This is the mail archive of the gcc-patches@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: [patch] passes.c: Run CSE only once after GCSE instead oftwice or more.


Hi,

> > I always use my tramp3d-v3 testcase to beat compilers into ground :)
> > It's fetchable from
> > http://www.tat.physik.uni-tuebingen.de/~rguenth/gcc/tramp3d-v3.cpp.gz
> 
> Thanks for the link.  Unfortunately, GCC generates the same assmebly
> code for tramp3d with or without my patch, so I cannot use this to do
> run-time performance comparison.
> 
> By the way, here is a timing in seconds for three runs of
> 
>   ./cc1plus -qiet -fno-random-seed -O2 -o /dev/null tramp3d-v3.ii
> 
>       original patched   diff%
> user   132.193 130.740 -1.099%
> real   130.636 128.853 -1.364%
> 
> I'll try POOMA.

OK, I just tried POOMA.  Six out of 28 assembly files differ.

One of them has trivial register allocation difference.  That is, some
registers are shuffled.

The other five have hash difference that I cannot disable even with
-fno-random-seed.

-	.long	_GLOBAL__I_DiskLayout.cmpl.ii_8D2A3925_5109C10D
+	.long	_GLOBAL__I_DiskLayout.cmpl.ii_8D2A3925_107B918A

I'll move on to other test cases.  I just found
http://gcc.gnu.org/testing/ useful for testing C++ stuff.

Kazu Hirata


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