[PATCH] D153323: [AArch64] Try to fold uaddlv and uaddlp

Tue Jun 20 04:11:23 PDT 2023

jaykang10 added inline comments.

================
Comment at: llvm/lib/Target/AArch64/AArch64InstrInfo.td:6331
+          (i64 (EXTRACT_SUBREG
+            (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), (UADDLVv8i16v V128:$op), ssub),
+            dsub))>;
----------------
dmgreen wrote:
> It is probably quite a minor point, but can you change this to a `(v4i32 (SUBREG_TO_REG (i64 0), (UADDLVv8i16v V128:$op), ssub))`. The EXTRACT_SUBREG is using the fact that the higher lanes will be implicitly zeroed.
Let me update the pattern.

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D153323/new/

https://reviews.llvm.org/D153323