181 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -S -passes='require<profile-summary>,function(codegenprepare)' < %s | FileCheck %s3; RUN: opt -S -passes='require<profile-summary>,function(codegenprepare)' -cgpp-huge-func=0 < %s | FileCheck %s4 5target datalayout =6"e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128"7target triple = "x86_64-unknown-linux-gnu"8 9%struct.a = type { i32, i32 }10@c = external dso_local global %struct.a, align 411@glob_array = internal unnamed_addr constant [16 x i32] [i32 1, i32 1, i32 2, i32 3, i32 5, i32 8, i32 13, i32 21, i32 34, i32 55, i32 89, i32 144, i32 233, i32 377, i32 610, i32 987], align 1612 13define <4 x i32> @splat_base(ptr %base, <4 x i64> %index) {14; CHECK-LABEL: @splat_base(15; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i32, ptr [[BASE:%.*]], <4 x i64> [[INDEX:%.*]]16; CHECK-NEXT: [[RES:%.*]] = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> align 4 [[TMP1]], <4 x i1> splat (i1 true), <4 x i32> undef)17; CHECK-NEXT: ret <4 x i32> [[RES]]18;19 %broadcast.splatinsert = insertelement <4 x ptr> poison, ptr %base, i32 020 %broadcast.splat = shufflevector <4 x ptr> %broadcast.splatinsert, <4 x ptr> poison, <4 x i32> zeroinitializer21 %gep = getelementptr i32, <4 x ptr> %broadcast.splat, <4 x i64> %index22 %res = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> %gep, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>, <4 x i32> undef)23 ret <4 x i32> %res24}25 26define <4 x i32> @splat_struct(ptr %base) {27; CHECK-LABEL: @splat_struct(28; CHECK-NEXT: [[TMP1:%.*]] = getelementptr [[STRUCT_A:%.*]], ptr [[BASE:%.*]], i64 0, i32 129; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i32, ptr [[TMP1]], <4 x i64> zeroinitializer30; CHECK-NEXT: [[RES:%.*]] = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> align 4 [[TMP2]], <4 x i1> splat (i1 true), <4 x i32> undef)31; CHECK-NEXT: ret <4 x i32> [[RES]]32;33 %gep = getelementptr %struct.a, ptr %base, <4 x i64> zeroinitializer, i32 134 %res = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> %gep, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>, <4 x i32> undef)35 ret <4 x i32> %res36}37 38define <4 x i32> @scalar_index(ptr %base, i64 %index) {39; CHECK-LABEL: @scalar_index(40; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i32, ptr [[BASE:%.*]], i64 [[INDEX:%.*]]41; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i32, ptr [[TMP1]], <4 x i64> zeroinitializer42; CHECK-NEXT: [[RES:%.*]] = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> align 4 [[TMP2]], <4 x i1> splat (i1 true), <4 x i32> undef)43; CHECK-NEXT: ret <4 x i32> [[RES]]44;45 %broadcast.splatinsert = insertelement <4 x ptr> poison, ptr %base, i32 046 %broadcast.splat = shufflevector <4 x ptr> %broadcast.splatinsert, <4 x ptr> poison, <4 x i32> zeroinitializer47 %gep = getelementptr i32, <4 x ptr> %broadcast.splat, i64 %index48 %res = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> %gep, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>, <4 x i32> undef)49 ret <4 x i32> %res50}51 52define <4 x i32> @splat_index(ptr %base, i64 %index) {53; CHECK-LABEL: @splat_index(54; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i32, ptr [[BASE:%.*]], i64 [[INDEX:%.*]]55; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i32, ptr [[TMP1]], <4 x i64> zeroinitializer56; CHECK-NEXT: [[RES:%.