1 ; RUN: opt < %s -sroa -S | FileCheck %s
2 ; RUN: opt < %s -sroa -force-ssa-updater -S | FileCheck %s
4 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-n8:16:32:64"
6 declare void @llvm.lifetime.start(i64, i8* nocapture)
7 declare void @llvm.lifetime.end(i64, i8* nocapture)
18 %a1.i8 = bitcast i32* %a1 to i8*
19 call void @llvm.lifetime.start(i64 4, i8* %a1.i8)
24 call void @llvm.lifetime.end(i64 4, i8* %a1.i8)
26 %a2.i8 = bitcast float* %a2 to i8*
27 call void @llvm.lifetime.start(i64 4, i8* %a2.i8)
29 store float 0.0, float* %a2
30 %v2 = load float * %a2
31 %v2.int = bitcast float %v2 to i32
32 %sum1 = add i32 %v1, %v2.int
34 call void @llvm.lifetime.end(i64 4, i8* %a2.i8)
45 %X = alloca { i32, float }
46 %Y = getelementptr { i32, float }* %X, i64 0, i32 0
52 define i64 @test2(i64 %X) {
59 %B = bitcast [8 x i8]* %A to i64*
68 define void @test3(i8* %dst, i8* %src) {
72 %a = alloca [300 x i8]
74 ; CHECK: %[[test3_a1:.*]] = alloca [42 x i8]
75 ; CHECK-NEXT: %[[test3_a2:.*]] = alloca [99 x i8]
76 ; CHECK-NEXT: %[[test3_a3:.*]] = alloca [16 x i8]
77 ; CHECK-NEXT: %[[test3_a4:.*]] = alloca [42 x i8]
78 ; CHECK-NEXT: %[[test3_a5:.*]] = alloca [7 x i8]
79 ; CHECK-NEXT: %[[test3_a6:.*]] = alloca [7 x i8]
80 ; CHECK-NEXT: %[[test3_a7:.*]] = alloca [85 x i8]
82 %b = getelementptr [300 x i8]* %a, i64 0, i64 0
83 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %b, i8* %src, i32 300, i32 1, i1 false)
84 ; CHECK-NEXT: %[[gep_dst:.*]] = getelementptr inbounds [42 x i8]* %[[test3_a1]], i64 0, i64 0
85 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep_dst]], i8* %src, i32 42
86 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds i8* %src, i64 42
87 ; CHECK-NEXT: %[[test3_r1:.*]] = load i8* %[[gep]]
88 ; CHECK-NEXT: %[[gep_src:.*]] = getelementptr inbounds i8* %src, i64 43
89 ; CHECK-NEXT: %[[gep_dst:.*]] = getelementptr inbounds [99 x i8]* %[[test3_a2]], i64 0, i64 0
90 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep_dst]], i8* %[[gep_src]], i32 99
91 ; CHECK-NEXT: %[[gep_src:.*]] = getelementptr inbounds i8* %src, i64 142
92 ; CHECK-NEXT: %[[gep_dst:.*]] = getelementptr inbounds [16 x i8]* %[[test3_a3]], i64 0, i64 0
93 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep_dst]], i8* %[[gep_src]], i32 16
94 ; CHECK-NEXT: %[[gep_src:.*]] = getelementptr inbounds i8* %src, i64 158
95 ; CHECK-NEXT: %[[gep_dst:.*]] = getelementptr inbounds [42 x i8]* %[[test3_a4]], i64 0, i64 0
96 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep_dst]], i8* %[[gep_src]], i32 42
97 ; CHECK-NEXT: %[[gep_src:.*]] = getelementptr inbounds i8* %src, i64 200
98 ; CHECK-NEXT: %[[gep_dst:.*]] = getelementptr inbounds [7 x i8]* %[[test3_a5]], i64 0, i64 0
99 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep_dst]], i8* %[[gep_src]], i32 7
100 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds i8* %src, i64 207
101 ; CHECK-NEXT: %[[test3_r2:.*]] = load i8* %[[gep]]
102 ; CHECK-NEXT: %[[gep_src:.*]] = getelementptr inbounds i8* %src, i64 208
103 ; CHECK-NEXT: %[[gep_dst:.*]] = getelementptr inbounds [7 x i8]* %[[test3_a6]], i64 0, i64 0
104 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep_dst]], i8* %[[gep_src]], i32 7
105 ; CHECK-NEXT: %[[gep_src:.*]] = getelementptr inbounds i8* %src, i64 215
106 ; CHECK-NEXT: %[[gep_dst:.*]] = getelementptr inbounds [85 x i8]* %[[test3_a7]], i64 0, i64 0
107 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep_dst]], i8* %[[gep_src]], i32 85
109 ; Clobber a single element of the array, this should be promotable.
110 %c = getelementptr [300 x i8]* %a, i64 0, i64 42
113 ; Make a sequence of overlapping stores to the array. These overlap both in
114 ; forward strides and in shrinking accesses.
115 %overlap.1.i8 = getelementptr [300 x i8]* %a, i64 0, i64 142
116 %overlap.2.i8 = getelementptr [300 x i8]* %a, i64 0, i64 143
117 %overlap.3.i8 = getelementptr [300 x i8]* %a, i64 0, i64 144
118 %overlap.4.i8 = getelementptr [300 x i8]* %a, i64 0, i64 145
119 %overlap.5.i8 = getelementptr [300 x i8]* %a, i64 0, i64 146
120 %overlap.6.i8 = getelementptr [300 x i8]* %a, i64 0, i64 147
121 %overlap.7.i8 = getelementptr [300 x i8]* %a, i64 0, i64 148
122 %overlap.8.i8 = getelementptr [300 x i8]* %a, i64 0, i64 149
123 %overlap.9.i8 = getelementptr [300 x i8]* %a, i64 0, i64 150
124 %overlap.1.i16 = bitcast i8* %overlap.1.i8 to i16*
125 %overlap.1.i32 = bitcast i8* %overlap.1.i8 to i32*
126 %overlap.1.i64 = bitcast i8* %overlap.1.i8 to i64*
127 %overlap.2.i64 = bitcast i8* %overlap.2.i8 to i64*
128 %overlap.3.i64 = bitcast i8* %overlap.3.i8 to i64*
129 %overlap.4.i64 = bitcast i8* %overlap.4.i8 to i64*
130 %overlap.5.i64 = bitcast i8* %overlap.5.i8 to i64*
131 %overlap.6.i64 = bitcast i8* %overlap.6.i8 to i64*
132 %overlap.7.i64 = bitcast i8* %overlap.7.i8 to i64*
133 %overlap.8.i64 = bitcast i8* %overlap.8.i8 to i64*
134 %overlap.9.i64 = bitcast i8* %overlap.9.i8 to i64*
135 store i8 1, i8* %overlap.1.i8
136 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds [16 x i8]* %[[test3_a3]], i64 0, i64 0
137 ; CHECK-NEXT: store i8 1, i8* %[[gep]]
138 store i16 1, i16* %overlap.1.i16
139 ; CHECK-NEXT: %[[bitcast:.*]] = bitcast [16 x i8]* %[[test3_a3]] to i16*
140 ; CHECK-NEXT: store i16 1, i16* %[[bitcast]]
141 store i32 1, i32* %overlap.1.i32
142 ; CHECK-NEXT: %[[bitcast:.*]] = bitcast [16 x i8]* %[[test3_a3]] to i32*
