On Sun, Jan 13, 2013 at 10:06:05PM -0800, Andi Kleen wrote:
> 2013-01-12 Andi Kleen <a...@linux.intel.com>
> Jonathan Wakely <jwakely....@gmail.com>
>
> PR libstdc++/55223
> * testsuite/29_atomics/atomic_flag/test_and_set/explicit-hle.cc:
> Add.
The test is broken on i?86-linux,
> --- /dev/null
> +++
> b/libstdc++-v3/testsuite/29_atomics/atomic_flag/test_and_set/explicit-hle.cc
> @@ -0,0 +1,120 @@
> +// { dg-options "-std=gnu++0x" }
1) you need // { dg-additional-options "-march=i486" { target ia32 } }
or something similar, otherwise for a pure i386 target it will not work right
> +// { dg-do compile { target i?86-*-* x86_64-*-* } }
> +// { dg-final { scan-assembler-times "\(xacquire\|\.byte.*0xf2\)" 14 } }
> +// { dg-final { scan-assembler-times "\(xrelease\|\.byte.*0xf3\)" 14 } }
2) \.byte.*0xf needs to be replaced by \.byte\[^\n\r]*0xf
otherwise it will happily match say
.byte 0
.uleb128 0x2d
.long 0xf23
in .debug_info
3) I guess you want to add "-g0 -fno-exceptions -fno-asynchronous-unwind-tables"
to double check that .byte 0xf2 or .byte 0xf3 won't suddenly appear in
.debug_* or .eh_frame sections.
Jakub
