1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt < %s -loop-vectorize -scalable-vectorization=on -riscv-v-vector-bits-min=128 -mtriple riscv64-linux-gnu -mattr=+v,+f -S 2>%t | FileCheck %s -check-prefix=CHECK 3 4; A collection of fairly basic functional tests when both fixed and scalable vectorization is 5; allowed. The primary goal of this is check for crashes during cost modeling, but it also 6; exercises the default heuristics in a useful way. 7 8target datalayout = "e-m:e-p:64:64-i64:64-i128:128-n64-S128" 9target triple = "riscv64" 10 11define void @vector_add(ptr noalias nocapture %a, i64 %v, i64 %n) { 12; CHECK-LABEL: @vector_add( 13; CHECK-NEXT: entry: 14; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 15; CHECK: vector.ph: 16; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i64> poison, i64 [[V:%.*]], i32 0 17; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT]], <2 x i64> poison, <2 x i32> zeroinitializer 18; CHECK-NEXT: [[BROADCAST_SPLATINSERT2:%.*]] = insertelement <2 x i64> poison, i64 [[V]], i32 0 19; CHECK-NEXT: [[BROADCAST_SPLAT3:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT2]], <2 x i64> poison, <2 x i32> zeroinitializer 20; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 21; CHECK: vector.body: 22; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 23; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 24; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 2 25; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[TMP0]] 26; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[TMP1]] 27; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i64, ptr [[TMP2]], i32 0 28; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i64>, ptr [[TMP4]], align 8 29; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i64, ptr [[TMP2]], i32 2 30; CHECK-NEXT: [[WIDE_LOAD1:%.*]] = load <2 x i64>, ptr [[TMP5]], align 8 31; CHECK-NEXT: [[TMP6:%.*]] = add <2 x i64> [[WIDE_LOAD]], [[BROADCAST_SPLAT]] 32; CHECK-NEXT: [[TMP7:%.*]] = add <2 x i64> [[WIDE_LOAD1]], [[BROADCAST_SPLAT3]] 33; CHECK-NEXT: store <2 x i64> [[TMP6]], ptr [[TMP4]], align 8 34; CHECK-NEXT: store <2 x i64> [[TMP7]], ptr [[TMP5]], align 8 35; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 36; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024 37; CHECK-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 38; CHECK: middle.block: 39; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, 1024 40; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] 41; CHECK: scalar.ph: 42; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1024, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 43; CHECK-NEXT: br label [[FOR_BODY:%.*]] 44; CHECK: for.body: 45; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] 46; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[IV]] 47; CHECK-NEXT: [[ELEM:%.*]] = load i64, ptr [[ARRAYIDX]], align 8 48; CHECK-NEXT: [[ADD:%.*]] = add i64 [[ELEM]], [[V]] 49; CHECK-NEXT: store i64 [[ADD]], ptr [[ARRAYIDX]], align 8 50; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 51; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], 1024 52; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP2:![0-9]+]] 53; CHECK: for.end: 54; CHECK-NEXT: ret void 55; 56entry: 57 br label %for.body 58 59for.body: 60 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] 61 %arrayidx = getelementptr inbounds i64, ptr %a, i64 %iv 62 %elem = load i64, ptr %arrayidx 63 %add = add i64 %elem, %v 64 store i64 %add, ptr %arrayidx 65 %iv.next = add nuw nsw i64 %iv, 1 66 %exitcond.not = icmp eq i64 %iv.next, 1024 67 br i1 %exitcond.not, label %for.end, label %for.body 68 69for.end: 70 ret void 71} 72 73define void @splat_int(ptr noalias nocapture %a, i64 %v, i64 %n) { 74; CHECK-LABEL: @splat_int( 75; CHECK-NEXT: entry: 76; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 77; CHECK: vector.ph: 78; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i64> poison, i64 [[V:%.