[InstCombine] Remove scalable vector restriction in InstCombineCasts

[junparser]

Summary

As is the title.

TestPlan: check-llvm

[junparser]

Remove the change that do not have testcases

[spatel]

Code changes look fairly mechanical at this point, so if the output is as expected, LGTM

llvm/test/Transforms/InstCombine/ptr-int-cast.ll

1

Could update this file using utils/update_test_checks.py as a preliminary commit?

[junparser]

Thanks for review the patch !

llvm/test/Transforms/InstCombine/ptr-int-cast.ll

1

I'll udpate this when commit this patch.

llvm/lib/IR/Verifier.cpp

void Verifier::visitAddrSpaceCastInst(AddrSpaceCastInst &I) {
 
   Assert(SrcTy->isPtrOrPtrVectorTy(), "AddrSpaceCast source must be a pointer",
          &I);
   Assert(DestTy->isPtrOrPtrVectorTy(), "AddrSpaceCast result must be a pointer",
          &I);
   Assert(SrcTy->getPointerAddressSpace() != DestTy->getPointerAddressSpace(),
          "AddrSpaceCast must be between different address spaces", &I);
   if (auto *SrcVTy = dyn_cast<VectorType>(SrcTy))
-    Assert(cast<FixedVectorType>(SrcVTy)->getNumElements() ==
-               cast<FixedVectorType>(DestTy)->getNumElements(),
+    Assert(SrcVTy->getElementCount() ==
+               cast<VectorType>(DestTy)->getElementCount(),
            "AddrSpaceCast vector pointer number of elements mismatch", &I);
   visitInstruction(I);
 }
 
 /// visitPHINode - Ensure that a PHI node is well formed.
 ///
 void Verifier::visitPHINode(PHINode &PN) {
   // Ensure that the PHI nodes are all grouped together at the top of the block.

llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp

Instruction *InstCombinerImpl::visitTrunc(TruncInst &Trunc) {
   //
   // Example (little endian):
   //   trunc (extractelement <4 x i64> %X, 0) to i32
   //   --->
   //   extractelement <8 x i32> (bitcast <4 x i64> %X to <8 x i32>), i32 0
   Value *VecOp;
   ConstantInt *Cst;
   if (match(Src, m_OneUse(m_ExtractElt(m_Value(VecOp), m_ConstantInt(Cst))))) {
-    auto *VecOpTy = cast<FixedVectorType>(VecOp->getType());
-    unsigned VecNumElts = VecOpTy->getNumElements();
+    auto *VecOpTy = cast<VectorType>(VecOp->getType());
+    auto VecElts = VecOpTy->getElementCount();
 
     // A badly fit destination size would result in an invalid cast.
     if (SrcWidth % DestWidth == 0) {
       uint64_t TruncRatio = SrcWidth / DestWidth;
-      uint64_t BitCastNumElts = VecNumElts * TruncRatio;
+      uint64_t BitCastNumElts = VecElts.getKnownMinValue() * TruncRatio;
       uint64_t VecOpIdx = Cst->getZExtValue();
       uint64_t NewIdx = DL.isBigEndian() ? (VecOpIdx + 1) * TruncRatio - 1
                                          : VecOpIdx * TruncRatio;
       assert(BitCastNumElts <= std::numeric_limits<uint32_t>::max() &&
              "overflow 32-bits");
 
-      auto *BitCastTo = FixedVectorType::get(DestTy, BitCastNumElts);
+      auto *BitCastTo =
+          VectorType::get(DestTy, BitCastNumElts, VecElts.isScalable());
       Value *BitCast = Builder.CreateBitCast(VecOp, BitCastTo);
       return ExtractElementInst::Create(BitCast, Builder.getInt32(NewIdx));
     }
   }
 
   return nullptr;
 }
 
static Type *shrinkFPConstant(ConstantFP *CFP) {
   return nullptr;
 }
 
 // Determine if this is a vector of ConstantFPs and if so, return the minimal
 // type we can safely truncate all elements to.
 // TODO: Make these support undef elements.
 static Type *shrinkFPConstantVector(Value *V) {
   auto *CV = dyn_cast<Constant>(V);
-  auto *CVVTy = dyn_cast<VectorType>(V->getType());
+  auto *CVVTy = dyn_cast<FixedVectorType>(V->getType());
   if (!CV || !CVVTy)
     return nullptr;
 
