|Bill Allombert on Wed, 6 Nov 2002 19:46:41 +0100|
[Date Prev] [Date Next] [Thread Prev] [Thread Next] [Date Index] [Thread Index]
|Re: CVS branch gmp-kernel-2-2-5|
Hello PARI developers, Thanks to Gerhart, ratlift is now implemented, and the gmp-kernel pass the bench. Please test, because I am not 100% sure my change is correct. Please not that due to the impossibility to know a good bound on the denominator of algebraic numbers, almost all function in PARI would still works (albeit more slowly) if ratlift just always return -1, so using bestappr() is a better way to test than make bench. For the following comparaison, I have tweaked the kernel so that gmp use the asm version of level0.h, so that odds are equal. PARI kernel w/ asm: make bench: 2278 GMP kernel w/ asm: make bench: 2368 GMP kernel w/o asm: make bench: 3114 So it looks like the gmp kernel make heavy use of the level0 kernel. It is probably due to the t_REAL functions that does not use gmp at all. Cheers, Bill.