xref: /llvm-project/llvm/test/Transforms/LoopVersioning/wrapping-pointer-non-integral-addrspace.ll (revision f719cfa8685a30a3f4115cc0ce446262daf81244)
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
2; RUN: opt -passes=loop-versioning -S < %s | FileCheck %s
3
4; NB: addrspaces 10-13 are non-integral
5target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128-ni:10:11:12:13"
6
7; This matches the test case from PR38290
8; Check that we expand the SCEV predicate check using GEP, rather
9; than ptrtoint.
10
11%jl_value_t = type opaque
12%jl_array_t = type { ptr addrspace(13), i64, i16, i16, i32 }
13
14declare i64 @julia_steprange_last_4949()
15
16define void @wrapping_ptr_nonint_addrspace(ptr %arg) {
17; CHECK-LABEL: define void @wrapping_ptr_nonint_addrspace(
18; CHECK-SAME: ptr [[ARG:%.*]]) {
19; CHECK-NEXT:  [[LOOP_LVER_CHECK:.*:]]
20; CHECK-NEXT:    [[LOAD0:%.*]] = load ptr addrspace(10), ptr [[ARG]], align 8
21; CHECK-NEXT:    [[LOAD1:%.*]] = load i32, ptr inttoptr (i64 12 to ptr), align 4
22; CHECK-NEXT:    [[SUB:%.*]] = sub i32 0, [[LOAD1]]
23; CHECK-NEXT:    [[CALL:%.*]] = call i64 @julia_steprange_last_4949()
24; CHECK-NEXT:    [[CAST0:%.*]] = addrspacecast ptr addrspace(10) [[LOAD0]] to ptr addrspace(11)
25; CHECK-NEXT:    [[LOAD2:%.*]] = load ptr addrspace(10), ptr addrspace(11) [[CAST0]], align 8
26; CHECK-NEXT:    [[CAST1:%.*]] = addrspacecast ptr addrspace(10) [[LOAD2]] to ptr addrspace(11)
27; CHECK-NEXT:    [[LOAD3:%.*]] = load ptr addrspace(13), ptr addrspace(11) [[CAST1]], align 8
28; CHECK-NEXT:    [[SEXT:%.*]] = sext i32 [[SUB]] to i64
29; CHECK-NEXT:    [[TMP0:%.*]] = shl i64 [[CALL]], 2
30; CHECK-NEXT:    [[TMP1:%.*]] = shl nsw i64 [[SEXT]], 2
31; CHECK-NEXT:    [[TMP2:%.*]] = add i64 [[TMP0]], [[TMP1]]
32; CHECK-NEXT:    [[TMP3:%.*]] = add i64 [[TMP2]], -4
33; CHECK-NEXT:    [[SCEVGEP:%.*]] = getelementptr i8, ptr addrspace(13) [[LOAD3]], i64 [[TMP3]]
34; CHECK-NEXT:    [[SCEVGEP1:%.*]] = getelementptr i8, ptr addrspace(13) [[LOAD3]], i64 [[TMP1]]
35; CHECK-NEXT:    [[TMP4:%.*]] = add i64 [[TMP0]], -4
36; CHECK-NEXT:    [[SCEVGEP2:%.*]] = getelementptr i8, ptr addrspace(13) [[LOAD3]], i64 [[TMP4]]
37; CHECK-NEXT:    [[BOUND0:%.*]] = icmp ult ptr addrspace(13) [[SCEVGEP]], [[LOAD3]]
38; CHECK-NEXT:    [[BOUND1:%.*]] = icmp ult ptr addrspace(13) [[SCEVGEP2]], [[SCEVGEP1]]
39; CHECK-NEXT:    [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
40; CHECK-NEXT:    br i1 [[FOUND_CONFLICT]], label %[[LOOP_PH_LVER_ORIG:.*]], label %[[LOOP_PH:.*]]
41; CHECK:       [[LOOP_PH_LVER_ORIG]]:
42; CHECK-NEXT:    br label %[[LOOP_LVER_ORIG:.*]]
43; CHECK:       [[LOOP_LVER_ORIG]]:
44; CHECK-NEXT:    [[VALUE_PHI3_LVER_ORIG:%.*]] = phi i64 [ 0, %[[LOOP_PH_LVER_ORIG]] ], [ [[ADD0_LVER_ORIG:%.*]], %[[LOOP_LVER_ORIG]] ]
45; CHECK-NEXT:    [[ADD0_LVER_ORIG]] = add i64 [[VALUE_PHI3_LVER_ORIG]], -1
46; CHECK-NEXT:    [[GEP0_LVER_ORIG:%.*]] = getelementptr inbounds i32, ptr addrspace(13) [[LOAD3]], i64 [[ADD0_LVER_ORIG]]
