202 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve -verify-machineinstrs < %s | FileCheck %s3 4;5; LDFF1B, LDFF1W, LDFF1H, LDFF1D: vector base + scalar offset (index)6; e.g. ldff1b { z0.d }, p0/z, [x0, z0.d]7;8 9; LDFF1B10define <vscale x 4 x i32> @gldff1b_s_scalar_offset(<vscale x 4 x i1> %pg, <vscale x 4 x i32> %base, i64 %offset) {11; CHECK-LABEL: gldff1b_s_scalar_offset:12; CHECK: // %bb.0:13; CHECK-NEXT: ldff1b { z0.s }, p0/z, [x0, z0.s, uxtw]14; CHECK-NEXT: ret15 %load = call <vscale x 4 x i8> @llvm.aarch64.sve.ldff1.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> @gldff1b_d_scalar_offset(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %base, i64 %offset) {23; CHECK-LABEL: gldff1b_d_scalar_offset:24; CHECK: // %bb.0:25; CHECK-NEXT: ldff1b { z0.d }, p0/z, [x0, z0.d]26; CHECK-NEXT: ret27 %load = call <vscale x 2 x i8> @llvm.aarch64.sve.ldff1.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; LDFF1H35define <vscale x 4 x i32> @gldff1h_s_scalar_offset(<vscale x 4 x i1> %pg, <vscale x 4 x i32> %base, i64 %offset) {36; CHECK-LABEL: gldff1h_s_scalar_offset:37; CHECK: // %bb.0:38; CHECK-NEXT: ldff1h { z0.s }, p0/z, [x0, z0.s, uxtw]39; CHECK-NEXT: ret40 %load = call <vscale x 4 x i16> @llvm.aarch64.sve.ldff1.gather.scalar.offset.nxv4i16.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> @gldff1h_d_scalar_offset(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %base, i64 %offset) {48; CHECK-LABEL: gldff1h_d_scalar_offset:49; CHECK: // %bb.0:50; CHECK-NEXT: ldff1h { z0.d }, p0/z, [x0, z0.d]51; CHECK-NEXT: ret52 %load = call <vscale x 2 x i16> @llvm.aarch64.sve.ldff1.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; LDFF1W60define <vscale x 4 x i32> @gldff1w_s_scalar_offset(<vscale x 4 x i1> %pg, <vscale x 4 x i32> %base, i64 %offset) {61; CHECK-LABEL: gldff1w_s_scalar_offset:62; CHECK: // %bb.0:63; CHECK-NEXT: ldff1w { z0.s }, p0/z, [x0, z0.s, uxtw]64; CHECK-NEXT: ret65 %load = call <vscale x 4 x i32> @llvm.aarch64.sve.ldff1.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 2 x i64> @gldff1w_d_scalar_offset(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %base, i64 %offset) {72; CHECK-LABEL: gldff1w_d_scalar_offset:73; CHECK: // %bb.0:74; CHECK-NEXT: ldff1w { z0.d }, p0/z, [x0, z0.d]75; CHECK-NEXT: ret76 %load = call <vscale x 2 x i32> @llvm.aarch64.sve.ldff1.gather.scalar.offset.nxv2i32.nxv2i64(<vscale x 2 x i1> %pg,77 <vscale x 2 x i64> %base,78 i64 %offset)79 %res = zext <vscale x 2 x i32> %load to <vscale x 2 x i64>80 ret <vscale x 2 x i64> %res81}82 83define <vscale x 4 x float> @gldff1w_s_scalar_offset_float(<vscale x 4 x i1> %pg, <vscale x 4 x i32> %base, i64 %offset) {84; CHECK-LABEL: gldff1w_s_scalar_offset_float:85; CHECK: // %bb.0:86; CHECK-NEXT: ldff1w { z0.s }, p0/z, [x0, z0.s, uxtw]87; CHECK-NEXT: ret88 %load = call <vscale x 4 x float> @llvm.aarch64.sve.ldff1.gather.scalar.offset.nxv4f32.nxv4i32(<vscale x 4 x i1> %pg,89 <vscale x 4 x i32> %base,90 i64 %offset)91 ret <vscale x 4 x float> %load92}93 94; LDFF1D95define <vscale x 2 x i64> @gldff1d_d_scalar_offset(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %base, i64 %offset) {96; CHECK-LABEL: gldff1d_d_scalar_offset:97; CHECK: // %bb.0:98; CHECK-NEXT: ldff1d { z0.d }, p0/z, [x0, z0.d]99; CHECK-NEXT: ret100 %load = call <vscale x 2 x i64> @llvm.aarch64.sve.ldff1.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 106define <vscale x 2 x double> @gldff1d_d_scalar_offset_double(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %base, i64 %offset) {107; CHECK-LABEL: gldff1d_d_scalar_offset_double:108; CHECK: // %bb.0:109; CHECK-NEXT: ldff1d { z0.d }, p0/z, [x0, z0.d]110; CHECK-NEXT: ret111 %load = call <vscale x 2 x double> @llvm.aarch64.sve.ldff1.gather.scalar.offset.nxv2f64.nxv2i64(<vscale x 2 x i1> %pg,112 <vscale x 2 x i64> %base,113 i64 %offset)114 ret <vscale x 2 x double> %load115}116 117; LDFF1SB, LDFF1SW, LDFF1SH: vector base + scalar offset (index)118; e.g. ldff1b { z0.d }, p0/z, [x0, z0.d]119;120 