204 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve2 < %s | FileCheck %s3 4;5; LDNT1B, LDNT1W, LDNT1H, LDNT1D: vector base + scalar offset6; ldnt1b { z0.s }, p0/z, [z0.s, x0]7;8 9; LDNT1B10define <vscale x 4 x i32> @gldnt1b_s(<vscale x 4 x i1> %pg, <vscale x 4 x i32> %base, i64 %offset) {11; CHECK-LABEL: gldnt1b_s:12; CHECK: // %bb.0:13; CHECK-NEXT: ldnt1b { z0.s }, p0/z, [z0.s, x0]14; CHECK-NEXT: ret15 %load = call <vscale x 4 x i8> @llvm.aarch64.sve.ldnt1.gather.scalar.offset.nxv4i8.nxv4i32(<vscale x 4 x i1> %pg,16 <vscale x 4 x i32> %base,17 i64 %offset)18 %res = zext <vscale x 4 x i8> %load to <vscale x 4 x i32>19 ret <vscale x 4 x i32> %res20}21 22define <vscale x 2 x i64> @gldnt1b_d(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %base, i64 %offset) {23; CHECK-LABEL: gldnt1b_d:24; CHECK: // %bb.0:25; CHECK-NEXT: ldnt1b { z0.d }, p0/z, [z0.d, x0]26; CHECK-NEXT: ret27 %load = call <vscale x 2 x i8> @llvm.aarch64.sve.ldnt1.gather.scalar.offset.nxv2i8.nxv2i64(<vscale x 2 x i1> %pg,28 <vscale x 2 x i64> %base,29 i64 %offset)30 %res = zext <vscale x 2 x i8> %load to <vscale x 2 x i64>31 ret <vscale x 2 x i64> %res32}33 34; LDNT1H35define <vscale x 4 x i32> @gldnt1h_s(<vscale x 4 x i1> %pg, <vscale x 4 x i32> %base, i64 %offset) {36; CHECK-LABEL: gldnt1h_s:37; CHECK: // %bb.0:38; CHECK-NEXT: ldnt1h { z0.s }, p0/z, [z0.s, x0]39; CHECK-NEXT: ret40 %load = call <vscale x 4 x i16> @llvm.aarch64.sve.ldnt1.gather.scalar.offset.nxv416.nxv4i32(<vscale x 4 x i1> %pg,41 <vscale x 4 x i32> %base,42 i64 %offset)43 %res = zext <vscale x 4 x i16> %load to <vscale x 4 x i32>44 ret <vscale x 4 x i32> %res45}46 47define <vscale x 2 x i64> @gldnt1h_d(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %base, i64 %offset) {48; CHECK-LABEL: gldnt1h_d:49; CHECK: // %bb.0:50; CHECK-NEXT: ldnt1h { z0.d }, p0/z, [z0.d, x0]51; CHECK-NEXT: ret52 %load = call <vscale x 2 x i16> @llvm.aarch64.sve.ldnt1.gather.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i1> %pg,53 <vscale x 2 x i64> %base,54 i64 %offset)55 %res = zext <vscale x 2 x i16> %load to <vscale x 2 x i64>56 ret <vscale x 2 x i64> %res57}58 59; LDNT1W60define <vscale x 4 x i32> @gldnt1w_s(<vscale x 4 x i1> %pg, <vscale x 4 x i32> %base, i64 %offset) {61; CHECK-LABEL: gldnt1w_s:62; CHECK: // %bb.0:63; CHECK-NEXT: ldnt1w { z0.s }, p0/z, [z0.s, x0]64; CHECK-NEXT: ret65 %load = call <vscale x 4 x i32> @llvm.aarch64.sve.ldnt1.gather.scalar.offset.nxv4i32.nxv4i32(<vscale x 4 x i1> %pg,66 <vscale x 4 x i32> %base,67 i64 %offset)68 ret <vscale x 4 x i32> %load69}70 71define <vscale x 4 x float> @gldnt1w_s_float(<vscale x 4 x i1> %pg, <vscale x 4 x i32> %base, i64 %offset) {72; CHECK-LABEL: gldnt1w_s_float:73; CHECK: // %bb.0:74; CHECK-NEXT: ldnt1w { z0.s }, p0/z, [z0.s, x0]75; CHECK-NEXT: