pre-commit: PR170835

[zyw-bot]

Link: llvm/llvm-project#170835
Requested by: @nikic

[zyw-bot]

Diff mode

runner: ariselab-64c-docker
baseline: llvm/llvm-project@2612dc9
patch: llvm/llvm-project#170835
sha256: 72705fff5769d1e96e260b0cdeb54bb09d836a47558f1ec701658e074c07d04d
commit: fc7189d

507 files changed, 1326968 insertions(+), 1261564 deletions(-)

Improvements:
  simple-loop-unswitch.NumCostMultiplierSkipped 17240 -> 17555 +1.83%
  local.NumPHICSEs 186111 -> 186743 +0.34%
  simple-loop-unswitch.NumBranches 104878 -> 105194 +0.30%
  instsimplify.NumSimplified 2467295 -> 2472421 +0.21%
  lcssa.NumLCSSA 15128817 -> 15146370 +0.12%
  gvn.NumGVNSimpl 4415379 -> 4420427 +0.11%
  local.NumRemoved 5303715 -> 5308837 +0.10%
  scalar-evolution.NumExitCountsNotComputed 11931824 -> 11938277 +0.05%
  gvn.NumGVNEqProp 427030 -> 427233 +0.05%
  mem2reg.NumPHIInsert 12839942 -> 12845402 +0.04%
Regressions:
  correlated-value-propagation.NumSDivs 16936 -> 11395 -32.72%
  instcombine.NumConstProp 168277 -> 159101 -5.45%
  correlated-value-propagation.NumSExt 47282 -> 46564 -1.52%
  correlated-value-propagation.NumSRems 1574 -> 1564 -0.64%
  simplifycfg.NumFoldBranchToCommonDest 748129 -> 743925 -0.56%
  correlated-value-propagation.NumAShrsConverted 3485 -> 3467 -0.52%
  indvars.NumSameSign 40295 -> 40097 -0.49%
  correlated-value-propagation.NumPhis 1237642 -> 1233030 -0.37%
  correlated-value-propagation.NumAddNUW 212244 -> 211468 -0.37%
  correlated-value-propagation.NumAddNW 428616 -> 427438 -0.27%

+6 llvm/RegAllocGreedy.ll
+4 boost/xml_grammar.ll
+4 verilator/V3OrderParallel.ll
+2 gromacs/tngio.ll
+2 meshlab/arap.ll
+2 meshlab/filter_parametrization.ll
+1 cxxopts/example.ll
+1 git/cache-tree.ll
+0 abc/abcExact.ll
+0 assimp/TriangulateProcess.ll
+0 brotli/compress_fragment.ll
+0 brotli/encode.ll
+0 casadi/cvodea.ll
+0 casadi/idaa.ll
+0 coremark/core_list_join.ll
+0 cpython/unicodeobject.ll
+0 duckdb/encode.ll
+0 duckdb/utf8proc.ll
+0 graphviz/blockpath.ll
+0 hwloc/topology-synthetic.ll
+0 libigl/exact_geodesic.ll
+0 libquic/s3_enc.ll
+0 linux/scsi.ll
+0 oiio/exrinput_c.ll
+0 openjdk/hb-blob.ll
+0 openjdk/hb-ot-layout.ll
+0 openssl/tls13encryptiontest.ll
+0 openusd/faceVertex.ll
+0 pola-rs/cztbx1zaxsgh33rhqav0blwq7.ll
+0 pola-rs/etaacdu7pg0nazmqnydgyjsn2.ll
+0 ruby/dir.ll
+0 ruff-rs/0jxokf4xx9lz4dh6x5yfoh8sv.ll
+0 ruff-rs/avzuiha0u13zvsyv1kglukilk.ll
+0 salsa-rs/bfvulplr01a8w79gvhbwzr5ai.ll
+0 sqlite/shell.ll
+0 sundials/cvodea.ll
+0 sundials/idaa.ll
+0 tev/StbiLdrImageSaver.ll
+0 wolfssl/benchmark.ll
-1 clamav/lzxd.ll
-1 clamav/mszipd.ll
-1 cpython/mpdecimal.ll
-1 ffmpeg/pngdec.ll
-1 llvm/SemaOverload.ll
-1 openjdk/check_classname.ll
-1 postgres/fastpath.ll
-1 qemu/virtio-pci.ll
-1 softposit-rs/5az6c15ag5q4gib5.ll
-1 typst-rs/5z4no3nnr5v1s13.ll
-2 bullet3/btContinuousConvexCollision.ll
-2 fmt/scan-test.ll
-2 hyperscan/ng_revacc.ll
-2 postgres/pl_gram.ll
-3 eastl/EASprintfCore.ll
-3 qemu/virtio.ll
-3 slurm/slurmdb_defs.ll
-4 abseil-cpp/distribution_test_util.ll
-4 clamav/tnef.ll
-4 cpython/longobject.ll
-4 git/tree-walk.ll
-4 llvm/ItaniumMangle.ll
-4 velox/BaseVector.ll
-4 zed-rs/3puy2qvhvlnhff2105q9h6j8i.ll
-6 postgres/fe-exec.ll
-6 velox/DenseHll.ll
-6 wireshark/androiddump.ll
-7 eastl/BenchmarkAlgorithm.ll
-8 git/walker.ll
-8 softposit-rs/4wrr62dd33canpwl.ll
-10 abseil-cpp/cord_rep_btree.ll
-13 llvm/APINotesReader.ll
-13 llvm/ASTReader.ll
-13 llvm/GlobalModuleIndex.ll
-15 lean4/name.ll

[github-actions]

Here's a concise summary of the major changes in this LLVM IR diff:

1. Loop Header Preheader Adjustments: Multiple loops had their preheader references updated (e.g., %.lr.ph.us.preheader.i, %.lr.ph.us.preheader.i80) to correctly point to newly introduced or renamed preheaders, improving loop structure consistency.

2. Phi Node Operand Updates: Phi instructions across several functions were updated to reference correct predecessor blocks and values—particularly after loop restructuring—ensuring accurate SSA form and dataflow (e.g., in cord_rep_btree.ll, core_list_join.ll, tnef.ll).

3. Memory Allocation & Initialization Refactoring: In cord_rep_btree.ll, heap allocation (@_Znwm) and initialization logic was reordered: the allocation is now performed after a conditional branch, and stores to the allocated memory use the new pointer instead of an old one—improving correctness and enabling better optimization.

4. Integer Arithmetic Refinements: Several places added nuw (no unsigned wrap) and/or nsw (no signed wrap) flags to arithmetic operations (e.g., shl, add, lshr) for stronger undefined-behavior guarantees and better optimization opportunities (e.g., shl nuw nsw, add nuw nsw, icmp samesign ult).

5. Control Flow Simplification & Cleanup: Redundant basic blocks (e.g., .thread, .split, .lr.ph.i.i.i.i.i.i) were removed or merged; unnecessary branches and phi inputs were pruned (e.g., in eastl/BenchmarkAlgorithm.ll, fmt/scan-test.ll, git/tree-walk.ll), reducing IR complexity and improving readability.

These changes collectively reflect ongoing compiler optimizations—especially around loop canonicalization, memory safety, integer semantics, and control-flow simplification—without altering observable program behavior.

model: qwen-plus-latest
CompletionUsage(completion_tokens=408, prompt_tokens=111002, total_tokens=111410, completion_tokens_details=None, prompt_tokens_details=None)