501 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc -mtriple=aarch64 %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-SD3; RUN: llc -mtriple=aarch64 -global-isel %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-GI4 5define <4 x i8> @concat1(<2 x i8> %A, <2 x i8> %B) {6; CHECK-SD-LABEL: concat1:7; CHECK-SD: // %bb.0:8; CHECK-SD-NEXT: uzp1 v0.4h, v0.4h, v1.4h9; CHECK-SD-NEXT: ret10;11; CHECK-GI-LABEL: concat1:12; CHECK-GI: // %bb.0:13; CHECK-GI-NEXT: // kill: def $d0 killed $d0 def $q014; CHECK-GI-NEXT: mov w8, v0.s[1]15; CHECK-GI-NEXT: // kill: def $d1 killed $d1 def $q116; CHECK-GI-NEXT: mov w9, v1.s[1]17; CHECK-GI-NEXT: mov v0.h[1], w818; CHECK-GI-NEXT: fmov w8, s119; CHECK-GI-NEXT: mov v0.h[2], w820; CHECK-GI-NEXT: mov v0.h[3], w921; CHECK-GI-NEXT: // kill: def $d0 killed $d0 killed $q022; CHECK-GI-NEXT: ret23 %v4i8 = shufflevector <2 x i8> %A, <2 x i8> %B, <4 x i32> <i32 0, i32 1, i32 2, i32 3>24 ret <4 x i8> %v4i825}26 27define <8 x i8> @concat2(<4 x i8> %A, <4 x i8> %B) {28; CHECK-SD-LABEL: concat2:29; CHECK-SD: // %bb.0:30; CHECK-SD-NEXT: uzp1 v0.8b, v0.8b, v1.8b31; CHECK-SD-NEXT: ret32;33; CHECK-GI-LABEL: concat2:34; CHECK-GI: // %bb.0:35; CHECK-GI-NEXT: uzp1 v1.8b, v1.8b, v0.8b36; CHECK-GI-NEXT: uzp1 v0.8b, v0.8b, v0.8b37; CHECK-GI-NEXT: fmov w8, s138; CHECK-GI-NEXT: mov v0.s[1], w839; CHECK-GI-NEXT: // kill: def $d0 killed $d0 killed $q040; CHECK-GI-NEXT: ret41 %v8i8 = shufflevector <4 x i8> %A, <4 x i8> %B, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>42 ret <8 x i8> %v8i843}44 45define <16 x i8> @concat3(<8 x i8> %A, <8 x i8> %B) {46; CHECK-LABEL: concat3:47; CHECK: // %bb.0:48; CHECK-NEXT: // kill: def $d0 killed $d0 def $q049; CHECK-NEXT: // kill: def $d1 killed $d1 def $q150; CHECK-NEXT: mov v0.d[1], v1.d[0]51; CHECK-NEXT: ret52 %v16i8 = shufflevector <8 x i8> %A, <8 x i8> %B, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>53 ret <16 x i8> %v16i854}55 56define <4 x i16> @concat4(<2 x i16> %A, <2 x i16> %B) {57; CHECK-SD-LABEL: concat4:58; CHECK-SD: // %bb.0:59; CHECK-SD-NEXT: uzp1 v0.4h, v0.4h, v1.4h60; CHECK-SD-NEXT: ret61;62; CHECK-GI-LABEL: concat4:63; CHECK-GI: // %bb.0:64; CHECK-GI-NEXT: uzp1 v1.4h, v1.4h, v0.4h65; CHECK-GI-NEXT: uzp1 v0.4h, v0.4h, v0.4h66; CHECK-GI-NEXT: fmov w8, s167; CHECK-GI-NEXT: mov v0.s[1], w868; CHECK-GI-NEXT: // kill: def $d0 killed $d0 killed $q069; CHECK-GI-NEXT: ret70 %v4i16 = shufflevector <2 x i16> %A, <2 x i16> %B, <4 x i32> <i32 0, i32 1, i32 2, i32 3>71 ret <4 x i16> %v4i1672}73 74define <8 x i16> @concat5(<4 x i16> %A, <4 x i16> %B) {75; CHECK-LABEL: concat5:76; CHECK: // %bb.0:77; CHECK-NEXT: // kill: def $d0 killed $d0 def $q078; CHECK-NEXT: // kill: def $d1 killed $d1 def $q179; CHECK-NEXT: mov v0.d[1], v1.d[0]80; CHECK-NEXT: ret81 %v8i16 = shufflevector <4 x i16> %A, <4 x i16> %B, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>82 ret <8 x i16> %v8i1683}84 85define <16 x i16> @concat6(ptr %A, ptr %B) {86; CHECK-LABEL: concat6:87; CHECK: // %bb.0:88; CHECK-NEXT: ldr q0, [x0]89; CHECK-NEXT: ldr q1, [x1]90; CHECK-NEXT: ret91 %tmp1 = load <8 x i16>, ptr %A92 %tmp2 = load <8 x i16>, ptr %B93 %v16i16 = shufflevector <8 x i16> %tmp1, <8 x i16> %tmp2, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>94 ret <16 x i16> %v16i1695}96 97define <4 x i32> @concat7(<2 x i32> %A, <2 x i32> %B) {98; CHECK-LABEL: concat7:99; CHECK: // %bb.0:100; CHECK-NEXT: // kill: def $d0 killed $d0 def $q0101; CHECK-NEXT: // kill: def $d1 killed $d1 def $q1102; CHECK-NEXT: mov v0.d[1], v1.d[0]103; CHECK-NEXT: ret104 %v4i32 = shufflevector <2 x i32> %A, <2 x i32> %B, <4 x i32> <i32 0, i32 1, i32 2, i32 3>105 ret <4 x i32> %v4i32106}107 108define <8 x i32> @concat8(ptr %A, ptr %B) {109; CHECK-LABEL: concat8:110; CHECK: // %bb.0:111; CHECK-NEXT: ldr q0, [x0]112; CHECK-NEXT: ldr q1, [x1]113; CHECK-NEXT: ret114 %tmp1 = load <4 x i32>, ptr %A115 %tmp2 = load <4 x i32>, ptr %B116 %v8i32 = shufflevector <4 x i32> %tmp1, <4 x i32> %tmp2, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>117 ret <8 x i32> %v8i32118}119 120define <4 x half> @concat9(<2 x half> %A, <2 x half> %B) {121; CHECK-SD-LABEL: concat9:122; CHECK-SD: // %bb.0:123; CHECK-SD-NEXT: zip1 v0.2s, v0.2s, v1.2s124; CHECK-SD-NEXT: ret125;126; CHECK-GI-LABEL: concat9:127; CHECK-GI: // %bb.0:128; CHECK-GI-NEXT: fmov w8, s1129; CHECK-GI-NEXT: // kill: def $d0 killed $d0 def $q0130; CHECK-GI-NEXT: mov v0.s[1], w8131; CHECK-GI-NEXT: // kill: def $d0 killed $d0 killed $q0132; CHECK-GI-NEXT: ret133 %v4half= shufflevector <2 x half> %A, <2 x half> %B, <4 x i32> <i32 0, i32 1, i32 2, i32 3>134 ret <4 x half> %v4half135}136 137define <8 x half> @concat10(<4 x half> %A, <4 x half> %B) {138; CHECK-LABEL: concat10:139; CHECK: // %bb.0:140; CHECK-NEXT: // kill: def $d0 killed $d0 def $q0141; CHECK-NEXT: // kill: def $d1 killed $d1 def $q1142; CHECK-NEXT: mov v0.d[1], v1.d[0]143; CHECK-NEXT: ret144 %v8half= shufflevector <4 x half> %A, <4 x half> %B, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>145 ret <8 x half> %v8half146}147 148define <16 x half> @concat11(<8 x half> %A, <8 x half> %B) {149; CHECK-LABEL: concat11:150; CHECK: // %bb.0:151; CHECK-NEXT: ret152 %v16half= shufflevector <8 x half> %A, <8 x half> %B, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>153 ret <16 x half> %v16half154}155 156define <8 x i16> @concat_v8s16_v2s16(ptr %ptr) {157; CHECK-SD-LABEL: concat_v8s16_v2s16:158; CHECK-SD: // %bb.0:159; CHECK-SD-NEXT: ldr s0, [x0]160; CHECK-SD-NEXT: ret161;162; CHECK-GI-LABEL: concat_v8s16_v2s16:163; CHECK-GI: // %bb.0:164; CHECK-GI-NEXT: ldrh w8, [x0]165; CHECK-GI-NEXT: ldrh w9, [x0, #2]166; CHECK-GI-NEXT: fmov s0, w8167; CHECK-GI-NEXT: mov v0.h[1], w9168; CHECK-GI-NEXT: ret169 %a = load <2 x i16>, ptr %ptr170 %b = shufflevector <2 x i16> %a, <2 x i16> %a, <8 x i32> <i32 