pre-commit: PR140954

[dtcxzyw]

Link: llvm/llvm-project#140954
Requested by: @dianqk

[dtcxzyw]

Diff mode

runner: ariselab-64c-docker
baseline: llvm/llvm-project@eec9431
patch: llvm/llvm-project#140954
sha256: 5f94f226371a64619f2eebb31f158aa68fba608e49f8df28f65b6af5e7931084
commit: cf1c19a

133 files changed, 125926 insertions(+), 126489 deletions(-)

Improvements:
  memcpyopt.NumCpyToSet 11025 -> 11300 +2.49%
  dse.NumCompletePartials 26009 -> 26013 +0.02%
  memcpyopt.NumMemSetInfer 385805 -> 385855 +0.01%
  dse.NumGetDomMemoryDefPassed 1408689 -> 1408817 +0.01%
  memcpyopt.NumCallSlot 1181139 -> 1181182 +0.00%
  dse.NumCFGTries 58312 -> 58314 +0.00%
  dse.NumFastStores 1198011 -> 1198045 +0.00%
  memcpyopt.NumMemCpyInstr 2027856 -> 2027888 +0.00%
  dse.NumCFGChecks 651895 -> 651905 +0.00%
  simplifycfg.NumHoistCommonInstrs 2489980 -> 2490002 +0.00%
Regressions:
  dse.NumRedundantStores 37594 -> 37591 -0.01%
  basicaa.SearchLimitReached 40218 -> 40216 -0.00%
  memcpyopt.NumStackMove 100779 -> 100776 -0.00%
  deadargelim.NumArgumentsReplacedWithPoison 105927 -> 105925 -0.00%
  simplifycfg.NumSinkCommonCode 389332 -> 389325 -0.00%
  simplifycfg.NumSinkCommonInstrs 833803 -> 833789 -0.00%
  correlated-value-propagation.NumNonNull 12816303 -> 12816247 -0.00%
  sroa.NumNewAllocas 30007668 -> 30007540 -0.00%
  sroa.NumAllocaPartitionUses 290732988 -> 290731967 -0.00%
  memdep.NumCacheCompleteNonLocalPtr 5614412 -> 5614394 -0.00%

2 2 bench/abseil-cpp/optimized/commandlineflag_test.ll
126 128 bench/abseil-cpp/optimized/usage_config.ll
4 4 bench/arrow/optimized/vector_selection.ll
63 78 bench/boost/optimized/ipv4_address_rule.ll
5 5 bench/boost/optimized/ipvfuture_rule.ll
14 16 bench/boost/optimized/matches_relation_factory.ll
9 21 bench/bullet3/optimized/btGImpactBvh.ll
55 81 bench/bullet3/optimized/btReducedDeformableBodyHelpers.ll
103 155 bench/bullet3/optimized/btSoftBodyHelpers.ll
356 367 bench/ceres/optimized/line_search.ll
13 33 bench/clamav/optimized/headers.ll
4 4 bench/cmake/optimized/cmCTest.ll
2 2 bench/cmake/optimized/cmCTestBuildHandler.ll
32 34 bench/cmake/optimized/cmNinjaTargetGenerator.ll
11 11 bench/cmake/optimized/cmProcessTools.ll
6 6 bench/cmake/optimized/cmSystemTools.ll
2 3 bench/fish-rs/optimized/13v7qjk9d4erkgv5w97enxlzt.ll
6 7 bench/lean4/optimized/ir_interpreter.ll
3 7 bench/linux/optimized/drm_modes.ll
1 1 bench/llvm/optimized/Attributor.ll
13 13 bench/llvm/optimized/BitcodeReader.ll
56 29 bench/llvm/optimized/FastISel.ll
11 14 bench/llvm/optimized/LoopVectorizationLegality.ll
1 6 bench/llvm/optimized/NativeFunctionSymbol.ll
32 37 bench/llvm/optimized/NativeInlineSiteSymbol.ll
14 16 bench/llvm/optimized/ObjCARCOpts.ll
14 16 bench/llvm/optimized/SemaLookup.ll
2 7 bench/llvm/optimized/SymbolDumper.ll
10 7 bench/mitsuba3/optimized/rectangle.ll
23 40 bench/open3d/optimized/ViewTrajectory.ll
65 75 bench/opencv/optimized/serialization.ll
2 3 bench/openusd/optimized/unitTestDelegate.ll
5 6 bench/openusd/optimized/xformCache.ll
21 31 bench/pingora-rs/optimized/2ov174hlbjn2zukn47jy5qk00.ll
21 31 bench/pingora-rs/optimized/acx590v32pn6eo20biyujqd8g.ll
4 5 bench/yosys/optimized/abc.ll
4 5 bench/yosys/optimized/cellmatch.ll
159 159 bench/z3/optimized/drat_frontend.ll
1 4 bench/zed-rs/optimized/11rrvqb0alhs5mh4wxxke3etr.ll

[github-actions]

Here is a high-level summary of the major changes in the provided LLVM IR diff:

1. Reduction in Memset Size:
   In commandlineflag_test.ll, the size of the llvm.memset call was reduced from 160 bytes to 128 bytes. This suggests an optimization in memory initialization, possibly due to more precise size analysis or struct layout changes.

2. TBAA Tag Updates:
   Several accesses previously tagged with TBAA (Type-Based Alias Analysis) metadata !tbaa !15 were updated to use !tbaa !14. This indicates that the type information for these accesses has been reclassified, which can improve alias analysis and enable further optimizations.

3. Elimination of Unused Allocations and Copies:
   In multiple files (ipvfuture_rule.ll, btSoftBodyHelpers.ll, etc.), unnecessary stack allocations like .sroa.0209 and associated memsets and memcpys were removed. These changes reflect SROA (Scalar Replacement of Aggregates) and dead code elimination improvements, reducing stack usage and simplifying control flow.

4. Loop Vectorization and Simplification:
   There are significant reductions in loop body complexity, particularly in btGImpactBvh.ll and btReducedDeformableBodyHelpers.ll, where manual vectorized memcpy/memset operations were replaced by fewer, larger calls. This implies better loop unrolling or vectorization passes.

5. Function Call and PHI Node Adjustments:
   Some landing pad PHIs and function calls had their operands adjusted or reordered (e.g., in btReducedDeformableBodyHelpers.ll). These changes may result from improved control flow graph (CFG) simplification or cleanup during exception handling lowering.

These changes primarily represent internal cleanups, better memory usage, and optimization pass improvements—none introduce new functionality but instead refine existing code generation and data layout.

model: qwen-plus-latest
CompletionUsage(completion_tokens=409, prompt_tokens=109110, total_tokens=109519, completion_tokens_details=None, prompt_tokens_details=None)