search

dtcxzyw / llvm-opt-benchmark

An LLVM IR dataset for data-driven compiler optimization research
https://dtcxzyw.github.io/llvm-opt-benchmark/
MIT License
19 stars 5 forks source link

Update diff October 4th 2024, 11:51:44 am #1434

Open github-actions[bot] opened 2 weeks ago

github-actions[bot] commented 2 weeks ago

from: https://github.com/llvm/llvm-project/commit/9cf8c094c77db1ed9e63322bedcf28b76e5c5a43 to: https://github.com/llvm/llvm-project/commit/dec4cfdb09596f34f17a653ad405ab2551b09039 commit: 3b463d5a5a15a4f87d060fdfdd6cfc885c049e62

Change Logs

from 9cf8c094c77db1ed9e63322bedcf28b76e5c5a43 to dec4cfdb09596f34f17a653ad405ab2551b09039

dec4cfdb09596f34f17a653ad405ab2551b09039 [LAA] Use loop guards when checking invariant accesses. 53b3c9e408b740addce630e806ccd143b3ca0fde [bazel] Fix libc/hdr past c63112a9118277a20ae440f3f69189c0937e8f4d (#111135) 948467bf7398f5882472ff30bf09e604e237f456 [AMDGPU] Use a range-based for loop. NFC. b837c9e289dab93c7f8a06876e3f70b6864f40ab Revert "[llvm] Mark win x64 SEH pseudo instruction as meta instructions (#110889)" 972353fdfad02ae3206c91c80dd4eaa997b3d499 [LAA] Add tests where results can be improved using loop guards. b5598c33f02732e805e50a3eb5c67b17952c5683 [X86] Regenerate test checks with vpternlog comments 60f3e67dae90f43698ae67167d0e024ff2ab0932 [libc++] Activate -Warray-zero when building (#110812) f6f4c177ef03ecd027d884af2cd3caf3ccafcfd1 [flang][debug] Use PROGRAM name for main function name (#111022) c734d77b9913052012faf91fdc19753f791421d9 [flang][semantics][OpenMP] no privatisation of stmt functions (#106550) ef8d5069e1b8d322cd6905c03bef1b42e2e86515 [AArch64]Fix FEAT_SME_LUTv2 to have FEAT_SME2 implemented. (#110474) 7be2ce7312191894c2458ecc08de7683fd2314fb [RISCV] fix SP recovery in a function epilogue (#110809) 26fca7256edc0013886717b2585e638a6100dcea [VPlan][NFC] Use patterns in test check (#111086) fcdb178bb477a7bd93048fa09a5ca6b1927bdb25 [doc] Add --verify-json to dwarfdump documentation (#110909) 23c02a48f1c7343149b94aafe350ec59d1b14999 [doc] Add llvm --error-display to dwarfdump documentation (#110922) d0756caedcf067860240bf31e8f9d371ba706757 [ARM][AArch64] Introduce the Armv9.6-A architecture version (#110825) d9d623310de196f2119fc593811204993916e26e [mlir][linalg] Add a new helper hook: hasVectorizationImpl (#110708) 992e75403f417d46ce9015147ff49d67009e736c Fix MSVC "not all control paths return a value" warning. NFC. 9e831d50a0d67c6779048742657edb437818b681 [AArch64] Pass -mabi option through to multilib (#110874) 856774dd4f5a80bca8e8d0bfd50ab630c1c2d49d [ConstantFPRange][UnitTest] Ignore NaN payloads when enumerating values in a range (#111083) 67d247a441572c046ccd82682ffe1102584f24ec [InstCombine] Decompose more icmps into masks (#110836)

Compilation time result (by files):
Top 5 improvements:
  abc/epd.c.ll 316231111 -> 291338118 -7.87%
  linux/mincore.ll 124073469 -> 123258421 -0.66%
  spike/fmax_s.ll 1150787762 -> 1144547318 -0.54%
  spike/fmin_s.ll 1149706817 -> 1143777236 -0.52%
  influxdb-rs/2gpesx480dqhic68.ll 6678221516 -> 6659669227 -0.28%
Top 5 regressions:
  lief/sha256.c.ll 514859469 -> 521557553 +1.30%
  openblas/dsytri_3x.c.ll 402080264 -> 405798852 +0.92%
  openblas/dsytri2x.c.ll 410064722 -> 413609993 +0.86%
  darktable/introspection_zonesystem.c.ll 548052743 -> 551414595 +0.61%
  lief/sha512.c.ll 356320710 -> 357540590 +0.34%

Overall: -0.02015955%
Compilation time result (by projects):
Top 5 improvements:
Top 5 regressions:

Overall: -0.02247979%
Improvements:
correlated-value-propagation.NumCmpIntr 15 -> 40 +166.67%
instcombine.NumGlobalCopies 66140 -> 134033 +102.65%
sroa.NumLoadsPredicated 1070 -> 2073 +93.74%
mem2reg.NumDeadAlloca 457518 -> 878044 +91.91%
memcpyopt.NumCallSlot 317680 -> 591983 +86.35%
elim-avail-extern.NumVariables 49155 -> 91400 +85.94%
correlated-value-propagation.NumCmps 128148 -> 236045 +84.20%
elim-avail-extern.NumRemovals 121269 -> 215677 +77.85%
memcpyopt.NumMemCpyInstr 642168 -> 1126944 +75.49%
function-attrs.NumCold 12812 -> 21390 +66.95%
Regressions:
globalopt.NumLocalized 19 -> 13 -31.58%
count-visits.MaxVisited 45146 -> 43182 -4.35%
function-attrs.NumNoUndefReturn 65883 -> 63060 -4.28%
build-libcalls.NumInaccessibleMemOnly 4820 -> 4636 -3.82%
globalsmodref-aa.NumIndirectGlobalVars 90 -> 87 -3.33%
globalopt.NumShrunkToBool 2332 -> 2263 -2.96%
build-libcalls.NumNoUndef 38151 -> 37150 -2.62%
sccp.NumGlobalConst 3303 -> 3223 -2.42%
move-auto-init.NumMoved 2941 -> 2874 -2.28%
build-libcalls.NumInaccessibleMemOrArgMemOnly 9383 -> 9175 -2.22%
dtcxzyw commented 2 weeks ago

See https://github.com/dtcxzyw/llvm-opt-benchmark/pull/1424#discussion_r1784849814