0, i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>171 ret <8 x i16> %b172}173 174define <16 x i8> @concat_v16s8_v4s8(ptr %ptr) {175; CHECK-LABEL: concat_v16s8_v4s8:176; CHECK: // %bb.0:177; CHECK-NEXT: ldr s0, [x0]178; CHECK-NEXT: ret179 %a = load <4 x i8>, ptr %ptr180 %b = shufflevector <4 x i8> %a, <4 x i8> %a, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>181 ret <16 x i8> %b182}183 184define <16 x i8> @concat_v16s8_v4s8_load(ptr %ptrA, ptr %ptrB, ptr %ptrC, ptr %ptrD) {185; CHECK-LABEL: concat_v16s8_v4s8_load:186; CHECK: // %bb.0:187; CHECK-NEXT: ldr s0, [x0]188; CHECK-NEXT: ld1 { v0.s }[1], [x1]189; CHECK-NEXT: ldr s1, [x2]190; CHECK-NEXT: ld1 { v1.s }[1], [x3]191; CHECK-NEXT: zip1 v0.2d, v0.2d, v1.2d192; CHECK-NEXT: ret193 %A = load <4 x i8>, ptr %ptrA194 %B = load <4 x i8>, ptr %ptrB195 %C = load <4 x i8>, ptr %ptrC196 %D = load <4 x i8>, ptr %ptrD197 %b = shufflevector <4 x i8> %A, <4 x i8> %B, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>198 %c = shufflevector <4 x i8> %C, <4 x i8> %D, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>199 %d = shufflevector <16 x i8> %b, <16 x i8> %c, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23>200 ret <16 x i8> %d201}202 203 204define <16 x i8> @concat_v16s8_v4s8_reg(<4 x i8> %A, <4 x i8> %B, <4 x i8> %C, <4 x i8> %D) {205; CHECK-SD-LABEL: concat_v16s8_v4s8_reg:206; CHECK-SD: // %bb.0:207; CHECK-SD-NEXT: // kill: def $d2 killed $d2 def $q2208; CHECK-SD-NEXT: // kill: def $d0 killed $d0 def $q0209; CHECK-SD-NEXT: // kill: def $d3 killed $d3 def $q3210; CHECK-SD-NEXT: // kill: def $d1 killed $d1 def $q1211; CHECK-SD-NEXT: mov v2.d[1], v3.d[0]212; CHECK-SD-NEXT: mov v0.d[1], v1.d[0]213; CHECK-SD-NEXT: uzp1 v0.16b, v0.16b, v2.16b214; CHECK-SD-NEXT: ret215;216; CHECK-GI-LABEL: concat_v16s8_v4s8_reg:217; CHECK-GI: // %bb.0:218; CHECK-GI-NEXT: uzp1 v1.8b, v1.8b, v0.8b219; CHECK-GI-NEXT: uzp1 v0.8b, v0.8b, v0.8b220; CHECK-GI-NEXT: fmov w8, s1221; CHECK-GI-NEXT: uzp1 v2.8b, v2.8b, v0.8b222; CHECK-GI-NEXT: mov v0.s[1], w8223; CHECK-GI-NEXT: fmov w8, s2224; CHECK-GI-NEXT: uzp1 v1.8b, v3.8b, v0.8b225; CHECK-GI-NEXT: mov v0.s[2], w8226; CHECK-GI-NEXT: fmov w8, s1227; CHECK-GI-NEXT: mov v0.s[3], w8228; CHECK-GI-NEXT: ret229 %b = shufflevector <4 x i8> %A, <4 x i8> %B, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>230 %c = shufflevector <4 x i8> %C, <4 x i8> %D, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>231 %d = shufflevector <16 x i8> %b, <16 x i8> %c, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23>232 ret <16 x i8> %d233}234 235define <8 x i16> @concat_v8s16_v2s16_reg(<2 x i16> %A, <2 x i16> %B, <2 x i16> %C, <2 x i16> %D) {236; CHECK-SD-LABEL: concat_v8s16_v2s16_reg:237; CHECK-SD: // %bb.0:238; CHECK-SD-NEXT: // kill: def $d3 killed $d3 killed $q0_q1_q2_q3 def $q0_q1_q2_q3239; CHECK-SD-NEXT: adrp x8, .LCPI15_0240; CHECK-SD-NEXT: // kill: def $d2 killed $d2 killed $q0_q1_q2_q3 def $q0_q1_q2_q3241; CHECK-SD-NEXT: