1 // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck %s
2 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s
3 // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s
4 // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck %s --check-prefix=TERM_DEBUG
5 
6 // RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck %s
7 // RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s
8 // RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s
9 // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck --check-prefix=TERM_DEBUG %s
10 // expected-no-diagnostics
11  #ifndef HEADER
12  #define HEADER
13 
14 // CHECK: [[SS_TY:%.+]] = type { i32 }
15 
16 long long get_val() { return 0; }
17 double *g_ptr;
18 
19 // CHECK-LABEL: define {{.*void}} @{{.*}}simple{{.*}}(float* {{.+}}, float* {{.+}}, float* {{.+}}, float* {{.+}})
20 void simple(float *a, float *b, float *c, float *d) {
21   #pragma omp simd
22 // CHECK: store i32 0, i32* [[OMP_IV:%[^,]+]]
23 
24 // CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
25 // CHECK-NEXT: [[CMP:%.+]] = icmp slt i32 [[IV]], 6
26 // CHECK-NEXT: br i1 [[CMP]], label %[[SIMPLE_LOOP1_BODY:.+]], label %[[SIMPLE_LOOP1_END:[^,]+]]
27   for (int i = 3; i < 32; i += 5) {
28 // CHECK: [[SIMPLE_LOOP1_BODY]]:
29 // Start of body: calculate i from IV:
30 // CHECK: [[IV1_1:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
31 // CHECK: [[CALC_I_1:%.+]] = mul nsw i32 [[IV1_1]], 5
32 // CHECK-NEXT: [[CALC_I_2:%.+]] = add nsw i32 3, [[CALC_I_1]]
33 // CHECK-NEXT: store i32 [[CALC_I_2]], i32* [[LC_I:.+]]{{.*}}!llvm.access.group
34 // ... loop body ...
35 // End of body: store into a[i]:
36 // CHECK: store float [[RESULT:%.+]], float* {{%.+}}{{.*}}!llvm.access.group
37     a[i] = b[i] * c[i] * d[i];
38 // CHECK: [[IV1_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
39 // CHECK-NEXT: [[ADD1_2:%.+]] = add nsw i32 [[IV1_2]], 1
40 // CHECK-NEXT: store i32 [[ADD1_2]], i32* [[OMP_IV]]{{.*}}!llvm.access.group
41 // br label %{{.+}}, !llvm.loop !{{.+}}
42   }
43 // CHECK: [[SIMPLE_LOOP1_END]]:
44 
45   long long k = get_val();
46 
47   #pragma omp simd linear(k : 3)
48 // CHECK: [[K0:%.+]] = call {{.*}}i64 @{{.*}}get_val
49 // CHECK-NEXT: store i64 [[K0]], i64* [[K_VAR:%[^,]+]]
50 // CHECK: store i32 0, i32* [[OMP_IV2:%[^,]+]]
51 // CHECK: [[K0LOAD:%.+]] = load i64, i64* [[K_VAR]]
52 // CHECK-NEXT: store i64 [[K0LOAD]], i64* [[LIN0:%[^,]+]]
53 
54 // CHECK: [[IV2:%.+]] = load i32, i32* [[OMP_IV2]]{{.*}}!llvm.access.group
55 // CHECK-NEXT: [[CMP2:%.+]] = icmp slt i32 [[IV2]], 9
56 // CHECK-NEXT: br i1 [[CMP2]], label %[[SIMPLE_LOOP2_BODY:.+]], label %[[SIMPLE_LOOP2_END:[^,]+]]
57   for (int i = 10; i > 1; i--) {
58 // CHECK: [[SIMPLE_LOOP2_BODY]]:
59 // Start of body: calculate i from IV:
60 // CHECK: [[IV2_0:%.+]] = load i32, i32* [[OMP_IV2]]{{.*}}!llvm.access.group
61 // FIXME: It is interesting, why the following "mul 1" was not constant folded?
62 // CHECK-NEXT: [[IV2_1:%.+]] = mul nsw i32 [[IV2_0]], 1
63 // CHECK-NEXT: [[LC_I_1:%.+]] = sub nsw i32 10, [[IV2_1]]
64 // CHECK-NEXT: store i32 [[LC_I_1]], i32* {{.+}}, !llvm.access.group
65 //
66 // CHECK-NEXT: [[LIN0_1:%.+]] = load i64, i64* [[LIN0]]{{.*}}!llvm.access.group
67 // CHECK-NEXT: [[IV2_2:%.+]] = load i32, i32* [[OMP_IV2]]{{.*}}!llvm.access.group
68 // CHECK-NEXT: [[LIN_MUL1:%.+]] = mul nsw i32 [[IV2_2]], 3
69 // CHECK-NEXT: [[LIN_EXT1:%.+]] = sext i32 [[LIN_MUL1]] to i64
70 // CHECK-NEXT: [[LIN_ADD1:%.+]] = add nsw i64 [[LIN0_1]], [[LIN_EXT1]]
71 // Update of the privatized version of linear variable!