*]] = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> align 4 [[TMP2]], <4 x i1> splat (i1 true), <4 x i32> undef)57; CHECK-NEXT: ret <4 x i32> [[RES]]58;59 %broadcast.splatinsert = insertelement <4 x i64> poison, i64 %index, i32 060 %broadcast.splat = shufflevector <4 x i64> %broadcast.splatinsert, <4 x i64> poison, <4 x i32> zeroinitializer61 %gep = getelementptr i32, ptr %base, <4 x i64> %broadcast.splat62 %res = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> %gep, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>, <4 x i32> undef)63 ret <4 x i32> %res64}65 66define <4 x i32> @test_global_array(<4 x i64> %indxs) {67; CHECK-LABEL: @test_global_array(68; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i32, ptr @glob_array, <4 x i64> [[INDXS:%.*]]69; CHECK-NEXT: [[G:%.*]] = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> align 4 [[TMP1]], <4 x i1> splat (i1 true), <4 x i32> undef)70; CHECK-NEXT: ret <4 x i32> [[G]]71;72 %p = getelementptr inbounds [16 x i32], ptr @glob_array, i64 0, <4 x i64> %indxs73 %g = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> %p, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>, <4 x i32> undef)74 ret <4 x i32> %g75}76 77define <4 x i32> @global_struct_splat() {78; CHECK-LABEL: @global_struct_splat(79; CHECK-NEXT: [[TMP1:%.*]] = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> align 4 <ptr getelementptr ([[STRUCT_A:%.*]], ptr @c, i64 0, i32 1), ptr getelementptr ([[STRUCT_A]], ptr @c, i64 0, i32 1), ptr getelementptr ([[STRUCT_A]], ptr @c, i64 0, i32 1), ptr getelementptr ([[STRUCT_A]], ptr @c, i64 0, i32 1)>, <4 x i1> splat (i1 true), <4 x i32> undef)80; CHECK-NEXT: ret <4 x i32> [[TMP1]]81;82 %1 = insertelement <4 x ptr> poison, ptr @c, i32 083 %2 = shufflevector <4 x ptr> %1, <4 x ptr> poison, <4 x i32> zeroinitializer84 %3 = getelementptr %struct.a, <4 x ptr> %2, <4 x i64> zeroinitializer, i32 185 %4 = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> %3, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>, <4 x i32> undef)86 ret <4 x i32> %487}88 89define <4 x i32> @splat_ptr_gather(ptr %ptr, <4 x i1> %mask, <4 x i32> %passthru) {90; CHECK-LABEL: @splat_ptr_gather(91; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i32, ptr [[PTR:%.*]], <4 x i64> zeroinitializer92; CHECK-NEXT: [[TMP2:%.*]] = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> align 4 [[TMP1]], <4 x i1> [[MASK:%.*]], <4 x i32> [[PASSTHRU:%.*]])93; CHECK-NEXT: ret <4 x i32> [[TMP2]]94;95 %1 = insertelement <4 x ptr> poison, ptr %ptr, i32 096 %2 = shufflevector <4 x ptr> %1, <4 x ptr> poison, <4 x i32> zeroinitializer97 %3 = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> %2, i32 4, <4 x i1> %mask, <4 x i32> %passthru)98 ret <4 x i32> %399}100 101define void @splat_ptr_scatter(ptr %ptr, <4 x i1> %mask, <4 x i32> %val) {102; CHECK-LABEL: @splat_ptr_scatter(103; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i32, ptr [[PTR:%.*]], <4 x i64> zeroinitializer104; CHECK-NEXT: call void @llvm.masked.scatter.v4i32.v4p0(<4 x i32> [[VAL:%.*]], <4 x ptr> align 4 [[TMP1]], <4 x i1> [[MASK:%.*]])105; CHECK-NEXT: ret void106;107 %1 = insertelement <4 x ptr> poison, ptr %ptr, i32 0108 %2 = shufflevector <4 x ptr> %1, <4 x ptr> poison, <4 x i32> zeroinitializer109 call void @llvm.masked.scatter.v4i32.v4p0(<4 x i32> %val, <4 x ptr> %2, i32 4, <4 x i1> %mask)110 ret void111}112 113define <4 x i32> @scalar_prefix(ptr %base, i64 %index, <4 x i64> %vecidx) {114; CHECK-LABEL: @scalar_prefix(115; CHECK-NEXT: [[TMP2:%.*]] = getelementptr [256 x i32], ptr [[BASE:%.*]], i64 [[INDEX:%.*]], <4 x i64> [[VECIDX:%.*]]116; CHECK-NEXT: [[RES:%.*]] = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> align 4 [[TMP2]], <4 x i1> splat (i1 true), <4 x i32> undef)117; CHECK-NEXT: ret <4 x i32> [[RES]]118;119 %gep = getelementptr [256 x i32], ptr %base, i64 %index, <4 x i64> %vecidx120 %res = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> %gep, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>, <4 x i32> undef)121 ret <4 x i32> %res122}123 124define <4 x i32> @scalar_prefix_with_splat(ptr %base, i64 %index, <4 x i64> %vecidx) {125; CHECK-LABEL: @scalar_prefix_with_splat(126; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[INDEX:%.