diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp --- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp +++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp @@ -8759,8 +8759,9 @@ } SmallPtrSet DeadInstructions; - VPlanTransforms::VPInstructionsToVPRecipes( - OrigLoop, Plan, Legal->getInductionVars(), DeadInstructions); + VPlanTransforms::VPInstructionsToVPRecipes(*PSE.getSE(), OrigLoop, Plan, + Legal->getInductionVars(), + DeadInstructions); return Plan; } diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.h b/llvm/lib/Transforms/Vectorize/VPlanTransforms.h --- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.h +++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.h @@ -23,7 +23,7 @@ /// Replaces the VPInstructions in \p Plan with corresponding /// widen recipes. static void VPInstructionsToVPRecipes( - Loop *OrigLoop, VPlanPtr &Plan, + ScalarEvolution &SE, Loop *OrigLoop, VPlanPtr &Plan, LoopVectorizationLegality::InductionList &Inductions, SmallPtrSetImpl &DeadInstructions); }; diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp --- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp +++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp @@ -17,7 +17,7 @@ using namespace llvm; void VPlanTransforms::VPInstructionsToVPRecipes( - Loop *OrigLoop, VPlanPtr &Plan, + ScalarEvolution &SE, Loop *OrigLoop, VPlanPtr &Plan, LoopVectorizationLegality::InductionList &Inductions, SmallPtrSetImpl &DeadInstructions) { @@ -71,6 +71,11 @@ } else if (GetElementPtrInst *GEP = dyn_cast(Inst)) { NewRecipe = new VPWidenGEPRecipe( GEP, Plan->mapToVPValues(GEP->operands()), OrigLoop); + } else if (SelectInst *SI = dyn_cast(Inst)) { + bool InvariantCond = + SE.isLoopInvariant(SE.getSCEV(SI->getOperand(0)), OrigLoop); + NewRecipe = new VPWidenSelectRecipe( + *SI, Plan->mapToVPValues(SI->operands()), InvariantCond); } else { NewRecipe = new VPWidenRecipe(*Inst, Plan->mapToVPValues(Inst->operands())); diff --git a/llvm/test/Transforms/LoopVectorize/vplan-widen-select-instruction.ll b/llvm/test/Transforms/LoopVectorize/vplan-widen-select-instruction.ll new file mode 100644 --- /dev/null +++ b/llvm/test/Transforms/LoopVectorize/vplan-widen-select-instruction.ll @@ -0,0 +1,144 @@ +; NOTE: Assertions have been autogenerated by utils/update_test_checks.py +; RUN: opt -loop-vectorize -force-vector-width=4 -enable-vplan-native-path -S %s | FileCheck %s + +; Test that VPlan native path is able to widen select instruction in the +; innermost loop under different conditions when outer loop is marked to be +; vectorized. These conditions include following: +; * Inner and outer loop invariant select condition +; * Select condition depending on outer loop iteration variable. +; * Select condidition depending on inner loop iteration variable. +; * Select conditition depending on both outer and inner loop iteration +; variables. + +define void @loop_invariant_select(double* noalias nocapture %c.out, i1 %select, double %a, double %b) { +; CHECK-LABEL: @loop_invariant_select( +; CHECK: %{{.*}} = select i1 %select, <4 x double> %{{.*}}, <4 x double> %{{.