[llvm-commits] [llvm] r155809 - in /llvm/trunk: lib/Transforms/Scalar/LoopStrengthReduce.cpp test/Transforms/LoopStrengthReduce/pr12691.ll

[Eli Friedman]

On Thu, May 3, 2012 at 11:33 AM, Bill Wendling <isanbard at gmail.com> wrote:
> Argh! I just kicked off rc2 testing last night. I suppose I can sneak this in. What is the risk of this patch and how often will this problem be hit?

This only affects vector code which uses vector registers of multiple
widths in the same function, where the backend doesn't have custom
concatenation code.  I don't think the difference is visible on x86 or
ARM (because we custom-lower the operation in question), but this
likely impacts other architectures with multiple vector register
widths.

-Eli