xref: /llvm-project/llvm/test/Transforms/MergeICmps/X86/multiple-blocks-does-work.ll (revision 9a3e39996320793ec5fe5c6bb4873f82cbc1bbf3)
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -passes=mergeicmps -verify-dom-info -mtriple=x86_64-unknown-unknown -S | FileCheck %s --check-prefix=X86
3
4%S = type { i32, i32, i32, i32 }
5
6declare void @foo(...)
7
8; We can discard %entry and %land.rhs.i, but still merge the last 2 blocks.
9define zeroext i1 @opeq1(
10; X86-LABEL: @opeq1(
11; X86-NEXT:  entry:
12; X86-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A:%.*]], align 4
13; X86-NEXT:    [[TMP1:%.*]] = load i32, ptr [[B:%.*]], align 4
14; X86-NEXT:    call void (...) @foo()
15; X86-NEXT:    [[CMP_I:%.*]] = icmp eq i32 [[TMP0]], [[TMP1]]
16; X86-NEXT:    br i1 [[CMP_I]], label [[LAND_RHS_I:%.*]], label [[OPEQ1_EXIT:%.*]]
17; X86:       land.rhs.i:
18; X86-NEXT:    [[SECOND_I:%.*]] = getelementptr inbounds [[S:%.*]], ptr [[A]], i64 0, i32 1
19; X86-NEXT:    [[TMP2:%.*]] = load i32, ptr [[SECOND_I]], align 4
20; X86-NEXT:    [[SECOND2_I:%.*]] = getelementptr inbounds [[S]], ptr [[B]], i64 0, i32 1
21; X86-NEXT:    [[TMP3:%.*]] = load i32, ptr [[SECOND2_I]], align 4
22; X86-NEXT:    call void (...) @foo()
23; X86-NEXT:    [[CMP2_I:%.*]] = icmp eq i32 [[TMP2]], [[TMP3]]
24; X86-NEXT:    br i1 [[CMP2_I]], label %"land.rhs.i.2+land.rhs.i.3", label [[OPEQ1_EXIT]]
25; X86:       "land.rhs.i.2+land.rhs.i.3":
26; X86-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [[S]], ptr [[A]], i64 0, i32 2
27; X86-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [[S]], ptr [[B]], i64 0, i32 2
28; X86-NEXT:    [[MEMCMP:%.*]] = call i32 @memcmp(ptr [[TMP4]], ptr [[TMP5]], i64 8)
29; X86-NEXT:    [[TMP6:%.*]] = icmp eq i32 [[MEMCMP]], 0
30; X86-NEXT:    br label [[OPEQ1_EXIT]]
31; X86:       opeq1.exit:
32; X86-NEXT:    [[TMP7:%.*]] = phi i1 [ false, [[ENTRY:%.*]] ], [ false, [[LAND_RHS_I]] ], [ [[TMP6]], %"land.rhs.i.2+land.rhs.i.3" ]
33; X86-NEXT:    ret i1 [[TMP7]]
34;
35  ptr nocapture readonly dereferenceable(16) %a,
36  ptr nocapture readonly dereferenceable(16) %b) local_unnamed_addr nofree nosync {
37entry:
38  %0 = load i32, ptr %a, align 4
39  %1 = load i32, ptr %b, align 4
40  ; Does other work.
41  call void (...) @foo()
42  %cmp.i = icmp eq i32 %0, %1
43  br i1 %cmp.i, label %land.rhs.i, label %opeq1.exit
44
45land.rhs.i:
46  %second.i = getelementptr inbounds %S, ptr %a, i64 0, i32 1
47  %2 = load i32, ptr %second.i, align 4
48  %second2.i = getelementptr inbounds %S, ptr %b, i64 0, i32 1
49  %3 = load i32, ptr %second2.i, align 4
50  ; Does other work.
51  call void (...) @foo()
52  %cmp2.i = icmp eq i32 %2, %3
53  br i1 %cmp2.i, label %land.rhs.i.2, label %opeq1.exit
54
55land.rhs.i.2:
56  %third.i = getelementptr inbounds %S, ptr %a, i64 0, i32 2
57  %4 = load i32, ptr %third.i, align 4
58  %third2.i = getelementptr inbounds %S, ptr %b, i64 0, i32 2
59  %5 = load i32, ptr %third2.i, align 4
60  %cmp3.i = icmp eq i32 %4, %5
61  br i1 %cmp3.i, label %land.rhs.i.3, label %opeq1.exit
62
63land.rhs.i.3:
64  %fourth.i = getelementptr inbounds %S, ptr %a, i64 0, i32 3
65  %6 = load i32, ptr %fourth.i, align 4
66  %fourth2.i = getelementptr inbounds %S, ptr %b, i64 0, i32 3
67  %7 = load i32, ptr %fourth2.i, align 4
68  %cmp4.i = icmp eq i32 %6, %7
69  br label %opeq1.exit
70
71opeq1.exit:
72  %8 = phi i1 [ false, %entry ], [ false, %land.rhs.i] , [ false, %land.rhs.i.2 ], [ %cmp4.i, %land.rhs.i.3 ]
73  ret i1 %8
74}
75