Re: FP transition
Am Donnerstag, 29. September 2005 20:05 schrieb Vincent Sanders:
> What I suggest is the creation of a "hard" softfloat library...wait it
> isnt insane ;-) this would allow the compiler switches to be changed
> to compile everything with libsoftfloat. There would be no issue with
> backwards compatability because there would (continue) be no mixing of
> soft and hard float. Then at some point down the line when the archive
> has rebuilt itself (and all the issues with having packages linking
> against libsoftfloat as well a glibc have been sorted) we can offer a
> transition package to change the dynamic library over to the real soft
> float library.
How do you want to do this? I think there will be a problem with the procedure
call standard/argument passing. Presently, we pass FP arguments on the stack.
That should be considered nonsense for a hardfloat ABI, but now it is quite
practical for transition to a softfloat ABI because it does not require a
change.
But unfortunately FP return values are presently passed in the floating-point
register F0, and should be passed in r0 or r0/r1 for the softfloat ABI. How
do you plan to solve this problem?
Peter Teichmann
Reply to: