From 54b6908755cc2ddd9dd1d4ea6cd318e6e0f56539 Mon Sep 17 00:00:00 2001 From: Florian Hahn Date: Sun, 2 Nov 2025 21:10:19 +0000 Subject: [PATCH 1/3] [VPlan] Create resume phis in scalar preheader early. (NFC) Create phi recipes for scalar resume value up front in addInitialSkeleton during initial construction. This will allow moving the remaining code dealing with resume values to VPlan transforms/construction. --- llvm/lib/Transforms/Vectorize/VPlan.h | 4 ++- .../Vectorize/VPlanConstruction.cpp | 9 +++++ .../Transforms/Vectorize/VPlanTransforms.cpp | 36 ++++++++----------- .../Transforms/Vectorize/VPlanHCFGTest.cpp | 6 ++-- .../Vectorize/VPlanVerifierTest.cpp | 4 +-- 5 files changed, 33 insertions(+), 26 deletions(-) diff --git a/llvm/lib/Transforms/Vectorize/VPlan.h b/llvm/lib/Transforms/Vectorize/VPlan.h index ea88eaa42d945..82c0e46278b2e 100644 --- a/llvm/lib/Transforms/Vectorize/VPlan.h +++ b/llvm/lib/Transforms/Vectorize/VPlan.h @@ -1084,7 +1084,7 @@ class LLVM_ABI_FOR_TEST VPInstruction : public VPRecipeWithIRFlags, OpcodeTy Opcode; /// An optional name that can be used for the generated IR instruction. - const std::string Name; + std::string Name; /// Returns true if we can generate a scalar for the first lane only if /// needed. @@ -1183,6 +1183,8 @@ class LLVM_ABI_FOR_TEST VPInstruction : public VPRecipeWithIRFlags, /// Returns the symbolic name assigned to the VPInstruction. StringRef getName() const { return Name; } + + void setName(StringRef NewName) { Name = NewName.str(); } }; /// A specialization of VPInstruction augmenting it with a dedicated result diff --git a/llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp b/llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp index 4ffd5577d31a4..f21e234ba2070 100644 --- a/llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp +++ b/llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp @@ -533,6 +533,15 @@ static void addInitialSkeleton(VPlan &Plan, Type *InductionTy, DebugLoc IVDL, Plan.getEntry()->swapSuccessors(); createExtractsForLiveOuts(Plan, MiddleVPBB); + + VPBuilder ScalarPHBuilder(ScalarPH); + for (const auto &[PhiR, ScalarPhiR] : zip_equal( + drop_begin(HeaderVPBB->phis()), Plan.getScalarHeader()->phis())) { + auto *VectorPhiR = cast(&PhiR); + auto *ResumePhiR = ScalarPHBuilder.createScalarPhi( + {VectorPhiR, VectorPhiR->getOperand(0)}, VectorPhiR->getDebugLoc()); + cast(&ScalarPhiR)->addOperand(ResumePhiR); + } } std::unique_ptr diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp index d05c22e3aeb61..b80c43661c53c 100644 --- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp +++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp @@ -4459,9 +4459,10 @@ void VPlanTransforms::addBranchWeightToMiddleTerminator( /// Create and return a ResumePhi for \p WideIV, unless it is truncated. If the /// induction recipe is not canonical, creates a VPDerivedIVRecipe to compute /// the end value of the induction. -static VPInstruction *addResumePhiRecipeForInduction( - VPWidenInductionRecipe *WideIV, VPBuilder &VectorPHBuilder, - VPBuilder &ScalarPHBuilder, VPTypeAnalysis &TypeInfo, VPValue *VectorTC) { +static VPValue *addResumePhiRecipeForInduction(VPWidenInductionRecipe *WideIV, + VPBuilder &VectorPHBuilder, + VPTypeAnalysis &TypeInfo, + VPValue *VectorTC) { auto *WideIntOrFp = dyn_cast(WideIV); // Truncated wide inductions resume from the last lane of their vector value // in the last vector iteration which is handled elsewhere. @@ -4487,9 +4488,7 @@ static VPInstruction *addResumePhiRecipeForInduction( WideIV->getDebugLoc()); } - auto *ResumePhiRecipe = ScalarPHBuilder.createScalarPhi( - {EndValue, Start}, WideIV->getDebugLoc(), "bc.resume.val"); - return ResumePhiRecipe; + return EndValue; } void VPlanTransforms::addScalarResumePhis( @@ -4502,21 +4501,18 @@ void VPlanTransforms::addScalarResumePhis( VPBuilder VectorPHBuilder( cast(VectorRegion->getSinglePredecessor())); VPBuilder MiddleBuilder(MiddleVPBB, MiddleVPBB->getFirstNonPhi()); - VPBuilder ScalarPHBuilder(ScalarPH); - for (VPRecipeBase &ScalarPhiR : Plan.getScalarHeader()->phis()) { - auto *ScalarPhiIRI = cast(&ScalarPhiR); + for (VPRecipeBase &PhiR : Plan.getScalarPreheader()->phis()) { + auto *ResumePhiR = cast(&PhiR); // TODO: Extract final value from induction recipe initially, optimize to // pre-computed end value together in optimizeInductionExitUsers. - auto *VectorPhiR = - cast(Builder.getRecipe(&ScalarPhiIRI->getIRPhi())); + auto *VectorPhiR = cast(ResumePhiR->getOperand(0)); if (auto *WideIVR = dyn_cast(VectorPhiR)) { - if (VPInstruction *ResumePhi = addResumePhiRecipeForInduction( - WideIVR, VectorPHBuilder, ScalarPHBuilder, TypeInfo, - &Plan.getVectorTripCount())) { - assert(isa(ResumePhi) && "Expected a phi"); - IVEndValues[WideIVR] = ResumePhi->getOperand(0); - ScalarPhiIRI->addOperand(ResumePhi); + if (VPValue *ResumeV = addResumePhiRecipeForInduction( + WideIVR, VectorPHBuilder, TypeInfo, &Plan.getVectorTripCount())) { + IVEndValues[WideIVR] = ResumeV; + ResumePhiR->setOperand(0, ResumeV); + ResumePhiR->setName("bc.resume.val"); continue; } // TODO: Also handle truncated inductions here. Computing end-values @@ -4538,10 +4534,8 @@ void VPlanTransforms::addScalarResumePhis( ResumeFromVectorLoop = MiddleBuilder.createNaryOp( VPInstruction::ExtractLastElement, {ResumeFromVectorLoop}, {}, "vector.recur.extract"); - StringRef Name = IsFOR ? "scalar.recur.init" : "bc.merge.rdx"; - auto *ResumePhiR = ScalarPHBuilder.createScalarPhi( - {ResumeFromVectorLoop, VectorPhiR->getStartValue()}, {}, Name); - ScalarPhiIRI->addOperand(ResumePhiR); + ResumePhiR->setName(IsFOR ? "scalar.recur.init" : "bc.merge.rdx"); + ResumePhiR->setOperand(0, ResumeFromVectorLoop); } } diff --git a/llvm/unittests/Transforms/Vectorize/VPlanHCFGTest.cpp b/llvm/unittests/Transforms/Vectorize/VPlanHCFGTest.cpp index b99d656c5c50f..2cc6fdb96c74e 100644 --- a/llvm/unittests/Transforms/Vectorize/VPlanHCFGTest.cpp +++ b/llvm/unittests/Transforms/Vectorize/VPlanHCFGTest.cpp @@ -113,12 +113,13 @@ compound=true N0 -> N2 [ label="F"] N1 [label = "scalar.ph:\l" + + " EMIT-SCALAR vp\<%6\> = phi [ ir\<%indvars.iv\>, middle.block ], [ ir\<0\>, ir-bb\ ]\l" + "Successor(s): ir-bb\\l" ] N1 -> N3 [ label=""] N3 [label = "ir-bb\:\l" + - " IR %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]\l" + + " IR %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] (extra operand: vp\<%6\> from scalar.ph)\l" + " IR %arr.idx = getelementptr inbounds i32, ptr %A, i64 %indvars.iv\l" + " IR %l1 = load i32, ptr %arr.idx, align 4\l" + " IR %res = add i32 %l1, 10\l" + @@ -282,12 +283,13 @@ compound=true N0 -> N2 [ label="F"] N1 [label = "scalar.ph:\l" + + " EMIT-SCALAR vp\<%6\> = phi [ ir\<%iv\>, middle.block ], [ ir\<0\>, ir-bb\ ]\l" + "Successor(s): ir-bb\\l" ] N1 -> N3 [ label=""] N3 [label = "ir-bb\:\l" + - " IR %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]\l" + + " IR %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ] (extra operand: vp\<%6\> from scalar.ph)\l" + " IR %arr.idx = getelementptr inbounds i32, ptr %A, i64 %iv\l" + " IR %l1 = load i32, ptr %arr.idx, align 4\l" + " IR %c = icmp eq i32 %l1, 0\l" + diff --git a/llvm/unittests/Transforms/Vectorize/VPlanVerifierTest.cpp b/llvm/unittests/Transforms/Vectorize/VPlanVerifierTest.cpp index 169114ed6c310..cca779142bce4 100644 --- a/llvm/unittests/Transforms/Vectorize/VPlanVerifierTest.cpp +++ b/llvm/unittests/Transforms/Vectorize/VPlanVerifierTest.cpp @@ -346,6 +346,8 @@ TEST_F(VPIRVerifierTest, testVerifyIRPhiInScalarHeaderVPIRBB) { Function *F = M.getFunction("f"); BasicBlock *LoopHeader = F->getEntryBlock().getSingleSuccessor(); auto Plan = buildVPlan(LoopHeader); + VPValue *Zero = Plan->getConstantInt(32, 0); + Plan->getScalarHeader()->front().addOperand(Zero); #if GTEST_HAS_STREAM_REDIRECTION ::testing::internal::CaptureStderr(); @@ -387,8 +389,6 @@ TEST_F(VPIRVerifierTest, testVerifyIRPhiInExitVPIRBB) { {HeaderBlock->front().getVPSingleValue()}); DefI->insertBefore(Plan->getMiddleBlock()->getTerminator()); Plan->getExitBlocks()[0]->front().addOperand(DefI); - VPValue *Zero = Plan->getConstantInt(32, 0); - Plan->getScalarHeader()->front().addOperand(Zero); #if GTEST_HAS_STREAM_REDIRECTION ::testing::internal::CaptureStderr(); From 34b3fcabb625169cdeedbb1de6ca3c4b9117ae3a Mon Sep 17 00:00:00 2001 From: Florian Hahn Date: Sat, 15 Nov 2025 20:57:58 +0000 Subject: [PATCH 2/3] [VPlan] Remove PtrIV::IsScalarAfterVectorization, use VPlan analysis. --- .../Transforms/Vectorize/LoopVectorize.cpp | 15 +++---- .../Transforms/Vectorize/VPRecipeBuilder.h | 3 +- llvm/lib/Transforms/Vectorize/VPlan.h | 11 ++---- .../lib/Transforms/Vectorize/VPlanRecipes.cpp | 2 +- .../epilog-vectorization-widen-inductions.ll | 39 ++++++++----------- .../LoopVectorize/AArch64/sve-widen-gep.ll | 6 +-- .../X86/interleave-opaque-pointers.ll | 11 ++---- ...ction-index-width-smaller-than-iv-width.ll | 17 ++++---- 8 files changed, 41 insertions(+), 63 deletions(-) diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp index cbfbc29360b0b..826f80acd9294 100644 --- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp +++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp @@ -7655,7 +7655,7 @@ createWidenInductionRecipes(VPInstruction *PhiR, } VPHeaderPHIRecipe * -VPRecipeBuilder::tryToOptimizeInductionPHI(VPInstruction *VPI, VFRange &Range) { +VPRecipeBuilder::tryToOptimizeInductionPHI(VPInstruction *VPI) { auto *Phi = cast(VPI->getUnderlyingInstr()); // Check if this is an integer or fp induction. If so, build the recipe that @@ -7666,14 +7666,9 @@ VPRecipeBuilder::tryToOptimizeInductionPHI(VPInstruction *VPI, VFRange &Range) { // Check if this is pointer induction. If so, build the recipe for it. if (auto *II = Legal->getPointerInductionDescriptor(Phi)) { VPValue *Step = vputils::getOrCreateVPValueForSCEVExpr(Plan, II->getStep()); - return new VPWidenPointerInductionRecipe( - Phi, VPI->getOperand(0), Step, &Plan.getVFxUF(), *II, - LoopVectorizationPlanner::getDecisionAndClampRange( - [&](ElementCount VF) { - return CM.isScalarAfterVectorization(Phi, VF); - }, - Range), - VPI->getDebugLoc()); + return new VPWidenPointerInductionRecipe(Phi, VPI->getOperand(0), Step, + &Plan.getVFxUF(), *II, + VPI->getDebugLoc()); } return nullptr; } @@ -8162,7 +8157,7 @@ VPRecipeBase *VPRecipeBuilder::tryToCreateWidenRecipe(VPSingleDefRecipe *R, "Non-header phis should have been handled during predication"); auto *Phi = cast(R->getUnderlyingInstr()); assert(R->getNumOperands() == 2 && "Must have 2 operands for header phis"); - if ((Recipe = tryToOptimizeInductionPHI(PhiR, Range))) + if ((Recipe = tryToOptimizeInductionPHI(PhiR))) return Recipe; VPHeaderPHIRecipe *PhiRecipe = nullptr; diff --git a/llvm/lib/Transforms/Vectorize/VPRecipeBuilder.h b/llvm/lib/Transforms/Vectorize/VPRecipeBuilder.h index a7000aff06379..c6019e733fe77 100644 --- a/llvm/lib/Transforms/Vectorize/VPRecipeBuilder.h +++ b/llvm/lib/Transforms/Vectorize/VPRecipeBuilder.h @@ -100,8 +100,7 @@ class VPRecipeBuilder { /// Check if an induction recipe should be constructed for \p VPI. If so build /// and return it. If not, return null. - VPHeaderPHIRecipe *tryToOptimizeInductionPHI(VPInstruction *VPI, - VFRange &Range); + VPHeaderPHIRecipe *tryToOptimizeInductionPHI(VPInstruction *VPI); /// Optimize the special case where the operand of \p VPI is a constant /// integer induction variable. diff --git a/llvm/lib/Transforms/Vectorize/VPlan.h b/llvm/lib/Transforms/Vectorize/VPlan.h index 82c0e46278b2e..cbb93156fcaf9 100644 --- a/llvm/lib/Transforms/Vectorize/VPlan.h +++ b/llvm/lib/Transforms/Vectorize/VPlan.h @@ -2213,19 +2213,15 @@ class VPWidenIntOrFpInductionRecipe : public VPWidenInductionRecipe { }; class VPWidenPointerInductionRecipe : public VPWidenInductionRecipe { - bool IsScalarAfterVectorization; - public: /// Create a new VPWidenPointerInductionRecipe for \p Phi with start value \p /// Start and the number of elements unrolled \p NumUnrolledElems, typically /// VF*UF. VPWidenPointerInductionRecipe(PHINode *Phi, VPValue *Start, VPValue *Step, VPValue *NumUnrolledElems, - const InductionDescriptor &IndDesc, - bool IsScalarAfterVectorization, DebugLoc DL) + const InductionDescriptor &IndDesc, DebugLoc DL) : VPWidenInductionRecipe(VPDef::VPWidenPointerInductionSC, Phi, Start, - Step, IndDesc, DL), - IsScalarAfterVectorization(IsScalarAfterVectorization) { + Step, IndDesc, DL) { addOperand(NumUnrolledElems); } @@ -2234,8 +2230,7 @@ class VPWidenPointerInductionRecipe : public VPWidenInductionRecipe { VPWidenPointerInductionRecipe *clone() override { return new VPWidenPointerInductionRecipe( cast(getUnderlyingInstr()), getOperand(0), getOperand(1), - getOperand(2), getInductionDescriptor(), IsScalarAfterVectorization, - getDebugLoc()); + getOperand(2), getInductionDescriptor(), getDebugLoc()); } VP_CLASSOF_IMPL(VPDef::VPWidenPointerInductionSC) diff --git a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp index aa85bd435ee9e..5071941eb1413 100644 --- a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp +++ b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp @@ -4310,7 +4310,7 @@ void VPCanonicalIVPHIRecipe::print(raw_ostream &O, const Twine &Indent, #endif bool VPWidenPointerInductionRecipe::onlyScalarsGenerated(bool IsScalable) { - return IsScalarAfterVectorization && + return vputils::onlyScalarValuesUsed(this) && (!IsScalable || vputils::onlyFirstLaneUsed(this)); } diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/epilog-vectorization-widen-inductions.ll b/llvm/test/Transforms/LoopVectorize/AArch64/epilog-vectorization-widen-inductions.ll index d23e3c29b59e5..28b2d30b03fd6 100644 --- a/llvm/test/Transforms/LoopVectorize/AArch64/epilog-vectorization-widen-inductions.ll +++ b/llvm/test/Transforms/LoopVectorize/AArch64/epilog-vectorization-widen-inductions.ll @@ -11,21 +11,14 @@ define void @test_widen_ptr_induction(ptr %ptr.start.1) { ; CHECK: vector.main.loop.iter.check: ; CHECK-NEXT: br i1 false, label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]] ; CHECK: vector.ph: +; CHECK-NEXT: [[TMP0:%.*]] = getelementptr i8, ptr [[PTR_START_1:%.*]], i64 10000 ; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK: vector.body: ; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] -; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 -; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 1 -; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 2 -; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 3 -; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[PTR_START_1:%.*]], i64 [[TMP0]] -; CHECK-NEXT: [[NEXT_GEP1:%.*]] = getelementptr i8, ptr [[PTR_START_1]], i64 [[TMP1]] -; CHECK-NEXT: [[TMP4:%.*]] = insertelement <2 x ptr> poison, ptr [[NEXT_GEP]], i32 0 -; CHECK-NEXT: [[TMP5:%.*]] = insertelement <2 x ptr> [[TMP4]], ptr [[NEXT_GEP1]], i32 1 -; CHECK-NEXT: [[NEXT_GEP2:%.*]] = getelementptr i8, ptr [[PTR_START_1]], i64 [[TMP2]] -; CHECK-NEXT: [[NEXT_GEP3:%.*]] = getelementptr i8, ptr [[PTR_START_1]], i64 [[TMP3]] -; CHECK-NEXT: [[TMP6:%.*]] = insertelement <2 x ptr> poison, ptr [[NEXT_GEP2]], i32 0 -; CHECK-NEXT: [[TMP7:%.*]] = insertelement <2 x ptr> [[TMP6]], ptr [[NEXT_GEP3]], i32 1 +; CHECK-NEXT: [[POINTER_PHI:%.*]] = phi ptr [ [[PTR_START_1]], [[VECTOR_PH]] ], [ [[PTR_IND:%.*]], [[VECTOR_BODY]] ] +; CHECK-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <2 x i64> +; CHECK-NEXT: [[NEXT_GEP:%.*]] = extractelement <2 x ptr> [[TMP5]], i32 0 +; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i8, <2 x ptr> [[TMP5]], <2 x i64> splat (i64 2) ; CHECK-NEXT: [[TMP8:%.*]] = icmp ne <2 x ptr> [[TMP5]], zeroinitializer ; CHECK-NEXT: [[TMP11:%.*]] = extractelement <2 x i1> [[TMP8]], i32 0 ; CHECK-NEXT: [[TMP12:%.