On Jan 2, 2008, at 11:56 AM, Christopher Lamb wrote: > For embedded devices with special address spaces often 0x0 is a > valid address in those spaces.
Ok, fair enough. > I knew this was going to be a troublesome issue as it goes against > a basic assumption for single address space compilation. I know > that this assumption is salted away in xforms scattered throughout > the compiler, and I'm resigned to addressing them as they become > evident. Ok. > I would appreciate some advice on how to provide target specific > information to analysis and transforms (alias analysis could also > benefit from target specific address space information). Is there > an example where this is already being done? There isn't a really good way to do this, as we don't want target- independent xforms (like instcombine) being overly parameterized by the target. I think your solution of disabling this for non-default pointers is a good way to go. Thanks! -Chris _______________________________________________ llvm-commits mailing list llvm-commits@cs.uiuc.edu http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
