218 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve < %s | FileCheck %s3 4; For all the functions below should the operation is a nop5define <vscale x 8 x i8> @trunc_i16toi8(<vscale x 8 x i16> %in) {6; CHECK-LABEL: trunc_i16toi8:7; CHECK: // %bb.0: // %entry8; CHECK-NEXT: ret9entry:10 %out = trunc <vscale x 8 x i16> %in to <vscale x 8 x i8>11 ret <vscale x 8 x i8> %out12}13 14define <vscale x 4 x i8> @trunc_i32toi8(<vscale x 4 x i32> %in) {15; CHECK-LABEL: trunc_i32toi8:16; CHECK: // %bb.0: // %entry17; CHECK-NEXT: ret18entry:19 %out = trunc <vscale x 4 x i32> %in to <vscale x 4 x i8>20 ret <vscale x 4 x i8> %out21}22 23define <vscale x 2 x i8> @trunc_i64toi8(<vscale x 2 x i64> %in) {24; CHECK-LABEL: trunc_i64toi8:25; CHECK: // %bb.0: // %entry26; CHECK-NEXT: ret27entry:28 %out = trunc <vscale x 2 x i64> %in to <vscale x 2 x i8>29 ret <vscale x 2 x i8> %out30}31 32define <vscale x 4 x i16> @trunc_i32toi16(<vscale x 4 x i32> %in) {33; CHECK-LABEL: trunc_i32toi16:34; CHECK: // %bb.0: // %entry35; CHECK-NEXT: ret36entry:37 %out = trunc <vscale x 4 x i32> %in to <vscale x 4 x i16>38 ret <vscale x 4 x i16> %out39}40 41define <vscale x 2 x i16> @trunc_i64toi16(<vscale x 2 x i64> %in) {42; CHECK-LABEL: trunc_i64toi16:43; CHECK: // %bb.0: // %entry44; CHECK-NEXT: ret45entry:46 %out = trunc <vscale x 2 x i64> %in to <vscale x 2 x i16>47 ret <vscale x 2 x i16> %out48}49 50define <vscale x 2 x i32> @trunc_i64toi32(<vscale x 2 x i64> %in) {51; CHECK-LABEL: trunc_i64toi32:52; CHECK: // %bb.0: // %entry53; CHECK-NEXT: ret54entry:55 %out = trunc <vscale x 2 x i64> %in to <vscale x 2 x i32>56 ret <vscale x 2 x i32> %out57}58 59; Truncating to i1 requires convert it to a cmp60 61define <vscale x 2 x i1> @trunc_i64toi1(<vscale x 2 x i64> %in) {62; CHECK-LABEL: trunc_i64toi1:63; CHECK: // %bb.0: // %entry64; CHECK-NEXT: and z0.d, z0.d, #0x165; CHECK-NEXT: ptrue p0.d66; CHECK-NEXT: cmpne p0.d, p0/z, z0.d, #067; CHECK-NEXT: ret68entry:69 %out = trunc <vscale x 2 x i64> %in to <vscale x 2 x i1>70 ret <vscale x 2 x i1> %out71}72 73define <vscale x 4 x i1> @trunc_i64toi1_split(<vscale x 4 x i64> %in) {74; CHECK-LABEL: trunc_i64toi1_split:75; CHECK: // %bb.0: // %entry76; CHECK-NEXT: and z1.d, z1.d, #0x177; CHECK-NEXT: and z0.d, z0.d, #0x178; CHECK-NEXT: ptrue p0.d79; CHECK-NEXT: cmpne p1.d, p0/z, z1.d, #080; CHECK-NEXT: cmpne p0.d, p0/z, z0.d, #081; CHECK-NEXT: uzp1 