1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc < %s --mattr=+sve -o - | FileCheck %s 3 4target triple = "aarch64" 5 6; a[i] * b[i] * (11.0 + 3.0.i); 7; 8define <vscale x 4 x double> @complex_mul_const(<vscale x 4 x double> %a, <vscale x 4 x double> %b) { 9; CHECK-LABEL: complex_mul_const: 10; CHECK: // %bb.0: // %entry 11; CHECK-NEXT: mov z4.d, #0 // =0x0 12; CHECK-NEXT: ptrue p0.d 13; CHECK-NEXT: fmov z7.d, #3.00000000 14; CHECK-NEXT: fmov z24.d, #11.00000000 15; CHECK-NEXT: mov z6.d, z4.d 16; CHECK-NEXT: mov z5.d, z4.d 17; CHECK-NEXT: fcmla z6.d, p0/m, z1.d, z3.d, #0 18; CHECK-NEXT: fcmla z5.d, p0/m, z0.d, z2.d, #0 19; CHECK-NEXT: fcmla z6.d, p0/m, z1.d, z3.d, #90 20; CHECK-NEXT: zip2 z1.d, z24.d, z7.d 21; CHECK-NEXT: fcmla z5.d, p0/m, z0.d, z2.d, #90 22; CHECK-NEXT: zip1 z2.d, z24.d, z7.d 23; CHECK-NEXT: mov z0.d, z4.d 24; CHECK-NEXT: fcmla z4.d, p0/m, z6.d, z1.d, #0 25; CHECK-NEXT: fcmla z0.d, p0/m, z5.d, z2.d, #0 26; CHECK-NEXT: fcmla z4.d, p0/m, z6.d, z1.d, #90 27; CHECK-NEXT: fcmla z0.d, p0/m, z5.d, z2.d, #90 28; CHECK-NEXT: mov z1.d, z4.d 29; CHECK-NEXT: ret 30entry: 31 %strided.vec = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %a) 32 %0 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec, 0 33 %1 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec, 1 34 %strided.vec48 = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %b) 35 %2 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec48, 0 36 %3 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec48, 1 37 %4 = fmul fast <vscale x 2 x double> %3, %0 38 %5 = fmul fast <vscale x 2 x double> %2, %1 39 %6 = fadd fast <vscale x 2 x double> %4, %5 40 %7 = fmul fast <vscale x 2 x double> %2, %0 41 %8 = fmul fast <vscale x 2 x double> %3, %1 42 %9 = fsub fast <vscale x 2 x double> %7, %8 43 %10 = fmul fast <vscale x 2 x double> %9, splat (double 3.000000e+00) 44 %11 = fmul fast <vscale x 2 x double> %6, splat (double 1.100000e+01) 45 %12 = fadd fast <vscale x 2 x double> %10, %11 46 %13 = fmul fast <vscale x 2 x double> %9, splat (double 1.100000e+01) 47 %14 = fmul fast <vscale x 2 x double> %6, splat (double 3.000000e+00) 48 %15 = fsub fast <vscale x 2 x double> %13, %14 49 %interleaved.vec = tail call <vscale x 4 x double> @llvm.vector.interleave2.nxv4f64(<vscale x 2 x double> %15, <vscale x 2 x double> %12) 50 ret <vscale x 4 x double> %interleaved.vec 51} 52 53; a[i] * b[i] * c; 54; 55define <vscale x 4 x double> @complex_mul_non_const(<vscale x 4 x double> %a, <vscale x 4 x double> %b, [2 x double] %c) { 56; CHECK-LABEL: complex_mul_non_const: 57; CHECK: // %bb.0: // %entry 58; CHECK-NEXT: mov z6.d, #0 // =0x0 59; CHECK-NEXT: ptrue p0.d 60; CHECK-NEXT: // kill: def $d5 killed $d5 def $z5 61; CHECK-NEXT: // kill: def $d4 killed $d4 def $z4 62; CHECK-NEXT: mov z5.d, d5 63; CHECK-NEXT: mov z4.d, d4 64; CHECK-NEXT: mov z24.d, z6.d 65; CHECK-NEXT: mov z7.d, z6.d 66; CHECK-NEXT: fcmla z24.d, p0/m, z1.d, z3.d, #0 67; CHECK-NEXT: fcmla z7.d, p0/m, z0.d, z2.d, #0 68; CHECK-NEXT: fcmla z24.d, p0/m, z1.d, z3.d, #90 69; CHECK-NEXT: zip2 z1.d, z4.d, z5.d 70; CHECK-NEXT: fcmla z7.d, p0/m, z0.d, z2.d, #90 71; CHECK-NEXT: zip1 z2.d, z4.d, z5.d 72; CHECK-NEXT: mov z0.d, z6.d 73; CHECK-NEXT: fcmla z6.d, p0/m, z24.d, z1.d, #0 74; CHECK-NEXT: fcmla z0.d, p0/m, z7.d, z2.d, #0 75; CHECK-NEXT: fcmla z6.d, p0/m, z24.d, z1.d, #90 76; CHECK-NEXT: fcmla z0.d, p0/m, z7.d, z2.d, #90 77; CHECK-NEXT: mov z1.d, z6.d 78; CHECK-NEXT: ret 79entry: 80 %c.coerce.fca.0.extract = extractvalue [2 x double] %c, 0 81 %c.coerce.fca.1.extract = extractvalue [2 x double] %c, 1 82 %broadcast.splatinsert = insertelement <vscale x 2 x double> poison, double %c.coerce.fca.1.extract, i64 0 83 %broadcast.splat = shufflevector <vscale x 2 x double> %broadcast.splatinsert, <vscale x 2 x double> poison, <vscale x 2 x i32> zeroinitializer 84 %broadcast.splatinsert49 = insertelement <vscale x 2 x double> poison, double %c.coerce.fca.0.extract, i64 0 85 %broadcast.splat50 = shufflevector <vscale x 2 x double> %broadcast.splatinsert49, <vscale x 2 x double> poison, <vscale x 2 x i32> zeroinitializer 86 %strided.vec = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %a) 87 %0 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec, 0 88 %1 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec, 1 89 %strided.vec48 = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %b) 90 %2 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec48, 0 91 %3 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec48, 1 92 %4 = fmul fast <vscale x 2 x double> %3, %0 93 %5 = fmul fast <vscale x 2 x double> %2, %1 94 %6 = fadd fast <vscale x 2 x double> %4, %5 95 %7 = fmul fast <vscale x 2 x double> %2, %0 96 %8 = fmul fast <vscale x 2 x double> %3, %1 97 %9 = fsub fast <vscale x 2 x double> %7, %8 98 %10 = fmul fast <vscale x 2 x double> %9, %broadcast.splat 99 %11 = fmul fast <vscale x 2 x double> %6, %broadcast.splat50 100 %12 = fadd fast <vscale x 2 x double> %10, %11 101 %13 = fmul fast <vscale x 2 x double> %9, %broadcast.splat50 102 %14 = fmul fast <vscale x 2 x double> %6, %broadcast.splat 103 %15 = fsub fast <vscale x 2 x double> %13, %14 104 %interleaved.vec = tail call <vscale x 4 x double> @llvm.vector.interleave2.nxv4f64(<vscale x 2 x double> %15, <vscale x 2 x double> %12) 105 ret <vscale x 4 x double> %interleaved.vec 106} 107 108declare { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double>) 109declare <vscale x 4 x double> @llvm.vector.interleave2.nxv4f64(<vscale x 2 x double>, <vscale x 2 x double>) 110