xref: /llvm-project/llvm/test/Transforms/SLPVectorizer/X86/memory-runtime-checks.ll (revision 9bf6365237f3a8a401afc0a69d2fb6d1b809ce68) 
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -aa-pipeline='basic-aa,scoped-noalias-aa' -passes=slp-vectorizer -mtriple=x86_64-apple-darwin -S %s | FileCheck %s
3
4define void @version_multiple(ptr nocapture %out_block, ptr nocapture readonly %counter) {
5; CHECK-LABEL: @version_multiple(
6; CHECK-NEXT:  entry:
7; CHECK-NEXT:    [[TMP0:%.*]] = load i32, ptr [[COUNTER:%.*]], align 4
8; CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr [[OUT_BLOCK:%.*]], align 4
9; CHECK-NEXT:    [[XOR:%.*]] = xor i32 [[TMP1]], [[TMP0]]
10; CHECK-NEXT:    store i32 [[XOR]], ptr [[OUT_BLOCK]], align 4
11; CHECK-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, ptr [[COUNTER]], i64 1
12; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr [[ARRAYIDX_1]], align 4
13; CHECK-NEXT:    [[ARRAYIDX2_1:%.*]] = getelementptr inbounds i32, ptr [[OUT_BLOCK]], i64 1
14; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[ARRAYIDX2_1]], align 4
15; CHECK-NEXT:    [[XOR_1:%.*]] = xor i32 [[TMP3]], [[TMP2]]
16; CHECK-NEXT:    store i32 [[XOR_1]], ptr [[ARRAYIDX2_1]], align 4
17; CHECK-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, ptr [[COUNTER]], i64 2
18; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr [[ARRAYIDX_2]], align 4
19; CHECK-NEXT:    [[ARRAYIDX2_2:%.*]] = getelementptr inbounds i32, ptr [[OUT_BLOCK]], i64 2
20; CHECK-NEXT:    [[TMP5:%.*]] = load i32, ptr [[ARRAYIDX2_2]], align 4
21; CHECK-NEXT:    [[XOR_2:%.*]] = xor i32 [[TMP5]], [[TMP4]]
22; CHECK-NEXT:    store i32 [[XOR_2]], ptr [[ARRAYIDX2_2]], align 4
23; CHECK-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, ptr [[COUNTER]], i64 3
24; CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[ARRAYIDX_3]], align 4
25; CHECK-NEXT:    [[ARRAYIDX2_3:%.*]] = getelementptr inbounds i32, ptr [[OUT_BLOCK]], i64 3
26; CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[ARRAYIDX2_3]], align 4
27; CHECK-NEXT:    [[XOR_3:%.*]] = xor i32 [[TMP7]], [[TMP6]]
28; CHECK-NEXT:    store i32 [[XOR_3]], ptr [[ARRAYIDX2_3]], align 4
29; CHECK-NEXT:    ret void
30;
31entry:
32  %0 = load i32, ptr %counter, align 4
33  %1 = load i32, ptr %out_block, align 4
34  %xor = xor i32 %1, %0
35  store i32 %xor, ptr %out_block, align 4
36  %arrayidx.1 = getelementptr inbounds i32, ptr %counter, i64 1
37  %2 = load i32, ptr %arrayidx.1, align 4
38  %arrayidx2.1 = getelementptr inbounds i32, ptr %out_block, i64 1
39  %3 = load i32, ptr %arrayidx2.1, align 4
40  %xor.1 = xor i32 %3, %2
41  store i32 %xor.1, ptr %arrayidx2.1, align 4
42  %arrayidx.2 = getelementptr inbounds i32, ptr %counter, i64 2
43  %4 = load i32, ptr %arrayidx.2, align 4
44  %arrayidx2.2 = getelementptr inbounds i32, ptr %out_block, i64 2
45  %5 = load i32, ptr %arrayidx2.2, align 4
46  %xor.2 = xor i32 %5, %4
47  store i32 %xor.2, ptr %arrayidx2.2, align 4
48  %arrayidx.3 = getelementptr inbounds i32, ptr %counter, i64 3
49  %6 = load i32, ptr %arrayidx.3, align 4
50  %arrayidx2.3 = getelementptr inbounds i32, ptr %out_block, i64 3
