Index: llvm/lib/Transforms/Vectorize/LoopVectorize.cpp =================================================================== --- llvm/lib/Transforms/Vectorize/LoopVectorize.cpp +++ llvm/lib/Transforms/Vectorize/LoopVectorize.cpp @@ -4742,9 +4742,9 @@ case InductionDescriptor::IK_PtrInduction: { // Handle the pointer induction variable case. assert(P->getType()->isPointerTy() && "Unexpected type."); - assert(!VF.isScalable() && "Currently unsupported for scalable vectors"); if (Cost->isScalarAfterVectorization(P, State.VF)) { + assert(!VF.isScalable() && "Currently unsupported for scalable vectors"); // This is the normalized GEP that starts counting at zero. Value *PtrInd = Builder.CreateSExtOrTrunc(Induction, II.getStep()->getType()); @@ -4753,11 +4753,11 @@ // first lane. Otherwise, we generate all VF values. unsigned Lanes = Cost->isUniformAfterVectorization(P, State.VF) ? 1 - : State.VF.getKnownMinValue(); + : State.VF.getFixedValue(); for (unsigned Part = 0; Part < UF; ++Part) { for (unsigned Lane = 0; Lane < Lanes; ++Lane) { Constant *Idx = ConstantInt::get( - PtrInd->getType(), Lane + Part * State.VF.getKnownMinValue()); + PtrInd->getType(), Lane + Part * State.VF.getFixedValue()); Value *GlobalIdx = Builder.CreateAdd(PtrInd, Idx); Value *SclrGep = emitTransformedIndex(Builder, GlobalIdx, PSE.getSE(), DL, II); @@ -4785,11 +4785,11 @@ SCEVExpander Exp(*PSE.getSE(), DL, "induction"); Value *ScalarStepValue = Exp.expandCodeFor(ScalarStep, PhiType, InductionLoc); + Value *RuntimeVF = getRuntimeVF(Builder, PhiType, VF); + Value *NumUnrolledElems = Builder.CreateMul(RuntimeVF, ConstantInt::get(PhiType, State.UF)); Value *InductionGEP = GetElementPtrInst::Create( ScStValueType->getPointerElementType(), NewPointerPhi, - Builder.CreateMul( - ScalarStepValue, - ConstantInt::get(PhiType, State.VF.getKnownMinValue() * State.UF)), + Builder.CreateMul(ScalarStepValue, NumUnrolledElems), "ptr.ind", InductionLoc); NewPointerPhi->addIncoming(InductionGEP, LoopLatch); @@ -4798,8 +4798,8 @@ // () as offset. for (unsigned Part = 0; Part < State.UF; ++Part) { Type *VecPhiType = VectorType::get(PhiType, State.VF); - Value *StartOffset = - ConstantInt::get(VecPhiType, Part * State.VF.getKnownMinValue()); + Value *StartOffsetScalar = Builder.CreateMul(RuntimeVF, ConstantInt::get(PhiType, Part)); + Value *StartOffset = Builder.CreateVectorSplat(State.VF, StartOffsetScalar); // Create a vector of consecutive numbers from zero to VF. StartOffset = Builder.CreateAdd(StartOffset, Builder.CreateStepVector(VecPhiType)); @@ -4808,7 +4808,7 @@ ScStValueType->getPointerElementType(), NewPointerPhi, Builder.CreateMul(StartOffset, Builder.CreateVectorSplat( - State.VF.getKnownMinValue(), ScalarStepValue), + State.VF, ScalarStepValue), "vector.gep")); State.set(Def, GEP, Part); } Index: llvm/test/Transforms/LoopVectorize/AArch64/sve-widen-phi.ll =================================================================== --- /dev/null +++ llvm/test/Transforms/LoopVectorize/AArch64/sve-widen-phi.ll @@ -0,0 +1,67 @@ +; RUN: opt -mtriple aarch64-linux-gnu -mattr=+sve -loop-vectorize -dce -instcombine -S < %s | FileCheck %s + +; Ensure that we can vectorize loops such as: +; int *ptr = c; +; for (long long i = 