Path: utzoo!utgpu!jarvis.csri.toronto.edu!rutgers!tut.cis.ohio-state.edu!cica!ctrsol!ginosko!infinet!rhorn From: rhorn@infinet.UUCP (Rob Horn) Newsgroups: comp.arch Subject: Re: Double Width Integer Multiplication and Division Message-ID: <1677@infinet.UUCP> Date: 12 Jul 89 02:33:12 GMT References: <1035@aber-cs.UUCP> <1370@l.cc.purdue.edu> <1333@sunset.MATH.UCLA.EDU> Reply-To: rhorn@infinet.UUCP (Rob Horn) Organization: Infinet, Inc. North Andover, MA Lines: 18 In article mac@mrk.ardent.com (Michael McNamara) writes: >[ A reference to GCC for a good way to integrate (Peter Montgomery)'s > need for access to the 68020 32x32->64bit multiply] I can also recommend using the GCC approach for this. I tweaked up one mpqs routine in a couple hours work (mostly paranoid testing because I found a gcc asm bug in the process) to replace zaddmulp. (The routine is basically a 31x31+31+carry->62 bit muladd.) I didn't go further primarily because rewriting more routines into inline asm macros then caused the optimizer to have fits and die on the large single source module. This was back on gcc rev 1.31. The current rev may well be better able to handle this. -- Rob Horn UUCP: ...harvard!adelie!infinet!rhorn ...ulowell!infinet!rhorn, ..decvax!infinet!rhorn Snail: Infinet, 40 High St., North Andover, MA