51  %7 = load i32, ptr %arrayidx2.3, align 4
52  %xor.3 = xor i32 %7, %6
53  store i32 %xor.3, ptr %arrayidx2.3, align 4
54  ret void
55}
56
57declare void @use(<8 x float>)
58define void @delete_pointer_bound(ptr %a, ptr %b, i1 %c) #0 {
59; CHECK-LABEL: @delete_pointer_bound(
60; CHECK-NEXT:  entry:
61; CHECK-NEXT:    [[B_10:%.*]] = getelementptr inbounds float, ptr [[B:%.*]], i64 10
62; CHECK-NEXT:    [[B_14:%.*]] = getelementptr inbounds float, ptr [[B]], i64 14
63; CHECK-NEXT:    br i1 [[C:%.*]], label [[THEN:%.*]], label [[ELSE:%.*]]
64; CHECK:       else:
65; CHECK-NEXT:    [[L0:%.*]] = load float, ptr [[B_10]], align 4
66; CHECK-NEXT:    [[L1:%.*]] = load float, ptr [[B_14]], align 4
67; CHECK-NEXT:    [[I2:%.*]] = insertelement <8 x float> undef, float [[L0]], i32 2
68; CHECK-NEXT:    [[I3:%.*]] = insertelement <8 x float> [[I2]], float [[L0]], i32 3
69; CHECK-NEXT:    [[I4:%.*]] = insertelement <8 x float> [[I3]], float [[L1]], i32 4
70; CHECK-NEXT:    [[I7:%.*]] = insertelement <8 x float> [[I4]], float [[L1]], i32 7
71; CHECK-NEXT:    call void @use(<8 x float> [[I7]])
72; CHECK-NEXT:    ret void
73; CHECK:       then:
74; CHECK-NEXT:    [[A_8:%.*]] = getelementptr inbounds float, ptr [[A:%.*]], i64 8
75; CHECK-NEXT:    store float 0.000000e+00, ptr [[A_8]], align 4
76; CHECK-NEXT:    [[L6:%.*]] = load float, ptr [[B_14]], align 4
77; CHECK-NEXT:    [[A_5:%.*]] = getelementptr inbounds float, ptr [[A]], i64 5
78; CHECK-NEXT:    store float [[L6]], ptr [[A_5]], align 4
79; CHECK-NEXT:    [[A_6:%.*]] = getelementptr inbounds float, ptr [[A]], i64 6
80; CHECK-NEXT:    store <2 x float> zeroinitializer, ptr [[A_6]], align 4
81; CHECK-NEXT:    ret void
82;
83entry:
84  %b.10 = getelementptr inbounds float, ptr %b, i64 10
85  %b.14 = getelementptr inbounds float, ptr %b, i64 14
86  br i1 %c, label %then, label %else
87
88else:
89  %l0 = load float, ptr %b.10, align 4
90  %l1 = load float, ptr %b.14, align 4
91  %i2 = insertelement <8 x float> undef, float %l0, i32 2
92  %i3 = insertelement <8 x float> %i2, float %l0, i32 3
93  %i4 = insertelement <8 x float> %i3, float %l1, i32 4
94  %i7 = insertelement <8 x float> %i4, float %l1, i32 7
95  call void @use(<8 x float> %i7)
96  ret void
97
98then:
99  %a.8 = getelementptr inbounds float, ptr %a, i64 8
100  store float 0.0, ptr %a.8, align 4
101  %l6 = load float, ptr %b.14, align 4
102  %a.5 = getelementptr inbounds float, ptr %a, i64 5
103  store float %l6, ptr %a.5, align 4
104  %a.6 = getelementptr inbounds float, ptr %a, i64 6
105  store float 0.0, ptr %a.6, align 4
106  %a.7 = getelementptr inbounds float, ptr %a, i64 7
107  store float 0.0, ptr %a.7, align 4
108  ret void
109}
110
111%struct.zot = type { i16, i16, i16, i32, float, float, float, ptr, ptr, ptr, %struct.wombat.0 }
112%struct.quux = type { i16, ptr, ptr }
113%struct.wombat = type { i32, i16, i8, i8, ptr }
114%struct.eggs = type { float, i8, %struct.ham }
115%struct.ham = type { [2 x double], [8 x i8] }
116%struct.wombat.0 = type { %struct.bar }
117%struct.bar = type { [3 x double], [3 x double], double, double, i16, ptr, i32, [3 x double] }
118
119define double @preserve_loop_info(ptr %arg, i1 %arg2) {
