zero extend) operations in LLVM IR code. Here's a high-level summary of the changes:

1. Addition of nneg Attribute to Zero-Extend Operations:
   The main change is adding the nneg flag to zext instructions across various functions. This indicates that the input value to be zero-extended is known to be non-negative, allowing the optimizer to avoid checks or transformations related to negative values.

2. Targeted Value Types:

   • Most of the modified zext operations convert i8 or i16 to i64, often after an add nsw operation subtracts 1 or another small constant from the original index.
   • These conversions typically occur as part of switch table indexing logic where bounds are expected to be controlled by earlier checks.

3. Context: Switch Table Indexing:

   • Many of the zext instructions appear when calculating indices into switch tables, which are used for efficient dispatch based on enum-like values.
   • By marking these indices as nneg, the compiler confirms they cannot be negative, reinforcing safety and potentially improving generated code efficiency.

4. Performance Optimization:

   • These changes enable better optimization opportunities by giving the backend more precise information about value ranges.
   • Specifically, it allows removal of redundant range checks and enables use of more optimal instruction sequences for indexing.

5. Wide Scope Across Benchmarks:

   • The modification appears in multiple benchmarks including:
     • abcNpn, btSoftBody, git, image-rs, glslang
     • LLVM's AArch64ISelLowering, AttrImpl, Consumed, DeclSpec
   • These files represent a diverse set of applications and libraries, indicating this is a general optimization pattern being applied systematically.

In summary:
This patch improves code generation and analysis by annotating zext instructions with nneg, reflecting that their inputs are guaranteed non-negative—especially useful when computing indices into switch tables. It enhances optimization potential without altering functionality.

model: qwen-plus-latest
CompletionUsage(completion_tokens=456, prompt_tokens=103363, total_tokens=103819, completion_tokens_details=None, prompt_tokens_details=None)