72 // CHECK-NEXT: store i64 [[LIN_ADD1]], i64* [[K_PRIVATIZED:%[^,]+]]
73     a[k]++;
74     k = k + 3;
75 // CHECK: [[IV2_2:%.+]] = load i32, i32* [[OMP_IV2]]{{.*}}!llvm.access.group
76 // CHECK-NEXT: [[ADD2_2:%.+]] = add nsw i32 [[IV2_2]], 1
77 // CHECK-NEXT: store i32 [[ADD2_2]], i32* [[OMP_IV2]]{{.*}}!llvm.access.group
78 // br label {{.+}}, !llvm.loop ![[SIMPLE_LOOP2_ID]]
79   }
80 // CHECK: [[SIMPLE_LOOP2_END]]:
81 //
82 // Update linear vars after loop, as the loop was operating on a private version.
83 // CHECK: [[LIN0_2:%.+]] = load i64, i64* [[LIN0]]
84 // CHECK-NEXT: [[LIN_ADD2:%.+]] = add nsw i64 [[LIN0_2]], 27
85 // CHECK-NEXT: store i64 [[LIN_ADD2]], i64* [[K_VAR]]
86 //
87 
88   int lin = 12;
89   #pragma omp simd linear(lin : get_val()), linear(g_ptr)
90 
91 // Init linear private var.
92 // CHECK: store i32 12, i32* [[LIN_VAR:%[^,]+]]
93 // CHECK: store i64 0, i64* [[OMP_IV3:%[^,]+]]
94 
95 // CHECK: [[LIN_LOAD:%.+]] = load i32, i32* [[LIN_VAR]]
96 // CHECK-NEXT: store i32 [[LIN_LOAD]], i32* [[LIN_START:%[^,]+]]
97 // Remember linear step.
98 // CHECK: [[CALL_VAL:%.+]] = invoke
99 // CHECK: store i64 [[CALL_VAL]], i64* [[LIN_STEP:%[^,]+]]
100 
101 // CHECK: [[GLIN_LOAD:%.+]] = load double*, double** [[GLIN_VAR:@[^,]+]]
102 // CHECK-NEXT: store double* [[GLIN_LOAD]], double** [[GLIN_START:%[^,]+]]
103 
104 // CHECK: [[IV3:%.+]] = load i64, i64* [[OMP_IV3]]{{.*}}!llvm.access.group
105 // CHECK-NEXT: [[CMP3:%.+]] = icmp ult i64 [[IV3]], 4
106 // CHECK-NEXT: br i1 [[CMP3]], label %[[SIMPLE_LOOP3_BODY:.+]], label %[[SIMPLE_LOOP3_END:[^,]+]]
107   for (unsigned long long it = 2000; it >= 600; it-=400) {
108 // CHECK: [[SIMPLE_LOOP3_BODY]]:
109 // Start of body: calculate it from IV:
110 // CHECK: [[IV3_0:%.+]] = load i64, i64* [[OMP_IV3]]{{.*}}!llvm.access.group
111 // CHECK-NEXT: [[LC_IT_1:%.+]] = mul i64 [[IV3_0]], 400
112 // CHECK-NEXT: [[LC_IT_2:%.+]] = sub i64 2000, [[LC_IT_1]]
113 // CHECK-NEXT: store i64 [[LC_IT_2]], i64* {{.+}}, !llvm.access.group
114 //
115 // Linear start and step are used to calculate current value of the linear variable.
116 // CHECK: [[LINSTART:.+]] = load i32, i32* [[LIN_START]]{{.*}}!llvm.access.group
117 // CHECK: [[LINSTEP:.+]] = load i64, i64* [[LIN_STEP]]{{.*}}!llvm.access.group
118 // CHECK-NOT: store i32 {{.+}}, i32* [[LIN_VAR]],{{.*}}!llvm.access.group
119 // CHECK: [[GLINSTART:.+]] = load double*, double** [[GLIN_START]]{{.*}}!llvm.access.group
120 // CHECK-NEXT: [[IV3_1:%.+]] = load i64, i64* [[OMP_IV3]]{{.*}}!llvm.access.group
121 // CHECK-NEXT: [[MUL:%.+]] = mul i64 [[IV3_1]], 1
122 // CHECK: [[GEP:%.+]] = getelementptr{{.*}}[[GLINSTART]]
123 // CHECK-NEXT: store double* [[GEP]], double** [[G_PTR_CUR:%[^,]+]]{{.*}}!llvm.access.group
124     *g_ptr++ = 0.0;
125 // CHECK: [[GEP_VAL:%.+]] = load double{{.*}}[[G_PTR_CUR]]{{.*}}!llvm.access.group
126 // CHECK: store double{{.*}}[[GEP_VAL]]{{.*}}!llvm.access.group
127     a[it + lin]++;
128 // CHECK: [[FLT_INC:%.+]] = fadd float
129 // CHECK-NEXT: store float [[FLT_INC]],{{.*}}!llvm.access.group
130 // CHECK: [[IV3_2:%.+]] = load i64, i64* [[OMP_IV3]]{{.*}}!llvm.access.group
131 // CHECK-NEXT: [[ADD3_2:%.+]] = add i64 [[IV3_2]], 1
132 // CHECK-NEXT: store i64 [[ADD3_2]], i64* [[OMP_IV3]]{{.*}}!llvm.access.group
133   }
134 // CHECK: [[SIMPLE_LOOP3_END]]:
135 //
136 // Linear start and step are used to calculate final value of the linear variables.
