1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt -passes=slp-vectorizer -mtriple=arm64-apple-ios -S %s | FileCheck %s 3; RUN: opt -aa-pipeline='basic-aa,scoped-noalias-aa' -passes=slp-vectorizer -mtriple=arm64-apple-darwin -S %s | FileCheck %s 4 5define void @loop1(ptr %A, ptr %B, i64 %N) { 6; CHECK-LABEL: @loop1( 7; CHECK-NEXT: entry: 8; CHECK-NEXT: br label [[LOOP:%.*]] 9; CHECK: loop: 10; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ] 11; CHECK-NEXT: [[B_GEP_0:%.*]] = getelementptr inbounds i32, ptr [[B:%.*]], i64 [[IV]] 12; CHECK-NEXT: [[B_0:%.*]] = load i32, ptr [[B_GEP_0]], align 4 13; CHECK-NEXT: [[A_GEP_0:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[IV]] 14; CHECK-NEXT: [[A_0:%.*]] = load i32, ptr [[A_GEP_0]], align 4 15; CHECK-NEXT: [[ADD_0:%.*]] = add i32 [[A_0]], 20 16; CHECK-NEXT: [[XOR_0:%.*]] = xor i32 [[ADD_0]], [[B_0]] 17; CHECK-NEXT: store i32 [[XOR_0]], ptr [[A_GEP_0]], align 4 18; CHECK-NEXT: [[IV_1:%.*]] = or i64 [[IV]], 1 19; CHECK-NEXT: [[B_GEP_1:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[IV_1]] 20; CHECK-NEXT: [[B_1:%.*]] = load i32, ptr [[B_GEP_1]], align 4 21; CHECK-NEXT: [[A_GEP_1:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[IV_1]] 22; CHECK-NEXT: [[A_1:%.*]] = load i32, ptr [[A_GEP_1]], align 4 23; CHECK-NEXT: [[ADD_1:%.*]] = add i32 [[A_1]], 20 24; CHECK-NEXT: [[XOR_1:%.*]] = xor i32 [[ADD_1]], [[B_1]] 25; CHECK-NEXT: store i32 [[XOR_1]], ptr [[A_GEP_1]], align 4 26; CHECK-NEXT: [[IV_2:%.*]] = or i64 [[IV]], 2 27; CHECK-NEXT: [[B_GEP_2:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[IV_2]] 28; CHECK-NEXT: [[B_2:%.*]] = load i32, ptr [[B_GEP_2]], align 4 29; CHECK-NEXT: [[A_GEP_2:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[IV_2]] 30; CHECK-NEXT: [[A_2:%.*]] = load i32, ptr [[A_GEP_2]], align 4 31; CHECK-NEXT: [[ADD_2:%.*]] = add i32 [[A_2]], 20 32; CHECK-NEXT: [[XOR_2:%.*]] = xor i32 [[ADD_2]], [[B_2]] 33; CHECK-NEXT: store i32 [[XOR_2]], ptr [[A_GEP_2]], align 4 34; CHECK-NEXT: [[IV_3:%.*]] = or i64 [[IV]], 3 35; CHECK-NEXT: [[B_GEP_3:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[IV_3]] 36; CHECK-NEXT: [[B_3:%.*]] = load i32, ptr [[B_GEP_3]], align 4 37; CHECK-NEXT: [[A_GEP_3:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[IV_3]] 38; CHECK-NEXT: [[A_3:%.*]] = load i32, ptr [[A_GEP_3]], align 4 39; CHECK-NEXT: [[ADD_3:%.*]] = add i32 [[A_3]], 20 40; CHECK-NEXT: [[XOR_3:%.*]] = xor i32 [[ADD_3]], [[B_3]] 41; CHECK-NEXT: store i32 [[XOR_3]], ptr [[A_GEP_3]], align 4 42; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 16 43; CHECK-NEXT: [[COND:%.