*}} +entry: + br label %for1.header + +for1.header: + %indvar1 = phi i64 [ 0, %entry ], [ %indvar11, %for1.latch ] + br label %for2.header + +for2.header: + %indvar2 = phi i64 [ 0, %for1.header ], [ %indvar21, %for2.header ] + %a.reduction = phi double [ %a, %for1.header ], [ %a.reduction1, %for2.header ] + ; Select condition is loop invariant for both inner and outer loop. + %select.b = select i1 %select, double %a, double %b + %a.reduction1 = fadd double %select.b, %a.reduction + %indvar21 = add nuw nsw i64 %indvar2, 1 + %for2.cond = icmp eq i64 %indvar21, 10000 + br i1 %for2.cond, label %for1.latch, label %for2.header + +for1.latch: + %c.ptr = getelementptr inbounds double, double* %c.out, i64 %indvar1 + store double %a.reduction1, double* %c.ptr, align 8 + %indvar11 = add nuw nsw i64 %indvar1, 1 + %for1.cond = icmp eq i64 %indvar11, 1000 + br i1 %for1.cond, label %exit, label %for1.header, !llvm.loop !0 + +exit: + ret void +} + +define void @outer_loop_dependant_select(double* noalias nocapture %c.out, double %a, double %b) { +; CHECK-LABEL: @outer_loop_dependant_select( +; CHECK: [[TMP0:%.*]] = trunc <4 x i64> %{{.*}} to <4 x i1> +; CHECK-NEXT: %{{.*}} = select <4 x i1> [[TMP0]], <4 x double> %{{.*}} <4 x double> %{{.*}} +entry: + br label %for1.header + +for1.header: + %indvar1 = phi i64 [ 0, %entry ], [ %indvar11, %for1.latch ] + br label %for2.header + +for2.header: + %indvar2 = phi i64 [ 0, %for1.header ], [ %indvar21, %for2.header ] + %a.reduction = phi double [ %a, %for1.header ], [ %a.reduction1, %for2.header ] + %select = trunc i64 %indvar1 to i1 + ; Select condition only depends on outer loop iteration variable. + %select.b = select i1 %select, double %a, double %b + %a.reduction1 = fadd double %select.b, %a.reduction + %indvar21 = add nuw nsw i64 %indvar2, 1 + %for2.cond = icmp eq i64 %indvar21, 10000 + br i1 %for2.cond, label %for1.latch, label %for2.header + +for1.latch: + %c.ptr = getelementptr inbounds double, double* %c.out, i64 %indvar1 + store double %a.reduction1, double* %c.ptr, align 8 + %indvar11 = add nuw nsw i64 %indvar1, 1 + %for1.cond = icmp eq i64 %indvar11, 1000 + br i1 %for1.cond, label %exit, label %for1.header, !llvm.loop !0 + +exit: + ret void +} + +define void @inner_loop_dependant_select(double* noalias nocapture %c.out, double %a, double %b) { +; CHECK-LABEL: @inner_loop_dependant_select( +; CHECK: [[TMP0:%.*]] = trunc <4 x i64> %{{.*}} to <4 x i1> +; CHECK-NEXT: %{{.*}} = select <4 x i1> [[TMP0]], <4 x double> %{{.*}}, <4 x double> %{{.*}} +entry: + br label %for1.header + +for1.header: + %indvar1 = phi i64 [ 0, %entry ], [ %indvar11, %for1.latch ] + br label %for2.header + +for2.header: + %indvar2 = phi i64 [ 0, %for1.header ], [ %indvar21, %for2.header ] + %a.reduction = phi double [ %a, %for1.header ], [ %a.reduction1, %for2.header ] + %select = trunc i64 %indvar2 to i1 + ; Select condition only depends on inner loop iteration variable. + %select.b = select i1 %select, double %a, double %b + %a.reduction1 = fadd double %select.b, %a.reduction + %indvar21 = add nuw nsw i64 %indvar2, 1 + %for2.cond = icmp eq i64 %indvar21, 10000 + br i1 %for2.cond, label %for1.latch, label %for2.header + +for1.latch: + %c.ptr = getelementptr inbounds double, double* %c.out, i64 %indvar1 + store double %a.reduction1, double* %c.ptr, align 8 + %indvar11 = add nuw nsw i64 %indvar1, 1 + %for1.cond = icmp eq i64 %indvar11, 1000 + br i1 %for1.cond, label %exit, label %for1.header, !llvm.loop !0 + +exit: + ret void +} + +define void @outer_and_inner_loop_dependant_select(double* noalias nocapture %c.out, double %a, double %b) { +; CHECK-LABEL: @outer_and_inner_loop_dependant_select( +; CHECK: [[TMP0:%.*]] = trunc <4 x i64> %{{.*}} to <4 x i1> +; CHECK-NEXT: %{{.*}} = select <4 x i1> [[TMP0]], <4 x double> %{{.*}} <4 x double> %{{.