[llvm-commits] [llvm] r62587 - /llvm/trunk/test/FrontendC/2009-01-20-k8.c

[Chris Lattner]

On Jan 20, 2009, at 11:51 PM, Anton Korobeynikov wrote:

> Hi, Chris
>
>> Does this work when run on a ppc host?
> The whole program is the follows: there is invalid -march processing
> logic in the apple gcc, so every x86 64 bit specified explicitly (e.g.
> k8, core2) will lead to invalid initialization of gcc's internal
> structures and lead, for example, to alignment of 4 of long double.
> There are two ways to check for this currently:
> 1. Feed empty file to fortran frontend (with march set to k8/core2),
> fortran fe constructs some aux trees during initialization and thus
> this will assert / ICE soon
> 2. Feed something using long double to C / C++ frontend.
>
> We really need some sort of test for this, otherwise I worry at some
> next merge the fix can be removed again. What if we just xfail it for
> everything non-x86 ?

I'm fine with xfailing it everywhere else.  I think it is unfortunate  
that stuff like this is so fragile :(

-Chris