Include shuffle arguments to improve cost analysis; remove undef tests

Author: kuroyukiasuna

llvm/lib/Transforms/Vectorize/VectorCombine.cpp

@@ -2999,11 +2999,11 @@ bool VectorCombine::foldPermuteOfIntrinsic(Instruction &I) {
   InstructionCost OldCost =
       TTI.getIntrinsicInstrCost(IntrinsicCostAttributes(IID, *II0), CostKind) +
       TTI.getShuffleCost(TargetTransformInfo::SK_PermuteSingleSrc, ShuffleDstTy,
-                         IntrinsicSrcTy, Mask, CostKind);
+                         IntrinsicSrcTy, Mask, CostKind, 0, nullptr, {V0, V1},
+                         &I);
 
   SmallVector<Type *> NewArgsTy;
   InstructionCost NewCost = 0;
-
   for (unsigned I = 0, E = II0->arg_size(); I != E; ++I) {
     if (isVectorIntrinsicWithScalarOpAtArg(IID, I, &TTI)) {
       NewArgsTy.push_back(II0->getArgOperand(I)->getType());
@@ -3013,7 +3013,8 @@ bool VectorCombine::foldPermuteOfIntrinsic(Instruction &I) {
                                          ShuffleDstTy->getNumElements());
       NewArgsTy.push_back(ArgTy);
       NewCost += TTI.getShuffleCost(TargetTransformInfo::SK_PermuteSingleSrc,
-                                    ArgTy, VecTy, Mask, CostKind);
+                                    ArgTy, VecTy, Mask, CostKind, 0, nullptr,
+                                    {II0->getArgOperand(I)});
     }
   }
   IntrinsicCostAttributes NewAttr(IID, ShuffleDstTy, NewArgsTy);

llvm/test/Transforms/VectorCombine/AArch64/shuffle-of-intrinsic-permute.ll

@@ -2,7 +2,7 @@
 ; RUN: opt -passes=vector-combine -S -mtriple=aarch64 %s | FileCheck %s
 
 ; This file tests the foldPermuteOfIntrinsic optimization which transforms:
-;   shuffle(intrinsic(args), poison/undef) -> intrinsic(shuffle(args))
+;   shuffle(intrinsic(args), poison) -> intrinsic(shuffle(args))
 ; when the shuffle is a permute (operates on single vector) and cost model
 ; determines the transformation is beneficial.
 
@@ -82,18 +82,6 @@ define <8 x i16> @extract_lower_i16(<16 x i16> %v1, <16 x i16> %v2) {
   ret <8 x i16> %result
 }
 
-define <4 x i32> @extract_lower_with_undef(<8 x i32> %v1, <8 x i32> %v2) {
-; CHECK-LABEL: @extract_lower_with_undef(
-; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <8 x i32> [[V1:%.*]], <8 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
-; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <8 x i32> [[V2:%.*]], <8 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
-; CHECK-NEXT:    [[RESULT:%.*]] = call <4 x i32> @llvm.sadd.sat.v4i32(<4 x i32> [[TMP1]], <4 x i32> [[TMP2]])
-; CHECK-NEXT:    ret <4 x i32> [[RESULT]]
-;
-  %sat = call <8 x i32> @llvm.sadd.sat.v8i32(<8 x i32> %v1, <8 x i32> %v2)
-  %result = shufflevector <8 x i32> %sat, <8 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
-  ret <4 x i32> %result
-}
-
 ;; ============================================================================
 ;; Negative Tests - Should NOT Optimize
 ;; ============================================================================