Diff mode

runner: ariselab-64c-docker
baseline: llvm/llvm-project@6a71938
patch: llvm/llvm-project#151132
sha256: af143e34730d8ed3c55701f7657d5d591b1ba29ab185425f82d84055994fda24
commit: 0f19182

2833 files changed, 1550598 insertions(+), 1550762 deletions(-)

Improvements:
  bdce.NumRemoved 387999 -> 406025 +4.65%
  sccp.NumInstRemoved 2016946 -> 2020614 +0.18%
  instcombine.NumDeadInst 42463020 -> 42473284 +0.02%
  instcombine.NegatorNumValuesVisited 22308691 -> 22310260 +0.01%
  instcombine.NegatorTotalNegationsAttempted 21353443 -> 21354924 +0.01%
  instcombine.NumOneIteration 63698974 -> 63701393 +0.00%
  instcombine.NumWorklistIterations 63698974 -> 63701393 +0.00%
  sccp.NumInstReplaced 133926 -> 133931 +0.00%
  gvn.NumGVNInstr 155528 -> 155532 +0.00%
  gvn.NumGVNPRE 155528 -> 155532 +0.00%
Regressions:
  correlated-value-propagation.NumSIToFP 1515 -> 1514 -0.07%
  correlated-value-propagation.NumMinMax 12484 -> 12482 -0.02%
  correlated-value-propagation.NumCmps 292266 -> 292227 -0.01%
  instcombine.NumConstProp 163716 -> 163698 -0.01%
  correlated-value-propagation.NumSExt 48149 -> 48145 -0.01%
  correlated-value-propagation.NumNUW 594770 -> 594721 -0.01%
  last-run-tracking.NumSkippedPasses 50279738 -> 50277319 -0.00%
  gvn.NumGVNEqProp 443254 -> 443234 -0.00%
  correlated-value-propagation.NumAddNUW 228705 -> 228695 -0.00%
  jump-threading.NumFolds 2648931 -> 2648818 -0.00%

+0 actix-rs/1kr0tdq4l6v38mbk.ll
+0 c3c/c_abi_x64.ll
+0 ceres/dynamic_sparse_normal_cholesky_solver.ll
+0 ceres/iterative_schur_complement_solver.ll
+0 coreutils-rs/icw7irkz6mu05a9.ll
+0 csmith/Constant.ll
+0 csmith/Function.ll
+0 cvc5/SimpSolver.ll
+0 cvc5/theory_sep_rewriter.ll
+0 delta-rs/4zvphat0q9a964bz.ll
+0 double_conversion/string-to-double.ll
+0 eastl/TestRingBuffer.ll
+0 flac/ogg_decoder_aspect.ll
+0 freetype/truetype.ll
+0 git/diffcore-rename.ll
+0 grpc/timeout_encoding.ll
+0 hwloc/base64.ll
+0 jsonnet/lexer.ll
+0 just-rs/1uppz7l1e89fp8n7.ll
+0 libevent/evdns.ll
+0 libjpeg-turbo/transupp.ll
+0 libpng/pngset.ll
+0 libquic/s3_pkt.ll
+0 libsodium/aegis256_soft.ll
+0 libsodium/crypto_secretbox_easy.ll
+0 libwebp/dwebp.ll
+0 lvgl/lv_sprintf_builtin.ll
+0 mold/cmdline.cc.X86_64.ll
+0 mold/passes.cc.X86_64.ll
+0 nghttp2/nghttp2_submit.ll
+0 nuttx/lib_base64.ll
+0 openblas/dggesx.ll
+0 opencc/Config.ll
+0 openmpi/pmix_pointer_array.ll
+0 pbrt-v4/parser.ll
+0 raylib/rmodels.ll
+0 re2/compile.ll
+0 ripgrep-rs/79zj4cmg2a4glb1.ll
+0 sentencepiece/builder.ll
+0 stb/stb_tilemap_editor.ll
+0 stb/stb_vorbis.ll
+0 verilator/V3Options.ll
+0 wasmi-rs/1z05d6joihj04oihu2i4irshf.ll
+0 zlib/deflate.ll
+0 zstd/zstd_opt.ll
-1 arrow/string.ll
-1 llvm/SemaDeclCXX.ll
-1 php/ir_check.ll
-2 typst-rs/3rk2ctuzbghb17s4.ll
-3 openjdk/lowMemoryDetector.ll
-14 rust-analyzer-rs/5fgtm3ees9sdsrsg.ll
-30 rust-analyzer-rs/44ayynfm08lnhjsw.ll

1. Improved memory comparison calls: Multiple bcmp calls now include noundef and dereferenceable attributes with known lengths, replacing variable length arguments (e.g., %2) with compile-time constants (e.g., 6, 7, 16). This enables better optimization by allowing LLVM to reason about memory access patterns.

2. Optimized string comparisons: In several benchmarks (arrow, mold, php), direct byte comparisons replaced bcmp calls when comparing against single characters. For example, loading a byte and comparing it directly to a constant (like 48 for '0') is more efficient than calling bcmp.

3. Enhanced loop induction variables: Several loops had their induction variables updated to reference new preheader blocks (e.g., from %.lr.ph212 to %.lr.ph213), indicating improved loop optimization and potentially better vectorization opportunities.

4. Improved exception handling labels: Many invoke instructions now target updated .noexc labels (e.g., from .noexc137 to .noexc138), reflecting changes in exception handling paths that may improve code layout and optimization.

5. Better phi node predecessor updates: Numerous phi nodes were updated to reference new block labels (e.g., from %._crit_edge to %._crit_edge190), indicating improved control flow graph transformations that can enable further optimizations like dead code elimination.

model: qwen-plus-latest
CompletionUsage(completion_tokens=301, prompt_tokens=84597, total_tokens=84898, completion_tokens_details=None, prompt_tokens_details=None)

cc @nikic