[llvm] [RISCV][VPlan] Cost vp.merges that are likely to be folded away as free (PR #209387)

[Luke Lau via llvm-commits]

================
@@ -2319,11 +2319,19 @@ InstructionCost VPWidenIntrinsicRecipe::computeCallCost(
         return toVectorTy(Op->getScalarType(), VF);
       });
 
+  VectorInstrContext VIC = VectorInstrContext::None;
+  for (const VPValue *Op : Operands)
+    if (isa<VPWidenRecipe>(Op) &&
+        Instruction::isBinaryOp(cast<VPWidenRecipe>(Op)->getOpcode())) {
+      VIC = VectorInstrContext::BinaryOp;
+      break;
+    }
+
   // TODO: Rework TTI interface to avoid reliance on underlying IntrinsicInst.
   IntrinsicCostAttributes CostAttrs(
       ID, RetTy, Arguments, ParamTys, R.getFastMathFlagsOrNone(),
----------------
lukel97 wrote:

Yeah. As we do more and more transformations in VPlan we end up with more recipes that don't exist in the original LLVM IR, and only exist at the VPlan level. 

It's also a slightly misleading API because we're querying the cost for a vector type, but we pass in scalar values. We can't pass in values of the actual vector type because they're not materialised yet. 

https://github.com/llvm/llvm-project/pull/209387