[llvm-testresults] lordcrumb-clang-x86 i386 nightly tester results

Thu Jun 4 14:47:00 PDT 2009

These test results are somewhat interesting because clang didn't used
to honor -mllvm -disable-llvm-optzns, which means that clang was
effective running -O3 twice for all inputs (once internally, and once
in the standard opt pass).

It turns out this was giving a huge speedup on objinst, I filed PR4327
in case someone wants to investigate. A number of other applications
were apparently being sped up quite a bit, for example Olden/bh/bh and
Dhrystone/fldry.

Even more interesting is the impact on .s compile time. It appears
that the assembler time for many many benchmarks was being sped up by
20-100%, seems like this could be worth investigating just from a
compile time / code size perspective.

Has anyone every done a study of, say, running -O3 level opts twice on
something we care about (SSL, SPEC, whatever)?

 - Daniel

On Thu, Jun 4, 2009 at 4:02 AM, Apache <apache at cs.uiuc.edu> wrote:
> http://llvm.org/nightlytest/test.php?machine=448&night=11041
> Name: lordcrumb.apple.com
> Nickname: lordcrumb-clang-x86
> Buildstatus: OK
>
> New Test Passes:
> Applications/JM/ldecod/ldecod [CBE]
> Applications/JM/lencod/lencod [CBE]
> Applications/SIBsim4/SIBsim4 [CBE]
> Benchmarks/ASCI_Purple/SMG2000/smg2000 [CBE]
> Benchmarks/MiBench/security-rijndael/security-rijndael [CBE]
> Benchmarks/Prolangs-C/agrep/agrep [JIT codegen, JIT]
> Benchmarks/Prolangs-C/football/football [JIT codegen, JIT]
> Benchmarks/Prolangs-C/TimberWolfMC/timberwolfmc [CBE]
>
>
> New Test Failures:
> Benchmarks/Prolangs-C/agrep/agrep [CBE]
> Benchmarks/Prolangs-C/football/football [CBE]
>
>
> Added Tests:
> test/CodeGen/Generic/2009-06-03-UnreachableSplitPad.ll for PR4317
> test/CodeGen/X86/2009-06-03-Win64DisableRedZone.ll
> test/CodeGen/X86/2009-06-03-Win64SpillXMM.ll
> test/Transforms/IndVarSimplify/variable-stride-ivs-0.ll
> test/Transforms/IndVarSimplify/variable-stride-ivs-1.ll for PR4315
>
>
> Removed Tests:
> test/Transforms/IndVarSimplify/variable-stride-ivs.ll
> UnitTests/2007-04-25-weak
> UnitTests/Threads/tls
>
>
> Significant changes in test results:
> Bytecode:
>  singlesource/Benchmarks/BenchmarkGame/recursive: 8.84% (5248 => 4784)
>  singlesource/Benchmarks/Dhrystone/dry: -10.75% (1488 => 1648)
>  singlesource/Benchmarks/Dhrystone/fldry: -12.12% (1584 => 1776)
>  singlesource/Benchmarks/Misc/ReedSolomon: 8.10% (14816 => 13616)
>  singlesource/Benchmarks/Misc/salsa20: -6.42% (4240 => 4512)
>  singlesource/Benchmarks/Shootout/lists: 5.88% (4624 => 4352)
>  singlesource/Benchmarks/Shootout/nestedloop: -38.46% (832 => 1152)
>  singlesource/Benchmarks/Stanford/Perm: 63.92% (11664 => 4208)
>  singlesource/Benchmarks/Stanford/Puzzle: 34.46% (8080 => 5296)
>  singlesource/Regression/C/bigstack: 20.68% (6112 => 4848)
>  singlesource/UnitTests/2002-10-09-ArrayResolution: -11.11% (576 => 640)
>  singlesource/UnitTests/Vector/SSE/sse.isamax: 8.08% (4160 => 3824)
>  singlesource/UnitTests/Vector/build2: -5.56% (1440 => 1520)
>  multisource/Applications/SIBsim4/SIBsim4: 6.59% (134240 => 125392)
>  multisource/Applications/lua/lua: 9.22% (995008 => 903280)
>  multisource/Benchmarks/FreeBench/fourinarow/fourinarow: 5.07% (27760 => 26352)
>  multisource/Benchmarks/MiBench/automotive-bitcount/automotive-bitcount: 11.49% (5152 => 4560)
>  multisource/Benchmarks/NPB-serial/is/is: 21.96% (8016 => 6256)
>  multisource/Benchmarks/Olden/bh/bh: 13.45% (27712 => 23984)
>  multisource/Benchmarks/Olden/bisort/bisort: 22.36% (6656 => 5168)
>  multisource/Benchmarks/Olden/perimeter/perimeter: 55.62% (48816 => 21664)
>  multisource/Benchmarks/Olden/power/power: 15.35% (14800 => 12528)
