154 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 4 x half> @complex_mul_v4f16(<vscale x 4 x half> %a, <vscale x 4 x half> %b) {8; CHECK-LABEL: complex_mul_v4f16:9; CHECK: // %bb.0: // %entry10; CHECK-NEXT: uzp2 z2.s, z0.s, z0.s11; CHECK-NEXT: uzp1 z0.s, z0.s, z0.s12; CHECK-NEXT: ptrue p0.d13; CHECK-NEXT: uzp2 z3.s, z1.s, z0.s14; CHECK-NEXT: uunpklo z0.d, z0.s15; CHECK-NEXT: uunpklo z2.d, z2.s16; CHECK-NEXT: uunpklo z3.d, z3.s17; CHECK-NEXT: uzp1 z1.s, z1.s, z0.s18; CHECK-NEXT: uunpklo z1.d, z1.s19; CHECK-NEXT: movprfx z4, z320; CHECK-NEXT: fmul z4.h, p0/m, z4.h, z0.h21; CHECK-NEXT: fmul z3.h, p0/m, z3.h, z2.h22; CHECK-NEXT: fmad z2.h, p0/m, z1.h, z4.h23; CHECK-NEXT: fnmsb z0.h, p0/m, z1.h, z3.h24; CHECK-NEXT: zip2 z1.d, z0.d, z2.d25; CHECK-NEXT: zip1 z0.d, z0.d, z2.d26; CHECK-NEXT: uzp1 z0.s, z0.s, z1.s27; CHECK-NEXT: ret28entry:29 %a.deinterleaved = tail call { <vscale x 2 x half>, <vscale x 2 x half> } @llvm.vector.deinterleave2.nxv4f16(<vscale x 4 x half> %a)30 %a.real = extractvalue { <vscale x 2 x half>, <vscale x 2 x half> } %a.deinterleaved, 031 %a.imag = extractvalue { <vscale x 2 x half>, <vscale x 2 x half> } %a.deinterleaved, 132 %b.deinterleaved = tail call { <vscale x 2 x half>, <vscale x 2 x half> } @llvm.vector.deinterleave2.nxv4f16(<vscale x 4 x half> %b)33 %b.real = extractvalue { <vscale x 2 x half>, <vscale x 2 x half> } %b.deinterleaved, 034 %b.imag = extractvalue { <vscale x 2 x half>, <vscale x 2 x half> } %b.deinterleaved, 135 %0 = fmul fast <vscale x 2 x half> %b.imag, %a.real36 %1 = fmul fast <vscale x 2 x half> %b.real, %a.imag37 %2 = fadd fast <vscale x 2 x half> %1, %038 %3 = fmul fast <vscale x 2 x half> %b.real, %a.real39 %4 = fmul fast <vscale x 2 x half> %a.imag, %b.imag40 %5 = fsub fast <vscale x 2 x half> %3, %441 %interleaved.vec = tail call <vscale x 4 x half> @llvm.vector.interleave2.nxv4f16(<vscale x 2 x half> %5, <vscale x 2 x half> %2)42 ret <vscale x 4 x half> %interleaved.vec43}44 45; Expected to transform46define <vscale x 8 x half> @complex_mul_v8f16(<vscale x 8 x half> %a, <vscale x 8 x half> %b) {47; CHECK-LABEL: complex_mul_v8f16:48; CHECK: // %bb.0: // %entry49; CHECK-NEXT: movi v2.2d, #000000000000000050; CHECK-NEXT: ptrue p0.h51; CHECK-NEXT: fcmla z2.h, p0/m, z1.h, z0.h, #052; CHECK-NEXT: fcmla z2.h, p0/m, z1.h, z0.h, #9053; CHECK-NEXT: mov z0.d, z2.d54; CHECK-NEXT: ret55entry:56 %a.deinterleaved = tail call { <vscale x 4 x half>, <vscale x 4 x half> } @llvm.vector.deinterleave2.nxv8f16(<vscale x 8 x half> %a)57 %a.real = extractvalue { <vscale x 4 x half>, <vscale x 4 x half> } %a.deinterleaved, 058 %a.imag = extractvalue { <vscale x 4 x half>, <vscale x 4 x half> } %a.deinterleaved, 159 %b.deinterleaved = tail call { <vscale