This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH,4.6/4.5.1,PR42776] Implement LTO for Windows (PE-COFF) targets.
On 17/04/2010 14:30, Dave Korn wrote:
> On 15/04/2010 18:40, Steven Bosscher wrote:
>> Could you commit these bits separately?
> I've just factored out the lto_elf_xxxx -> lto_obj_xxxx changes.
> Results tomorrow sometime.
Something strange, unexpected and a bit alarming happened:
> --- summary.clean.log 2010-04-18 18:53:44.093750000 +0100
> +++ summary.patched.log 2010-04-18 18:54:04.625000000 +0100
> @@ -44,12 +44,12 @@ WARNING: gcc.target/i386/avx-vmovntps-25
>
> === gcc Summary ===
>
> -# of expected passes 72800
> +# of expected passes 72799
> # of unexpected failures 13
> # of unexpected successes 19
> # of expected failures 192
> -# of unsupported tests 797
> -/gnu/gcc/obj.clean/gcc/xgcc version 4.6.0 20100417 (experimental) (GCC)
> +# of unsupported tests 798
> +/gnu/gcc/obj.patched/gcc/xgcc version 4.6.0 20100417 (experimental) (GCC)
... which turns out to be because:
> --- obj.clean/gcc/testsuite/gcc/gcc.sum 2010-04-18 18:32:09.000000000 +0100
> +++ obj.patched/gcc/testsuite/gcc/gcc.sum 2010-04-18 06:07:32.000000000 +0100
> @@ -1,4 +1,4 @@
> -Test Run By davek on Sun Apr 18 15:01:13 2010
> +Test Run By davek on Sun Apr 18 03:08:06 2010
> Native configuration is i686-pc-linux-gnu
>
> === gcc tests ===
> @@ -56264,7 +56264,7 @@ PASS: gcc.dg/guality/pr43479.c -O0 lin
> PASS: gcc.dg/guality/pr43479.c -O0 line 12 n == 9
> PASS: gcc.dg/guality/pr43479.c -O1 (test for excess errors)
> PASS: gcc.dg/guality/pr43479.c -O1 execution test
> -PASS: gcc.dg/guality/pr43479.c -O1 line 13 i == 6
> +UNSUPPORTED: gcc.dg/guality/pr43479.c -O1 line 13 i == 6
> UNSUPPORTED: gcc.dg/guality/pr43479.c -O1 line 13 h == 9
> UNSUPPORTED: gcc.dg/guality/pr43479.c -O1 line 13 n == 9
> UNSUPPORTED: gcc.dg/guality/pr43479.c -O1 line 18 j == 8
I don't know what exactly the guality tests are testing (until I read the
source some) but it must be something to do with the accuracy of VTA debug
info, and I don't see that renaming a few internal subroutines inside the
compiler ought to change the generated debug info. Particularly when LTO
isn't even in use. This is not good, I think, but won't know for sure until
I've done some digging. (I suspect it can't actually be directly caused by
the patch but is just a side effect of something using uninitialised memory or
perhaps hashing pointers when it shouldn't do.)
cheers,
DaveK