ldr q4, [x8, :lo12:.LCPI15_0]242; CHECK-SD-NEXT: // kill: def $d1 killed $d1 killed $q0_q1_q2_q3 def $q0_q1_q2_q3243; CHECK-SD-NEXT: // kill: def $d0 killed $d0 killed $q0_q1_q2_q3 def $q0_q1_q2_q3244; CHECK-SD-NEXT: tbl v0.16b, { v0.16b, v1.16b, v2.16b, v3.16b }, v4.16b245; CHECK-SD-NEXT: ret246;247; CHECK-GI-LABEL: concat_v8s16_v2s16_reg:248; CHECK-GI: // %bb.0:249; CHECK-GI-NEXT: uzp1 v1.4h, v1.4h, v0.4h250; CHECK-GI-NEXT: uzp1 v0.4h, v0.4h, v0.4h251; CHECK-GI-NEXT: fmov w8, s1252; CHECK-GI-NEXT: uzp1 v2.4h, v2.4h, v0.4h253; CHECK-GI-NEXT: mov v0.s[1], w8254; CHECK-GI-NEXT: fmov w8, s2255; CHECK-GI-NEXT: uzp1 v1.4h, v3.4h, v0.4h256; CHECK-GI-NEXT: mov v0.s[2], w8257; CHECK-GI-NEXT: fmov w8, s1258; CHECK-GI-NEXT: mov v0.s[3], w8259; CHECK-GI-NEXT: ret260 %b = shufflevector <2 x i16> %A, <2 x i16> %B, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef>261 %c = shufflevector <2 x i16> %C, <2 x i16> %D, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef>262 %d = shufflevector <8 x i16> %b, <8 x i16> %c, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 10, i32 11>263 ret <8 x i16> %d264}265 266define <4 x i16> @concat_undef_first_use_first(ptr %p1, ptr %p2) {267; CHECK-SD-LABEL: concat_undef_first_use_first:268; CHECK-SD: // %bb.0:269; CHECK-SD-NEXT: ld1r { v0.2s }, [x0]270; CHECK-SD-NEXT: ret271;272; CHECK-GI-LABEL: concat_undef_first_use_first:273; CHECK-GI: // %bb.0:274; CHECK-GI-NEXT: ldrh w8, [x0]275; CHECK-GI-NEXT: ldrh w9, [x0, #2]276; CHECK-GI-NEXT: fmov s1, w8277; CHECK-GI-NEXT: mov v1.h[1], w9278; CHECK-GI-NEXT: mov v0.s[1], v1.s[0]279; CHECK-GI-NEXT: // kill: def $d0 killed $d0 killed $q0280; CHECK-GI-NEXT: ret281 %l1 = load <2 x i16>, ptr %p1282 %l2 = load <2 x i16>, ptr %p2283 %ext1 = shufflevector <2 x i16> %l1, <2 x i16> poison, <8 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>284 %ext2 = shufflevector <2 x i16> %l1, <2 x i16> poison, <8 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>285 %t = shufflevector <8 x i16> %ext1, <8 x i16> %ext2, <4 x i32> <i32 poison, i32 poison, i32 0, i32 1>286 ret <4 x i16> %t287}288 289define <4 x i16> @concat_undef_first_use_second(ptr %p1, ptr %p2) {290; CHECK-SD-LABEL: concat_undef_first_use_second:291; CHECK-SD: // %bb.0:292; CHECK-SD-NEXT: ld1r { v0.2s }, [x0]293; CHECK-SD-NEXT: ret294;295; CHECK-GI-LABEL: concat_undef_first_use_second:296; CHECK-GI: // %bb.0:297; CHECK-GI-NEXT: ldrh w8, [x0]298; CHECK-GI-NEXT: ldrh w9, [x0, #2]299; CHECK-GI-NEXT: fmov s1, w8300; CHECK-GI-NEXT: mov v1.h[1], w9301; CHECK-GI-NEXT: mov v0.s[1], v1.s[0]302; CHECK-GI-NEXT: // kill: def $d0 killed $d0 killed $q0303; CHECK-GI-NEXT: ret304 %l1 = load <2 x i16>, ptr %p1305 %l2 = load <2 x i16>, ptr %p2306 %ext1 = shufflevector <2 x i16> %l1, <2 x i16> poison, <8 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>307 %ext2 = shufflevector <2 x i16> %l1, <2 x i16> poison, <8 