*]] = icmp ult i64 [[IV_NEXT]], [[N:%.*]] 44; CHECK-NEXT: br i1 [[COND]], label [[LOOP]], label [[EXIT:%.*]] 45; CHECK: exit: 46; CHECK-NEXT: ret void 47; 48entry: 49 br label %loop 50 51loop: 52 %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ] 53 %B.gep.0 = getelementptr inbounds i32, ptr %B, i64 %iv 54 %B.0 = load i32, ptr %B.gep.0, align 4 55 %A.gep.0 = getelementptr inbounds i32, ptr %A, i64 %iv 56 %A.0 = load i32, ptr %A.gep.0, align 4 57 %add.0 = add i32 %A.0, 20 58 %xor.0 = xor i32 %add.0, %B.0 59 store i32 %xor.0, ptr %A.gep.0, align 4 60 %iv.1 = or i64 %iv, 1 61 %B.gep.1 = getelementptr inbounds i32, ptr %B, i64 %iv.1 62 %B.1 = load i32, ptr %B.gep.1, align 4 63 %A.gep.1 = getelementptr inbounds i32, ptr %A, i64 %iv.1 64 %A.1 = load i32, ptr %A.gep.1, align 4 65 %add.1 = add i32 %A.1, 20 66 %xor.1 = xor i32 %add.1, %B.1 67 store i32 %xor.1, ptr %A.gep.1, align 4 68 %iv.2 = or i64 %iv, 2 69 %B.gep.2 = getelementptr inbounds i32, ptr %B, i64 %iv.2 70 %B.2 = load i32, ptr %B.gep.2, align 4 71 %A.gep.2 = getelementptr inbounds i32, ptr %A, i64 %iv.2 72 %A.2 = load i32, ptr %A.gep.2, align 4 73 %add.2 = add i32 %A.2, 20 74 %xor.2 = xor i32 %add.2, %B.2 75 store i32 %xor.2, ptr %A.gep.2, align 4 76 %iv.3 = or i64 %iv, 3 77 %B.gep.3 = getelementptr inbounds i32, ptr %B, i64 %iv.3 78 %B.3 = load i32, ptr %B.gep.3, align 4 79 %A.gep.3 = getelementptr inbounds i32, ptr %A, i64 %iv.3 80 %A.3 = load i32, ptr %A.gep.3, align 4 81 %add.3 = add i32 %A.3, 20 82 %xor.3 = xor i32 %add.3, %B.3 83 store i32 %xor.3, ptr %A.gep.3, align 4 84 %iv.next = add nuw nsw i64 %iv, 16 85 %cond = icmp ult i64 %iv.next, %N 86 br i1 %cond, label %loop, label %exit 87 88exit: 89 ret void 90} 91 92define void @loop_iv_update_at_start(ptr %src, ptr %dst) #0 { 93; CHECK-LABEL: @loop_iv_update_at_start( 94; CHECK-NEXT: entry: 95; CHECK-NEXT: br label [[LOOP:%.*]] 96; CHECK: loop: 97; CHECK-NEXT: [[IV:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ] 98; CHECK-NEXT: [[IV_NEXT]] = add i32 [[IV]], 1 99; CHECK-NEXT: [[COND:%.*]] = icmp ult i32 [[IV]], 2000 100; CHECK-NEXT: [[SRC_0:%.*]] = load float, ptr [[SRC:%.*]], align 8 101; CHECK-NEXT: [[ADD_0:%.*]] = fadd float [[SRC_0]], 1.000000e+00 102; CHECK-NEXT: [[MUL_0:%.*]] = fmul float [[ADD_0]], [[SRC_0]] 103; CHECK-NEXT: store float [[MUL_0]], ptr [[DST:%.*]], align 8 104; CHECK-NEXT: [[SRC_GEP_1:%.*]] = getelementptr inbounds float, ptr [[SRC]], i64 1 105; CHECK-NEXT: [[SRC_1:%.*]] = load float, ptr [[SRC_GEP_1]], align 8 106; CHECK-NEXT: [[ADD_1:%.