[llvm] [AArch64] Lower partial add reduction to udot or svdot (PR #101010)

[Sander de Smalen via llvm-commits]

================
@@ -1971,6 +1971,57 @@ bool AArch64TargetLowering::shouldExpandGetActiveLaneMask(EVT ResVT,
   return false;
 }
 
+bool AArch64TargetLowering::shouldExpandPartialReductionIntrinsic(
+    const CallInst *CI) const {
+  const bool TargetLowers = false;
+  const bool GenericLowers = true;
+
+  auto *I = dyn_cast<IntrinsicInst>(CI);
+  if (!I)
+    return GenericLowers;
+
+  ScalableVectorType *RetTy = dyn_cast<ScalableVectorType>(I->getType());
----------------
sdesmalen-arm wrote:

Am I right that this is just an artificial limitation at the moment? (i.e. we can make this work for fixed-length vectors too)

https://github.com/llvm/llvm-project/pull/101010