[llvm-testresults] Grawp-NO-LTO-PIC i386 nightly tester results

[Evan Cheng]

On Jan 8, 2009, at 6:32 PM, Evan Cheng wrote:

> This is a makefile bug. I'll fix it.

Actually it's not. It's doing exactly what's intended. When it's  
producing the bc file for individual files, it's running -O3 instead  
of -O0. That means it's running the llvm inliner on individual .c files.

So yes, we are seeing significant performance regressions as a result  
of disabling gcc inlining. We need to look into these.

Evan

>
>
> Evan
> On Jan 7, 2009, at 5:09 PM, Dale Johannesen wrote:
>
>>
>> On Jan 7, 2009, at 10:55 AMPST, Evan Cheng wrote:
>>
>>> I see significant loses (and wins). Is this due to disabling of gcc
>>> inliner?
>>>
>>> To reproduce do something like:
>>> make DISABLE_LTO=1 TARGET_FLAGS= TARGET_LLCFLAGS="-relocation-
>>> model=pic -disable-fp-elim"
>>
>> When invoked this way the makefile invokes opt, and llvm-ld, with -
>> disable-inlining.  So, yes, but I don't think it necessarily  
>> indicates
>> a problem.
>>
>> _______________________________________________
>> 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