*]] = fadd float [[SRC_1]], 1.000000e+00 107; CHECK-NEXT: [[MUL_1:%.*]] = fmul float [[ADD_1]], [[SRC_1]] 108; CHECK-NEXT: [[DST_GEP_1:%.*]] = getelementptr inbounds float, ptr [[DST]], i64 1 109; CHECK-NEXT: store float [[MUL_1]], ptr [[DST_GEP_1]], align 8 110; CHECK-NEXT: [[SRC_GEP_2:%.*]] = getelementptr inbounds float, ptr [[SRC]], i64 2 111; CHECK-NEXT: [[SRC_2:%.*]] = load float, ptr [[SRC_GEP_2]], align 8 112; CHECK-NEXT: [[ADD_2:%.*]] = fadd float [[SRC_2]], 1.000000e+00 113; CHECK-NEXT: [[MUL_2:%.*]] = fmul float [[ADD_2]], [[SRC_2]] 114; CHECK-NEXT: [[DST_GEP_2:%.*]] = getelementptr inbounds float, ptr [[DST]], i64 2 115; CHECK-NEXT: store float [[MUL_2]], ptr [[DST_GEP_2]], align 8 116; CHECK-NEXT: [[SRC_GEP_3:%.*]] = getelementptr inbounds float, ptr [[SRC]], i64 3 117; CHECK-NEXT: [[SRC_3:%.*]] = load float, ptr [[SRC_GEP_3]], align 8 118; CHECK-NEXT: [[ADD_3:%.*]] = fadd float [[SRC_3]], 1.000000e+00 119; CHECK-NEXT: [[MUL_3:%.*]] = fmul float [[ADD_3]], [[SRC_3]] 120; CHECK-NEXT: [[DST_GEP_3:%.*]] = getelementptr inbounds float, ptr [[DST]], i64 3 121; CHECK-NEXT: store float [[MUL_3]], ptr [[DST_GEP_3]], align 8 122; CHECK-NEXT: [[SRC_GEP_4:%.*]] = getelementptr inbounds float, ptr [[SRC]], i64 4 123; CHECK-NEXT: [[SRC_4:%.*]] = load float, ptr [[SRC_GEP_4]], align 8 124; CHECK-NEXT: [[ADD_4:%.*]] = fadd float [[SRC_4]], 1.000000e+00 125; CHECK-NEXT: [[MUL_4:%.*]] = fmul float [[ADD_4]], [[SRC_4]] 126; CHECK-NEXT: [[DST_GEP_4:%.*]] = getelementptr inbounds float, ptr [[DST]], i64 4 127; CHECK-NEXT: store float [[MUL_4]], ptr [[DST_GEP_4]], align 8 128; CHECK-NEXT: br i1 [[COND]], label [[LOOP]], label [[EXIT:%.*]] 129; CHECK: exit: 130; CHECK-NEXT: ret void 131; 132entry: 133 br label %loop 134 135loop: 136 %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ] 137 %iv.next = add i32 %iv, 1 138 %cond = icmp ult i32 %iv, 2000 139 140 %src.0 = load float, ptr %src, align 8 141 %add.0 = fadd float %src.0, 1.0 142 %mul.0 = fmul float %add.0, %src.0 143 store float %mul.0, ptr %dst, align 8 144 145 %src.gep.1 = getelementptr inbounds float, ptr %src, i64 1 146 %src.1 = load float, ptr %src.gep.1, align 8 147 %add.1 = fadd float %src.1, 1.0 148 %mul.1 = fmul float %add.1, %src.1 149 %dst.gep.1 = getelementptr inbounds float, ptr %dst, i64 1 150 store float %mul.1, ptr %dst.gep.1, align 8 151 %src.gep.2 = getelementptr inbounds float, ptr %src, i64 2 152 %src.2 = load float, ptr %src.gep.2, align 8 153 %add.2 = fadd float %src.2, 1.0 154 %mul.2 = fmul float %add.2, %src.2 155 %dst.gep.2 = getelementptr inbounds float, ptr %dst, i64 2 156 store float %mul.2, ptr %dst.gep.2, align 8 157 %src.gep.3 = getelementptr inbounds float, ptr %src, i64 3 158 %src.3 = load float, ptr %src.gep.3, align 8 159 %add.3 = fadd float %src.3, 1.0 160 %mul.3 = fmul float %add.3, %src.3 161 %dst.gep.3 = getelementptr inbounds float, ptr %dst, i64 3 162 store float %mul.3, ptr %dst.gep.3, align 8 163 %src.gep.4 = getelementptr inbounds float, ptr %src, i64 4 164 %src.4 = load float, ptr %src.gep.4, align 8 165 %add.4 = fadd float %src.4, 1.0 166 %mul.4 = fmul float %add.4, %src.4 167 %dst.gep.4 = getelementptr inbounds float, ptr %dst, i64 4 168 store float %mul.4, ptr %dst.gep.4, align 8 169 br i1 %cond, label %loop, label %exit 170 171exit: 172 ret void 173} 174 175; Similar to @loop1, but a load is used in a phi in the same basic block. 176define i32 @value_used_in_phi(ptr %A, ptr %B, i64 %N) { 177; CHECK-LABEL: @value_used_in_phi( 178; CHECK-NEXT: entry: 179; CHECK-NEXT: br label [[LOOP:%.*]] 180; CHECK: loop: 181; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ] 182; CHECK-NEXT: [[P:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[A_3:%.*]], [[LOOP]] ] 183; CHECK-NEXT: [[B_GEP_0:%.*]] = getelementptr inbounds i32, ptr [[B:%.*]], i64 [[IV]] 184; CHECK-NEXT: [[B_0:%.*]] = load i32, ptr [[B_GEP_0]], align 4 185; CHECK-NEXT: [[A_GEP_0:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[IV]] 186; CHECK-NEXT: [[A_0:%.*]] = load i32, ptr [[A_GEP_0]], align 4 187; CHECK-NEXT: [[ADD_0:%.*]] = add i32 [[A_0]], 20 188; CHECK-NEXT: [[XOR_0:%.*]] = xor i32 [[ADD_0]], [[B_0]] 189; CHECK-NEXT: store i32 [[XOR_0]], ptr [[A_GEP_0]], align 4 190; CHECK-NEXT: [[IV_1:%.*]] = or i64 [[IV]], 1 191; CHECK-NEXT: [[B_GEP_1:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[IV_1]] 192; CHECK-NEXT: [[B_1:%.*]] = load i32, ptr [[B_GEP_1]], align 4 193; CHECK-NEXT: [[A_GEP_1:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[IV_1]] 194; CHECK-NEXT: [[A_1:%.*]] = load i32, ptr [[A_GEP_1]], align 4 195; CHECK-NEXT: [[ADD_1:%.*]] = add i32 [[A_1]], 20 196; CHECK-NEXT: [[XOR_1:%.*]] = xor i32 [[ADD_1]], [[B_1]] 197; CHECK-NEXT: store i32 [[XOR_1]], ptr [[A_GEP_1]], align 4 198; CHECK-NEXT: [[IV_2:%.*]] = or i64 [[IV]], 2 199; CHECK-NEXT: [[B_GEP_2:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[IV_2]] 200; CHECK-NEXT: [[B_2:%.*]] = load i32, ptr [[B_GEP_2]], align 4 201; CHECK-NEXT: [[A_GEP_2:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[IV_2]] 202; CHECK-NEXT: [[A_2:%.*]] = load i32, ptr [[A_GEP_2]], align 4 203; CHECK-NEXT: [[ADD_2:%.