1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 3
2 ; RUN: opt -passes="print<scalar-evolution>,loop-vectorize" --verify-scev -S < %s -force-vector-width=4 2>/dev/null | FileCheck %s
4 ; Make sure users of SCEVUnknowns from the scalar loop are invalidated.
6 define void @pr66616(ptr %ptr) {
7 ; CHECK-LABEL: define void @pr66616(
8 ; CHECK-SAME: ptr [[PTR:%.*]]) {
10 ; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
12 ; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
14 ; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
15 ; CHECK-NEXT: [[TMP0:%.*]] = load i32, ptr [[PTR]], align 4
16 ; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[TMP0]], i64 0
17 ; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT]], <4 x i32> poison, <4 x i32> zeroinitializer
18 ; CHECK-NEXT: [[TMP1:%.*]] = add <4 x i32> [[BROADCAST_SPLAT]], <i32 1, i32 1, i32 1, i32 1>
19 ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
20 ; CHECK-NEXT: [[TMP2:%.*]] = icmp eq i32 [[INDEX_NEXT]], 256
21 ; CHECK-NEXT: br i1 [[TMP2]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
22 ; CHECK: middle.block:
23 ; CHECK-NEXT: [[TMP3:%.*]] = extractelement <4 x i32> [[TMP1]], i32 3
24 ; CHECK-NEXT: br i1 true, label [[PREHEADER:%.*]], label [[SCALAR_PH]]
26 ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i8 [ 0, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
27 ; CHECK-NEXT: br label [[LOOP_1:%.*]]
29 ; CHECK-NEXT: [[IV_1:%.*]] = phi i8 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INC:%.*]], [[LOOP_1]] ]
30 ; CHECK-NEXT: [[LOAD:%.*]] = load i32, ptr [[PTR]], align 4
31 ; CHECK-NEXT: [[ADD3:%.*]] = add i32 [[LOAD]], 1
32 ; CHECK-NEXT: [[INC]] = add i8 [[IV_1]], 1
33 ; CHECK-NEXT: [[COND1:%.*]] = icmp eq i8 [[INC]], 0
34 ; CHECK-NEXT: br i1 [[COND1]], label [[PREHEADER]], label [[LOOP_1]], !llvm.loop [[LOOP3:![0-9]+]]
36 ; CHECK-NEXT: [[ADD3_LCSSA:%.*]] = phi i32 [ [[ADD3]], [[LOOP_1]] ], [ [[TMP3]], [[MIDDLE_BLOCK]] ]
37 ; CHECK-NEXT: [[TMP4:%.*]] = sub i32 0, [[ADD3_LCSSA]]
38 ; CHECK-NEXT: [[TMP5:%.*]] = zext i32 [[TMP4]] to i64
39 ; CHECK-NEXT: [[TMP6:%.*]] = add nuw nsw i64 [[TMP5]], 1
40 ; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP6]], 4
41 ; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH2:%.*]], label [[VECTOR_PH3:%.*]]
43 ; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP6]], 4
44 ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP6]], [[N_MOD_VF]]
45 ; CHECK-NEXT: [[DOTCAST:%.*]] = trunc i64 [[N_VEC]] to i32
46 ; CHECK-NEXT: [[IND_END:%.*]] = add i32 [[ADD3_LCSSA]], [[DOTCAST]]
47 ; CHECK-NEXT: [[IND_END5:%.*]] = getelementptr i8, ptr [[PTR]], i64 [[N_VEC]]
48 ; CHECK-NEXT: br label [[VECTOR_BODY7:%.*]]
49 ; CHECK: vector.body7:
50 ; CHECK-NEXT: [[INDEX8:%.*]] = phi i64 [ 0, [[VECTOR_PH3]] ], [ [[INDEX_NEXT9:%.*]], [[VECTOR_BODY7]] ]
51 ; CHECK-NEXT: [[INDEX_NEXT9]] = add nuw i64 [[INDEX8]], 4
52 ; CHECK-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT9]], [[N_VEC]]
53 ; CHECK-NEXT: br i1 [[TMP7]], label [[MIDDLE_BLOCK1:%.*]], label [[VECTOR_BODY7]], !llvm.loop [[LOOP4:![0-9]+]]
54 ; CHECK: middle.block1:
55 ; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP6]], [[N_VEC]]
56 ; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH2]]
58 ; CHECK-NEXT: [[BC_RESUME_VAL4:%.*]] = phi i32 [ [[IND_END]], [[MIDDLE_BLOCK1]] ], [ [[ADD3_LCSSA]], [[PREHEADER]] ]
59 ; CHECK-NEXT: [[BC_RESUME_VAL6:%.*]] = phi ptr [ [[IND_END5]], [[MIDDLE_BLOCK1]] ], [ [[PTR]], [[PREHEADER]] ]
60 ; CHECK-NEXT: br label [[LOOP_2:%.*]]
62 ; CHECK-NEXT: [[IV_2:%.*]] = phi i32 [ [[IV_2_I:%.*]], [[LOOP_2]] ], [ [[BC_RESUME_VAL4]], [[SCALAR_PH2]] ]
63 ; CHECK-NEXT: [[IV_3:%.*]] = phi ptr [ [[IV_3_I:%.*]], [[LOOP_2]] ], [ [[BC_RESUME_VAL6]], [[SCALAR_PH2]] ]
64 ; CHECK-NEXT: [[IV_2_I]] = add i32 [[IV_2]], 1
65 ; CHECK-NEXT: [[IV_3_I]] = getelementptr i8, ptr [[IV_3]], i64 1
66 ; CHECK-NEXT: [[COND2:%.*]] = icmp eq i32 [[IV_2]], 0
67 ; CHECK-NEXT: br i1 [[COND2]], label [[EXIT]], label [[LOOP_2]], !llvm.loop [[LOOP5:![0-9]+]]
69 ; CHECK-NEXT: ret void
75 %iv.1 = phi i8 [ 0, %entry ], [ %inc, %loop.1 ]
76 %load = load i32, ptr %ptr, align 4
77 %add3 = add i32 %load, 1
78 %inc = add i8 %iv.1, 1
79 %cond1 = icmp eq i8 %inc, 0
80 br i1 %cond1, label %preheader, label %loop.1
86 %iv.2 = phi i32 [ %iv.2.i, %loop.2 ], [ %add3, %preheader ]
87 %iv.3 = phi ptr [ %iv.3.i, %loop.2 ], [ %ptr, %preheader ]
88 %iv.2.i = add i32 %iv.2, 1
89 %iv.3.i = getelementptr i8, ptr %iv.3, i64 1
90 %cond2 = icmp eq i32 %iv.2, 0
91 br i1 %cond2, label %exit, label %loop.2