137 // CHECK: [[LINSTART:.+]] = load i32, i32* [[LIN_START]]
138 // CHECK: [[LINSTEP:.+]] = load i64, i64* [[LIN_STEP]]
139 // CHECK: store i32 {{.+}}, i32* [[LIN_VAR]],
140 // CHECK: [[GLINSTART:.+]] = load double*, double** [[GLIN_START]]
141 // CHECK: store double* {{.*}}[[GLIN_VAR]]
142 
143   #pragma omp simd
144 // CHECK: store i32 0, i32* [[OMP_IV4:%[^,]+]]
145 
146 // CHECK: [[IV4:%.+]] = load i32, i32* [[OMP_IV4]]{{.*}}!llvm.access.group
147 // CHECK-NEXT: [[CMP4:%.+]] = icmp slt i32 [[IV4]], 4
148 // CHECK-NEXT: br i1 [[CMP4]], label %[[SIMPLE_LOOP4_BODY:.+]], label %[[SIMPLE_LOOP4_END:[^,]+]]
149   for (short it = 6; it <= 20; it-=-4) {
150 // CHECK: [[SIMPLE_LOOP4_BODY]]:
151 // Start of body: calculate it from IV:
152 // CHECK: [[IV4_0:%.+]] = load i32, i32* [[OMP_IV4]]{{.*}}!llvm.access.group
153 // CHECK-NEXT: [[LC_IT_1:%.+]] = mul nsw i32 [[IV4_0]], 4
154 // CHECK-NEXT: [[LC_IT_2:%.+]] = add nsw i32 6, [[LC_IT_1]]
155 // CHECK-NEXT: [[LC_IT_3:%.+]] = trunc i32 [[LC_IT_2]] to i16
156 // CHECK-NEXT: store i16 [[LC_IT_3]], i16* {{.+}}, !llvm.access.group
157 
158 // CHECK: [[IV4_2:%.+]] = load i32, i32* [[OMP_IV4]]{{.*}}!llvm.access.group
159 // CHECK-NEXT: [[ADD4_2:%.+]] = add nsw i32 [[IV4_2]], 1
160 // CHECK-NEXT: store i32 [[ADD4_2]], i32* [[OMP_IV4]]{{.*}}!llvm.access.group
161   }
162 // CHECK: [[SIMPLE_LOOP4_END]]:
163 
164   #pragma omp simd
165 // CHECK: store i32 0, i32* [[OMP_IV5:%[^,]+]]
166 
167 // CHECK: [[IV5:%.+]] = load i32, i32* [[OMP_IV5]]{{.*}}!llvm.access.group
168 // CHECK-NEXT: [[CMP5:%.+]] = icmp slt i32 [[IV5]], 26
169 // CHECK-NEXT: br i1 [[CMP5]], label %[[SIMPLE_LOOP5_BODY:.+]], label %[[SIMPLE_LOOP5_END:[^,]+]]
170   for (unsigned char it = 'z'; it >= 'a'; it+=-1) {
171 // CHECK: [[SIMPLE_LOOP5_BODY]]:
172 // Start of body: calculate it from IV:
173 // CHECK: [[IV5_0:%.+]] = load i32, i32* [[OMP_IV5]]{{.*}}!llvm.access.group
174 // CHECK-NEXT: [[IV5_1:%.+]] = mul nsw i32 [[IV5_0]], 1
175 // CHECK-NEXT: [[LC_IT_1:%.+]] = sub nsw i32 122, [[IV5_1]]
176 // CHECK-NEXT: [[LC_IT_2:%.+]] = trunc i32 [[LC_IT_1]] to i8
177 // CHECK-NEXT: store i8 [[LC_IT_2]], i8* {{.+}}, !llvm.access.group
178 
179 // CHECK: [[IV5_2:%.+]] = load i32, i32* [[OMP_IV5]]{{.*}}!llvm.access.group
180 // CHECK-NEXT: [[ADD5_2:%.+]] = add nsw i32 [[IV5_2]], 1
181 // CHECK-NEXT: store i32 [[ADD5_2]], i32* [[OMP_IV5]]{{.*}}!llvm.access.group
182   }
183 // CHECK: [[SIMPLE_LOOP5_END]]:
184 
185 // CHECK-NOT: mul i32 %{{.+}}, 10
186   #pragma omp simd
187   for (unsigned i=100; i<10; i+=10) {
188   }
189 
190   int A;
191   // CHECK: store i32 -1, i32* [[A:%.+]],
192   A = -1;
193   #pragma omp simd lastprivate(A)
194 // CHECK: store i64 0, i64* [[OMP_IV7:%[^,]+]]
195 // CHECK: br label %[[SIMD_LOOP7_COND:[^,]+]]
196 // CHECK: [[SIMD_LOOP7_COND]]:
197 // CHECK-NEXT: [[IV7:%.+]] = load i64, i64* [[OMP_IV7]]{{.*}}!llvm.access.group
198 // CHECK-NEXT: [[CMP7:%.+]] = icmp slt i64 [[IV7]], 7
199 // CHECK-NEXT: br i1 [[CMP7]], label %[[SIMPLE_LOOP7_BODY:.+]], label %[[SIMPLE_LOOP7_END:[^,]+]]
200   for (long long i = -10; i < 10; i += 3) {
201 // CHECK: [[SIMPLE_LOOP7_BODY]]:
202 // Start of body: calculate i from IV:
