Author: Dieter Buerssner
Date: 13:59:09 08/06/03
Go up one level in this thread
On August 05, 2003 at 14:27:03, Eugene Nalimov wrote: >(1) Some ancient decisions were made in gcc design when VAX and mc68k were >primary gcc targets; the worst one is poor memory disambiguation. Now that >decisions hurt optimizations a lot, especially on CPUs with large number of >registers, but fixing those decisions require *highly coordinated* rewrite of >lot of code for lot of targets. I don't want to doubt, what you say. But for earlier versions of gcc, I made some other observations. For identical functions, where it produced seemingly good machine code on Alpha, gcc produced onx x86 stupid code like mov reg, some_memory_on_stack ; nothing in between mov some_memory_on_stack, reg Perhaps I should mention, that it is an integer reg (for floating point, one could see reasons). Some functions were full of such things. I reported it to Gcc-maintainers. "Sorry, no time to look into that", "It's because x86 has too few registers" (paraphrasing answers I got). So, it gave me the impression, that when Gcc runs out of registers, it started to produce such things. Regards, Dieter
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.