187 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 4define <vscale x 2 x i64> @masked_gather_nxv2i8(<vscale x 2 x ptr> %bases, <vscale x 2 x i1> %mask) {5; CHECK-LABEL: masked_gather_nxv2i8:6; CHECK: // %bb.0:7; CHECK-NEXT: ld1b { z0.d }, p0/z, [z0.d, #1]8; CHECK-NEXT: ret9 %ptrs = getelementptr i8, <vscale x 2 x ptr> %bases, i32 110 %vals = call <vscale x 2 x i8> @llvm.masked.gather.nxv2i8(<vscale x 2 x ptr> %ptrs, i32 1, <vscale x 2 x i1> %mask, <vscale x 2 x i8> poison)11 %vals.zext = zext <vscale x 2 x i8> %vals to <vscale x 2 x i64>12 ret <vscale x 2 x i64> %vals.zext13}14 15define <vscale x 2 x i64> @masked_gather_nxv2i16(<vscale x 2 x ptr> %bases, <vscale x 2 x i1> %mask) {16; CHECK-LABEL: masked_gather_nxv2i16:17; CHECK: // %bb.0:18; CHECK-NEXT: ld1h { z0.d }, p0/z, [z0.d, #2]19; CHECK-NEXT: ret20 %ptrs = getelementptr i16, <vscale x 2 x ptr> %bases, i32 121 %vals = call <vscale x 2 x i16> @llvm.masked.gather.nxv2i16(<vscale x 2 x ptr> %ptrs, i32 2, <vscale x 2 x i1> %mask, <vscale x 2 x i16> poison)22 %vals.zext = zext <vscale x 2 x i16> %vals to <vscale x 2 x i64>23 ret <vscale x 2 x i64> %vals.zext24}25 26define <vscale x 2 x i64> @masked_gather_nxv2i32(<vscale x 2 x ptr> %bases, <vscale x 2 x i1> %mask) {27; CHECK-LABEL: masked_gather_nxv2i32:28; CHECK: // %bb.0:29; CHECK-NEXT: ld1w { z0.d }, p0/z, [z0.d, #4]30; CHECK-NEXT: ret31 %ptrs = getelementptr i32, <vscale x 2 x ptr> %bases, i32 132 %vals = call <vscale x 2 x i32> @llvm.masked.gather.nxv2i32(<vscale x 2 x ptr> %ptrs, i32 4, <vscale x 2 x i1> %mask, <vscale x 2 x i32> poison)33 %vals.zext = zext <vscale x 2 x i32> %vals to <vscale x 2 x i64>34 ret <vscale x 2 x i64> %vals.zext35}36 37define <vscale x 2 x i64> @masked_gather_nxv2i64(<vscale x 2 x ptr> %bases, <vscale x 2 x i1> %mask) {38; CHECK-LABEL: masked_gather_nxv2i64:39; CHECK: // %bb.0:40; CHECK-NEXT: ld1d { z0.d }, p0/z, [z0.d, #8]41; CHECK-NEXT: ret42 %ptrs = getelementptr i64, <vscale x 2 x ptr> %bases, i32 143 %vals.zext = call <vscale x 2 x i64> @llvm.masked.gather.nxv2i64(<vscale x 2 x ptr> %ptrs, i32 8, <vscale x 2 x i1> %mask, <vscale x 2 x i64> poison)44 ret <vscale x 2 x i64> %vals.zext45}46 47define <vscale x 2 x half> @masked_gather_nxv2f16(<vscale x 2 x ptr> %bases, <vscale x 2 x i1> %mask) {48; CHECK-LABEL: masked_gather_nxv2f16:49; CHECK: // %bb.0:50; CHECK-NEXT: ld1h { z0.d }, p0/z, [z0.d, #4]51; CHECK-NEXT: ret52 %ptrs = getelementptr half, <vscale x 2 x ptr> %bases, i32 253 %vals = call <vscale x 2 x half> @llvm.masked.gather.nxv2f16(<vscale x 2 x ptr> %ptrs, i32 2, <vscale x 2 x i1> %mask, <vscale x 2 x half> poison)54 ret <vscale x 2 x half> %vals55}56 57define <vscale x 2 x bfloat> @masked_gather_nxv2bf16(<vscale x 2 x ptr> %bases, <vscale x 2 x i1> %mask) #0 {58; CHECK-LABEL: masked_gather_nxv2bf16:59; CHECK: // %bb.0:60; CHECK-NEXT: ld1h { z0.d }, p0/z, [z0.d, #4]61; CHECK-NEXT: ret62 %ptrs = getelementptr bfloat, <vscale x 2 x ptr> %bases, i32 263 %vals = call <vscale x 2 x bfloat> @llvm.masked.gather.nxv2bf16(<vscale x 2 x ptr> %ptrs, i32 2, <vscale x 2 x i1> %mask, <vscale x 2 x bfloat> poison)64 ret <vscale x 2 x bfloat> %vals65}66 67define <vscale x 2 x float> @masked_gather_nxv2f32(<vscale x 