Gitiles
Code Review Sign In
gerrit-public.fairphone.software / toolchain / llvm-project / 91be65be656072a68b51a8c4e7bb751ea475d896 / . / llvm / test / CodeGen / X86 / swap.ll
blob: e74ed248c60caeb3d5f4904c9161b849da4e4833 [file] [log] [blame]
Clement Courbetf3da6ab2019-02-05 08:00:17 +0000[diff] [blame]1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=haswell | FileCheck %s -check-prefix=NOAA
3; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=haswell -combiner-global-alias-analysis=1 | FileCheck %s -check-prefix=AA
4
5declare void @llvm.lifetime.start.p0i8(i64, i8* nocapture)
6declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture writeonly, i8* nocapture readonly, i64, i1)
7declare void @llvm.lifetime.end.p0i8(i64, i8* nocapture)
8
9%struct.S = type { [16 x i8] }
10
11define dso_local void @_Z4SwapP1SS0_(%struct.S* nocapture %a, %struct.S* nocapture %b) local_unnamed_addr {
12; NOAA-LABEL: _Z4SwapP1SS0_:
13; NOAA: # %bb.0: # %entry
14; NOAA-NEXT: vmovups (%rdi), %xmm0
15; NOAA-NEXT: vmovaps %xmm0, -{{[0-9]+}}(%rsp)
16; NOAA-NEXT: vmovups (%rsi), %xmm0
17; NOAA-NEXT: vmovups %xmm0, (%rdi)
18; NOAA-NEXT: vmovaps -{{[0-9]+}}(%rsp), %xmm0
19; NOAA-NEXT: vmovups %xmm0, (%rsi)
20; NOAA-NEXT: retq
21;
22; AA-LABEL: _Z4SwapP1SS0_:
23; AA: # %bb.0: # %entry
24; AA-NEXT: vmovups (%rdi), %xmm0
25; AA-NEXT: vmovaps %xmm0, -{{[0-9]+}}(%rsp)
26; AA-NEXT: vmovups (%rsi), %xmm1
27; AA-NEXT: vmovups %xmm1, (%rdi)
28; AA-NEXT: vmovups %xmm0, (%rsi)
29; AA-NEXT: retq
30entry:
31 %tmp.sroa.0 = alloca [16 x i8], align 1
32 %tmp.sroa.0.0..sroa_idx6 = getelementptr inbounds [16 x i8], [16 x i8]* %tmp.sroa.0, i64 0, i64 0
33 call void @llvm.lifetime.start.p0i8(i64 16, i8* nonnull %tmp.sroa.0.0..sroa_idx6)
34 %tmp.sroa.0.0..sroa_idx1 = getelementptr inbounds %struct.S, %struct.S* %a, i64 0, i32 0, i64 0
35 call void @llvm.memcpy.p0i8.p0i8.i64(i8* nonnull align 1 %tmp.sroa.0.0..sroa_idx6, i8* align 1 %tmp.sroa.0.0..sroa_idx1, i64 16, i1 false)
36 %0 = getelementptr inbounds %struct.S, %struct.S* %b, i64 0, i32 0, i64 0
37 tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 1 %tmp.sroa.0.0..sroa_idx1, i8* align 1 %0, i64 16, i1 false), !tbaa.struct !2
38 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 1 %0, i8* nonnull align 1 %tmp.sroa.0.0..sroa_idx6, i64 16, i1 false)
39 call void @llvm.lifetime.end.p0i8(i64 16, i8* nonnull %tmp.sroa.0.0..sroa_idx6)
40 ret void
41}
42
43define dso_local void @onealloc_noreadback(i8* nocapture %a, i8* nocapture %b) local_unnamed_addr {