*]], i32 0 79; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT]], <2 x i64> poison, <2 x i32> zeroinitializer 80; CHECK-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <2 x i64> poison, i64 [[V]], i32 0 81; CHECK-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT1]], <2 x i64> poison, <2 x i32> zeroinitializer 82; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 83; CHECK: vector.body: 84; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 85; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 86; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 2 87; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[TMP0]] 88; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[TMP1]] 89; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i64, ptr [[TMP2]], i32 0 90; CHECK-NEXT: store <2 x i64> [[BROADCAST_SPLAT]], ptr [[TMP4]], align 8 91; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i64, ptr [[TMP2]], i32 2 92; CHECK-NEXT: store <2 x i64> [[BROADCAST_SPLAT2]], ptr [[TMP5]], align 8 93; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 94; CHECK-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024 95; CHECK-NEXT: br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] 96; CHECK: middle.block: 97; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, 1024 98; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] 99; CHECK: scalar.ph: 100; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1024, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 101; CHECK-NEXT: br label [[FOR_BODY:%.*]] 102; CHECK: for.body: 103; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] 104; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[IV]] 105; CHECK-NEXT: store i64 [[V]], ptr [[ARRAYIDX]], align 8 106; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 107; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], 1024 108; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] 109; CHECK: for.end: 110; CHECK-NEXT: ret void 111; 112entry: 113 br label %for.body 114 115for.body: 116 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] 117 %arrayidx = getelementptr inbounds i64, ptr %a, i64 %iv 118 store i64 %v, ptr %arrayidx 119 %iv.next = add nuw nsw i64 %iv, 1 120 %exitcond.not = icmp eq i64 %iv.next, 1024 121 br i1 %exitcond.not, label %for.end, label %for.body 122 123for.end: 124 ret void 125} 126 127define void @splat_ptr(ptr noalias nocapture %a, ptr %v, i64 %n) { 128; CHECK-LABEL: @splat_ptr( 129; CHECK-NEXT: entry: 130; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 131; CHECK: vector.ph: 132; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x ptr> poison, ptr [[V:%.*]], i32 0 133; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x ptr> [[BROADCAST_SPLATINSERT]], <2 x ptr> poison, <2 x i32> zeroinitializer 134; CHECK-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <2 x ptr> poison, ptr [[V]], i32 0 135; CHECK-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <2 x ptr> [[BROADCAST_SPLATINSERT1]], <2 x ptr> poison, <2 x i32> zeroinitializer 136; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 137; CHECK: vector.body: 138; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 139; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 140; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 2 141; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[TMP0]] 142; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[TMP1]] 143; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds ptr, ptr [[TMP2]], i32 0 144; CHECK-NEXT: store <2 x ptr> [[BROADCAST_SPLAT]], ptr [[TMP4]], align 8 145; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds ptr, ptr [[TMP2]], i32 2 146; CHECK-NEXT: store <2 x ptr> [[BROADCAST_SPLAT2]], ptr [[TMP5]], align 8 147; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 148; CHECK-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024 149; CHECK-NEXT: br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] 150; CHECK: middle.block: 151; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, 1024 152; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] 153; CHECK: scalar.ph: 154; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1024, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 155; CHECK-NEXT: br label [[FOR_BODY:%.*]] 156; CHECK: for.body: 157; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] 158; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[IV]] 159; CHECK-NEXT: store ptr [[V]], ptr [[ARRAYIDX]], align 8 160; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 161; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], 1024 162; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]] 163; CHECK: for.end: 164; CHECK-NEXT: ret void 165; 166entry: 167 br label %for.body 168 169for.body: 170 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] 171 %arrayidx = getelementptr inbounds i64, ptr %a, i64 %iv 172 store ptr %v, ptr %arrayidx 173 %iv.next = add nuw nsw i64 %iv, 1 174 %exitcond.not = icmp eq i64 %iv.next, 1024 175 br i1 %exitcond.not, label %for.end, label %for.body 176 177for.end: 178 ret void 179} 180 181