Diff mode

runner: ariselab-64c-docker
baseline: llvm/llvm-project@613804c
patch: llvm/llvm-project#123518
sha256: 65df05721a33e13e53a6b690681ec59c8d67a7d3b24c08d67c1fc9aa640892ac
commit: 0833417

3056 files changed, 676708 insertions(+), 675365 deletions(-)

Improvements:
  indvars.NumSimplifiedSRem 16 -> 40 +150.00%
  licm.NumAddSubHoisted 1316 -> 1881 +42.93%
  indvars.NumSimplifiedSDiv 79 -> 108 +36.71%
  correlated-value-propagation.NumSDivSRemsNarrowed 2973 -> 3041 +2.29%
  indvars.NumElimCmp 57996 -> 59284 +2.22%
  correlated-value-propagation.NumSMinMax 9236 -> 9342 +1.15%
  correlated-value-propagation.NumSExt 51128 -> 51706 +1.13%
  instcombine.NumReassoc 286872 -> 289584 +0.95%
  licm.NumBOAssociationsHoisted 7181 -> 7234 +0.74%
  simple-loop-unswitch.NumSwitches 2032 -> 2045 +0.64%
Regressions:
  alignment-from-assumptions.NumStoreAlignChanged 50 -> 1 -98.00%
  loop-simplifycfg.NumLoopExitsDeleted 3987 -> 3959 -0.70%
  sccp.NumInstReplaced 175387 -> 174227 -0.66%
  loop-instsimplify.NumSimplified 196913 -> 196398 -0.26%
  loop-simplifycfg.NumTerminatorsFolded 10668 -> 10645 -0.22%
  aggressive-instcombine.NumInstrsReduced 73601 -> 73461 -0.19%
  aggressive-instcombine.NumExprsReduced 22613 -> 22575 -0.17%
  simple-loop-unswitch.NumCostMultiplierSkipped 17337 -> 17312 -0.14%
  licm.NumIntAssociationsHoisted 4100 -> 4096 -0.10%
  correlated-value-propagation.NumMulNUW 63088 -> 63033 -0.09%

