randy@WHEATIES.AI.MIT.EDU (09/29/88)
This would be fairly easy to do, seeing as -fswitch (under sun): 1) Compiles code as with -msoft_float. 2) Links with a different crt1.o file. 3) That's it. The new crt1.o file forces library calls to be switched based on checks done at startup time. It would be easy to get gcc to do the above; I just haven't done it. If somebody else wants to look into it, well, it probably wouldn't take up too much of your time to implement. If nobody else does, I'll probably do it one of these years (however, I'm afraid this translates to: don't hold your breath). -- Randy (Programmer, Freesoft) randy@wheaties.ai.mit.edu P.S. Followups directly to me or to bug-gcc if they are of a technical nature; I don't want to bore the info-gcc list.
bob@allosaur.cis.ohio-state.edu (Bob Sutterfield) (09/30/88)
Are there any thoughts toward adding floating point support to gcc that would be switchable at run time between -m68881 and -msoft-float? It would eliminate some compile-time gymnastics. Here's an excerpt from Sun's cc(1), describing a similar option: -fswitch Run-time-switched floating-point calls. The compiled object code is linked at runtime to routines that support one of the above types of floating point code. [-f68881, -ffpa, -fsky, -fsoft] This was the default in previous releases. Only for use with programs that are floating-point intensive, and must be portable to machines with various floating-point hardware options. -=- Zippy sez, --Bob Our father who art in heaven.. I sincerely pray that SOMEBODY at this table will PAY for my SHREDDED WHAT and ENGLISH MUFFIN.. and also leave a GENEROUS TIP...