[X86] Update avx512vbmi2 intrinsic tests to avoid adds
authorLuo, Yuanke <yuanke.luo@intel.com>
Tue, 8 Mar 2022 09:46:54 +0000 (17:46 +0800)
committerLuo, Yuanke <yuanke.luo@intel.com>
Tue, 8 Mar 2022 09:49:44 +0000 (17:49 +0800)
commit5494769e323a73a559143a518bca5fee5fbed41b
tree24664a203759afb9db5f3629b18c87e793b6f775
parentf831d6fc800ccf22c1c09888fce3e3c8ebc2c992
[X86] Update avx512vbmi2 intrinsic tests to avoid adds

As noticed in D119654, by adding the masked intrinsics results together
we can end up with the selects being canonicalized away from the
intrinsic - this isn't what we want to test here so replace with a
insertvalue chain into a aggregate instead to retain all the results.

Differential Revision: https://reviews.llvm.org/D121196
llvm/test/CodeGen/X86/avx512vbmi2-intrinsics.ll