147 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s --mattr=+sve2 -o - | FileCheck %s3 4target triple = "aarch64"5 6; Expected to transform7define <vscale x 2 x i64> @complex_mul_v2i64(<vscale x 2 x i64> %a, <vscale x 2 x i64> %b) {8; CHECK-LABEL: complex_mul_v2i64:9; CHECK: // %bb.0: // %entry10; CHECK-NEXT: movi v2.2d, #000000000000000011; CHECK-NEXT: cmla z2.d, z1.d, z0.d, #012; CHECK-NEXT: cmla z2.d, z1.d, z0.d, #9013; CHECK-NEXT: mov z0.d, z2.d14; CHECK-NEXT: ret15entry:16 %a.deinterleaved = tail call { <vscale x 1 x i64>, <vscale x 1 x i64> } @llvm.vector.deinterleave2.nxv2i64(<vscale x 2 x i64> %a)17 %a.real = extractvalue { <vscale x 1 x i64>, <vscale x 1 x i64> } %a.deinterleaved, 018 %a.imag = extractvalue { <vscale x 1 x i64>, <vscale x 1 x i64> } %a.deinterleaved, 119 %b.deinterleaved = tail call { <vscale x 1 x i64>, <vscale x 1 x i64> } @llvm.vector.deinterleave2.nxv2i64(<vscale x 2 x i64> %b)20 %b.real = extractvalue { <vscale x 1 x i64>, <vscale x 1 x i64> } %b.deinterleaved, 021 %b.imag = extractvalue { <vscale x 1 x i64>, <vscale x 1 x i64> } %b.deinterleaved, 122 %0 = mul <vscale x 1 x i64> %b.imag, %a.real23 %1 = mul <vscale x 1 x i64> %b.real, %a.imag24 %2 = add <vscale x 1 x i64> %1, %025 %3 = mul <vscale x 1 x i64> %b.real, %a.real26 %4 = mul <vscale x 1 x i64> %a.imag, %b.imag27 %5 = sub <vscale x 1 x i64> %3, %428 %interleaved.vec = tail call <vscale x 2 x i64> @llvm.vector.interleave2.nxv2i64(<vscale x 1 x i64> %5, <vscale x 1 x i64> %2)29 ret <vscale x 2 x i64> %interleaved.vec30}31 32; Expected to transform33define <vscale x 4 x i64> @complex_mul_v4i64(<vscale x 4 x i64> %a, <vscale x 4 x i64> %b) {34; CHECK-LABEL: complex_mul_v4i64:35; CHECK: // %bb.0: // %entry36; CHECK-NEXT: movi v4.2d, #000000000000000037; CHECK-NEXT: movi v5.2d, #000000000000000038; CHECK-NEXT: cmla z5.d, z2.d, z0.d, #039; CHECK-NEXT: cmla z4.d, z3.d, z1.d, #040; CHECK-NEXT: cmla z5.d, z2.d, z0.d, #9041; CHECK-NEXT: cmla z4.d, z3.d, z1.d, #9042; CHECK-NEXT: mov z0.d, z5.d43; CHECK-NEXT: mov z1.d, z4.d44; CHECK-NEXT: ret45entry:46 %a.deinterleaved = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.vector.deinterleave2.nxv4i64(<vscale x 4 x i64> %a)47 %a.real = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } %a.deinterleaved, 048 %a.imag = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } %a.deinterleaved, 149 %b.deinterleaved = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.vector.deinterleave2.nxv4i64(<vscale x 4 x i64> %b)50 %b.real = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } %b.deinterleaved, 051 %b.imag = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } %b.deinterleaved, 152 %0 = mul <vscale x 2 x i64> %b.imag, %a.real53 %1 = mul <vscale x 2 x i64> %b.real, %a.imag54 %2 = add <vscale x 2 x i64> %1, %055 %3 = mul <vscale x 2 x i64> %b.real, %a.real56 %4 = mul <vscale x 2 x i64> %a.imag, %b.imag57 %5 = sub <vscale x 2 x i64> %3, %458 %interleaved.vec = tail call <vscale x 4 x i64> @llvm.vector.interleave2.nxv4i64(<vscale x 2 x i64> %5, <vscale x 2 x i64> %2)59 ret <vscale x 4 x i64> %interleaved.vec60}61 62; Expected to transform63define <vscale x 8 x i64> @complex_mul_v8i64(<vscale x 8 x i64> %a, <vscale x 8 x i64> %b) {64; CHECK-LABEL: complex_mul_v8i64:65; CHECK: // %bb.0: // %entry66; CHECK-NEXT: movi v24.2d, #000000000000000067; CHECK-NEXT: movi v25.2d, #000000000000000068; CHECK-NEXT: movi v26.2d, #000000000000000069; CHECK-NEXT: movi v27.2d, #000000000000000070; CHECK-NEXT: cmla z24.d, z4.d, z0.d, #071; CHECK-NEXT: cmla z25.d, z5.d, z1.d, #072; CHECK-NEXT: cmla z27.d, z6.d, z2.d, #073; CHECK-NEXT: cmla z26.d, z7.d, z3.d, #074; CHECK-NEXT: cmla z24.d, z4.d, z0.d, #9075; CHECK-NEXT: cmla z25.d, z5.d, z1.d, #9076; CHECK-NEXT: cmla z27.d, z6.d, z2.d, #9077; CHECK-NEXT: cmla z26.d, z7.d, z3.d, #9078; CHECK-NEXT: mov