*]] = add i32 [[A_2]], 20 204; CHECK-NEXT: [[XOR_2:%.*]] = xor i32 [[ADD_2]], [[B_2]] 205; CHECK-NEXT: store i32 [[XOR_2]], ptr [[A_GEP_2]], align 4 206; CHECK-NEXT: [[IV_3:%.*]] = or i64 [[IV]], 3 207; CHECK-NEXT: [[B_GEP_3:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[IV_3]] 208; CHECK-NEXT: [[B_3:%.*]] = load i32, ptr [[B_GEP_3]], align 4 209; CHECK-NEXT: [[A_GEP_3:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[IV_3]] 210; CHECK-NEXT: [[A_3]] = load i32, ptr [[A_GEP_3]], align 4 211; CHECK-NEXT: [[ADD_3:%.*]] = add i32 [[A_3]], 20 212; CHECK-NEXT: [[XOR_3:%.*]] = xor i32 [[ADD_3]], [[B_3]] 213; CHECK-NEXT: store i32 [[XOR_3]], ptr [[A_GEP_3]], align 4 214; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 16 215; CHECK-NEXT: [[COND:%.*]] = icmp ult i64 [[IV_NEXT]], [[N:%.*]] 216; CHECK-NEXT: br i1 [[COND]], label [[LOOP]], label [[EXIT:%.*]] 217; CHECK: exit: 218; CHECK-NEXT: ret i32 [[P]] 219; 220entry: 221 br label %loop 222 223loop: 224 %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ] 225 %p = phi i32 [ 0, %entry ], [ %A.3, %loop ] 226 %B.gep.0 = getelementptr inbounds i32, ptr %B, i64 %iv 227 %B.0 = load i32, ptr %B.gep.0, align 4 228 %A.gep.0 = getelementptr inbounds i32, ptr %A, i64 %iv 229 %A.0 = load i32, ptr %A.gep.0, align 4 230 %add.0 = add i32 %A.0, 20 231 %xor.0 = xor i32 %add.0, %B.0 232 store i32 %xor.0, ptr %A.gep.0, align 4 233 %iv.1 = or i64 %iv, 1 234 %B.gep.1 = getelementptr inbounds i32, ptr %B, i64 %iv.1 235 %B.1 = load i32, ptr %B.gep.1, align 4 236 %A.gep.1 = getelementptr inbounds i32, ptr %A, i64 %iv.1 237 %A.1 = load i32, ptr %A.gep.1, align 4 238 %add.1 = add i32 %A.1, 20 239 %xor.1 = xor i32 %add.1, %B.1 240 store i32 %xor.1, ptr %A.gep.1, align 4 241 %iv.2 = or i64 %iv, 2 242 %B.gep.2 = getelementptr inbounds i32, ptr %B, i64 %iv.2 243 %B.2 = load i32, ptr %B.gep.2, align 4 244 %A.gep.2 = getelementptr inbounds i32, ptr %A, i64 %iv.2 245 %A.2 = load i32, ptr %A.gep.2, align 4 246 %add.2 = add i32 %A.2, 20 247 %xor.2 = xor i32 %add.2, %B.2 248 store i32 %xor.2, ptr %A.gep.2, align 4 249 %iv.3 = or i64 %iv, 3 250 %B.gep.3 = getelementptr inbounds i32, ptr %B, i64 %iv.3 251 %B.3 = load i32, ptr %B.gep.3, align 4 252 %A.gep.3 = getelementptr inbounds i32, ptr %A, i64 %iv.3 253 %A.3 = load i32, ptr %A.gep.3, align 4 254 %add.3 = add i32 %A.3, 20 255 %xor.3 = xor i32 %add.3, %B.3 256 store i32 %xor.3, ptr %A.gep.3, align 4 257 %iv.next = add nuw nsw i64 %iv, 16 258 %cond = icmp ult i64 %iv.next, %N 259 br i1 %cond, label %loop, label %exit 260 261exit: 262 ret i32 %p 263} 264 265 266