ret76 %load = call <vscale x 4 x float> @llvm.aarch64.sve.ldnt1.gather.scalar.offset.nxv4f32.nxv4i32(<vscale x 4 x i1> %pg,77 <vscale x 4 x i32> %base,78 i64 %offset)79 ret <vscale x 4 x float> %load80}81 82define <vscale x 2 x i64> @gldnt1w_d(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %base, i64 %offset) {83; CHECK-LABEL: gldnt1w_d:84; CHECK: // %bb.0:85; CHECK-NEXT: ldnt1w { z0.d }, p0/z, [z0.d, x0]86; CHECK-NEXT: ret87 %load = call <vscale x 2 x i32> @llvm.aarch64.sve.ldnt1.gather.scalar.offset.nxv2i32.nxv2i64(<vscale x 2 x i1> %pg,88 <vscale x 2 x i64> %base,89 i64 %offset)90 %res = zext <vscale x 2 x i32> %load to <vscale x 2 x i64>91 ret <vscale x 2 x i64> %res92}93 94; LDNT1D95define <vscale x 2 x i64> @gldnt1d_d(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %base, i64 %offset) {96; CHECK-LABEL: gldnt1d_d:97; CHECK: // %bb.0:98; CHECK-NEXT: ldnt1d { z0.d }, p0/z, [z0.d, x0]99; CHECK-NEXT: ret100 %load = call <vscale x 2 x i64> @llvm.aarch64.sve.ldnt1.gather.scalar.offset.nxv2i64.nxv2i64(<vscale x 2 x i1> %pg,101 <vscale x 2 x i64> %base,102 i64 %offset)103 ret <vscale x 2 x i64> %load104}105 106; LDNT1D107define <vscale x 2 x double> @gldnt1d_d_double(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %base, i64 %offset) {108; CHECK-LABEL: gldnt1d_d_double:109; CHECK: // %bb.0:110; CHECK-NEXT: ldnt1d { z0.d }, p0/z, [z0.d, x0]111; CHECK-NEXT: ret112 %load = call <vscale x 2 x double> @llvm.aarch64.sve.ldnt1.gather.scalar.offset.nxv2f64.nxv2i64(<vscale x 2 x i1> %pg,113 <vscale x 2 x i64> %base,114 i64 %offset)115 ret <vscale x 2 x double> %load116}117 118;119; LDNT1SB, LDNT1SW, LDNT1SH, LDNT1SD: vector base + scalar offset120; ldnt1sb { z0.s }, p0/z, [z0.s, x0]121;122 123; LDNT1SB124define <vscale x 4 x i32> @gldnt1sb_s(<vscale x 4 x i1> %pg, <vscale x 4 x i32> %base, i64 %offset) {125; CHECK-LABEL: gldnt1sb_s:126; CHECK: // %bb.0:127; CHECK-NEXT: ldnt1sb { z0.s }, p0/z, [z0.s, x0]128; CHECK-NEXT: ret129 %load = call <vscale x 4 x i8> @llvm.aarch64.sve.ldnt1.gather.scalar.offset.nxv4i8.nxv4i32(<vscale x 4 x i1> %pg,130 <vscale x 4 x i32> %base,131 i64 %offset)132 %res = sext <vscale x 4 x i8> %load to <vscale x 4 x i32>133 ret <vscale x 4 x i32> %res134}135 136define <vscale x 2 x i64> @gldnt1sb_d(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %base, i64 %offset) {137; CHECK-LABEL: gldnt1sb_d:138; CHECK: // %bb.0:139; CHECK-NEXT: ldnt1sb { z0.d }, p0/z, [z0.d, x0]140; CHECK-NEXT: ret141 %load = call <vscale x 2 x i8> @llvm.aarch64.sve.ldnt1.gather.scalar.offset.nxv2i8.nxv2i64(<vscale x 2 x i1> %pg,142 <vscale x 2 x i64> %base,143 i64 %offset)144 %res = sext <vscale x 2 x i8> %load to <vscale x 2 x i64>145 ret <vscale x 2 x i64> %res146}147 148; LDNT1SH149define <vscale x 4 x i32> @gldnt1sh_s(<vscale