[PATCH, i386] Optionally use %xmm0 to return float and/or double values (take 2)

[Paolo Bonzini]

> However, tomorrow (unless Uros beats me to it :-) I'll benchmark povray: 
> a much simpler patch that only does this for local functions might still 
> provide a good bang for the buck.

Here's the simpler patch, bootstrapped/regtested C/C++ on 
i686-pc-linux-gnu (a Pentium 4).  Like in the previous patch, the 
special calling convention is used as long as the ISA supports it (e.g. 
for floats on SSE processors, or for doubles on SSE2), but this time 
only for local functions and with -mfpmath=sse.

Again, the code does not apply to x86-64 platforms, for which 
-mfpmath=sse is the default, so it requires a special command-line option.

I'll provide benchmark numbers tomorrow, however I still ask if this is 
ok for mainline?

Paolo
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: fp-ret-sse-2.patch
URL: <http://gcc.gnu.org/pipermail/gcc-patches/attachments/20050125/ca6c22dd/attachment.ksh>