This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Big differences on SpecFP results for gcc and icc


	We have investigated these benchmarks for PowerPC.  The high-level
analysis is:

>>>>> Daniel Berlin writes:

>> An interesting examples are:
>> -177.mesa (this is a c test), where icc is almost 40% faster

	FP to Int conversion.

Dan> SSE Vectorization, I believe.
>> -178.galgel, where icc is again 40% faster

Dan> This is because of a better fortran *library*, not compiler.
Dan> IIRC, it's all in matmul.

	Agreed, Fortran90 arrays and matmul.

>> -179.art, where llvm is more than 1.5x faster than both gcc and icc

Dan> taken care of on struct-reorg.

	Agreed, art hack on struct-reorg branch.

>> -187.facerec, where icc is 100% faster than gcc
Dan> No idea.

	Fortran90 arrays.

>> -189.lucas, where icc is 60% faster
Dan> No idea, though i'd imagine its the same issue.

	Fortran90 arrays and FP to Int conversion.

David


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]