Re: [PATCH]: Promote float args on i386 (attn Hurd, Cygwin, DJGPP...)

[Michael Snyder <msnyder at cygnus dot com>]

Mark Kettenis wrote:

> It's got nothing to do with the Linux version, and probably everything
> with the compiler.  Could you post the assembler output that your compiler
> generates for the following fragment:
> 
> extern int foo (float f, float g);
> 
> void
> bar (void)
> {
>   foo (3.14, 2.72);
> }

As you'll see below, I get something completely different.

> In my case this yields:
> 
>         .file   "foo.c"
>         .version        "01.01"
> gcc2_compiled.:
> .section        .rodata
>         .align 4
> .LC0:
>         .long 0x402e147b
>         .align 4
> .LC1:
>         .long 0x4048f5c3
> .text
>         .align 4
> .globl bar
>         .type    bar,@function
> bar:
>         pushl %ebp
>         movl %esp,%ebp
>         subl $8,%esp
>         addl $-8,%esp
>         flds .LC0
>         subl $4,%esp
>         fstps (%esp)
>         flds .LC1
>         subl $4,%esp
>         fstps (%esp)
>         call foo
>         addl $16,%esp
> .L2:
>         movl %ebp,%esp
>         popl %ebp
>         ret
> .Lfe1:
>         .size    bar,.Lfe1-bar
>         .ident  "GCC: (GNU) 2.95.2 19991024 (release)"
> 
> The "subl $4,%esp" and "fstps (%esp)" instructions show that the
> arguments passed to foo() are both four bytes in size.  So they can't
> be doubles.  This is what I'd expect after reading the processor
> specific SystemV ABI, which AFAIK is used on Linux.

Here's what I get:

        .file   "foo.c"
        .version        "01.01"
gcc2_compiled.:
.text
        .align 4
.globl bar
        .type    bar,@function
bar:
        pushl %ebp
        movl %esp,%ebp
        pushl $0x402e147b
        pushl $0x4048f5c3
        call foo
        addl $8,%esp
.L1:
        leave
        ret
.Lfe1:
        .size    bar,.Lfe1-bar
        .ident  "GCC: (GNU) egcs-2.91.66 19990314/Linux (egcs-1.1.2
release)"

This is on a Redhat 6.2 distribution with arch == i686.

Michael