*]], i32 0127; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer128; CHECK-NEXT: [[TMP2:%.*]] = getelementptr [256 x i32], ptr [[BASE:%.*]], <4 x i64> [[BROADCAST_SPLAT]], <4 x i64> [[VECIDX:%.*]]129; CHECK-NEXT: [[RES:%.*]] = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> align 4 [[TMP2]], <4 x i1> splat (i1 true), <4 x i32> undef)130; CHECK-NEXT: ret <4 x i32> [[RES]]131;132 %broadcast.splatinsert = insertelement <4 x i64> poison, i64 %index, i32 0133 %broadcast.splat = shufflevector <4 x i64> %broadcast.splatinsert, <4 x i64> poison, <4 x i32> zeroinitializer134 135 %gep = getelementptr [256 x i32], ptr %base, <4 x i64> %broadcast.splat, <4 x i64> %vecidx136 %res = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> %gep, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>, <4 x i32> undef)137 ret <4 x i32> %res138}139 140define <4 x i32> @scalar_prefix_with_constant_splat(ptr %base, <4 x i64> %vecidx) {141; CHECK-LABEL: @scalar_prefix_with_constant_splat(142; CHECK-NEXT: [[TMP2:%.*]] = getelementptr [256 x i32], ptr [[BASE:%.*]], <4 x i64> splat (i64 20), <4 x i64> [[VECIDX:%.*]]143; CHECK-NEXT: [[RES:%.*]] = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> align 4 [[TMP2]], <4 x i1> splat (i1 true), <4 x i32> undef)144; CHECK-NEXT: ret <4 x i32> [[RES]]145;146 %gep = getelementptr [256 x i32], ptr %base, <4 x i64> splat (i64 20), <4 x i64> %vecidx147 %res = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> %gep, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>, <4 x i32> undef)148 ret <4 x i32> %res149}150 151define <4 x i32> @reassociate(ptr %base, i64 %index, <4 x i64> %vecidx) {152; CHECK-LABEL: @reassociate(153; CHECK-NEXT: [[GEP:%.*]] = getelementptr [256 x i32], ptr [[BASE:%.*]], <4 x i64> [[VECIDX:%.*]], i64 [[INDEX:%.*]]154; CHECK-NEXT: [[RES:%.*]] = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> align 4 [[GEP]], <4 x i1> splat (i1 true), <4 x i32> undef)155; CHECK-NEXT: ret <4 x i32> [[RES]]156;157 %gep = getelementptr [256 x i32], ptr %base, <4 x i64> %vecidx, i64 %index158 %res = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> %gep, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>, <4 x i32> undef)159 ret <4 x i32> %res160}161 162define <4 x i32> @reassociate_with_splat(ptr %base, i64 %index, <4 x i64> %vecidx) {163; CHECK-LABEL: @reassociate_with_splat(164; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[INDEX:%.*]], i32 0165; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer166; CHECK-NEXT: [[GEP:%.*]] = getelementptr [256 x i32], ptr [[BASE:%.*]], <4 x i64> [[VECIDX:%.*]], <4 x i64> [[BROADCAST_SPLAT]]167; CHECK-NEXT: [[RES:%.*]] = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> align 4 [[GEP]], <4 x i1> splat (i1 true), <4 x i32> undef)168; CHECK-NEXT: ret <4 x i32> [[RES]]169;170 %broadcast.splatinsert = insertelement <4 x i64> poison, i64 %index, i32 0171 %broadcast.splat = shufflevector <4 x i64> %broadcast.splatinsert, <4 x i64> poison, <4 x i32> zeroinitializer172 173 %gep = getelementptr [256 x i32], ptr %base, <4 x i64> %vecidx, <4 x i64> %broadcast.splat174 %res = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> %gep, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>, <4 x i32> undef)175 ret <4 x i32> %res176}177 178 179declare <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr>, i32, <4 x i1>, <4 x i32>)180declare void @llvm.masked.scatter.v4i32.v4p0(<4 x i32>, <4 x ptr>, i32, <4 x i1>)181