[LLVMdev] alias information in codegen

[Pertti Kellomäki]

Daniel Berlin wrote:
> Trying to do pointer analysis on very lowered forms is both expensive
> and hard.

This is actually something that has been bothering me. There
is a lot of existing work on parallelizing Fortran compilers,
and the dependence analysis there focuses on figuring out
whether two references to the same array alias or not. The
analysis methods basically solve systems of equations formed
from the index expressions in the array references.

It would seem that this kind of analysis should be done
way before converting to anything like the LLVM IR. Are there
any plans or ideas as to how such alias information could be
propagated e.g. from clang to LLVM?
-- 
Pertti