   Type *MinType = nullptr;
 
-  unsigned NumElts = cast<FixedVectorType>(CVVTy)->getNumElements();
+  unsigned NumElts = CVVTy->getNumElements();
   for (unsigned i = 0; i != NumElts; ++i) {
     auto *CFP = dyn_cast_or_null<ConstantFP>(CV->getAggregateElement(i));
     if (!CFP)
       return nullptr;
 
     Type *T = shrinkFPConstant(CFP);
     if (!T)
       return nullptr;
Instruction *InstCombinerImpl::visitPtrToInt(PtrToIntInst &CI) {
   // to be exposed to other transforms.
   Value *SrcOp = CI.getPointerOperand();
   Type *Ty = CI.getType();
   unsigned AS = CI.getPointerAddressSpace();
   unsigned TySize = Ty->getScalarSizeInBits();
   unsigned PtrSize = DL.getPointerSizeInBits(AS);
   if (TySize != PtrSize) {
     Type *IntPtrTy = DL.getIntPtrType(CI.getContext(), AS);
-    if (auto *VecTy = dyn_cast<VectorType>(Ty)) {
-      // Handle vectors of pointers.
-      // FIXME: what should happen for scalable vectors?
-      IntPtrTy = FixedVectorType::get(
-          IntPtrTy, cast<FixedVectorType>(VecTy)->getNumElements());
-    }
+    if (auto *VecTy = dyn_cast<VectorType>(Ty))
+      IntPtrTy = VectorType::get(IntPtrTy, VecTy->getElementCount());
 
     Value *P = Builder.CreatePtrToInt(SrcOp, IntPtrTy);
     return CastInst::CreateIntegerCast(P, Ty, /*isSigned=*/false);
   }
 
   Value *Vec, *Scalar, *Index;
   if (match(SrcOp, m_OneUse(m_InsertElt(m_IntToPtr(m_Value(Vec)),
                                         m_Value(Scalar), m_Value(Index)))) &&

 /// The source and destination vector types may have different element types.
 static Instruction *
 optimizeVectorResizeWithIntegerBitCasts(Value *InVal, VectorType *DestTy,
                                         InstCombinerImpl &IC) {
   // We can only do this optimization if the output is a multiple of the input
   // element size, or the input is a multiple of the output element size.
   // Convert the input type to have the same element type as the output.
   VectorType *SrcTy = cast<VectorType>(InVal->getType());
+  if (!isa<FixedVectorType>(SrcTy) || !isa<FixedVectorType>(DestTy))
+    return nullptr;
 
   if (SrcTy->getElementType() != DestTy->getElementType()) {
     // The input types don't need to be identical, but for now they must be the
     // same size.  There is no specific reason we couldn't handle things like
     // <4 x i16> -> <4 x i32> by bitcasting to <2 x i32> but haven't gotten
     // there yet.
     if (SrcTy->getElementType()->getPrimitiveSizeInBits() !=
         DestTy->getElementType()->getPrimitiveSizeInBits())
if (FixedVectorType *SrcVTy = dyn_cast<FixedVectorType>(SrcTy)) {
     }
   }
 
   if (auto *Shuf = dyn_cast<ShuffleVectorInst>(Src)) {
     // Okay, we have (bitcast (shuffle ..)).  Check to see if this is
     // a bitcast to a vector with the same # elts.
     Value *ShufOp0 = Shuf->getOperand(0);
     Value *ShufOp1 = Shuf->getOperand(1);
-    unsigned NumShufElts =
-        cast<FixedVectorType>(Shuf->getType())->getNumElements();
-    unsigned NumSrcVecElts =
-        cast<FixedVectorType>(ShufOp0->getType())->getNumElements();
+    auto ShufElts = cast<VectorType>(Shuf->getType())->getElementCount();
+    auto SrcVecElts = cast<VectorType>(ShufOp0->getType())->getElementCount();
     if (Shuf->hasOneUse() && DestTy->isVectorTy() &&
-        cast<FixedVectorType>(DestTy)->getNumElements() == NumShufElts &&
-        NumShufElts == NumSrcVecElts) {
+        cast<VectorType>(DestTy)->getElementCount() == ShufElts &&
+        ShufElts == SrcVecElts) {
       BitCastInst *Tmp;
       // If either of the operands is a cast from CI.getType(), then
       // evaluating the shuffle in the casted destination's type will allow
       // us to eliminate at least one cast.
       if (((Tmp = dyn_cast<BitCastInst>(ShufOp0)) &&
            Tmp->getOperand(0)->getType() == DestTy) ||
           ((Tmp = dyn_cast<BitCastInst>(ShufOp1)) &&
            Tmp->getOperand(0)->getType() == DestTy)) {
         Value *LHS = Builder.CreateBitCast(ShufOp0, DestTy);
         Value *RHS = Builder.CreateBitCast(ShufOp1, DestTy);
         // Return a new shuffle vector.  Use the same element ID's, as we
         // know the vector types match #elts.
         return new ShuffleVectorInst(LHS, RHS, Shuf->getShuffleMask());
       }
     }
 
     // A bitcasted-to-scalar and byte-reversing shuffle is better recognized as
     // a byte-swap:
     // bitcast <N x i8> (shuf X, undef, <N, N-1,...0>) --> bswap (bitcast X)
     // TODO: We should match the related pattern for bitreverse.
     if (DestTy->isIntegerTy() &&
         DL.isLegalInteger(DestTy->getScalarSizeInBits()) &&
-        SrcTy->getScalarSizeInBits() == 8 && NumShufElts % 2 == 0 &&
-        Shuf->hasOneUse() && Shuf->isReverse()) {
+        SrcTy->getScalarSizeInBits() == 8 &&
+        ShufElts.getKnownMinValue() % 2 == 0 && Shuf->hasOneUse() &&
+        Shuf->isReverse()) {
       assert(ShufOp0->getType() == SrcTy && "Unexpected shuffle mask");
       assert(isa<UndefValue>(ShufOp1) && "Unexpected shuffle op");
       Function *Bswap =
           Intrinsic::getDeclaration(CI.getModule(), Intrinsic::bswap, DestTy);
       Value *ScalarX = Builder.CreateBitCast(ShufOp0, DestTy);
       return IntrinsicInst::Create(Bswap, { ScalarX });
     }
   }
Instruction *InstCombinerImpl::visitAddrSpaceCast(AddrSpaceCastInst &CI) {
   // This allows the cast to be exposed to other transforms.
   Value *Src = CI.getOperand(0);
   PointerType *SrcTy = cast<PointerType>(Src->getType()->getScalarType());
   PointerType *DestTy = cast<PointerType>(CI.getType()->getScalarType());
 
   Type *DestElemTy = DestTy->getElementType();
   if (SrcTy->getElementType() != DestElemTy) {
     Type *MidTy = PointerType::get(DestElemTy, SrcTy->getAddressSpace());
-    if (VectorType *VT = dyn_cast<VectorType>(CI.getType())) {
-      // Handle vectors of pointers.
-      // FIXME: what should happen for scalable vectors?
-      MidTy = FixedVectorType::get(MidTy,
-                                   cast<FixedVectorType>(VT)->getNumElements());
-    }
+    if (VectorType *VT = dyn_cast<VectorType>(CI.getType()))
+      MidTy = VectorType::get(MidTy, VT->getElementCount());
 
     Value *NewBitCast = Builder.CreateBitCast(Src, MidTy);
     return new AddrSpaceCastInst(NewBitCast, CI.getType());
   }
 
   return commonPointerCastTransforms(CI);
 }

llvm/test/Transforms/InstCombine/addrspacecast.ll

 ; CHECK-LABEL: @combine_addrspacecast_types_vector(
 ; CHECK-NEXT: bitcast <4 x i32 addrspace(1)*> %x to <4 x float addrspace(1)*>
 ; CHECK-NEXT: addrspacecast <4 x float addrspace(1)*> %1 to <4 x float addrspace(2)*>
 ; CHECK-NEXT: ret
   %y = addrspacecast <4 x i32 addrspace(1)*> %x to <4 x float addrspace(2)*>
   ret <4 x float addrspace(2)*> %y
 }
 
+define <vscale x 4 x float addrspace(2)*> @combine_addrspacecast_types_scalevector(<vscale x 4 x i32 addrspace(1)*> %x) nounwind {
+; CHECK-LABEL: @combine_addrspacecast_types_scalevector(
+; CHECK-NEXT: bitcast <vscale x 4 x i32 addrspace(1)*> %x to <vscale x 4 x float addrspace(1)*>
+; CHECK-NEXT: addrspacecast <vscale x 4 x float addrspace(1)*> %1 to <vscale x 4 x float addrspace(2)*>
+; CHECK-NEXT: ret
+  %y = addrspacecast <vscale x 4 x i32 addrspace(1)*> %x to <vscale x 4 x float addrspace(2)*>
+  ret <vscale x 4 x float addrspace(2)*> %y
+}
+
+
 define i32 @canonicalize_addrspacecast([16 x i32] addrspace(1)* %arr) {
 ; CHECK-LABEL: @canonicalize_addrspacecast(
 ; CHECK-NEXT: getelementptr [16 x i32], [16 x i32] addrspace(1)* %arr, i32 0, i32 0
 ; CHECK-NEXT: addrspacecast i32 addrspace(1)* %{{[a-zA-Z0-9]+}} to i32*
 ; CHECK-NEXT: load i32, i32*
 ; CHECK-NEXT: ret i32
   %p = addrspacecast [16 x i32] addrspace(1)* %arr to i32*
   %v = load i32, i32* %p

llvm/test/Transforms/InstCombine/ptr-int-cast.ll

 ; RUN: opt < %s -instcombine -S | FileCheck %s
 target datalayout = "E-p:64:64:64-a0:0:8-f32:32:32-f64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-v64:64:64-v128:128:128"
 
 define i1 @test1(i32 *%x) nounwind {
 entry:
 ; CHECK: test1
 ; CHECK: ptrtoint i32* %x to i64
   %0 = ptrtoint i32* %x to i1
   ret i1 %0

 define <4 x i32> @test4(<4 x i8*> %arg) nounwind {
 ; CHECK-LABEL: @test4(
 ; CHECK: ptrtoint <4 x i8*> %arg to <4 x i64>
 ; CHECK: trunc <4 x i64> %1 to <4 x i32>
   %p1 = ptrtoint <4 x i8*> %arg to <4 x i32>
   ret <4 x i32> %p1
 }
 
+define <vscale x 4 x i32> @testvscale4(<vscale x 4 x i8*> %arg) nounwind {
+; CHECK-LABEL: @testvscale4(
+; CHECK: ptrtoint <vscale x 4 x i8*> %arg to <vscale x 4 x i64>
+; CHECK: trunc <vscale x 4 x i64> %1 to <vscale x 4 x i32>
+  %p1 = ptrtoint <vscale x 4 x i8*> %arg to <vscale x 4 x i32>
+  ret <vscale x 4 x i32> %p1
+}
+
 define <4 x i128> @test5(<4 x i8*> %arg) nounwind {
 ; CHECK-LABEL: @test5(
 ; CHECK: ptrtoint <4 x i8*> %arg to <4 x i64>
 ; CHECK: zext <4 x i64> %1 to <4 x i128>
   %p1 = ptrtoint <4 x i8*> %arg to <4 x i128>
   ret <4 x i128> %p1
 }
 

llvm/test/Transforms/InstCombine/trunc-extractelement.ll

 ; BE-NEXT:    [[T:%.*]] = extractelement <6 x i32> [[TMP1]], i32 1
 ; BE-NEXT:    ret i32 [[T]]
 ;
   %e = extractelement <3 x i64> %x, i32 0
   %t = trunc i64 %e to i32
   ret i32 %t
 }
 
