Gitiles
Code Review Sign In
gerrit-public.fairphone.software / toolchain / llvm-project / a69bbe02a2352271e8b14542073f177e24c499c1 / . / llvm / test / Transforms / LoopIdiom / bcmp-widening.ll
blob: 4c8b7f3f030399d5e3c136b58e4e4fec774eb30d [file] [log] [blame]
Roman Lebedev445c22b2019-04-25 08:33:47 +0000[diff] [blame]1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
Roman Lebedev76cdcf22019-10-12 15:35:32 +0000[diff] [blame]2; RUN: opt -loop-idiom -verify -verify-each -verify-dom-info -verify-loop-info < %s -S | FileCheck %s
Roman Lebedev445c22b2019-04-25 08:33:47 +0000[diff] [blame]3
4target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
5
6; We do not necessarily have a loop with comparison of two loaded values.
7; The loop may already be doing bcmp() itself. We just need to widen it.
8
9; FIXME: -memcpyopt does not promote memcmp() source into memcmp() / bcmp().
10
11; #include <algorithm>
12;
13; // FIXME
14; bool widening_of_bcmp(char const* ptr0, char const* ptr1, size_t count) {
15; static constexpr auto Size = 32, Step = 32;
16; for(size_t i = 0; i < count; i++, ptr0 += Step, ptr1 += Step) {
17; if(bcmp(ptr0, ptr1, Size) != 0)
18; return false;
19; }
20; return true;
21; }
22;
23; // FIXME
24; bool widening_of_overlapping_bcmp(char const* ptr0, char const* ptr1, size_t count) {
25; static constexpr auto Size = 32, Step = 16;
26; for(size_t i = 0; i < count; i++, ptr0 += Step, ptr1 += Step) {
27; if(bcmp(ptr0, ptr1, Size) != 0)
28; return false;
29; }
30; return true;
31; }
32
33declare i32 @bcmp(i8*, i8*, i64)
34
35; FIXME
36define i1 @_Z18widening_of_bcmpPKcS0_m(i8* %ptr0, i8* %ptr1, i64 %count) {
37; CHECK-LABEL: @_Z18widening_of_bcmpPKcS0_m(
38; CHECK-NEXT: entry:
39; CHECK-NEXT: [[CMP7:%.*]] = icmp eq i64 [[COUNT:%.*]], 0
40; CHECK-NEXT: br i1 [[CMP7]], label [[CLEANUP:%.*]], label [[FOR_BODY_PREHEADER:%.*]]
41; CHECK: for.body.preheader:
42; CHECK-NEXT: br label [[FOR_BODY:%.*]]
43; CHECK: for.body:
44; CHECK-NEXT: [[I_010:%.*]] = phi i64 [ [[INC:%.*]], [[FOR_INC:%.*]] ], [ 0, [[FOR_BODY_PREHEADER]] ]
45; CHECK-NEXT: [[PTR1_ADDR_09:%.*]] = phi i8* [ [[ADD_PTR2:%.*]], [[FOR_INC]] ], [ [[PTR1:%.*]], [[FOR_BODY_PREHEADER]] ]
46; CHECK-NEXT: [[PTR0_ADDR_08:%.*]] = phi i8* [ [[ADD_PTR:%.*]], [[FOR_INC]] ], [ [[PTR0:%.*]], [[FOR_BODY_PREHEADER]] ]
47; CHECK-NEXT: [[BCMP:%.*]] = tail call i32 @bcmp(i8* [[PTR0_ADDR_08]], i8* [[PTR1_ADDR_09]], i64 32)
48; CHECK-NEXT: [[CMP1:%.*]] = icmp eq i32 [[BCMP]], 0
49; CHECK-NEXT: br i1 [[CMP1]], label [[FOR_INC]], label [[CLEANUP_LOOPEXIT:%.*]]
50; CHECK: for.inc:
51; CHECK-NEXT: [[INC]] = add nuw i64 [[I_010]], 1
52; CHECK-NEXT: [[ADD_PTR]] = getelementptr inbounds i8, i8* [[PTR0_ADDR_08]], i64 32
53; CHECK-NEXT: [[ADD_PTR2]] = getelementptr inbounds i8, i8* [[PTR1_ADDR_09]], i64 32
54; CHECK-NEXT: [[CMP:%.*]] = icmp ult i64 [[INC]], [[COUNT]]
55; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label [[CLEANUP_LOOPEXIT]]
56; CHECK: cleanup.loopexit:
57; CHECK-NEXT: [[RES_PH:%.*]] = phi i1 [ false, [[FOR_BODY]] ], [ true, [[FOR_INC]] ]
58; CHECK-NEXT: br label [[CLEANUP]]
59; CHECK: cleanup:
60; CHECK-NEXT: [[RES:%.*]] = phi i1 [ true, [[ENTRY:%.*]] ], [ [[RES_PH]], [[CLEANUP_LOOPEXIT]] ]
61; CHECK-NEXT: ret i1 [[RES]]
62;
63entry:
64 %cmp7 = icmp eq i64 %count, 0
65 br i1 %cmp7, label %cleanup, label %for.body
66
67for.body: ; preds = %entry, %for.inc
68 %i.010 = phi i64 [ %inc, %for.inc ], [ 0, %entry ]
