Author: Robert Hyatt
Date: 15:11:29 09/26/02
Go up one level in this thread
On September 26, 2002 at 15:05:39, Dieter Buerssner wrote: >On September 25, 2002 at 16:51:17, Dann Corbit wrote: > >>Here is what my compiler spit out (I could probably choose better flags too): > >When I look at the generated code, it seems, you couldn't choose better flags. >To me, it looks essentially identical to the hand written assembly. The few >differences (slightly other order of things, add eax, 1 instead of inc eax) I >cannot judge. With my assembler knowledge mainly from 386 days and before, I >would even trust the compiler more here, than myself writing this in assembler >(I would come up with practically identical code as well). I could imagine, that >the compiler knows something about pairing of the instructions, which I don't. >He could also know better, about which labels should be aligned (perhaps >dependent on specific processor switches). He can know easily, how many bytes >each instruction needs. Perhaps using add eax, 1 makes a better alignment in the >end, than the shorter inc eax. > >Gcc also produced practically the same code. > >Regards, >Dieter Don't forget that since the P6 (pentium pro) "pairing instructions" is no longer a big issue. The CPU does this internally by fetching a wad of instructions, and thru dataflow analysis, figures out which can proceed in parallel without the older (original pentium) pairing/scheduling as required back then... The hardware guys are making some things easier for the compiler guys. :)
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.