244 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s --mattr=+sve -o - | FileCheck %s3 4target triple = "aarch64"5 6; a * b + c7define <vscale x 4 x double> @mull_add(<vscale x 4 x double> %a, <vscale x 4 x double> %b, <vscale x 4 x double> %c) {8; CHECK-LABEL: mull_add:9; CHECK: // %bb.0: // %entry10; CHECK-NEXT: ptrue p0.d11; CHECK-NEXT: fcmla z4.d, p0/m, z0.d, z2.d, #012; CHECK-NEXT: fcmla z5.d, p0/m, z1.d, z3.d, #013; CHECK-NEXT: fcmla z4.d, p0/m, z0.d, z2.d, #9014; CHECK-NEXT: fcmla z5.d, p0/m, z1.d, z3.d, #9015; CHECK-NEXT: mov z0.d, z4.d16; CHECK-NEXT: mov z1.d, z5.d17; CHECK-NEXT: ret18entry:19 %strided.vec = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %a)20 %0 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec, 021 %1 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec, 122 %strided.vec29 = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %b)23 %2 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec29, 024 %3 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec29, 125 %4 = fmul fast <vscale x 2 x double> %3, %026 %5 = fmul fast <vscale x 2 x double> %2, %127 %6 = fadd fast <vscale x 2 x double> %4, %528 %7 = fmul fast <vscale x 2 x double> %2, %029 %strided.vec31 = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %c)30 %8 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec31, 031 %9 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec31, 132 %10 = fadd fast <vscale x 2 x double> %8, %733 %11 = fmul fast <vscale x 2 x double> %3, %134 %12 = fsub fast <vscale x 2 x double> %10, %1135 %13 = fadd fast <vscale x 2 x double> %6, %936 %interleaved.vec = tail call <vscale x 4 x double> @llvm.vector.interleave2.nxv4f64(<vscale x 2 x double> %12, <vscale x 2 x double> %13)37 ret <vscale x 4 x double> %interleaved.vec38}39 40; a * b + c * d41define <vscale x 4 x double> @mul_add_mull(<vscale x 4 x double> %a, <vscale x 4 x double> %b, <vscale x 4 x double> %c, <vscale x 4 x double> %d) {42; CHECK-LABEL: mul_add_mull:43; CHECK: // %bb.0: // %entry44; CHECK-NEXT: movi v24.2d, #000000000000000045; CHECK-NEXT: movi v25.2d, #000000000000000046; CHECK-NEXT: ptrue p0.d47; CHECK-NEXT: fcmla z25.d, p0/m, z6.d, z4.d, #048; CHECK-NEXT: fcmla z24.d, p0/m, z7.d, z5.d, #049; CHECK-NEXT: fcmla z25.d, p0/m, z0.d, z2.d, #050; CHECK-NEXT: fcmla z24.d, p0/m, z1.d, z3.d, #051; CHECK-NEXT: fcmla z25.d, p0/m, z6.d, z4.d, #9052; CHECK-NEXT: fcmla z24.d, p0/m, z7.d, z5.d, #9053; CHECK-NEXT: fcmla z25.d, p0/m, z0.d, z2.d, #9054; CHECK-NEXT: fcmla z24.d, p0/m, z1.d, z3.d, #9055; CHECK-NEXT: mov z0.d, z25.d56; CHECK-NEXT: mov z1.d, z24.d57; CHECK-NEXT: ret58entry:59 %strided.vec = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %a)60 %0 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec, 061 %1 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec, 162 %strided.vec52 = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %b)63 %2 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec52, 064 %3 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec52, 165 %4 = fmul fast <vscale x 2 x double> %3, %066 %5 = fmul fast <vscale x 2 x double> %2, %167 %6 = fmul fast <vscale x 2 x double> %2, %068 %7 = fmul fast <vscale x 2 x double> %3, %169 %strided.vec54 = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %c)70 %8 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec54, 071 %9 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec54, 172 %strided.vec56 = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %d)73 %10 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec56, 074 %11 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec56, 175 %12 = fmul fast <vscale x 2 x double> %11, %876 %13 = fmul fast <vscale x 2 x double> %10, %977 %14 = fmul fast <vscale x 2 x double> %10, %878 %15 = fmul fast <vscale x 2 x double> %11, %979 %16 = fadd fast <vscale x 2 x double> %15, %780 %17 = fadd fast <vscale x 2 x double> %14, %681 %18 = fsub fast <vscale x 2 x double> %17, %1682 %19 = fadd fast <vscale x 2 x double> %4, %583 %20 = fadd fast <vscale x 2 x double> %19, %1384 %21 = fadd fast <vscale x 2 x double> %20, %1285 %interleaved.vec = tail call <vscale x 4 x double> @llvm.vector.interleave2.nxv4f64(<vscale x 2 x double> %18, <vscale x 2 x double> %21)86 ret <vscale x 4 x double> %interleaved.vec87}88 89; a * b - c * d90define <vscale x 4 x double> @mul_sub_mull(<vscale x 4 x double> %a, <vscale x 4 x double> %b, <vscale x 4 x double> %c, <vscale x 4 x double> %d) {91; CHECK-LABEL: mul_sub_mull:92; CHECK: // %bb.0: // %entry93; CHECK-NEXT: movi v24.2d, #000000000000000094; CHECK-NEXT: movi v25.2d, #000000000000000095; CHECK-NEXT: ptrue p0.d96; CHECK-NEXT: fcmla z25.d, p0/m, z6.d, z4.d, #27097; CHECK-NEXT: fcmla z24.d, p0/m, z7.d, z5.d, #27098; CHECK-NEXT: fcmla z25.d, p0/m, z0.d, z2.d, #099; CHECK-NEXT: fcmla z24.d, p0/m, z1.d, z3.d, #0100; CHECK-NEXT: fcmla z25.d, p0/m, z6.d, z4.d, #180101; CHECK-NEXT: fcmla z24.d, p0/m, z7.d, z5.d, #180102; CHECK-NEXT: fcmla z25.d, p0/m, z0.d, z2.d, #90103; CHECK-NEXT: fcmla z24.d, p0/m, z1.d, z3.d, #90104; CHECK-NEXT: mov z0.d, z25.d105; CHECK-NEXT: mov z1.d, z24.d106; CHECK-NEXT: ret107entry:108 %strided.vec = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %a)109 %0 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec, 0110 %1 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec, 1111 %strided.vec54 = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %b)112 %2 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec54, 0113 %3 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec54, 1114 %4 = fmul fast <vscale x 2 x double> %3, %0115 %5 = fmul fast <vscale x 2 x double> %2, %1116 %6 = fmul fast <vscale x 2 x double> %2, %0117 %7 = fmul fast <vscale x 2 x double> %3, %1118 %strided.vec56 = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %c)119 %8 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec56, 0120 %9 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec56, 1121 %strided.vec58 = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %d)122 %10 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec58, 0123 %11 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec58, 1124 %12 = fmul fast <vscale x 2 x double> %11, %9125 %13 = fmul fast <vscale x 2 x double> %10, %8126 %14 = fadd fast <vscale x 2 x double> %13, %7127 %15 = fadd fast <vscale x 2 x double> %12, %6128 %16 = fsub fast <vscale x 2 x double> %15, %14129 %17 = fmul fast <vscale x 2 x double> %10, %9130 %18 = fmul fast <vscale x 2 x double> %11, %8131 %19 = fadd fast <vscale x 2 x double> %18, %17132 %20 = fadd fast <vscale x 2 x double> %4, %5133 %21 = fsub fast <vscale x 2 x double> %20, %19134 %interleaved.vec = tail call <vscale x 4 x double> @llvm.vector.interleave2.nxv4f64(<vscale x 2 x double> %16, <vscale x 2 x double> %21)135 ret <vscale x 4 x double> %interleaved.vec136}137 138; a * b + conj(c) * d139define <vscale x 4 x double> @mul_conj_mull(<vscale x 4 x double> %a, <vscale x 4 x double> %b, <vscale x 4 x double> %c, <vscale x 4 x double> %d) {140; CHECK-LABEL: mul_conj_mull:141; CHECK: // %bb.0: // %entry142; CHECK-NEXT: movi v24.2d, #0000000000000000143; CHECK-NEXT: movi v25.2d, #0000000000000000144; CHECK-NEXT: ptrue p0.d145; CHECK-NEXT: fcmla z25.d, p0/m, z0.d, z2.d, #0146; CHECK-NEXT: fcmla z24.d, p0/m, z1.d, z3.d, #0147; CHECK-NEXT: fcmla z25.d, p0/m, z0.d, z2.d, #90148; CHECK-NEXT: fcmla z24.d, p0/m, z1.d, z3.d, #90149; CHECK-NEXT: fcmla z25.d, p0/m, z4.d, z6.d, #0150; CHECK-NEXT: fcmla z24.d, p0/m, z5.d, z7.d, #0151; CHECK-NEXT: fcmla z25.d, p0/m, z4.d, z6.d, #270152; CHECK-NEXT: fcmla z24.d, p0/m, z5.d, z7.d, #270153; CHECK-NEXT: mov z0.d, z25.d154; CHECK-NEXT: mov z1.d, z24.d155; CHECK-NEXT: ret156entry:157 %strided.vec = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %a)158 %0 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec, 0159 %1 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec, 1160 %strided.vec60 = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %b)161 %2 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec60, 0162 %3 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec60, 1163 %4 = fmul fast <vscale x 2 x double> %3, %0164 %5 = fmul fast <vscale x 2 x double> %2, %1165 %6 = fmul