203 // CHECK: [[IV7_0:%.+]] = load i64, i64* [[OMP_IV7]]{{.*}}!llvm.access.group
204 // CHECK-NEXT: [[LC_IT_1:%.+]] = mul nsw i64 [[IV7_0]], 3
205 // CHECK-NEXT: [[LC_IT_2:%.+]] = add nsw i64 -10, [[LC_IT_1]]
206 // CHECK-NEXT: store i64 [[LC_IT_2]], i64* [[LC:%[^,]+]],{{.+}}!llvm.access.group
207 // CHECK-NEXT: [[LC_VAL:%.+]] = load i64, i64* [[LC]]{{.+}}!llvm.access.group
208 // CHECK-NEXT: [[CONV:%.+]] = trunc i64 [[LC_VAL]] to i32
209 // CHECK-NEXT: store i32 [[CONV]], i32* [[A_PRIV:%[^,]+]],{{.+}}!llvm.access.group
210     A = i;
211 // CHECK: [[IV7_2:%.+]] = load i64, i64* [[OMP_IV7]]{{.*}}!llvm.access.group
212 // CHECK-NEXT: [[ADD7_2:%.+]] = add nsw i64 [[IV7_2]], 1
213 // CHECK-NEXT: store i64 [[ADD7_2]], i64* [[OMP_IV7]]{{.*}}!llvm.access.group
214   }
215 // CHECK: [[SIMPLE_LOOP7_END]]:
216 // CHECK-NEXT: store i64 11, i64*
217 // CHECK-NEXT: [[A_PRIV_VAL:%.+]] = load i32, i32* [[A_PRIV]],
218 // CHECK-NEXT: store i32 [[A_PRIV_VAL]], i32* [[A]],
219   int R;
220   // CHECK: store i32 -1, i32* [[R:%[^,]+]],
221   R = -1;
222 // CHECK: store i64 0, i64* [[OMP_IV8:%[^,]+]],
223 // CHECK: store i32 1, i32* [[R_PRIV:%[^,]+]],
224   #pragma omp simd reduction(*:R)
225 // CHECK: br label %[[SIMD_LOOP8_COND:[^,]+]]
226 // CHECK: [[SIMD_LOOP8_COND]]:
227 // CHECK-NEXT: [[IV8:%.+]] = load i64, i64* [[OMP_IV8]]{{.*}}!llvm.access.group
228 // CHECK-NEXT: [[CMP8:%.+]] = icmp slt i64 [[IV8]], 7
229 // CHECK-NEXT: br i1 [[CMP8]], label %[[SIMPLE_LOOP8_BODY:.+]], label %[[SIMPLE_LOOP8_END:[^,]+]]
230   for (long long i = -10; i < 10; i += 3) {
231 // CHECK: [[SIMPLE_LOOP8_BODY]]:
232 // Start of body: calculate i from IV:
233 // CHECK: [[IV8_0:%.+]] = load i64, i64* [[OMP_IV8]]{{.*}}!llvm.access.group
234 // CHECK-NEXT: [[LC_IT_1:%.+]] = mul nsw i64 [[IV8_0]], 3
235 // CHECK-NEXT: [[LC_IT_2:%.+]] = add nsw i64 -10, [[LC_IT_1]]
236 // CHECK-NEXT: store i64 [[LC_IT_2]], i64* [[LC:%[^,]+]],{{.+}}!llvm.access.group
237 // CHECK-NEXT: [[LC_VAL:%.+]] = load i64, i64* [[LC]]{{.+}}!llvm.access.group
238 // CHECK: store i32 %{{.+}}, i32* [[R_PRIV]],{{.+}}!llvm.access.group
239     R *= i;
240 // CHECK: [[IV8_2:%.+]] = load i64, i64* [[OMP_IV8]]{{.*}}!llvm.access.group
241 // CHECK-NEXT: [[ADD8_2:%.+]] = add nsw i64 [[IV8_2]], 1
242 // CHECK-NEXT: store i64 [[ADD8_2]], i64* [[OMP_IV8]]{{.*}}!llvm.access.group
243   }
244 // CHECK: [[SIMPLE_LOOP8_END]]:
245 // CHECK-DAG: [[R_VAL:%.+]] = load i32, i32* [[R]],
246 // CHECK-DAG: [[R_PRIV_VAL:%.+]] = load i32, i32* [[R_PRIV]],
247 // CHECK: [[RED:%.+]] = mul nsw i32 [[R_VAL]], [[R_PRIV_VAL]]
248 // CHECK-NEXT: store i32 [[RED]], i32* [[R]],
249 // CHECK-NEXT: ret void
250 }
251 
252 template <class T, unsigned K> T tfoo(T a) { return a + K; }
253 
254 template <typename T, unsigned N>
255 int templ1(T a, T *z) {
256   #pragma omp simd collapse(N)
257   for (int i = 0; i < N * 2; i++) {
258     for (long long j = 0; j < (N + N + N + N); j += 2) {
259       z[i + j] = a + tfoo<T, N>(i + j);
260     }
261   }
262   return 0;
263 }
264 
265 // Instatiation templ1<float,2>
266 // CHECK-LABEL: define {{.*i32}} @{{.*}}templ1{{.*}}(float {{.+}}, float* {{.+}})
267 // CHECK: store i64 0, i64* [[T1_OMP_IV:[^,]+]]
268 // ...
