[llvm-testresults] evchen-ppc32 nightly tester results

[Dale Johannesen]

On Jul 2, 2009, at 6:15 PM, Daniel Dunbar wrote:

> Are the 7 LLC failures here expected? This is the list:
> --
> MultiSource/Applications/ClamAV/clamscan
> MultiSource/Applications/oggenc/oggenc
> MultiSource/Benchmarks/VersaBench/dbms/dbms
> SingleSource/UnitTests/AtomicOps
> SingleSource/UnitTests/Vector/Altivec/alti_expandfft
> SingleSource/UnitTests/Vector/Altivec/alti_sdot
> SingleSource/UnitTests/Vector/Altivec/alti_stepfft

If you look closer at AtomicOps, you'll see it's actually gcc that's  
failing; LLC and JIT work, but miscompare because gcc failed.
All the others except oggenc failed for me on the last run.  I suspect  
dbms is inherently little-endian and is supposed to fail, but haven't  
proved it.
I suspect the Altivec failures are due to non-standard-conformant  use  
of multiply-add instructions, which we do because our numerics guys  
think it is the right thing, but haven't proved that either.  oggenc  
succeeds and fails somewhat arbitrarily, and I suspect nondeterminancy  
in the source, but, you guessed it (it's quite difficult to get  
anywhere analyzing it, as the output is a gigantic binary file).   
Clamscan is probably a bug.  All of these (except oggenc) have been  
failing for quite a while, at least a month, if you look back.

It annoys me to leave things unsettled like this, but I really can't  
justify taking more time on PPC at the moment...let's see what the PPC  
SVR4 guys can do...

> - Daniel
>
> On Thu, Jul 2, 2009 at 5:58 PM, <daniel_dunbar at apple.com> wrote:
>> http://smooshlab.apple.com/perf.cgi/db_nt_internal/nightlytest//456
>> Name:  evchen.apple.com
>> Nickname: evchen-ppc32:1
>> Buildstatus: OK
>>
>> Run: 456, Start Time: 2009-07-02 12:39:39
>> Comparing To: 443, Start Time: 2009-07-02 01:05:11
>>
>> --- Changes Summary ---
>> New Test Passes: 0
>> New Test Failures: 0
>> Added Tests: 0
>> Removed Tests: 0
>> Significant Changes: 11
>>
>> --- Tests Summary ---
>> Total Tests: 3573
>> Total Test Failures: 97
>>
>> Total Test Failures By Type:
>>  CBE: 35
>>  JIT: 17
>>  JIT codegen: 17
>>  LLC: 7
>>  LLC compile: 7
>>  LLC-BETA: 7
>>  LLC_BETA compile: 7
>>
>> --- Changes Detail ---
>> New Test Passes:
>>
>> New Test Failures:
>>
>> Added Tests:
>>
>> Removed Tests:
>>
>> Significant Changes in Test Results:
>> CBE:
>>  MultiSource/Applications/kimwitu++/kc: 5.56% (0.3600 => 0.3800)
>>  SingleSource/Benchmarks/Misc/flops: -10.55% (17.5400 => 15.6900)
>>  SingleSource/Benchmarks/Misc/himenobmtxpa: -51.18% (65.9200 =>  
>> 32.1800)
>> LLC-BETA:
>>  SingleSource/Benchmarks/Misc/himenobmtxpa: 507.48% (11.2300 =>  
>> 68.2200)
>> LLC compile:
>>  Externals/SPEC/CINT2000/253_perlbmk/253_perlbmk: 5.56% (21.8264 =>  
>> 23.0395)
>> JIT:
>>  SingleSource/Benchmarks/Adobe-C++/loop_unroll: -14.53% (7.0900 =>  
>> 6.0600)
>>  SingleSource/Benchmarks/CoyoteBench/lpbench: -7.49% (15.7600 =>  
>> 14.5800)
>> LLC:
>>  MultiSource/Applications/kimwitu++/kc: 5.56% (0.3600 => 0.3800)
>>  MultiSource/Benchmarks/Prolangs-C++/primes/primes: 5.00% (1.4000  
>> => 1.4700)
>>  SingleSource/Benchmarks/Misc/himenobmtxpa: 19.07% (56.9900 =>  
>> 67.8600)
>> LLC-BETA compile:
>>  Externals/SPEC/CINT2000/253_perlbmk/253_perlbmk: -5.07% (11.0455  
>> => 10.4853)
>> _______________________________________________
>> llvm-testresults mailing list
>> llvm-testresults at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-testresults
>>
>
> _______________________________________________
> llvm-testresults mailing list
> llvm-testresults at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-testresults