[llvm-commits] [llvm] r152590 - /llvm/trunk/lib/Analysis/InstructionSimplify.cpp

[Duncan Sands]

Hi Chandler,

> Address some review comments from Duncan. This moves the iterative
> offset accumulation to use a boring APInt instead of ConstantExprs.

I know I suggested this, but in fact is there any reason to only accept
ConstantInt GEP indices?  If a GEP index is a ConstantExpr that should be
fine for your purposes since the pointer difference still becomes a constant.

I also have gripes about the computation, both for the do-it-with-constants
version and this version, but it seems pointless to worry about that without
first deciding about this ConstantInt indices vs Constant indices point.

Ciao, Duncan.

PS: Do you plan to do the "trunc" generalization I suggested?