1; REQUIRES: asserts 2 3; RUN: opt -passes='loop-vectorize' -mtriple=x86_64-unknown-linux -S -debug %s 2>&1 | FileCheck %s 4 5target 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" 6 7target triple = "x86_64-unknown-linux" 8 9declare double @llvm.pow.f64(double, double) 10 11; Test case where the memory runtime checks and vector body is more expensive 12; than running the scalar loop. 13define void @test(double* nocapture %A, double* nocapture %B, double* nocapture %C, double* nocapture %D, double* nocapture %E) { 14 15; CHECK: Calculating cost of runtime checks: 16; CHECK-NEXT: 0 for {{.+}} = getelementptr double, double* %A, i64 16 17; CHECK-NEXT: 0 for {{.+}} = bitcast double* 18; CHECK-NEXT: 0 for {{.+}} = getelementptr double, double* %B, i64 16 19; CHECK-NEXT: 0 for {{.+}} = bitcast double* 20; CHECK-NEXT: 0 for {{.+}} = getelementptr double, double* %E, i64 16 21; CHECK-NEXT: 0 for {{.+}} = bitcast double* 22; CHECK-NEXT: 0 for {{.+}} = getelementptr double, double* %C, i64 16 23; CHECK-NEXT: 0 for {{.+}} = bitcast double* 24; CHECK-NEXT: 0 for {{.+}} = getelementptr double, double* %D, i64 16 25; CHECK-NEXT: 0 for {{.+}} = bitcast double* 26; CHECK-NEXT: 1 for {{.+}} = icmp ult i8* 27; CHECK-NEXT: 1 for {{.+}} = icmp ult i8* 28; CHECK-NEXT: 1 for {{.+}} = and i1 29; CHECK-NEXT: 1 for {{.+}} = icmp ult i8* 30; CHECK-NEXT: 1 for {{.+}} = icmp ult i8* 31; CHECK-NEXT: 1 for {{.+}} = and i1 32; CHECK-NEXT: 1 for {{.+}} = or i1 33; CHECK-NEXT: 1 for {{.+}} = icmp ult i8* 34; CHECK-NEXT: 1 for {{.+}} = icmp ult i8* 35; CHECK-NEXT: 1 for {{.+}} = and i1 36; CHECK-NEXT: 1 for {{.+}} = or i1 37; CHECK-NEXT: 1 for {{.+}} = icmp ult i8* 38; CHECK-NEXT: 1 for {{.+}} = icmp ult i8* 39; CHECK-NEXT: 1 for {{.+}} = and i1 40; CHECK-NEXT: 1 for {{.+}} = or i1 41; CHECK-NEXT: 1 for {{.+}} = icmp ult i8* 42; CHECK-NEXT: 1 for {{.+}} = icmp ult i8* 43; CHECK-NEXT: 1 for {{.+}} = and i1 44; CHECK-NEXT: 1 for {{.+}} = or i1 45; CHECK-NEXT: 1 for {{.+}} = icmp ult i8* 46; CHECK-NEXT: 1 for {{.+}} = icmp ult i8* 47; CHECK-NEXT: 1 for {{.+}} = and i1 48; CHECK-NEXT: 1 for {{.+}} = or i1 49; CHECK-NEXT: 1 for {{.+}} = icmp ult i8* 50; CHECK-NEXT: 1 for {{.+}} = icmp ult i8* 51; CHECK-NEXT: 1 for {{.+}} = and i1 52; CHECK-NEXT: 1 for {{.+}} = or i1 53; CHECK-NEXT: 1 for {{.+}} = icmp ult i8* 54; CHECK-NEXT: 1 for {{.+}} = icmp ult i8* 55; CHECK-NEXT: 1 for {{.+}} = and i1 56; CHECK-NEXT: 1 for {{.+}} = or i1 57; CHECK-NEXT: 1 for {{.+}} = icmp ult i8* 58; CHECK-NEXT: 1 for {{.+}} = icmp ult i8* 59; CHECK-NEXT: 1 for {{.+}} = and i1 60; CHECK-NEXT: 1 for {{.+}} = or i1 61; CHECK-NEXT: Total cost of runtime checks: 35 62 63; CHECK: LV: Vectorization is not beneficial: expected trip count < minimum profitable VF (16 < 70) 64; 65; CHECK-LABEL: @test( 66; CHECK-NEXT: entry: 67; CHECK-NEXT: br label %for.body 68; CHECK-NOT: vector.memcheck 69; CHECK-NOT: vector.body 70; 71entry: 72 br label %for.body 73 74for.body: 75 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] 76 %gep.A = getelementptr inbounds double, double* %A, i64 %iv 77 %l.A = load double, double* %gep.A, align 4 78 store double 0.0, double* %gep.A, align 4 79 %p.1 = call double @llvm.pow.f64(double %l.A, double 2.0) 80 81 %gep.B = getelementptr inbounds double, double* %B, i64 %iv 82 %l.B = load double, double* %gep.B, align 4 83 %p.2 = call double @llvm.pow.f64(double %l.B, double %p.1) 84 store double 0.0, double* %gep.B, align 4 85 86 %gep.C = getelementptr inbounds double, double* %C, i64 %iv 87 %l.C = load double, double* %gep.C, align 4 88 %p.3 = call double @llvm.pow.f64(double %p.1, double %l.C) 89 90 %gep.D = getelementptr inbounds double, double* %D, i64 %iv 91 %l.D = load double, double* %gep.D 92 %p.4 = call double @llvm.pow.f64(double %p.3, double %l.D) 93 %p.5 = call double @llvm.pow.f64(double %p.4, double %p.3) 94 %mul = fmul double 2.0, %p.5 95 %mul.2 = fmul double %mul, 2.0 96 %mul.3 = fmul double %mul, %mul.2 97 %gep.E = getelementptr inbounds double, double* %E, i64 %iv 98 store double %mul.3, double* %gep.E, align 4 99 %iv.next = add i64 %iv, 1 100 %exitcond = icmp eq i64 %iv.next, 16 101 br i1 %exitcond, label %for.end, label %for.body 102 103for.end: 104 ret void 105} 106