Path: utzoo!utgpu!jarvis.csri.toronto.edu!cs.utexas.edu!rice!uw-beaver!uw-june!pardo From: pardo@cs.washington.edu (David Keppel) Newsgroups: comp.sys.m88k Subject: Re: Register Allocation (was Re: Info about 88open & standards) Message-ID: <9972@june.cs.washington.edu> Date: 29 Nov 89 02:15:57 GMT References: <1989Nov16.212149.9770@paris.ics.uci.edu> <100050002@hpcuhc.HP.COM> Reply-To: pardo@june.cs.washington.edu (David Keppel) Organization: University of Washington, Computer Science, Seattle Lines: 28 thomas@uplog.se (Thomas Hameenaho) writes: >[Idea: caller register mask, callee register mask, save only > registers indicated by AND of masks.] There is a potential problem. It takes extra code to decide whether or not to save certain registers. The time you spend executing that code could be spent naively saving registers and you'd come out about the same. On the VAX, where it already has save-by-mask microcode, the tme penalty might be OK. On the RISC-family processors, you'll probably lose a lot. Is it worth adding save-by-mask hardware? Probaly not. The last time this came around I saved a bunch of articles. You can get them via anonymous ftp: ftp june.cs.washington.edu (128.95.1.4) login anonymous passwd username cd tmp/pardo get proc.call.txt About 750 lines. ;-D on ( USENET by RPC ) Pardo -- pardo@cs.washington.edu {rutgers,cornell,ucsd,ubc-cs,tektronix}!uw-beaver!june!pardo