0; i < n; i++) { +; int X1 = *ptr++; +; int X2 = *ptr++; +; a[i] = X1 + 1; +; b[i] = X2 + 1; +; } +; with scalable vectors, including unrolling. The test below makes sure +; that we can use gather instructions with the correct offsets, taking +; vscale into account. + +define void @widen_ptr_phi_unrolled(i32* noalias nocapture %a, i32* noalias nocapture %b, i32* nocapture readonly %c, i64 %n) { +; CHECK-LABEL: @widen_ptr_phi_unrolled( +; CHECK: vector.body: +; CHECK-NEXT: [[POINTER_PHI:%.*]] = phi i32* [ %c, %vector.ph ], [ %[[PTR_IND:.*]], %vector.body ] +; CHECK: [[TMP5:%.*]] = call i64 @llvm.vscale.i64() +; CHECK-NEXT: [[TMP6:%.*]] = shl i64 [[TMP5]], 2 +; CHECK-NEXT: [[TMP7:%.*]] = shl i64 [[TMP5]], 4 +; CHECK-NEXT: [[TMP8:%.*]] = call @llvm.experimental.stepvector.nxv4i64() +; CHECK-NEXT: [[VECTOR_GEP:%.*]] = shl [[TMP8]], shufflevector ( insertelement ( undef, i64 1, i32 0), undef, zeroinitializer) +; CHECK-NEXT: [[TMP9:%.*]] = getelementptr i32, i32* [[POINTER_PHI]], [[VECTOR_GEP]] +; CHECK-NEXT: [[DOTSPLATINSERT2:%.*]] = insertelement poison, i64 [[TMP6]], i32 0 +; CHECK-NEXT: [[DOTSPLAT3:%.*]] = shufflevector [[DOTSPLATINSERT2]], poison, zeroinitializer +; CHECK-NEXT: [[TMP10:%.*]] = call @llvm.experimental.stepvector.nxv4i64() +; CHECK-NEXT: [[TMP11:%.*]] = add [[DOTSPLAT3]], [[TMP10]] +; CHECK-NEXT: [[VECTOR_GEP4:%.*]] = shl [[TMP11]], shufflevector ( insertelement ( undef, i64 1, i32 0), undef, zeroinitializer) +; CHECK-NEXT: [[TMP12:%.*]] = getelementptr i32, i32* [[POINTER_PHI]], [[VECTOR_GEP4]] +; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds i32, [[TMP9]], i64 1 +; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds i32, [[TMP12]], i64 1 +; CHECK-NEXT: {{%.*}} = call @llvm.masked.gather.nxv4i32.nxv4p0i32( [[TMP9]], +; CHECK-NEXT: {{%.*}} = call @llvm.masked.gather.nxv4i32.nxv4p0i32( [[TMP12]], +; CHECK-NEXT: {{%.*}} = call @llvm.masked.gather.nxv4i32.nxv4p0i32( [[TMP13]], +; CHECK-NEXT: {{%.*}} = call @llvm.masked.gather.nxv4i32.nxv4p0i32( [[TMP14]], +; CHECK: [[PTR_IND]] = getelementptr i32, i32* [[POINTER_PHI]], i64 [[TMP7]] +entry: + br label %for.body + +for.body: ; preds = %entry, %for.body + %ptr.014 = phi i32* [ %incdec.ptr1, %for.body ], [ %c, %entry ] + %i.013 = phi i64 [ %inc, %for.body ], [ 0, %entry ] + %incdec.ptr = getelementptr inbounds i32, i32* %ptr.014, i64 1 + %0 = load i32, i32* %ptr.014, align 4 + %incdec.ptr1 = getelementptr inbounds i32, i32* %ptr.014, i64 2 + %1 = load i32, i32* %incdec.ptr, align 4 + %add = add nsw i32 %0, 1 + %arrayidx = getelementptr inbounds i32, i32* %a, i64 %i.013 + store i32 %add, i32* %arrayidx, align 4 + %add2 = add nsw i32 %1, 1 + %arrayidx3 = getelementptr inbounds i32, i32* %b, i64 %i.013 + store i32 %add2, i32* %arrayidx3, align 4 + %inc = add nuw nsw i64 %i.013, 1 + %exitcond.not = icmp eq i64 %inc, %n + br i1 %exitcond.not, label %for.exit, label %for.body, !llvm.loop !0 + +for.exit: ; preds = %for.body + ret void +} + +!0 = distinct !{!0, !1, !2, !3, !4, !5} +!1 = !{!"llvm.loop.mustprogress"} +!2 = !{!"llvm.loop.vectorize.width", i32 4} +!3 = !{!"llvm.loop.vectorize.scalable.enable", i1 true} +!4 = !{!"llvm.loop.vectorize.enable", i1 true} +!5 = !{!"llvm.loop.interleave.count", i32 2}