218 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s --mattr=+complxnum,+neon -o - | FileCheck %s3 4target triple = "aarch64"5 6; a * b + c7define <4 x double> @mull_add(<4 x double> %a, <4 x double> %b, <4 x double> %c) {8; CHECK-LABEL: mull_add:9; CHECK: // %bb.0: // %entry10; CHECK-NEXT: fcmla v4.2d, v0.2d, v2.2d, #011; CHECK-NEXT: fcmla v5.2d, v1.2d, v3.2d, #012; CHECK-NEXT: fcmla v4.2d, v0.2d, v2.2d, #9013; CHECK-NEXT: fcmla v5.2d, v1.2d, v3.2d, #9014; CHECK-NEXT: mov v0.16b, v4.16b15; CHECK-NEXT: mov v1.16b, v5.16b16; CHECK-NEXT: ret17entry:18 %strided.vec = shufflevector <4 x double> %a, <4 x double> poison, <2 x i32> <i32 0, i32 2>19 %strided.vec28 = shufflevector <4 x double> %a, <4 x double> poison, <2 x i32> <i32 1, i32 3>20 %strided.vec30 = shufflevector <4 x double> %b, <4 x double> poison, <2 x i32> <i32 0, i32 2>21 %strided.vec31 = shufflevector <4 x double> %b, <4 x double> poison, <2 x i32> <i32 1, i32 3>22 %0 = fmul fast <2 x double> %strided.vec31, %strided.vec23 %1 = fmul fast <2 x double> %strided.vec30, %strided.vec2824 %2 = fadd fast <2 x double> %0, %125 %3 = fmul fast <2 x double> %strided.vec30, %strided.vec26 %strided.vec33 = shufflevector <4 x double> %c, <4 x double> poison, <2 x i32> <i32 0, i32 2>27 %strided.vec34 = shufflevector <4 x double> %c, <4 x double> poison, <2 x i32> <i32 1, i32 3>28 %4 = fadd fast <2 x double> %strided.vec33, %329 %5 = fmul fast <2 x double> %strided.vec31, %strided.vec2830 %6 = fsub fast <2 x double> %4, %531 %7 = fadd fast <2 x double> %2, %strided.vec3432 %interleaved.vec = shufflevector <2 x double> %6, <2 x double> %7, <4 x i32> <i32 0, i32 2, i32 1, i32 3>33 ret <4 x double> %interleaved.vec34}35 36; a * b + c * d37define <4 x double> @mul_add_mull(<4 x double> %a, <4 x double> %b, <4 x double> %c, <4 x double> %d) {38; CHECK-LABEL: mul_add_mull:39; CHECK: // %bb.0: // %entry40; CHECK-NEXT: movi v16.2d, #000000000000000041; CHECK-NEXT: movi v17.2d, #000000000000000042; CHECK-NEXT: fcmla v17.2d, v6.2d, v4.2d, #043; CHECK-NEXT: fcmla v16.2d, v7.2d, v5.2d, #044; CHECK-NEXT: fcmla v17.2d, v0.2d, v2.2d, #045; CHECK-NEXT: fcmla v16.2d, v1.2d, v3.2d, #046; CHECK-NEXT: fcmla v17.2d, v6.2d, v4.2d, #9047; CHECK-NEXT: fcmla v16.2d, v7.2d, v5.2d, #9048; CHECK-NEXT: fcmla v17.2d, v0.2d, v2.2d, #9049; CHECK-NEXT: fcmla v16.2d, v1.2d, v3.2d, #9050; CHECK-NEXT: mov v0.16b, v17.16b51; CHECK-NEXT: mov v1.16b, v16.16b52; CHECK-NEXT: ret53entry:54 %strided.vec = shufflevector <4 x double> %a, <4 x double> poison, <2 x i32> <i32 0, i32 2>55 %strided.vec51 = shufflevector <4 x double> %a, <4 x double> poison, <2 x i32> <i32 1, i32 3>56 %strided.vec53 = shufflevector <4 x double> %b, <4 x double> poison, <2 x i32> <i32 0, i32 2>57 %strided.vec54 = shufflevector <4 x double> %b, <4 x double> poison, <2 x i32> <i32 1, i32 3>58 %0 = fmul fast <2 x double> %strided.vec54, %strided.vec59 %1 = fmul