69 %ptr1.addr.09 = phi i8* [ %add.ptr2, %for.inc ], [ %ptr1, %entry ]
70 %ptr0.addr.08 = phi i8* [ %add.ptr, %for.inc ], [ %ptr0, %entry ]
71 %bcmp = tail call i32 @bcmp(i8* %ptr0.addr.08, i8* %ptr1.addr.09, i64 32)
72 %cmp1 = icmp eq i32 %bcmp, 0
73 br i1 %cmp1, label %for.inc, label %cleanup
74
75for.inc: ; preds = %for.body
76 %inc = add nuw i64 %i.010, 1
77 %add.ptr = getelementptr inbounds i8, i8* %ptr0.addr.08, i64 32
78 %add.ptr2 = getelementptr inbounds i8, i8* %ptr1.addr.09, i64 32
79 %cmp = icmp ult i64 %inc, %count
80 br i1 %cmp, label %for.body, label %cleanup
81
82cleanup: ; preds = %for.body, %for.inc, %entry
83 %res = phi i1 [ true, %entry ], [ true, %for.inc ], [ false, %for.body ]
84 ret i1 %res
85}
86
87; FIXME
88define i1 @_Z30widening_of_overlapping_bcmpPKcS0_m(i8* %ptr0, i8* %ptr1, i64 %count) {
89; CHECK-LABEL: @_Z30widening_of_overlapping_bcmpPKcS0_m(
90; CHECK-NEXT: entry:
91; CHECK-NEXT: [[CMP7:%.*]] = icmp eq i64 [[COUNT:%.*]], 0
92; CHECK-NEXT: br i1 [[CMP7]], label [[CLEANUP:%.*]], label [[FOR_BODY_PREHEADER:%.*]]
93; CHECK: for.body.preheader:
94; CHECK-NEXT: br label [[FOR_BODY:%.*]]
95; CHECK: for.body:
96; CHECK-NEXT: [[I_010:%.*]] = phi i64 [ [[INC:%.*]], [[FOR_INC:%.*]] ], [ 0, [[FOR_BODY_PREHEADER]] ]
97; CHECK-NEXT: [[PTR1_ADDR_09:%.*]] = phi i8* [ [[ADD_PTR2:%.*]], [[FOR_INC]] ], [ [[PTR1:%.*]], [[FOR_BODY_PREHEADER]] ]
98; CHECK-NEXT: [[PTR0_ADDR_08:%.*]] = phi i8* [ [[ADD_PTR:%.*]], [[FOR_INC]] ], [ [[PTR0:%.*]], [[FOR_BODY_PREHEADER]] ]
99; CHECK-NEXT: [[BCMP:%.*]] = tail call i32 @bcmp(i8* [[PTR0_ADDR_08]], i8* [[PTR1_ADDR_09]], i64 32)
100; CHECK-NEXT: [[CMP1:%.*]] = icmp eq i32 [[BCMP]], 0
101; CHECK-NEXT: br i1 [[CMP1]], label [[FOR_INC]], label [[CLEANUP_LOOPEXIT:%.*]]
102; CHECK: for.inc:
103; CHECK-NEXT: [[INC]] = add nuw i64 [[I_010]], 1
104; CHECK-NEXT: [[ADD_PTR]] = getelementptr inbounds i8, i8* [[PTR0_ADDR_08]], i64 16
105; CHECK-NEXT: [[ADD_PTR2]] = getelementptr inbounds i8, i8* [[PTR1_ADDR_09]], i64 16
106; CHECK-NEXT: [[CMP:%.*]] = icmp ult i64 [[INC]], [[COUNT]]
107; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label [[CLEANUP_LOOPEXIT]]
108; CHECK: cleanup.loopexit:
109; CHECK-NEXT: [[RES_PH:%.*]] = phi i1 [ false, [[FOR_BODY]] ], [ true, [[FOR_INC]] ]
110; CHECK-NEXT: br label [[CLEANUP]]
111; CHECK: cleanup:
112; CHECK-NEXT: [[RES:%.*]] = phi i1 [ true, [[ENTRY:%.*]] ], [ [[RES_PH]], [[CLEANUP_LOOPEXIT]] ]
113; CHECK-NEXT: ret i1 [[RES]]
114;
115entry:
116 %cmp7 = icmp eq i64 %count, 0
117 br i1 %cmp7, label %cleanup, label %for.body
118
119for.body: ; preds = %entry, %for.inc
120 %i.010 = phi i64 [ %inc, %for.inc ], [ 0, %entry ]
121 %ptr1.addr.09 = phi i8* [ %add.ptr2, %for.inc ], [ %ptr1, %entry ]
122 %ptr0.addr.08 = phi i8* [ %add.ptr, %for.inc ], [ %ptr0, %entry ]
123 %bcmp = tail call i32 @bcmp(i8* %ptr0.addr.08, i8* %ptr1.addr.09, i64 32)
124 %cmp1 = icmp eq i32 %bcmp, 0
125 br i1 %cmp1, label %for.inc, label %cleanup
126
127for.inc: ; preds = %for.body
128 %inc = add nuw i64 %i.010, 1
129 %add.ptr = getelementptr inbounds i8, i8* %ptr0.addr.08, i64 16
130 %add.ptr2 = getelementptr inbounds i8, i8* %ptr1.addr.09, i64 16
131 %cmp = icmp ult i64 %inc, %count
132 br i1 %cmp, label %for.body, label %cleanup
133
134cleanup: ; preds = %for.body, %for.inc, %entry
135 %res = phi i1 [ true, %entry ], [ true, %for.inc ], [ false, %for.body ]
136 ret i1 %res
137}