114 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 4target triple = "aarch64-unknown-linux-gnu"5 6%struct.a = type { i32, i32 }7@c = external dso_local global %struct.a, align 48@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 169 10define <vscale x 4 x i32> @splat_base(ptr %base, <vscale x 4 x i64> %index, <vscale x 4 x i1> %mask) #0 {11; CHECK-LABEL: @splat_base(12; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i32, ptr [[BASE:%.*]], <vscale x 4 x i64> [[INDEX:%.*]]13; CHECK-NEXT: [[RES:%.*]] = call <vscale x 4 x i32> @llvm.masked.gather.nxv4i32.nxv4p0(<vscale x 4 x ptr> align 4 [[TMP1]], <vscale x 4 x i1> [[MASK:%.*]], <vscale x 4 x i32> undef)14; CHECK-NEXT: ret <vscale x 4 x i32> [[RES]]15;16 %broadcast.splatinsert = insertelement <vscale x 4 x ptr> undef, ptr %base, i32 017 %broadcast.splat = shufflevector <vscale x 4 x ptr> %broadcast.splatinsert, <vscale x 4 x ptr> undef, <vscale x 4 x i32> zeroinitializer18 %gep = getelementptr i32, <vscale x 4 x ptr> %broadcast.splat, <vscale x 4 x i64> %index19 %res = call <vscale x 4 x i32> @llvm.masked.gather.nxv4i32.nxv4p0(<vscale x 4 x ptr> %gep, i32 4, <vscale x 4 x i1> %mask, <vscale x 4 x i32> undef)20 ret <vscale x 4 x i32> %res21}22 23define <vscale x 4 x i32> @splat_struct(ptr %base, <vscale x 4 x i1> %mask) #0 {24; CHECK-LABEL: @splat_struct(25; CHECK-NEXT: [[TMP1:%.*]] = getelementptr [[STRUCT_A:%.*]], ptr [[BASE:%.*]], i64 0, i32 126; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i32, ptr [[TMP1]], <vscale x 4 x i64> zeroinitializer27; CHECK-NEXT: [[RES:%.*]] = call <vscale x 4 x i32> @llvm.masked.gather.nxv4i32.nxv4p0(<vscale x 4 x ptr> align 4 [[TMP2]], <vscale x 4 x i1> [[MASK:%.*]], <vscale x 4 x i32> undef)28; CHECK-NEXT: ret <vscale x 4 x i32> [[RES]]29;30 %gep = getelementptr %struct.a, ptr %base, <vscale x 4 x i64> zeroinitializer, i32 131 %res = call <vscale x 4 x i32> @llvm.masked.gather.nxv4i32.nxv4p0(<vscale x 4 x ptr> %gep, i32 4, <vscale x 4 x i1> %mask, <vscale x 4 x i32> undef)32 ret <vscale x 4 x i32> %res33}34 35define <vscale x 4 x i32> @scalar_index(ptr %base, i64 %index, <vscale x 4 x i1> %mask) #0 {36; CHECK-LABEL: @scalar_index(37; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i32, ptr [[BASE:%.*]], i64 [[INDEX:%.*]]38; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i32, ptr [[TMP1]], <vscale x 4 x i64> zeroinitializer39; CHECK-NEXT: [[RES:%.*]] = call <vscale x 4 x i32> @llvm.masked.gather.nxv4i32.nxv4p0(<vscale x 4 x ptr> align 4 [[TMP2]], <vscale x 4 x i1> [[MASK:%.*]], <vscale x 4 x i32> undef)40; CHECK-NEXT: ret <vscale x 4 x i32> [[RES]]41;42 %broadcast.splatinsert = insertelement <vscale x 4 x ptr> undef, ptr %base, i32 043 %broadcast.splat = shufflevector <vscale x 4 x ptr> %broadcast.splatinsert, <vscale x 4 x ptr> undef, <vscale x 4 x i32> zeroinitializer44 %gep = getelementptr i32, <vscale x 4 x ptr> %broadcast.splat, i64 %index45 %res = call <vscale x 4 x i32> @llvm.masked.gather.nxv4i32.nxv4p0(<vscale x 4 x ptr> %gep, i32 4, <vscale x 4 x i1> %mask, <vscale x 4 x i32> undef)46 ret <vscale x 4 x i32> %res47}48 49define <vscale x 4 x i32> @splat_index(ptr %base, i64 %index, <vscale x 4 x i1> %mask) #0 {50; CHECK-LABEL: @splat_index(51; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i32, ptr [[BASE:%.*]], i64 [[INDEX:%.*]]52; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i32, ptr [[TMP1]], <vscale x 4 x i64> zeroinitializer53; CHECK-NEXT: [[RES:%.*]] = call <vscale x 4 x i32> @llvm.masked.gather.nxv4i32.nxv4p0(<vscale x 4 x ptr> align 4 [[TMP2]], <vscale x 4 x i1> [[MASK:%.