fast <2 x double> %strided.vec53, %strided.vec5160 %2 = fmul fast <2 x double> %strided.vec53, %strided.vec61 %3 = fmul fast <2 x double> %strided.vec54, %strided.vec5162 %strided.vec56 = shufflevector <4 x double> %c, <4 x double> poison, <2 x i32> <i32 0, i32 2>63 %strided.vec57 = shufflevector <4 x double> %c, <4 x double> poison, <2 x i32> <i32 1, i32 3>64 %strided.vec59 = shufflevector <4 x double> %d, <4 x double> poison, <2 x i32> <i32 0, i32 2>65 %strided.vec60 = shufflevector <4 x double> %d, <4 x double> poison, <2 x i32> <i32 1, i32 3>66 %4 = fmul fast <2 x double> %strided.vec60, %strided.vec5667 %5 = fmul fast <2 x double> %strided.vec59, %strided.vec5768 %6 = fmul fast <2 x double> %strided.vec59, %strided.vec5669 %7 = fmul fast <2 x double> %strided.vec60, %strided.vec5770 %8 = fadd fast <2 x double> %7, %371 %9 = fadd fast <2 x double> %6, %272 %10 = fsub fast <2 x double> %9, %873 %11 = fadd fast <2 x double> %0, %174 %12 = fadd fast <2 x double> %11, %575 %13 = fadd fast <2 x double> %12, %476 %interleaved.vec = shufflevector <2 x double> %10, <2 x double> %13, <4 x i32> <i32 0, i32 2, i32 1, i32 3>77 ret <4 x double> %interleaved.vec78}79 80; a * b - c * d81define <4 x double> @mul_sub_mull(<4 x double> %a, <4 x double> %b, <4 x double> %c, <4 x double> %d) {82; CHECK-LABEL: mul_sub_mull:83; CHECK: // %bb.0: // %entry84; CHECK-NEXT: movi v16.2d, #000000000000000085; CHECK-NEXT: movi v17.2d, #000000000000000086; CHECK-NEXT: fcmla v17.2d, v6.2d, v4.2d, #27087; CHECK-NEXT: fcmla v16.2d, v7.2d, v5.2d, #27088; CHECK-NEXT: fcmla v17.2d, v0.2d, v2.2d, #089; CHECK-NEXT: fcmla v16.2d, v1.2d, v3.2d, #090; CHECK-NEXT: fcmla v17.2d, v6.2d, v4.2d, #18091; CHECK-NEXT: fcmla v16.2d, v7.2d, v5.2d, #18092; CHECK-NEXT: fcmla v17.2d, v0.2d, v2.2d, #9093; CHECK-NEXT: fcmla v16.2d, v1.2d, v3.2d, #9094; CHECK-NEXT: mov v0.16b, v17.16b95; CHECK-NEXT: mov v1.16b, v16.16b96; CHECK-NEXT: ret97entry:98 %strided.vec = shufflevector <4 x double> %a, <4 x double> poison, <2 x i32> <i32 0, i32 2>99 %strided.vec53 = shufflevector <4 x double> %a, <4 x double> poison, <2 x i32> <i32 1, i32 3>100 %strided.vec55 = shufflevector <4 x double> %b, <4 x double> poison, <2 x i32> <i32 0, i32 2>101 %strided.vec56 = shufflevector <4 x double> %b, <4 x double> poison, <2 x i32> <i32 1, i32 3>102 %0 = fmul fast <2 x double> %strided.vec56, %strided.vec103 %1 = fmul fast <2 x double> %strided.vec55, %strided.vec53104 %2 = fmul fast <2 x double> %strided.vec55, %strided.vec105 %3 = fmul fast <2 x double> %strided.vec56, %strided.vec53106 %strided.vec58 = shufflevector <4 x double> %c, <4 x double> poison, <2 x i32> <i32 0, i32 2>107 %strided.vec59 = shufflevector <4 x double> %c, <4 x double> poison, <2 x i32> <i32 1, i32 3>108 %strided.vec61 = shufflevector <4 x double> %d, <4 x double> poison, <2 x i32> <i32 0, i32 2>109 %strided.vec62 = shufflevector <4 x