[Bug target/35491] New: wrong ABI for some struct passing with vector code

[Bug target/35491] New: wrong ABI for some struct passing with vector code

[pinskia at gcc dot gnu dot org]

Take the following code:
#define vector __vector
struct data {
     float f;  //0 - 3
     int i;    // 4 - 7
     double d; // 8 - 15
     vector float v;  // 16 - 31
   };// __attribute__ ((d64_abi));  //size is 32

   extern int bar (int a, struct data b, void *c);

   int foo (struct data *inp_r3, void *inp_r4)
   {
     return bar(10, *inp_r3, ((void *) inp_r4));
   }

--- CUT ---
We get:
_foo:
        mflr r0
        std r0,16(r1)
        stdu r1,-128(r1)
        mfspr r0,256
        stw r0,124(r1)
        oris r0,r0,0x2000
        mtspr 256,r0
        li r0,16
        lfs f1,0(r3)
        lvx v2,r3,r0
        mr r10,r4  <--- note r10
        lfd f2,8(r3)
        lwz r5,4(r3)
        li r3,10
        bl L_bar$stub
        addi r1,r1,128
        lwz r12,-4(r1)
        mtspr 256,r12
        ld r0,16(r1)
        mtlr r0
        blr

--- CUT ---
With Apple's compiler we get:
_foo:
        mfspr r0,256
        stw r0,-12(r1)
        oris r0,r0,0x2000
        mtspr 256,r0
        mflr r0
        mr r9,r4  <--- NOTE r9
        std r0,16(r1)
        stdu r1,-128(r1)
        li r0,16
        nop
        lfs f1,0(r3)
        lfd f2,8(r3)
        lwz r5,4(r3)
        lvx v2,r3,r0
        li r3,10
        bl L_bar$stub
        addi r1,r1,128
        lwz r12,-12(r1)
        mtspr 256,r12
        ld r0,16(r1)
        mtlr r0
        blr


-- 
           Summary: wrong ABI for some struct passing with vector code
           Product: gcc
           Version: 4.3.0
            Status: UNCONFIRMED
          Keywords: wrong-code, ABI
          Severity: normal
          Priority: P3
         Component: target
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: pinskia at gcc dot gnu dot org
GCC target triplet: powerpc64-darwin


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35491

[Bug target/35491] wrong ABI for some struct passing with vector code

[pinskia at gcc dot gnu dot org]

------- Comment #1 from pinskia at gcc dot gnu dot org  2008-03-11 00:56 -------
The fix for this ABI issue was radar 4028089 in Apple's bug database.  I can
attach the patch which I extracted which fixes this from Apple's GCC if someone
wants.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35491

[Bug target/35491] wrong ABI for some struct passing with vector code

[pinskia at gcc dot gnu dot org]

------- Comment #2 from pinskia at gcc dot gnu dot org  2008-03-11 00:57 -------
Created an attachment (id=15293)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=15293&action=view)
Patch which comes from Apple's GCC


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35491

[Bug target/35491] wrong ABI for some struct passing with vector code

[iains at gcc dot gnu dot org]

------- Comment #3 from iains at gcc dot gnu dot org  2010-07-24 08:41 -------
http://gcc.gnu.org/ml/gcc-patches/2010-07/msg01944.html


-- 

iains at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |iains at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35491

[Bug target/35491] wrong ABI for some struct passing with vector code

[iains at gcc dot gnu dot org]

------- Comment #4 from iains at gcc dot gnu dot org  2010-07-27 13:24 -------
Subject: Bug 35491

Author: iains
Date: Tue Jul 27 13:24:08 2010
New Revision: 162567

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=162567
Log:

        PR target/35491
        PR target/29090

        Merge from Apple local 4.2.1.
        2005-05-11  Stan Shebs  <shebs@apple.com>
        Fix 64-bit varargs for Darwin (Radar 4028089).
        * config/rs6000/rs6000.h (rs6000_args): New field floats_in_gpr.
        * config/rs6000/rs6000.c (rs6000_darwin64_record_arg_advance_flush):
        Add argument, add case for 8-byte register half-filled with a float.
        (rs6000_darwin64_record_arg_advance_recurse): Detect and handle
        single-precision floats specially.


Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/rs6000/rs6000.c
    trunk/gcc/config/rs6000/rs6000.h


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35491