*]], <vscale x 4 x i32> undef)54; CHECK-NEXT: ret <vscale x 4 x i32> [[RES]]55;56 %broadcast.splatinsert = insertelement <vscale x 4 x i64> undef, i64 %index, i32 057 %broadcast.splat = shufflevector <vscale x 4 x i64> %broadcast.splatinsert, <vscale x 4 x i64> undef, <vscale x 4 x i32> zeroinitializer58 %gep = getelementptr i32, ptr %base, <vscale x 4 x i64> %broadcast.splat59 %res = call <vscale x 4 x i32> @llvm.masked.gather.nxv4i32.nxv4p0(<vscale x 4 x ptr> %gep, i32 4, <vscale x 4 x i1> %mask, <vscale x 4 x i32> undef)60 ret <vscale x 4 x i32> %res61}62 63define <vscale x 4 x i32> @test_global_array(<vscale x 4 x i64> %indxs, <vscale x 4 x i1> %mask) #0 {64; CHECK-LABEL: @test_global_array(65; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i32, ptr @glob_array, <vscale x 4 x i64> [[INDXS:%.*]]66; CHECK-NEXT: [[G:%.*]] = call <vscale x 4 x i32> @llvm.masked.gather.nxv4i32.nxv4p0(<vscale x 4 x ptr> align 4 [[TMP1]], <vscale x 4 x i1> [[MASK:%.*]], <vscale x 4 x i32> undef)67; CHECK-NEXT: ret <vscale x 4 x i32> [[G]]68;69 %p = getelementptr inbounds [16 x i32], ptr @glob_array, i64 0, <vscale x 4 x i64> %indxs70 %g = call <vscale x 4 x i32> @llvm.masked.gather.nxv4i32.nxv4p0(<vscale x 4 x ptr> %p, i32 4, <vscale x 4 x i1> %mask, <vscale x 4 x i32> undef)71 ret <vscale x 4 x i32> %g72}73 74define <vscale x 4 x i32> @global_struct_splat(<vscale x 4 x i1> %mask) #0 {75; CHECK-LABEL: @global_struct_splat(76; CHECK-NEXT: [[TMP1:%.*]] = call <vscale x 4 x i32> @llvm.masked.gather.nxv4i32.nxv4p0(<vscale x 4 x ptr> align 4 shufflevector (<vscale x 4 x ptr> insertelement (<vscale x 4 x ptr> poison, ptr getelementptr ([[STRUCT_A:%.*]], ptr @c, i64 0, i32 1), i64 0), <vscale x 4 x ptr> poison, <vscale x 4 x i32> zeroinitializer), <vscale x 4 x i1> [[MASK:%.*]], <vscale x 4 x i32> undef)77; CHECK-NEXT: ret <vscale x 4 x i32> [[TMP1]]78;79 %1 = insertelement <vscale x 4 x ptr> undef, ptr @c, i32 080 %2 = shufflevector <vscale x 4 x ptr> %1, <vscale x 4 x ptr> undef, <vscale x 4 x i32> zeroinitializer81 %3 = getelementptr %struct.a, <vscale x 4 x ptr> %2, <vscale x 4 x i64> zeroinitializer, i32 182 %4 = call <vscale x 4 x i32> @llvm.masked.gather.nxv4i32.nxv4p0(<vscale x 4 x ptr> %3, i32 4, <vscale x 4 x i1> %mask, <vscale x 4 x i32> undef)83 ret <vscale x 4 x i32> %484}85 86define <vscale x 4 x i32> @splat_ptr_gather(ptr %ptr, <vscale x 4 x i1> %mask, <vscale x 4 x i32> %passthru) #0 {87; CHECK-LABEL: @splat_ptr_gather(88; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i32, ptr [[PTR:%.*]], <vscale x 4 x i64> zeroinitializer89; CHECK-NEXT: [[TMP2:%.*]] = call <vscale x 4 x i32> @llvm.masked.gather.nxv4i32.nxv4p0(<vscale x 4 x ptr> align 4 [[TMP1]], <vscale x 4 x i1> [[MASK:%.*]], <vscale x 4 x i32> [[PASSTHRU:%.*]])90; CHECK-NEXT: ret <vscale x 4 x i32> [[TMP2]]91;92 %1 = insertelement <vscale x 4 x ptr> undef, ptr %ptr, i32 093 %2 = shufflevector <vscale x 4 x ptr> %1, <vscale x 4 x ptr> undef, <vscale x 4 x i32> zeroinitializer94 %3 = call <vscale x 4 x i32> @llvm.masked.gather.nxv4i32.nxv4p0(<vscale x 4 x ptr> %2, i32 4, <vscale x 4 x i1> %mask, <vscale x 4 x i32> %passthru)95 ret <vscale x 4 x i32> %396}97 98define void @splat_ptr_scatter(ptr %ptr, <vscale x 4 x i1> %mask, <vscale x 4 x i32> %val) #0 {99; CHECK-LABEL: @splat_ptr_scatter(100; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i32, ptr [[PTR:%.*]], <vscale x 4 x i64> zeroinitializer101; CHECK-NEXT: call void @llvm.masked.scatter.nxv4i32.nxv4p0(<vscale x 4 x i32> [[VAL:%.*]], <vscale x 4 x ptr> align 4 [[TMP1]], <vscale x 4 x i1> [[MASK:%.*]])102; CHECK-NEXT: ret void103;104 %1 = insertelement <vscale x 4 x ptr> undef, ptr %ptr, i32 0105 %2 = shufflevector <vscale x 4 x ptr> %1, <vscale x 4 x ptr> undef, <vscale x 4 x i32> zeroinitializer106 call void @llvm.masked.scatter.nxv4i32.nxv4p0(<vscale x 4 x i32> %val, <vscale x 4 x ptr> %2, i32 4, <vscale x 4 x i1> %mask)107 ret void108}109 110declare <vscale x 4 x i32> @llvm.masked.gather.nxv4i32.nxv4p0(<vscale x 4 x ptr>, i32, <vscale x 4 x i1>, <vscale x 4 x i32>)111declare void @llvm.masked.scatter.nxv4i32.nxv4p0(<vscale x 4 x i32>, <vscale x 4 x ptr>, i32, <vscale x 4 x i1>)112 113attributes #0 = { "target-features"="+sve" }114