This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
Re: [RFC] Linux-specific ppc32 ABI
- From: Mark Kettenis <mark dot kettenis at xs4all dot nl>
- To: brobecker at adacore dot com
- Cc: bauerman at br dot ibm dot com, luisgpm at linux dot vnet dot ibm dot com, drow at false dot org, gdb-patches at sourceware dot org
- Date: Sat, 2 Feb 2008 01:08:47 +0100 (CET)
- Subject: Re: [RFC] Linux-specific ppc32 ABI
- References: <1199991624.3343.19.camel@gargoyle> <20080111060629.GC12954@adacore.com> <1200066920.26270.9.camel@gargoyle> <20080111155733.GA3240@caradoc.them.org> <1200086736.26270.35.camel@gargoyle> <1201277155.11950.134.camel@localhost.localdomain> <200801261535.m0QFZddH026156@brahms.sibelius.xs4all.nl> <20080201223839.GA27578@adacore.com>
> Date: Fri, 1 Feb 2008 14:38:39 -0800
> From: Joel Brobecker <brobecker@adacore.com>
>
> Hi Mark,
>
> > How about the attached diff?
>
> Sounds like everyone is happy with the patch you suggested. Thiago
> just mentioned one nit in your comments (repeated below). Can the patch
> be checked in? We'd like to have it for the release.
>
I tweaked the comment a bit more. See below what I comitted.
Thiago/Luis, can you commit the testsuite change?
Index: ChangeLog
from Mark Kettenis <kettenis@gnu.org>
Luis Machado <luisgpm@br.ibm.com>
Thiago Jung Bauermann <bauerman@br.ibm.com>
* ppc-sysv-tdep.c (ppc_sysv_abi_push_dummy_call): Pass floats that
don't fit into registerson the stack the way GCC does.
Index: ppc-sysv-tdep.c
===================================================================
RCS file: /cvs/src/src/gdb/ppc-sysv-tdep.c,v
retrieving revision 1.47
diff -u -p -r1.47 ppc-sysv-tdep.c
--- ppc-sysv-tdep.c 1 Feb 2008 15:04:18 -0000 1.47
+++ ppc-sysv-tdep.c 2 Feb 2008 00:07:14 -0000
@@ -129,17 +129,21 @@ ppc_sysv_abi_push_dummy_call (struct gdb
}
else
{
- /* SysV ABI converts floats to doubles before
- writing them to an 8 byte aligned stack location. */
- argoffset = align_up (argoffset, 8);
+ /* The SysV ABI tells us to convert floats to
+ doubles before writing them to an 8 byte aligned
+ stack location. Unfortunately GCC does not do
+ that, and stores floats into 4 byte aligned
+ locations without converting them to doubles.
+ Since there is no know compiler that actually
+ follows the ABI here, we implement the GCC
+ convention. */
+
+ /* Align to 4 bytes or 8 bytes depending on the type of
+ the argument (float or double). */
+ argoffset = align_up (argoffset, len);
if (write_pass)
- {
- char memval[8];
- convert_typed_floating (val, type, memval,
- builtin_type_ieee_double);
write_memory (sp + argoffset, val, len);
- }
- argoffset += 8;
+ argoffset += len;
}
}
else if (TYPE_CODE (type) == TYPE_CODE_FLT