Author: Vincent Diepeveen
Date: 07:37:34 01/08/01
Go up one level in this thread
On January 07, 2001 at 05:31:23, Gian-Carlo Pascutto wrote:
>On January 06, 2001 at 22:34:59, Robert Hyatt wrote:
>
>>Always has been.. Doesn't optimize as well as Microsoft's compiler...
>
>Have you ever experimented with a snapshot of gcc-2.96/3.0 ?
>
>It's supposed to have a completely new x86 backend and several
>additional optimizations as well. Maybe it would be interesting
>to experiment with it.
Yes i did regurarly.
First of all the IA-32 backbone doesn't give much extra.
Basically some datastructure initialization code before it
starts searching it is using a FEW of those instructions.
So imagine a program that's real big. around 2 MB source and
from that 2 MB source only a few lines in the initialization code
where speed is not so important as it only gets done once, only
there it uses P6 instructions!
Not very good.
Secondly gcc-2.96 has had since integration of the ia-32 backbone
bugs in optimizing 2 dimensional arrays from typedefs with many
loops around it (datastructure initialization).
I have a define:
#if GCCBUG
#else
#endif
I rewrote that 2 dimensional array code back to 1 dimensional array
code by using an extra pointer.
then it doesn't crash on that.
It's not faster as 2.95.2 not very surprising of course if it hardly
uses the IA-32 backbone.
At a PII gcc is 8.5% slower as msvc
at a PIII gcc is 10.8% slower as msvc
I don't know why it's so much slower. I lack the insight in assembly
to really see where MSVC gets the 10.8%, but i guess it's register
efficiency.
>Also, have you ever figured out why Crafty's preformance drops
>with higher optimization levels in GCC? If you could find this
>out it could be very interesting for the GCC folks...
Register efficiency?
If you unroll more you get more variables. A compiler that's already
bad in using registers efficient has perhaps a big problem then.
>--
>GCP
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.