1 ; REQUIRES: asserts 2 ; RUN: llc < %s -march=x86-64 -mcpu=core2 -pre-RA-sched=source -enable-misched -stats 2>&1 | FileCheck %s 3 ; 4 ; Verify that register pressure heuristics are working in MachineScheduler. 5 ; 6 ; We can further reduce spills in this case with a global register 7 ; pressure heuristic, like sethi-ullman numbers or biasing toward 8 ; scheduled subtrees. However, these heuristics are marginally 9 ; beneficial on x86_64 and exacerbate register pressure in other 10 ; more complex cases. 11 ; 12 ; CHECK: @wrap_mul4 13 ; CHECK: 23 regalloc - Number of spills inserted 14 15 define void @wrap_mul4(double* nocapture %Out, [4 x double]* nocapture %A, [4 x double]* nocapture %B) #0 { 16 entry: 17 %arrayidx1.i = getelementptr inbounds [4 x double], [4 x double]* %A, i64 0, i64 0 18 %0 = load double, double* %arrayidx1.i, align 8 19 %arrayidx3.i = getelementptr inbounds [4 x double], [4 x double]* %B, i64 0, i64 0 20 %1 = load double, double* %arrayidx3.i, align 8 21 %mul.i = fmul double %0, %1 22 %arrayidx5.i = getelementptr inbounds [4 x double], [4 x double]* %A, i64 0, i64 1 23 %2 = load double, double* %arrayidx5.i, align 8 24 %arrayidx7.i = getelementptr inbounds [4 x double], [4 x double]* %B, i64 1, i64 0 25 %3 = load double, double* %arrayidx7.i, align 8 26 %mul8.i = fmul double %2, %3 27 %add.i = fadd double %mul.i, %mul8.i 28 %arrayidx10.i = getelementptr inbounds [4 x double], [4 x double]* %A, i64 0, i64 2 29 %4 = load double, double* %arrayidx10.i, align 8 30 %arrayidx12.i = getelementptr inbounds [4 x double], [4 x double]* %B, i64 2, i64 0 31 %5 = load double, double* %arrayidx12.i, align 8 32 %mul13.i = fmul double %4, %5 33 %add14.i = fadd double %add.i, %mul13.i 34 %arrayidx16.i = getelementptr inbounds [4 x double], [4 x double]* %A, i64 0, i64 3 35 %6 = load double, double* %arrayidx16.i, align 8 36 %arrayidx18.i = getelementptr inbounds [4 x double], [4 x double]* %B, i64 3, i64 0 37 %7 = load double, double* %arrayidx18.i, align 8 38 %mul19.i = fmul double %6, %7 39 %add20.i = fadd double %add14.i, %mul19.i 40 %arrayidx25.i = getelementptr inbounds [4 x double], [4 x double]* %B, i64 0, i64 1 41 %8 = load double, double* %arrayidx25.i, align 8 42 %mul26.i = fmul double %0, %8 43 %arrayidx30.i = getelementptr inbounds [4 x double], [4 x double]* %B, i64 1, i64 1 44 %9 = load double, double* %arrayidx30.i, align 8 45 %mul31.i = fmul double %2, %9 46 %add32.i = fadd double %mul26.i, %mul31.i 47 %arrayidx36.i = getelementptr inbounds [4 x double], [4 x double]* %B, i64 2, i64 1 48 %10 = load double, double* %arrayidx36.i, align 8 49 %mul37.i = fmul double %4, %10 50 %add38.i = fadd double %add32.i, %mul37.i 51 %arrayidx42.i = getelementptr inbounds [4 x double], [4 x double]* %B, i64 3, i64 1 52 %11 = load double, double* %arrayidx42.i, align 8 53 %mul43.i = fmul double %6, %11 54 %add44.i = fadd double %add38.i, %mul43.i 55 %arrayidx49.i = getelementptr inbounds [4 x double], [4 x double]* %B, i64 0, i64 2 56 %12 = load double, double* %arrayidx49.i, align 8 57 %mul50.i = fmul double %0, %12 58 %arrayidx54.i = getelementptr inbounds [4 x double], [4 x double]* %B, i64 1, i64 2 59 %13 = load double, double* %arrayidx54.i, align 8 60 %mul55.i = fmul double %2, %13 61 %add56.i = fadd double %mul50.i, %mul55.i 62 %arrayidx60.i = getelementptr inbounds [4 x double], [4 x double]* %B, i64 2, i64 2 63 %14 = load double, double* %arrayidx60.i, align 8 64 %mul61.i = fmul double %4, %14 65 %add62.i = fadd double %add56.i, %mul61.i 66 %arrayidx66.i = getelementptr inbounds [4 x double], [4 x double]* %B, i64 3, i64 2 67 %15 = load double, double* %arrayidx66.i, align 8 68 %mul67.i = fmul double %6, %15 69 %add68.i = fadd double %add62.i, %mul67.i 70 %arrayidx73.i = getelementptr inbounds [4 x double], [4 x double]* %B, i64 0, i64 3 71 %16 = load double, double* %arrayidx73.i, align 8 72 %mul74.i = fmul double %0, %16 73 %arrayidx78.i = getelementptr inbounds [4 x double], [4 x double]* %B, i64 1, i64 3 74 %17 = load double, double* %arrayidx78.i, align 8 75 %mul79.i = fmul double %2, %17 76 %add80.i = fadd double %mul74.i, %mul79.i 77 %arrayidx84.i = getelementptr inbounds [4 x double], [4 x double]* %B, i64 2, i64 3 78 %18 = load double, double* %arrayidx84.i, align 8 79 %mul85.i = fmul double %4, %18 80 %add86.i = fadd double %add80.i, %mul85.i 81 %arrayidx90.i = getelementptr inbounds [4 x double], [4 x double]* %B, i64 3, i64 3 82 %19 = load double, double* %arrayidx90.i, align 8 83 %mul91.i = fmul double %6, %19 84 %add92.i = fadd double %add86.i, %mul91.i 85 %arrayidx95.i = getelementptr inbounds [4 x double], [4 x double]* %A, i64 1, i64 0 86 %20 = load double, double* %arrayidx95.i, align 8 87 %mul98.i = fmul double %1, %20 88 %arrayidx100.i = getelementptr inbounds [4 x double], [4 x double]* %A, i64 1, i64 1 89 %21 = load double, double* %arrayidx100.i, align 8 90 %mul103.i = fmul double %3, %21 91 %add104.i = fadd double %mul98.i, %mul103.i 92 %arrayidx106.i = getelementptr inbounds [4 x double], [4 x double]* %A, i64 1, i64 2 93 %22 = load double, double* %arrayidx106.i, align 8 94 %mul109.i = fmul double %5, %22 95 %add110.i = fadd double %add104.i, %mul109.i 96 %arrayidx112.i = getelementptr inbounds [4 x double], [4 x double]* %A, i64 1, i64 3 97 %23 = load double, double* %arrayidx112.i, align 8 98 %mul115.i = fmul double %7, %23 99 %add116.i = fadd double %add110.i, %mul115.i 100 %mul122.i = fmul double %8, %20 101 %mul127.i = fmul double %9, %21 102 %add128.i = fadd double %mul122.i, %mul127.i 103 %mul133.i = fmul double %10, %22 104 %add134.i = fadd double %add128.i, %mul133.i 105 %mul139.i = fmul double %11, %23 106 %add140.i = fadd double %add134.i, %mul139.i 107 %mul146.i = fmul double %12, %20 108 %mul151.i = fmul double %13, %21 109 %add152.i = fadd double %mul146.i, %mul151.i 110 %mul157.i = fmul double %14, %22 111 %add158.i = fadd double %add152.i, %mul157.i 112 %mul163.i = fmul double %15, %23 113 %add164.i = fadd double %add158.i, %mul163.i 114 %mul170.i = fmul double %16, %20 115 %mul175.i = fmul double %17, %21 116 %add176.i = fadd double %mul170.i, %mul175.i 117 %mul181.i = fmul double %18, %22 118 %add182.i = fadd double %add176.i, %mul181.i 119 %mul187.i = fmul double %19, %23 120 %add188.i = fadd double %add182.i, %mul187.i 121 %arrayidx191.i = getelementptr inbounds [4 x double], [4 x double]* %A, i64 2, i64 0 122 %24 = load double, double* %arrayidx191.i, align 8 123 %mul194.i = fmul double %1, %24 124 %arrayidx196.i = getelementptr inbounds [4 x double], [4 x double]* %A, i64 2, i64 1 125 %25 = load double, double* %arrayidx196.i, align 8 126 %mul199.i = fmul double %3, %25 127 %add200.i = fadd double %mul194.i, %mul199.i 128 %arrayidx202.i = getelementptr inbounds [4 x double], [4 x double]* %A, i64 2, i64 2 129 %26 = load double, double* %arrayidx202.i, align 8 130 %mul205.i = fmul double %5, %26 131 %add206.i = fadd double %add200.i, %mul205.i 132 %arrayidx208.i = getelementptr inbounds [4 x double], [4 x double]* %A, i64 2, i64 3 133 %27 = load double, double* %arrayidx208.i, align 8 134 %mul211.i = fmul double %7, %27 135 %add212.i = fadd double %add206.i, %mul211.i 136 %mul218.i = fmul double %8, %24 137 %mul223.i = fmul double %9, %25 138 %add224.i = fadd double %mul218.i, %mul223.i 139 %mul229.i = fmul double %10, %26 140 %add230.i = fadd double %add224.i, %mul229.i 141 %mul235.i = fmul double %11, %27 142 %add236.i = fadd double %add230.i, %mul235.i 143 %mul242.i = fmul double %12, %24 144 %mul247.i = fmul double %13, %25 145 %add248.i = fadd double %mul242.i, %mul247.i 146 %mul253.i = fmul double %14, %26 147 %add254.i = fadd double %add248.i, %mul253.i 148 %mul259.i = fmul double %15, %27 149 %add260.i = fadd double %add254.i, %mul259.i 150 %mul266.i = fmul double %16, %24 151 %mul271.i = fmul double %17, %25 152 %add272.i = fadd double %mul266.i, %mul271.i 153 %mul277.i = fmul double %18, %26 154 %add278.i = fadd double %add272.i, %mul277.i 155 %mul283.i = fmul double %19, %27 156 %add284.i = fadd double %add278.i, %mul283.i 157 %arrayidx287.i = getelementptr inbounds [4 x double], [4 x double]* %A, i64 3, i64 0 158 %28 = load double, double* %arrayidx287.i, align 8 159 %mul290.i = fmul double %1, %28 160 %arrayidx292.i = getelementptr inbounds [4 x double], [4 x double]* %A, i64 3, i64 1 161 %29 = load double, double* %arrayidx292.i, align 8 162 %mul295.i = fmul double %3, %29 163 %add296.i = fadd double %mul290.i, %mul295.i 164 %arrayidx298.i = getelementptr inbounds [4 x double], [4 x double]* %A, i64 3, i64 2 165 %30 = load double, double* %arrayidx298.i, align 8 166 %mul301.i = fmul double %5, %30 167 %add302.i = fadd double %add296.i, %mul301.i 168 %arrayidx304.i = getelementptr inbounds [4 x double], [4 x double]* %A, i64 3, i64 3 169 %31 = load double, double* %arrayidx304.i, align 8 170 %mul307.i = fmul double %7, %31 171 %add308.i = fadd double %add302.i, %mul307.i 172 %mul314.i = fmul double %8, %28 173 %mul319.i = fmul double %9, %29 174 %add320.i = fadd double %mul314.i, %mul319.i 175 %mul325.i = fmul double %10, %30 176 %add326.i = fadd double %add320.i, %mul325.i 177 %mul331.i = fmul double %11, %31 178 %add332.i = fadd double %add326.i, %mul331.i 