100 lines · plain
1; RUN: opt -passes="loop-vectorize" -mtriple=x86_64-unknown-linux -S -debug %s 2>&1 | FileCheck %s2; REQUIRES: asserts3 4target 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"5 6target triple = "x86_64-unknown-linux"7 8declare double @llvm.pow.f64(double, double)9 10; Test case where the memory runtime checks and vector body is more expensive11; than running the scalar loop.12define void @test(ptr nocapture %A, ptr nocapture %B, ptr nocapture %C, ptr nocapture %D, ptr nocapture %E) {13 14; CHECK: Calculating cost of runtime checks:15; CHECK-NEXT: 0 for {{.+}} = getelementptr i8, ptr %A, i64 12816; CHECK-NEXT: 0 for {{.+}} = getelementptr i8, ptr %B, i64 12817; CHECK-NEXT: 0 for {{.+}} = getelementptr i8, ptr %E, i64 12818; CHECK-NEXT: 0 for {{.+}} = getelementptr i8, ptr %C, i64 12819; CHECK-NEXT: 0 for {{.+}} = getelementptr i8, ptr %D, i64 12820; CHECK-NEXT: 1 for {{.+}} = icmp ult ptr21; CHECK-NEXT: 1 for {{.+}} = icmp ult ptr22; CHECK-NEXT: 1 for {{.+}} = and i123; CHECK-NEXT: 1 for {{.+}} = icmp ult ptr24; CHECK-NEXT: 1 for {{.+}} = icmp ult ptr25; CHECK-NEXT: 1 for {{.+}} = and i126; CHECK-NEXT: 1 for {{.+}} = or i127; CHECK-NEXT: 1 for {{.+}} = icmp ult ptr28; CHECK-NEXT: 1 for {{.+}} = icmp ult ptr29; CHECK-NEXT: 1 for {{.+}} = and i130; CHECK-NEXT: 1 for {{.+}} = or i131; CHECK-NEXT: 1 for {{.+}} = icmp ult ptr32; CHECK-NEXT: 1 for {{.+}} = icmp ult ptr33; CHECK-NEXT: 1 for {{.+}} = and i134; CHECK-NEXT: 1 for {{.+}} = or i135; CHECK-NEXT: 1 for {{.+}} = icmp ult ptr36; CHECK-NEXT: 1 for {{.+}} = icmp ult ptr37; CHECK-NEXT: 1 for {{.+}} = and i138; CHECK-NEXT: 1 for {{.+}} = or i139; CHECK-NEXT: 1 for {{.+}} = icmp ult ptr40; CHECK-NEXT: 1 for {{.+}} = icmp ult ptr41; CHECK-NEXT: 1 for {{.+}} = and i142; CHECK-NEXT: 1 for {{.+}} = or i143; CHECK-NEXT: 1 for {{.+}} = icmp ult ptr44; CHECK-NEXT: 1 for {{.+}} = icmp ult ptr45; CHECK-NEXT: 1 for {{.+}} = and i146; CHECK-NEXT: 1 for {{.+}} = or i147; CHECK-NEXT: 1 for {{.+}} = icmp ult ptr48; CHECK-NEXT: 1 for {{.+}} = icmp ult ptr49; CHECK-NEXT: 1 for {{.+}} = and i150; CHECK-NEXT: 1 for {{.+}} = or i151; CHECK-NEXT: 1 for {{.+}} = icmp ult ptr52; CHECK-NEXT: 1 for {{.+}} = icmp ult ptr53; CHECK-NEXT: 1 for {{.+}} = and i154; CHECK-NEXT: 1 for {{.+}} = or i155; CHECK-NEXT: Total cost of runtime checks: 3556 57; CHECK: LV: Vectorization is not beneficial: expected trip count < minimum profitable VF (16 < 24)58;59; CHECK-LABEL: @test(60; CHECK-NEXT: entry:61; CHECK-NEXT: br label %for.body62; CHECK-NOT: vector.memcheck63; CHECK-NOT: vector.body64;65entry:66 br label %for.body67 68for.body:69 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]70 %gep.A = getelementptr inbounds double, ptr %A, i64 %iv71 %l.A = load double, ptr %gep.A, align 472 store double 0.0, ptr %gep.A, align 473 %p.1 = call double @llvm.pow.f64(double %l.A, double 2.0)74 75 %gep.B = getelementptr inbounds double, ptr %B, i64 %iv76 %l.B = load double, ptr %gep.B, align 477 %p.2 = call double @llvm.pow.f64(double %l.B, double %p.1)78 store double 0.0, ptr %gep.B, align 479 80 %gep.C = getelementptr inbounds double, ptr %C, i64 %iv81 %l.C = load double, ptr %gep.C, align 482 %p.3 = call double @llvm.pow.f64(double %p.1, double %l.C)83 84 %gep.D = getelementptr inbounds double, ptr %D, i64 %iv85 %l.D = load double, ptr %gep.D86 %p.4 = call double @llvm.pow.f64(double %p.2, double %l.D)87 %p.5 = call double @llvm.pow.f64(double %p.4, double %p.3)88 %mul = fmul double 2.0, %p.589 %mul.2 = fmul double %mul, 2.090 %mul.3 = fmul double %mul, %mul.291 %gep.E = getelementptr inbounds double, ptr %E, i64 %iv92 store double %mul.3, ptr %gep.E, align 493 %iv.next = add i64 %iv, 194 %exitcond = icmp eq i64 %iv.next, 1695 br i1 %exitcond, label %for.end, label %for.body96 97for.end:98 ret void99}100