On Sat, Oct 22, 2011 at 06:54:05PM -0400, David Miller wrote:
>
> Just let the compiler do it for you, please... the compiler absolutely cannot
> optimize away a function that uses a volatile asm statement. And it absolutely
> cannot make a function that clobbers %o7 a leaf function.
That's not what I see. I tried writing up a patch for Ruby using this,
and here's what I get:
1. Preprocessed source of cont.c contains flush_register_windows
definition:
# 219 "./include/ruby/defines.h"
static void
flush_register_windows(void)
{
asm __volatile__ ("flushw" : : : "o7");
}
2. Preprocessed code of cont_save_machine_stack contains a call to
flush_register_windows.
3. In cont.s both cont_save_machine_stack and flush_register_windows
are completely optimized out when compiled with -O2. Relevant
snippet:
.LL47:
call ruby_xmalloc2, 0
mov 4, %o1
ld [%fp-12], %g3
st %o0, [%g3+28]
.LLBB115:
.LLBB114:
.loc 2 222 0
#APP
! 222 "./include/ruby/defines.h" 1
flushw
! 0 "" 2
#NO_APP
.LLBE114:
.LLBE115:
.loc 1 353 0
ld [%fp-12], %g1
sll %i4, 2, %o2
call memcpy, 0
ld [%g1+32], %o1
In assembled form (objdump -d) it looks like this:
54c: 40 00 00 00 call 54c <cont_capture+0x24c>
550: 92 10 20 04 mov 4, %o1
554: c6 07 bf f4 ld [ %fp + -12 ], %g3
558: d0 20 e0 1c st %o0, [ %g3 + 0x1c ]
55c: 81 58 00 00 flushw
560: c2 07 bf f4 ld [ %fp + -12 ], %g1
564: 95 2f 20 02 sll %i4, 2, %o2
568: 40 00 00 00 call 568 <cont_capture+0x268>
56c: d2 00 60 20 ld [ %g1 + 0x20 ], %o1
I also tried listing the clobbered register as "%o7" instead of "o7",
as the examples I could find prepend the register names with % in the
clobber list. The results were the same.
Best regards,
--
Jurij Smakov jurij@xxxxxxxxx
Key: http://www.wooyd.org/pgpkey/ KeyID: C99E03CC
--
To unsubscribe from this list: send the line "unsubscribe sparclinux" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
