260 lines · plain
1; RUN: llc < %s -O=3 -mtriple=arm64-apple-ios -mcpu=cyclone -mattr=+fullfp16 -verify-machineinstrs | FileCheck %s2 3define void @foo_2d(ptr %src) {4; CHECK-LABEL: %entry5; CHECK: fmul {{d[0-9]+}}, {{d[0-9]+}}, {{d[0-9]+}}6; CHECK: fmadd {{d[0-9]+}}, {{d[0-9]+}}, {{d[0-9]+}}, {{d[0-9]+}}7entry:8 %arrayidx1 = getelementptr inbounds double, ptr %src, i64 59 %arrayidx2 = getelementptr inbounds double, ptr %src, i64 1110 %tmp1 = load double, ptr %arrayidx2, align 811 %tmp2 = load double, ptr %arrayidx1, align 812 %fmul = fmul fast double %tmp1, %tmp113 %fmul2 = fmul fast double %tmp2, 0x3F94AFD6A052BF5B14 %fadd = fadd fast double %fmul, %fmul215 br label %for.body16 17; CHECK-LABEL: %for.body18; CHECK: fmla.2d {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}19; CHECK: fmla.2d {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}[0]20; CHECK: fmadd {{d[0-9]+}}, {{d[0-9]+}}, {{d[0-9]+}}, {{d[0-9]+}}21for.body: ; preds = %for.body, %entry22 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]23 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 124 %arrayidx3 = getelementptr inbounds double, ptr %src, i64 %indvars.iv.next25 %tmp3 = load double, ptr %arrayidx3, align 826 %add = fadd fast double %tmp3, %tmp327 %mul = fmul fast double %add, %fadd28 %e1 = insertelement <2 x double> undef, double %add, i32 029 %e2 = insertelement <2 x double> %e1, double %add, i32 130 %add2 = fadd fast <2 x double> %e2, <double 3.000000e+00, double -3.000000e+00>31 %e3 = insertelement <2 x double> undef, double %mul, i32 032 %e4 = insertelement <2 x double> %e3, double %mul, i32 133 %mul2 = fmul fast <2 x double> %add2,<double 3.000000e+00, double -3.000000e+00>34 %e5 = insertelement <2 x double> undef, double %add, i32 035 %e6 = insertelement <2 x double> %e5, double %add, i32 136 %add3 = fadd fast <2 x double> %mul2, <double 3.000000e+00, double -3.000000e+00>37 %mulx = fmul fast <2 x double> %add2, %e238 %addx = fadd fast <2 x double> %mulx, %e439 %e7 = insertelement <2 x double> undef, double %mul, i32 040 %e8 = insertelement <2 x double> %e7, double %mul, i32 141 %e9 = fmul fast <2 x double> %addx, %add342 store <2 x double> %e9, ptr %arrayidx1, align 843 %e10 = extractelement <2 x double> %add3, i32 044 %mul3 = fmul fast double %mul, %e1045 %add4 = fadd fast double %mul3, %mul46 store double %add4, ptr %arrayidx2, align 847 %exitcond = icmp eq i64 %indvars.iv.next, 2548 br i1 %exitcond, label %for.end, label %for.body49 50for.end: ; preds = %for.body51 ret void52}53define void @foo_2s(ptr %src) {54entry:55 %arrayidx1 = getelementptr inbounds float, ptr %src, i64 556 %arrayidx2 = getelementptr inbounds float, ptr %src, i64 1157 br label %for.body58 59; CHECK-LABEL: %for.body60; CHECK: fmla.2s {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}61; CHECK: fmla.2s {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}[0]62; CHECK: fmadd {{s[0-9]+}}, {{s[0-9]+}}, {{s[0-9]+}}, {{s[0-9]+}}63for.body: ; preds = %for.body, %entry64 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]65 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 166 %arrayidx3 = getelementptr inbounds float, ptr %src, i64 %indvars.iv.next67 %tmp1 = load float, ptr %arrayidx3, align 868 %add = fadd fast float %tmp1, %tmp169 %mul = fmul fast float %add, %add70 %e1 = insertelement <2 x float> undef, float %add, i32 071 %e2 = insertelement <2 x float> %e1, float %add, i32 172 %add2 = fadd fast <2 x float> %e2, <float 3.000000e+00, float -3.000000e+00>73 %e3 = insertelement <2 x float> undef, float %mul, i32 074 %e4 = insertelement <2 x float> %e3, float %mul, i32 175 %mul2 = fmul fast <2 x float> %add2,<float 3.000000e+00, float -3.000000e+00>76 %e5 = insertelement <2 x float> undef, float %add, i32 077 %e6 = insertelement <2 x float> %e5, float %add, i32 178 %add3 = fadd fast <2 x float> %mul2, <float 3.000000e+00, float -3.000000e+00>79 %mulx = fmul fast <2 x float> %add2, %e280 %addx = fadd fast <2 x float> %mulx, %e481 %e7 = insertelement <2 x float> undef, float %mul, i32 082 %e8 = insertelement <2 x float> %e7, float %mul, i32 183 %e9 = fmul fast <2 x float> %addx, %add384 store <2 x float> %e9, ptr %arrayidx1, align 885 %e10 = extractelement <2 x float> %add3, i32 086 %mul3 = fmul fast float %mul, %e1087 %add4 = fadd fast float %mul3, %mul88 store float %add4, ptr %arrayidx2, align 889 %exitcond = icmp eq i64 %indvars.iv.next, 2590 br i1 %exitcond, label %for.end, label %for.body91 92for.end: ; preds = %for.body93 ret void94}95define void @foo_4s(ptr %src) {96entry:97 %arrayidx1 = getelementptr inbounds float, ptr %src, i64 598 %arrayidx2 = getelementptr inbounds float, ptr %src, i64 1199 br label %for.body100 101; CHECK-LABEL: %for.body102; CHECK: fmla.4s {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}103; CHECK: fmla.4s {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}[0]104for.body: ; preds = %for.body, %entry105 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]106 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1107 %arrayidx3 = getelementptr inbounds float, ptr %src, i64 %indvars.iv.next108 %tmp1 = load float, ptr %arrayidx3, align 8109 %add = fadd fast float %tmp1, %tmp1110 %mul = fmul fast float %add, %add111 %e1 = insertelement <4 x float> undef, float %add, i32 0112 %e2 = insertelement <4 x float> %e1, float %add, i32 1113 %add2 = fadd fast <4 x float> %e2, <float 3.000000e+00, float -3.000000e+00, float 5.000000e+00, float 7.000000e+00>114 %e3 = insertelement <4 x float> undef, float %mul, i32 0115 %e4 = insertelement <4 x float> %e3, float %mul, i32 1116 %mul2 = fmul fast <4 x float> %add2,<float 3.000000e+00, float -3.000000e+00, float 5.000000e+00, float 7.000000e+00>117 %e5 = insertelement <4 x float> undef, float %add, i32 0118 %e6 = insertelement <4 x float> %e5, float %add, i32 1119 %add3 = fadd fast <4 x float> %mul2, <float 3.000000e+00, float -3.000000e+00, float 5.000000e+00, float 7.000000e+00> 120 %mulx = fmul fast <4 x float> %add2, %e2121 %addx = fadd fast <4 x float> %mulx, %e4122 %e7 = insertelement <4 x float> undef, float %mul, i32 0123 %e8 = insertelement <4 x float> %e7, float %mul, i32 1124 %e9 = fmul fast <4 x float> %addx, %add3125 store <4 x float> %e9, ptr %arrayidx1, align 8126 %e10 = extractelement <4 x float> %add3, i32 0127 %mul3 = fmul fast float %mul, %e10128 store float %mul3, ptr %arrayidx2, align 8129 %exitcond = icmp eq i64 %indvars.iv.next, 25130 br i1 %exitcond, label %for.end, label %for.body131 132for.end: ; preds = %for.body133 ret void134}135 136define void @indexed_2s(<2 x float> %shuf, <2 x float> %add,137 ptr %pmul, ptr %pret) {138; CHECK-LABEL: %entry139; CHECK: for.body140; CHECK: fmla.2s {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}[0]141;142entry:143 %shuffle = shufflevector <2 x float> %shuf, <2 x float> undef, <2 x i32> zeroinitializer144 br label %for.body145 146for.body:147 %i = phi i64 [ 0, %entry ], [ %inext, %for.body ]148 %pmul_i = getelementptr inbounds <2 x float>, ptr %pmul, i64 %i149 %pret_i = getelementptr inbounds <2 x float>, ptr %pret, i64 %i150 151 %mul_i = load <2 x float>, ptr %pmul_i152 153 %mul = fmul fast <2 x float> %mul_i, %shuffle154 %muladd = fadd fast <2 x float> %mul, %add155 156 store <2 x float> %muladd, ptr %pret_i, align 16157 %inext = add i64 %i, 1158 br label %for.body159}160 161define void @indexed_2d(<2 x double> %shuf, <2 x double> %add,162 ptr %pmul, ptr %pret) {163; CHECK-LABEL: %entry164; CHECK: for.body165; CHECK: fmla.2d {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}[0]166;167entry:168 %shuffle = shufflevector <2 x double> %shuf, <2 x double> undef, <2 x i32> zeroinitializer169 br label %for.body170 171for.body:172 %i = phi i64 [ 0, %entry ], [ %inext, %for.body ]173 %pmul_i = getelementptr inbounds <2 x double>, ptr %pmul, i64 %i174 %pret_i = getelementptr inbounds <2 x double>, ptr %pret, i64 %i175 176 %mul_i = load <2 x double>, ptr %pmul_i177 178 %mul = fmul fast <2 x double> %mul_i, %shuffle179 %muladd = fadd fast <2 x double> %mul, %add180 181 store <2 x double> %muladd, ptr %pret_i, align 16182 %inext = add i64 %i, 1183 br label %for.body184}185 186define void @indexed_4s(<4 x float> %shuf, <4 x float> %add,187 ptr %pmul, ptr %pret) {188; CHECK-LABEL: %entry189; CHECK: for.body190; CHECK: fmla.4s {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}[0]191;192entry:193 %shuffle = shufflevector <4 x float> %shuf, <4 x float> undef, <4 x i32> zeroinitializer194 br label %for.body195 196for.body:197 %i = phi i64 [ 0, %entry ], [ %inext, %for.body ]198 %pmul_i = getelementptr inbounds <4 x float>, ptr %pmul, i64 %i199 %pret_i = getelementptr inbounds <4 x float>, ptr %pret, i64 %i200 201 %mul_i = load <4 x float>, ptr %pmul_i202 203 %mul = fmul fast <4 x float> %mul_i, %shuffle204 %muladd = fadd fast <4 x float> %mul, %add205 206 store <4 x float> %muladd, ptr %pret_i, align 16207 %inext = add i64 %i, 1208 br label %for.body209}210 211define void @indexed_4h(<4 x half> %shuf, <4 x half> %add,212 ptr %pmul, ptr %pret) {213; CHECK-LABEL: %entry214; CHECK: for.body215; CHECK: fmla.4h {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}[0]216;217entry:218 %shuffle = shufflevector <4 x half> %shuf, <4 x half> undef, <4 x i32> zeroinitializer219 br label %for.body220 221for.body:222 %i = phi i64 [ 0, %entry ], [ %inext, %for.body ]223 %pmul_i = getelementptr inbounds <4 x half>, ptr %pmul, i64 %i224 %pret_i = getelementptr inbounds <4 x half>, ptr %pret, i64 %i225 226 %mul_i = load <4 x half>, ptr %pmul_i227 228 %mul = fmul fast <4 x half> %mul_i, %shuffle229 %muladd = fadd fast <4 x half> %mul, %add230 231 store <4 x half> %muladd, ptr %pret_i, align 16232 %inext = add i64 %i, 1233 br label %for.body234}235 236define void @indexed_8h(<8 x half> %shuf, <8 x half> %add,237 ptr %pmul, ptr %pret) {238; CHECK-LABEL: %entry239; CHECK: for.body240; CHECK: fmla.8h {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}[0]241;242entry:243 %shuffle = shufflevector <8 x half> %shuf, <8 x half> undef, <8 x i32> zeroinitializer244 br label %for.body245 246for.body:247 %i = phi i64 [ 0, %entry ], [ %inext, %for.body ]248 %pmul_i = getelementptr inbounds <8 x half>, ptr %pmul, i64 %i249 %pret_i = getelementptr inbounds <8 x half>, ptr %pret, i64 %i250 251 %mul_i = load <8 x half>, ptr %pmul_i252 253 %mul = fmul fast <8 x half> %mul_i, %shuffle254 %muladd = fadd fast <8 x half> %mul, %add255 256 store <8 x half> %muladd, ptr %pret_i, align 16257 %inext = add i64 %i, 1258 br label %for.body259}260