[llvm-testresults] buildbot failure in smooshlab on llvm-gcc-powerpc-darwin9

[Dale Johannesen]

On Feb 26, 2010, at 2:40 PMPST, Dale Johannesen wrote:
> On Feb 26, 2010, at 2:35 PMPST, Evan Cheng wrote:
> 
>> No, it's not 97262. It's this:
>>> 
>>> I wonder if r97262 has fixed this issue. PPC could accidentally store a f32 to a f64 spill slot, so floating point could definitely get badly corrupted. I don't think it was possible to store a f64 to a f32 spill slot. That would be catastrophic, overwriting the neighboring slot.
> 
> I do have hopes it will fix the other PPC regression in Povray that started at the same time, though.  That does lots of floating point.  Let you know.

Both 175.vpr and Povray are fixed.  oggenc is also working; in the past it has succeeded and failed pseudo-randomly, so this may not be permanent.  OTOH that's exactly the kind of behavior this bug could cause, so maybe it is fixed.  A good one to catch.