+define i32 @vscale_shrinkExtractElt_i64_to_i32_0(<vscale x 3 x i64> %x) {
+; LE-LABEL: @vscale_shrinkExtractElt_i64_to_i32_0(
+; LE-NEXT:    [[TMP1:%.*]] = bitcast <vscale x 3 x i64> [[X:%.*]] to <vscale x 6 x i32>
+; LE-NEXT:    [[T:%.*]] = extractelement <vscale x 6 x i32> [[TMP1]], i32 0
+; LE-NEXT:    ret i32 [[T]]
+;
+; BE-LABEL: @vscale_shrinkExtractElt_i64_to_i32_0(
+; BE-NEXT:    [[TMP1:%.*]] = bitcast <vscale x 3 x i64> [[X:%.*]] to <vscale x 6 x i32>
+; BE-NEXT:    [[T:%.*]] = extractelement <vscale x 6 x i32> [[TMP1]], i32 1
+; BE-NEXT:    ret i32 [[T]]
+;
+  %e = extractelement <vscale x 3 x i64> %x, i32 0
+  %t = trunc i64 %e to i32
+  ret i32 %t
+}
+
+
 define i32 @shrinkExtractElt_i64_to_i32_1(<3 x i64> %x) {
 ; LE-LABEL: @shrinkExtractElt_i64_to_i32_1(
 ; LE-NEXT:    [[TMP1:%.*]] = bitcast <3 x i64> [[X:%.*]] to <6 x i32>
 ; LE-NEXT:    [[T:%.*]] = extractelement <6 x i32> [[TMP1]], i32 2
 ; LE-NEXT:    ret i32 [[T]]
 ;
 ; BE-LABEL: @shrinkExtractElt_i64_to_i32_1(
 ; BE-NEXT:    [[TMP1:%.*]] = bitcast <3 x i64> [[X:%.*]] to <6 x i32>

llvm/test/Transforms/InstCombine/vec_shuffle.ll

 ;
   %X1 = bitcast <4 x float> %X to <4 x i32>
   %t = shufflevector <4 x i32> %X1, <4 x i32> undef, <4 x i32> zeroinitializer
   %t2 = bitcast <4 x i32> %t to <4 x float>
   %r = extractelement <4 x float> %t2, i32 0
   ret float %r
 }
 
+define float @testvscale6(<vscale x 4 x float> %X) {
+; CHECK-LABEL: @testvscale6(
+; CHECK-NEXT:    [[T:%.*]] = shufflevector <vscale x 4 x float> [[X:%.*]], <vscale x 4 x float> undef, <vscale x 4 x i32> zeroinitializer
+; CHECK-NEXT:    [[R:%.*]] = extractelement <vscale x 4 x float> [[T]], i32 0
+; CHECK-NEXT:    ret float [[R]]
+;
+  %X1 = bitcast <vscale x 4 x float> %X to <vscale x 4 x i32>
+  %t = shufflevector <vscale x 4 x i32> %X1, <vscale x 4 x i32> undef, <vscale x 4 x i32> zeroinitializer
+  %t2 = bitcast <vscale x 4 x i32> %t to <vscale x 4 x float>
+  %r = extractelement <vscale x 4 x float> %t2, i32 0
+  ret float %r
+}
+
+
 define <4 x float> @test7(<4 x float> %x) {
 ; CHECK-LABEL: @test7(
 ; CHECK-NEXT:    [[R:%.*]] = shufflevector <4 x float> [[X:%.*]], <4 x float> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
 ; CHECK-NEXT:    ret <4 x float> [[R]]
 ;
   %r = shufflevector <4 x float> %x, <4 x float> undef, <4 x i32> < i32 0, i32 1, i32 6, i32 7 >
   ret <4 x float> %r
 }