210 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 22; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve < %s | FileCheck %s3; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sme -force-streaming < %s | FileCheck %s4 5;6; Converting to svbool_t (<vscale x 16 x i1>)7;8 9define <vscale x 16 x i1> @reinterpret_bool_from_b(<vscale x 16 x i1> %pg) {10; CHECK-LABEL: reinterpret_bool_from_b:11; CHECK: // %bb.0:12; CHECK-NEXT: ret13 %out = call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv16i1(<vscale x 16 x i1> %pg)14 ret <vscale x 16 x i1> %out15}16 17define <vscale x 16 x i1> @reinterpret_bool_from_h(<vscale x 8 x i1> %pg) {18; CHECK-LABEL: reinterpret_bool_from_h:19; CHECK: // %bb.0:20; CHECK-NEXT: ptrue p1.h21; CHECK-NEXT: and p0.b, p0/z, p0.b, p1.b22; CHECK-NEXT: ret23 %out = call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv8i1(<vscale x 8 x i1> %pg)24 ret <vscale x 16 x i1> %out25}26 27define <vscale x 16 x i1> @reinterpret_bool_from_s(<vscale x 4 x i1> %pg) {28; CHECK-LABEL: reinterpret_bool_from_s:29; CHECK: // %bb.0:30; CHECK-NEXT: ptrue p1.s31; CHECK-NEXT: and p0.b, p0/z, p0.b, p1.b32; CHECK-NEXT: ret33 %out = call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> %pg)34 ret <vscale x 16 x i1> %out35}36 37define <vscale x 16 x i1> @reinterpret_bool_from_d(<vscale x 2 x i1> %pg) {38; CHECK-LABEL: reinterpret_bool_from_d:39; CHECK: // %bb.0:40; CHECK-NEXT: ptrue p1.d41; CHECK-NEXT: and p0.b, p0/z, p0.b, p1.b42; CHECK-NEXT: ret43 %out = call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %pg)44 ret <vscale x 16 x i1> %out45}46 47define <vscale x 16 x i1> @reinterpret_bool_from_q(<vscale x 1 x i1> %arg) {48; CHECK-LABEL: reinterpret_bool_from_q:49; CHECK: // %bb.0:50; CHECK-NEXT: ptrue p1.d51; CHECK-NEXT: punpklo p1.h, p1.b52; CHECK-NEXT: and p0.b, p0/z, p0.b, p1.b53; CHECK-NEXT: ret54 %res = call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv1i1(<vscale x 1 x i1> %arg)55 ret <vscale x 16 x i1> %res56}57 58define <vscale x 16 x i1> @reinterpret_bool_from_svcount(target("aarch64.svcount") %pg) "target-features"="+sme2" {59; CHECK-LABEL: reinterpret_bool_from_svcount:60; CHECK: // %bb.0:61; CHECK-NEXT: ret62 %out = call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.taarch64.svcountt(target("aarch64.svcount") %pg)63 ret <vscale x 16 x i1> %out64}65 66;67; Converting from svbool_t68;69 70define <vscale x 16 x i1> @reinterpret_bool_to_b(<vscale x 16 x i1> %pg) {71; CHECK-LABEL: reinterpret_bool_to_b:72; CHECK: // %bb.0:73; CHECK-NEXT: ret74 %out = call <vscale x 16 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv16i1(<vscale x 16 x i1> %pg)75 ret <vscale x 16 x i1> %out76}77 78define <vscale x 8 x i1> @reinterpret_bool_to_h(<vscale x 16 x i1> %pg) {79; CHECK-LABEL: reinterpret_bool_to_h:80; CHECK: // %bb.0:81; CHECK-NEXT: ret82 %out = call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> %pg)83 ret <vscale x 8 x i1> %out84}85 86define <vscale x 4 x i1> @reinterpret_bool_to_s(<vscale x 16 x i1> %pg) {87; CHECK-LABEL: reinterpret_bool_to_s:88; CHECK: // %bb.0:89; CHECK-NEXT: ret90 %out = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)91 ret <vscale x 4 x i1> %out92}93 94define <vscale x 2 x i1> @reinterpret_bool_to_d(<vscale x 16 x i1> %pg) {95; CHECK-LABEL: reinterpret_bool_to_d:96; CHECK: // %bb.0:97; CHECK-NEXT: ret98 %out = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)99 ret <vscale x 2 x i1> %out100}101 102define <vscale x 1 x i1> @reinterpret_bool_to_q(<vscale x 16 x i1> %pg) {103; CHECK-LABEL: reinterpret_bool_to_q:104; CHECK: // %bb.0:105; CHECK-NEXT: ret106 %out = call <vscale x 1 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv1i1(<vscale x 16 x i1> %pg)107 ret <vscale x 1 x i1> %out108}109 110define target("aarch64.svcount") @reinterpret_bool_to_svcount(<vscale x 16 x i1> %pg) "target-features"="+sme2" {111; CHECK-LABEL: reinterpret_bool_to_svcount:112; CHECK: // %bb.0:113; CHECK-NEXT: ret114 %out = call target("aarch64.svcount") @llvm.aarch64.sve.convert.from.svbool.taarch64.svcountt(<vscale x 16 x i1> %pg)115 ret target("aarch64.svcount") %out116}117 118 119; Reinterpreting a ptrue should not introduce an `and` instruction.120define <vscale x 16 x i1> @reinterpret_ptrue() {121; CHECK-LABEL: reinterpret_ptrue:122; CHECK: // %bb.0:123; CHECK-NEXT: ptrue