Author: Gerd Isenberg
Date: 15:22:03 06/15/04
Go up one level in this thread
>I see no way to convince MSVC to produce inline code for the multiplication, it >always calls that __allmul function (it has no problem to produce good code for >32-bit*32-bit -> 64-bit multiplications). The relative slowness seems just to >come from those 3 multiplications for Gcc. Perhaps there are some magic numbers >where doing the multiplication by the constant with the typical >shift/add/sub/lea tricks would be faster. But I doubt it. Who knows, i guess the ones with the most "ones" near the bottom are fastest ;-) Simply run my de Bruijn generator. With a found de Bruijn call a routine and stop the time with performance counters of some muls in a loop, store fastest, slowest and sum for average times etc. of all 2**27 de Bruijns in 2**26 runs ;-) Or try the extremes: 0x0218a392cd3d5dbf; // the 1. 0x07ef3ae369961512; // the 134217728. Gerd > >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.