x 4 x i1> %pg, <vscale x 4 x i32> %base, i64 %offset) {150; CHECK-LABEL: gldnt1sh_s:151; CHECK: // %bb.0:152; CHECK-NEXT: ldnt1sh { z0.s }, p0/z, [z0.s, x0]153; CHECK-NEXT: ret154 %load = call <vscale x 4 x i16> @llvm.aarch64.sve.ldnt1.gather.scalar.offset.nxv416.nxv4i32(<vscale x 4 x i1> %pg,155 <vscale x 4 x i32> %base,156 i64 %offset)157 %res = sext <vscale x 4 x i16> %load to <vscale x 4 x i32>158 ret <vscale x 4 x i32> %res159}160 161define <vscale x 2 x i64> @gldnt1sh_d(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %base, i64 %offset) {162; CHECK-LABEL: gldnt1sh_d:163; CHECK: // %bb.0:164; CHECK-NEXT: ldnt1sh { z0.d }, p0/z, [z0.d, x0]165; CHECK-NEXT: ret166 %load = call <vscale x 2 x i16> @llvm.aarch64.sve.ldnt1.gather.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i1> %pg,167 <vscale x 2 x i64> %base,168 i64 %offset)169 %res = sext <vscale x 2 x i16> %load to <vscale x 2 x i64>170 ret <vscale x 2 x i64> %res171}172 173; LDNT1SW174define <vscale x 2 x i64> @gldnt1sw_d(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %base, i64 %offset) {175; CHECK-LABEL: gldnt1sw_d:176; CHECK: // %bb.0:177; CHECK-NEXT: ldnt1sw { z0.d }, p0/z, [z0.d, x0]178; CHECK-NEXT: ret179 %load = call <vscale x 2 x i32> @llvm.aarch64.sve.ldnt1.gather.scalar.offset.nxv2i32.nxv2i64(<vscale x 2 x i1> %pg,180 <vscale x 2 x i64> %base,181 i64 %offset)182 %res = sext <vscale x 2 x i32> %load to <vscale x 2 x i64>183 ret <vscale x 2 x i64> %res184}185 186; LDNT1B/LDNT1SB187declare <vscale x 4 x i8> @llvm.aarch64.sve.ldnt1.gather.scalar.offset.nxv4i8.nxv4i32(<vscale x 4 x i1>, <vscale x 4 x i32>, i64)188declare <vscale x 2 x i8> @llvm.aarch64.sve.ldnt1.gather.scalar.offset.nxv2i8.nxv2i64(<vscale x 2 x i1>, <vscale x 2 x i64>, i64)189 190; LDNT1H/LDNT1SH191declare <vscale x 4 x i16> @llvm.aarch64.sve.ldnt1.gather.scalar.offset.nxv416.nxv4i32(<vscale x 4 x i1>, <vscale x 4 x i32>, i64)192declare <vscale x 2 x i16> @llvm.aarch64.sve.ldnt1.gather.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i1>, <vscale x 2 x i64>, i64)193 194; LDNT1W/LDNT1SW195declare <vscale x 4 x i32> @llvm.aarch64.sve.ldnt1.gather.scalar.offset.nxv4i32.nxv4i32(<vscale x 4 x i1>, <vscale x 4 x i32>, i64)196declare <vscale x 2 x i32> @llvm.aarch64.sve.ldnt1.gather.scalar.offset.nxv2i32.nxv2i64(<vscale x 2 x i1>, <vscale x 2 x i64>, i64)197 198declare <vscale x 4 x float> @llvm.aarch64.sve.ldnt1.gather.scalar.offset.nxv4f32.nxv4i32(<vscale x 4 x i1>, <vscale x 4 x i32>, i64)199 200; LDNT1D201declare <vscale x 2 x i64> @llvm.aarch64.sve.ldnt1.gather.scalar.offset.nxv2i64.nxv2i64(<vscale x 2 x i1>, <vscale x 2 x i64>, i64)202 203declare <vscale x 2 x double> @llvm.aarch64.sve.ldnt1.gather.scalar.offset.nxv2f64.nxv2i64(<vscale x 2 x i1>, <vscale x 2 x i64>, i64)204