*}} +entry: + br label %for1.header + +for1.header: + %indvar1 = phi i64 [ 0, %entry ], [ %indvar11, %for1.latch ] + br label %for2.header + +for2.header: + %indvar2 = phi i64 [ 0, %for1.header ], [ %indvar21, %for2.header ] + %a.reduction = phi double [ %a, %for1.header ], [ %a.reduction1, %for2.header ] + %sum = add nuw nsw i64 %indvar1, %indvar2 + %select = trunc i64 %sum to i1 + ; Select condition depends on both inner and outer loop iteration variables. + %select.b = select i1 %select, double %a, double %b + %a.reduction1 = fadd double %select.b, %a.reduction + %indvar21 = add nuw nsw i64 %indvar2, 1 + %for2.cond = icmp eq i64 %indvar21, 10000 + br i1 %for2.cond, label %for1.latch, label %for2.header + +for1.latch: + %c.ptr = getelementptr inbounds double, double* %c.out, i64 %indvar1 + store double %a.reduction1, double* %c.ptr, align 8 + %indvar11 = add nuw nsw i64 %indvar1, 1 + %for1.cond = icmp eq i64 %indvar11, 1000 + br i1 %for1.cond, label %exit, label %for1.header, !llvm.loop !0 + +exit: + ret void +} +!0 = distinct !{!0, !1} +!1 = !{!"llvm.loop.vectorize.enable", i1 true} diff --git a/llvm/unittests/Transforms/Vectorize/VPlanHCFGTest.cpp b/llvm/unittests/Transforms/Vectorize/VPlanHCFGTest.cpp --- a/llvm/unittests/Transforms/Vectorize/VPlanHCFGTest.cpp +++ b/llvm/unittests/Transforms/Vectorize/VPlanHCFGTest.cpp @@ -130,8 +130,8 @@ LoopVectorizationLegality::InductionList Inductions; SmallPtrSet DeadInstructions; - VPlanTransforms::VPInstructionsToVPRecipes(LI->getLoopFor(LoopHeader), Plan, - Inductions, DeadInstructions); + VPlanTransforms::VPInstructionsToVPRecipes( + *SE, LI->getLoopFor(LoopHeader), Plan, Inductions, DeadInstructions); } TEST_F(VPlanHCFGTest, testVPInstructionToVPRecipesInner) { @@ -160,8 +160,8 @@ LoopVectorizationLegality::InductionList Inductions; SmallPtrSet DeadInstructions; - VPlanTransforms::VPInstructionsToVPRecipes(LI->getLoopFor(LoopHeader), Plan, - Inductions, DeadInstructions); + VPlanTransforms::VPInstructionsToVPRecipes( + *SE, LI->getLoopFor(LoopHeader), Plan, Inductions, DeadInstructions); VPBlockBase *Entry = Plan->getEntry()->getEntryBasicBlock(); EXPECT_NE(nullptr, Entry->getSingleSuccessor()); diff --git a/llvm/unittests/Transforms/Vectorize/VPlanTestBase.h b/llvm/unittests/Transforms/Vectorize/VPlanTestBase.h --- a/llvm/unittests/Transforms/Vectorize/VPlanTestBase.h +++ b/llvm/unittests/Transforms/Vectorize/VPlanTestBase.h @@ -17,6 +17,7 @@ #include "llvm/Analysis/AssumptionCache.h" #include "llvm/Analysis/BasicAliasAnalysis.h" #include "llvm/Analysis/LoopInfo.h" +#include "llvm/Analysis/TargetLibraryInfo.h" #include "llvm/AsmParser/Parser.h" #include "llvm/IR/Dominators.h" #include "llvm/Support/SourceMgr.h" @@ -28,12 +29,23 @@ /// given loop entry block. class VPlanTestBase : public testing::Test { protected: + TargetLibraryInfoImpl TLII; + TargetLibraryInfo TLI; + DataLayout DL; + std::unique_ptr Ctx; std::unique_ptr M; std::unique_ptr LI; std::unique_ptr DT; + std::unique_ptr AC; + std::unique_ptr SE; - VPlanTestBase() : Ctx(new LLVMContext) {} + VPlanTestBase() + : TLII(), TLI(TLII), + DL("e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-" + "f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:" + "16:32:64-S128"), + Ctx(new LLVMContext) {} Module &parseModule(const char *ModuleString) { SMDiagnostic Err; @@ -45,6 +57,8 @@ void doAnalysis(Function &F) { DT.reset(new DominatorTree(F)); LI.reset(new LoopInfo(*DT)); + AC.reset(new AssumptionCache(F)); + SE.reset(new ScalarEvolution(F, TLI, *AC, *DT, *LI)); } VPlanPtr buildHCFG(BasicBlock *LoopHeader) {