47; CHECK-NEXT:    [[LOAD4_LVER_ORIG:%.*]] = load i32, ptr addrspace(13) [[GEP0_LVER_ORIG]], align 4
48; CHECK-NEXT:    [[ADD1_LVER_ORIG:%.*]] = add i64 [[ADD0_LVER_ORIG]], [[SEXT]]
49; CHECK-NEXT:    [[GEP1_LVER_ORIG:%.*]] = getelementptr inbounds i32, ptr addrspace(13) [[LOAD3]], i64 [[ADD1_LVER_ORIG]]
50; CHECK-NEXT:    store i32 [[LOAD4_LVER_ORIG]], ptr addrspace(13) [[GEP1_LVER_ORIG]], align 4
51; CHECK-NEXT:    [[CMP_LVER_ORIG:%.*]] = icmp eq i64 [[VALUE_PHI3_LVER_ORIG]], [[CALL]]
52; CHECK-NEXT:    br i1 [[CMP_LVER_ORIG]], label %[[EXIT_LOOPEXIT:.*]], label %[[LOOP_LVER_ORIG]]
53; CHECK:       [[LOOP_PH]]:
54; CHECK-NEXT:    br label %[[LOOP:.*]]
55; CHECK:       [[LOOP]]:
56; CHECK-NEXT:    [[VALUE_PHI3:%.*]] = phi i64 [ 0, %[[LOOP_PH]] ], [ [[ADD0:%.*]], %[[LOOP]] ]
57; CHECK-NEXT:    [[ADD0]] = add i64 [[VALUE_PHI3]], -1
58; CHECK-NEXT:    [[GEP0:%.*]] = getelementptr inbounds i32, ptr addrspace(13) [[LOAD3]], i64 [[ADD0]]
59; CHECK-NEXT:    [[LOAD4:%.*]] = load i32, ptr addrspace(13) [[GEP0]], align 4, !alias.scope [[META0:![0-9]+]]
60; CHECK-NEXT:    [[ADD1:%.*]] = add i64 [[ADD0]], [[SEXT]]
61; CHECK-NEXT:    [[GEP1:%.*]] = getelementptr inbounds i32, ptr addrspace(13) [[LOAD3]], i64 [[ADD1]]
62; CHECK-NEXT:    store i32 [[LOAD4]], ptr addrspace(13) [[GEP1]], align 4, !alias.scope [[META3:![0-9]+]], !noalias [[META0]]
63; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i64 [[VALUE_PHI3]], [[CALL]]
64; CHECK-NEXT:    br i1 [[CMP]], label %[[EXIT_LOOPEXIT3:.*]], label %[[LOOP]]
65; CHECK:       [[EXIT_LOOPEXIT]]:
66; CHECK-NEXT:    br label %[[EXIT:.*]]
67; CHECK:       [[EXIT_LOOPEXIT3]]:
68; CHECK-NEXT:    br label %[[EXIT]]
69; CHECK:       [[EXIT]]:
70; CHECK-NEXT:    ret void
71;
72top:
73  %load0 = load ptr addrspace(10), ptr %arg, align 8
74  %load1 = load i32, ptr inttoptr (i64 12 to ptr), align 4
75  %sub = sub i32 0, %load1
76  %call = call i64 @julia_steprange_last_4949()
77  %cast0 = addrspacecast ptr addrspace(10) %load0 to ptr addrspace(11)
78  %load2 = load ptr addrspace(10), ptr addrspace(11) %cast0, align 8
79  %cast1 = addrspacecast ptr addrspace(10) %load2 to ptr addrspace(11)
80  %load3 = load ptr addrspace(13), ptr addrspace(11) %cast1, align 8
81  %sext = sext i32 %sub to i64
82  br label %loop
83
84loop:
85  %value_phi3 = phi i64 [ 0, %top ], [ %add0, %loop ]
86  %add0 = add i64 %value_phi3, -1
87  %gep0 = getelementptr inbounds i32, ptr addrspace(13) %load3, i64 %add0
88  %load4 = load i32, ptr addrspace(13) %gep0, align 4
89  %add1 = add i64 %add0, %sext
90  %gep1 = getelementptr inbounds i32, ptr addrspace(13) %load3, i64 %add1
91  store i32 %load4, ptr addrspace(13) %gep1, align 4
92  %cmp = icmp eq i64 %value_phi3, %call
93  br i1 %cmp, label %exit, label %loop
94
95exit:
96  ret void
97}
98;.
99; CHECK: [[META0]] = !{[[META1:![0-9]+]]}
100; CHECK: [[META1]] = distinct !{[[META1]], [[META2:![0-9]+]]}
101; CHECK: [[META2]] = distinct !{[[META2]], !"LVerDomain"}
102; CHECK: [[META3]] = !{[[META4:![0-9]+]]}
103; CHECK: [[META4]] = distinct !{[[META4]], [[META2]]}
104;.
105