269 // CHECK: [[IV:%.+]] = load i64, i64* [[T1_OMP_IV]]{{.*}}!llvm.access.group
270 // CHECK-NEXT: [[CMP1:%.+]] = icmp slt i64 [[IV]], 16
271 // CHECK-NEXT: br i1 [[CMP1]], label %[[T1_BODY:.+]], label %[[T1_END:[^,]+]]
272 // CHECK: [[T1_BODY]]:
273 // Loop counters i and j updates:
274 // CHECK: [[IV1:%.+]] = load i64, i64* [[T1_OMP_IV]]{{.*}}!llvm.access.group
275 // CHECK-NEXT: [[I_1:%.+]] = sdiv i64 [[IV1]], 4
276 // CHECK-NEXT: [[I_1_MUL1:%.+]] = mul nsw i64 [[I_1]], 1
277 // CHECK-NEXT: [[I_1_ADD0:%.+]] = add nsw i64 0, [[I_1_MUL1]]
278 // CHECK-NEXT: [[I_2:%.+]] = trunc i64 [[I_1_ADD0]] to i32
279 // CHECK-NEXT: store i32 [[I_2]], i32* {{%.+}}{{.*}}!llvm.access.group
280 // CHECK: [[IV2:%.+]] = load i64, i64* [[T1_OMP_IV]]{{.*}}!llvm.access.group
281 // CHECK: [[IV2_1:%.+]] = load i64, i64* [[T1_OMP_IV]]{{.*}}!llvm.access.group
282 // CHECK-NEXT: [[J_1_DIV1:%.+]] = sdiv i64 [[IV2_1]], 4
283 // CHECK-NEXT: [[J_1_MUL1:%.+]] = mul nsw i64 [[J_1_DIV1]], 4
284 // CHECK-NEXT: [[J_1_SUB0:%.+]] = sub nsw i64 [[IV2]], [[J_1_MUL1]]
285 // CHECK-NEXT: [[J_2:%.+]] = mul nsw i64 [[J_1_SUB0]], 2
286 // CHECK-NEXT: [[J_2_ADD0:%.+]] = add nsw i64 0, [[J_2]]
287 // CHECK-NEXT: store i64 [[J_2_ADD0]], i64* {{%.+}}{{.*}}!llvm.access.group
288 // simd.for.inc:
289 // CHECK: [[IV3:%.+]] = load i64, i64* [[T1_OMP_IV]]{{.*}}!llvm.access.group
290 // CHECK-NEXT: [[INC:%.+]] = add nsw i64 [[IV3]], 1
291 // CHECK-NEXT: store i64 [[INC]], i64* [[T1_OMP_IV]]{{.*}}!llvm.access.group
292 // CHECK-NEXT: br label {{%.+}}
293 // CHECK: [[T1_END]]:
294 // CHECK: ret i32 0
295 //
296 void inst_templ1() {
297   float a;
298   float z[100];
299   templ1<float,2> (a, z);
300 }
301 
302 
303 typedef int MyIdx;
304 
305 class IterDouble {
306   double *Ptr;
307 public:
308   IterDouble operator++ () const {
309     IterDouble n;
310     n.Ptr = Ptr + 1;
311     return n;
312   }
313   bool operator < (const IterDouble &that) const {
314     return Ptr < that.Ptr;
315   }
316   double & operator *() const {
317     return *Ptr;
318   }
319   MyIdx operator - (const IterDouble &that) const {
320     return (MyIdx) (Ptr - that.Ptr);
321   }
322   IterDouble operator + (int Delta) {
323     IterDouble re;
324     re.Ptr = Ptr + Delta;
325     return re;
326   }
327 
328   ///~IterDouble() {}
329 };
330 
331 // CHECK-LABEL: define {{.*void}} @{{.*}}iter_simple{{.*}}
332 void iter_simple(IterDouble ia, IterDouble ib, IterDouble ic) {
333 //
334 // Calculate number of iterations before the loop body.