2 x ptr> %bases, <vscale x 2 x i1> %mask) {68; CHECK-LABEL: masked_gather_nxv2f32:69; CHECK: // %bb.0:70; CHECK-NEXT: ld1w { z0.d }, p0/z, [z0.d, #12]71; CHECK-NEXT: ret72 %ptrs = getelementptr float, <vscale x 2 x ptr> %bases, i32 373 %vals = call <vscale x 2 x float> @llvm.masked.gather.nxv2f32(<vscale x 2 x ptr> %ptrs, i32 4, <vscale x 2 x i1> %mask, <vscale x 2 x float> poison)74 ret <vscale x 2 x float> %vals75}76 77define <vscale x 2 x double> @masked_gather_nxv2f64(<vscale x 2 x ptr> %bases, <vscale x 2 x i1> %mask) {78; CHECK-LABEL: masked_gather_nxv2f64:79; CHECK: // %bb.0:80; CHECK-NEXT: ld1d { z0.d }, p0/z, [z0.d, #32]81; CHECK-NEXT: ret82 %ptrs = getelementptr double, <vscale x 2 x ptr> %bases, i32 483 %vals = call <vscale x 2 x double> @llvm.masked.gather.nxv2f64(<vscale x 2 x ptr> %ptrs, i32 8, <vscale x 2 x i1> %mask, <vscale x 2 x double> poison)84 ret <vscale x 2 x double> %vals85}86 87define <vscale x 2 x i64> @masked_sgather_nxv2i8(<vscale x 2 x ptr> %bases, <vscale x 2 x i1> %mask) {88; CHECK-LABEL: masked_sgather_nxv2i8:89; CHECK: // %bb.0:90; CHECK-NEXT: ld1sb { z0.d }, p0/z, [z0.d, #5]91; CHECK-NEXT: ret92 %ptrs = getelementptr i8, <vscale x 2 x ptr> %bases, i32 593 %vals = call <vscale x 2 x i8> @llvm.masked.gather.nxv2i8(<vscale x 2 x ptr> %ptrs, i32 1, <vscale x 2 x i1> %mask, <vscale x 2 x i8> poison)94 %vals.sext = sext <vscale x 2 x i8> %vals to <vscale x 2 x i64>95 ret <vscale x 2 x i64> %vals.sext96}97 98define <vscale x 2 x i64> @masked_sgather_nxv2i16(<vscale x 2 x ptr> %bases, <vscale x 2 x i1> %mask) {99; CHECK-LABEL: masked_sgather_nxv2i16:100; CHECK: // %bb.0:101; CHECK-NEXT: ld1sh { z0.d }, p0/z, [z0.d, #12]102; CHECK-NEXT: ret103 %ptrs = getelementptr i16, <vscale x 2 x ptr> %bases, i32 6104 %vals = call <vscale x 2 x i16> @llvm.masked.gather.nxv2i16(<vscale x 2 x ptr> %ptrs, i32 2, <vscale x 2 x i1> %mask, <vscale x 2 x i16> poison)105 %vals.sext = sext <vscale x 2 x i16> %vals to <vscale x 2 x i64>106 ret <vscale x 2 x i64> %vals.sext107}108 109define <vscale x 2 x i64> @masked_sgather_nxv2i32(<vscale x 2 x ptr> %bases, <vscale x 2 x i1> %mask) {110; CHECK-LABEL: masked_sgather_nxv2i32:111; CHECK: // %bb.0:112; CHECK-NEXT: ld1sw { z0.d }, p0/z, [z0.d, #28]113; CHECK-NEXT: ret114 %ptrs = getelementptr i32, <vscale x 2 x ptr> %bases, i32 7115 %vals = call <vscale x 2 x i32> @llvm.masked.gather.nxv2i32(<vscale x 2 x ptr> %ptrs, i32 4, <vscale x 2 x i1> %mask, <vscale x 2 x i32> poison)116 %vals.sext = sext <vscale x 2 x i32> %vals to <vscale x 2 x i64>117 ret <vscale x 2 x i64> %vals.sext118}119 120; Tests where the immediate is out of range121 122define <vscale x 2 x i64> @masked_gather_nxv2i8_range(<vscale x 2 x ptr> %bases, <vscale x 2 x i1> %mask) {123; CHECK-LABEL: masked_gather_nxv2i8_range:124; CHECK: // %bb.0:125; CHECK-NEXT: mov w8, #32126; CHECK-NEXT: ld1b { z0.d }, p0/z, [x8, z0.d]127; CHECK-NEXT: ret128 %ptrs = getelementptr i8, <vscale x 2 x ptr> %bases, i32 32129 %vals = call <vscale x 2 x i8> @llvm.masked.gather.nxv2i8(<vscale x 2 x ptr> %ptrs, i32 1, <vscale x 2 x i1> %mask, <vscale x 2 x i8> poison)130 %vals.zext = zext <vscale x 2 x i8> %vals to <vscale x 2 x i64>131 ret <vscale x 2 x