From mboxrd@z Thu Jan 1 00:00:00 1970 From: Stan Shebs To: dberlin@msn.com Cc: jtc@redback.com, gdb@cygnus.com, jepler@inetnebr.com, jimb@cygnus.com Subject: Re: MMX registers on x86? Date: Wed, 14 Apr 1999 18:55:00 -0000 Message-ID: <199904150058.RAA17793@andros.cygnus.com> References: <001d01be8799$59947480$8d8f9780@DANIELBE> X-SW-Source: 1999-04/msg00036.html Message-ID: <19990414185500.YlBvzFOr7irZ5PgiTMherexc7pDJy-2ENIb6B0qQMSI@z> From: "Daniel Berlin" Date: Thu, 15 Apr 1999 19:40:26 -0400 However, it might be better to just add some mechanism to define "special" registers (Fer instance, ,3dnow/KNI/MMX) that are in some way different from the rest of the processor registers, be it in size/usage/whatever. I'm about to add KNI support to the BeOS port, as well as MMX and 3DNow support, and i'd rather have some standard supported way of doing it than making another hack that someone will have to futz with and learn the reasoning behind a year or two down the line. Or when something like Altivec comes along, it wouldn't take major retrofitting and port hacking (every single target that can run on that processor with a different -nat file) to do. Andrew Cagney has been working on some infrastructure that will, among other things, greatly simplify handling of special registers and multiple register sets. He's supposed to send a long detailed message to this list in a few days. We're hoping to roll the code out into the sources this month, give people a crack at it. Stan