*]] = extractelement <2 x i1> [[TMP8]], i32 1 @@ -40,25 +33,24 @@ define void @test_widen_ptr_induction(ptr %ptr.start.1) { ; CHECK-NEXT: store <2 x i8> zeroinitializer, ptr [[NEXT_GEP]], align 1 ; CHECK-NEXT: store <2 x i8> zeroinitializer, ptr [[TMP15]], align 1 ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 +; CHECK-NEXT: [[PTR_IND]] = getelementptr i8, ptr [[POINTER_PHI]], i64 4 ; CHECK-NEXT: [[TMP16:%.*]] = icmp eq i64 [[INDEX_NEXT]], 10000 ; CHECK-NEXT: br i1 [[TMP16]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], {{!llvm.loop ![0-9]+}} ; CHECK: middle.block: ; CHECK-NEXT: br i1 false, label [[EXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]] ; CHECK: vec.epilog.iter.check: ; CHECK-NEXT: [[IND_END4:%.*]] = getelementptr i8, ptr [[PTR_START_1]], i64 10000 -; CHECK-NEXT: br i1 true, label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]] +; CHECK-NEXT: br i1 true, label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF3:![0-9]+]] ; CHECK: vec.epilog.ph: ; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ 10000, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ] +; CHECK-NEXT: [[BC_RESUME_VAL1:%.*]] = phi ptr [ [[TMP0]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[PTR_START_1]], [[VECTOR_MAIN_LOOP_ITER_CHECK]] ] ; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[PTR_START_1]], i64 10000 ; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]] ; CHECK: vec.epilog.vector.body: ; CHECK-NEXT: [[INDEX6:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT9:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ] -; CHECK-NEXT: [[TMP17:%.*]] = add i64 [[INDEX6]], 0 -; CHECK-NEXT: [[TMP18:%.*]] = add i64 [[INDEX6]], 1 -; CHECK-NEXT: [[NEXT_GEP7:%.*]] = getelementptr i8, ptr [[PTR_START_1]], i64 [[TMP17]] -; CHECK-NEXT: [[NEXT_GEP8:%.*]] = getelementptr i8, ptr [[PTR_START_1]], i64 [[TMP18]] -; CHECK-NEXT: [[TMP19:%.*]] = insertelement <2 x ptr> poison, ptr [[NEXT_GEP7]], i32 0 -; CHECK-NEXT: [[TMP20:%.*]] = insertelement <2 x ptr> [[TMP19]], ptr [[NEXT_GEP8]], i32 1 +; CHECK-NEXT: [[POINTER_PHI2:%.*]] = phi ptr [ [[BC_RESUME_VAL1]], [[VEC_EPILOG_PH]] ], [ [[PTR_IND5:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ] +; CHECK-NEXT: [[TMP20:%.*]] = getelementptr i8, ptr [[POINTER_PHI2]], <2 x i64> +; CHECK-NEXT: [[NEXT_GEP7:%.*]] = extractelement <2 x ptr> [[TMP20]], i32 0 ; CHECK-NEXT: [[TMP21:%.*]] = icmp ne <2 x ptr> [[TMP20]], zeroinitializer ; CHECK-NEXT: [[TMP22:%.*]] = extractelement <2 x i1> [[TMP21]], i32 0 ; CHECK-NEXT: [[TMP23:%.*]] = extractelement <2 x i1> [[TMP21]], i32 1 @@ -66,6 +58,7 @@ define void @test_widen_ptr_induction(ptr %ptr.start.1) { ; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP23]]) ; CHECK-NEXT: store <2 x i8> zeroinitializer, ptr [[NEXT_GEP7]], align 1 ; CHECK-NEXT: [[INDEX_NEXT9]] = add nuw i64 [[INDEX6]], 2 +; CHECK-NEXT: [[PTR_IND5]] = getelementptr i8, ptr [[POINTER_PHI2]], i64 2 ; CHECK-NEXT: [[TMP25:%.*]] = icmp eq i64 [[INDEX_NEXT9]], 10000 ; CHECK-NEXT: br i1 [[TMP25]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], {{!llvm.loop ![0-9]+}} ; CHECK: vec.epilog.middle.block: @@ -136,7 +129,7 @@ define void @test_widen_induction(ptr %A, i64 %N) { ; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]] ; CHECK: vec.epilog.iter.check: ; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_MOD_VF]], 2 -; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]] +; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF3]] ; CHECK: vec.epilog.ph: ; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ] ; CHECK-NEXT: [[N_MOD_VF2:%.*]] = urem i64 [[N]], 2 @@ -221,7 +214,7 @@ define void @test_widen_induction_variable_start(ptr %A, i64 %N, i64 %start) { ; CHECK: vec.epilog.iter.check: ; CHECK-NEXT: [[IND_END5:%.*]] = add i64 [[START]], [[N_VEC]] ; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_MOD_VF]], 2 -; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]] +; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF3]] ; CHECK: vec.epilog.ph: ; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ] ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[START]], [[VECTOR_MAIN_LOOP_ITER_CHECK]] ] @@ -304,7 +297,7 @@ define void @test_widen_induction_step_2(ptr %A, i64 %N, i32 %step) { ; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]] ; CHECK: vec.epilog.iter.check: ; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_MOD_VF]], 2 -; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]] +; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF3]] ; CHECK: vec.epilog.ph: ; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[IND_END4]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ] ; CHECK-NEXT: [[N_MOD_VF2:%.*]] = urem i64 [[N]], 2 @@ -416,7 +409,7 @@ define void @test_widen_truncated_induction(ptr %A) { ; CHECK: middle.block: ; CHECK-NEXT: br i1 true, label [[EXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]] ; CHECK: vec.epilog.iter.check: -; CHECK-NEXT: br i1 true, label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]] +; CHECK-NEXT: br i1 true, label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF3]] ; CHECK: vec.epilog.ph: ; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ 10000, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ] ; CHECK-NEXT: [[TMP5:%.