Add new intrinsics for Neon VTRN, VZIP and VUZP operations. Modeling these
authorBob Wilson <bob.wilson@apple.com>
Fri, 7 Aug 2009 23:53:05 +0000 (23:53 +0000)
committerBob Wilson <bob.wilson@apple.com>
Fri, 7 Aug 2009 23:53:05 +0000 (23:53 +0000)
commit9e699cc72639f25b20e73f51ab06ab7a0884ea62
tree8e0082e7544b47805f1f4cad5ba444dd66308f26
parentc3c9239d76e2acb01dac8c72323a0b28fadebe4b
Add new intrinsics for Neon VTRN, VZIP and VUZP operations.  Modeling these
as vector shuffles did not work out well.  Shuffles that produce double-wide
vectors accurately represent the operation but make it hard to do anything
with the results.  I considered splitting them up into 2 shuffles, one to
write each register separately, but there doesn't seem to be a good way to
reunite them for codegen.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78437 91177308-0d34-0410-b5e6-96231b3b80d8
include/llvm/IntrinsicsARM.td