------- Comment #8 from howarth at nitro dot med dot uc dot edu 2010-09-04
19:13 -------
(In reply to comment #7)
> I fixed the test case to not expect the unimplemented optimization in r157197,
> however, it would be nice to ensure the async signal handlers can handle
> unaligned stacks and to perform this optimization. I'm fairly certain the
> signal handers have to cope as code gen can do:
>
> _h:
> subl $12, %esp
> movl $1, %eax
> addl $12, %esp
> ret
>
> for int h () { return 1; } and certainly we can take a signal at _h+0, where
> esp isn't aligned. Given that, we can omit sp alignments for leaf functions
> (and no tail calls either), even on machines where otherwise an alignment is
> required, as long as any variables on the stack are correctly aligned.
>
> When this optimization is added, we can undo the r157197 change.
>
My proposed patch to fix PR36502...
http://gcc.gnu.org/ml/gcc-patches/2010-09/msg00237.html
that enables stack realignment on intel darwin also solves this PR as well.
Comparing the output from gcc trunk before and after my patch, I see...
--- builtin-unreachable.s 2010-09-04 15:12:40.000000000 -0400
+++ builtin-unreachable.trunk_patched 2010-09-04 15:02:45.000000000 -0400
@@ -3,11 +3,7 @@
.globl _h
_h:
LFB0:
- subl $12, %esp
-LCFI0:
movl $1, %eax
- addl $12, %esp
-LCFI1:
ret
LFE0:
.section
__TEXT,__eh_frame,coalesced,no_toc+strip_static_syms+live_support
@@ -39,16 +35,6 @@
.set L$set$2,LFE0-LFB0
.long L$set$2
.byte 0
- .byte 0x4
- .set L$set$3,LCFI0-LFB0
- .long L$set$3
- .byte 0xe
- .byte 0x10
- .byte 0x4
- .set L$set$4,LCFI1-LCFI0
- .long L$set$4
- .byte 0xe
- .byte 0x4
.align 2
LEFDE1:
.subsections_via_symbols
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42313
