This is the mail archive of the
mailing list for the GCC project.
Re: PATCH] PPC: Correct SYSV ABI libcall handling
- From: Franz Sirl <Franz dot Sirl-kernel at lauterbach dot com>
- To: David Edelsohn <dje at watson dot ibm dot com>
- Cc: gcc-patches at gcc dot gnu dot org
- Date: Tue, 17 Jun 2003 23:36:06 +0200
- Subject: Re: PATCH] PPC: Correct SYSV ABI libcall handling
- References: <200306172108.RAA27298@makai.watson.ibm.com>
On Tuesday 17 June 2003 23:08, David Edelsohn wrote:
> >>>>> Franz Sirl writes:
> Franz> Where did ABI_AIX ever use it for the mode == VOIDmode case? This is
> only used Franz> on ABI_V4, cause only ABI_V4 needs to set/clear a CR bit
> to signal a vararg Franz> callee if FP args where passed (at least I don't
> know of a similar Franz> requirement for ABI_AIX).
> ABI_AIX does not use it for the exact same situation, but it does
> use type to determine whether FP parameters can be passed in just FPRs or
> need to be duplicated in GPRs as well.
Ah, ok. So the question was not directly targeted at my patch, rather you
wonder if similar measures are needed/correctly handled on ABI_AIX?
This would be easy to see in the generated assembly if you can recognize the
typical FPR->GPR sequences generated by the compiler. If they are not there
for libcalls, all should be fine (obviously type being NULL didn't cause any
problems so far, so it's enough to look for useless code pessimizations).
Hmm, it seems ABI_AIX does the handling outside of the call/call_value
patterns (which get the call_cookie passed passed as op2 and op3), so
certainly the code of a potential fix would be different as well.