335 // CHECK: [[DIFF1:%.+]] = invoke {{.*}}i32 @{{.*}}IterDouble{{.*}}
336 // CHECK: [[DIFF2:%.+]] = sub nsw i32 [[DIFF1]], 1
337 // CHECK-NEXT: [[DIFF3:%.+]] = add nsw i32 [[DIFF2]], 1
338 // CHECK-NEXT: [[DIFF4:%.+]] = sdiv i32 [[DIFF3]], 1
339 // CHECK-NEXT: [[DIFF5:%.+]] = sub nsw i32 [[DIFF4]], 1
340 // CHECK-NEXT: store i32 [[DIFF5]], i32* [[OMP_LAST_IT:%[^,]+]]{{.+}}
341 // CHECK: store i32 0, i32* [[IT_OMP_IV:%[^,]+]]
342   #pragma omp simd
343 
344 // CHECK: [[IV:%.+]] = load i32, i32* [[IT_OMP_IV]]{{.+}} !llvm.access.group
345 // CHECK-NEXT: [[LAST_IT:%.+]] = load i32, i32* [[OMP_LAST_IT]]{{.+}}!llvm.access.group
346 // CHECK-NEXT: [[NUM_IT:%.+]] = add nsw i32 [[LAST_IT]], 1
347 // CHECK-NEXT: [[CMP:%.+]] = icmp slt i32 [[IV]], [[NUM_IT]]
348 // CHECK-NEXT: br i1 [[CMP]], label %[[IT_BODY:[^,]+]], label %[[IT_END:[^,]+]]
349   for (IterDouble i = ia; i < ib; ++i) {
350 // CHECK: [[IT_BODY]]:
351 // Start of body: calculate i from index:
352 // CHECK: [[IV1:%.+]] = load i32, i32* [[IT_OMP_IV]]{{.+}}!llvm.access.group
353 // Call of operator+ (i, IV).
354 // CHECK: {{%.+}} = invoke {{.+}} @{{.*}}IterDouble{{.*}}
355 // ... loop body ...
356    *i = *ic * 0.5;
357 // Float multiply and save result.
358 // CHECK: [[MULR:%.+]] = fmul double {{%.+}}, 5.000000e-01
359 // CHECK-NEXT: invoke {{.+}} @{{.*}}IterDouble{{.*}}
360 // CHECK: store double [[MULR:%.+]], double* [[RESULT_ADDR:%.+]], !llvm.access.group
361    ++ic;
362 //
363 // CHECK: [[IV2:%.+]] = load i32, i32* [[IT_OMP_IV]]{{.+}}!llvm.access.group
364 // CHECK-NEXT: [[ADD2:%.+]] = add nsw i32 [[IV2]], 1
365 // CHECK-NEXT: store i32 [[ADD2]], i32* [[IT_OMP_IV]]{{.+}}!llvm.access.group
366 // br label %{{.*}}, !llvm.loop ![[ITER_LOOP_ID]]
367   }
368 // CHECK: [[IT_END]]:
369 // CHECK: ret void
370 }
371 
372 
373 // CHECK-LABEL: define {{.*void}} @{{.*}}collapsed{{.*}}
374 void collapsed(float *a, float *b, float *c, float *d) {
375   int i; // outer loop counter
376   unsigned j; // middle loop couter, leads to unsigned icmp in loop header.
377   // k declared in the loop init below
378   short l; // inner loop counter
379 // CHECK: store i32 0, i32* [[OMP_IV:[^,]+]]
380 //
381   #pragma omp simd collapse(4)
382 
383 // CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group
384 // CHECK-NEXT: [[CMP:%.+]] = icmp ult i32 [[IV]], 120
385 // CHECK-NEXT: br i1 [[CMP]], label %[[COLL1_BODY:[^,]+]], label %[[COLL1_END:[^,]+]]
386   for (i = 1; i < 3; i++) // 2 iterations
387     for (j = 2u; j < 5u; j++) //3 iterations
388       for (int k = 3; k <= 6; k++) // 4 iterations
389         for (l = 4; l < 9; ++l) // 5 iterations
390         {
391 // CHECK: [[COLL1_BODY]]:
392 // Start of body: calculate i from index:
393 // CHECK: [[IV1:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group
394 // Calculation of the loop counters values.
