[PATCH] D48323: Derive GEP index type from Data Layout (cont)

[Elena Demikhovsky via Phabricator via llvm-commits]

delena added inline comments.


================
Comment at: lib/Transforms/InstCombine/InstCombineCasts.cpp:1735
 
     Value *P = Builder.CreateZExtOrTrunc(CI.getOperand(0), Ty);
     return new IntToPtrInst(P, CI.getType());
----------------
lebedev.ri wrote:
> I realize that it is existing behaviour, but this creates new extra instruction,
> and does not check for uses of previous instruction.
> This is not good for instcombine, although i suppose this is a very special case.
> 
> You //might// want to check that there are tests that verify this behavior, or add such tests.
It shouldn't create an extra instruction. I fixed incompatibility between  IntPtrType and expected type of the source.
The current version just entered to an endless loop, because the  IntPtrType  is the index type and not the pointer type.


================
Comment at: test/Transforms/InstCombine/ptr-int-cast_custom_gep.ll:1
+; RUN: opt < %s -instcombine -S | FileCheck %s
+target datalayout = "E-p:40:64:64:32-a0:0:8-f32:32:32-f64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-v64:64:64-v128:128:128"
----------------
lebedev.ri wrote:
> Instcombine tests use `utils/update_test_checks.py`.
> (And ideally (at least when committing) the diff should be a diff to the previous results)
I can generate. I just took the original code and changed it.


Repository:
  rL LLVM

https://reviews.llvm.org/D48323