[llvm-testresults] Grawp-PIC i386 nightly tester results

[Chris Lattner]

On Jan 12, 2008, at 7:45 AM, Apache wrote:
> http://llvm.org/nightlytest/test.php?machine=205&night=4546
> Name: grawp.apple.com
> Nickname: Grawp-PIC
> Buildstatus: OK

This was the first run with Evan's x86-32 byval stuff:

> New Test Failures:
> Benchmarks/MiBench/consumer-lame/consumer-lame [LLC compile, LLC- 
> BETA compile, , CBE, LLC]

This is fixed now.

> LLC:
> singlesource/Benchmarks/Misc-C++/stepanov_v1p2: -306.24% (7.37 =>  
> 29.94)
> singlesource/Benchmarks/Misc/mandel: 35.67% (1.71 => 1.10)
> multisource/Benchmarks/Olden/bh/bh: 32.50% (3.60 => 2.43)

Very nice on mandel/bh.  I fixed the stepanov_v1p2 regression, maybe  
it will help other apps too.

> GCCAS:
> multisource/Applications/JM/lencod/lencod: 30.38% (7.8074 => 5.4353)
> multisource/Benchmarks/MiBench/consumer-lame/consumer-lame: 33.88%  
> (1.0792 => 0.7136)
> LLC compile:
> multisource/Applications/JM/lencod/lencod: 11.59% (7.3839 => 6.5280)

This is pretty nice and not something I expected.  Yay for faster  
compiles.  Many bc files also shrunk, which makes sense.

> CBE:
> singlesource/Benchmarks/Misc-C++/stepanov_v1p2: -39.90% (14.46 =>  
> 20.23)

fixed.

> multisource/Applications/SIBsim4/SIBsim4: 7.35% (5.44 => 5.04)
> multisource/Applications/aha/aha: 8.11% (2.96 => 2.72)
> multisource/Applications/minisat/minisat: 7.27% (8.94 => 8.29)
> multisource/Benchmarks/Olden/bh/bh: 14.08% (3.41 => 2.93)
> multisource/Benchmarks/Olden/treeadd/treeadd: 5.86% (5.97 => 5.62)
> external/SPEC/CINT2006/462.libquantum/462.libquantum: 7.13% (5.05 =>  
> 4.69)

Ok, good deal.

> external/SPEC/CINT2000/175.vpr/175.vpr: -5.08% (5.32 => 5.59)
> external/SPEC/CINT2000/252.eon/252.eon: -186.27% (0.51 => 1.46)

Evan, please investigate these.  While performance of the CBE is not  
critical, it is important to understand what is going on here.  Eon  
also regressed significantly on your non-pic tester, so this is not  
noise.

Thanks,

-Chris