*]] = trunc i64 [[VEC_EPILOG_RESUME_VAL]] to i8 diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/sve-widen-gep.ll b/llvm/test/Transforms/LoopVectorize/AArch64/sve-widen-gep.ll index f223786a07cdf..456c03824106a 100644 --- a/llvm/test/Transforms/LoopVectorize/AArch64/sve-widen-gep.ll +++ b/llvm/test/Transforms/LoopVectorize/AArch64/sve-widen-gep.ll @@ -112,15 +112,11 @@ define void @pointer_induction(ptr noalias %start, i64 %N) { ; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK: vector.body: ; CHECK-NEXT: [[INDEX2:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] -; CHECK-NEXT: [[POINTER_PHI:%.*]] = phi ptr [ [[START]], [[VECTOR_PH]] ], [ [[PTR_IND:%.*]], [[VECTOR_BODY]] ] -; CHECK-NEXT: [[TMP12:%.*]] = call @llvm.stepvector.nxv2i64() -; CHECK-NEXT: [[VECTOR_GEP:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], [[TMP12]] -; CHECK-NEXT: [[TMP15:%.*]] = extractelement [[VECTOR_GEP]], i32 0 +; CHECK-NEXT: [[TMP15:%.*]] = getelementptr i8, ptr [[START]], i64 [[INDEX2]] ; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load , ptr [[TMP15]], align 1 ; CHECK-NEXT: [[TMP17:%.*]] = add [[WIDE_LOAD]], splat (i8 1) ; CHECK-NEXT: store [[TMP17]], ptr [[TMP15]], align 1 ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX2]], [[TMP4]] -; CHECK-NEXT: [[PTR_IND]] = getelementptr i8, ptr [[POINTER_PHI]], i64 [[TMP4]] ; CHECK-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] ; CHECK-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] ; CHECK: middle.block: diff --git a/llvm/test/Transforms/LoopVectorize/X86/interleave-opaque-pointers.ll b/llvm/test/Transforms/LoopVectorize/X86/interleave-opaque-pointers.ll index 3a07bcca523ce..13da121fe2dc2 100644 --- a/llvm/test/Transforms/LoopVectorize/X86/interleave-opaque-pointers.ll +++ b/llvm/test/Transforms/LoopVectorize/X86/interleave-opaque-pointers.ll @@ -25,17 +25,14 @@ define void @test_pr55375_interleave_opaque_ptr(ptr %start, ptr %end) { ; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK: vector.body: ; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] -; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 16 -; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[OFFSET_IDX]], 0 -; CHECK-NEXT: [[TMP6:%.*]] = add i64 [[OFFSET_IDX]], 16 -; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP5]] -; CHECK-NEXT: [[TMP8:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP6]] -; CHECK-NEXT: [[TMP9:%.*]] = insertelement <2 x ptr> poison, ptr [[TMP7]], i32 0 -; CHECK-NEXT: [[TMP10:%.*]] = insertelement <2 x ptr> [[TMP9]], ptr [[TMP8]], i32 1 +; CHECK-NEXT: [[POINTER_PHI:%.*]] = phi ptr [ [[START]], [[VECTOR_PH]] ], [ [[PTR_IND:%.*]], [[VECTOR_BODY]] ] +; CHECK-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <2 x i64> +; CHECK-NEXT: [[TMP7:%.*]] = extractelement <2 x ptr> [[TMP10]], i32 0 ; CHECK-NEXT: [[TMP12:%.*]] = shufflevector <2 x ptr> zeroinitializer, <2 x ptr> [[TMP10]], <4 x i32> ; CHECK-NEXT: [[INTERLEAVED_VEC:%.*]] = shufflevector <4 x ptr> [[TMP12]], <4 x ptr> poison, <4 x i32> ; CHECK-NEXT: store <4 x ptr> [[INTERLEAVED_VEC]], ptr [[TMP7]], align 8 ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 +; CHECK-NEXT: [[PTR_IND]] = getelementptr i8, ptr [[POINTER_PHI]], i64 32 ; CHECK-NEXT: [[TMP13:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] ; CHECK-NEXT: br i1 [[TMP13]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] ; CHECK: middle.block: diff --git a/llvm/test/Transforms/LoopVectorize/pointer-induction-index-width-smaller-than-iv-width.ll b/llvm/test/Transforms/LoopVectorize/pointer-induction-index-width-smaller-than-iv-width.ll index cebf90ae9f8ce..94392f856c386 100644 --- a/llvm/test/Transforms/LoopVectorize/pointer-induction-index-width-smaller-than-iv-width.ll +++ b/llvm/test/Transforms/LoopVectorize/pointer-induction-index-width-smaller-than-iv-width.ll @@ -15,16 +15,20 @@ define void @wide_ptr_induction_index_width_smaller_than_iv_width(ptr noalias %s ; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] ; CHECK: [[VECTOR_BODY]]: ; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] -; CHECK-NEXT: [[POINTER_PHI:%.*]] = phi ptr [ [[SRC]], %[[VECTOR_PH]] ], [ [[PTR_IND:%.*]], %[[VECTOR_BODY]] ] -; CHECK-NEXT: [[VECTOR_GEP:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <4 x i32> -; CHECK-NEXT: [[TMP5:%.*]] = extractelement <4 x ptr> [[VECTOR_GEP]], i32 0 -; CHECK-NEXT: [[TMP12:%.*]] = extractelement <4 x ptr> [[VECTOR_GEP]], i32 1 -; CHECK-NEXT: [[TMP13:%.*]] = extractelement <4 x ptr> [[VECTOR_GEP]], i32 2 -; CHECK-NEXT: [[TMP14:%.*]] = extractelement <4 x ptr> [[VECTOR_GEP]], i32 3 ; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 0 ; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 1 ; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 2 ; CHECK-NEXT: [[TMP4:%.*]] = add i64 [[INDEX]], 3 +; CHECK-NEXT: [[DOTCAST:%.*]] = trunc i64 [[INDEX]] to i32 +; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i32 [[DOTCAST]], 8 +; CHECK-NEXT: [[TMP11:%.*]] = add i32 [[OFFSET_IDX]], 0 +; CHECK-NEXT: [[TMP6:%.*]] = add i32 [[OFFSET_IDX]], 8 +; CHECK-NEXT: [[TMP16:%.*]] = add i32 [[OFFSET_IDX]], 16 +; CHECK-NEXT: [[TMP17:%.*]] = add i32 [[OFFSET_IDX]], 24 +; CHECK-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[SRC]], i32 [[TMP11]] +; CHECK-NEXT: [[TMP12:%.