debugging plasma ?

[Marc Espie]

On Sun, Jan 27, 2008 at 01:50:25PM +0100, Marc Espie wrote:
> I have some work-around in place, and I'm making progress on actually fixing
> the issue.

And we have some fix which is going to get committed soon.

If you want to know the gory details, the arch-dependent parts of our
pthread library is mixing up frame and stack alignment when building stacks
for new threads. As a result, the frame ends up being aligned on a 16 bytes
boundaries, and the stack is offset by 8 bytes from this.

Of course, sse2 values don't agree with that...