i64> %vals.zext132}133 134define <vscale x 2 x half> @masked_gather_nxv2f16_range(<vscale x 2 x ptr> %bases, <vscale x 2 x i1> %mask) {135; CHECK-LABEL: masked_gather_nxv2f16_range:136; CHECK: // %bb.0:137; CHECK-NEXT: mov w8, #64138; CHECK-NEXT: ld1h { z0.d }, p0/z, [x8, z0.d]139; CHECK-NEXT: ret140 %ptrs = getelementptr half, <vscale x 2 x ptr> %bases, i32 32141 %vals = call <vscale x 2 x half> @llvm.masked.gather.nxv2f16(<vscale x 2 x ptr> %ptrs, i32 2, <vscale x 2 x i1> %mask, <vscale x 2 x half> poison)142 ret <vscale x 2 x half> %vals143}144 145define <vscale x 2 x bfloat> @masked_gather_nxv2bf16_range(<vscale x 2 x ptr> %bases, <vscale x 2 x i1> %mask) #0 {146; CHECK-LABEL: masked_gather_nxv2bf16_range:147; CHECK: // %bb.0:148; CHECK-NEXT: mov w8, #64149; CHECK-NEXT: ld1h { z0.d }, p0/z, [x8, z0.d]150; CHECK-NEXT: ret151 %ptrs = getelementptr bfloat, <vscale x 2 x ptr> %bases, i32 32152 %vals = call <vscale x 2 x bfloat> @llvm.masked.gather.nxv2bf16(<vscale x 2 x ptr> %ptrs, i32 2, <vscale x 2 x i1> %mask, <vscale x 2 x bfloat> poison)153 ret <vscale x 2 x bfloat> %vals154}155 156define <vscale x 2 x float> @masked_gather_nxv2f32_range(<vscale x 2 x ptr> %bases, <vscale x 2 x i1> %mask) {157; CHECK-LABEL: masked_gather_nxv2f32_range:158; CHECK: // %bb.0:159; CHECK-NEXT: mov w8, #128160; CHECK-NEXT: ld1w { z0.d }, p0/z, [x8, z0.d]161; CHECK-NEXT: ret162 %ptrs = getelementptr float, <vscale x 2 x ptr> %bases, i32 32163 %vals = call <vscale x 2 x float> @llvm.masked.gather.nxv2f32(<vscale x 2 x ptr> %ptrs, i32 4, <vscale x 2 x i1> %mask, <vscale x 2 x float> poison)164 ret <vscale x 2 x float> %vals165}166 167define <vscale x 2 x double> @masked_gather_nxv2f64_range(<vscale x 2 x ptr> %bases, <vscale x 2 x i1> %mask) {168; CHECK-LABEL: masked_gather_nxv2f64_range:169; CHECK: // %bb.0:170; CHECK-NEXT: mov w8, #256171; CHECK-NEXT: ld1d { z0.d }, p0/z, [x8, z0.d]172; CHECK-NEXT: ret173 %ptrs = getelementptr double, <vscale x 2 x ptr> %bases, i32 32174 %vals = call <vscale x 2 x double> @llvm.masked.gather.nxv2f64(<vscale x 2 x ptr> %ptrs, i32 8, <vscale x 2 x i1> %mask, <vscale x 2 x double> poison)175 ret <vscale x 2 x double> %vals176}177 178declare <vscale x 2 x i8> @llvm.masked.gather.nxv2i8(<vscale x 2 x ptr>, i32, <vscale x 2 x i1>, <vscale x 2 x i8>)179declare <vscale x 2 x i16> @llvm.masked.gather.nxv2i16(<vscale x 2 x ptr>, i32, <vscale x 2 x i1>, <vscale x 2 x i16>)180declare <vscale x 2 x i32> @llvm.masked.gather.nxv2i32(<vscale x 2 x ptr>, i32, <vscale x 2 x i1>, <vscale x 2 x i32>)181declare <vscale x 2 x i64> @llvm.masked.gather.nxv2i64(<vscale x 2 x ptr>, i32, <vscale x 2 x i1>, <vscale x 2 x i64>)182declare <vscale x 2 x half> @llvm.masked.gather.nxv2f16(<vscale x 2 x ptr>, i32, <vscale x 2 x i1>, <vscale x 2 x half>)183declare <vscale x 2 x bfloat> @llvm.masked.gather.nxv2bf16(<vscale x 2 x ptr>, i32, <vscale x 2 x i1>, <vscale x 2 x bfloat>)184declare <vscale x 2 x float> @llvm.masked.gather.nxv2f32(<vscale x 2 x ptr>, i32, <vscale x 2 x i1>, <vscale x 2 x float>)185declare <vscale x 2 x double> @llvm.masked.gather.nxv2f64(<vscale x 2 x ptr>, i32, <vscale x 2 x i1>, <vscale x 2 x double>)186attributes #0 = { "target-features"="+sve,+bf16" }187