[PATCH] Disable indvar widening if arithmetics on the wider type are more expensive

[Jingyue Wu]

Hi atrick, jholewinski, eliben, meheff,

We should not widen an indvar if arithmetics on the wider indvar are more
expensive than those on the narrower indvar. For instance, although
NVPTX64 treats i64 as a legal type, an ADD on i64 is twice as expensive
as that on i32, because the hardware needs to simulate a 64-bit integer
using two 32-bit integers.

This patch
1. adds a NVPTX-specific TargetTransformInfo with specialized computing
of cost of arithmetic instructions;
2. compares the cost of arithmetics on the wider type and that on the
narrower type before widening an indvar.

Fixes PR21148.

http://reviews.llvm.org/D6188

Files:
  lib/Target/NVPTX/CMakeLists.txt
  lib/Target/NVPTX/NVPTX.h
  lib/Target/NVPTX/NVPTXTargetMachine.cpp
  lib/Target/NVPTX/NVPTXTargetMachine.h
  lib/Target/NVPTX/NVPTXTargetTransformInfo.cpp
  lib/Transforms/Scalar/IndVarSimplify.cpp
  test/Transforms/IndVarSimplify/no-widen-expensive.ll
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D6188.15965.patch
Type: text/x-patch
Size: 13123 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20141110/950018d2/attachment.bin>