[PATCH]: Add more altivec builtins
Daniel Berlin
dan@cgsoftware.com
Wed Dec 5 11:47:00 GMT 2001
On 5 Dec 2001, Aldy Hernandez wrote:
> >>>>> "Daniel" == Daniel Berlin <dan@cgsoftware.com> writes:
>
> > Past some that are trivial (I.E. Don't even have an overloaded function
> > that could generate them), some that are faked (abs, for instance), and a
> > few conversion functions, this is all of them.
>
> Can you send me a list of the ones that are missing?
Sure.
I've not listed the ones that builtins can't generate
(only because i don't have that list in front of me).
vsum4ubs
vsum4sbs
vsum4shs
vsum2sws
vcfux
vcfsx
vctuxs
vctsxs
vexptefp
vlogefp
vrsqrtefp
vrfin
vrfip
vrfim
vsel
I'll get to them today, hopefully.
> > There are a few more that could be converted from unspecs (the fp
> > compares), i just haven't had time to do it.
>
> Thanks so much.
>
> Now for the status report-- what's missing (volunteers highly welcome,
> so we can get this in time for 3.1):
>
> - darwin varargs abi changes (stan shebs)
>
> - vector_size attribute (done, testing patch)
>
> - __builtin_types_compatible_p for overloaded functions in C
> (done, but i need to correct lots of documentation and
> testing issues that Joseph pointed out).
>
> - provide altivec.h with overloaded functions, etc (undone)
>
> - vector initializers (undone)
>
> - DWARF_FRAME_REGISTERS: fix usage of this throughout the
> compiler to be an expression, not a constant. Eg. Stuff
> like "int foo[DWARF_FRAME_REGISTERS]" to use alloca. (undone)
> cheers
> all-dee
>
More information about the Gcc-patches
mailing list