5 6 bench/abc/optimized/abcDetect.ll
13 14 bench/abc/optimized/fraigVec.ll
17 13 bench/abc/optimized/giaIso.ll
17 21 bench/abc/optimized/sclLoad.ll
11 9 bench/abc/optimized/utilBSet.ll
4 3 bench/abc/optimized/verCore.ll
1 1 bench/abseil-cpp/optimized/cord_test.ll
2 2 bench/actix-rs/optimized/2nm6r5hc2u1hk1e1.ll
39 39 bench/annoy/optimized/annoymodule.ll
21 22 bench/arrow/optimized/chunked_array.ll
20 21 bench/bdwgc/optimized/gc.ll
29 15 bench/box2d/optimized/solver.ll
27 27 bench/c3c/optimized/context.ll
25 23 bench/cmake/optimized/trees.ll
27 26 bench/cpython/optimized/_randommodule.ll
35 33 bench/darktable/optimized/AbstractLJpegDecoder.ll
16 14 bench/darktable/optimized/IiqDecoder.ll
6 4 bench/duckdb/optimized/decode.ll
27 24 bench/eastl/optimized/EAString.ll
16 12 bench/faiss/optimized/Heap.ll
14 13 bench/ffmpeg/optimized/af_adeclick.ll
22 18 bench/ffmpeg/optimized/celp_filters.ll
10 8 bench/ffmpeg/optimized/lossless_videodsp.ll
8 7 bench/ffmpeg/optimized/mobiclip.ll
4 3 bench/ffmpeg/optimized/motionpixels.ll
17 15 bench/ffmpeg/optimized/snow_dwt.ll
10 10 bench/folly/optimized/Benchmark.ll
22 22 bench/g2o/optimized/edge_se2_lotsofxy.ll
16 16 bench/g2o/optimized/vertex_ellipse.ll
7 7 bench/glslang/optimized/doc.ll
3 3 bench/glslang/optimized/hlslParseHelper.ll
13 11 bench/graphviz/optimized/lu.ll
10 11 bench/graphviz/optimized/maze.ll
26 23 bench/gromacs/optimized/hxprops.ll
103 93 bench/grpc/optimized/histogram_view.ll
9 8 bench/hdf5/optimized/H5Dbtree2.ll
4 3 bench/hdf5/optimized/H5Dearray.ll
29 28 bench/hdf5/optimized/H5Pencdec.ll
11 10 bench/icu/optimized/collationiterator.ll
9 5 bench/icu/optimized/rbtz.ll
6 7 bench/icu/optimized/reslist.ll
5 5 bench/image-rs/optimized/2mngkegtim1o10y3.ll
12 10 bench/ipopt/optimized/IpDenseGenMatrix.ll
30 25 bench/ipopt/optimized/IpDenseSymMatrix.ll
26 24 bench/ipopt/optimized/IpLimMemQuasiNewtonUpdater.ll
17 17 bench/libdeflate/optimized/deflate_compress.ll
23 21 bench/libquic/optimized/quic_packet_creator.ll
15 14 bench/libwebp/optimized/cost_enc.ll
41 38 bench/llama.cpp/optimized/llama-kv-cache.ll
2 1 bench/llvm/optimized/DataFlowSanitizer.ll
12 10 bench/llvm/optimized/OptionParserEmitter.ll
9 9 bench/lodepng/optimized/lodepng.ll
24 13 bench/meshlab/optimized/shell.ll
11 10 bench/miniaudio/optimized/unity.ll
24 21 bench/mitsuba3/optimized/rgb2spec_opt.ll
64 59 bench/nuttx/optimized/fs_files.ll
20 19 bench/oiio/optimized/imagebufalgo.ll
15 15 bench/open3d/optimized/ColoredICP.ll
29 28 bench/openblas/optimized/lapacke_ctp_trans.ll
10 9 bench/openblas/optimized/lapacke_dtp_trans.ll
19 17 bench/openblas/optimized/sscal_k.ll
3 2 bench/opencv/optimized/descriptor_database.ll
48 44 bench/opencv/optimized/lrn_layer.ll
6 5 bench/opencv/optimized/matrix_sparse.ll
14 11 bench/openjdk/optimized/c1_ValueMap.ll
11 8 bench/openjdk/optimized/jfrJdkJfrEvent.ll
12 9 bench/openjdk/optimized/shenandoahSupport.ll
6 5 bench/openspiel/optimized/random_uci_bot.ll
44 40 bench/openssl/optimized/extensions_srvr.ll
4 4 bench/openvdb/optimized/PoissonSolver.ll
41 38 bench/pbrt-v4/optimized/rgb2spec_opt.ll
3 5 bench/postgres/optimized/gistscan.ll
85 76 bench/postgres/optimized/numeric.ll
22 24 bench/postgres/optimized/regcomp.ll
12 14 bench/quantlib/optimized/knuthuniformrng.ll
5 4 bench/quiche-rs/optimized/4epksim1e613ria3d5gxmneiw.ll
32 34 bench/raylib/optimized/rtextures.ll
20 23 bench/rust-analyzer-rs/optimized/1n6t6n716s6vtscg.ll
16 19 bench/rust-analyzer-rs/optimized/2jezmucvqvfiuhvx.ll
19 22 bench/rust-analyzer-rs/optimized/4tstt0w6z5grnfcy.ll
28 31 bench/rust-analyzer-rs/optimized/84mgk0fwp2eksq1.ll
8 8 bench/rustfmt-rs/optimized/3n0n7c4xrusd0kzl.ll
33 32 bench/slurm/optimized/env.ll
8 7 bench/slurm/optimized/hostlist.ll
20 19 bench/sqlite/optimized/shell.ll
29 31 bench/stb/optimized/stb_image_write.ll
11 11 bench/stockfish/optimized/uci.ll
7 6 bench/sundials/optimized/sunnonlinsol_fixedpoint.ll
24 24 bench/syn/optimized/2khi0xu1ufmhwoo.ll
32 34 bench/tinygltf/optimized/tiny_gltf.ll
6 6 bench/tinympc/optimized/cartpole_example.ll
30 29 bench/tomlplusplus/optimized/toml.ll
8 6 bench/velox/optimized/MapZipWith.ll
29 18 bench/velox/optimized/PrestoSerializer.ll
21 14 bench/velox/optimized/SparseHll.ll
7 5 bench/velox/optimized/Variant.ll
5 4 bench/yalantinglibs/optimized/conformance.struct_pb.ll
7 8 bench/yosys/optimized/genrtlil.ll
12 16 bench/zfp/optimized/encode4d.ll

Here is a brief summary of up to 5 major changes in the patch:

1. Loop Phi Node Adjustments: In abcDetect.ll, the %245 phi node has been modified so that instead of taking %372 from predecessor %367, it now takes value %.reass. This change likely represents a transformation where a truncation and comparison sequence was replaced by a simpler addition operation, shifting some logic into earlier parts of the loop.

2. Truncation and Comparison Simplification: In abcDetect.ll, the instruction %368 = or disjoint i64 %indvars.iv.next359, 1 was replaced with a truncation followed by an add (%368 = trunc nsw i64 %indvars.iv358 to i32 and %.reass = add i32 %368, 3). The subsequent comparison was also changed from icmp slt i64 to icmp slt i32, reflecting a narrowing of types and potentially reducing unnecessary widening operations.

3. Integer Extension Changes: Several files (e.g., fraigVec.ll, giaIso.ll, celp_filters.ll) show a shift from zero extension (zext) to sign extension (sext) or vice versa. For example, in fraigVec.ll, %17 is now computed using sext instead of zext. This indicates better handling of signedness assumptions in loops or data flow.

4. Load Alignment Improvements: Across several files (e.g., cord_test.ll, annoymodule.ll, IiqDecoder.ll), there are changes increasing alignment on load instructions — such as from align 1 to align 4 or align 8. These changes suggest performance improvements due to more efficient memory access patterns.

5. Loop Control Condition Revisions: In multiple places (e.g., sclLoad.ll, libdeflate/deflate_compress.ll), the condition controlling loop exits has been updated from using icmp samesign ult or icmp ugt to icmp slt or other simpler conditions. This simplifies loop control logic and may improve branch prediction and readability without altering semantics.

These changes reflect optimizations in integer extensions, alignment, loop control structure, and phi usage — all aimed at improving performance and correctness.

model: qwen-plus-latest
CompletionUsage(completion_tokens=506, prompt_tokens=110254, total_tokens=110760, completion_tokens_details=None, prompt_tokens_details=None)