121; LDFF1SB122define <vscale x 4 x i32> @gldff1sb_s_scalar_offset(<vscale x 4 x i1> %pg, <vscale x 4 x i32> %base, i64 %offset) {123; CHECK-LABEL: gldff1sb_s_scalar_offset:124; CHECK: // %bb.0:125; CHECK-NEXT: ldff1sb { z0.s }, p0/z, [x0, z0.s, uxtw]126; CHECK-NEXT: ret127 %load = call <vscale x 4 x i8> @llvm.aarch64.sve.ldff1.gather.scalar.offset.nxv4i8.nxv4i32(<vscale x 4 x i1> %pg,128 <vscale x 4 x i32> %base,129 i64 %offset)130 %res = sext <vscale x 4 x i8> %load to <vscale x 4 x i32>131 ret <vscale x 4 x i32> %res132}133 134define <vscale x 2 x i64> @gldff1sb_d_scalar_offset(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %base, i64 %offset) {135; CHECK-LABEL: gldff1sb_d_scalar_offset:136; CHECK: // %bb.0:137; CHECK-NEXT: ldff1sb { z0.d }, p0/z, [x0, z0.d]138; CHECK-NEXT: ret139 %load = call <vscale x 2 x i8> @llvm.aarch64.sve.ldff1.gather.scalar.offset.nxv2i8.nxv2i64(<vscale x 2 x i1> %pg,140 <vscale x 2 x i64> %base,141 i64 %offset)142 %res = sext <vscale x 2 x i8> %load to <vscale x 2 x i64>143 ret <vscale x 2 x i64> %res144}145 146; LDFF1SH147define <vscale x 4 x i32> @gldff1sh_s_scalar_offset(<vscale x 4 x i1> %pg, <vscale x 4 x i32> %base, i64 %offset) {148; CHECK-LABEL: gldff1sh_s_scalar_offset:149; CHECK: // %bb.0:150; CHECK-NEXT: ldff1sh { z0.s }, p0/z, [x0, z0.s, uxtw]151; CHECK-NEXT: ret152 %load = call <vscale x 4 x i16> @llvm.aarch64.sve.ldff1.gather.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i1> %pg,153 <vscale x 4 x i32> %base,154 i64 %offset)155 %res = sext <vscale x 4 x i16> %load to <vscale x 4 x i32>156 ret <vscale x 4 x i32> %res157}158 159define <vscale x 2 x i64> @gldff1sh_d_scalar_offset(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %base, i64 %offset) {160; CHECK-LABEL: gldff1sh_d_scalar_offset:161; CHECK: // %bb.0:162; CHECK-NEXT: ldff1sh { z0.d }, p0/z, [x0, z0.d]163; CHECK-NEXT: ret164 %load = call <vscale x 2 x i16> @llvm.aarch64.sve.ldff1.gather.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i1> %pg,165 <vscale x 2 x i64> %base,166 i64 %offset)167 %res = sext <vscale x 2 x i16> %load to <vscale x 2 x i64>168 ret <vscale x 2 x i64> %res169}170 171; LDFF1SW172define <vscale x 2 x i64> @gldff1sw_d_scalar_offset(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %base, i64 %offset) {173; CHECK-LABEL: gldff1sw_d_scalar_offset:174; CHECK: // %bb.0:175; CHECK-NEXT: ldff1sw { z0.d }, p0/z, [x0, z0.d]176; CHECK-NEXT: ret177 %load = call <vscale x 2 x i32> @llvm.aarch64.sve.ldff1.gather.scalar.offset.nxv2i32.nxv2i64(<vscale x 2 x i1> %pg,178 <vscale x 2 x i64> %base,179 i64 %offset)180 %res = sext <vscale x 2 x i32> %load to <vscale x 2 x i64>181 ret <vscale x 2 x i64> %res182}183 184; LDFF1B/LDFF1SB185declare <vscale x 4 x i8> @llvm.aarch64.sve.ldff1.gather.scalar.offset.nxv4i8.nxv4i32(<vscale x 4 x i1>, <vscale x 4 x i32>, i64)186declare <vscale x 2 x i8> @llvm.aarch64.sve.ldff1.gather.scalar.offset.nxv2i8.nxv2i64(<vscale x 2 x i1>, <vscale x 2 x i64>, i64)187 188; LDFF1H/LDFF1SH189declare <vscale x 4 x i16> @llvm.aarch64.sve.ldff1.gather.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i1>, <vscale x 4 x i32>, i64)190declare <vscale x 2 x i16> @llvm.aarch64.sve.ldff1.gather.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i1>, <vscale x 2 x i64>, i64)191 192; LDFF1W/LDFF1SW193declare <vscale x 4 x i32> @llvm.aarch64.sve.ldff1.gather.scalar.offset.nxv4i32.nxv4i32(<vscale x 4 x i1>, <vscale x 4 x i32>, i64)194declare <vscale x 2 x i32> @llvm.aarch64.sve.ldff1.gather.scalar.offset.nxv2i32.nxv2i64(<vscale x 2 x i1>, <vscale x 2 x i64>, i64)195 196declare <vscale x 4 x float> @llvm.aarch64.sve.ldff1.gather.scalar.offset.nxv4f32.nxv4i32(<vscale x 4 x i1>, <vscale x 4 x i32>, i64)197 198; LDFF1D199declare <vscale x 2 x i64> @llvm.aarch64.sve.ldff1.gather.scalar.offset.nxv2i64.nxv2i64(<vscale x 2 x i1>, <vscale x 2 x i64>, i64)200 201declare <vscale x 2 x double> @llvm.aarch64.sve.ldff1.gather.scalar.offset.nxv2f64.nxv2i64(<vscale x 2 x i1>, <vscale x 2 x i64>, i64)202