1 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" 2 ; RUN: opt < %s -bb-vectorize -bb-vectorize-req-chain-depth=3 -bb-vectorize-ignore-target-info -instcombine -gvn -S | FileCheck %s 3 4 declare double @llvm.fma.f64(double, double, double) 5 declare double @llvm.fmuladd.f64(double, double, double) 6 declare double @llvm.cos.f64(double) 7 declare double @llvm.powi.f64(double, i32) 8 9 ; Basic depth-3 chain with fma 10 define double @test1(double %A1, double %A2, double %B1, double %B2, double %C1, double %C2) { 11 %X1 = fsub double %A1, %B1 12 %X2 = fsub double %A2, %B2 13 %Y1 = call double @llvm.fma.f64(double %X1, double %A1, double %C1) 14 %Y2 = call double @llvm.fma.f64(double %X2, double %A2, double %C2) 15 %Z1 = fadd double %Y1, %B1 16 %Z2 = fadd double %Y2, %B2 17 %R = fmul double %Z1, %Z2 18 ret double %R 19 ; CHECK: @test1 20 ; CHECK: %X1.v.i1.1 = insertelement <2 x double> undef, double %B1, i32 0 21 ; CHECK: %X1.v.i1.2 = insertelement <2 x double> %X1.v.i1.1, double %B2, i32 1 22 ; CHECK: %X1.v.i0.1 = insertelement <2 x double> undef, double %A1, i32 0 23 ; CHECK: %X1.v.i0.2 = insertelement <2 x double> %X1.v.i0.1, double %A2, i32 1 24 ; CHECK: %X1 = fsub <2 x double> %X1.v.i0.2, %X1.v.i1.2 25 ; CHECK: %Y1.v.i2.1 = insertelement <2 x double> undef, double %C1, i32 0 26 ; CHECK: %Y1.v.i2.2 = insertelement <2 x double> %Y1.v.i2.1, double %C2, i32 1 27 ; CHECK: %Y1 = call <2 x double> @llvm.fma.v2f64(<2 x double> %X1, <2 x double> %X1.v.i0.2, <2 x double> %Y1.v.i2.2) 28 ; CHECK: %Z1 = fadd <2 x double> %Y1, %X1.v.i1.2 29 ; CHECK: %Z1.v.r1 = extractelement <2 x double> %Z1, i32 0 30 ; CHECK: %Z1.v.r2 = extractelement <2 x double> %Z1, i32 1 31 ; CHECK: %R = fmul double %Z1.v.r1, %Z1.v.r2 32 ; CHECK: ret double %R 33 } 34 35 ; Basic depth-3 chain with fmuladd 36 define double @test1a(double %A1, double %A2, double %B1, double %B2, double %C1, double %C2) { 37 %X1 = fsub double %A1, %B1 38 %X2 = fsub double %A2, %B2 39 %Y1 = call double @llvm.fmuladd.f64(double %X1, double %A1, double %C1) 40 %Y2 = call double @llvm.fmuladd.f64(double %X2, double %A2, double %C2) 41 %Z1 = fadd double %Y1, %B1 42 %Z2 = fadd double %Y2, %B2 43 %R = fmul double %Z1, %Z2 44 ret double %R 45 ; CHECK: @test1a 46 ; CHECK: %X1.v.i1.1 = insertelement <2 x double> undef, double %B1, i32 0 47 ; CHECK: %X1.v.i1.2 = insertelement <2 x double> %X1.v.i1.1, double %B2, i32 1 48 ; CHECK: %X1.v.i0.1 = insertelement <2 x double> undef, double %A1, i32 0 49 ; CHECK: %X1.v.i0.2 = insertelement <2 x double> %X1.v.i0.1, double %A2, i32 1 50 ; CHECK: %X1 = fsub <2 x double> %X1.v.i0.2, %X1.v.i1.2 51 ; CHECK: %Y1.v.i2.1 = insertelement <2 x double> undef, double %C1, i32 0 52 ; CHECK: %Y1.v.i2.2 = insertelement <2 x double> %Y1.v.i2.1, double %C2, i32 1 53 ; CHECK: %Y1 = call <2 x double> @llvm.fmuladd.v2f64(<2 x double> %X1, <2 x double> %X1.v.i0.2, <2 x double> %Y1.v.i2.2) 54 ; CHECK: %Z1 = fadd <2 x double> %Y1, %X1.v.i1.2 55 ; CHECK: %Z1.v.r1 = extractelement <2 x double> %Z1, i32 0 56 ; CHECK: %Z1.v.r2 = extractelement <2 x double> %Z1, i32 1 57 ; CHECK: %R = fmul double %Z1.v.r1, %Z1.v.r2 58 ; CHECK: ret double %R 59 } 60 61 ; Basic depth-3 chain with cos 62 define