*]] = getelementptr i8, ptr [[SRC]], i32 [[TMP6]] +; CHECK-NEXT: [[TMP13:%.*]] = getelementptr i8, ptr [[SRC]], i32 [[TMP16]] +; CHECK-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[SRC]], i32 [[TMP17]] ; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP5]], align 1 ; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i64, ptr [[DST_0]], i64 [[TMP1]] ; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds i64, ptr [[DST_0]], i64 [[TMP2]] @@ -36,7 +40,6 @@ define void @wide_ptr_induction_index_width_smaller_than_iv_width(ptr noalias %s ; CHECK-NEXT: store ptr [[TMP13]], ptr [[TMP9]], align 8 ; CHECK-NEXT: store ptr [[TMP14]], ptr [[TMP10]], align 8 ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 -; CHECK-NEXT: [[PTR_IND]] = getelementptr i8, ptr [[POINTER_PHI]], i32 32 ; CHECK-NEXT: [[TMP15:%.*]] = icmp eq i64 [[INDEX_NEXT]], 100 ; CHECK-NEXT: br i1 [[TMP15]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] ; CHECK: [[MIDDLE_BLOCK]]: From 2c266d7f1663247fbcca72473557803cc376f44f Mon Sep 17 00:00:00 2001 From: Florian Hahn Date: Sun, 16 Nov 2025 21:37:24 +0000 Subject: [PATCH 3/3] [VPlan] Create header phis once, after constructing VPlan0 (NFC). --- .../Transforms/Vectorize/LoopVectorize.cpp | 118 +++--------------- .../Transforms/Vectorize/VPRecipeBuilder.h | 4 - llvm/lib/Transforms/Vectorize/VPlan.h | 6 +- .../Vectorize/VPlanConstruction.cpp | 87 +++++++++++++ .../Transforms/Vectorize/VPlanTransforms.h | 11 ++ 5 files changed, 123 insertions(+), 103 deletions(-) diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp index 826f80acd9294..9b91b520b4333 100644 --- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp +++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp @@ -1423,6 +1423,11 @@ class LoopVectorizationCostModel { return InLoopReductions.contains(Phi); } + /// Returns the set of in-loop reduction PHIs. + const SmallPtrSetImpl &getInLoopReductions() const { + return InLoopReductions; + } + /// Returns true if the predicated reduction select should be used to set the /// incoming value for the reduction phi. bool usePredicatedReductionSelect() const { @@ -7626,53 +7631,6 @@ VPWidenMemoryRecipe *VPRecipeBuilder::tryToWidenMemory(VPInstruction *VPI, VPIRMetadata(*Store, LVer), VPI->getDebugLoc()); } -/// Creates a VPWidenIntOrFpInductionRecipe for \p PhiR. If needed, it will -/// also insert a recipe to expand the step for the induction recipe. -static VPWidenIntOrFpInductionRecipe * -createWidenInductionRecipes(VPInstruction *PhiR, - const InductionDescriptor &IndDesc, VPlan &Plan, - ScalarEvolution &SE, Loop &OrigLoop) { - assert(SE.isLoopInvariant(IndDesc.getStep(), &OrigLoop) && - "step must be loop invariant"); - - VPValue *Start = PhiR->getOperand(0); - assert(Plan.getLiveIn(IndDesc.getStartValue()) == Start && - "Start VPValue must match IndDesc's start value"); - - VPValue *Step = - vputils::getOrCreateVPValueForSCEVExpr(Plan, IndDesc.getStep()); - - // Update wide induction increments to use the same step as the corresponding - // wide induction. This enables detecting induction increments directly in - // VPlan and removes redundant splats. - using namespace llvm::VPlanPatternMatch; - if (match(PhiR->getOperand(1), m_Add(m_Specific(PhiR), m_VPValue()))) - PhiR->getOperand(1)->getDefiningRecipe()->setOperand(1, Step); - - PHINode *Phi = cast(PhiR->getUnderlyingInstr()); - return new VPWidenIntOrFpInductionRecipe(Phi, Start, Step, &Plan.getVF(), - IndDesc, PhiR->getDebugLoc()); -} - -VPHeaderPHIRecipe * -VPRecipeBuilder::tryToOptimizeInductionPHI(VPInstruction *VPI) { - auto *Phi = cast(VPI->getUnderlyingInstr()); - - // Check if this is an integer or fp induction. If so, build the recipe that - // produces its scalar and vector values. - if (auto *II = Legal->getIntOrFpInductionDescriptor(Phi)) - return createWidenInductionRecipes(VPI, *II, Plan, *PSE.getSE(), *OrigLoop); - - // Check if this is pointer induction. If so, build the recipe for it. - if (auto *II = Legal->getPointerInductionDescriptor(Phi)) { - VPValue *Step = vputils::getOrCreateVPValueForSCEVExpr(Plan, II->getStep()); - return new VPWidenPointerInductionRecipe(Phi, VPI->getOperand(0), Step, - &Plan.getVFxUF(), *II, - VPI->getDebugLoc()); - } - return nullptr; -} - VPWidenIntOrFpInductionRecipe * VPRecipeBuilder::tryToOptimizeInductionTruncate(VPInstruction *VPI, VFRange &Range) { @@ -8149,45 +8107,7 @@ VPRecipeBase *VPRecipeBuilder::tryToCreateWidenRecipe(VPSingleDefRecipe *R, // First, check for specific widening recipes that deal with inductions, Phi // nodes, calls and memory operations. VPRecipeBase *Recipe; - if (auto *PhiR = dyn_cast(R)) { - VPBasicBlock *Parent = PhiR->getParent(); - [[maybe_unused]] VPRegionBlock *LoopRegionOf = - Parent->getEnclosingLoopRegion(); - assert(LoopRegionOf && LoopRegionOf->getEntry() == Parent && - "Non-header phis should have been handled during predication"); - auto *Phi = cast(R->getUnderlyingInstr()); - assert(R->getNumOperands() == 2 && "Must have 2 operands for header phis"); - if ((Recipe = tryToOptimizeInductionPHI(PhiR))) - return Recipe; - - VPHeaderPHIRecipe *PhiRecipe = nullptr; - assert((Legal->isReductionVariable(Phi) || - Legal->isFixedOrderRecurrence(Phi)) && - "can only widen