111 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s --mattr=+sve -o - | FileCheck %s3; RUN: llc -use-constant-int-for-scalable-splat -use-constant-fp-for-scalable-splat < %s --mattr=+sve -o - | FileCheck %s4 5target triple = "aarch64"6 7; a[i] * b[i] * (11.0 + 3.0.i);8;9define <vscale x 4 x double> @complex_mul_const(<vscale x 4 x double> %a, <vscale x 4 x double> %b) {10; CHECK-LABEL: complex_mul_const:11; CHECK: // %bb.0: // %entry12; CHECK-NEXT: movi v4.2d, #000000000000000013; CHECK-NEXT: movi v5.2d, #000000000000000014; CHECK-NEXT: ptrue p0.d15; CHECK-NEXT: fmov z6.d, #3.0000000016; CHECK-NEXT: fmov z7.d, #11.0000000017; CHECK-NEXT: fcmla z4.d, p0/m, z0.d, z2.d, #018; CHECK-NEXT: fcmla z5.d, p0/m, z1.d, z3.d, #019; CHECK-NEXT: fcmla z4.d, p0/m, z0.d, z2.d, #9020; CHECK-NEXT: movi v2.2d, #000000000000000021; CHECK-NEXT: fcmla z5.d, p0/m, z1.d, z3.d, #9022; CHECK-NEXT: zip2 z1.d, z7.d, z6.d23; CHECK-NEXT: movi v0.2d, #000000000000000024; CHECK-NEXT: zip1 z3.d, z7.d, z6.d25; CHECK-NEXT: fcmla z2.d, p0/m, z5.d, z1.d, #026; CHECK-NEXT: fcmla z0.d, p0/m, z4.d, z3.d, #027; CHECK-NEXT: fcmla z2.d, p0/m, z5.d, z1.d, #9028; CHECK-NEXT: fcmla z0.d, p0/m, z4.d, z3.d, #9029; CHECK-NEXT: mov z1.d, z2.d30; CHECK-NEXT: ret31entry:32 %strided.vec = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %a)33 %0 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec, 034 %1 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec, 135 %strided.vec48 = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %b)36 %2 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec48, 037 %3 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec48, 138 %4 = fmul fast <vscale x 2 x double> %3, %039 %5 = fmul fast <vscale x 2 x double> %2, %140 %6 = fadd fast <vscale x 2 x double> %4, %541 %7 = fmul fast <vscale x 2 x double> %2, %042 %8 = fmul fast <vscale x 2 x double> %3, %143 %9 = fsub fast <vscale x 2 x double> %7, %844 %10 = fmul fast <vscale x 2 x double> %9, splat (double 3.000000e+00)45 %11 = fmul fast <vscale x 2 x double> %6, splat (double 1.100000e+01)46 %12 = fadd fast <vscale x 2 x double> %10, %1147 %13 = fmul fast <vscale x 2 x double> %9, splat (double 1.100000e+01)48 %14 = fmul fast <vscale x 2 x double> %6, splat (double 3.000000e+00)49 %15 = fsub fast <vscale x 2 x double> %13, %1450 %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)51 ret <vscale x 4 x double> %interleaved.vec52}53 54; a[i] * b[i] * c;55;56define <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) {57; CHECK-LABEL: complex_mul_non_const:58; CHECK: // %bb.0: // %entry59; CHECK-NEXT: movi v7.2d, #000000000000000060; CHECK-NEXT: ptrue p0.d61; CHECK-NEXT: // kill: def $d5 killed $d5 def $z562; CHECK-NEXT: // kill: def $d4 killed $d4 def $z463; CHECK-NEXT: movi v6.2d, #000000000000000064; CHECK-NEXT: mov z5.d, d565; CHECK-NEXT: mov z24.d, d466; CHECK-NEXT: movi v4.2d, #000000000000000067; CHECK-NEXT: fcmla z7.d, p0/m, z1.d, z3.d, #068; CHECK-NEXT: fcmla z6.d, p0/m, z0.d, z2.d, #069; CHECK-NEXT: fcmla z7.d, p0/m, z1.d, z3.d, #9070; CHECK-NEXT: zip2 z1.d, z24.d, z5.d71; CHECK-NEXT: fcmla z6.d, p0/m, z0.d, z2.d, #9072; CHECK-NEXT: movi v0.2d, #000000000000000073; CHECK-NEXT: zip1 z2.d, z24.d, z5.d74; CHECK-NEXT: fcmla z4.d, p0/m, z7.d, z1.d, #075; CHECK-NEXT: fcmla z0.d, p0/m, z6.d, z2.d, #076; CHECK-NEXT: fcmla z4.d, p0/m, z7.d, z1.d, #9077; CHECK-NEXT: fcmla z0.d, p0/m, z6.d, z2.d, #9078; CHECK-NEXT: mov z1.d, z4.d79; CHECK-NEXT: ret80entry:81 %c.coerce.fca.0.extract = extractvalue [2 x double] %c, 082 %c.coerce.fca.1.extract = extractvalue [2 x double] %c, 183 %broadcast.splatinsert = insertelement <vscale x 2 x double> poison, double %c.coerce.fca.1.extract, i64 084 %broadcast.splat = shufflevector <vscale x 2 x double> %broadcast.splatinsert, <vscale x 2 x double> poison, <vscale x 2 x i32> zeroinitializer85 %broadcast.splatinsert49 = insertelement <vscale x 2 x double> poison, double %c.coerce.fca.0.extract, i64 086 %broadcast.splat50 = shufflevector <vscale x 2 x double> %broadcast.splatinsert49, <vscale x 2 x double> poison, <vscale x 2 x i32> zeroinitializer87 %strided.vec = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %a)88 %0 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec, 089 %1 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec, 190 %strided.vec48 = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %b)91 %2 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec48, 092 %3 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec48, 193 %4 = fmul fast <vscale x 2 x double> %3, %094 %5 = fmul fast <vscale x 2 x double> %2, %195 %6 = fadd fast <vscale x 2 x double> %4, %596 %7 = fmul fast <vscale x 2 x double> %2, %097 %8 = fmul fast <vscale x 2 x double> %3, %198 %9 = fsub fast <vscale x 2 x double> %7, %899 %10 = fmul fast <vscale x 2 x double> %9, %broadcast.splat100 %11 = fmul fast <vscale x 2 x double> %6, %broadcast.splat50101 %12 = fadd fast <vscale x 2 x double> %10, %11102 %13 = fmul fast <vscale x 2 x double> %9, %broadcast.splat50103 %14 = fmul fast <vscale x 2 x double> %6, %broadcast.splat104 %15 = fsub fast <vscale x 2 x double> %13, %14105 %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)106 ret <vscale x 4 x double> %interleaved.vec107}108 109declare { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double>)110declare <vscale x 4 x double> @llvm.vector.interleave2.nxv4f64(<vscale x 2 x double>, <vscale x 2 x double>)111