Index: llvm/include/llvm/Analysis/TargetTransformInfoImpl.h =================================================================== --- llvm/include/llvm/Analysis/TargetTransformInfoImpl.h +++ llvm/include/llvm/Analysis/TargetTransformInfoImpl.h @@ -505,6 +505,11 @@ case Intrinsic::is_constant: case Intrinsic::lifetime_start: case Intrinsic::lifetime_end: + case Intrinsic::noalias_decl: + case Intrinsic::noalias: + case Intrinsic::provenance_noalias: + case Intrinsic::noalias_arg_guard: + case Intrinsic::noalias_copy_guard: case Intrinsic::objectsize: case Intrinsic::ptr_annotation: case Intrinsic::var_annotation: Index: llvm/lib/Analysis/VectorUtils.cpp =================================================================== --- llvm/lib/Analysis/VectorUtils.cpp +++ llvm/lib/Analysis/VectorUtils.cpp @@ -119,7 +119,8 @@ if (isTriviallyVectorizable(ID) || ID == Intrinsic::lifetime_start || ID == Intrinsic::lifetime_end || ID == Intrinsic::assume || - ID == Intrinsic::sideeffect) + ID == Intrinsic::sideeffect || ID == Intrinsic::noalias || + ID == Intrinsic::provenance_noalias) return ID; return Intrinsic::not_intrinsic; } Index: llvm/lib/Transforms/Vectorize/LoopVectorize.cpp =================================================================== --- llvm/lib/Transforms/Vectorize/LoopVectorize.cpp +++ llvm/lib/Transforms/Vectorize/LoopVectorize.cpp @@ -3302,11 +3302,6 @@ if (VF == 1) return ScalarCallCost; - // Compute corresponding vector type for return value and arguments. - Type *RetTy = ToVectorTy(ScalarRetTy, VF); - for (Type *ScalarTy : ScalarTys) - Tys.push_back(ToVectorTy(ScalarTy, VF)); - // Compute costs of unpacking argument values for the scalar calls and // packing the return values to a vector. unsigned ScalarizationCost = getScalarizationOverhead(CI, VF); @@ -3322,6 +3317,11 @@ if (!TLI || CI->isNoBuiltin() || !VecFunc) return Cost; + // Compute corresponding vector type for return value and arguments. + Type *RetTy = ToVectorTy(ScalarRetTy, VF); + for (Type *ScalarTy : ScalarTys) + Tys.push_back(ToVectorTy(ScalarTy, VF)); + // If the corresponding vector cost is cheaper, return its cost. unsigned VectorCallCost = TTI.getCallInstrCost(nullptr, RetTy, Tys, TTI::TCK_RecipThroughput); @@ -6970,7 +6970,8 @@ Intrinsic::ID ID = getVectorIntrinsicIDForCall(CI, TLI); if (ID && (ID == Intrinsic::assume || ID == Intrinsic::lifetime_end || - ID == Intrinsic::lifetime_start || ID == Intrinsic::sideeffect)) + ID == Intrinsic::lifetime_start || ID == Intrinsic::sideeffect || + ID == Intrinsic::noalias || ID == Intrinsic::provenance_noalias)) return nullptr; auto willWiden = [&](unsigned VF) -> bool { Index: llvm/test/Transforms/LoopVectorize/noalias.ll =================================================================== --- /dev/null +++ llvm/test/Transforms/LoopVectorize/noalias.ll @@ -0,0 +1,90 @@ +; RUN: opt -S -loop-vectorize -force-vector-width=2 -force-vector-interleave=1 < %s | FileCheck %s + +target datalayout = "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" + +; Make sure we can vectorize loops which contain noalias intrinsics + + +; A not-used llvm.noalias should not interfere +; CHECK-LABEL: @test_noalias_not_connected( +; CHECK: @llvm.noalias.p0i32 +; CHECK: store <2 x i32> +; CHECK: ret +define void @test_noalias_not_connected(i32 *%d) { +entry: + br label %for.body + +for.body: + %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] + %d2 = call i32* @llvm.noalias.p0i32.p0i8.p0p0i32.i32(i32* %d, i8* null, i32** null, i32 0, metadata !1) + %arrayidx = getelementptr inbounds i32, i32* %d, i64 %indvars.iv + %v1 = load i32, i32* %arrayidx, align 8 + store i32 100, i32* %arrayidx, align 8 + %indvars.iv.next = add i64 %indvars.iv, 1 + %lftr.wideiv = trunc i64 %indvars.iv.next to i32 + %exitcond = icmp ne i32 %lftr.wideiv, 128 + br i1 %exitcond, label %for.body, label %for.end + +for.end: + ret void +} + +; A used llvm.noalias should block vectorization. +; CHECK-LABEL: @test_noalias_connected( +; CHECK: @llvm.noalias.p0i32 +; CHECK-NOT: store <2 x i32> +; CHECK: ret +define void @test_noalias_connected(i32 *%d) { +entry: + br label %for.body + +for.body: + %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] + %d2 = call i32* @llvm.noalias.p0i32.p0i8.p0p0i32.i32(i32* %d, i8* null, i32** null, i32 0, metadata !1) + %arrayidx = getelementptr inbounds i32, i32* %d2, i64 %indvars.iv + %v1 = load i32, i32* %arrayidx, align 8 + store i32 100, i32* %arrayidx, align 8 + %indvars.iv.next = add i64 %indvars.iv, 1 + %lftr.wideiv = trunc i64 %indvars.iv.next to i32 + %exitcond = icmp ne i32 %lftr.wideiv, 128 + br i1 %exitcond, label %for.body, label %for.end + +for.end: + ret void +} + +; A used llvm.provenance.noalias should NOT block vectorization. +; CHECK-LABEL: @test_provenance.noalias( +; CHECK: @llvm.provenance.noalias.p0i32 +; NOTE: the ptr_provenance is omitted +; CHECK: store <2 x i32> , <2 x i32>* {{%[0-9.a-zA-Z]*}}, align 8 +; CHECK: ret + +define void @test_provenance.noalias(i32 *%d) { +entry: + br label %for.body + +for.body: + %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] + %prov.d = call i32* @llvm.provenance.noalias.p0i32.p0i8.p0p0i32.p0p0i32.i32(i32* %d, i8* null, i32** null, i32** null, i32 0, metadata !1) + %arrayidx = getelementptr inbounds i32, i32* %d, i64 %indvars.iv + %v1 = load i32, i32* %arrayidx, ptr_provenance i32* %prov.d, align 8 + store i32 100, i32* %arrayidx, ptr_provenance i32* %prov.d, align 8 + %indvars.iv.next = add i64 %indvars.iv, 1 + %lftr.wideiv = trunc i64 %indvars.iv.next to i32 + %exitcond = icmp ne i32 %lftr.wideiv, 128 + br i1 %exitcond, label %for.body, label %for.end + +for.end: + ret void +} + +declare i8* @llvm.noalias.decl.p0i8.p0p0i32.i32(i8**, i32, metadata) argmemonly nounwind +declare i32* @llvm.noalias.p0i32.p0i8.p0p0i32.i32(i32*, i8*, i32**, i32, metadata ) nounwind +declare i32* @llvm.provenance.noalias.p0i32.p0i8.p0p0i32.p0p0i32.i32(i32*, i8*, i32**, i32**, i32, metadata ) nounwind + + +;declare i32* @llvm.noalias.p0i32(i32*, metadata) nounwind argmemonly + +!0 = !{!0, !"some domain"} +!1 = !{!1, !0, !"some scope"}