chuckp@bach@ccncsu.colostate.edu (Chuck Phillips) (11/15/89)
This isn't a gcc bug, just an unfortunate consequence of the various floating point options for Sun3's. PROBLEM: When you compile gnulib and gnulib2 with the native compiler, you have to choose a floating point option for some of the routines in gnulib. You're then stuck with that choice regardless of the floating point option you pass to gcc later on. PROPOSAL: Separate the floating point functions from the rest of the gnulib functions into gnulibm_68881.a, gnulibm_soft.a, gnulibm_fpa.a, etc. (The user could select the targets.) Then when the user selects a floating point option, _all_ floating point operations would be affected consistantly. Although the only machine I'm personally familiar with that has this problem is the Sun3 series, I would imagine it applies to others. (e.g. An 80386 with either an 80387 or a Wietek math coprocessor or none at all) I'm willing to help on implementing this, but I'd like an OK on it before I start. Chuck Phillips -- chuckp%bach.ncr-fc.FtCollins.NCR.COM or phillips%holst.cs.ColoState.edu