143 ; CHECK-NEXT: store i32 1, i32* %[[bitcast]]
144 store i64 1, i64* %overlap.1.i64
145 ; CHECK-NEXT: %[[bitcast:.*]] = bitcast [16 x i8]* %[[test3_a3]] to i64*
146 ; CHECK-NEXT: store i64 1, i64* %[[bitcast]]
147 store i64 2, i64* %overlap.2.i64
148 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds [16 x i8]* %[[test3_a3]], i64 0, i64 1
149 ; CHECK-NEXT: %[[bitcast:.*]] = bitcast i8* %[[gep]] to i64*
150 ; CHECK-NEXT: store i64 2, i64* %[[bitcast]]
151 store i64 3, i64* %overlap.3.i64
152 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds [16 x i8]* %[[test3_a3]], i64 0, i64 2
153 ; CHECK-NEXT: %[[bitcast:.*]] = bitcast i8* %[[gep]] to i64*
154 ; CHECK-NEXT: store i64 3, i64* %[[bitcast]]
155 store i64 4, i64* %overlap.4.i64
156 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds [16 x i8]* %[[test3_a3]], i64 0, i64 3
157 ; CHECK-NEXT: %[[bitcast:.*]] = bitcast i8* %[[gep]] to i64*
158 ; CHECK-NEXT: store i64 4, i64* %[[bitcast]]
159 store i64 5, i64* %overlap.5.i64
160 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds [16 x i8]* %[[test3_a3]], i64 0, i64 4
161 ; CHECK-NEXT: %[[bitcast:.*]] = bitcast i8* %[[gep]] to i64*
162 ; CHECK-NEXT: store i64 5, i64* %[[bitcast]]
163 store i64 6, i64* %overlap.6.i64
164 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds [16 x i8]* %[[test3_a3]], i64 0, i64 5
165 ; CHECK-NEXT: %[[bitcast:.*]] = bitcast i8* %[[gep]] to i64*
166 ; CHECK-NEXT: store i64 6, i64* %[[bitcast]]
167 store i64 7, i64* %overlap.7.i64
168 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds [16 x i8]* %[[test3_a3]], i64 0, i64 6
169 ; CHECK-NEXT: %[[bitcast:.*]] = bitcast i8* %[[gep]] to i64*
170 ; CHECK-NEXT: store i64 7, i64* %[[bitcast]]
171 store i64 8, i64* %overlap.8.i64
172 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds [16 x i8]* %[[test3_a3]], i64 0, i64 7
173 ; CHECK-NEXT: %[[bitcast:.*]] = bitcast i8* %[[gep]] to i64*
174 ; CHECK-NEXT: store i64 8, i64* %[[bitcast]]
175 store i64 9, i64* %overlap.9.i64
176 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds [16 x i8]* %[[test3_a3]], i64 0, i64 8
177 ; CHECK-NEXT: %[[bitcast:.*]] = bitcast i8* %[[gep]] to i64*
178 ; CHECK-NEXT: store i64 9, i64* %[[bitcast]]
180 ; Make two sequences of overlapping stores with more gaps and irregularities.
181 %overlap2.1.0.i8 = getelementptr [300 x i8]* %a, i64 0, i64 200
182 %overlap2.1.1.i8 = getelementptr [300 x i8]* %a, i64 0, i64 201
183 %overlap2.1.2.i8 = getelementptr [300 x i8]* %a, i64 0, i64 202
184 %overlap2.1.3.i8 = getelementptr [300 x i8]* %a, i64 0, i64 203
186 %overlap2.2.0.i8 = getelementptr [300 x i8]* %a, i64 0, i64 208
187 %overlap2.2.1.i8 = getelementptr [300 x i8]* %a, i64 0, i64 209
188 %overlap2.2.2.i8 = getelementptr [300 x i8]* %a, i64 0, i64 210
189 %overlap2.2.3.i8 = getelementptr [300 x i8]* %a, i64 0, i64 211
191 %overlap2.1.0.i16 = bitcast i8* %overlap2.1.0.i8 to i16*
192 %overlap2.1.0.i32 = bitcast i8* %overlap2.1.0.i8 to i32*
193 %overlap2.1.1.i32 = bitcast i8* %overlap2.1.1.i8 to i32*
194 %overlap2.1.2.i32 = bitcast i8* %overlap2.1.2.i8 to i32*
195 %overlap2.1.3.i32 = bitcast i8* %overlap2.1.3.i8 to i32*
196 store i8 1, i8* %overlap2.1.0.i8
197 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds [7 x i8]* %[[test3_a5]], i64 0, i64 0
198 ; CHECK-NEXT: store i8 1, i8* %[[gep]]
199 store i16 1, i16* %overlap2.1.0.i16
200 ; CHECK-NEXT: %[[bitcast:.*]] = bitcast [7 x i8]* %[[test3_a5]] to i16*
201 ; CHECK-NEXT: store i16 1, i16* %[[bitcast]]
202 store i32 1, i32* %overlap2.1.0.i32
203 ; CHECK-NEXT: %[[bitcast:.*]] = bitcast [7 x i8]* %[[test3_a5]] to i32*
204 ; CHECK-NEXT: store i32 1, i32* %[[bitcast]]
205 store i32 2, i32* %overlap2.1.1.i32
206 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds [7 x i8]* %[[test3_a5]], i64 0, i64 1
207 ; CHECK-NEXT: %[[bitcast:.*]] = bitcast i8* %[[gep]] to i32*
208 ; CHECK-NEXT: store i32 2, i32* %[[bitcast]]
209 store i32 3, i32* %overlap2.1.2.i32
210 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds [7 x i8]* %[[test3_a5]], i64 0, i64 2
211 ; CHECK-NEXT: %[[bitcast:.*]] = bitcast i8* %[[gep]] to i32*
212 ; CHECK-NEXT: store i32 3, i32* %[[bitcast]]
213 store i32 4, i32* %overlap2.1.3.i32
214 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds [7 x i8]* %[[test3_a5]], i64 0, i64 3
215 ; CHECK-NEXT: %[[bitcast:.*]] = bitcast i8* %[[gep]] to i32*
216 ; CHECK-NEXT: store i32 4, i32* %[[bitcast]]
218 %overlap2.2.0.i32 = bitcast i8* %overlap2.2.0.i8 to i32*
219 %overlap2.2.1.i16 = bitcast i8* %overlap2.2.1.i8 to i16*
220 %overlap2.2.1.i32 = bitcast i8* %overlap2.2.1.i8 to i32*
221 %overlap2.2.2.i32 = bitcast i8* %overlap2.2.2.i8 to i32*
222 %overlap2.2.3.i32 = bitcast i8* %overlap2.2.3.i8 to i32*
223 store i32 1, i32* %overlap2.2.0.i32
224 ; CHECK-NEXT: %[[bitcast:.*]] = bitcast [7 x i8]* %[[test3_a6]] to i32*
225 ; CHECK-NEXT: store i32 1, i32* %[[bitcast]]
226 store i8 1, i8* %overlap2.2.1.i8
227 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds [7 x i8]* %[[test3_a6]], i64 0, i64 1
228 ; CHECK-NEXT: store i8 1, i8* %[[gep]]
229 store i16 1, i16* %overlap2.2.1.i16
230 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds [7 x i8]* %[[test3_a6]], i64 0, i64 1
231 ; CHECK-NEXT: %[[bitcast:.*]] = bitcast i8* %[[gep]] to i16*
232 ; CHECK-NEXT: store i16 1, i16* %[[bitcast]]
