144 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;5; ST1B, ST1W, ST1H, ST1D: vector base + scalar offset6; e.g. st1h { z0.s }, p0, [x0, z1.d]7;8 9; ST1B10define void @sst1b_s_scalar_offset(<vscale x 4 x i32> %data, <vscale x 4 x i1> %pg, <vscale x 4 x i32> %base, i64 %offset) {11; CHECK-LABEL: sst1b_s_scalar_offset:12; CHECK: // %bb.0:13; CHECK-NEXT: st1b { z0.s }, p0, [x0, z1.s, uxtw]14; CHECK-NEXT: ret15 %data_trunc = trunc <vscale x 4 x i32> %data to <vscale x 4 x i8>16 call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv4i8.nxv4i32(<vscale x 4 x i8> %data_trunc,17 <vscale x 4 x i1> %pg,18 <vscale x 4 x i32> %base,19 i64 %offset)20 ret void21}22 23define void @sst1b_d_scalar_offset(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, <vscale x 2 x i64> %base, i64 %offset) {24; CHECK-LABEL: sst1b_d_scalar_offset:25; CHECK: // %bb.0:26; CHECK-NEXT: st1b { z0.d }, p0, [x0, z1.d]27; CHECK-NEXT: ret28 %data_trunc = trunc <vscale x 2 x i64> %data to <vscale x 2 x i8>29 call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i8.nxv2i64(<vscale x 2 x i8> %data_trunc,30 <vscale x 2 x i1> %pg,31 <vscale x 2 x i64> %base,32 i64 %offset)33 ret void34}35 36; ST1H37define void @sst1h_s_scalar_offset(<vscale x 4 x i32> %data, <vscale x 4 x i1> %pg, <vscale x 4 x i32> %base, i64 %offset) {38; CHECK-LABEL: sst1h_s_scalar_offset:39; CHECK: // %bb.0:40; CHECK-NEXT: st1h { z0.s }, p0, [x0, z1.s, uxtw]41; CHECK-NEXT: ret42 %data_trunc = trunc <vscale x 4 x i32> %data to <vscale x 4 x i16>43 call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i16> %data_trunc,44 <vscale x 4 x i1> %pg,45 <vscale x 4 x i32> %base,46 i64 %offset)47 ret void48}49 50define void @sst1h_d_scalar_offset(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, <vscale x 2 x i64> %base, i64 %offset) {51; CHECK-LABEL: sst1h_d_scalar_offset:52; CHECK: // %bb.0:53; CHECK-NEXT: st1h { z0.d }, p0, [x0, z1.d]54; CHECK-NEXT: ret55 %data_trunc = trunc <vscale x 2 x i64> %data to <vscale x 2 x i16>56 call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i16> %data_trunc,57 <vscale x 2 x i1> %pg,58 <vscale x 2 x i64> %base,59 i64 %offset)60 ret void61}62 63; ST1W64define void @sst1w_s_scalar_offset(<vscale x 4 x i32> %data, <vscale x 4 x i1> %pg, <vscale x 4 x i32> %base, i64 %offset) {65; CHECK-LABEL: sst1w_s_scalar_offset:66; CHECK: // %bb.0:67; CHECK-NEXT: st1w { z0.s }, p0, [x0, z1.s, uxtw]68; CHECK-NEXT: ret69 call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv4i32.nxv4i32(<vscale x 4 x i32> %data,70 <vscale x 4 x i1> %pg,71 <vscale x 4 x i32> %base,72 i64 %offset)73 ret void74}75 76define void @sst1w_d_scalar_offset(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, <vscale x 2 x i64> %base, i64 %offset) {77; CHECK-LABEL: sst1w_d_scalar_offset:78; CHECK: // %bb.0:79; CHECK-NEXT: st1w { z0.d }, p0, [x0, z1.d]80; CHECK-NEXT: ret81 %data_trunc = trunc <vscale x 2 x i64> %data to <vscale x 2 x i32>82 call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i32.nxv2i64(<vscale x 2 x i32> %data_trunc,83 <vscale x 2 x i1> %pg,84 <vscale x 2 x i64> %base,85 i64 %offset)86 ret void87}88 89define void @sst1w_s_scalar_offset_float(<vscale x 4 x float> %data, <vscale x 4 x i1> %pg, <vscale x 4 x i32> %base, i64 %offset) {90; CHECK-LABEL: sst1w_s_scalar_offset_float:91; CHECK: // %bb.0:92; CHECK-NEXT: st1w { z0.s }, p0, [x0, z1.s, uxtw]93; CHECK-NEXT: ret94 call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv4f32.nxv4i32(<vscale x 4 x float> %data,95 <vscale x 4 x i1> %pg,96 <vscale x 4 x i32> %base,97 i64 %offset)98 ret void99}100 101; ST1D102define void @sst1d_d_scalar_offset(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, <vscale x 2 x i64> %base, i64 %offset) {103; CHECK-LABEL: sst1d_d_scalar_offset:104; CHECK: // %bb.0:105; CHECK-NEXT: st1d { z0.d }, p0, [x0, z1.d]106; CHECK-NEXT: ret107 call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i64.nxv2i64(<vscale x 2 x i64> %data,108 <vscale x 2 x i1> %pg,109 <vscale x 2 x i64> %base,110 i64 %offset)111 ret void112}113 114define void @sst1d_d_scalar_offset_double(<vscale x 2 x double> %data, <vscale x 2 x i1> %pg, <vscale x 2 x i64> %base, i64 %offset) {115; CHECK-LABEL: sst1d_d_scalar_offset_double:116; CHECK: // %bb.0:117; CHECK-NEXT: st1d { z0.d }, p0, [x0, z1.d]118; CHECK-NEXT: ret119 call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2f64.nxv2i64(<vscale x 2 x double> %data,120 <vscale x 2 x i1> %pg,121 <vscale x 2 x i64> %base,122 i64 %offset)123 ret void124}125 126; ST1B127declare void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv4i8.nxv4i32(<vscale x 4 x i8>, <vscale x 4 x i1>, <vscale x 4 x i32>, i64)128declare void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i8.nxv2i64(<vscale x 2 x i8>, <vscale x 2 x i1>, <vscale x 2 x i64>, i64)129 130; ST1H131declare void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i16>, <vscale x 4 x i1>, <vscale x 4 x i32>, i64)132declare void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i16>, <vscale x 2 x i1>, <vscale x 2 x i64>, i64)133 134; ST1W135declare void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv4i32.nxv4i32(<vscale x 4 x i32>, <vscale x 4 x i1>, <vscale x 4 x i32>, i64)136declare void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i32.nxv2i64(<vscale x 2 x i32>, <vscale x 2 x i1>, <vscale x 2 x i64>, i64)137 138declare void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv4f32.nxv4i32(<vscale x 4 x float>, <vscale x 4 x i1>, <vscale x 4 x i32>, i64)139 140; ST1D141declare void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i64.nxv2i64(<vscale x 2 x i64>, <vscale x 2 x i1>, <vscale x 2 x i64>, i64)142 143declare void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2f64.nxv2i64(<vscale x 2 x double>, <vscale x 2 x i1>, <vscale x 2 x i64>, i64)144