Author: Gerd Isenberg
Date: 11:35:44 03/30/04
Go up one level in this thread
On March 30, 2004 at 13:49:15, Gian-Carlo Pascutto wrote: >On March 30, 2004 at 13:43:38, Gerd Isenberg wrote: > >>As i saw so far, the 64-bit speedup is not as good as expected. >>I can't imagine that the signed index issue is that expensive. > >Fixing it gives about a 5-7% overall speedup for SOS and Deep Sjeng. Oups - that's a considerable amount of course! MOVSX instead of MOV very often with one byte longer opcode each? > >>With similar optimization abilities the 64-bit code should be much faster, >>simply due to the eight additional registers and (per default? fastcall instead >>of cdecl or pascal parameter passing). Alignment and Load/Store forewarding >>stalls are important issues too, but even true for 32-bit code. >> >>My guess is, that the 64-bit ms-compiler you use is yet too safe or paranoid >>with aliasing issues and tends to write locals inside registers to memory >>permanently. > >Yes. Speedups with GCC are higher than with MSVC betas. Especially new >GCC 3.4 is impressive. > >-- >GCP Interesting news. I guess the ms guys are working hard ;-) Gerd
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.