1 // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --function-signature --include-generated-funcs
2 // RUN: %clang_cc1 -no-opaque-pointers -fopenmp-enable-irbuilder -verify -fopenmp -fopenmp-version=51 -x c -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck %s
3 // expected-no-diagnostics
4 
5 #ifndef HEADER
6 #define HEADER
7 
8 // CHECK-LABEL: define {{.*}}@unroll_partial_factor_for(
9 // CHECK-NEXT:  [[ENTRY:.*]]:
10 // CHECK-NEXT:    %[[A_ADDR:.+]] = alloca float*, align 8
11 // CHECK-NEXT:    %[[B_ADDR:.+]] = alloca float*, align 8
12 // CHECK-NEXT:    %[[C_ADDR:.+]] = alloca float*, align 8
13 // CHECK-NEXT:    %[[D_ADDR:.+]] = alloca float*, align 8
14 // CHECK-NEXT:    %[[I:.+]] = alloca i32, align 4
15 // CHECK-NEXT:    %[[AGG_CAPTURED:.+]] = alloca %struct.anon, align 8
16 // CHECK-NEXT:    %[[AGG_CAPTURED1:.+]] = alloca %struct.anon.0, align 4
17 // CHECK-NEXT:    %[[DOTCOUNT_ADDR:.+]] = alloca i32, align 4
18 // CHECK-NEXT:    %[[P_LASTITER:.+]] = alloca i32, align 4
19 // CHECK-NEXT:    %[[P_LOWERBOUND:.+]] = alloca i32, align 4
20 // CHECK-NEXT:    %[[P_UPPERBOUND:.+]] = alloca i32, align 4
21 // CHECK-NEXT:    %[[P_STRIDE:.+]] = alloca i32, align 4
22 // CHECK-NEXT:    store float* %[[A:.+]], float** %[[A_ADDR]], align 8
23 // CHECK-NEXT:    store float* %[[B:.+]], float** %[[B_ADDR]], align 8
24 // CHECK-NEXT:    store float* %[[C:.+]], float** %[[C_ADDR]], align 8
25 // CHECK-NEXT:    store float* %[[D:.+]], float** %[[D_ADDR]], align 8
26 // CHECK-NEXT:    store i32 0, i32* %[[I]], align 4
27 // CHECK-NEXT:    %[[TMP0:.+]] = getelementptr inbounds %struct.anon, %struct.anon* %[[AGG_CAPTURED]], i32 0, i32 0
28 // CHECK-NEXT:    store i32* %[[I]], i32** %[[TMP0]], align 8
29 // CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon.0, %struct.anon.0* %[[AGG_CAPTURED1]], i32 0, i32 0
30 // CHECK-NEXT:    %[[TMP2:.+]] = load i32, i32* %[[I]], align 4
31 // CHECK-NEXT:    store i32 %[[TMP2]], i32* %[[TMP1]], align 4
32 // CHECK-NEXT:    call void @__captured_stmt(i32* %[[DOTCOUNT_ADDR]], %struct.anon* %[[AGG_CAPTURED]])
33 // CHECK-NEXT:    %[[DOTCOUNT:.+]] = load i32, i32* %[[DOTCOUNT_ADDR]], align 4
34 // CHECK-NEXT:    br label %[[OMP_LOOP_PREHEADER:.+]]
35 // CHECK-EMPTY:
36 // CHECK-NEXT:  [[OMP_LOOP_PREHEADER]]:
37 // CHECK-NEXT:    %[[TMP3:.+]] = udiv i32 %[[DOTCOUNT]], 2
38 // CHECK-NEXT:    %[[TMP4:.+]] = urem i32 %[[DOTCOUNT]], 2
39 // CHECK-NEXT:    %[[TMP5:.+]] = icmp ne i32 %[[TMP4]], 0
40 // CHECK-NEXT:    %[[TMP6:.+]] = zext i1 %[[TMP5]] to i32
41 // CHECK-NEXT:    %[[OMP_FLOOR0_TRIPCOUNT:.+]] = add nuw i32 %[[TMP3]], %[[TMP6]]