233 store i32 1, i32* %overlap2.2.1.i32
234 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds [7 x i8]* %[[test3_a6]], i64 0, i64 1
235 ; CHECK-NEXT: %[[bitcast:.*]] = bitcast i8* %[[gep]] to i32*
236 ; CHECK-NEXT: store i32 1, i32* %[[bitcast]]
237 store i32 3, i32* %overlap2.2.2.i32
238 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds [7 x i8]* %[[test3_a6]], i64 0, i64 2
239 ; CHECK-NEXT: %[[bitcast:.*]] = bitcast i8* %[[gep]] to i32*
240 ; CHECK-NEXT: store i32 3, i32* %[[bitcast]]
241 store i32 4, i32* %overlap2.2.3.i32
242 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds [7 x i8]* %[[test3_a6]], i64 0, i64 3
243 ; CHECK-NEXT: %[[bitcast:.*]] = bitcast i8* %[[gep]] to i32*
244 ; CHECK-NEXT: store i32 4, i32* %[[bitcast]]
246 %overlap2.prefix = getelementptr i8* %overlap2.1.1.i8, i64 -4
247 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %overlap2.prefix, i8* %src, i32 8, i32 1, i1 false)
248 ; CHECK-NEXT: %[[gep_dst:.*]] = getelementptr inbounds [42 x i8]* %[[test3_a4]], i64 0, i64 39
249 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep_dst]], i8* %src, i32 3
250 ; CHECK-NEXT: %[[gep_src:.*]] = getelementptr inbounds i8* %src, i64 3
251 ; CHECK-NEXT: %[[gep_dst:.*]] = getelementptr inbounds [7 x i8]* %[[test3_a5]], i64 0, i64 0
252 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep_dst]], i8* %[[gep_src]], i32 5
254 ; Bridge between the overlapping areas
255 call void @llvm.memset.p0i8.i32(i8* %overlap2.1.2.i8, i8 42, i32 8, i32 1, i1 false)
256 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds [7 x i8]* %[[test3_a5]], i64 0, i64 2
257 ; CHECK-NEXT: call void @llvm.memset.p0i8.i32(i8* %[[gep]], i8 42, i32 5
258 ; ...promoted i8 store...
259 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds [7 x i8]* %[[test3_a6]], i64 0, i64 0
260 ; CHECK-NEXT: call void @llvm.memset.p0i8.i32(i8* %[[gep]], i8 42, i32 2
262 ; Entirely within the second overlap.
263 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %overlap2.2.1.i8, i8* %src, i32 5, i32 1, i1 false)
264 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds [7 x i8]* %[[test3_a6]], i64 0, i64 1
265 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep]], i8* %src, i32 5
267 ; Trailing past the second overlap.
268 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %overlap2.2.2.i8, i8* %src, i32 8, i32 1, i1 false)
269 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds [7 x i8]* %[[test3_a6]], i64 0, i64 2
270 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep]], i8* %src, i32 5
271 ; CHECK-NEXT: %[[gep_src:.*]] = getelementptr inbounds i8* %src, i64 5
272 ; CHECK-NEXT: %[[gep_dst:.*]] = getelementptr inbounds [85 x i8]* %[[test3_a7]], i64 0, i64 0
273 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep_dst]], i8* %[[gep_src]], i32 3
275 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dst, i8* %b, i32 300, i32 1, i1 false)
276 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds [42 x i8]* %[[test3_a1]], i64 0, i64 0
277 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dst, i8* %[[gep]], i32 42
278 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds i8* %dst, i64 42
279 ; CHECK-NEXT: store i8 0, i8* %[[gep]]
280 ; CHECK-NEXT: %[[gep_dst:.*]] = getelementptr inbounds i8* %dst, i64 43
281 ; CHECK-NEXT: %[[gep_src:.*]] = getelementptr inbounds [99 x i8]* %[[test3_a2]], i64 0, i64 0
282 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep_dst]], i8* %[[gep_src]], i32 99
283 ; CHECK-NEXT: %[[gep_dst:.*]] = getelementptr inbounds i8* %dst, i64 142
284 ; CHECK-NEXT: %[[gep_src:.*]] = getelementptr inbounds [16 x i8]* %[[test3_a3]], i64 0, i64 0
285 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep_dst]], i8* %[[gep_src]], i32 16
286 ; CHECK-NEXT: %[[gep_dst:.*]] = getelementptr inbounds i8* %dst, i64 158
287 ; CHECK-NEXT: %[[gep_src:.*]] = getelementptr inbounds [42 x i8]* %[[test3_a4]], i64 0, i64 0
288 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep_dst]], i8* %[[gep_src]], i32 42
289 ; CHECK-NEXT: %[[gep_dst:.*]] = getelementptr inbounds i8* %dst, i64 200
290 ; CHECK-NEXT: %[[gep_src:.*]] = getelementptr inbounds [7 x i8]* %[[test3_a5]], i64 0, i64 0
291 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep_dst]], i8* %[[gep_src]], i32 7
292 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds i8* %dst, i64 207
293 ; CHECK-NEXT: store i8 42, i8* %[[gep]]
294 ; CHECK-NEXT: %[[gep_dst:.*]] = getelementptr inbounds i8* %dst, i64 208
295 ; CHECK-NEXT: %[[gep_src:.*]] = getelementptr inbounds [7 x i8]* %[[test3_a6]], i64 0, i64 0
296 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep_dst]], i8* %[[gep_src]], i32 7
297 ; CHECK-NEXT: %[[gep_dst:.*]] = getelementptr inbounds i8* %dst, i64 215
298 ; CHECK-NEXT: %[[gep_src:.*]] = getelementptr inbounds [85 x i8]* %[[test3_a7]], i64 0, i64 0
299 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep_dst]], i8* %[[gep_src]], i32 85
304 define void @test4(i8* %dst, i8* %src) {
308 %a = alloca [100 x i8]
310 ; CHECK: %[[test4_a1:.*]] = alloca [20 x i8]
311 ; CHECK-NEXT: %[[test4_a2:.*]] = alloca [7 x i8]
312 ; CHECK-NEXT: %[[test4_a3:.*]] = alloca [10 x i8]
313 ; CHECK-NEXT: %[[test4_a4:.*]] = alloca [7 x i8]
314 ; CHECK-NEXT: %[[test4_a5:.*]] = alloca [7 x i8]
315 ; CHECK-NEXT: %[[test4_a6:.*]] = alloca [40 x i8]