p0.s, p0.s, p1.s82; CHECK-NEXT: ret83entry:84 %out = trunc <vscale x 4 x i64> %in to <vscale x 4 x i1>85 ret <vscale x 4 x i1> %out86}87 88define <vscale x 8 x i1> @trunc_i64toi1_split2(<vscale x 8 x i64> %in) {89; CHECK-LABEL: trunc_i64toi1_split2:90; CHECK: // %bb.0: // %entry91; CHECK-NEXT: and z3.d, z3.d, #0x192; CHECK-NEXT: and z2.d, z2.d, #0x193; CHECK-NEXT: and z1.d, z1.d, #0x194; CHECK-NEXT: and z0.d, z0.d, #0x195; CHECK-NEXT: ptrue p0.d96; CHECK-NEXT: cmpne p1.d, p0/z, z3.d, #097; CHECK-NEXT: cmpne p2.d, p0/z, z2.d, #098; CHECK-NEXT: cmpne p3.d, p0/z, z1.d, #099; CHECK-NEXT: cmpne p0.d, p0/z, z0.d, #0100; CHECK-NEXT: uzp1 p1.s, p2.s, p1.s101; CHECK-NEXT: uzp1 p0.s, p0.s, p3.s102; CHECK-NEXT: uzp1 p0.h, p0.h, p1.h103; CHECK-NEXT: ret104entry:105 %out = trunc <vscale x 8 x i64> %in to <vscale x 8 x i1>106 ret <vscale x 8 x i1> %out107}108 109define <vscale x 16 x i1> @trunc_i64toi1_split3(<vscale x 16 x i64> %in) {110; CHECK-LABEL: trunc_i64toi1_split3:111; CHECK: // %bb.0: // %entry112; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill113; CHECK-NEXT: addvl sp, sp, #-1114; CHECK-NEXT: str p7, [sp, #4, mul vl] // 2-byte Spill115; CHECK-NEXT: str p6, [sp, #5, mul vl] // 2-byte Spill116; CHECK-NEXT: str p5, [sp, #6, mul vl] // 2-byte Spill117; CHECK-NEXT: str p4, [sp, #7, mul vl] // 2-byte Spill118; CHECK-NEXT: .cfi_escape 0x0f, 0x08, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x38, 0x1e, 0x22 // sp + 16 + 8 * VG119; CHECK-NEXT: .cfi_offset w29, -16120; CHECK-NEXT: and z7.d, z7.d, #0x1121; CHECK-NEXT: and z6.d, z6.d, #0x1122; CHECK-NEXT: and z5.d, z5.d, #0x1123; CHECK-NEXT: and z4.d, z4.d, #0x1124; CHECK-NEXT: and z3.d, z3.d, #0x1125; CHECK-NEXT: and z2.d, z2.d, #0x1126; CHECK-NEXT: and z1.d, z1.d, #0x1127; CHECK-NEXT: and z0.d, z0.d, #0x1128; CHECK-NEXT: ptrue p0.d129; CHECK-NEXT: cmpne p1.d, p0/z, z7.d, #0130; CHECK-NEXT: cmpne p2.d, p0/z, z6.d, #0131; CHECK-NEXT: cmpne p3.d, p0/z, z5.d, #0132; CHECK-NEXT: cmpne p4.d, p0/z, z4.d, #0133; CHECK-NEXT: cmpne p5.d, p0/z, z3.d, #0134; CHECK-NEXT: cmpne p6.d, p0/z, z2.d, #0135; CHECK-NEXT: cmpne p7.d, p0/z, z1.d, #0136; CHECK-NEXT: cmpne p0.d, p0/z, z0.d, #0137; CHECK-NEXT: uzp1 p1.s, p2.s, p1.s138; CHECK-NEXT: uzp1 p2.s, p4.s, p3.s139; CHECK-NEXT: ldr p4, [sp, #7, mul vl] // 2-byte Reload140; CHECK-NEXT: uzp1 p3.s, p6.s, p5.s141; CHECK-NEXT: ldr p6, [sp, #5, mul vl] // 2-byte Reload142; CHECK-NEXT: uzp1 