185 lines · plain
1; RUN: llc < %s -O3 -mtriple=arm64-apple-ios -mattr=+fullfp16 | FileCheck %s2; RUN: llc < %s -O3 -mtriple=arm64-apple-ios -fp-contract=fast -mattr=+fullfp16 | FileCheck %s3 4define void @foo_2d(ptr %src) {5entry:6 %arrayidx1 = getelementptr inbounds double, ptr %src, i64 57 %arrayidx2 = getelementptr inbounds double, ptr %src, i64 118 br label %for.body9 10; CHECK-LABEL: %for.body11; CHECK: fmls.2d {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}12; CHECK: fmls.2d {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}[0]13; CHECK: fmsub {{d[0-9]+}}, {{d[0-9]+}}, {{d[0-9]+}}, {{d[0-9]+}}14for.body: ; preds = %for.body, %entry15 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]16 %indvars.iv.next = sub nuw nsw i64 %indvars.iv, 117 %arrayidx3 = getelementptr inbounds double, ptr %src, i64 %indvars.iv.next18 %tmp1 = load double, ptr %arrayidx3, align 819 %add = fadd fast double %tmp1, %tmp120 %mul = fmul fast double %add, %add21 %e1 = insertelement <2 x double> undef, double %add, i32 022 %e2 = insertelement <2 x double> %e1, double %add, i32 123 %sub2 = fsub fast <2 x double> %e2, <double 3.000000e+00, double -3.000000e+00>24 %e3 = insertelement <2 x double> undef, double %mul, i32 025 %e4 = insertelement <2 x double> %e3, double %mul, i32 126 %mul2 = fmul fast <2 x double> %sub2,<double 3.000000e+00, double -3.000000e+00>27 %e5 = insertelement <2 x double> undef, double %add, i32 028 %e6 = insertelement <2 x double> %e5, double %add, i32 129 %sub3 = fsub fast <2 x double> <double 3.000000e+00, double -3.000000e+00>, %mul230 %mulx = fmul fast <2 x double> %sub2, %e231 %subx = fsub fast <2 x double> %e4, %mulx32 %e7 = insertelement <2 x double> undef, double %mul, i32 033 %e8 = insertelement <2 x double> %e7, double %mul, i32 134 %e9 = fmul fast <2 x double> %subx, %sub335 store <2 x double> %e9, ptr %arrayidx1, align 836 %e10 = extractelement <2 x double> %sub3, i32 037 %mul3 = fmul fast double %mul, %e1038 %sub4 = fsub fast double %mul, %mul339 store double %sub4, ptr %arrayidx2, align 840 %exitcond = icmp eq i64 %indvars.iv.next, 2541 br i1 %exitcond, label %for.end, label %for.body42 43for.end: ; preds = %for.body44 ret void45}46define void @foo_2s(ptr %src) {47entry:48 %arrayidx1 = getelementptr inbounds float, ptr %src, i64 549 %arrayidx2 = getelementptr inbounds float, ptr %src, i64 1150 br label %for.body51 52; CHECK-LABEL: %for.body53; CHECK: fmls.2s {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}54; CHECK: fmls.2s {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}[0]55; CHECK: fmsub {{s[0-9]+}}, {{s[0-9]+}}, {{s[0-9]+}}, {{s[0-9]+}}56for.body: ; preds = %for.body, %entry57 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]58 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 159 %arrayidx3 = getelementptr inbounds float, ptr %src, i64 %indvars.iv.next60 %tmp1 = load float, ptr %arrayidx3, align 861 %add = fadd fast float %tmp1, %tmp162 %mul = fmul fast float %add, %add63 %e1 = insertelement <2 x float> undef, float %add, i32 064 %e2 = insertelement <2 x float> %e1, float %add, i32 165 %add2 = fsub fast <2 x float> %e2, <float 3.000000e+00, float -3.000000e+00>66 %e3 = insertelement <2 x float> undef, float %mul, i32 067 %e4 = insertelement <2 x float> %e3, float %mul, i32 168 %mul2 = fmul fast <2 x float> %add2,<float 3.000000e+00, float -3.000000e+00>69 %e5 = insertelement <2 x float> undef, float %add, i32 070 %e6 = insertelement <2 x float> %e5, float %add, i32 171 %add3 = fsub fast <2 x float> <float 3.000000e+00, float -3.000000e+00>, %mul272 %mulx = fmul fast <2 x float> %add2, %e273 %addx = fsub fast <2 x float> %e4, %mulx74 %e7 = insertelement <2 x float> undef, float %mul, i32 075 %e8 = insertelement <2 x float> %e7, float %mul, i32 176 %e9 = fmul fast <2 x float> %addx, %add377 store <2 x float> %e9, ptr %arrayidx1, align 878 %e10 = extractelement <2 x float> %add3, i32 079 %mul3 = fmul fast float %mul, %e1080 %add4 = fsub fast float %mul, %mul381 store float %add4, ptr %arrayidx2, align 882 %exitcond = icmp eq i64 %indvars.iv.next, 2583 br i1 %exitcond, label %for.end, label %for.body84 85for.end: ; preds = %for.body86 ret void87}88define void @foo_4s(ptr %src) {89entry:90 %arrayidx1 = getelementptr inbounds float, ptr %src, i64 591 %arrayidx2 = getelementptr inbounds float, ptr %src, i64 1192 br label %for.body93 94; CHECK-LABEL: %for.body95; CHECK: fmls.4s {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}96; CHECK: fmls.4s {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}[0]97for.body: ; preds = %for.body, %entry98 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]99 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1100 %arrayidx3 = getelementptr inbounds float, ptr %src, i64 %indvars.iv.next101 %tmp1 = load float, ptr %arrayidx3, align 8102 %add = fadd fast float %tmp1, %tmp1103 %mul = fmul fast float %add, %add104 %e1 = insertelement <4 x float> undef, float %add, i32 0105 %e2 = insertelement <4 x float> %e1, float %add, i32 1106 %add2 = fadd fast <4 x float> %e2, <float 3.000000e+00, float -3.000000e+00, float 5.000000e+00, float 7.000000e+00>107 %e3 = insertelement <4 x float> undef, float %mul, i32 0108 %e4 = insertelement <4 x float> %e3, float %mul, i32 1109 %mul2 = fmul fast <4 x float> %add2,<float 3.000000e+00, float -3.000000e+00, float 5.000000e+00, float 7.000000e+00>110 %e5 = insertelement <4 x float> undef, float %add, i32 0111 %e6 = insertelement <4 x float> %e5, float %add, i32 1112 %add3 = fsub fast <4 x float> <float 3.000000e+00, float -3.000000e+00, float 5.000000e+00, float 7.000000e+00> , %mul2113 %mulx = fmul fast <4 x float> %add2, %e2114 %addx = fsub fast <4 x float> %e4, %mulx115 %e7 = insertelement <4 x float> undef, float %mul, i32 0116 %e8 = insertelement <4 x float> %e7, float %mul, i32 1117 %e9 = fmul fast <4 x float> %addx, %add3118 store <4 x float> %e9, ptr %arrayidx1, align 8119 %e10 = extractelement <4 x float> %add3, i32 0120 %mul3 = fmul fast float %mul, %e10121 store float %mul3, ptr %arrayidx2, align 8122 %exitcond = icmp eq i64 %indvars.iv.next, 25123 br i1 %exitcond, label %for.end, label %for.body124 125for.end: ; preds = %for.body126 ret void127}128 129; CHECK-LABEL: test0:130; CHECK: fnmadd h0, h0, h1, h2131define half @test0(half %a, half %b, half %c) {132entry:133 %0 = fmul contract half %a, %b134 %mul = fsub contract half -0.000000e+00, %0135 %sub1 = fsub contract half %mul, %c136 ret half %sub1137}138 139; CHECK-LABEL: test1:140; CHECK: fnmadd s0, s0, s1, s2141define float @test1(float %a, float %b, float %c) {142entry:143 %0 = fmul contract float %a, %b144 %mul = fsub contract float -0.000000e+00, %0145 %sub1 = fsub contract float %mul, %c146 ret float %sub1147}148 149; CHECK-LABEL: test2:150; CHECK: fnmadd d0, d0, d1, d2151define double @test2(double %a, double %b, double %c) {152entry:153 %0 = fmul contract double %a, %b154 %mul = fsub contract double -0.000000e+00, %0155 %sub1 = fsub contract double %mul, %c156 ret double %sub1157}158 159; CHECK-LABEL: test3:160; CHECK: fnmadd h0, h0, h1, h2161define half @test3(half %0, half %1, half %2) {162 %4 = fneg fast half %0163 %5 = fmul fast half %4, %1164 %6 = fsub fast half %5, %2165 ret half %6166}167 168; CHECK-LABEL: test4:169; CHECK: fnmadd s0, s0, s1, s2170define float @test4(float %0, float %1, float %2) {171 %4 = fneg fast float %0172 %5 = fmul fast float %4, %1173 %6 = fsub fast float %5, %2174 ret float %6175}176 177; CHECK-LABEL: test5:178; CHECK: fnmadd d0, d0, d1, d2179define double @test5(double %0, double %1, double %2) {180 %4 = fneg fast double %0181 %5 = fmul fast double %4, %1182 %6 = fsub fast double %5, %2183 ret double %6184}185