317 %b = getelementptr [100 x i8]* %a, i64 0, i64 0
318 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %b, i8* %src, i32 100, i32 1, i1 false)
319 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds [20 x i8]* %[[test4_a1]], i64 0, i64 0
320 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep]], i8* %src, i32 20
321 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds i8* %src, i64 20
322 ; CHECK-NEXT: %[[bitcast:.*]] = bitcast i8* %[[gep]] to i16*
323 ; CHECK-NEXT: %[[test4_r1:.*]] = load i16* %[[bitcast]]
324 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds i8* %src, i64 22
325 ; CHECK-NEXT: %[[test4_r2:.*]] = load i8* %[[gep]]
326 ; CHECK-NEXT: %[[gep_src:.*]] = getelementptr inbounds i8* %src, i64 23
327 ; CHECK-NEXT: %[[gep_dst:.*]] = getelementptr inbounds [7 x i8]* %[[test4_a2]], i64 0, i64 0
328 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep_dst]], i8* %[[gep_src]], i32 7
329 ; CHECK-NEXT: %[[gep_src:.*]] = getelementptr inbounds i8* %src, i64 30
330 ; CHECK-NEXT: %[[gep_dst:.*]] = getelementptr inbounds [10 x i8]* %[[test4_a3]], i64 0, i64 0
331 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep_dst]], i8* %[[gep_src]], i32 10
332 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds i8* %src, i64 40
333 ; CHECK-NEXT: %[[bitcast:.*]] = bitcast i8* %[[gep]] to i16*
334 ; CHECK-NEXT: %[[test4_r3:.*]] = load i16* %[[bitcast]]
335 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds i8* %src, i64 42
336 ; CHECK-NEXT: %[[test4_r4:.*]] = load i8* %[[gep]]
337 ; CHECK-NEXT: %[[gep_src:.*]] = getelementptr inbounds i8* %src, i64 43
338 ; CHECK-NEXT: %[[gep_dst:.*]] = getelementptr inbounds [7 x i8]* %[[test4_a4]], i64 0, i64 0
339 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep_dst]], i8* %[[gep_src]], i32 7
340 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds i8* %src, i64 50
341 ; CHECK-NEXT: %[[bitcast:.*]] = bitcast i8* %[[gep]] to i16*
342 ; CHECK-NEXT: %[[test4_r5:.*]] = load i16* %[[bitcast]]
343 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds i8* %src, i64 52
344 ; CHECK-NEXT: %[[test4_r6:.*]] = load i8* %[[gep]]
345 ; CHECK-NEXT: %[[gep_src:.*]] = getelementptr inbounds i8* %src, i64 53
346 ; CHECK-NEXT: %[[gep_dst:.*]] = getelementptr inbounds [7 x i8]* %[[test4_a5]], i64 0, i64 0
347 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep_dst]], i8* %[[gep_src]], i32 7
348 ; CHECK-NEXT: %[[gep_src:.*]] = getelementptr inbounds i8* %src, i64 60
349 ; CHECK-NEXT: %[[gep_dst:.*]] = getelementptr inbounds [40 x i8]* %[[test4_a6]], i64 0, i64 0
350 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep_dst]], i8* %[[gep_src]], i32 40
352 %a.src.1 = getelementptr [100 x i8]* %a, i64 0, i64 20
353 %a.dst.1 = getelementptr [100 x i8]* %a, i64 0, i64 40
354 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %a.dst.1, i8* %a.src.1, i32 10, i32 1, i1 false)
355 ; CHECK-NEXT: %[[gep_dst:.*]] = getelementptr inbounds [7 x i8]* %[[test4_a4]], i64 0, i64 0
356 ; CHECK-NEXT: %[[gep_src:.*]] = getelementptr inbounds [7 x i8]* %[[test4_a2]], i64 0, i64 0
357 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep_dst]], i8* %[[gep_src]], i32 7
359 ; Clobber a single element of the array, this should be promotable, and be deleted.
360 %c = getelementptr [100 x i8]* %a, i64 0, i64 42
363 %a.src.2 = getelementptr [100 x i8]* %a, i64 0, i64 50
364 call void @llvm.memmove.p0i8.p0i8.i32(i8* %a.dst.1, i8* %a.src.2, i32 10, i32 1, i1 false)
365 ; CHECK-NEXT: %[[gep_dst:.*]] = getelementptr inbounds [7 x i8]* %[[test4_a4]], i64 0, i64 0
366 ; CHECK-NEXT: %[[gep_src:.*]] = getelementptr inbounds [7 x i8]* %[[test4_a5]], i64 0, i64 0
367 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep_dst]], i8* %[[gep_src]], i32 7
369 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dst, i8* %b, i32 100, i32 1, i1 false)
370 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds [20 x i8]* %[[test4_a1]], i64 0, i64 0
371 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dst, i8* %[[gep]], i32 20
372 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds i8* %dst, i64 20
373 ; CHECK-NEXT: %[[bitcast:.*]] = bitcast i8* %[[gep]] to i16*
374 ; CHECK-NEXT: store i16 %[[test4_r1]], i16* %[[bitcast]]
375 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds i8* %dst, i64 22
376 ; CHECK-NEXT: store i8 %[[test4_r2]], i8* %[[gep]]
377 ; CHECK-NEXT: %[[gep_dst:.*]] = getelementptr inbounds i8* %dst, i64 23
378 ; CHECK-NEXT: %[[gep_src:.*]] = getelementptr inbounds [7 x i8]* %[[test4_a2]], i64 0, i64 0
379 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep_dst]], i8* %[[gep_src]], i32 7
380 ; CHECK-NEXT: %[[gep_dst:.*]] = getelementptr inbounds i8* %dst, i64 30
381 ; CHECK-NEXT: %[[gep_src:.*]] = getelementptr inbounds [10 x i8]* %[[test4_a3]], i64 0, i64 0
382 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep_dst]], i8* %[[gep_src]], i32 10
383 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds i8* %dst, i64 40
384 ; CHECK-NEXT: %[[bitcast:.*]] = bitcast i8* %[[gep]] to i16*
385 ; CHECK-NEXT: store i16 %[[test4_r5]], i16* %[[bitcast]]
386 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds i8* %dst, i64 42
387 ; CHECK-NEXT: store i8 %[[test4_r6]], i8* %[[gep]]
388 ; CHECK-NEXT: %[[gep_dst:.*]] = getelementptr inbounds i8* %dst, i64 43
389 ; CHECK-NEXT: %[[gep_src:.*]] = getelementptr inbounds [7 x i8]* %[[test4_a4]], i64 0, i64 0