>  multisource/Benchmarks/Prolangs-C/TimberWolfMC/timberwolfmc: -9.45% (813328 => 890224)
>  multisource/Benchmarks/SciMark2-C/scimark2: 11.49% (26320 => 23296)
>  multisource/Benchmarks/VersaBench/bmm/bmm: 8.29% (3472 => 3184)
>  multisource/Benchmarks/mediabench/g721/g721encode/encode: 7.26% (21808 => 20224)
> LLC-BETA:
>  singlesource/Benchmarks/Dhrystone/dry: -8.43% (2.61 => 2.83)
>  singlesource/Benchmarks/Dhrystone/fldry: -12.69% (2.60 => 2.93)
>  singlesource/Benchmarks/Shootout/objinst: -92.65% (4.90 => 9.44)
>  multisource/Benchmarks/Olden/bh/bh: -21.94% (1.96 => 2.39)
> CBE:
>  singlesource/Benchmarks/Dhrystone/fldry: -56.30% (1.35 => 2.11)
>  singlesource/Benchmarks/Misc/oourafft: -6.64% (9.04 => 9.64)
>  singlesource/Benchmarks/Misc/salsa20: -8.44% (11.49 => 12.46)
>  singlesource/Benchmarks/Shootout-C++/ary3: -7.41% (3.51 => 3.77)
>  singlesource/Benchmarks/Shootout/lists: -13.86% (5.99 => 6.82)
>  singlesource/Benchmarks/Shootout/objinst: -99.58% (4.79 => 9.56)
>  multisource/Applications/viterbi/viterbi: -5.83% (15.95 => 16.88)
>  multisource/Benchmarks/ASC_Sequoia/AMGmk/AMGmk: 6.02% (17.29 => 16.25)
> JIT:
>  singlesource/Benchmarks/Dhrystone/fldry: -20.51% (2.34 => 2.82)
>  singlesource/Benchmarks/Shootout/matrix: -6.52% (3.68 => 3.92)
>  singlesource/Benchmarks/Shootout/objinst: -101.96% (4.59 => 9.27)
>  multisource/Applications/JM/lencod/lencod: -13.54% (14.18 => 16.10)
>  multisource/Benchmarks/Olden/bh/bh: -18.50% (2.00 => 2.37)
>  multisource/Benchmarks/Olden/power/power: 38.63% (2.33 => 1.43)
>  multisource/Benchmarks/VersaBench/dbms/dbms: 35.44% (2.37 => 1.53)
>  multisource/Benchmarks/mafft/pairlocalalign: -18.36% (35.63 => 42.17)
> LLC:
>  singlesource/Benchmarks/Dhrystone/fldry: -23.14% (2.29 => 2.82)
>  singlesource/Benchmarks/Shootout/objinst: -93.32% (4.94 => 9.55)
>  multisource/Benchmarks/Olden/bh/bh: -23.62% (1.99 => 2.46)
> GCCAS:
>  multisource/Applications/ClamAV/clamscan: -50.17% (3.4895 => 5.2401)
>  multisource/Applications/JM/ldecod/ldecod: -23.42% (2.4369 => 3.0076)
>  multisource/Applications/JM/lencod/lencod: -40.99% (4.5061 => 6.3532)
>  multisource/Applications/SPASS/SPASS: -68.46% (4.1106 => 6.9246)
>  multisource/Applications/d/make_dparser: -93.22% (0.6667 => 1.2882)
>  multisource/Applications/lua/lua: -27.39% (1.3817 => 1.7602)
>  multisource/Applications/oggenc/oggenc: -41.33% (1.2297 => 1.7379)
>  multisource/Applications/sqlite3/sqlite3: -54.33% (2.5469 => 3.9306)
>  multisource/Applications/treecc/treecc: -36.47% (0.6310 => 0.8611)
>  multisource/Benchmarks/ASCI_Purple/SMG2000/smg2000: -45.01% (2.3504 => 3.4082)
>  multisource/Benchmarks/MiBench/consumer-jpeg/consumer-jpeg: -55.80% (0.9062 => 1.4119)
>  multisource/Benchmarks/MiBench/consumer-lame/consumer-lame: -45.98% (1.0155 => 1.4824)
>  multisource/Benchmarks/MiBench/consumer-typeset/consumer-typeset: -74.11% (3.2834 => 5.7168)
>  multisource/Benchmarks/MiBench/telecomm-gsm/telecomm-gsm: -116.47% (0.2192 => 0.4745)
>  multisource/Benchmarks/Prolangs-C/TimberWolfMC/timberwolfmc: -32.46% (1.4213 => 1.8827)
>  multisource/Benchmarks/mafft/pairlocalalign: -54.42% (2.9751 => 4.5942)
>  multisource/Benchmarks/mediabench/gsm/toast/toast: -115.38% (0.2197 => 0.4732)
>  multisource/Benchmarks/mediabench/jpeg/jpeg-6a/cjpeg: -53.42% (0.9272 => 1.4225)
> LLC compile:
>  multisource/Applications/lua/lua: 10.84% (2.6742 => 2.3844)
> LLC-BETA compile:
>  multisource/Applications/lua/lua: 10.93% (2.9112 => 2.5929)
>
> _______________________________________________
> llvm-testresults mailing list
> llvm-testresults at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-testresults
>