On Dec 1, 2008, at 5:28 PM, Daniel Dunbar wrote:

> This sounds like the kind of test case I would write. :)
>
> At the time this was my attempt to get a test case which had some
> stronger connection to correctness than just "IRgen didn't crash/give
> an error", but perhaps the right thing to do is just make an
> executable test case out of it and move it into llvm-test? OTOH, it is
> also nice to know that the optimizer does end up cleaning up these
> things for the kind of code we generate.
>
> Chris, do you have an opinion on whether using the optimizer to
> canonicalize test cases is good or bad?

This is a QOI issue.  For that specific case, I think it's better to  
improve Clang's codegen than it is to fix the optimizer to handle that  
one specific case.  I'm sure that there are lots of other cases that  
are pessimized by Clang's current approach as well.

-Chris

_______________________________________________
cfe-commits mailing list
[email protected]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits

Reply via email to