390 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep_dst]], i8* %[[gep_src]], i32 7
391 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds i8* %dst, i64 50
392 ; CHECK-NEXT: %[[bitcast:.*]] = bitcast i8* %[[gep]] to i16*
393 ; CHECK-NEXT: store i16 %[[test4_r5]], i16* %[[bitcast]]
394 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds i8* %dst, i64 52
395 ; CHECK-NEXT: store i8 %[[test4_r6]], i8* %[[gep]]
396 ; CHECK-NEXT: %[[gep_dst:.*]] = getelementptr inbounds i8* %dst, i64 53
397 ; CHECK-NEXT: %[[gep_src:.*]] = getelementptr inbounds [7 x i8]* %[[test4_a5]], i64 0, i64 0
398 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep_dst]], i8* %[[gep_src]], i32 7
399 ; CHECK-NEXT: %[[gep_dst:.*]] = getelementptr inbounds i8* %dst, i64 60
400 ; CHECK-NEXT: %[[gep_src:.*]] = getelementptr inbounds [40 x i8]* %[[test4_a6]], i64 0, i64 0
401 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[gep_dst]], i8* %[[gep_src]], i32 40
406 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* nocapture, i8* nocapture, i32, i32, i1) nounwind
407 declare void @llvm.memmove.p0i8.p0i8.i32(i8* nocapture, i8* nocapture, i32, i32, i1) nounwind
408 declare void @llvm.memset.p0i8.i32(i8* nocapture, i8, i32, i32, i1) nounwind
410 define i16 @test5() {
412 ; CHECK-NOT: alloca float
413 ; CHECK: %[[cast:.*]] = bitcast float 0.0{{.*}} to i32
414 ; CHECK-NEXT: %[[shr:.*]] = lshr i32 %[[cast]], 16
415 ; CHECK-NEXT: %[[trunc:.*]] = trunc i32 %[[shr]] to i16
416 ; CHECK-NEXT: ret i16 %[[trunc]]
420 %fptr = bitcast [4 x i8]* %a to float*
421 store float 0.0, float* %fptr
422 %ptr = getelementptr [4 x i8]* %a, i32 0, i32 2
423 %iptr = bitcast i8* %ptr to i16*
424 %val = load i16* %iptr
428 define i32 @test6() {
431 ; CHECK-NEXT: store volatile i32
432 ; CHECK-NEXT: load i32*
433 ; CHECK-NEXT: ret i32
437 %ptr = getelementptr [4 x i8]* %a, i32 0, i32 0
438 call void @llvm.memset.p0i8.i32(i8* %ptr, i8 42, i32 4, i32 1, i1 true)
439 %iptr = bitcast i8* %ptr to i32*
440 %val = load i32* %iptr
444 define void @test7(i8* %src, i8* %dst) {
447 ; CHECK-NEXT: bitcast i8* %src to i32*
448 ; CHECK-NEXT: load volatile i32*
449 ; CHECK-NEXT: store volatile i32
450 ; CHECK-NEXT: bitcast i8* %dst to i32*
451 ; CHECK-NEXT: load volatile i32*
452 ; CHECK-NEXT: store volatile i32
457 %ptr = getelementptr [4 x i8]* %a, i32 0, i32 0
458 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %ptr, i8* %src, i32 4, i32 1, i1 true)
459 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dst, i8* %ptr, i32 4, i32 1, i1 true)
464 %S1 = type { i32, i32, [16 x i8] }
465 %S2 = type { %S1*, %S2* }
467 define %S2 @test8(%S2* %s2) {
473 %s2.next.ptr = getelementptr %S2* %s2, i64 0, i32 1
474 %s2.next = load %S2** %s2.next.ptr
475 ; CHECK: %[[gep:.*]] = getelementptr %S2* %s2, i64 0, i32 1
476 ; CHECK-NEXT: %[[next:.*]] = load %S2** %[[gep]]
478 %s2.next.s1.ptr = getelementptr %S2* %s2.next, i64 0, i32 0
479 %s2.next.s1 = load %S1** %s2.next.s1.ptr
480 %new.s1.ptr = getelementptr %S2* %new, i64 0, i32 0
481 store %S1* %s2.next.s1, %S1** %new.s1.ptr
482 %s2.next.next.ptr = getelementptr %S2* %s2.next, i64 0, i32 1
483 %s2.next.next = load %S2** %s2.next.next.ptr
484 %new.next.ptr = getelementptr %S2* %new, i64 0, i32 1
485 store %S2* %s2.next.next, %S2** %new.next.ptr
486 ; CHECK-NEXT: %[[gep:.*]] = getelementptr %S2* %[[next]], i64 0, i32 0
487 ; CHECK-NEXT: %[[next_s1:.*]] = load %S1** %[[gep]]
488 ; CHECK-NEXT: %[[gep:.*]] = getelementptr %S2* %[[next]], i64 0, i32 1
489 ; CHECK-NEXT: %[[next_next:.*]] = load %S2** %[[gep]]
491 %new.s1 = load %S1** %new.s1.ptr
492 %result1 = insertvalue %S2 undef, %S1* %new.s1, 0
493 ; CHECK-NEXT: %[[result1:.*]] = insertvalue %S2 undef, %S1* %[[next_s1]], 0
494 %new.next = load %S2** %new.next.ptr
495 %result2 = insertvalue %S2 %result1, %S2* %new.next, 1
496 ; CHECK-NEXT: %[[result2:.*]] = insertvalue %S2 %[[result1]], %S2* %[[next_next]], 1
498 ; CHECK-NEXT: ret %S2 %[[result2]]
501 define i64 @test9() {
502 ; Ensure we can handle loads off the end of an alloca even when wrapped in
503 ; weird bit casts and types. The result is undef, but this shouldn't crash
507 ; CHECK: ret i64 undef
510 %a = alloca { [3 x i8] }
511 %gep1 = getelementptr inbounds { [3 x i8] }* %a, i32 0, i32 0, i32 0
512 store i8 0, i8* %gep1, align 1
513 %gep2 = getelementptr inbounds { [3 x i8] }* %a, i32 0, i32 0, i32 1
514 store i8 0, i8* %gep2, align 1
515 %gep3 = getelementptr inbounds { [3 x i8] }* %a, i32 0, i32 0, i32 2
516 store i8 26, i8* %gep3, align 1
517 %cast = bitcast { [3 x i8] }* %a to { i64 }*
518 %elt = getelementptr inbounds { i64 }* %cast, i32 0, i32 0
519 %result = load i64* %elt
523 define %S2* @test10() {
525 ; CHECK-NOT: alloca %S2*
526 ; CHECK: ret %S2* null
530 %ptr = getelementptr [8 x i8]* %a, i32 0, i32 0
531 call void @llvm.memset.p0i8.i32(i8* %ptr, i8 0, i32 8, i32 1, i1 false)
532 %s2ptrptr = bitcast i8* %ptr to %S2**
533 %s2ptr = load %S2** %s2ptrptr
537 define i32 @test11() {
544 br i1 undef, label %good, label %bad
547 %Y = getelementptr i32* %X, i64 0
553 %Y2 = getelementptr i32* %X, i64 1
554 store i32 0, i32* %Y2
559 define i8 @test12() {
560 ; We fully promote these to the i24 load or store size, resulting in just masks
561 ; and other operations that instcombine will fold, but no alloca.