395 // CHECK: [[CALC_I_1:%.+]] = udiv i32 [[IV1]], 60
396 // CHECK-NEXT: [[CALC_I_1_MUL1:%.+]] = mul i32 [[CALC_I_1]], 1
397 // CHECK-NEXT: [[CALC_I_2:%.+]] = add i32 1, [[CALC_I_1_MUL1]]
398 // CHECK-NEXT: store i32 [[CALC_I_2]], i32* [[LC_I:.+]]
399 
400 // CHECK: [[IV1_2:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group
401 // CHECK: [[IV1_2_1:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group
402 // CHECK-NEXT: [[CALC_J_1:%.+]] = udiv i32 [[IV1_2_1]], 60
403 // CHECK-NEXT: [[MUL_1:%.+]] = mul i32 [[CALC_J_1]], 60
404 // CHECK-NEXT: [[SUB_3:%.+]] = sub i32 [[IV1_2]], [[MUL_1]]
405 // CHECK-NEXT: [[CALC_J_2:%.+]] = udiv i32 [[SUB_3]], 20
406 // CHECK-NEXT: [[CALC_J_2_MUL1:%.+]] = mul i32 [[CALC_J_2]], 1
407 // CHECK-NEXT: [[CALC_J_3:%.+]] = add i32 2, [[CALC_J_2_MUL1]]
408 // CHECK-NEXT: store i32 [[CALC_J_3]], i32* [[LC_J:.+]]
409 
410 // CHECK: [[IV1_3:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group
411 // CHECK: [[IV1_3_1:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group
412 // CHECK-NEXT: [[DIV_1:%.+]] = udiv i32 [[IV1_3_1]], 60
413 // CHECK-NEXT: [[MUL_2:%.+]] = mul i32 [[DIV_1]], 60
414 // CHECK-NEXT: [[ADD_3:%.+]] = sub i32 [[IV1_3]], [[MUL_2]]
415 
416 // CHECK: [[IV1_4:%.+]] = load i32, i32* [[OMP_IV]]
417 // CHECK: [[IV1_4_1:%.+]] = load i32, i32* [[OMP_IV]]
418 // CHECK-NEXT: [[DIV_2:%.+]] = udiv i32 [[IV1_4_1]], 60
419 // CHECK-NEXT: [[MUL_3:%.+]] = mul i32 [[DIV_2]], 60
420 // CHECK-NEXT: [[SUB_6:%.+]] = sub i32 [[IV1_4]], [[MUL_3]]
421 // CHECK-NEXT: [[DIV_3:%.+]] = udiv i32 [[SUB_6]], 20
422 // CHECK-NEXT: [[MUL_4:%.+]] = mul i32 [[DIV_3]], 20
423 // CHECK-NEXT: [[ADD_5:%.+]] = sub i32 [[ADD_3]], [[MUL_4]]
424 // CHECK-NEXT: [[DIV_4:%.+]] = udiv i32 [[ADD_5]], 5
425 // CHECK-NEXT: [[MUL_5:%.+]] = mul i32 [[DIV_4]], 1
426 // CHECK-NEXT: [[ADD_6:%.+]] = add i32 3, [[MUL_5]]
427 // CHECK-NEXT: store i32 [[ADD_6]], i32* [[LC_K:.+]]
428 
429 // CHECK: [[IV1_5:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group
430 // CHECK: [[IV1_5_1:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group
431 // CHECK-NEXT: [[DIV_5:%.+]] = udiv i32 [[IV1_5_1]], 60
432 // CHECK-NEXT: [[MUL_6:%.+]] = mul i32 [[DIV_5]], 60
433 // CHECK-NEXT: [[SUB_7:%.+]] = sub i32 [[IV1_5]], [[MUL_6]]
434 
435 // CHECK: [[IV1_6:%.+]] = load i32, i32* [[OMP_IV]]
436 // CHECK: [[IV1_6_1:%.+]] = load i32, i32* [[OMP_IV]]
437 // CHECK-NEXT: [[DIV_6:%.+]] = udiv i32 [[IV1_6_1]], 60
438 // CHECK-NEXT: [[MUL_7:%.+]] = mul i32 [[DIV_6]], 60
439 // CHECK-NEXT: [[SUB_10:%.+]] = sub i32 [[IV1_6]], [[MUL_7]]
440 // CHECK-NEXT: [[DIV_7:%.+]] = udiv i32 [[SUB_10]], 20
441 // CHECK-NEXT: [[MUL_8:%.+]] = mul i32 [[DIV_7]], 20
442 // CHECK-NEXT: [[SUB_11:%.+]] = sub i32 [[SUB_7]], [[MUL_8]]
443 
444 // CHECK: [[IV1_7:%.+]] = load i32, i32* [[OMP_IV]]
445 // CHECK: [[IV1_7_1:%.+]] = load i32, i32* [[OMP_IV]]
446 // CHECK-NEXT: [[DIV_8:%.+]] = udiv i32 [[IV1_7_1]], 60
447 // CHECK-NEXT: [[MUL_9:%.+]] = mul i32 [[DIV_8]], 60
448 // CHECK-NEXT: [[SUB_12:%.+]] = sub i32 [[IV1_7]], [[MUL_9]]
449 
450 // CHECK: [[IV1_8:%.+]] = load i32, i32* [[OMP_IV]]
451 // CHECK: [[IV1_8_1:%.+]] = load i32, i32* [[OMP_IV]]
452 // CHECK-NEXT: [[DIV_3:%.+]] = udiv i32 [[IV1_8_1]], 60
453 // CHECK-NEXT: [[MUL_4:%.+]] = mul i32 [[DIV_3]], 60
454 // CHECK-NEXT: [[SUB_7:%.+]] = sub i32 [[IV1_8]], [[MUL_4]]
455 // CHECK-NEXT: [[DIV_4:%.+]] = udiv i32 [[SUB_7]], 20
456 // CHECK-NEXT: [[MUL_5:%.+]] = mul i32 [[DIV_4]], 20
457 // CHECK-NEXT: [[SUB_8:%.+]] = sub i32 [[SUB_12]], [[MUL_5]]
458 // CHECK-NEXT: [[DIV_5:%.+]] = udiv i32 [[SUB_8]], 5
459 // CHECK-NEXT: [[MUL_6:%.+]] = mul i32 [[DIV_5]], 5
460 // CHECK-NEXT: [[SUB_9:%.+]] = sub i32 [[SUB_11]], [[MUL_6]]
461 // CHECK-NEXT: [[MUL_6:%.+]] = mul i32 [[SUB_9]], 1
462 // CHECK-NEXT: [[CALC_L_2:%.+]] = add i32 4, [[MUL_6]]
463 // CHECK-NEXT: [[CALC_L_3:%.+]] = trunc i32 [[CALC_L_2]] to i16
464 // CHECK-NEXT: store i16 [[CALC_L_3]], i16* [[LC_L:.+]]
465 // ... loop body ...
