new RPMs
Matthew Fluet
mfluet@intertrust.com
Tue, 24 Jul 2001 23:35:38 -0700 (PDT)
> > Unfortunately, the latest x86-codegen bug is still outstanding. It turns
> > out to be a little more insidious than I originally thought. I have a
> > "fix" that works with count-graphs, but it's not a real solution. And I'm
> > not going to be able to fix it before the contest. However, I've written
> > the fix in a way that if the bug were triggered in a way that the patch
> > doesn't cover, we'll get an unresolved symbol error from the linker.
>
> You can do that but you can't raise a compile time exception? Weird.
It's just a little easier this way. I'm introducing a new temporary
memory location which is just serving as a placeholder in the register
file -- I never want to def it (and write it to memory) or attempt to
fetch it from memory. I could extend the MemLoc.t datatype with something
for this case and check these invariants, but that's a lot of work.
> > That
> > should work for the contest, in that even though there is a know bug,
> > we'll know if it's triggered.
>
> Anyways, it should be fine enough, since it's very rare, and we can always turn
> off CSE.
Well, there's nothing about CSE that really triggered it, so I think this
is a little safer.