570 %a0ptr = getelementptr [3 x i8]* %a, i64 0, i32 0
571 store i8 0, i8* %a0ptr
572 %a1ptr = getelementptr [3 x i8]* %a, i64 0, i32 1
573 store i8 0, i8* %a1ptr
574 %a2ptr = getelementptr [3 x i8]* %a, i64 0, i32 2
575 store i8 0, i8* %a2ptr
576 %aiptr = bitcast [3 x i8]* %a to i24*
577 %ai = load i24* %aiptr
580 ; CHECK: %[[mask0:.*]] = and i24 undef, -256
581 ; CHECK-NEXT: %[[mask1:.*]] = and i24 %[[mask0]], -65281
582 ; CHECK-NEXT: %[[mask2:.*]] = and i24 %[[mask1]], 65535
584 %biptr = bitcast [3 x i8]* %b to i24*
585 store i24 %ai, i24* %biptr
586 %b0ptr = getelementptr [3 x i8]* %b, i64 0, i32 0
587 %b0 = load i8* %b0ptr
588 %b1ptr = getelementptr [3 x i8]* %b, i64 0, i32 1
589 %b1 = load i8* %b1ptr
590 %b2ptr = getelementptr [3 x i8]* %b, i64 0, i32 2
591 %b2 = load i8* %b2ptr
594 ; CHECK: %[[trunc0:.*]] = trunc i24 %[[mask2]] to i8
595 ; CHECK-NEXT: %[[shift1:.*]] = lshr i24 %[[mask2]], 8
596 ; CHECK-NEXT: %[[trunc1:.*]] = trunc i24 %[[shift1]] to i8
597 ; CHECK-NEXT: %[[shift2:.*]] = lshr i24 %[[mask2]], 16
598 ; CHECK-NEXT: %[[trunc2:.*]] = trunc i24 %[[shift2]] to i8
600 %bsum0 = add i8 %b0, %b1
601 %bsum1 = add i8 %bsum0, %b2
603 ; CHECK: %[[sum0:.*]] = add i8 %[[trunc0]], %[[trunc1]]
604 ; CHECK-NEXT: %[[sum1:.*]] = add i8 %[[sum0]], %[[trunc2]]
605 ; CHECK-NEXT: ret i8 %[[sum1]]
608 define i32 @test13() {
609 ; Ensure we don't crash and handle undefined loads that straddle the end of the
612 ; CHECK: %[[ret:.*]] = zext i16 undef to i32
613 ; CHECK: ret i32 %[[ret]]
617 %b0ptr = getelementptr [3 x i8]* %a, i64 0, i32 0
618 store i8 0, i8* %b0ptr
619 %b1ptr = getelementptr [3 x i8]* %a, i64 0, i32 1
620 store i8 0, i8* %b1ptr
621 %b2ptr = getelementptr [3 x i8]* %a, i64 0, i32 2
622 store i8 0, i8* %b2ptr
623 %iptrcast = bitcast [3 x i8]* %a to i16*
624 %iptrgep = getelementptr i16* %iptrcast, i64 1
625 %i = load i16* %iptrgep
626 %ret = zext i16 %i to i32
630 %test14.struct = type { [3 x i32] }
632 define void @test14(...) nounwind uwtable {
633 ; This is a strange case where we split allocas into promotable partitions, but
634 ; also gain enough data to prove they must be dead allocas due to GEPs that walk
635 ; across two adjacent allocas. Test that we don't try to promote or otherwise
636 ; do bad things to these dead allocas, they should just be removed.
639 ; CHECK-NEXT: ret void
642 %a = alloca %test14.struct
643 %p = alloca %test14.struct*
644 %0 = bitcast %test14.struct* %a to i8*
645 %1 = getelementptr i8* %0, i64 12
646 %2 = bitcast i8* %1 to %test14.struct*
647 %3 = getelementptr inbounds %test14.struct* %2, i32 0, i32 0
648 %4 = getelementptr inbounds %test14.struct* %a, i32 0, i32 0
649 %5 = bitcast [3 x i32]* %3 to i32*
650 %6 = bitcast [3 x i32]* %4 to i32*
651 %7 = load i32* %6, align 4
652 store i32 %7, i32* %5, align 4
653 %8 = getelementptr inbounds i32* %5, i32 1
654 %9 = getelementptr inbounds i32* %6, i32 1
655 %10 = load i32* %9, align 4
656 store i32 %10, i32* %8, align 4
657 %11 = getelementptr inbounds i32* %5, i32 2
658 %12 = getelementptr inbounds i32* %6, i32 2
659 %13 = load i32* %12, align 4
660 store i32 %13, i32* %11, align 4
664 define i32 @test15(i1 %flag) nounwind uwtable {
665 ; Ensure that when there are dead instructions using an alloca that are not
666 ; loads or stores we still delete them during partitioning and rewriting.
667 ; Otherwise we'll go to promote them while thy still have unpromotable uses.
670 ; CHECK-NEXT: br label %loop
672 ; CHECK-NEXT: br label %loop
682 %dead3 = phi i8* [ %gep3, %loop ], [ null, %entry ]
684 store i64 1879048192, i64* %l0, align 8
685 %bc0 = bitcast i64* %l0 to i8*
686 %gep0 = getelementptr i8* %bc0, i64 3
687 %dead0 = bitcast i8* %gep0 to i64*
689 store i64 1879048192, i64* %l1, align 8
690 %bc1 = bitcast i64* %l1 to i8*
691 %gep1 = getelementptr i8* %bc1, i64 3
692 %dead1 = getelementptr i8* %gep1, i64 1
694 store i64 1879048192, i64* %l2, align 8
695 %bc2 = bitcast i64* %l2 to i8*
696 %gep2.1 = getelementptr i8* %bc2, i64 1
697 %gep2.2 = getelementptr i8* %bc2, i64 3
698 ; Note that this select should get visited multiple times due to using two
699 ; different GEPs off the same alloca. We should only delete it once.
700 %dead2 = select i1 %flag, i8* %gep2.1, i8* %gep2.2
702 store i64 1879048192, i64* %l3, align 8
703 %bc3 = bitcast i64* %l3 to i8*
704 %gep3 = getelementptr i8* %bc3, i64 3
709 define void @test16(i8* %src, i8* %dst) {
710 ; Ensure that we can promote an alloca of [3 x i8] to an i24 SSA value.
713 ; CHECK: %[[srccast:.*]] = bitcast i8* %src to i24*
714 ; CHECK-NEXT: load i24* %[[srccast]]
715 ; CHECK-NEXT: %[[dstcast:.*]] = bitcast i8* %dst to i24*
716 ; CHECK-NEXT: store i24 0, i24* %[[dstcast]]
717 ; CHECK-NEXT: ret void
721 %ptr = getelementptr [3 x i8]* %a, i32 0, i32 0
722 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %ptr, i8* %src, i32 4, i32 1, i1 false)
723 %cast = bitcast i8* %ptr to i24*
724 store i24 0, i24* %cast
725 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dst, i8* %ptr, i32 4, i32 1, i1 false)
729 define void @test17(i8* %src, i8* %dst) {
730 ; Ensure that we can rewrite unpromotable memcpys which extend past the end of
733 ; CHECK: %[[a:.*]] = alloca [3 x i8]
734 ; CHECK-NEXT: %[[ptr:.*]] = getelementptr [3 x i8]* %[[a]], i32 0, i32 0
735 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[ptr]], i8* %src,
736 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dst, i8* %[[ptr]],
737 ; CHECK-NEXT: ret void
741 %ptr = getelementptr [3 x i8]* %a, i32 0, i32 0
742 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %ptr, i8* %src, i32 4, i32 1, i1 true)
743 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dst, i8* %ptr, i32 4, i32 1, i1 true)
747 define void @test18(i8* %src, i8* %dst, i32 %size) {
748 ; Preserve transfer instrinsics with a variable size, even if they overlap with
749 ; fixed size operations. Further, continue to split and promote allocas preceding
750 ; the variable sized intrinsic.