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>308 %t = shufflevector <8 x i16> %ext1, <8 x i16> %ext2, <4 x i32> <i32 poison, i32 poison, i32 8, i32 9>309 ret <4 x i16> %t310}311 312define <4 x i16> @concat_undef_first_use_undef(ptr %p1, ptr %p2) {313; CHECK-SD-LABEL: concat_undef_first_use_undef:314; CHECK-SD: // %bb.0:315; CHECK-SD-NEXT: ldr s0, [x0]316; CHECK-SD-NEXT: ret317;318; CHECK-GI-LABEL: concat_undef_first_use_undef:319; CHECK-GI: // %bb.0:320; CHECK-GI-NEXT: ret321 %l1 = load <2 x i16>, ptr %p1322 %l2 = load <2 x i16>, ptr %p2323 %ext1 = shufflevector <2 x i16> %l1, <2 x i16> poison, <8 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>324 %ext2 = shufflevector <2 x i16> %l1, <2 x i16> poison, <8 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>325 %t = shufflevector <8 x i16> %ext1, <8 x i16> %ext2, <4 x i32> <i32 poison, i32 poison, i32 2, i32 3>326 ret <4 x i16> %t327}328 329define <8 x i16> @concat_low_low_v8i16(<8 x i16> %a_vec, <8 x i16> %b_vec) {330; CHECK-LABEL: concat_low_low_v8i16:331; CHECK: // %bb.0: // %entry332; CHECK-NEXT: mov v0.d[1], v1.d[0]333; CHECK-NEXT: ret334entry:335 %shuffle.i3 = shufflevector <8 x i16> %a_vec, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>336 %shuffle.i = shufflevector <8 x i16> %b_vec, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>337 %shuffle.i4 = shufflevector <4 x i16> %shuffle.i3, <4 x i16> %shuffle.i, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>338 ret <8 x i16> %shuffle.i4339}340 341define <8 x i16> @concat_high_low_v8i16(<8 x i16> %a_vec, <8 x i16> %b_vec) {342; CHECK-SD-LABEL: concat_high_low_v8i16:343; CHECK-SD: // %bb.0: // %entry344; CHECK-SD-NEXT: ext v0.16b, v0.16b, v1.16b, #8345; CHECK-SD-NEXT: ret346;347; CHECK-GI-LABEL: concat_high_low_v8i16:348; CHECK-GI: // %bb.0: // %entry349; CHECK-GI-NEXT: mov d0, v0.d[1]350; CHECK-GI-NEXT: mov v0.d[1], v1.d[0]351; CHECK-GI-NEXT: ret352entry:353 %shuffle.i3 = shufflevector <8 x i16> %a_vec, <8 x i16> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>354 %shuffle.i = shufflevector <8 x i16> %b_vec, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>355 %shuffle.i4 = shufflevector <4 x i16> %shuffle.i3, <4 x i16> %shuffle.i, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>356 ret <8 x i16> %shuffle.i4357}358 359define <8 x i16> @concat_low_high_v8i16(<8 x i16> %a_vec, <8 x i16> %b_vec) {360; CHECK-SD-LABEL: concat_low_high_v8i16:361; CHECK-SD: // %bb.0: // %entry362; CHECK-SD-NEXT: ext v1.16b, v1.16b, v1.16b, #8363; CHECK-SD-NEXT: mov v0.d[1], v1.d[0]364; CHECK-SD-NEXT: ret365;366; CHECK-GI-LABEL: concat_low_high_v8i16:367; CHECK-GI: // %bb.0: // %entry368; CHECK-GI-NEXT: mov d1, v1.d[1]369; CHECK-GI-NEXT: mov v0.d[1], v1.d[0]370; CHECK-GI-NEXT: ret371entry:372 %shuffle.i3 = shufflevector <8 x i16> %a_vec, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>373 %shuffle.i = shufflevector <8 x i16> %b_vec, <8 x