x 4 x half>, <vscale x 4 x half> } @llvm.vector.deinterleave2.nxv8f16(<vscale x 8 x half> %b)60 %b.real = extractvalue { <vscale x 4 x half>, <vscale x 4 x half> } %b.deinterleaved, 061 %b.imag = extractvalue { <vscale x 4 x half>, <vscale x 4 x half> } %b.deinterleaved, 162 %0 = fmul fast <vscale x 4 x half> %b.imag, %a.real63 %1 = fmul fast <vscale x 4 x half> %b.real, %a.imag64 %2 = fadd fast <vscale x 4 x half> %1, %065 %3 = fmul fast <vscale x 4 x half> %b.real, %a.real66 %4 = fmul fast <vscale x 4 x half> %a.imag, %b.imag67 %5 = fsub fast <vscale x 4 x half> %3, %468 %interleaved.vec = tail call <vscale x 8 x half> @llvm.vector.interleave2.nxv8f16(<vscale x 4 x half> %5, <vscale x 4 x half> %2)69 ret <vscale x 8 x half> %interleaved.vec70}71; Expected to transform72define <vscale x 16 x half> @complex_mul_v16f16(<vscale x 16 x half> %a, <vscale x 16 x half> %b) {73; CHECK-LABEL: complex_mul_v16f16:74; CHECK: // %bb.0: // %entry75; CHECK-NEXT: movi v4.2d, #000000000000000076; CHECK-NEXT: movi v5.2d, #000000000000000077; CHECK-NEXT: ptrue p0.h78; CHECK-NEXT: fcmla z5.h, p0/m, z2.h, z0.h, #079; CHECK-NEXT: fcmla z4.h, p0/m, z3.h, z1.h, #080; CHECK-NEXT: fcmla z5.h, p0/m, z2.h, z0.h, #9081; CHECK-NEXT: fcmla z4.h, p0/m, z3.h, z1.h, #9082; CHECK-NEXT: mov z0.d, z5.d83; CHECK-NEXT: mov z1.d, z4.d84; CHECK-NEXT: ret85entry:86 %a.deinterleaved = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.vector.deinterleave2.nxv16f16(<vscale x 16 x half> %a)87 %a.real = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } %a.deinterleaved, 088 %a.imag = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } %a.deinterleaved, 189 %b.deinterleaved = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.vector.deinterleave2.nxv16f16(<vscale x 16 x half> %b)90 %b.real = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } %b.deinterleaved, 091 %b.imag = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } %b.deinterleaved, 192 %0 = fmul fast <vscale x 8 x half> %b.imag, %a.real93 %1 = fmul fast <vscale x 8 x half> %b.real, %a.imag94 %2 = fadd fast <vscale x 8 x half> %1, %095 %3 = fmul fast <vscale x 8 x half> %b.real, %a.real96 %4 = fmul fast <vscale x 8 x half> %a.imag, %b.imag97 %5 = fsub fast <vscale x 8 x half> %3, %498 %interleaved.vec = tail call <vscale x 16 x half> @llvm.vector.interleave2.nxv16f16(<vscale x 8 x half> %5, <vscale x 8 x half> %2)99 ret <vscale x 16 x half> %interleaved.vec100}101 102; Expected to transform103define <vscale x 32 x half> @complex_mul_v32f16(<vscale x 32 x half> %a, <vscale x 32 x half> %b) {104; CHECK-LABEL: complex_mul_v32f16:105; CHECK: // %bb.0: // %entry106; CHECK-NEXT: movi v24.2d, #0000000000000000107; CHECK-NEXT: movi v25.2d, #0000000000000000108; CHECK-NEXT: movi v26.2d, #0000000000000000109; CHECK-NEXT: movi v27.2d, #0000000000000000110; CHECK-NEXT: ptrue p0.h111; CHECK-NEXT: fcmla z24.h, p0/m, z4.h, z0.h, #0112; CHECK-NEXT: fcmla z25.h, p0/m, z5.h, z1.h, #0113; CHECK-NEXT: fcmla z27.h, p0/m, z6.h, z2.h, #0114; CHECK-NEXT: fcmla z26.h, p0/m, z7.h, z3.h, #0115; CHECK-NEXT: fcmla z24.h, p0/m, z4.h, z0.h, #90116; CHECK-NEXT: fcmla z25.h, p0/m, z5.h, z1.h, #90117; CHECK-NEXT: fcmla z27.h, p0/m, z6.h, z2.h, #90118; CHECK-NEXT: fcmla z26.h, p0/m, z7.h, z3.h, #90119; CHECK-NEXT: mov z0.d, z24.d120; CHECK-NEXT: mov z1.d, z25.d121; CHECK-NEXT: mov z2.d, z27.d122; CHECK-NEXT: mov z3.d, z26.d123; CHECK-NEXT: ret124entry:125 %a.deinterleaved = tail call { <vscale x 16 x half>, <vscale x 16 x half> } @llvm.vector.deinterleave2.nxv32f16(<vscale x 32 x half> %a)126 %a.real = extractvalue { <vscale x 16 x half>, <vscale x 16 x half> } %a.deinterleaved, 0127 %a.imag = extractvalue { <vscale x 16 x half>, <vscale x 16 x half> } %a.deinterleaved, 1128 %b.deinterleaved = tail call { <vscale x 16 x half>, <vscale x 16 x half> } @llvm.vector.deinterleave2.nxv32f16(<vscale x 32 x half> %b)129 %b.real = extractvalue { <vscale x 16 x half>, <vscale x 16 x half> } %b.deinterleaved, 0130 %b.imag = extractvalue { <vscale x 16 x half>, <vscale x 16 x half> } %b.deinterleaved, 1131 %0 = fmul fast <vscale x 16 x half> %b.imag, %a.real132 %1 = fmul fast <vscale x 16 x half> %b.real, %a.imag133 %2 = fadd fast <vscale x 16 x half> %1, %0134 %3 = fmul fast <vscale x 16 x half> %b.real, %a.real135 %4 = fmul fast <vscale x 16 x half> %a.imag, %b.imag136 %5 = fsub fast <vscale x 16 x half> %3, %4137 %interleaved.vec = tail call <vscale x 32 x half> @llvm.vector.interleave2.nxv32f16(<vscale x 16 x half> %5, <vscale x 16 x half> %2)138 ret <vscale x 32 x half> %interleaved.vec139}140 141declare { <vscale x 2 x half>, <vscale x 2 x half> } @llvm.vector.deinterleave2.nxv4f16(<vscale x 4 x half>)142declare <vscale x 4 x half> @llvm.vector.interleave2.nxv4f16(<vscale x 2 x half>, <vscale x 2 x half>)143 144declare { <vscale x 4 x half>, <vscale x 4 x half> } @llvm.vector.deinterleave2.nxv8f16(<vscale x 8 x half>)145declare <vscale x 8 x half> @llvm.vector.interleave2.nxv8f16(<vscale x 4 x half>, <vscale x 4 x half>)146 147declare { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.vector.deinterleave2.nxv16f16(<vscale x 16 x half>)148declare <vscale x 16 x half> @llvm.vector.interleave2.nxv16f16(<vscale x 8 x half>, <vscale x 8 x half>)149 150declare { <vscale x 16 x half>, <vscale x 16 x half> } @llvm.vector.deinterleave2.nxv32f16(<vscale x 32 x half>)151declare <vscale x 32 x half> @llvm.vector.interleave2.nxv32f16(<vscale x 16 x half>, <vscale x 16 x half>)152 153 154