752 ; CHECK: %[[a:.*]] = alloca [34 x i8]
753 ; CHECK: %[[srcgep1:.*]] = getelementptr inbounds i8* %src, i64 4
754 ; CHECK-NEXT: %[[srccast1:.*]] = bitcast i8* %[[srcgep1]] to i32*
755 ; CHECK-NEXT: %[[srcload:.*]] = load i32* %[[srccast1]]
756 ; CHECK-NEXT: %[[agep1:.*]] = getelementptr inbounds [34 x i8]* %[[a]], i64 0, i64 0
757 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %[[agep1]], i8* %src, i32 %size,
758 ; CHECK-NEXT: %[[agep2:.*]] = getelementptr inbounds [34 x i8]* %[[a]], i64 0, i64 0
759 ; CHECK-NEXT: call void @llvm.memset.p0i8.i32(i8* %[[agep2]], i8 42, i32 %size,
760 ; CHECK-NEXT: %[[dstcast1:.*]] = bitcast i8* %dst to i32*
761 ; CHECK-NEXT: store i32 42, i32* %[[dstcast1]]
762 ; CHECK-NEXT: %[[dstgep1:.*]] = getelementptr inbounds i8* %dst, i64 4
763 ; CHECK-NEXT: %[[dstcast2:.*]] = bitcast i8* %[[dstgep1]] to i32*
764 ; CHECK-NEXT: store i32 %[[srcload]], i32* %[[dstcast2]]
765 ; CHECK-NEXT: %[[agep3:.*]] = getelementptr inbounds [34 x i8]* %[[a]], i64 0, i64 0
766 ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dst, i8* %[[agep3]], i32 %size,
767 ; CHECK-NEXT: ret void
770 %a = alloca [42 x i8]
771 %ptr = getelementptr [42 x i8]* %a, i32 0, i32 0
772 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %ptr, i8* %src, i32 8, i32 1, i1 false)
773 %ptr2 = getelementptr [42 x i8]* %a, i32 0, i32 8
774 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %ptr2, i8* %src, i32 %size, i32 1, i1 false)
775 call void @llvm.memset.p0i8.i32(i8* %ptr2, i8 42, i32 %size, i32 1, i1 false)
776 %cast = bitcast i8* %ptr to i32*
777 store i32 42, i32* %cast
778 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dst, i8* %ptr, i32 8, i32 1, i1 false)
779 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dst, i8* %ptr2, i32 %size, i32 1, i1 false)
783 %opaque = type opaque
785 define i32 @test19(%opaque* %x) {
786 ; This input will cause us to try to compute a natural GEP when rewriting
787 ; pointers in such a way that we try to GEP through the opaque type. Previously,
788 ; a check for an unsized type was missing and this crashed. Ensure it behaves
792 ; CHECK: ret i32 undef
795 %a = alloca { i64, i8* }
796 %cast1 = bitcast %opaque* %x to i8*
797 %cast2 = bitcast { i64, i8* }* %a to i8*
798 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %cast2, i8* %cast1, i32 16, i32 1, i1 false)
799 %gep = getelementptr inbounds { i64, i8* }* %a, i32 0, i32 0
800 %val = load i64* %gep
804 define i32 @test20() {
805 ; Ensure we can track negative offsets (before the beginning of the alloca) and
806 ; negative relative offsets from offsets starting past the end of the alloca.
809 ; CHECK: %[[sum1:.*]] = add i32 1, 2
810 ; CHECK: %[[sum2:.*]] = add i32 %[[sum1]], 3
811 ; CHECK: ret i32 %[[sum2]]
814 %a = alloca [3 x i32]
815 %gep1 = getelementptr [3 x i32]* %a, i32 0, i32 0
816 store i32 1, i32* %gep1
817 %gep2.1 = getelementptr [3 x i32]* %a, i32 0, i32 -2
818 %gep2.2 = getelementptr i32* %gep2.1, i32 3
819 store i32 2, i32* %gep2.2
820 %gep3.1 = getelementptr [3 x i32]* %a, i32 0, i32 14
821 %gep3.2 = getelementptr i32* %gep3.1, i32 -12
822 store i32 3, i32* %gep3.2
824 %load1 = load i32* %gep1
825 %load2 = load i32* %gep2.2
826 %load3 = load i32* %gep3.2
827 %sum1 = add i32 %load1, %load2
828 %sum2 = add i32 %sum1, %load3
832 declare void @llvm.memset.p0i8.i64(i8* nocapture, i8, i64, i32, i1) nounwind
834 define i8 @test21() {
835 ; Test allocations and offsets which border on overflow of the int64_t used
836 ; internally. This is really awkward to really test as LLVM doesn't really
837 ; support such extreme constructs cleanly.
840 ; CHECK: or i8 -1, -1
843 %a = alloca [2305843009213693951 x i8]
844 %gep0 = getelementptr [2305843009213693951 x i8]* %a, i64 0, i64 2305843009213693949
845 store i8 255, i8* %gep0
846 %gep1 = getelementptr [2305843009213693951 x i8]* %a, i64 0, i64 -9223372036854775807
847 %gep2 = getelementptr i8* %gep1, i64 -1
848 call void @llvm.memset.p0i8.i64(i8* %gep2, i8 0, i64 18446744073709551615, i32 1, i1 false)
849 %gep3 = getelementptr i8* %gep1, i64 9223372036854775807
850 %gep4 = getelementptr i8* %gep3, i64 9223372036854775807
851 %gep5 = getelementptr i8* %gep4, i64 -6917529027641081857
852 store i8 255, i8* %gep5
853 %cast1 = bitcast i8* %gep4 to i32*
854 store i32 0, i32* %cast1
855 %load = load i8* %gep0
856 %gep6 = getelementptr i8* %gep0, i32 1
857 %load2 = load i8* %gep6
858 %result = or i8 %load, %load2
862 %PR13916.struct = type { i8 }
864 define void @PR13916.1() {
865 ; Ensure that we handle overlapping memcpy intrinsics correctly, especially in
866 ; the case where there is a directly identical value for both source and dest.
873 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %a, i8* %a, i32 1, i32 1, i1 false)
878 define void @PR13916.2() {
879 ; Check whether we continue to handle them correctly when they start off with
880 ; different pointer value chains, but during rewriting we coalesce them into the
887 %a = alloca %PR13916.struct, align 1
888 br i1 undef, label %if.then, label %if.end
891 %tmp0 = bitcast %PR13916.struct* %a to i8*
892 %tmp1 = bitcast %PR13916.struct* %a to i8*
893 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %tmp0, i8* %tmp1, i32 1, i32 1, i1 false)
897 %gep = getelementptr %PR13916.struct* %a, i32 0, i32 0
898 %tmp2 = load i8* %gep
902 define void @PR13990() {
903 ; Ensure we can handle cases where processing one alloca causes the other
904 ; alloca to become dead and get deleted. This might crash or fail under
905 ; Valgrind if we regress.