i16> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>374 %shuffle.i4 = shufflevector <4 x i16> %shuffle.i3, <4 x i16> %shuffle.i, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>375 ret <8 x i16> %shuffle.i4376}377 378define <8 x i16> @concat_high_high_v8i16(<8 x i16> %a_vec, <8 x i16> %b_vec) {379; CHECK-LABEL: concat_high_high_v8i16:380; CHECK: // %bb.0: // %entry381; CHECK-NEXT: mov v1.d[0], v0.d[1]382; CHECK-NEXT: mov v0.16b, v1.16b383; CHECK-NEXT: ret384entry:385 %shuffle.i3 = shufflevector <8 x i16> %a_vec, <8 x i16> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>386 %shuffle.i = shufflevector <8 x i16> %b_vec, <8 x i16> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>387 %shuffle.i4 = shufflevector <4 x i16> %shuffle.i3, <4 x i16> %shuffle.i, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>388 ret <8 x i16> %shuffle.i4389}390 391define <8 x half> @concat_high_high_v8f16(<8 x half> %a_vec, <8 x half> %b_vec) {392; CHECK-LABEL: concat_high_high_v8f16:393; CHECK: // %bb.0: // %entry394; CHECK-NEXT: mov v1.d[0], v0.d[1]395; CHECK-NEXT: mov v0.16b, v1.16b396; CHECK-NEXT: ret397entry:398 %shuffle.i3 = shufflevector <8 x half> %a_vec, <8 x half> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>399 %shuffle.i = shufflevector <8 x half> %b_vec, <8 x half> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>400 %shuffle.i4 = shufflevector <4 x half> %shuffle.i3, <4 x half> %shuffle.i, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>401 ret <8 x half> %shuffle.i4402}403 404define <8 x bfloat> @concat_high_high_v8bf16(<8 x bfloat> %a_vec, <8 x bfloat> %b_vec) {405; CHECK-LABEL: concat_high_high_v8bf16:406; CHECK: // %bb.0: // %entry407; CHECK-NEXT: mov v1.d[0], v0.d[1]408; CHECK-NEXT: mov v0.16b, v1.16b409; CHECK-NEXT: ret410entry:411 %shuffle.i3 = shufflevector <8 x bfloat> %a_vec, <8 x bfloat> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>412 %shuffle.i = shufflevector <8 x bfloat> %b_vec, <8 x bfloat> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>413 %shuffle.i4 = shufflevector <4 x bfloat> %shuffle.i3, <4 x bfloat> %shuffle.i, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>414 ret <8 x bfloat> %shuffle.i4415}416 417define <4 x i32> @concat_high_high_v4i32(<4 x i32> %a_vec, <4 x i32> %b_vec) {418; CHECK-SD-LABEL: concat_high_high_v4i32:419; CHECK-SD: // %bb.0: // %entry420; CHECK-SD-NEXT: zip2 v0.2d, v0.2d, v1.2d421; CHECK-SD-NEXT: ret422;423; CHECK-GI-LABEL: concat_high_high_v4i32:424; CHECK-GI: // %bb.0: // %entry425; CHECK-GI-NEXT: mov v1.d[0], v0.d[1]426; CHECK-GI-NEXT: mov v0.16b, v1.16b427; CHECK-GI-NEXT: ret428entry:429 %shuffle.i3 = shufflevector <4 x i32> %a_vec, <4 x i32> poison, <2 x i32> <i32 2, i32 3>430 %shuffle.i = shufflevector <4 x i32> %b_vec, <4 x i32> poison, <2 x i32> <i32 2, i32 3>431 %shuffle.i4 = shufflevector <2 x i32> %shuffle.i3, <2 x i32> %shuffle.i, <4 x i32> <i32 0, i32 1, i32 2, i32 3>432 ret <4 x