42 // CHECK-NEXT:    br label %[[OMP_FLOOR0_PREHEADER:.+]]
43 // CHECK-EMPTY:
44 // CHECK-NEXT:  [[OMP_FLOOR0_PREHEADER]]:
45 // CHECK-NEXT:    store i32 0, i32* %[[P_LOWERBOUND]], align 4
46 // CHECK-NEXT:    %[[TMP7:.+]] = sub i32 %[[OMP_FLOOR0_TRIPCOUNT]], 1
47 // CHECK-NEXT:    store i32 %[[TMP7]], i32* %[[P_UPPERBOUND]], align 4
48 // CHECK-NEXT:    store i32 1, i32* %[[P_STRIDE]], align 4
49 // CHECK-NEXT:    %[[OMP_GLOBAL_THREAD_NUM:.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @1)
50 // CHECK-NEXT:    call void @__kmpc_for_static_init_4u(%struct.ident_t* @1, i32 %[[OMP_GLOBAL_THREAD_NUM]], i32 34, i32* %[[P_LASTITER]], i32* %[[P_LOWERBOUND]], i32* %[[P_UPPERBOUND]], i32* %[[P_STRIDE]], i32 1, i32 0)
51 // CHECK-NEXT:    %[[TMP8:.+]] = load i32, i32* %[[P_LOWERBOUND]], align 4
52 // CHECK-NEXT:    %[[TMP9:.+]] = load i32, i32* %[[P_UPPERBOUND]], align 4
53 // CHECK-NEXT:    %[[TMP10:.+]] = sub i32 %[[TMP9]], %[[TMP8]]
54 // CHECK-NEXT:    %[[TMP11:.+]] = add i32 %[[TMP10]], 1
55 // CHECK-NEXT:    br label %[[OMP_FLOOR0_HEADER:.+]]
56 // CHECK-EMPTY:
57 // CHECK-NEXT:  [[OMP_FLOOR0_HEADER]]:
58 // CHECK-NEXT:    %[[OMP_FLOOR0_IV:.+]] = phi i32 [ 0, %[[OMP_FLOOR0_PREHEADER]] ], [ %[[OMP_FLOOR0_NEXT:.+]], %[[OMP_FLOOR0_INC:.+]] ]
59 // CHECK-NEXT:    br label %[[OMP_FLOOR0_COND:.+]]
60 // CHECK-EMPTY:
61 // CHECK-NEXT:  [[OMP_FLOOR0_COND]]:
62 // CHECK-NEXT:    %[[OMP_FLOOR0_CMP:.+]] = icmp ult i32 %[[OMP_FLOOR0_IV]], %[[TMP11]]
63 // CHECK-NEXT:    br i1 %[[OMP_FLOOR0_CMP]], label %[[OMP_FLOOR0_BODY:.+]], label %[[OMP_FLOOR0_EXIT:.+]]
64 // CHECK-EMPTY:
65 // CHECK-NEXT:  [[OMP_FLOOR0_BODY]]:
66 // CHECK-NEXT:    %[[TMP12:.+]] = add i32 %[[OMP_FLOOR0_IV]], %[[TMP8]]
67 // CHECK-NEXT:    %[[TMP13:.+]] = icmp eq i32 %[[TMP12]], %[[OMP_FLOOR0_TRIPCOUNT]]
68 // CHECK-NEXT:    %[[TMP14:.+]] = select i1 %[[TMP13]], i32 %[[TMP4]], i32 2
69 // CHECK-NEXT:    br label %[[OMP_TILE0_PREHEADER:.+]]
70 // CHECK-EMPTY:
71 // CHECK-NEXT:  [[OMP_TILE0_PREHEADER]]:
72 // CHECK-NEXT:    br label %[[OMP_TILE0_HEADER:.+]]
73 // CHECK-EMPTY:
74 // CHECK-NEXT:  [[OMP_TILE0_HEADER]]:
75 // CHECK-NEXT:    %[[OMP_TILE0_IV:.+]] = phi i32 [ 0, %[[OMP_TILE0_PREHEADER]] ], [ %[[OMP_TILE0_NEXT:.+]], %[[OMP_TILE0_INC:.+]] ]
76 // CHECK-NEXT:    br label %[[OMP_TILE0_COND:.+]]
77 // CHECK-EMPTY:
78 // CHECK-NEXT:  [[OMP_TILE0_COND]]:
79 // CHECK-NEXT:    %[[OMP_TILE0_CMP:.+]] = icmp ult i32 %[[OMP_TILE0_IV]], %[[TMP14]]
80 // CHECK-NEXT:    br i1 %[[OMP_TILE0_CMP]], label %[[OMP_TILE0_BODY:.+]], label %[[OMP_TILE0_EXIT:.+]]
81 // CHECK-EMPTY:
82 // CHECK-NEXT:  [[OMP_TILE0_BODY]]:
83 // CHECK-NEXT:    %[[TMP15:.+]] = mul nuw i32 2, %[[TMP12]]