z0.d, z24.d79; CHECK-NEXT: mov z1.d, z25.d80; CHECK-NEXT: mov z2.d, z27.d81; CHECK-NEXT: mov z3.d, z26.d82; CHECK-NEXT: ret83entry:84 %a.deinterleaved = tail call { <vscale x 4 x i64>, <vscale x 4 x i64> } @llvm.vector.deinterleave2.nxv8i64(<vscale x 8 x i64> %a)85 %a.real = extractvalue { <vscale x 4 x i64>, <vscale x 4 x i64> } %a.deinterleaved, 086 %a.imag = extractvalue { <vscale x 4 x i64>, <vscale x 4 x i64> } %a.deinterleaved, 187 %b.deinterleaved = tail call { <vscale x 4 x i64>, <vscale x 4 x i64> } @llvm.vector.deinterleave2.nxv8i64(<vscale x 8 x i64> %b)88 %b.real = extractvalue { <vscale x 4 x i64>, <vscale x 4 x i64> } %b.deinterleaved, 089 %b.imag = extractvalue { <vscale x 4 x i64>, <vscale x 4 x i64> } %b.deinterleaved, 190 %0 = mul <vscale x 4 x i64> %b.imag, %a.real91 %1 = mul <vscale x 4 x i64> %b.real, %a.imag92 %2 = add <vscale x 4 x i64> %1, %093 %3 = mul <vscale x 4 x i64> %b.real, %a.real94 %4 = mul <vscale x 4 x i64> %a.imag, %b.imag95 %5 = sub <vscale x 4 x i64> %3, %496 %interleaved.vec = tail call <vscale x 8 x i64> @llvm.vector.interleave2.nxv8i64(<vscale x 4 x i64> %5, <vscale x 4 x i64> %2)97 ret <vscale x 8 x i64> %interleaved.vec98}99 100; Expected to transform101define <vscale x 8 x i64> @complex_minus_mul_v8i64(<vscale x 8 x i64> %a, <vscale x 8 x i64> %b) {102; CHECK-LABEL: complex_minus_mul_v8i64:103; CHECK: // %bb.0: // %entry104; CHECK-NEXT: movi v24.2d, #0000000000000000105; CHECK-NEXT: movi v25.2d, #0000000000000000106; CHECK-NEXT: movi v26.2d, #0000000000000000107; CHECK-NEXT: movi v27.2d, #0000000000000000108; CHECK-NEXT: cmla z24.d, z4.d, z0.d, #270109; CHECK-NEXT: cmla z25.d, z5.d, z1.d, #270110; CHECK-NEXT: cmla z27.d, z6.d, z2.d, #270111; CHECK-NEXT: cmla z26.d, z7.d, z3.d, #270112; CHECK-NEXT: cmla z24.d, z4.d, z0.d, #180113; CHECK-NEXT: cmla z25.d, z5.d, z1.d, #180114; CHECK-NEXT: cmla z27.d, z6.d, z2.d, #180115; CHECK-NEXT: cmla z26.d, z7.d, z3.d, #180116; CHECK-NEXT: mov z0.d, z24.d117; CHECK-NEXT: mov z1.d, z25.d118; CHECK-NEXT: mov z2.d, z27.d119; CHECK-NEXT: mov z3.d, z26.d120; CHECK-NEXT: ret121entry:122 %a.deinterleaved = tail call { <vscale x 4 x i64>, <vscale x 4 x i64> } @llvm.vector.deinterleave2.nxv8i64(<vscale x 8 x i64> %a)123 %a.real = extractvalue { <vscale x 4 x i64>, <vscale x 4 x i64> } %a.deinterleaved, 0124 %a.imag = extractvalue { <vscale x 4 x i64>, <vscale x 4 x i64> } %a.deinterleaved, 1125 %0 = sub <vscale x 4 x i64> zeroinitializer, %a.real126 %b.deinterleaved = tail call { <vscale x 4 x i64>, <vscale x 4 x i64> } @llvm.vector.deinterleave2.nxv8i64(<vscale x 8 x i64> %b)127 %b.real = extractvalue { <vscale x 4 x i64>, <vscale x 4 x i64> } %b.deinterleaved, 0128 %b.imag = extractvalue { <vscale x 4 x i64>, <vscale x 4 x i64> } %b.deinterleaved, 1129 %1 = mul <vscale x 4 x i64> %b.real, %0130 %2 = mul <vscale x 4 x i64> %b.imag, %a.imag131 %3 = add <vscale x 4 x i64> %2, %1132 %4 = mul <vscale x 4 x i64> %b.real, %a.imag133 %5 = mul <vscale x 4 x i64> %b.imag, %0134 %6 = sub <vscale x 4 x i64> %5, %4135 %interleaved.vec = tail call <vscale x 8 x i64> @llvm.vector.interleave2.nxv8i64(<vscale x 4 x i64> %3, <vscale x 4 x i64> %6)136 ret <vscale x 8 x i64> %interleaved.vec137}138 139declare { <vscale x 1 x i64>, <vscale x 1 x i64> } @llvm.vector.deinterleave2.nxv2i64(<vscale x 2 x i64>)140declare <vscale x 2 x i64> @llvm.vector.interleave2.nxv2i64(<vscale x 1 x i64>, <vscale x 1 x i64>)141 142declare { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.vector.deinterleave2.nxv4i64(<vscale x 4 x i64>)143declare <vscale x 4 x i64> @llvm.vector.interleave2.nxv4i64(<vscale x 2 x i64>, <vscale x 2 x i64>)144 145declare { <vscale x 4 x i64>, <vscale x 4 x i64> } @llvm.vector.deinterleave2.nxv8i64(<vscale x 8 x i64>)146declare <vscale x 8 x i64> @llvm.vector.interleave2.nxv8i64(<vscale x 4 x i64>, <vscale x 4 x i64>)147