i32> %shuffle.i4433}434 435define <4 x float> @concat_high_high_v4f32(<4 x float> %a_vec, <4 x float> %b_vec) {436; CHECK-SD-LABEL: concat_high_high_v4f32:437; CHECK-SD: // %bb.0: // %entry438; CHECK-SD-NEXT: zip2 v0.2d, v0.2d, v1.2d439; CHECK-SD-NEXT: ret440;441; CHECK-GI-LABEL: concat_high_high_v4f32:442; CHECK-GI: // %bb.0: // %entry443; CHECK-GI-NEXT: mov v1.d[0], v0.d[1]444; CHECK-GI-NEXT: mov v0.16b, v1.16b445; CHECK-GI-NEXT: ret446entry:447 %shuffle.i3 = shufflevector <4 x float> %a_vec, <4 x float> poison, <2 x i32> <i32 2, i32 3>448 %shuffle.i = shufflevector <4 x float> %b_vec, <4 x float> poison, <2 x i32> <i32 2, i32 3>449 %shuffle.i4 = shufflevector <2 x float> %shuffle.i3, <2 x float> %shuffle.i, <4 x i32> <i32 0, i32 1, i32 2, i32 3>450 ret <4 x float> %shuffle.i4451}452 453define <16 x i8> @concat_high_high_v16i8(<16 x i8> %a_vec, <16 x i8> %b_vec) {454; CHECK-LABEL: concat_high_high_v16i8:455; CHECK: // %bb.0: // %entry456; CHECK-NEXT: mov v1.d[0], v0.d[1]457; CHECK-NEXT: mov v0.16b, v1.16b458; CHECK-NEXT: ret459entry:460 %shuffle.i3 = shufflevector <16 x i8> %a_vec, <16 x i8> poison, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>461 %shuffle.i = shufflevector <16 x i8> %b_vec, <16 x i8> poison, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>462 %shuffle.i4 = shufflevector <8 x i8> %shuffle.i3, <8 x i8> %shuffle.i, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>463 ret <16 x i8> %shuffle.i4464}465 466define <2 x i64> @concat_high_high_v2i64(<2 x i64> %a_vec, <2 x i64> %b_vec) {467; CHECK-SD-LABEL: concat_high_high_v2i64:468; CHECK-SD: // %bb.0: // %entry469; CHECK-SD-NEXT: zip2 v0.2d, v0.2d, v1.2d470; CHECK-SD-NEXT: ret471;472; CHECK-GI-LABEL: concat_high_high_v2i64:473; CHECK-GI: // %bb.0: // %entry474; CHECK-GI-NEXT: mov d0, v0.d[1]475; CHECK-GI-NEXT: mov v0.d[1], v1.d[1]476; CHECK-GI-NEXT: ret477entry:478 %shuffle.i3 = shufflevector <2 x i64> %a_vec, <2 x i64> poison, <1 x i32> <i32 1>479 %shuffle.i = shufflevector <2 x i64> %b_vec, <2 x i64> poison, <1 x i32> <i32 1>480 %shuffle.i4 = shufflevector <1 x i64> %shuffle.i3, <1 x i64> %shuffle.i, <2 x i32> <i32 0, i32 1>481 ret <2 x i64> %shuffle.i4482}483 484define <2 x double> @concat_high_high_v2f64(<2 x double> %a_vec, <2 x double> %b_vec) {485; CHECK-SD-LABEL: concat_high_high_v2f64:486; CHECK-SD: // %bb.0: // %entry487; CHECK-SD-NEXT: zip2 v0.2d, v0.2d, v1.2d488; CHECK-SD-NEXT: ret489;490; CHECK-GI-LABEL: concat_high_high_v2f64:491; CHECK-GI: // %bb.0: // %entry492; CHECK-GI-NEXT: mov d0, v0.d[1]493; CHECK-GI-NEXT: mov v0.d[1], v1.d[1]494; CHECK-GI-NEXT: ret495entry:496 %shuffle.i3 = shufflevector <2 x double> %a_vec, <2 x double> poison, <1 x i32> <i32 1>497 %shuffle.i = shufflevector <2 x double> %b_vec, <2 x double> poison, <1 x i32> <i32 1>498 %shuffle.i4 = shufflevector <1 x double> %shuffle.i3, <1 x double> %shuffle.i, <2 x i32> <i32 0, i32 1>499 ret <2 x double> %shuffle.i4500}501