84 // CHECK-NEXT:    %[[TMP16:.+]] = add nuw i32 %[[TMP15]], %[[OMP_TILE0_IV]]
85 // CHECK-NEXT:    br label %[[OMP_LOOP_BODY:.+]]
86 // CHECK-EMPTY:
87 // CHECK-NEXT:  [[OMP_LOOP_BODY]]:
88 // CHECK-NEXT:    call void @__captured_stmt.1(i32* %[[I]], i32 %[[TMP16]], %struct.anon.0* %[[AGG_CAPTURED1]])
89 // CHECK-NEXT:    %[[TMP17:.+]] = load float*, float** %[[B_ADDR]], align 8
90 // CHECK-NEXT:    %[[TMP18:.+]] = load i32, i32* %[[I]], align 4
91 // CHECK-NEXT:    %[[IDXPROM:.+]] = sext i32 %[[TMP18]] to i64
92 // CHECK-NEXT:    %[[ARRAYIDX:.+]] = getelementptr inbounds float, float* %[[TMP17]], i64 %[[IDXPROM]]
93 // CHECK-NEXT:    %[[TMP19:.+]] = load float, float* %[[ARRAYIDX]], align 4
94 // CHECK-NEXT:    %[[TMP20:.+]] = load float*, float** %[[C_ADDR]], align 8
95 // CHECK-NEXT:    %[[TMP21:.+]] = load i32, i32* %[[I]], align 4
96 // CHECK-NEXT:    %[[IDXPROM2:.+]] = sext i32 %[[TMP21]] to i64
97 // CHECK-NEXT:    %[[ARRAYIDX3:.+]] = getelementptr inbounds float, float* %[[TMP20]], i64 %[[IDXPROM2]]
98 // CHECK-NEXT:    %[[TMP22:.+]] = load float, float* %[[ARRAYIDX3]], align 4
99 // CHECK-NEXT:    %[[MUL:.+]] = fmul float %[[TMP19]], %[[TMP22]]
100 // CHECK-NEXT:    %[[TMP23:.+]] = load float*, float** %[[D_ADDR]], align 8
101 // CHECK-NEXT:    %[[TMP24:.+]] = load i32, i32* %[[I]], align 4
102 // CHECK-NEXT:    %[[IDXPROM4:.+]] = sext i32 %[[TMP24]] to i64
103 // CHECK-NEXT:    %[[ARRAYIDX5:.+]] = getelementptr inbounds float, float* %[[TMP23]], i64 %[[IDXPROM4]]
104 // CHECK-NEXT:    %[[TMP25:.+]] = load float, float* %[[ARRAYIDX5]], align 4
105 // CHECK-NEXT:    %[[MUL6:.+]] = fmul float %[[MUL]], %[[TMP25]]
106 // CHECK-NEXT:    %[[TMP26:.+]] = load float*, float** %[[A_ADDR]], align 8
107 // CHECK-NEXT:    %[[TMP27:.+]] = load i32, i32* %[[I]], align 4
108 // CHECK-NEXT:    %[[IDXPROM7:.+]] = sext i32 %[[TMP27]] to i64
109 // CHECK-NEXT:    %[[ARRAYIDX8:.+]] = getelementptr inbounds float, float* %[[TMP26]], i64 %[[IDXPROM7]]
110 // CHECK-NEXT:    store float %[[MUL6]], float* %[[ARRAYIDX8]], align 4
111 // CHECK-NEXT:    br label %[[OMP_TILE0_INC]]
112 // CHECK-EMPTY:
113 // CHECK-NEXT:  [[OMP_TILE0_INC]]:
114 // CHECK-NEXT:    %[[OMP_TILE0_NEXT]] = add nuw i32 %[[OMP_TILE0_IV]], 1
115 // CHECK-NEXT:    br label %[[OMP_TILE0_HEADER]], !llvm.loop ![[LOOP3:[0-9]+]]
116 // CHECK-EMPTY:
117 // CHECK-NEXT:  [[OMP_TILE0_EXIT]]:
118 // CHECK-NEXT:    br label %[[OMP_TILE0_AFTER:.+]]
119 // CHECK-EMPTY:
120 // CHECK-NEXT:  [[OMP_TILE0_AFTER]]:
121 // CHECK-NEXT:    br label %[[OMP_FLOOR0_INC]]
122 // CHECK-EMPTY:
123 // CHECK-NEXT:  [[OMP_FLOOR0_INC]]:
124 // CHECK-NEXT:    %[[OMP_FLOOR0_NEXT]] = add nuw i32 %[[OMP_FLOOR0_IV]], 1