44; NOAA-LABEL: onealloc_noreadback:
45; NOAA: # %bb.0: # %entry
46; NOAA-NEXT: vmovups (%rdi), %xmm0
47; NOAA-NEXT: vmovaps %xmm0, -{{[0-9]+}}(%rsp)
48; NOAA-NEXT: vmovups (%rsi), %xmm0
49; NOAA-NEXT: vmovaps %xmm0, -{{[0-9]+}}(%rsp)
50; NOAA-NEXT: retq
51;
52; AA-LABEL: onealloc_noreadback:
53; AA: # %bb.0: # %entry
54; AA-NEXT: vmovups (%rdi), %xmm0
55; AA-NEXT: vmovaps %xmm0, -{{[0-9]+}}(%rsp)
56; AA-NEXT: vmovups (%rsi), %xmm0
57; AA-NEXT: vmovaps %xmm0, -{{[0-9]+}}(%rsp)
58; AA-NEXT: retq
59entry:
60 %alloc = alloca [16 x i8], i8 2, align 1
61 %part1 = getelementptr inbounds [16 x i8], [16 x i8]* %alloc, i64 0, i64 0
62 %part2 = getelementptr inbounds [16 x i8], [16 x i8]* %alloc, i64 1, i64 0
63 call void @llvm.lifetime.start.p0i8(i64 16, i8* nonnull %part1)
64 call void @llvm.lifetime.start.p0i8(i64 16, i8* nonnull %part2)
65 call void @llvm.memcpy.p0i8.p0i8.i64(i8* nonnull align 1 %part1, i8* align 1 %a, i64 16, i1 false)
66 tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* nonnull align 1 %part2, i8* align 1 %b, i64 16, i1 false)
67 call void @llvm.lifetime.end.p0i8(i64 16, i8* nonnull %part1)
68 call void @llvm.lifetime.end.p0i8(i64 16, i8* nonnull %part2)
69 ret void
70}
71
72
73define dso_local void @twoallocs_trivial(i8* nocapture %a, i8* nocapture %b) local_unnamed_addr {
74; NOAA-LABEL: twoallocs_trivial:
75; NOAA: # %bb.0: # %entry
76; NOAA-NEXT: vmovups (%rdi), %xmm0
77; NOAA-NEXT: vmovaps %xmm0, -{{[0-9]+}}(%rsp)
78; NOAA-NEXT: vmovups (%rsi), %xmm0
79; NOAA-NEXT: vmovaps %xmm0, -{{[0-9]+}}(%rsp)
80; NOAA-NEXT: retq
81;
82; AA-LABEL: twoallocs_trivial:
83; AA: # %bb.0: # %entry
84; AA-NEXT: vmovups (%rdi), %xmm0
85; AA-NEXT: vmovaps %xmm0, -{{[0-9]+}}(%rsp)
86; AA-NEXT: vmovups (%rsi), %xmm0
87; AA-NEXT: vmovaps %xmm0, -{{[0-9]+}}(%rsp)
88; AA-NEXT: retq
89entry:
90 %alloc1 = alloca [16 x i8], align 1
91 %alloc2 = alloca [16 x i8], align 1
92 %part1 = getelementptr inbounds [16 x i8], [16 x i8]* %alloc1, i64 0, i64 0
93 %part2 = getelementptr inbounds [16 x i8], [16 x i8]* %alloc2, i64 0, i64 0
94 call void @llvm.lifetime.start.p0i8(i64 16, i8* nonnull %part1)
95 call void @llvm.lifetime.start.p0i8(i64 16, i8* nonnull %part2)
96 call void @llvm.memcpy.p0i8.p0i8.i64(i8* nonnull align 1 %part1, i8* align 1 %a, i64 16, i1 false)
97 call void @llvm.memcpy.p0i8.p0i8.i64(i8* nonnull align 1 %part2, i8* align 1 %b, i64 16, i1 false)
98 call void @llvm.lifetime.end.p0i8(i64 16, i8* nonnull %part1)
99 call void @llvm.lifetime.end.p0i8(i64 16, i8* nonnull %part2)