fast <vscale x 2 x double> %2, %0166 %strided.vec62 = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %c)167 %7 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec62, 0168 %8 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec62, 1169 %strided.vec64 = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %d)170 %9 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec64, 0171 %10 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec64, 1172 %11 = fmul fast <vscale x 2 x double> %10, %7173 %12 = fmul fast <vscale x 2 x double> %9, %7174 %13 = fmul fast <vscale x 2 x double> %10, %8175 %14 = fmul fast <vscale x 2 x double> %3, %1176 %15 = fsub fast <vscale x 2 x double> %6, %14177 %16 = fadd fast <vscale x 2 x double> %15, %12178 %17 = fadd fast <vscale x 2 x double> %16, %13179 %18 = fadd fast <vscale x 2 x double> %4, %5180 %19 = fmul fast <vscale x 2 x double> %9, %8181 %20 = fsub fast <vscale x 2 x double> %18, %19182 %21 = fadd fast <vscale x 2 x double> %20, %11183 %interleaved.vec = tail call <vscale x 4 x double> @llvm.vector.interleave2.nxv4f64(<vscale x 2 x double> %17, <vscale x 2 x double> %21)184 ret <vscale x 4 x double> %interleaved.vec185}186 187; a + b + 1i * c * d188define <vscale x 4 x double> @mul_add_rot_mull(<vscale x 4 x double> %a, <vscale x 4 x double> %b, <vscale x 4 x double> %c, <vscale x 4 x double> %d) {189; CHECK-LABEL: mul_add_rot_mull:190; CHECK: // %bb.0: // %entry191; CHECK-NEXT: uzp2 z24.d, z2.d, z3.d192; CHECK-NEXT: uzp2 z25.d, z0.d, z1.d193; CHECK-NEXT: uzp1 z2.d, z2.d, z3.d194; CHECK-NEXT: uzp1 z0.d, z0.d, z1.d195; CHECK-NEXT: uzp2 z1.d, z4.d, z5.d196; CHECK-NEXT: uzp1 z26.d, z6.d, z7.d197; CHECK-NEXT: ptrue p0.d198; CHECK-NEXT: uzp1 z4.d, z4.d, z5.d199; CHECK-NEXT: uzp2 z5.d, z6.d, z7.d200; CHECK-NEXT: fmul z3.d, z2.d, z25.d201; CHECK-NEXT: fmul z25.d, z24.d, z25.d202; CHECK-NEXT: fmla z3.d, p0/m, z24.d, z0.d203; CHECK-NEXT: fmla z25.d, p0/m, z26.d, z1.d204; CHECK-NEXT: fmla z25.d, p0/m, z5.d, z4.d205; CHECK-NEXT: fmla z3.d, p0/m, z26.d, z4.d206; CHECK-NEXT: fnmsb z2.d, p0/m, z0.d, z25.d207; CHECK-NEXT: fmsb z1.d, p0/m, z5.d, z3.d208; CHECK-NEXT: zip1 z0.d, z2.d, z1.d209; CHECK-NEXT: zip2 z1.d, z2.d, z1.d210; CHECK-NEXT: ret211entry:212 %strided.vec = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %a)213 %0 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec, 0214 %1 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec, 1215 %strided.vec80 = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %b)216 %2 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec80, 0217 %3 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec80, 1218 %4 = fmul fast <vscale x 2 x double> %3, %0219 %5 = fmul fast <vscale x 2 x double> %2, %1220 %6 = fmul fast <vscale x 2 x double> %2, %0221 %7 = fmul fast <vscale x 2 x double> %3, %1222 %strided.vec82 = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %c)223 %8 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec82, 0224 %9 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec82, 1225 %strided.vec84 = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %d)226 %10 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec84, 0227 %11 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec84, 1228 %12 = fmul fast <vscale x 2 x double> %10, %8229 %13 = fmul fast <vscale x 2 x double> %10, %9230 %14 = fmul fast <vscale x 2 x double> %11, %8231 %15 = fadd fast <vscale x 2 x double> %13, %7232 %16 = fadd fast <vscale x 2 x double> %15, %14233 %17 = fsub fast <vscale x 2 x double> %6, %16234 %18 = fadd fast <vscale x 2 x double> %4, %5235 %19 = fadd fast <vscale x 2 x double> %18, %12236 %20 = fmul fast <vscale x 2 x double> %11, %9237 %21 = fsub fast <vscale x 2 x double> %19, %20238 %interleaved.vec = tail call <vscale x 4 x double> @llvm.vector.interleave2.nxv4f64(<vscale x 2 x double> %17, <vscale x 2 x double> %21)239 ret <vscale x 4 x double> %interleaved.vec240}241 242declare { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double>)243declare <vscale x 4 x double> @llvm.vector.interleave2.nxv4f64(<vscale x 2 x double>, <vscale x 2 x double>)244