double> %d, <4 x double> poison, <2 x i32> <i32 1, i32 3>110 %4 = fmul fast <2 x double> %strided.vec62, %strided.vec59111 %5 = fmul fast <2 x double> %strided.vec61, %strided.vec58112 %6 = fadd fast <2 x double> %5, %3113 %7 = fadd fast <2 x double> %4, %2114 %8 = fsub fast <2 x double> %7, %6115 %9 = fmul fast <2 x double> %strided.vec61, %strided.vec59116 %10 = fmul fast <2 x double> %strided.vec62, %strided.vec58117 %11 = fadd fast <2 x double> %10, %9118 %12 = fadd fast <2 x double> %0, %1119 %13 = fsub fast <2 x double> %12, %11120 %interleaved.vec = shufflevector <2 x double> %8, <2 x double> %13, <4 x i32> <i32 0, i32 2, i32 1, i32 3>121 ret <4 x double> %interleaved.vec122}123 124; a * b + conj(c) * d125define <4 x double> @mul_conj_mull(<4 x double> %a, <4 x double> %b, <4 x double> %c, <4 x double> %d) {126; CHECK-LABEL: mul_conj_mull:127; CHECK: // %bb.0: // %entry128; CHECK-NEXT: movi v16.2d, #0000000000000000129; CHECK-NEXT: movi v17.2d, #0000000000000000130; CHECK-NEXT: fcmla v17.2d, v0.2d, v2.2d, #0131; CHECK-NEXT: fcmla v16.2d, v1.2d, v3.2d, #0132; CHECK-NEXT: fcmla v17.2d, v0.2d, v2.2d, #90133; CHECK-NEXT: fcmla v16.2d, v1.2d, v3.2d, #90134; CHECK-NEXT: fcmla v17.2d, v4.2d, v6.2d, #0135; CHECK-NEXT: fcmla v16.2d, v5.2d, v7.2d, #0136; CHECK-NEXT: fcmla v17.2d, v4.2d, v6.2d, #270137; CHECK-NEXT: fcmla v16.2d, v5.2d, v7.2d, #270138; CHECK-NEXT: mov v0.16b, v17.16b139; CHECK-NEXT: mov v1.16b, v16.16b140; CHECK-NEXT: ret141entry:142 %strided.vec = shufflevector <4 x double> %a, <4 x double> poison, <2 x i32> <i32 0, i32 2>143 %strided.vec59 = shufflevector <4 x double> %a, <4 x double> poison, <2 x i32> <i32 1, i32 3>144 %strided.vec61 = shufflevector <4 x double> %b, <4 x double> poison, <2 x i32> <i32 0, i32 2>145 %strided.vec62 = shufflevector <4 x double> %b, <4 x double> poison, <2 x i32> <i32 1, i32 3>146 %0 = fmul fast <2 x double> %strided.vec62, %strided.vec147 %1 = fmul fast <2 x double> %strided.vec61, %strided.vec59148 %2 = fmul fast <2 x double> %strided.vec61, %strided.vec149 %strided.vec64 = shufflevector <4 x double> %c, <4 x double> poison, <2 x i32> <i32 0, i32 2>150 %strided.vec65 = shufflevector <4 x double> %c, <4 x double> poison, <2 x i32> <i32 1, i32 3>151 %strided.vec67 = shufflevector <4 x double> %d, <4 x double> poison, <2 x i32> <i32 0, i32 2>152 %strided.vec68 = shufflevector <4 x double> %d, <4 x double> poison, <2 x i32> <i32 1, i32 3>153 %3 = fmul fast <2 x double> %strided.vec68, %strided.vec64154 %4 = fmul fast <2 x double> %strided.vec67, %strided.vec64155 %5 = fmul fast <2 x double> %strided.vec68, %strided.vec65156 %6 = fmul fast <2 x double> %strided.vec62, %strided.vec59157 %7 = fsub fast <2 x double> %2, %6158 %8 = fadd fast <2 x double> %7, %4159 %9 = fadd fast <2 x double> %8, %5160 %10 = fadd fast <2 x double> %0, %1161 %11 = fmul fast <2 x double> %strided.vec67, %strided.vec65162 %12 = fsub