Re: rs6000_flag_pic

[Alan Modra <amodra at bigpond dot net dot au>]

On Wed, Mar 12, 2003 at 05:32:51PM -0500, David Edelsohn wrote:
> 	The current behavior is correct and necessary, not overly
> pessimistic. 

Please explain.  I've been using this patch for quite some time with
no apparent ill effects.  The following simple example shows a case
where the current code is overly pessimistic:

$ cat > const.c <<EOF
const char *const msg = "hi!";
EOF
$ gcc64-curr/gcc/xgcc -Bgcc64-curr/gcc/ -S -O2 const.c 
$ cat const.s
        .file   "const.c"
        .section        .rodata.str1.8,"aMS",@progbits,1
        .align 3
.LC0:
        .string "hi!"
        .globl msg
        .section        .data.rel.ro.local,"aw",@progbits
        .align 3
        .type   msg, @object
        .size   msg, 8
msg:
        .quad   .LC0
        .ident  "GCC: (GNU) 3.4 20030311 (experimental)"
$ gcc64-rs6000/gcc/xgcc -Bgcc64-rs6000/gcc/ -S -O2 const.c 
$ cat const.s
        .file   "const.c"
        .section        .rodata.str1.8,"aMS",@progbits,1
        .align 3
.LC0:
        .string "hi!"
        .globl msg
        .section        .rodata
        .align 3
        .type   msg, @object
        .size   msg, 8
msg:
        .quad   .LC0
        .ident  "GCC: (GNU) 3.4 20030311 (experimental)"

Note the selection of a writable section for "msg" in the unpatched
(first) case.

-- 
Alan Modra
IBM OzLabs - Linux Technology Centre