reductions and fixed-order recurrences here"); - VPValue *StartV = R->getOperand(0); - if (Legal->isReductionVariable(Phi)) { - const RecurrenceDescriptor &RdxDesc = Legal->getRecurrenceDescriptor(Phi); - assert(RdxDesc.getRecurrenceStartValue() == - Phi->getIncomingValueForBlock(OrigLoop->getLoopPreheader())); - - // If the PHI is used by a partial reduction, set the scale factor. - unsigned ScaleFactor = - getScalingForReduction(RdxDesc.getLoopExitInstr()).value_or(1); - PhiRecipe = new VPReductionPHIRecipe( - Phi, RdxDesc.getRecurrenceKind(), *StartV, CM.isInLoopReduction(Phi), - CM.useOrderedReductions(RdxDesc), ScaleFactor); - } else { - // TODO: Currently fixed-order recurrences are modeled as chains of - // first-order recurrences. If there are no users of the intermediate - // recurrences in the chain, the fixed order recurrence should be modeled - // directly, enabling more efficient codegen. - PhiRecipe = new VPFirstOrderRecurrencePHIRecipe(Phi, *StartV); - } - // Add backedge value. - PhiRecipe->addOperand(R->getOperand(1)); - return PhiRecipe; - } - assert(!R->isPhi() && "only VPPhi nodes expected at this point"); + assert(!R->isPhi() && "phis must be handled earlier"); auto *VPI = cast(R); Instruction *Instr = R->getUnderlyingInstr(); @@ -8244,6 +8164,9 @@ VPRecipeBuilder::tryToCreatePartialReduction(VPInstruction *Reduction, if (isa(BinOp) || isa(BinOp)) std::swap(BinOp, Accumulator); + if (auto *RedPhiR = dyn_cast(Accumulator)) + RedPhiR->setVFScaleFactor(ScaleFactor); + assert(ScaleFactor == vputils::getVFScaleFactor(Accumulator->getDefiningRecipe()) && "all accumulators in chain must have same scale factor"); @@ -8290,6 +8213,12 @@ void LoopVectorizationPlanner::buildVPlansWithVPRecipes(ElementCount MinVF, OrigLoop, *LI, Legal->getWidestInductionType(), getDebugLocFromInstOrOperands(Legal->getPrimaryInduction()), PSE); + // Create recipes for header phis. + VPlanTransforms::createHeaderPhiRecipes( + *VPlan0, *PSE.getSE(), *OrigLoop, Legal->getInductionVars(), + Legal->getReductionVars(), Legal->getFixedOrderRecurrences(), + CM.getInLoopReductions(), Hints.allowReordering()); + auto MaxVFTimes2 = MaxVF * 2; for (ElementCount VF = MinVF; ElementCount::isKnownLT(VF, MaxVFTimes2);) { VFRange SubRange = {VF, MaxVFTimes2}; @@ -8410,25 +8339,18 @@ VPlanPtr LoopVectorizationPlanner::tryToBuildVPlanWithVPRecipes( // Mapping from VPValues in the initial plan to their widened VPValues. Needed // temporarily to update created block masks. DenseMap Old2New; + + // Now process all other blocks and instructions. for (VPBasicBlock *VPBB : VPBlockUtils::blocksOnly(RPOT)) { // Convert input VPInstructions to widened recipes. for (VPRecipeBase &R : make_early_inc_range(*VPBB)) { - auto *SingleDef = cast(&R); - auto *UnderlyingValue = SingleDef->getUnderlyingValue(); - // Skip recipes that do not need transforming, including canonical IV, - // wide canonical IV and VPInstructions without underlying values. The - // latter are added above for masking. - // FIXME: Migrate code relying on the underlying instruction from VPlan0 - // to construct recipes below to not use the underlying instruction. - if (isa( - &R) || - (isa(&R) && !UnderlyingValue)) + auto *SingleDef = dyn_cast(&R); + if (!SingleDef || !SingleDef->getUnderlyingValue()) continue; - assert(isa(&R) && UnderlyingValue && "unsupported recipe"); // TODO: Gradually replace uses of underlying instruction by analyses on // VPlan. - Instruction *Instr = cast(UnderlyingValue); + Instruction *Instr = cast(SingleDef->getUnderlyingValue()); Builder.setInsertPoint(SingleDef); // The stores with invariant address inside the loop will be deleted, and diff --git a/llvm/lib/Transforms/Vectorize/VPRecipeBuilder.h b/llvm/lib/Transforms/Vectorize/VPRecipeBuilder.h index c6019e733fe77..367b42d72633d 100644 --- a/llvm/lib/Transforms/Vectorize/VPRecipeBuilder.h +++ b/llvm/lib/Transforms/Vectorize/VPRecipeBuilder.h @@ -98,10 +98,6 @@ class VPRecipeBuilder { /// recipe that takes an additional VPInstruction for the mask. VPWidenMemoryRecipe *tryToWidenMemory(VPInstruction *VPI, VFRange &Range); - /// Check if an induction recipe should be constructed for \p VPI. If so build - /// and return it. If not, return null. - VPHeaderPHIRecipe *tryToOptimizeInductionPHI(VPInstruction *VPI); - /// Optimize the special case where the operand of \p VPI is a constant /// integer induction variable. VPWidenIntOrFpInductionRecipe * diff --git a/llvm/lib/Transforms/Vectorize/VPlan.h b/llvm/lib/Transforms/Vectorize/VPlan.h index cbb93156fcaf9..b8522e7305ca0 100644 --- a/llvm/lib/Transforms/Vectorize/VPlan.h +++ b/llvm/lib/Transforms/Vectorize/VPlan.h @@ -2306,8 +2306,10 @@ struct VPFirstOrderRecurrencePHIRecipe : public VPHeaderPHIRecipe { VP_CLASSOF_IMPL(VPDef::VPFirstOrderRecurrencePHISC) VPFirstOrderRecurrencePHIRecipe *clone() override { - return new VPFirstOrderRecurrencePHIRecipe( + auto *R = new VPFirstOrderRecurrencePHIRecipe( cast(getUnderlyingInstr()), *getOperand(0)); + R->addOperand(getOperand(1)); + return R; } void execute(VPTransformState &State) override; @@ -2376,6 +2378,8 @@ class VPReductionPHIRecipe : public VPHeaderPHIRecipe, /// Get the factor that the VF of this recipe's output should be scaled by. unsigned getVFScaleFactor() const { return VFScaleFactor; } + void setVFScaleFactor(unsigned ScaleFactor) { VFScaleFactor = ScaleFactor; } + #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP) /// Print the recipe. void print(raw_ostream &O, const Twine &Indent, diff --git a/llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp b/llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp index f21e234ba2070..d517271b868f0 100644 --- a/llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp +++ b/llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp @@ -553,6 +553,93 @@ VPlanTransforms::buildVPlan0(Loop *TheLoop, LoopInfo &LI, Type *InductionTy, return VPlan0; } +/// Creates a VPWidenIntOrFpInductionRecipe or VPWidenPointerInductionRecipe +/// for \p Phi based on \p IndDesc. +static VPHeaderPHIRecipe * +createWidenInductionRecipe(PHINode *Phi, VPPhi *PhiR, + const InductionDescriptor &IndDesc, VPlan &Plan, + ScalarEvolution &SE, Loop &OrigLoop) { + assert(SE.isLoopInvariant(IndDesc.getStep(), &OrigLoop) && + "step must be loop invariant"); + + VPValue *Start = PhiR->getOperand(0); + assert(Plan.getLiveIn(IndDesc.getStartValue()) == Start && + "Start VPValue must match IndDesc's start value"); + VPValue *Step = + vputils::getOrCreateVPValueForSCEVExpr(Plan, IndDesc.getStep()); + + if (IndDesc.getKind() == InductionDescriptor::IK_PtrInduction) + return new VPWidenPointerInductionRecipe(Phi, Start, Step, &Plan.getVFxUF(), + IndDesc, PhiR->getDebugLoc()); + + // Update wide induction increments to use the same step as the corresponding + // wide induction. This enables detecting induction increments directly in + // VPlan and removes redundant splats. + using namespace llvm::VPlanPatternMatch; + if (match(PhiR->getOperand(1), m_Add(m_Specific(PhiR), m_VPValue()))) + PhiR->getOperand(1)->getDefiningRecipe()->setOperand(1, Step); + + return new VPWidenIntOrFpInductionRecipe(Phi, Start, Step, &Plan.getVF(), + IndDesc, PhiR->getDebugLoc()); +} + +void VPlanTransforms::createHeaderPhiRecipes( + VPlan &Plan, ScalarEvolution &SE, Loop &OrigLoop, + const MapVector &Inductions, + const MapVector &Reductions, + const SmallPtrSetImpl &FixedOrderRecurrences, + const SmallPtrSetImpl &InLoopReductions, bool AllowReordering) { + + VPBasicBlock *HeaderVPBB = cast( + Plan.getEntry()->getSuccessors()[1]->getSingleSuccessor()); + + for (VPRecipeBase &R : make_early_inc_range(*HeaderVPBB)) { + if (isa(&R)) + continue; + auto *PhiR = dyn_cast(&R); + if (!PhiR) + break; + + // TODO: Gradually replace uses of underlying instruction by analyses on + // VPlan. + auto *Phi = cast(PhiR->getUnderlyingInstr()); + assert(PhiR->getNumOperands() == 2 && + "Must have 2 operands for header phis"); + + VPHeaderPHIRecipe *HeaderPhiR = nullptr; + auto InductionIt = Inductions.find(Phi); + if (InductionIt != Inductions.end()) { + HeaderPhiR = createWidenInductionRecipe(Phi, PhiR, InductionIt->second, + Plan, SE, OrigLoop); + } else { + VPValue *Start = PhiR->getOperand(0); + auto ReductionIt = Reductions.find(Phi); + if (ReductionIt != Reductions.end()) { + const RecurrenceDescriptor &RdxDesc = ReductionIt->second; + assert(RdxDesc.getRecurrenceStartValue() == + Phi->getIncomingValueForBlock(OrigLoop.getLoopPreheader())); + + bool UseOrderedReductions = !AllowReordering && RdxDesc.isOrdered(); + HeaderPhiR = new VPReductionPHIRecipe( + Phi, RdxDesc.getRecurrenceKind(), *Start, + InLoopReductions.contains(Phi), UseOrderedReductions); + } else { + assert(FixedOrderRecurrences.contains(Phi) && + "can only widen reductions and fixed-order recurrences here"); + // TODO: Currently fixed-order recurrences are modeled as chains of + // first-order recurrences. If there are no users of the intermediate + // recurrences in the chain, the fixed order recurrence should be + // modeled directly, enabling more efficient codegen. + HeaderPhiR = new VPFirstOrderRecurrencePHIRecipe(Phi, *Start); + } + HeaderPhiR->addOperand(PhiR->getOperand(1)); + } + HeaderPhiR->insertBefore(PhiR); + PhiR->replaceAllUsesWith(HeaderPhiR); + PhiR->eraseFromParent(); + } +} + void VPlanTransforms::handleEarlyExits(VPlan &Plan, bool HasUncountableEarlyExit) { auto *MiddleVPBB = cast( diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.h b/llvm/lib/Transforms/Vectorize/VPlanTransforms.h index e3bde8a47dcbc..84224be96ab68 100644 --- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.h +++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.h @@ -101,6 +101,17 @@ struct VPlanTransforms { buildVPlan0(Loop *TheLoop, LoopInfo &LI, Type *InductionTy, DebugLoc IVDL, PredicatedScalarEvolution &PSE); + /// Replace VPPhi recipes in \p Plan's header with corresponding + /// VPHeaderPHIRecipe subclasses for inductions, reductions, and + /// fixed-order recurrences. This processes all header phis and creates + /// the appropriate widened recipe for each one. + static void createHeaderPhiRecipes( + VPlan &Plan, ScalarEvolution &SE, Loop &OrigLoop, + const MapVector &Inductions, + const MapVector &Reductions, + const SmallPtrSetImpl &FixedOrderRecurrences, + const SmallPtrSetImpl &InLoopReductions, bool AllowReordering); + /// Update \p Plan to account for all early exits. LLVM_ABI_FOR_TEST static void handleEarlyExits(VPlan &Plan, bool HasUncountableExit);