86 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc -mattr=+sve < %s | FileCheck %s3 4 5target triple = "aarch64-unknown-linux-gnu"6 7;8; Masked Load9;10 11define <16 x i8> @masked_load_v16i8(ptr %src, <16 x i1> %mask) {12; CHECK-LABEL: masked_load_v16i8:13; CHECK: // %bb.0:14; CHECK-NEXT: shl v0.16b, v0.16b, #715; CHECK-NEXT: ptrue p0.b, vl1616; CHECK-NEXT: cmpne p0.b, p0/z, z0.b, #017; CHECK-NEXT: ld1b { z0.b }, p0/z, [x0]18; CHECK-NEXT: // kill: def $q0 killed $q0 killed $z019; CHECK-NEXT: ret20 %load = call <16 x i8> @llvm.masked.load.v16i8(ptr %src, i32 8, <16 x i1> %mask, <16 x i8> zeroinitializer)21 ret <16 x i8> %load22}23 24define <8 x half> @masked_load_v8f16(ptr %src, <8 x i1> %mask) {25; CHECK-LABEL: masked_load_v8f16:26; CHECK: // %bb.0:27; CHECK-NEXT: ushll v0.8h, v0.8b, #028; CHECK-NEXT: ptrue p0.h, vl829; CHECK-NEXT: shl v0.8h, v0.8h, #1530; CHECK-NEXT: cmpne p0.h, p0/z, z0.h, #031; CHECK-NEXT: ld1h { z0.h }, p0/z, [x0]32; CHECK-NEXT: // kill: def $q0 killed $q0 killed $z033; CHECK-NEXT: ret34 %load = call <8 x half> @llvm.masked.load.v8f16(ptr %src, i32 8, <8 x i1> %mask, <8 x half> zeroinitializer)35 ret <8 x half> %load36}37 38define <4 x float> @masked_load_v4f32(ptr %src, <4 x i1> %mask) {39; CHECK-LABEL: masked_load_v4f32:40; CHECK: // %bb.0:41; CHECK-NEXT: ushll v0.4s, v0.4h, #042; CHECK-NEXT: ptrue p0.s, vl443; CHECK-NEXT: shl v0.4s, v0.4s, #3144; CHECK-NEXT: cmpne p0.s, p0/z, z0.s, #045; CHECK-NEXT: ld1w { z0.s }, p0/z, [x0]46; CHECK-NEXT: // kill: def $q0 killed $q0 killed $z047; CHECK-NEXT: ret48 %load = call <4 x float> @llvm.masked.load.v4f32(ptr %src, i32 8, <4 x i1> %mask, <4 x float> zeroinitializer)49 ret <4 x float> %load50}51 52define <2 x double> @masked_load_v2f64(ptr %src, <2 x i1> %mask) {53; CHECK-LABEL: masked_load_v2f64:54; CHECK: // %bb.0:55; CHECK-NEXT: ushll v0.2d, v0.2s, #056; CHECK-NEXT: ptrue p0.d, vl257; CHECK-NEXT: shl v0.2d, v0.2d, #6358; CHECK-NEXT: cmpne p0.d, p0/z, z0.d, #059; CHECK-NEXT: ld1d { z0.d }, p0/z, [x0]60; CHECK-NEXT: // kill: def $q0 killed $q0 killed $z061; CHECK-NEXT: ret62 %load = call <2 x double> @llvm.masked.load.v2f64(ptr %src, i32 8, <2 x i1> %mask, <2 x double> zeroinitializer)63 ret <2 x double> %load64}65 66define <2 x double> @masked_load_passthru_v2f64(ptr %src, <2 x i1> %mask, <2 x double> %passthru) {67; CHECK-LABEL: masked_load_passthru_v2f64:68; CHECK: // %bb.0:69; CHECK-NEXT: ushll v0.2d, v0.2s, #070; CHECK-NEXT: ptrue p0.d, vl271; CHECK-NEXT: // kill: def $q1 killed $q1 def $z172; CHECK-NEXT: shl v0.2d, v0.2d, #6373; CHECK-NEXT: cmpne p0.d, p0/z, z0.d, #074; CHECK-NEXT: ld1d { z0.d }, p0/z, [x0]75; CHECK-NEXT: sel z0.d, p0, z0.d, z1.d76; CHECK-NEXT: // kill: def $q0 killed $q0 killed $z077; CHECK-NEXT: ret78 %load = call <2 x double> @llvm.masked.load.v2f64(ptr %src, i32 8, <2 x i1> %mask, <2 x double> %passthru)79 ret <2 x double> %load80}81 82declare <16 x i8> @llvm.masked.load.v16i8(ptr, i32, <16 x i1>, <16 x i8>)83declare <8 x half> @llvm.masked.load.v8f16(ptr, i32, <8 x i1>, <8 x half>)84declare <4 x float> @llvm.masked.load.v4f32(ptr, i32, <4 x i1>, <4 x float>)85declare <2 x double> @llvm.masked.load.v2f64(ptr, i32, <2 x i1>, <2 x double>)86