139 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 42; RUN: llc -mtriple=aarch64 < %s | FileCheck %s3 4define <16 x i8> @lower_trunc_16xi8(i16 %a, i16 %b, i16 %c, i16 %d, i16 %e, i16 %f, i16 %g, i16 %h, i16 %i, i16 %j, i16 %k, i16 %l, i16 %m, i16 %n, i16 %o, i16 %p) {5; CHECK-LABEL: lower_trunc_16xi8:6; CHECK: // %bb.0:7; CHECK-NEXT: fmov s0, w08; CHECK-NEXT: mov x8, sp9; CHECK-NEXT: mov v0.b[1], w110; CHECK-NEXT: mov v0.b[2], w211; CHECK-NEXT: mov v0.b[3], w312; CHECK-NEXT: mov v0.b[4], w413; CHECK-NEXT: mov v0.b[5], w514; CHECK-NEXT: mov v0.b[6], w615; CHECK-NEXT: mov v0.b[7], w716; CHECK-NEXT: ld1 { v0.b }[8], [x8]17; CHECK-NEXT: add x8, sp, #818; CHECK-NEXT: ld1 { v0.b }[9], [x8]19; CHECK-NEXT: add x8, sp, #1620; CHECK-NEXT: ld1 { v0.b }[10], [x8]21; CHECK-NEXT: add x8, sp, #2422; CHECK-NEXT: ld1 { v0.b }[11], [x8]23; CHECK-NEXT: add x8, sp, #3224; CHECK-NEXT: ld1 { v0.b }[12], [x8]25; CHECK-NEXT: add x8, sp, #4026; CHECK-NEXT: ld1 { v0.b }[13], [x8]27; CHECK-NEXT: add x8, sp, #4828; CHECK-NEXT: ld1 { v0.b }[14], [x8]29; CHECK-NEXT: add x8, sp, #5630; CHECK-NEXT: ld1 { v0.b }[15], [x8]31; CHECK-NEXT: add v1.16b, v0.16b, v0.16b32; CHECK-NEXT: eor v0.16b, v0.16b, v1.16b33; CHECK-NEXT: ret34 %a1 = insertelement <16 x i16> poison, i16 %a, i16 035 %b1 = insertelement <16 x i16> %a1, i16 %b, i16 136 %c1 = insertelement <16 x i16> %b1, i16 %c, i16 237 %d1 = insertelement <16 x i16> %c1, i16 %d, i16 338 %e1 = insertelement <16 x i16> %d1, i16 %e, i16 439 %f1 = insertelement <16 x i16> %e1, i16 %f, i16 540 %g1 = insertelement <16 x i16> %f1, i16 %g, i16 641 %h1 = insertelement <16 x i16> %g1, i16 %h, i16 742 %i1 = insertelement <16 x i16> %h1, i16 %i, i16 843 %j1 = insertelement <16 x i16> %i1, i16 %j, i16 944 %k1 = insertelement <16 x i16> %j1, i16 %k, i16 1045 %l1 = insertelement <16 x i16> %k1, i16 %l, i16 1146 %m1 = insertelement <16 x i16> %l1, i16 %m, i16 1247 %n1 = insertelement <16 x i16> %m1, i16 %n, i16 1348 %o1 = insertelement <16 x i16> %n1, i16 %o, i16 1449 %p1 = insertelement <16 x i16> %o1, i16 %p, i16 1550 %t = trunc <16 x i16> %p1 to <16 x i8>51 %s = add <16 x i16> %p1, %p152 %t2 = trunc <16 x i16> %s to <16 x i8>53 %pro = xor <16 x i8> %t, %t254 ret <16 x i8> %pro55}56 57define <8 x i16> @lower_trunc_8xi16(i32 %a, i32 %b, i32 %c, i32 %d, i32 %e, i32 %f, i32 %g, i32 %h) {58; CHECK-LABEL: lower_trunc_8xi16:59; CHECK: // %bb.0:60; CHECK-NEXT: fmov s0, w061; CHECK-NEXT: mov v0.h[1], w162; CHECK-NEXT: mov v0.h[2], w263; CHECK-NEXT: mov v0.h[3], w364; CHECK-NEXT: mov v0.h[4], w465; CHECK-NEXT: mov v0.h[5], w566; CHECK-NEXT: mov v0.h[6], w667; CHECK-NEXT: mov v0.h[7], w768; CHECK-NEXT: add v1.8h, v0.8h, v0.8h69; CHECK-NEXT: eor v0.16b, v0.16b, v1.16b70; CHECK-NEXT: ret71 %a1 = insertelement <8 x i32> poison, i32 %a, i32 072 %b1 = insertelement <8 x i32> %a1, i32 %b, i32 173 %c1 = insertelement <8 x i32> %b1, i32 %c, i32 274 %d1 = insertelement <8 x i32> %c1, i32 %d, i32 375 %e1 = insertelement <8 x i32> %d1, i32 %e, i32 476 %f1 = insertelement <8 x i32> %e1, i32 %f, i32 577 %g1 = insertelement <8 x i32> %f1, i32 %g, i32 678 %h1 = insertelement <8 x i32> %g1, i32 %h, i32 779 %t = trunc <8 x i32> %h1 to <8 x i16>80 %s = add <8 x i32> %h1, %h181 %t2 = trunc <8 x i32> %s to <8 x i16>82 %o = xor <8 x i16> %t, %t283 ret <8 x i16> %o84}85 86define <4 x i32> @lower_trunc_4xi32(i64 %a, i64 %b, i64 %c, i64 %d) {87; CHECK-LABEL: lower_trunc_4xi32:88; CHECK: // %bb.0:89; CHECK-NEXT: fmov s0, w090; CHECK-NEXT: mov v0.s[1], w191; CHECK-NEXT: mov v0.s[2], w292; CHECK-NEXT: mov v0.s[3], w393; CHECK-NEXT: add v1.4s, v0.4s, v0.4s94; CHECK-NEXT: eor v0.16b, v0.16b, v1.16b95; CHECK-NEXT: ret96 %a1 = insertelement <4 x i64> poison, i64 %a, i64 097 %b1 = insertelement <4 x i64> %a1, i64 %b, i64 198 %c1 = insertelement <4 x i64> %b1, i64 %c, i64 299 %d1 = insertelement <4 x i64> %c1, i64 %d, i64 3100 %t = trunc <4 x i64> %d1 to <4 x i32>101 %s = add <4 x i64> %d1, %d1102 %t2 = trunc <4 x i64> %s to <4 x i32>103 %o = xor <4 x i32> %t, %t2104 ret <4 x i32> %o105}106 107define <8 x i32> @lower_trunc_8xi32(i64 %a, i64 %b, i64 %c, i64 %d, i64 %e, i64 %f, i64 %g, i64 %h) {108; CHECK-LABEL: lower_trunc_8xi32:109; CHECK: // %bb.0:110; CHECK-NEXT: fmov d0, x6111; CHECK-NEXT: fmov d1, x4112; CHECK-NEXT: fmov d2, x2113; CHECK-NEXT: fmov d3, x0114; CHECK-NEXT: mov v0.d[1], x7115; CHECK-NEXT: mov v1.d[1], x5116; CHECK-NEXT: mov v2.d[1], x3117; CHECK-NEXT: mov v3.d[1], x1118; CHECK-NEXT: uzp1 v1.4s, v1.4s, v0.4s119; CHECK-NEXT: uzp1 v2.4s, v3.4s, v2.4s120; CHECK-NEXT: add v3.4s, v1.4s, v1.4s121; CHECK-NEXT: add v0.4s, v2.4s, v2.4s122; CHECK-NEXT: eor v1.16b, v1.16b, v3.16b123; CHECK-NEXT: eor v0.16b, v2.16b, v0.16b124; CHECK-NEXT: ret125 %a1 = insertelement <8 x i64> poison, i64 %a, i64 0126 %b1 = insertelement <8 x i64> %a1, i64 %b, i64 1127 %c1 = insertelement <8 x i64> %b1, i64 %c, i64 2128 %d1 = insertelement <8 x i64> %c1, i64 %d, i64 3129 %e1 = insertelement <8 x i64> %d1, i64 %e, i64 4130 %f1 = insertelement <8 x i64> %e1, i64 %f, i64 5131 %g1 = insertelement <8 x i64> %f1, i64 %g, i64 6132 %h1 = insertelement <8 x i64> %g1, i64 %h, i64 7133 %t = trunc <8 x i64> %h1 to <8 x i32>134 %s = add <8 x i64> %h1, %h1135 %t2 = trunc <8 x i64> %s to <8 x i32>136 %o = xor <8 x i32> %t, %t2137 ret <8 x i32> %o138}139