[llvm] [RISCV] Migrate zvqdotq reduce matching to use partial_reduce infrastructure (PR #142212)

[Philip Reames via llvm-commits]

================
@@ -18455,61 +18430,62 @@ static SDValue foldReduceOperandViaVQDOT(SDValue InVec, const SDLoc &DL,
     }
   }
 
-  // reduce (zext a) <--> reduce (mul zext a. zext 1)
-  // reduce (sext a) <--> reduce (mul sext a. sext 1)
+  // reduce (zext a) <--> partial_reduce_umla 0, a, 1
----------------
preames wrote:

Actually, staring as it, not sure how to really make the comment more clear.  The reduce here is the outermost reduce - we might be well into an interior add in the reduce tree.  Any suggestions?

https://github.com/llvm/llvm-project/pull/142212