179 %mul338.i = fmul double %12, %28 180 %mul343.i = fmul double %13, %29 181 %add344.i = fadd double %mul338.i, %mul343.i 182 %mul349.i = fmul double %14, %30 183 %add350.i = fadd double %add344.i, %mul349.i 184 %mul355.i = fmul double %15, %31 185 %add356.i = fadd double %add350.i, %mul355.i 186 %mul362.i = fmul double %16, %28 187 %mul367.i = fmul double %17, %29 188 %add368.i = fadd double %mul362.i, %mul367.i 189 %mul373.i = fmul double %18, %30 190 %add374.i = fadd double %add368.i, %mul373.i 191 %mul379.i = fmul double %19, %31 192 %add380.i = fadd double %add374.i, %mul379.i 193 store double %add20.i, double* %Out, align 8 194 %Res.i.sroa.1.8.idx2 = getelementptr inbounds double, double* %Out, i64 1 195 store double %add44.i, double* %Res.i.sroa.1.8.idx2, align 8 196 %Res.i.sroa.2.16.idx4 = getelementptr inbounds double, double* %Out, i64 2 197 store double %add68.i, double* %Res.i.sroa.2.16.idx4, align 8 198 %Res.i.sroa.3.24.idx6 = getelementptr inbounds double, double* %Out, i64 3 199 store double %add92.i, double* %Res.i.sroa.3.24.idx6, align 8 200 %Res.i.sroa.4.32.idx8 = getelementptr inbounds double, double* %Out, i64 4 201 store double %add116.i, double* %Res.i.sroa.4.32.idx8, align 8 202 %Res.i.sroa.5.40.idx10 = getelementptr inbounds double, double* %Out, i64 5 203 store double %add140.i, double* %Res.i.sroa.5.40.idx10, align 8 204 %Res.i.sroa.6.48.idx12 = getelementptr inbounds double, double* %Out, i64 6 205 store double %add164.i, double* %Res.i.sroa.6.48.idx12, align 8 206 %Res.i.sroa.7.56.idx14 = getelementptr inbounds double, double* %Out, i64 7 207 store double %add188.i, double* %Res.i.sroa.7.56.idx14, align 8 208 %Res.i.sroa.8.64.idx16 = getelementptr inbounds double, double* %Out, i64 8 209 store double %add212.i, double* %Res.i.sroa.8.64.idx16, align 8 210 %Res.i.sroa.9.72.idx18 = getelementptr inbounds double, double* %Out, i64 9 211 store double %add236.i, double* %Res.i.sroa.9.72.idx18, align 8 212 %Res.i.sroa.10.80.idx20 = getelementptr inbounds double, double* %Out, i64 10 213 store double %add260.i, double* %Res.i.sroa.10.80.idx20, align 8 214 %Res.i.sroa.11.88.idx22 = getelementptr inbounds double, double* %Out, i64 11 215 store double %add284.i, double* %Res.i.sroa.11.88.idx22, align 8 216 %Res.i.sroa.12.96.idx24 = getelementptr inbounds double, double* %Out, i64 12 217 store double %add308.i, double* %Res.i.sroa.12.96.idx24, align 8 218 %Res.i.sroa.13.104.idx26 = getelementptr inbounds double, double* %Out, i64 13 219 store double %add332.i, double* %Res.i.sroa.13.104.idx26, align 8 220 %Res.i.sroa.14.112.idx28 = getelementptr inbounds double, double* %Out, i64 14 221 store double %add356.i, double* %Res.i.sroa.14.112.idx28, align 8 222 %Res.i.sroa.15.120.idx30 = getelementptr inbounds double, double* %Out, i64 15 223 store double %add380.i, double* %Res.i.sroa.15.120.idx30, align 8 224 ret void 225 } 226 227 attributes #0 = { noinline nounwind ssp uwtable "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "unsafe-fp-math"="false" "use-soft-float"="false" } 228