fast <2 x double> %10, %11163 %13 = fadd fast <2 x double> %12, %3164 %interleaved.vec = shufflevector <2 x double> %9, <2 x double> %13, <4 x i32> <i32 0, i32 2, i32 1, i32 3>165 ret <4 x double> %interleaved.vec166}167 168; a + b + 1i * c * d169define <4 x double> @mul_add_rot_mull(<4 x double> %a, <4 x double> %b, <4 x double> %c, <4 x double> %d) {170; CHECK-LABEL: mul_add_rot_mull:171; CHECK: // %bb.0: // %entry172; CHECK-NEXT: zip2 v16.2d, v2.2d, v3.2d173; CHECK-NEXT: zip2 v17.2d, v0.2d, v1.2d174; CHECK-NEXT: zip1 v2.2d, v2.2d, v3.2d175; CHECK-NEXT: zip2 v18.2d, v4.2d, v5.2d176; CHECK-NEXT: zip1 v19.2d, v6.2d, v7.2d177; CHECK-NEXT: zip1 v0.2d, v0.2d, v1.2d178; CHECK-NEXT: zip1 v1.2d, v4.2d, v5.2d179; CHECK-NEXT: zip2 v5.2d, v6.2d, v7.2d180; CHECK-NEXT: fmul v3.2d, v16.2d, v17.2d181; CHECK-NEXT: fmul v4.2d, v2.2d, v17.2d182; CHECK-NEXT: fmla v3.2d, v18.2d, v19.2d183; CHECK-NEXT: fmla v4.2d, v0.2d, v16.2d184; CHECK-NEXT: fmla v3.2d, v1.2d, v5.2d185; CHECK-NEXT: fmla v4.2d, v1.2d, v19.2d186; CHECK-NEXT: fneg v3.2d, v3.2d187; CHECK-NEXT: fmls v4.2d, v18.2d, v5.2d188; CHECK-NEXT: fmla v3.2d, v0.2d, v2.2d189; CHECK-NEXT: zip1 v0.2d, v3.2d, v4.2d190; CHECK-NEXT: zip2 v1.2d, v3.2d, v4.2d191; CHECK-NEXT: ret192entry:193 %strided.vec = shufflevector <4 x double> %a, <4 x double> poison, <2 x i32> <i32 0, i32 2>194 %strided.vec79 = shufflevector <4 x double> %a, <4 x double> poison, <2 x i32> <i32 1, i32 3>195 %strided.vec81 = shufflevector <4 x double> %b, <4 x double> poison, <2 x i32> <i32 0, i32 2>196 %strided.vec82 = shufflevector <4 x double> %b, <4 x double> poison, <2 x i32> <i32 1, i32 3>197 %0 = fmul fast <2 x double> %strided.vec82, %strided.vec198 %1 = fmul fast <2 x double> %strided.vec81, %strided.vec79199 %2 = fmul fast <2 x double> %strided.vec81, %strided.vec200 %3 = fmul fast <2 x double> %strided.vec82, %strided.vec79201 %strided.vec84 = shufflevector <4 x double> %c, <4 x double> poison, <2 x i32> <i32 0, i32 2>202 %strided.vec85 = shufflevector <4 x double> %c, <4 x double> poison, <2 x i32> <i32 1, i32 3>203 %strided.vec87 = shufflevector <4 x double> %d, <4 x double> poison, <2 x i32> <i32 0, i32 2>204 %strided.vec88 = shufflevector <4 x double> %d, <4 x double> poison, <2 x i32> <i32 1, i32 3>205 %4 = fmul fast <2 x double> %strided.vec87, %strided.vec84206 %5 = fmul fast <2 x double> %strided.vec87, %strided.vec85207 %6 = fmul fast <2 x double> %strided.vec88, %strided.vec84208 %7 = fadd fast <2 x double> %5, %3209 %8 = fadd fast <2 x double> %7, %6210 %9 = fsub fast <2 x double> %2, %8211 %10 = fadd fast <2 x double> %0, %1212 %11 = fadd fast <2 x double> %10, %4213 %12 = fmul fast <2 x double> %strided.vec88, %strided.vec85214 %13 = fsub fast <2 x double> %11, %12215 %interleaved.vec = shufflevector <2 x double> %9, <2 x double> %13, <4 x i32> <i32 0, i32 2, i32 1, i32 3>216 ret <4 x double> %interleaved.vec217}218