120; CHECK-LABEL: @preserve_loop_info(
121; CHECK-NEXT:  entry:
122; CHECK-NEXT:    [[TMP:%.*]] = alloca [3 x double], align 16
123; CHECK-NEXT:    br label [[OUTER_HEADER:%.*]]
124; CHECK:       outer.header:
125; CHECK-NEXT:    br label [[INNER:%.*]]
126; CHECK:       inner:
127; CHECK-NEXT:    br i1 %arg2, label [[OUTER_LATCH:%.*]], label [[INNER]]
128; CHECK:       outer.latch:
129; CHECK-NEXT:    br i1 %arg2, label [[BB:%.*]], label [[OUTER_HEADER]]
130; CHECK:       bb:
131; CHECK-NEXT:    [[TMP5:%.*]] = load ptr, ptr undef, align 8
132; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [3 x double], ptr [[TMP]], i64 0, i64 1
133; CHECK-NEXT:    br label [[LOOP_3HEADER:%.*]]
134; CHECK:       loop.3header:
135; CHECK-NEXT:    br i1 %arg2, label [[LOOP_3LATCH:%.*]], label [[BB9:%.*]]
136; CHECK:       bb9:
137; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [3 x double], ptr [[TMP5]], i64 undef, i64 1
138; CHECK-NEXT:    store double undef, ptr [[TMP]], align 16
139; CHECK-NEXT:    [[TMP12:%.*]] = load double, ptr [[TMP10]], align 8
140; CHECK-NEXT:    store double [[TMP12]], ptr [[TMP7]], align 8
141; CHECK-NEXT:    br label [[LOOP_3LATCH]]
142; CHECK:       loop.3latch:
143; CHECK-NEXT:    br i1 %arg2, label [[BB14:%.*]], label [[LOOP_3HEADER]]
144; CHECK:       bb14:
145; CHECK-NEXT:    [[TMP15:%.*]] = call double undef(ptr [[TMP]], ptr [[ARG:%.*]])
146; CHECK-NEXT:    ret double undef
147;
148entry:
149  %tmp = alloca [3 x double], align 16
150  br label %outer.header
151
152outer.header:                                              ; preds = %bb3, %bb
153  br label %inner
154
155inner:
156  br i1 %arg2, label %outer.latch, label %inner
157
158outer.latch:                                              ; preds = %bb16
159  br i1 %arg2, label %bb, label %outer.header
160
161bb:                                              ; preds = %bb3
162  %tmp5 = load ptr, ptr undef, align 8
163  %tmp7 = getelementptr inbounds [3 x double], ptr %tmp, i64 0, i64 1
164  br label %loop.3header
165
166loop.3header:                                              ; preds = %bb13, %bb4
167  br i1 %arg2, label %loop.3latch, label %bb9
168
169bb9:                                              ; preds = %bb8
170  %tmp10 = getelementptr inbounds [3 x double], ptr %tmp5, i64 undef, i64 1
171  store double undef, ptr %tmp, align 16
172  %tmp12 = load double, ptr %tmp10, align 8
173  store double %tmp12, ptr %tmp7, align 8
174  br label %loop.3latch
175
176loop.3latch:                                             ; preds = %bb11, %bb8
177  br i1 %arg2, label %bb14, label %loop.3header
178
179bb14:                                             ; preds = %bb13
180  %tmp15 = call double undef(ptr %tmp, ptr %arg)
181  ret double undef
182}
183
184define void @gather_sequence_crash(<2 x float> %arg, ptr %arg1, float %arg2, ptr %arg3, ptr %arg4, ptr %arg5, i1 %c.1, i1 %c.2) {
185; CHECK-LABEL: @gather_sequence_crash(
186; CHECK-NEXT:  bb:
187; CHECK-NEXT:    br i1 [[C_1:%.*]], label [[BB16:%.*]], label [[BB6:%.*]]
188; CHECK:       bb6:
189; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds float, ptr [[ARG1:%.*]], i32 3
