138 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, i64 %offset, <vscale x 2 x i1> %mask) {5; CHECK-LABEL: masked_gather_nxv2i8:6; CHECK: // %bb.0:7; CHECK-NEXT: ld1b { z0.d }, p0/z, [x0, z0.d]8; CHECK-NEXT: ret9 %ptrs = getelementptr i8, <vscale x 2 x ptr> %bases, i64 %offset10 %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, i64 %offset, <vscale x 2 x i1> %mask) {16; CHECK-LABEL: masked_gather_nxv2i16:17; CHECK: // %bb.0:18; CHECK-NEXT: ld1h { z0.d }, p0/z, [x0, z0.d]19; CHECK-NEXT: ret20 %byte_ptrs = getelementptr i8, <vscale x 2 x ptr> %bases, i64 %offset21 %ptrs = bitcast <vscale x 2 x ptr> %byte_ptrs to <vscale x 2 x ptr>22 %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)23 %vals.zext = zext <vscale x 2 x i16> %vals to <vscale x 2 x i64>24 ret <vscale x 2 x i64> %vals.zext25}26 27define <vscale x 2 x i64> @masked_gather_nxv2i32(<vscale x 2 x ptr> %bases, i64 %offset, <vscale x 2 x i1> %mask) {28; CHECK-LABEL: masked_gather_nxv2i32:29; CHECK: // %bb.0:30; CHECK-NEXT: ld1w { z0.d }, p0/z, [x0, z0.d]31; CHECK-NEXT: ret32 %byte_ptrs = getelementptr i8, <vscale x 2 x ptr> %bases, i64 %offset33 %ptrs = bitcast <vscale x 2 x ptr> %byte_ptrs to <vscale x 2 x ptr>34 %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)35 %vals.zext = zext <vscale x 2 x i32> %vals to <vscale x 2 x i64>36 ret <vscale x 2 x i64> %vals.zext37}38 39define <vscale x 2 x i64> @masked_gather_nxv2i64(<vscale x 2 x ptr> %bases, i64 %offset, <vscale x 2 x i1> %mask) {40; CHECK-LABEL: masked_gather_nxv2i64:41; CHECK: // %bb.0:42; CHECK-NEXT: ld1d { z0.d }, p0/z, [x0, z0.d]43; CHECK-NEXT: ret44 %byte_ptrs = getelementptr i8, <vscale x 2 x ptr> %bases, i64 %offset45 %ptrs = bitcast <vscale x 2 x ptr> %byte_ptrs to <vscale x 2 x ptr>46 %vals = 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)47 ret <vscale x 2 x i64> %vals48}49 50define <vscale x 2 x half> @masked_gather_nxv2f16(<vscale x 2 x ptr> %bases, i64 %offset, <vscale x 2 x i1> %mask) {51; CHECK-LABEL: masked_gather_nxv2f16:52; CHECK: // %bb.0:53; CHECK-NEXT: ld1h { z0.d }, p0/z, [x0, z0.d]54; CHECK-NEXT: ret55 %byte_ptrs = getelementptr i8, <vscale x 2 x ptr> %bases, i64 %offset56 %ptrs = bitcast <vscale x 2 x ptr> %byte_ptrs to <vscale x 2 x ptr>57 %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)58 ret <vscale x 2 x half> %vals59}60 61define <vscale x 2 x bfloat> @masked_gather_nxv2bf16(<vscale x 2 x ptr> %bases, i64 %offset, <vscale x 2 x i1> %mask) #0 {62; CHECK-LABEL: masked_gather_nxv2bf16:63; CHECK: // %bb.0:64; CHECK-NEXT: ld1h { z0.d }, p0/z, [x0, z0.d]65; CHECK-NEXT: ret66 %byte_ptrs = getelementptr i8, <vscale x 2 x ptr> %bases, i64 %offset67 %ptrs = bitcast <vscale x 2 x ptr> %byte_ptrs to <vscale x 2 x ptr>68 %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)69 ret <vscale x 2 x bfloat> %vals70}71 72define <vscale x 2 x float> @masked_gather_nxv2f32(<vscale x 2 x ptr> %bases, i64 %offset, <vscale x 2 x i1> %mask) {73; CHECK-LABEL: masked_gather_nxv2f32:74; CHECK: // %bb.0:75; CHECK-NEXT: ld1w { z0.d }, p0/z, [x0, z0.d]76; CHECK-NEXT: ret77 %byte_ptrs = getelementptr i8, <vscale x 2 x ptr> %bases, i64 %offset78 %ptrs = bitcast <vscale x 2 x ptr> %byte_ptrs to <vscale x 2 x ptr>79 %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)80 ret <vscale x 2 x float> %vals81}82 83define <vscale x 2 x double> @masked_gather_nxv2f64(<vscale x 2 x ptr> %bases, i64 %offset, <vscale x 2 x i1> %mask) {84; CHECK-LABEL: masked_gather_nxv2f64:85; CHECK: // %bb.0:86; CHECK-NEXT: ld1d { z0.d }, p0/z, [x0, z0.d]87; CHECK-NEXT: ret88 %byte_ptrs = getelementptr i8, <vscale x 2 x ptr> %bases, i64 %offset89 %ptrs = bitcast <vscale x 2 x ptr> %byte_ptrs to <vscale x 2 x ptr>90 %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)91 ret <vscale x 2 x double> %vals92}93 94define <vscale x 2 x i64> @masked_sgather_nxv2i8(<vscale x 2 x ptr> %bases, i64 %offset, <vscale x 2 x i1> %mask) {95; CHECK-LABEL: masked_sgather_nxv2i8:96; CHECK: // %bb.0:97; CHECK-NEXT: ld1sb { z0.d }, p0/z, [x0, z0.d]98; CHECK-NEXT: ret99 %ptrs = getelementptr i8, <vscale x 2 x ptr> %bases, i64 %offset100 %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)101 %vals.sext = sext <vscale x 2 x i8> %vals to <vscale x 2 x i64>102 ret <vscale x 2 x i64> %vals.sext103}104 105define <vscale x 2 x i64> @masked_sgather_nxv2i16(<vscale x 2 x ptr> %bases, i64 %offset, <vscale x 2 x i1> %mask) {106; CHECK-LABEL: masked_sgather_nxv2i16:107; CHECK: // %bb.0:108; CHECK-NEXT: ld1sh { z0.d }, p0/z, [x0, z0.d]109; CHECK-NEXT: ret110 %byte_ptrs = getelementptr i8, <vscale x 2 x ptr> %bases, i64 %offset111 %ptrs = bitcast <vscale x 2 x ptr> %byte_ptrs to <vscale x 2 x ptr>112 %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)113 %vals.sext = sext <vscale x 2 x i16> %vals to <vscale x 2 x i64>114 ret <vscale x 2 x i64> %vals.sext115}116 117define <vscale x 2 x i64> @masked_sgather_nxv2i32(<vscale x 2 x ptr> %bases, i64 %offset, <vscale x 2 x i1> %mask) {118; CHECK-LABEL: masked_sgather_nxv2i32:119; CHECK: // %bb.0:120; CHECK-NEXT: ld1sw { z0.d }, p0/z, [x0, z0.d]121; CHECK-NEXT: ret122 %byte_ptrs = getelementptr i8, <vscale x 2 x ptr> %bases, i64 %offset123 %ptrs = bitcast <vscale x 2 x ptr> %byte_ptrs to <vscale x 2 x ptr>124 %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)125 %vals.sext = sext <vscale x 2 x i32> %vals to <vscale x 2 x i64>126 ret <vscale x 2 x i64> %vals.sext127}128 129declare <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>)130declare <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>)131declare <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>)132declare <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>)133declare <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>)134declare <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>)135declare <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>)136declare <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>)137attributes #0 = { "target-features"="+sve,+bf16" }138