pre-commit: PR173134

[zyw-bot]

Link: llvm/llvm-project#173134
Requested by: @nikic

[zyw-bot]

Diff mode

runner: ariselab-64c-docker
baseline: llvm/llvm-project@3eee451
patch: llvm/llvm-project#173134
sha256: 338be5c0cc9b89a6d591b43eb599d9197e0e89912af9a122a280ae43cd20fb4c
commit: 63e03b9

704 files changed, 245886 insertions(+), 245885 deletions(-)

Improvements:
  correlated-value-propagation.NumPhiCommon 52041 -> 52884 +1.62%
  correlated-value-propagation.NumPhis 1237050 -> 1237907 +0.07%
  dse.NumRedundantStores 36288 -> 36309 +0.06%
  licm.NumLoadPromoted 80248 -> 80267 +0.02%
  instcombine.NumCombined 120654744 -> 120668013 +0.01%
  licm.NumPromotionCandidates 586442 -> 586499 +0.01%
  memcpyopt.NumMemSetInfer 360877 -> 360909 +0.01%
  aggressive-instcombine.NumInstrsReduced 44901 -> 44904 +0.01%
  aggressive-instcombine.NumExprsReduced 15545 -> 15546 +0.01%
  adce.NumRemoved 95993 -> 95998 +0.01%
Regressions:
  indvars.NumReplaced 69579 -> 67325 -3.24%
  gvn.NumGVNEqProp 427689 -> 416757 -2.56%
  simple-loop-unswitch.NumTrivial 3256 -> 3237 -0.58%
  loop-simplifycfg.NumLoopExitsDeleted 4226 -> 4207 -0.45%
  loop-simplifycfg.NumTerminatorsFolded 10829 -> 10809 -0.18%
  simple-loop-unswitch.NumCostMultiplierSkipped 17255 -> 17235 -0.12%
  gvn.NumGVNSimpl 4425749 -> 4423009 -0.06%
  memdep.NumCacheCompleteNonLocalPtr 5339776 -> 5337502 -0.04%
  simple-loop-unswitch.NumBranches 104887 -> 104848 -0.04%
  lcssa.NumLCSSA 15147718 -> 15143663 -0.03%

+14 llvm/CoverageMappingReader.ll
+10 arrow/diff.ll
+10 arrow/scalar_cast_string.ll
+6 ruff-rs/eflamytq0vbr5latbim0v068w.ll
+4 ruff-rs/9jp9cbp5emd24n0w99xw04urm.ll
+1 abc/ioReadPlaMo.ll
+1 coreutils-rs/3xc4e7mmntf4esc6.ll
+1 ockam-rs/2ngtaq92gcad4v6j.ll
+1 yosys/sim.ll
+0 abseil-cpp/inlined_vector_test.ll
+0 actix-rs/22x16e3cd4musvfe.ll
+0 arrow/file.ll
+0 cpython/lexer.ll
+0 delta-rs/47qjbhol909h8zu7.ll
+0 delta-rs/4say4x9grcidoih4.ll
+0 duckdb/format.ll
+0 entt/adjacency_matrix.ll
+0 ffmpeg/ffmetadec.ll
+0 ffmpeg/mpegvideo_enc.ll
+0 fish-rs/5o5xgybh9xo8acbeboom17uhq.ll
+0 fmt/format-impl-test.ll
+0 g2o/solver_pcg.ll
+0 git/config.ll
+0 hdf5/H5Shyper.ll
+0 libjpeg-turbo/rdswitch.ll
+0 libquic/inflate.ll
+0 libquic/vlog.ll
+0 lightgbm/linear_tree_learner.ll
+0 meshlab/cube_style_precomputation.ll
+0 mold/crc32.ll
+0 ockam-rs/u4cgd14nodyt386.ll
+0 pbrt-v4/shapes.ll
+0 php/iconv.ll
+0 pingora-rs/1u078lli1cyrvy3iacncgr0sz.ll
+0 pingora-rs/24oggy68gf6cx0i1vzuuyyts4.ll
+0 postgres/hashsearch.ll
+0 quiche-rs/dcln9cwp955y6zcrpmqhoqx85.ll
+0 quickjs/quickjs.ll
+0 recastnavigation/catch_amalgamated.ll
+0 regex-rs/11vfjke4utuj478u.ll
+0 sdl/SDL_blit_slow.ll
+0 spdlog/bundled_fmtlib_format.ll
+0 tev/Channel.ll
+0 tree-sitter-rs/2fq1z2kxn3lpndsy.ll
+0 wasmtime-rs/48myxw210mngc99c.ll
+0 wolfssl/internal.ll
-1 cpython/_testcapimodule.ll
-1 curl/var.ll
-1 hyperscan/mcclellancompile.ll
-1 hyperscan/ng_haig.ll
-1 llvm/TGParser.ll
-1 quantlib/blackkarasinski.ll
-1 rocksdb/db_impl.ll
-1 xgboost/comm.ll
-2 ceres/line_search_minimizer.ll
-2 hyperscan/limex_compile.ll
-2 libigl/orient_outward.ll
-2 open3d/NonRigidOptimizer.ll
-2 open3d/TriangleMeshDeformation.ll
-2 ozz-animation/sampling_job.ll
-2 postgres/spgscan.ll
-2 quantlib/mcpathbasketengine.ll
-2 quantlib/sparseilupreconditioner.ll
-3 openjdk/c1_MacroAssembler_x86.ll
-3 quantlib/svd.ll
-6 hwloc/topology.ll
-6 linux/syscall_user_dispatch.ll
-10 hermes/g_fmt.ll

[github-actions]

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

1. Phi Node Simplification & Constant Folding: Several phi instructions were updated to use constant values (e.g., 0, 1, 9, 10, 30, 134217728) instead of variable operands, indicating improved constant propagation and simplification of control-flow merge points—especially in loops and exception-handling paths.

2. Store-to-Memset Optimization: Multiple sequences of scalar stores (e.g., store i64 %x, ptr %p, store i64 %x, ptr %q) were replaced with llvm.memset.p0.i64 calls initializing memory regions (e.g., 16–25 bytes) to zero. This reflects aggressive dead-store elimination and memset lowering, notably in Eigen matrix initialization, Open3D, HWLOC, and libigl.

3. TBAA Metadata Updates: Type-based alias analysis (TBAA) node definitions were reordered or corrected (e.g., swapping field offsets in !25/!28, updating !273/!275/!276), improving precision for memory access disambiguation across Eigen and Open3D code.

4. Loop & Branch Target Fixups: Switch and branch targets were updated to reference correct predecessor blocks (e.g., .lr.ph387.preheader instead of .lr.ph388.preheader, %._crit_edge421 instead of %._crit_edge420), ensuring accurate CFG reconstruction after loop optimizations or block merging.

5. Redundant Load/Store Elimination & Pointer Arithmetic Fixes: Stores of computed pointers (e.g., %84 = getelementptr ..., store i64 %23, ptr %84) were replaced with direct uses of getelementptr results; loads from incorrect offsets (e.g., %207 → %206) were fixed; and pointer arithmetic (e.g., shl vs add for scaling) was refined in ozz-animation and regex-rs, improving correctness and optimization potential.

These changes collectively reflect improvements in constant propagation, memory layout optimization, TBAA modeling, CFG maintenance, and low-level addressing logic—consistent with mature backend and middle-end optimizations.

model: qwen-plus-latest
CompletionUsage(completion_tokens=504, prompt_tokens=113156, total_tokens=113660, completion_tokens_details=None, prompt_tokens_details=None)