190; CHECK-NEXT:    [[TMP0:%.*]] = shufflevector <2 x float> [[ARG:%.*]], <2 x float> poison, <4 x i32> <i32 poison, i32 0, i32 1, i32 poison>
191; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <4 x float> [[TMP0]], <4 x float> <float poison, float poison, float poison, float 0.000000e+00>, <4 x i32> <i32 poison, i32 1, i32 2, i32 7>
192; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <4 x float> [[TMP1]], float [[ARG2:%.*]], i32 0
193; CHECK-NEXT:    [[TMP3:%.*]] = fmul <4 x float> [[TMP2]], zeroinitializer
194; CHECK-NEXT:    store <4 x float> [[TMP3]], ptr [[TMP8]], align 4
195; CHECK-NEXT:    ret void
196; CHECK:       bb16:
197; CHECK-NEXT:    br label [[BB17:%.*]]
198; CHECK:       bb17:
199; CHECK-NEXT:    br label [[BB18:%.*]]
200; CHECK:       bb18:
201; CHECK-NEXT:    br label [[BB19:%.*]]
202; CHECK:       bb19:
203; CHECK-NEXT:    br label [[BB20:%.*]]
204; CHECK:       bb20:
205; CHECK-NEXT:    br label [[BB21:%.*]]
206; CHECK:       bb21:
207; CHECK-NEXT:    br label [[BB22:%.*]]
208; CHECK:       bb22:
209; CHECK-NEXT:    [[TMP24:%.*]] = getelementptr float, ptr [[ARG4:%.*]], i64 7
210; CHECK-NEXT:    br i1 [[C_2:%.*]], label [[BB25:%.*]], label [[BB22]]
211; CHECK:       bb25:
212; CHECK-NEXT:    [[TMP26:%.*]] = getelementptr float, ptr [[ARG4]], i64 6
213; CHECK-NEXT:    store float 0.000000e+00, ptr [[TMP24]], align 4
214; CHECK-NEXT:    [[TMP27:%.*]] = load float, ptr [[ARG5:%.*]], align 4
215; CHECK-NEXT:    [[TMP29:%.*]] = fadd float 0.000000e+00, 0.000000e+00
216; CHECK-NEXT:    store float 0.000000e+00, ptr [[TMP26]], align 4
217; CHECK-NEXT:    [[TMP30:%.*]] = getelementptr float, ptr [[ARG4]], i64 4
218; CHECK-NEXT:    [[TMP31:%.*]] = fadd float 0.000000e+00, 0.000000e+00
219; CHECK-NEXT:    store <2 x float> zeroinitializer, ptr [[TMP30]], align 4
220; CHECK-NEXT:    br label [[BB33:%.*]]
221; CHECK:       bb33:
222; CHECK-NEXT:    br label [[BB34:%.*]]
223; CHECK:       bb34:
224; CHECK-NEXT:    [[TMP35:%.*]] = getelementptr float, ptr [[ARG4]], i64 3
225; CHECK-NEXT:    [[TMP37:%.*]] = load float, ptr [[TMP35]], align 4
226; CHECK-NEXT:    [[TMP38:%.*]] = fadd float 0.000000e+00, [[TMP37]]
227; CHECK-NEXT:    store float [[TMP38]], ptr [[TMP35]], align 4
228; CHECK-NEXT:    [[TMP39:%.*]] = getelementptr float, ptr [[ARG4]], i64 1
229; CHECK-NEXT:    [[TMP4:%.*]] = load <2 x float>, ptr [[TMP39]], align 4
230; CHECK-NEXT:    [[TMP5:%.*]] = fadd <2 x float> zeroinitializer, [[TMP4]]
231; CHECK-NEXT:    store <2 x float> [[TMP5]], ptr [[TMP39]], align 4
232; CHECK-NEXT:    [[TMP44:%.*]] = load float, ptr [[ARG3:%.*]], align 4
233; CHECK-NEXT:    [[TMP45:%.*]] = load float, ptr [[ARG4]], align 4
234; CHECK-NEXT:    [[TMP46:%.*]] = fadd float 0.000000e+00, [[TMP45]]
235; CHECK-NEXT:    store float [[TMP46]], ptr [[ARG4]], align 4
236; CHECK-NEXT:    call void @quux()
237; CHECK-NEXT:    br label [[BB47:%.*]]
238; CHECK:       bb47:
239; CHECK-NEXT:    br label [[BB17]]
240;
241bb:
242  br i1 %c.1, label %bb16, label %bb6
243
244bb6:                                              ; preds = %bb
245  %tmp = getelementptr inbounds float, ptr %arg1, i32 4