double @test2(double %A1, double %A2, double %B1, double %B2) { 63 %X1 = fsub double %A1, %B1 64 %X2 = fsub double %A2, %B2 65 %Y1 = call double @llvm.cos.f64(double %X1) 66 %Y2 = call double @llvm.cos.f64(double %X2) 67 %Z1 = fadd double %Y1, %B1 68 %Z2 = fadd double %Y2, %B2 69 %R = fmul double %Z1, %Z2 70 ret double %R 71 ; CHECK: @test2 72 ; CHECK: %X1.v.i1.1 = insertelement <2 x double> undef, double %B1, i32 0 73 ; CHECK: %X1.v.i1.2 = insertelement <2 x double> %X1.v.i1.1, double %B2, i32 1 74 ; CHECK: %X1.v.i0.1 = insertelement <2 x double> undef, double %A1, i32 0 75 ; CHECK: %X1.v.i0.2 = insertelement <2 x double> %X1.v.i0.1, double %A2, i32 1 76 ; CHECK: %X1 = fsub <2 x double> %X1.v.i0.2, %X1.v.i1.2 77 ; CHECK: %Y1 = call <2 x double> @llvm.cos.v2f64(<2 x double> %X1) 78 ; CHECK: %Z1 = fadd <2 x double> %Y1, %X1.v.i1.2 79 ; CHECK: %Z1.v.r1 = extractelement <2 x double> %Z1, i32 0 80 ; CHECK: %Z1.v.r2 = extractelement <2 x double> %Z1, i32 1 81 ; CHECK: %R = fmul double %Z1.v.r1, %Z1.v.r2 82 ; CHECK: ret double %R 83 } 84 85 ; Basic depth-3 chain with powi 86 define double @test3(double %A1, double %A2, double %B1, double %B2, i32 %P) { 87 88 %X1 = fsub double %A1, %B1 89 %X2 = fsub double %A2, %B2 90 %Y1 = call double @llvm.powi.f64(double %X1, i32 %P) 91 %Y2 = call double @llvm.powi.f64(double %X2, i32 %P) 92 %Z1 = fadd double %Y1, %B1 93 %Z2 = fadd double %Y2, %B2 94 %R = fmul double %Z1, %Z2 95 ret double %R 96 ; CHECK: @test3 97 ; CHECK: %X1.v.i1.1 = insertelement <2 x double> undef, double %B1, i32 0 98 ; CHECK: %X1.v.i1.2 = insertelement <2 x double> %X1.v.i1.1, double %B2, i32 1 99 ; CHECK: %X1.v.i0.1 = insertelement <2 x double> undef, double %A1, i32 0 100 ; CHECK: %X1.v.i0.2 = insertelement <2 x double> %X1.v.i0.1, double %A2, i32 1 101 ; CHECK: %X1 = fsub <2 x double> %X1.v.i0.2, %X1.v.i1.2 102 ; CHECK: %Y1 = call <2 x double> @llvm.powi.v2f64(<2 x double> %X1, i32 %P) 103 ; CHECK: %Z1 = fadd <2 x double> %Y1, %X1.v.i1.2 104 ; CHECK: %Z1.v.r1 = extractelement <2 x double> %Z1, i32 0 105 ; CHECK: %Z1.v.r2 = extractelement <2 x double> %Z1, i32 1 106 ; CHECK: %R = fmul double %Z1.v.r1, %Z1.v.r2 107 ; CHECK: ret double %R 108 } 109 110 ; Basic depth-3 chain with powi (different powers: should not vectorize) 111 define double @test4(double %A1, double %A2, double %B1, double %B2, i32 %P) { 112 113 %X1 = fsub double %A1, %B1 114 %X2 = fsub double %A2, %B2 115 %P2 = add i32 %P, 1 116 %Y1 = call double @llvm.powi.f64(double %X1, i32 %P) 117 %Y2 = call double @llvm.powi.f64(double %X2, i32 %P2) 118 %Z1 = fadd double %Y1, %B1 119 %Z2 = fadd double %Y2, %B2 120 %R = fmul double %Z1, %Z2 121 ret double %R 122 ; CHECK: @test4 123 ; CHECK-NOT: <2 x double> 124 ; CHECK: ret double %R 125 } 126 127 ; CHECK: declare <2 x double> @llvm.fma.v2f64(<2 x double>, <2 x double>, <2 x double>) #0 128 ; CHECK: declare <2 x double> @llvm.fmuladd.v2f64(<2 x double>, <2 x double>, <2 x double>) #0 129 ; CHECK: declare <2 x double> @llvm.cos.v2f64(<2 x double>) #1 130 ; CHECK: declare <2 x double> @llvm.powi.v2f64(<2 x double>, i32) #1 131 132 ; CHECK: attributes #0 = { nounwind readnone } 133 ; CHECK: attributes #1 = { nounwind readonly } 134