p0.s, p0.s, p7.s143; CHECK-NEXT: ldr p7, [sp, #4, mul vl] // 2-byte Reload144; CHECK-NEXT: uzp1 p1.h, p2.h, p1.h145; CHECK-NEXT: ldr p5, [sp, #6, mul vl] // 2-byte Reload146; CHECK-NEXT: uzp1 p0.h, p0.h, p3.h147; CHECK-NEXT: uzp1 p0.b, p0.b, p1.b148; CHECK-NEXT: addvl sp, sp, #1149; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload150; CHECK-NEXT: ret151entry:152 %out = trunc <vscale x 16 x i64> %in to <vscale x 16 x i1>153 ret <vscale x 16 x i1> %out154}155 156 157define <vscale x 4 x i1> @trunc_i32toi1(<vscale x 4 x i32> %in) {158; CHECK-LABEL: trunc_i32toi1:159; CHECK: // %bb.0: // %entry160; CHECK-NEXT: and z0.s, z0.s, #0x1161; CHECK-NEXT: ptrue p0.s162; CHECK-NEXT: cmpne p0.s, p0/z, z0.s, #0163; CHECK-NEXT: ret164entry:165 %out = trunc <vscale x 4 x i32> %in to <vscale x 4 x i1>166 ret <vscale x 4 x i1> %out167}168 169define <vscale x 8 x i1> @trunc_i16toi1(<vscale x 8 x i16> %in) {170; CHECK-LABEL: trunc_i16toi1:171; CHECK: // %bb.0: // %entry172; CHECK-NEXT: and z0.h, z0.h, #0x1173; CHECK-NEXT: ptrue p0.h174; CHECK-NEXT: cmpne p0.h, p0/z, z0.h, #0175; CHECK-NEXT: ret176entry:177 %out = trunc <vscale x 8 x i16> %in to <vscale x 8 x i1>178 ret <vscale x 8 x i1> %out179}180 181define <vscale x 16 x i1> @trunc_i8toi1(<vscale x 16 x i8> %in) {182; CHECK-LABEL: trunc_i8toi1:183; CHECK: // %bb.0: // %entry184; CHECK-NEXT: and z0.b, z0.b, #0x1185; CHECK-NEXT: ptrue p0.b186; CHECK-NEXT: cmpne p0.b, p0/z, z0.b, #0187; CHECK-NEXT: ret188entry:189 %out = trunc <vscale x 16 x i8> %in to <vscale x 16 x i1>190 ret <vscale x 16 x i1> %out191}192 193define <vscale x 1 x i1> @trunc_nxv1i32_to_nxv1i1(<vscale x 1 x i32> %in) {194; CHECK-LABEL: trunc_nxv1i32_to_nxv1i1:195; CHECK: // %bb.0:196; CHECK-NEXT: and z0.s, z0.s, #0x1197; CHECK-NEXT: ptrue p0.s198; CHECK-NEXT: cmpne p0.s, p0/z, z0.s, #0199; CHECK-NEXT: punpklo p0.h, p0.b200; CHECK-NEXT: punpklo p0.h, p0.b201; CHECK-NEXT: ret202 %out = trunc <vscale x 1 x i32> %in to <vscale x 1 x i1>203 ret <vscale x 1 x i1> %out204}205 206define void @trunc_promoteIntRes(<vscale x 4 x i64> %0, ptr %ptr) {207; CHECK-LABEL: trunc_promoteIntRes:208; CHECK: // %bb.0: // %entry209; CHECK-NEXT: uzp1 z0.s, z0.s, z1.s210; CHECK-NEXT: ptrue p0.s211; CHECK-NEXT: st1h { z0.s }, p0, [x0]212; CHECK-NEXT: ret213entry:214 %1 = trunc <vscale x 4 x i64> %0 to <vscale x 4 x i16>215 store <vscale x 4 x i16> %1, ptr %ptr, align 2216 ret void217}218