This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: Patch: fix regression in PPC sibcalls
- From: Dale Johannesen <dalej at apple dot com>
- To: Franz Sirl <Franz dot Sirl-kernel at lauterbach dot com>
- Cc: Dale Johannesen <dalej at apple dot com>, gcc-patches at gcc dot gnu dot org, joern dot rennecke at superh dot com
- Date: Fri, 3 Jan 2003 12:03:17 -0800
- Subject: Re: Patch: fix regression in PPC sibcalls
On Friday, December 27, 2002, at 08:49 AM, Franz Sirl wrote:
On Saturday 21 December 2002 20:21, Dale Johannesen wrote:
This patch breaks ppc (also sparc and probably others, I think)
sibcalls:
Wed Apr 24 21:58:09 2002 J"orn Rennecke <joern.rennecke@superh.com>
* calls.c (expand_call): Take
current_function_pretend_args_size
into account when setting argblock for sibcalls.
It is not right to do the adjustment in that patch when the caller has
already allocated
the stack space for partially-in-regs params, that is, when
REG_PARM_STACK_SPACE is defined.
Bootstrapped and tested on darwin.
2002-12-21 Dale Johannesen <dalej@apple.com>
* calls.c (expand_call): Fix sibcalls on
REG_PARM_STACK_SPACE
machines.
Does this cover the REG_PARM_STACK_SPACE defined to 0 case as well?
After staring at this a while longer, I think the answer is no (which
means
the doc for REG_PARM_STACK_SPACE does not match the actual usage). I've
redone the patch to check for only the nonzero case, which I'm sure is
correct:
Index: calls.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/calls.c,v
retrieving revision 1.247
diff -u -d -b -w -r1.247 calls.c
--- calls.c 18 Dec 2002 05:57:43 -0000 1.247
+++ calls.c 3 Jan 2003 19:54:11 -0000
@@ -2699,12 +2716,15 @@
if (pass == 0)
{
argblock = virtual_incoming_args_rtx;
+#if !defined (REG_PARM_STACK_SPACE) || \
+ REG_PARM_STACK_SPACE (current_function_fndecl) == 0
argblock
#ifdef STACK_GROWS_DOWNWARD
= plus_constant (argblock,
current_function_pretend_args_size);
#else
= plus_constant (argblock,
-current_function_pretend_args_size);
#endif
+#endif
stored_args_map = sbitmap_alloc (args_size.constant);
sbitmap_zero (stored_args_map);
}
I'm not convinced the code above is correct for all targets, but I am
not really
equipped to test elsewhere, and the above should not break anything.
Perhaps I
should just request that J"orn Rennecke's patch above be reverted; it
clearly
breaks ppc (darwin and AIX). If this doesn't get approved I'll do
that, I guess.