914 br i1 undef, label %bb1, label %bb2
917 store i8* undef, i8** %tmp2
918 br i1 undef, label %bb2, label %bb3
921 %tmp50 = select i1 undef, i8** %tmp2, i8** %tmp1
922 br i1 undef, label %bb3, label %bb4
931 define double @PR13969(double %x) {
932 ; Check that we detect when promotion will un-escape an alloca and iterate to
933 ; re-try running SROA over that alloca. Without that, the two allocas that are
934 ; stored into a dead alloca don't get rewritten and promoted.
943 store double %x, double* %a
944 store double* %c, double** %b
945 store double* %a, double** %b
946 store double %x, double* %c
947 %ret = load double* %a
952 ; CHECK: ret double %x
955 %PR14034.struct = type { { {} }, i32, %PR14034.list }
956 %PR14034.list = type { %PR14034.list*, %PR14034.list* }
958 define void @PR14034() {
959 ; This test case tries to form GEPs into the empty leading struct members, and
960 ; subsequently crashed (under valgrind) before we fixed the PR. The important
961 ; thing is to handle empty structs gracefully.
965 %a = alloca %PR14034.struct
966 %list = getelementptr %PR14034.struct* %a, i32 0, i32 2
967 %prev = getelementptr %PR14034.list* %list, i32 0, i32 1
968 store %PR14034.list* undef, %PR14034.list** %prev
969 %cast0 = bitcast %PR14034.struct* undef to i8*
970 %cast1 = bitcast %PR14034.struct* %a to i8*
971 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %cast0, i8* %cast1, i32 12, i32 0, i1 false)
975 define i32 @test22(i32 %x) {
976 ; Test that SROA and promotion is not confused by a grab bax mixture of pointer
977 ; types involving wrapper aggregates and zero-length aggregate members.
981 %a1 = alloca { { [1 x { i32 }] } }
982 %a2 = alloca { {}, { float }, [0 x i8] }
983 %a3 = alloca { [0 x i8], { [0 x double], [1 x [1 x <4 x i8>]], {} }, { { {} } } }
986 %wrap1 = insertvalue [1 x { i32 }] undef, i32 %x, 0, 0
987 %gep1 = getelementptr { { [1 x { i32 }] } }* %a1, i32 0, i32 0, i32 0
988 store [1 x { i32 }] %wrap1, [1 x { i32 }]* %gep1
990 %gep2 = getelementptr { { [1 x { i32 }] } }* %a1, i32 0, i32 0
991 %ptrcast1 = bitcast { [1 x { i32 }] }* %gep2 to { [1 x { float }] }*
992 %load1 = load { [1 x { float }] }* %ptrcast1
993 %unwrap1 = extractvalue { [1 x { float }] } %load1, 0, 0
995 %wrap2 = insertvalue { {}, { float }, [0 x i8] } undef, { float } %unwrap1, 1
996 store { {}, { float }, [0 x i8] } %wrap2, { {}, { float }, [0 x i8] }* %a2
998 %gep3 = getelementptr { {}, { float }, [0 x i8] }* %a2, i32 0, i32 1, i32 0
999 %ptrcast2 = bitcast float* %gep3 to <4 x i8>*
1000 %load3 = load <4 x i8>* %ptrcast2
1001 %valcast1 = bitcast <4 x i8> %load3 to i32
1003 %wrap3 = insertvalue [1 x [1 x i32]] undef, i32 %valcast1, 0, 0
1004 %wrap4 = insertvalue { [1 x [1 x i32]], {} } undef, [1 x [1 x i32]] %wrap3, 0
1005 %gep4 = getelementptr { [0 x i8], { [0 x double], [1 x [1 x <4 x i8>]], {} }, { { {} } } }* %a3, i32 0, i32 1
1006 %ptrcast3 = bitcast { [0 x double], [1 x [1 x <4 x i8>]], {} }* %gep4 to { [1 x [1 x i32]], {} }*
1007 store { [1 x [1 x i32]], {} } %wrap4, { [1 x [1 x i32]], {} }* %ptrcast3
1009 %gep5 = getelementptr { [0 x i8], { [0 x double], [1 x [1 x <4 x i8>]], {} }, { { {} } } }* %a3, i32 0, i32 1, i32 1, i32 0
1010 %ptrcast4 = bitcast [1 x <4 x i8>]* %gep5 to { {}, float, {} }*
1011 %load4 = load { {}, float, {} }* %ptrcast4
1012 %unwrap2 = extractvalue { {}, float, {} } %load4, 1
1013 %valcast2 = bitcast float %unwrap2 to i32
1019 define void @PR14059.1(double* %d) {
1020 ; In PR14059 a peculiar construct was identified as something that is used
1021 ; pervasively in ARM's ABI-calling-convention lowering: the passing of a struct
1022 ; of doubles via an array of i32 in order to place the data into integer
1023 ; registers. This in turn was missed as an optimization by SROA due to the
1024 ; partial loads and stores of integers to the double alloca we were trying to
1025 ; form and promote. The solution is to widen the integer operations to be
1026 ; whole-alloca operations, and perform the appropriate bitcasting on the
1027 ; *values* rather than the pointers. When this works, partial reads and writes
1028 ; via integers can be promoted away.
1034 %X.sroa.0.i = alloca double, align 8
1035 %0 = bitcast double* %X.sroa.0.i to i8*
1036 call void @llvm.lifetime.start(i64 -1, i8* %0)
1038 ; Store to the low 32-bits...
1039 %X.sroa.0.0.cast2.i = bitcast double* %X.sroa.0.i to i32*
1040 store i32 0, i32* %X.sroa.0.0.cast2.i, align 8
1042 ; Also use a memset to the middle 32-bits for fun.
1043 %X.sroa.0.2.raw_idx2.i = getelementptr inbounds i8* %0, i32 2
1044 call void @llvm.memset.p0i8.i64(i8* %X.sroa.0.2.raw_idx2.i, i8 0, i64 4, i32 1, i1 false)
1046 ; Or a memset of the whole thing.
1047 call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 8, i32 1, i1 false)
1049 ; Store to the high 32-bits...
1050 %X.sroa.0.4.raw_idx4.i = getelementptr inbounds i8* %0, i32 4
1051 %X.sroa.0.4.cast5.i = bitcast i8* %X.sroa.0.4.raw_idx4.i to i32*
1052 store i32 1072693248, i32* %X.sroa.0.4.cast5.i, align 4
1054 ; Do the actual math...
1055 %X.sroa.0.0.load1.i = load double* %X.sroa.0.i, align 8
1056 %accum.real.i = load double* %d, align 8
1057 %add.r.i = fadd double %accum.real.i, %X.sroa.0.0.load1.i
1058 store double %add.r.i, double* %d, align 8
1059 call void @llvm.lifetime.end(i64 -1, i8* %0)