[llvm] [AArch64] Add partial reduce patterns for new fdot instructions (PR #184659)

[Paul Walker via llvm-commits]

================
@@ -2031,6 +2038,8 @@ AArch64TargetLowering::AArch64TargetLowering(const TargetMachine &TM,
       // We can use SVE2p1 fdot to emulate the fixed-length variant.
       setPartialReduceMLAAction(ISD::PARTIAL_REDUCE_FMLA, MVT::v4f32,
                                 MVT::v8f16, Custom);
+      setPartialReduceMLAAction(ISD::PARTIAL_REDUCE_FMLA, MVT::v2f32,
+                                MVT::v4f16, Custom);
----------------
paulwalker-arm wrote:

I'm not beholden to this designed decision so if the AArch64 maintainers don't like it then it's your call to make. I only ask that you weigh up the value of reversing direction given we've survived this long with minimal impact.

https://github.com/llvm/llvm-project/pull/184659