[llvm] [AArch64][CostModel] Add constraints on which partial reductions are (PR #163728)

[Sander de Smalen via llvm-commits]

================
@@ -5721,6 +5721,38 @@ InstructionCost AArch64TTIImpl::getPartialReductionCost(
       return Cost;
   }
 
+  // FIXME:
+  // 1. Do cost modelling for USDOT.
+  // 2. Refactor the whole code here.
+  if (ST->isSVEorStreamingSVEAvailable() && !IsUSDot) {
+    if (AccumLT.second.getScalarType() == MVT::i32 &&
+        InputLT.second.getScalarType() == MVT::i16) {
+      // i16 -> i32 is supported in SVE 2.1
+      if (ST->hasSVE2p1())
+        return Cost;
+      // umlalt + umlalb. Same goes for signed types.
+      return Cost + 1;
+    }
+    if (AccumLT.second.getScalarType() == MVT::i64 &&
+        InputLT.second.getScalarType() == MVT::i32)
+      return Cost + 1;
+  }
+  if (AccumLT.second.isFixedLengthVector() && ST->isNeonAvailable() &&
+      ST->hasDotProd() && !IsUSDot) {
+    // umull + umull2 + (2 * uaddw) + (2 * uaddw2). Same goes for signed types.
+    if (AccumLT.second.getScalarType() == MVT::i64 &&
+        InputLT.second.getScalarType() == MVT::i16)
+      return Cost + 5;
+    // umlal + umlal2. Same goes for signed types.
+    if ((AccumLT.second.getScalarType() == MVT::i32 &&
+         InputLT.second.getScalarType() == MVT::i16) ||
+        (AccumLT.second.getScalarType() == MVT::i64 &&
+         InputLT.second.getScalarType() == MVT::i32))
+      return Cost + 1;
+  }
+
+  // FIXME: This should be more expensive for NEON as we see fmov instructions
+  // with very low throughput.
   // Add additional cost for the extends that would need to be inserted.
   return Cost + 4;
----------------
sdesmalen-arm wrote:

My suggestion was merely to change the `return Cost + 4;` at the bottom of this function to `return Cost + 2;`.

https://github.com/llvm/llvm-project/pull/163728