p0.h124; CHECK-NEXT: ret125 %in = tail call <vscale x 8 x i1> @llvm.aarch64.sve.ptrue.nxv8i1(i32 31)126 %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv8i1(<vscale x 8 x i1> %in)127 ret <vscale x 16 x i1> %out128}129 130; Reinterpreting a comparison not introduce an `and` instruction.131define <vscale x 16 x i1> @reinterpret_cmpgt(<vscale x 8 x i1> %p, <vscale x 8 x i16> %a, <vscale x 8 x i16> %b) {132; CHECK-LABEL: reinterpret_cmpgt:133; CHECK: // %bb.0:134; CHECK-NEXT: cmpgt p0.h, p0/z, z0.h, z1.h135; CHECK-NEXT: ret136 %1 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.cmpgt.nxv8i16(<vscale x 8 x i1> %p, <vscale x 8 x i16> %a, <vscale x 8 x i16> %b)137 %2 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv8i1(<vscale x 8 x i1> %1)138 ret <vscale x 16 x i1> %2139}140 141; The first reinterpret should prevent the second one from being simplified as a nop142define <vscale x 16 x i1> @chained_reinterpret() {143; CHECK-LABEL: chained_reinterpret:144; CHECK: // %bb.0:145; CHECK-NEXT: ptrue p0.d146; CHECK-NEXT: ret147 %in = tail call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 31)148 %cast2 = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %in)149 %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %cast2)150 ret <vscale x 16 x i1> %out151}152 153define <vscale x 16 x i1> @reinterpret_scalar_bool_h(i1 %x){154; CHECK-LABEL: reinterpret_scalar_bool_h:155; CHECK: // %bb.0:156; CHECK-NEXT: // kill: def $w0 killed $w0 def $x0157; CHECK-NEXT: sbfx x8, x0, #0, #1158; CHECK-NEXT: whilelo p0.h, xzr, x8159; CHECK-NEXT: ret160 %.splatinsert = insertelement <vscale x 8 x i1> poison, i1 %x, i64 0161 %.splat = shufflevector <vscale x 8 x i1> %.splatinsert, <vscale x 8 x i1> poison, <vscale x 8 x i32> zeroinitializer162 %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv8i1(<vscale x 8 x i1> %.splat)163 ret <vscale x 16 x i1> %out164}165 166define <vscale x 16 x i1> @reinterpret_scalar_bool_s(i1 %x){167; CHECK-LABEL: reinterpret_scalar_bool_s:168; CHECK: // %bb.0:169; CHECK-NEXT: // kill: def $w0 killed $w0 def $x0170; CHECK-NEXT: sbfx x8, x0, #0, #1171; CHECK-NEXT: whilelo p0.s, xzr, x8172; CHECK-NEXT: ret173 %.splatinsert = insertelement <vscale x 4 x i1> poison, i1 %x, i64 0174 %.splat = shufflevector <vscale x 4 x i1> %.splatinsert, <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer175 %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> %.splat)176 ret <vscale x 16 x i1> %out177}178 179define <vscale x 16 x i1> @reinterpret_scalar_bool_q(i1 %x){180; CHECK-LABEL: reinterpret_scalar_bool_q:181; CHECK: // %bb.0:182; CHECK-NEXT: // kill: def $w0 killed $w0 def $x0183; CHECK-NEXT: sbfx x8, x0, #0, #1184; CHECK-NEXT: whilelo p0.d, xzr, x8185; CHECK-NEXT: ret186 %.splatinsert = insertelement <vscale x 2 x i1> poison, i1 %x, i64 0187 %.splat = shufflevector <vscale x 2 x i1> %.splatinsert, <vscale x 2 x i1> poison, <vscale x 2 x i32> zeroinitializer188 %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %.splat)189 ret <vscale x 16 x i1> %out190}191 192 193declare <vscale x 8 x i1> @llvm.aarch64.sve.ptrue.nxv8i1(i32 immarg)194declare <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 immarg)195declare <vscale x 8 x i1> @llvm.aarch64.sve.cmpgt.nxv8i16(<vscale x 8 x i1>, <vscale x 8 x i16>, <vscale x 8 x i16>)196 197declare <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv16i1(<vscale x 16 x i1>)198declare <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv8i1(<vscale x 8 x i1>)199declare <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1>)200declare <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1>)201declare <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv1i1(<vscale x 1 x i1>)202declare <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.taarch64.svcountt(target("aarch64.svcount"))203 204declare <vscale x 16 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv16i1(<vscale x 16 x i1>)205declare <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1>)206declare <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1>)207declare <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1>)208declare <vscale x 1 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv1i1(<vscale x 16 x i1>)209declare target("aarch64.svcount") @llvm.aarch64.sve.convert.from.svbool.taarch64.svcountt(<vscale x 16 x i1>)210