246  %tmp7 = getelementptr inbounds float, ptr %arg1, i32 5
247  %tmp8 = getelementptr inbounds float, ptr %arg1, i32 3
248  %tmp9 = getelementptr inbounds float, ptr %arg1, i32 6
249  %tmp10 = extractelement <2 x float> %arg, i32 0
250  %tmp11 = fmul float %tmp10, 0.000000e+00
251  store float %tmp11, ptr %tmp, align 4
252  %tmp12 = extractelement <2 x float> %arg, i32 1
253  %tmp13 = fmul float %tmp12, 0.000000e+00
254  store float %tmp13, ptr %tmp7, align 4
255  %tmp14 = fmul float %arg2, 0.000000e+00
256  store float %tmp14, ptr %tmp8, align 4
257  %tmp15 = fmul float 0.000000e+00, 0.000000e+00
258  store float %tmp15, ptr %tmp9, align 4
259  ret void
260
261bb16:                                             ; preds = %bb
262  br label %bb17
263
264bb17:                                             ; preds = %bb47, %bb16
265  br label %bb18
266
267bb18:                                             ; preds = %bb17
268  br label %bb19
269
270bb19:                                             ; preds = %bb18
271  br label %bb20
272
273bb20:                                             ; preds = %bb19
274  br label %bb21
275
276bb21:                                             ; preds = %bb20
277  br label %bb22
278
279bb22:                                             ; preds = %bb22, %bb21
280  %tmp24 = getelementptr float, ptr %arg4, i64 7
281  br i1 %c.2, label %bb25, label %bb22
282
283bb25:                                             ; preds = %bb22
284  %tmp26 = getelementptr float, ptr %arg4, i64 6
285  store float 0.000000e+00, ptr %tmp24, align 4
286  %tmp27 = load float, ptr %arg5, align 4
287  %tmp28 = getelementptr float, ptr %arg4, i64 5
288  %tmp29 = fadd float 0.000000e+00, 0.000000e+00
289  store float 0.000000e+00, ptr %tmp26, align 4
290  %tmp30 = getelementptr float, ptr %arg4, i64 4
291  store float 0.000000e+00, ptr %tmp28, align 4
292  %tmp31 = fadd float 0.000000e+00, 0.000000e+00
293  store float 0.000000e+00, ptr %tmp30, align 4
294  br label %bb33
295
296bb33:                                             ; preds = %bb25
297  br label %bb34
298
299bb34:                                             ; preds = %bb33
300  %tmp35 = getelementptr float, ptr %arg4, i64 3
301  %tmp36 = getelementptr float, ptr %arg4, i64 2
302  %tmp37 = load float, ptr %tmp35, align 4
303  %tmp38 = fadd float 0.000000e+00, %tmp37
304  store float %tmp38, ptr %tmp35, align 4
305  %tmp39 = getelementptr float, ptr %arg4, i64 1
306  %tmp40 = load float, ptr %tmp36, align 4
307  %tmp41 = fadd float 0.000000e+00, %tmp40
308  store float %tmp41, ptr %tmp36, align 4
309  %tmp42 = load float, ptr %tmp39, align 4
310  %tmp43 = fadd float 0.000000e+00, %tmp42
311  store float %tmp43, ptr %tmp39, align 4
312  %tmp44 = load float, ptr %arg3, align 4
313  %tmp45 = load float, ptr %arg4, align 4
314  %tmp46 = fadd float 0.000000e+00, %tmp45
315  store float %tmp46, ptr %arg4, align 4
316  call void @quux()
317  br label %bb47
318
319bb47:                                             ; preds = %bb34
320  br label %bb17
321}
322
323declare void @quux()
324attributes #0 = { "target-features"="+avx2" }
325