This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: [lno] 50% runtime performance regression since yesterday
On Wed, Jun 23, 2004 at 05:11:17PM +0200, Richard Guenther wrote:
>
> Didn't change anything. Diffstat on a diff -cb from yesterdays .s to
> todays .s with your patch says (uh, ia64 asm is ugly!)
>
I cannot reproduce the regression even with the leafify patch.
The version including the patch
2004-06-21 Zdenek Dvorak <rakdver@atrey.karlin.mff.cuni.cz>
produces identical code on i686 as the last version of LNO in cvs.
The run time is almost the same for older (2004-06-22) or newer
versions (today) on a P4 3GHz.
i = 1 t = 0.00407921 dt = 0.00407921 (0.050695s/it)
i = 2 t = 0.00679089 dt = 0.00271169 (1.51221s/it)
i = 3 t = 0.00937923 dt = 0.00258833 (1.52281s/it)
i = 4 t = 0.0120634 dt = 0.0026842 (1.52458s/it)
i = 5 t = 0.0150383 dt = 0.00297491 (1.52104s/it)
i = 6 t = 0.0183952 dt = 0.00335687 (1.52298s/it)
i = 7 t = 0.0217122 dt = 0.00331702 (1.51668s/it)
i = 8 t = 0.0250681 dt = 0.00335591 (1.51319s/it)
i = 9 t = 0.0284961 dt = 0.00342798 (1.52687s/it)
i = 10 t = 0.032036 dt = 0.00353984 (1.51509s/it)
The line that I use for compiling the tramp3d-v3 is
cc1plus -O2 -funroll-loops -ffast-math tramp3d-v3.cpp
as requested. I tried with and without -ffast-math with
the same quite short diff:
--- ../20040621/tramp3d-v3.s 2004-06-23 19:13:04.000000000 -0400
+++ ../20040623new/tramp3d-v3.s 2004-06-23 19:15:40.000000000 -0400
@@ -2645,5 +2645,5 @@ _ZTS11CreateEvent:
.weak pthread_key_delete
.weak pthread_key_create
.weak pthread_once
- .ident "GCC: (GNU) 3.5-tree-ssa-lno 20040622 (merged 20040612)"
+ .ident "GCC: (GNU) 3.5-tree-ssa-lno 20040623 (merged 20040612)"
.section .note.GNU-stack,"",@progbits
Maybe I have to try on ia64... but I no longer have access to one of
these beasts for the moment.
Sebastian