100 ret void
101}
102
103define dso_local void @twoallocs(i8* nocapture %a, i8* nocapture %b) local_unnamed_addr {
104; NOAA-LABEL: twoallocs:
105; NOAA: # %bb.0: # %entry
106; NOAA-NEXT: vmovups (%rdi), %xmm0
107; NOAA-NEXT: vmovaps %xmm0, -{{[0-9]+}}(%rsp)
108; NOAA-NEXT: vmovaps -{{[0-9]+}}(%rsp), %xmm0
109; NOAA-NEXT: vmovups %xmm0, (%rsi)
110; NOAA-NEXT: retq
111;
112; AA-LABEL: twoallocs:
113; AA: # %bb.0: # %entry
114; AA-NEXT: vmovups (%rdi), %xmm0
115; AA-NEXT: vmovaps %xmm0, -{{[0-9]+}}(%rsp)
116; AA-NEXT: vmovaps -{{[0-9]+}}(%rsp), %xmm0
117; AA-NEXT: vmovups %xmm0, (%rsi)
118; AA-NEXT: retq
119entry:
120 %alloc1 = alloca [16 x i8], align 1
121 %alloc2 = alloca [16 x i8], align 1
122 %part1 = getelementptr inbounds [16 x i8], [16 x i8]* %alloc1, i64 0, i64 0
123 %part2 = getelementptr inbounds [16 x i8], [16 x i8]* %alloc2, i64 0, i64 0
124 %part2_alias = getelementptr inbounds [16 x i8], [16 x i8]* %alloc2, i64 0, i64 0
125 call void @llvm.lifetime.start.p0i8(i64 16, i8* nonnull %part1)
126 call void @llvm.lifetime.start.p0i8(i64 16, i8* nonnull %part2)
127 call void @llvm.memcpy.p0i8.p0i8.i64(i8* nonnull align 1 %part2, i8* align 1 %a, i64 16, i1 false)
128 call void @llvm.lifetime.end.p0i8(i64 16, i8* nonnull %part1)
129 tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* nonnull align 1 %b, i8* align 1 %part2_alias, i64 16, i1 false)
130 call void @llvm.lifetime.end.p0i8(i64 16, i8* nonnull %part2)
131 ret void
132}
133
134define dso_local void @onealloc_readback_1(i8* nocapture %a, i8* nocapture %b) local_unnamed_addr {
135; NOAA-LABEL: onealloc_readback_1:
136; NOAA: # %bb.0: # %entry
137; NOAA-NEXT: vmovups (%rdi), %xmm0
138; NOAA-NEXT: vmovaps %xmm0, -{{[0-9]+}}(%rsp)
139; NOAA-NEXT: vmovups (%rsi), %xmm0
140; NOAA-NEXT: vmovaps %xmm0, -{{[0-9]+}}(%rsp)
141; NOAA-NEXT: vmovaps -{{[0-9]+}}(%rsp), %xmm0
142; NOAA-NEXT: vmovups %xmm0, (%rdi)
143; NOAA-NEXT: retq
144;
145; AA-LABEL: onealloc_readback_1:
146; AA: # %bb.0: # %entry
147; AA-NEXT: vmovups (%rdi), %xmm0
148; AA-NEXT: vmovaps %xmm0, -{{[0-9]+}}(%rsp)
149; AA-NEXT: vmovups (%rsi), %xmm0
150; AA-NEXT: vmovaps %xmm0, -{{[0-9]+}}(%rsp)
151; AA-NEXT: vmovaps -{{[0-9]+}}(%rsp), %xmm0
152; AA-NEXT: vmovups %xmm0, (%rdi)
153; AA-NEXT: retq
154entry:
155 %alloc = alloca [16 x i8], i8 2, align 1
156 %part2 = getelementptr inbounds [16 x i8], [16 x i8]* %alloc, i64 0, i64 0
157 %part1 = getelementptr inbounds [16 x i8], [16 x i8]* %alloc, i64 1, i64 0
