1; RUN: opt < %s -basicaa -dse -S | FileCheck %s
2target datalayout = "E-p:64:64:64-a0:0:8-f32:32:32-f64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-v64:64:64-v128:128:128"
3
4declare void @llvm.memset.p0i8.i64(i8* nocapture, i8, i64, i32, i1) nounwind
5declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture, i64, i32, i1) nounwind
6declare i8* @llvm.init.trampoline(i8*, i8*, i8*)
7
8define void @test1(i32* %Q, i32* %P) {
9        %DEAD = load i32* %Q
10        store i32 %DEAD, i32* %P
11        store i32 0, i32* %P
12        ret void
13; CHECK: @test1
14; CHECK-NEXT: store i32 0, i32* %P
15; CHECK-NEXT: ret void
16}
17
18; PR8576 - Should delete store of 10 even though p/q are may aliases.
19define void @test2(i32 *%p, i32 *%q) {
20  store i32 10, i32* %p, align 4
21  store i32 20, i32* %q, align 4
22  store i32 30, i32* %p, align 4
23  ret void
24; CHECK: @test2
25; CHECK-NEXT: store i32 20
26}
27
28
29; PR8677
30@g = global i32 1
31
32define i32 @test3(i32* %g_addr) nounwind {
33; CHECK: @test3
34; CHECK: load i32* %g_addr
35  %g_value = load i32* %g_addr, align 4
36  store i32 -1, i32* @g, align 4
37  store i32 %g_value, i32* %g_addr, align 4
38  %tmp3 = load i32* @g, align 4
39  ret i32 %tmp3
40}
41
42
43define void @test4(i32* %Q) {
44        %a = load i32* %Q
45        store volatile i32 %a, i32* %Q
46        ret void
47; CHECK: @test4
48; CHECK-NEXT: load i32
49; CHECK-NEXT: store volatile
50; CHECK-NEXT: ret void
51}
52
53define void @test5(i32* %Q) {
54        %a = load volatile i32* %Q
55        store i32 %a, i32* %Q
56        ret void
57; CHECK: @test5
58; CHECK-NEXT: load volatile
59; CHECK-NEXT: ret void
60}
61
62; Should delete store of 10 even though memset is a may-store to P (P and Q may
63; alias).
64define void @test6(i32 *%p, i8 *%q) {
65  store i32 10, i32* %p, align 4       ;; dead.
66  call void @llvm.memset.p0i8.i64(i8* %q, i8 42, i64 900, i32 1, i1 false)
67  store i32 30, i32* %p, align 4
68  ret void
69; CHECK: @test6
70; CHECK-NEXT: call void @llvm.memset
71}
72
73; Should delete store of 10 even though memcpy is a may-store to P (P and Q may
74; alias).
75define void @test7(i32 *%p, i8 *%q, i8* noalias %r) {
76  store i32 10, i32* %p, align 4       ;; dead.
77  call void @llvm.memcpy.p0i8.p0i8.i64(i8* %q, i8* %r, i64 900, i32 1, i1 false)
78  store i32 30, i32* %p, align 4
79  ret void
80; CHECK: @test7
81; CHECK-NEXT: call void @llvm.memcpy
82}
83
84; Do not delete stores that are only partially killed.
85define i32 @test8() {
86        %V = alloca i32
87        store i32 1234567, i32* %V
88        %V2 = bitcast i32* %V to i8*
89        store i8 0, i8* %V2
90        %X = load i32* %V
91        ret i32 %X
92
93; CHECK: @test8
94; CHECK: store i32 1234567
95}
96
97
98; Test for byval handling.
99%struct.x = type { i32, i32, i32, i32 }
100define void @test9(%struct.x* byval  %a) nounwind  {
101	%tmp2 = getelementptr %struct.x* %a, i32 0, i32 0
102	store i32 1, i32* %tmp2, align 4
103	ret void
104; CHECK: @test9
105; CHECK-NEXT: ret void
106}
107
108; va_arg has fuzzy dependence, the store shouldn't be zapped.
109define double @test10(i8* %X) {
110        %X_addr = alloca i8*
111        store i8* %X, i8** %X_addr
112        %tmp.0 = va_arg i8** %X_addr, double
113        ret double %tmp.0
114; CHECK: @test10
115; CHECK: store
116}
117
118
119; DSE should delete the dead trampoline.
