111 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; Expected to transform7define <vscale x 2 x double> @complex_mul_v2f64(<vscale x 2 x double> %a, <vscale x 2 x double> %b) {8; CHECK-LABEL: complex_mul_v2f64:9; CHECK: // %bb.0: // %entry10; CHECK-NEXT: movi v2.2d, #000000000000000011; CHECK-NEXT: ptrue p0.d12; CHECK-NEXT: fcmla z2.d, p0/m, z1.d, z0.d, #013; CHECK-NEXT: fcmla z2.d, p0/m, z1.d, z0.d, #9014; CHECK-NEXT: mov z0.d, z2.d15; CHECK-NEXT: ret16entry:17 %a.deinterleaved = tail call { <vscale x 1 x double>, <vscale x 1 x double> } @llvm.vector.deinterleave2.nxv2f64(<vscale x 2 x double> %a)18 %a.real = extractvalue { <vscale x 1 x double>, <vscale x 1 x double> } %a.deinterleaved, 019 %a.imag = extractvalue { <vscale x 1 x double>, <vscale x 1 x double> } %a.deinterleaved, 120 %b.deinterleaved = tail call { <vscale x 1 x double>, <vscale x 1 x double> } @llvm.vector.deinterleave2.nxv2f64(<vscale x 2 x double> %b)21 %b.real = extractvalue { <vscale x 1 x double>, <vscale x 1 x double> } %b.deinterleaved, 022 %b.imag = extractvalue { <vscale x 1 x double>, <vscale x 1 x double> } %b.deinterleaved, 123 %0 = fmul fast <vscale x 1 x double> %b.imag, %a.real24 %1 = fmul fast <vscale x 1 x double> %b.real, %a.imag25 %2 = fadd fast <vscale x 1 x double> %1, %026 %3 = fmul fast <vscale x 1 x double> %b.real, %a.real27 %4 = fmul fast <vscale x 1 x double> %a.imag, %b.imag28 %5 = fsub fast <vscale x 1 x double> %3, %429 %interleaved.vec = tail call <vscale x 2 x double> @llvm.vector.interleave2.nxv2f64(<vscale x 1 x double> %5, <vscale x 1 x double> %2)30 ret <vscale x 2 x double> %interleaved.vec31}32 33; Expected to transform34define <vscale x 4 x double> @complex_mul_v4f64(<vscale x 4 x double> %a, <vscale x 4 x double> %b) {35; CHECK-LABEL: complex_mul_v4f64:36; CHECK: // %bb.0: // %entry37; CHECK-NEXT: movi v4.2d, #000000000000000038; CHECK-NEXT: movi v5.2d, #000000000000000039; CHECK-NEXT: ptrue p0.d40; CHECK-NEXT: fcmla z5.d, p0/m, z2.d, z0.d, #041; CHECK-NEXT: fcmla z4.d, p0/m, z3.d, z1.d, #042; CHECK-NEXT: fcmla z5.d, p0/m, z2.d, z0.d, #9043; CHECK-NEXT: fcmla z4.d, p0/m, z3.d, z1.d, #9044; CHECK-NEXT: mov z0.d, z5.d45; CHECK-NEXT: mov z1.d, z4.d46; CHECK-NEXT: ret47entry:48 %a.deinterleaved = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %a)49 %a.real = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %a.deinterleaved, 050 %a.imag = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %a.deinterleaved, 151 %b.deinterleaved = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %b)52 %b.real = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %b.deinterleaved, 053 %b.imag = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %b.deinterleaved, 154 %0 = fmul fast <vscale x 2 x double> %b.imag, %a.real55 %1 = fmul fast <vscale x 2 x double> %b.real, %a.imag56 %2 = fadd fast <vscale x 2 x double> %1, %057 %3 = fmul fast <vscale x 2 x double> %b.real, %a.real58 %4 = fmul fast <vscale x 2 x double> %a.imag, %b.imag59 %5 = fsub fast <vscale x 2 x double> %3, %460 %interleaved.vec = tail call <vscale x 4 x double> @llvm.vector.interleave2.nxv4f64(<vscale x 2 x double> %5, <vscale x 2 x double> %2)61 ret <vscale x 4 x double> %interleaved.vec62}63 64; Expected to transform65define <vscale x 8 x double> @complex_mul_v8f64(<vscale x 8 x double> %a, <vscale x 8 x double> %b) {66; CHECK-LABEL: complex_mul_v8f64:67; CHECK: // %bb.0: // %entry68; CHECK-NEXT: movi v24.2d, #000000000000000069; CHECK-NEXT: movi v25.2d, #000000000000000070; CHECK-NEXT: movi v26.2d, #000000000000000071; CHECK-NEXT: movi v27.2d, #000000000000000072; CHECK-NEXT: ptrue p0.d73; CHECK-NEXT: fcmla z24.d, p0/m, z4.d, z0.d, #074; CHECK-NEXT: fcmla z25.d, p0/m, z5.d, z1.d, #075; CHECK-NEXT: fcmla z27.d, p0/m, z6.d, z2.d, #076; CHECK-NEXT: fcmla z26.d, p0/m, z7.d, z3.d, #077; CHECK-NEXT: fcmla z24.d, p0/m, z4.d, z0.d, #9078; CHECK-NEXT: fcmla z25.d, p0/m, z5.d, z1.d, #9079; CHECK-NEXT: fcmla z27.d, p0/m, z6.d, z2.d, #9080; CHECK-NEXT: fcmla z26.d, p0/m, z7.d, z3.d, #9081; CHECK-NEXT: mov z0.d, z24.d82; CHECK-NEXT: mov z1.d, z25.d83; CHECK-NEXT: mov z2.d, z27.d84; CHECK-NEXT: mov z3.d, z26.d85; CHECK-NEXT: ret86entry:87 %a.deinterleaved = tail call { <vscale x 4 x double>, <vscale x 4 x double> } @llvm.vector.deinterleave2.nxv8f64(<vscale x 8 x double> %a)88 %a.real = extractvalue { <vscale x 4 x double>, <vscale x 4 x double> } %a.deinterleaved, 089 %a.imag = extractvalue { <vscale x 4 x double>, <vscale x 4 x double> } %a.deinterleaved, 190 %b.deinterleaved = tail call { <vscale x 4 x double>, <vscale x 4 x double> } @llvm.vector.deinterleave2.nxv8f64(<vscale x 8 x double> %b)91 %b.real = extractvalue { <vscale x 4 x double>, <vscale x 4 x double> } %b.deinterleaved, 092 %b.imag = extractvalue { <vscale x 4 x double>, <vscale x 4 x double> } %b.deinterleaved, 193 %0 = fmul fast <vscale x 4 x double> %b.imag, %a.real94 %1 = fmul fast <vscale x 4 x double> %b.real, %a.imag95 %2 = fadd fast <vscale x 4 x double> %1, %096 %3 = fmul fast <vscale x 4 x double> %b.real, %a.real97 %4 = fmul fast <vscale x 4 x double> %a.imag, %b.imag98 %5 = fsub fast <vscale x 4 x double> %3, %499 %interleaved.vec = tail call <vscale x 8 x double> @llvm.vector.interleave2.nxv8f64(<vscale x 4 x double> %5, <vscale x 4 x double> %2)100 ret <vscale x 8 x double> %interleaved.vec101}102 103declare { <vscale x 1 x double>, <vscale x 1 x double> } @llvm.vector.deinterleave2.nxv2f64(<vscale x 2 x double>)104declare <vscale x 2 x double> @llvm.vector.interleave2.nxv2f64(<vscale x 1 x double>, <vscale x 1 x double>)105 106declare { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double>)107declare <vscale x 4 x double> @llvm.vector.interleave2.nxv4f64(<vscale x 2 x double>, <vscale x 2 x double>)108 109declare { <vscale x 4 x double>, <vscale x 4 x double> } @llvm.vector.deinterleave2.nxv8f64(<vscale x 8 x double>)110declare <vscale x 8 x double> @llvm.vector.interleave2.nxv8f64(<vscale x 4 x double>, <vscale x 4 x double>)111