125 // CHECK-NEXT:    br label %[[OMP_FLOOR0_HEADER]]
126 // CHECK-EMPTY:
127 // CHECK-NEXT:  [[OMP_FLOOR0_EXIT]]:
128 // CHECK-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @1, i32 %[[OMP_GLOBAL_THREAD_NUM]])
129 // CHECK-NEXT:    %[[OMP_GLOBAL_THREAD_NUM9:.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @1)
130 // CHECK-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @2, i32 %[[OMP_GLOBAL_THREAD_NUM9]])
131 // CHECK-NEXT:    br label %[[OMP_FLOOR0_AFTER:.+]]
132 // CHECK-EMPTY:
133 // CHECK-NEXT:  [[OMP_FLOOR0_AFTER]]:
134 // CHECK-NEXT:    br label %[[OMP_LOOP_AFTER:.+]]
135 // CHECK-EMPTY:
136 // CHECK-NEXT:  [[OMP_LOOP_AFTER]]:
137 // CHECK-NEXT:    ret void
138 // CHECK-NEXT:  }
139 
unroll_partial_factor_for(float * a,float * b,float * c,float * d)140 void unroll_partial_factor_for(float *a, float *b, float *c, float *d) {
141 #pragma omp for
142 #pragma omp unroll partial(2)
143   for (int i = 0; i < 2; i++) {
144     a[i] = b[i] * c[i] * d[i];
145   }
146 }
147 
148 #endif // HEADER
149 
150 // CHECK-LABEL: define {{.*}}@__captured_stmt(
151 // CHECK-NEXT:  [[ENTRY:.*]]:
152 // CHECK-NEXT:    %[[DISTANCE_ADDR:.+]] = alloca i32*, align 8
153 // CHECK-NEXT:    %[[__CONTEXT_ADDR:.+]] = alloca %struct.anon*, align 8
154 // CHECK-NEXT:    %[[DOTSTART:.+]] = alloca i32, align 4
155 // CHECK-NEXT:    %[[DOTSTOP:.+]] = alloca i32, align 4
156 // CHECK-NEXT:    %[[DOTSTEP:.+]] = alloca i32, align 4
157 // CHECK-NEXT:    store i32* %[[DISTANCE:.+]], i32** %[[DISTANCE_ADDR]], align 8
158 // CHECK-NEXT:    store %struct.anon* %[[__CONTEXT:.+]], %struct.anon** %[[__CONTEXT_ADDR]], align 8
159 // CHECK-NEXT:    %[[TMP0:.+]] = load %struct.anon*, %struct.anon** %[[__CONTEXT_ADDR]], align 8
160 // CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon, %struct.anon* %[[TMP0]], i32 0, i32 0
161 // CHECK-NEXT:    %[[TMP2:.+]] = load i32*, i32** %[[TMP1]], align 8
162 // CHECK-NEXT:    %[[TMP3:.+]] = load i32, i32* %[[TMP2]], align 4
163 // CHECK-NEXT:    store i32 %[[TMP3]], i32* %[[DOTSTART]], align 4
164 // CHECK-NEXT:    store i32 2, i32* %[[DOTSTOP]], align 4
165 // CHECK-NEXT:    store i32 1, i32* %[[DOTSTEP]], align 4
166 // CHECK-NEXT:    %[[TMP4:.+]] = load i32, i32* %[[DOTSTART]], align 4
167 // CHECK-NEXT:    %[[TMP5:.+]] = load i32, i32* %[[DOTSTOP]], align 4
168 // CHECK-NEXT:    %[[CMP:.+]] = icmp slt i32 %[[TMP4]], %[[TMP5]]
169 // CHECK-NEXT:    br i1 %[[CMP]], label %[[COND_TRUE:.+]], label %[[COND_FALSE:.+]]
170 // CHECK-EMPTY:
171 // CHECK-NEXT:  [[COND_TRUE]]:
172 // CHECK-NEXT:    %[[TMP6:.+]] = load i32, i32* %[[DOTSTOP]], align 4
