------- Additional Comments From ebotcazou at gcc dot gnu dot org 2005-01-04
06:46 -------
> MAIN__:
> save %sp, -112, %sp
> st %g0, [%fp-16]
> st %g0, [%fp-12]
> call cabsf, 0
> add %fp, -16, %o0
> ^^^^ This should be an ld.
No, this is correct, see the 32-bit calling conventions for complex types as
documented in config/sparc/sparc.c.
Here's the diff between the assembly output on Solaris and Linux:
@@ -9,11 +9,11 @@
.type MAIN__, #function
.proc 020
MAIN__:
- save %sp, -120, %sp
- st %g0, [%fp-24]
- st %g0, [%fp-20]
+ save %sp, -112, %sp
+ st %g0, [%fp-16]
+ st %g0, [%fp-12]
call cabsf, 0
- add %fp, -24, %o0
+ add %fp, -16, %o0
sethi %hi(.LLC1), %g1
ld [%g1+%lo(.LLC1)], %f8
fcmpes %f0, %f8
So it's basically the same assembly and the testcase passes on Solaris.
> I have no clue what is causing this.
The runtime (specifically cabsf) was probably compiled with the old calling
conventions for complex types; it must be recompiled with GCC 3.4.x.
--
What |Removed |Added
----------------------------------------------------------------------------
CC| |ebotcazou at gcc dot gnu dot
| |org
Status|UNCONFIRMED |RESOLVED
Resolution| |INVALID
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19247
