[PATCH] D53268: [X86] WIP: Stop promoting and/or/xor/andn to vXi64.

[Craig Topper via Phabricator via llvm-commits]

craig.topper added inline comments.


================
Comment at: lib/Target/X86/X86InstrAVX512.td:11437
+              (i8 15)), sub_xmm)>;
+  def : Pat<(xor VR128X:$src, (bc_v2i64 (v4i32 immAllOnesV))),
             (EXTRACT_SUBREG
----------------
nhaehnle wrote:
> This and many other cases of duplication in the .td files seem amenable to using foreach. Something like:
> ```
>   foreach bc_ty = [bc_v16i8, bc_v8i16, bc_v4i32, bc_v2i64]<PatFrag> in
>   def : Pat<(xor VR128X:$src, (bc_ty (v4i32 immAllOnesV))),
>             (EXTRACT_SUBREG
>              (VPTERNLOGQZrri
>               (INSERT_SUBREG (v8i64 (IMPLICIT_DEF)), VR128X:$src, sub_xmm),
>               (INSERT_SUBREG (v8i64 (IMPLICIT_DEF)), VR128X:$src, sub_xmm),
>               (INSERT_SUBREG (v8i64 (IMPLICIT_DEF)), VR128X:$src, sub_xmm),
>               (i8 15)), sub_xmm)>;
> ```
I agree we need to reduce the repetitiveness. My immediate focus is on fixing the test regressions. Then I want to see if should do anything to do custom selection in X86ISelDAGToDAG.cpp to avoid growing the static isel table.


https://reviews.llvm.org/D53268