173 // CHECK-NEXT:    %[[TMP7:.+]] = load i32, i32* %[[DOTSTART]], align 4
174 // CHECK-NEXT:    %[[SUB:.+]] = sub nsw i32 %[[TMP6]], %[[TMP7]]
175 // CHECK-NEXT:    %[[TMP8:.+]] = load i32, i32* %[[DOTSTEP]], align 4
176 // CHECK-NEXT:    %[[SUB1:.+]] = sub i32 %[[TMP8]], 1
177 // CHECK-NEXT:    %[[ADD:.+]] = add i32 %[[SUB]], %[[SUB1]]
178 // CHECK-NEXT:    %[[TMP9:.+]] = load i32, i32* %[[DOTSTEP]], align 4
179 // CHECK-NEXT:    %[[DIV:.+]] = udiv i32 %[[ADD]], %[[TMP9]]
180 // CHECK-NEXT:    br label %[[COND_END:.+]]
181 // CHECK-EMPTY:
182 // CHECK-NEXT:  [[COND_FALSE]]:
183 // CHECK-NEXT:    br label %[[COND_END]]
184 // CHECK-EMPTY:
185 // CHECK-NEXT:  [[COND_END]]:
186 // CHECK-NEXT:    %[[COND:.+]] = phi i32 [ %[[DIV]], %[[COND_TRUE]] ], [ 0, %[[COND_FALSE]] ]
187 // CHECK-NEXT:    %[[TMP10:.+]] = load i32*, i32** %[[DISTANCE_ADDR]], align 8
188 // CHECK-NEXT:    store i32 %[[COND]], i32* %[[TMP10]], align 4
189 // CHECK-NEXT:    ret void
190 // CHECK-NEXT:  }
191 
192 
193 // CHECK-LABEL: define {{.*}}@__captured_stmt.1(
194 // CHECK-NEXT:  [[ENTRY:.*]]:
195 // CHECK-NEXT:    %[[LOOPVAR_ADDR:.+]] = alloca i32*, align 8
196 // CHECK-NEXT:    %[[LOGICAL_ADDR:.+]] = alloca i32, align 4
197 // CHECK-NEXT:    %[[__CONTEXT_ADDR:.+]] = alloca %struct.anon.0*, align 8
198 // CHECK-NEXT:    store i32* %[[LOOPVAR:.+]], i32** %[[LOOPVAR_ADDR]], align 8
199 // CHECK-NEXT:    store i32 %[[LOGICAL:.+]], i32* %[[LOGICAL_ADDR]], align 4
200 // CHECK-NEXT:    store %struct.anon.0* %[[__CONTEXT:.+]], %struct.anon.0** %[[__CONTEXT_ADDR]], align 8
201 // CHECK-NEXT:    %[[TMP0:.+]] = load %struct.anon.0*, %struct.anon.0** %[[__CONTEXT_ADDR]], align 8
202 // CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon.0, %struct.anon.0* %[[TMP0]], i32 0, i32 0
203 // CHECK-NEXT:    %[[TMP2:.+]] = load i32, i32* %[[TMP1]], align 4
204 // CHECK-NEXT:    %[[TMP3:.+]] = load i32, i32* %[[LOGICAL_ADDR]], align 4
205 // CHECK-NEXT:    %[[MUL:.+]] = mul i32 1, %[[TMP3]]
206 // CHECK-NEXT:    %[[ADD:.+]] = add i32 %[[TMP2]], %[[MUL]]
207 // CHECK-NEXT:    %[[TMP4:.+]] = load i32*, i32** %[[LOOPVAR_ADDR]], align 8
208 // CHECK-NEXT:    store i32 %[[ADD]], i32* %[[TMP4]], align 4
209 // CHECK-NEXT:    ret void
210 // CHECK-NEXT:  }
211 
212 
213 // CHECK: ![[META0:[0-9]+]] = !{i32 1, !"wchar_size", i32 4}
214 // CHECK: ![[META1:[0-9]+]] = !{i32 7, !"openmp", i32 51}
215 // CHECK: ![[META2:[0-9]+]] =
216 // CHECK: ![[LOOP3]] = distinct !{![[LOOP3]], ![[LOOPPROP4:[0-9]+]], ![[LOOPPROP5:[0-9]+]]}
217 // CHECK: ![[LOOPPROP4]] = !{!"llvm.loop.unroll.enable"}
218 // CHECK: ![[LOOPPROP5]] = !{!"llvm.loop.unroll.count", i32 2}
219