158 call void @llvm.lifetime.start.p0i8(i64 16, i8* nonnull %part1)
159 call void @llvm.lifetime.start.p0i8(i64 16, i8* nonnull %part2)
160 call void @llvm.memcpy.p0i8.p0i8.i64(i8* nonnull align 1 %part1, i8* align 1 %a, i64 16, i1 false)
161 call void @llvm.memcpy.p0i8.p0i8.i64(i8* nonnull align 1 %part2, i8* align 1 %b, i64 16, i1 false)
162 call void @llvm.lifetime.end.p0i8(i64 16, i8* nonnull %part1)
163 tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* nonnull align 1 %a, i8* align 1 %part2, i64 16, i1 false)
164 call void @llvm.lifetime.end.p0i8(i64 16, i8* nonnull %part2)
165 ret void
166}
167
168define dso_local void @onealloc_readback_2(i8* nocapture %a, i8* nocapture %b) local_unnamed_addr {
169; NOAA-LABEL: onealloc_readback_2:
170; NOAA: # %bb.0: # %entry
171; NOAA-NEXT: vmovups (%rdi), %xmm0
172; NOAA-NEXT: vmovaps %xmm0, -{{[0-9]+}}(%rsp)
173; NOAA-NEXT: vmovups (%rsi), %xmm0
174; NOAA-NEXT: vmovaps %xmm0, -{{[0-9]+}}(%rsp)
175; NOAA-NEXT: vmovaps -{{[0-9]+}}(%rsp), %xmm0
176; NOAA-NEXT: vmovups %xmm0, (%rdi)
177; NOAA-NEXT: retq
178;
179; AA-LABEL: onealloc_readback_2:
180; AA: # %bb.0: # %entry
181; AA-NEXT: vmovups (%rdi), %xmm0
182; AA-NEXT: vmovaps %xmm0, -{{[0-9]+}}(%rsp)
183; AA-NEXT: vmovups (%rsi), %xmm0
184; AA-NEXT: vmovaps %xmm0, -{{[0-9]+}}(%rsp)
185; AA-NEXT: vmovaps -{{[0-9]+}}(%rsp), %xmm0
186; AA-NEXT: vmovups %xmm0, (%rdi)
187; AA-NEXT: retq
188entry:
189 %alloc = alloca [16 x i8], i8 2, align 1
190 %part1 = getelementptr inbounds [16 x i8], [16 x i8]* %alloc, i64 0, i64 0
191 %part2 = getelementptr inbounds [16 x i8], [16 x i8]* %alloc, i64 1, i64 0
192 call void @llvm.lifetime.start.p0i8(i64 16, i8* nonnull %part1)
193 call void @llvm.lifetime.start.p0i8(i64 16, i8* nonnull %part2)
194 call void @llvm.memcpy.p0i8.p0i8.i64(i8* nonnull align 1 %part1, i8* align 1 %a, i64 16, i1 false)
195 call void @llvm.memcpy.p0i8.p0i8.i64(i8* nonnull align 1 %part2, i8* align 1 %b, i64 16, i1 false)
196 call void @llvm.lifetime.end.p0i8(i64 16, i8* nonnull %part1)
197 tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* nonnull align 1 %a, i8* align 1 %part2, i64 16, i1 false)
198 call void @llvm.lifetime.end.p0i8(i64 16, i8* nonnull %part2)
199 ret void
200}
201
202
203!llvm.module.flags = !{!0}
204!llvm.ident = !{!1}
205
206!0 = !{i32 1, !"wchar_size", i32 4}
207!1 = !{!"clang version 9.0.0 (trunk 352631) (llvm/trunk 352632)"}
208!2 = !{i64 0, i64 16, !3}
209!3 = !{!4, !4, i64 0}
210!4 = !{!"omnipotent char", !5, i64 0}
211!5 = !{!"Simple C++ TBAA"}