120declare void @test11f()
121define void @test11() {
122; CHECK: @test11
123	%storage = alloca [10 x i8], align 16		; <[10 x i8]*> [#uses=1]
124; CHECK-NOT: alloca
125	%cast = getelementptr [10 x i8]* %storage, i32 0, i32 0		; <i8*> [#uses=1]
126	%tramp = call i8* @llvm.init.trampoline( i8* %cast, i8* bitcast (void ()* @test11f to i8*), i8* null )		; <i8*> [#uses=1]
127; CHECK-NOT: trampoline
128	ret void
129; CHECK: ret void
130}
131
132
133; PR2599 - load -> store to same address.
134define void @test12({ i32, i32 }* %x) nounwind  {
135	%tmp4 = getelementptr { i32, i32 }* %x, i32 0, i32 0
136	%tmp5 = load i32* %tmp4, align 4
137	%tmp7 = getelementptr { i32, i32 }* %x, i32 0, i32 1
138	%tmp8 = load i32* %tmp7, align 4
139	%tmp17 = sub i32 0, %tmp8
140	store i32 %tmp5, i32* %tmp4, align 4
141	store i32 %tmp17, i32* %tmp7, align 4
142	ret void
143; CHECK: @test12
144; CHECK-NOT: tmp5
145; CHECK: ret void
146}
147
148
149; %P doesn't escape, the DEAD instructions should be removed.
150declare void @test13f()
151define i32* @test13() {
152        %p = tail call i8* @malloc(i32 4)
153        %P = bitcast i8* %p to i32*
154        %DEAD = load i32* %P
155        %DEAD2 = add i32 %DEAD, 1
156        store i32 %DEAD2, i32* %P
157        call void @test13f( )
158        store i32 0, i32* %P
159        ret i32* %P
160; CHECK: @test13()
161; CHECK-NEXT: malloc
162; CHECK-NEXT: bitcast
163; CHECK-NEXT: call void
164}
165
166declare noalias i8* @malloc(i32)
167
168
169
170define void @test14(i32* %Q) {
171        %P = alloca i32
172        %DEAD = load i32* %Q
173        store i32 %DEAD, i32* %P
174        ret void
175
176; CHECK: @test14
177; CHECK-NEXT: ret void
178}
179
180
181; PR8701
182
183;; Fully dead overwrite of memcpy.
184define void @test15(i8* %P, i8* %Q) nounwind ssp {
185  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
186  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
187  ret void
188; CHECK: @test15
189; CHECK-NEXT: call void @llvm.memcpy
190; CHECK-NEXT: ret
191}
192
193;; Full overwrite of smaller memcpy.
194define void @test16(i8* %P, i8* %Q) nounwind ssp {
195  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 8, i32 1, i1 false)
196  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
197  ret void
198; CHECK: @test16
199; CHECK-NEXT: call void @llvm.memcpy
200; CHECK-NEXT: ret
201}
202
203;; Overwrite of memset by memcpy.
204define void @test17(i8* %P, i8* noalias %Q) nounwind ssp {
205  tail call void @llvm.memset.p0i8.i64(i8* %P, i8 42, i64 8, i32 1, i1 false)
206  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
207  ret void
208; CHECK: @test17
209; CHECK-NEXT: call void @llvm.memcpy
210; CHECK-NEXT: ret
211}
212
213; Should not delete the volatile memset.
214define void @test17v(i8* %P, i8* %Q) nounwind ssp {
215  tail call void @llvm.memset.p0i8.i64(i8* %P, i8 42, i64 8, i32 1, i1 true)
216  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
217  ret void
218; CHECK: @test17v
219; CHECK-NEXT: call void @llvm.memset
220; CHECK-NEXT: call void @llvm.memcpy
221; CHECK-NEXT: ret
222}
223
224; PR8728
225; Do not delete instruction where possible situation is:
226; A = B
227; A = A
228define void @test18(i8* %P, i8* %Q, i8* %R) nounwind ssp {
229  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
230  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %R, i64 12, i32 1, i1 false)
231  ret void
232; CHECK: @test18
233; CHECK-NEXT: call void @llvm.memcpy
234; CHECK-NEXT: call void @llvm.memcpy
235; CHECK-NEXT: ret
236}
237
238
239; The store here is not dead because the byval call reads it.
240declare void @test19f({i32}* byval align 4 %P)
241
242define void @test19({i32} * nocapture byval align 4 %arg5) nounwind ssp {
243bb:
244  %tmp7 = getelementptr inbounds {i32}* %arg5, i32 0, i32 0
245  store i32 912, i32* %tmp7
246  call void @test19f({i32}* byval align 4 %arg5)
247  ret void
248
249; CHECK: @test19(
250; CHECK: store i32 912
251; CHECK: call void @test19f
252}
253
254