466 // End of body: store into a[i]:
467 // CHECK: store float [[RESULT:%.+]], float* [[RESULT_ADDR:%.+]]{{.+}}!llvm.access.group
468     float res = b[j] * c[k];
469     a[i] = res * d[l];
470 // CHECK: [[IV2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
471 // CHECK-NEXT: [[ADD2:%.+]] = add i32 [[IV2]], 1
472 // CHECK-NEXT: store i32 [[ADD2]], i32* [[OMP_IV]]{{.*}}!llvm.access.group
473 // br label %{{[^,]+}}, !llvm.loop ![[COLL1_LOOP_ID]]
474 // CHECK: [[COLL1_END]]:
475   }
476 // i,j,l are updated; k is not updated.
477 // CHECK: store i32 3, i32*
478 // CHECK-NEXT: store i32 5, i32*
479 // CHECK-NEXT: store i32 7, i32*
480 // CHECK-NEXT: store i16 9, i16*
481 // CHECK: ret void
482 }
483 
484 extern char foo();
485 extern double globalfloat;
486 
487 // CHECK-LABEL: define {{.*void}} @{{.*}}widened{{.*}}
488 void widened(float *a, float *b, float *c, float *d) {
489   int i; // outer loop counter
490   short j; // inner loop counter
491   globalfloat = 1.0;
492   int localint = 1;
493 // CHECK: store double {{.+}}, double* [[GLOBALFLOAT:@.+]]
494 // Counter is widened to 64 bits.
495 // CHECK: store i64 0, i64* [[OMP_IV:[^,]+]]
496 //
497   #pragma omp simd collapse(2) private(globalfloat, localint)
498 
499 // CHECK: [[IV:%.+]] = load i64, i64* [[OMP_IV]]{{.+}}!llvm.access.group
500 // CHECK-NEXT: [[LI:%.+]] = load i64, i64* [[OMP_LI:%[^,]+]]{{.+}}!llvm.access.group
501 // CHECK-NEXT: [[NUMIT:%.+]] = add nsw i64 [[LI]], 1
502 // CHECK-NEXT: [[CMP:%.+]] = icmp slt i64 [[IV]], [[NUMIT]]
503 // CHECK-NEXT: br i1 [[CMP]], label %[[WIDE1_BODY:[^,]+]], label %[[WIDE1_END:[^,]+]]
504   for (i = 1; i < 3; i++) // 2 iterations
505     for (j = 0; j < foo(); j++) // foo() iterations
506   {
507 // CHECK: [[WIDE1_BODY]]:
508 // Start of body: calculate i from index:
509 // CHECK: [[IV1:%.+]] = load i64, i64* [[OMP_IV]]{{.+}}!llvm.access.group
510 // Calculation of the loop counters values...
511 // CHECK: store i32 {{[^,]+}}, i32* [[LC_I:.+]]
512 // CHECK: [[IV1_2:%.+]] = load i64, i64* [[OMP_IV]]{{.+}}!llvm.access.group
513 // CHECK: store i16 {{[^,]+}}, i16* [[LC_J:.+]]
514 // ... loop body ...
515 //
516 // Here we expect store into private double var, not global
517 // CHECK-NOT: store double {{.+}}, double* [[GLOBALFLOAT]]
518     globalfloat = (float)j/i;
519     float res = b[j] * c[j];
520 // Store into a[i]:
521 // CHECK: store float [[RESULT:%.+]], float* [[RESULT_ADDR:%.+]]{{.+}}!llvm.access.group
522     a[i] = res * d[i];
523 // Then there's a store into private var localint:
524 // CHECK: store i32 {{.+}}, i32* [[LOCALINT:%[^,]+]]{{.+}}!llvm.access.group
525     localint = (int)j;
526 // CHECK: [[IV2:%.+]] = load i64, i64* [[OMP_IV]]{{.*}}!llvm.access.group
527 // CHECK-NEXT: [[ADD2:%.+]] = add nsw i64 [[IV2]], 1
528 // CHECK-NEXT: store i64 [[ADD2]], i64* [[OMP_IV]]{{.*}}!llvm.access.group
529 //
530 // br label %{{[^,]+}}, !llvm.loop ![[WIDE1_LOOP_ID]]
531 // CHECK: [[WIDE1_END]]:
532   }
533 // i,j are updated.
534 // CHECK: store i32 3, i32* [[I:%[^,]+]]
535 // CHECK: store i16
536 //
537 // Here we expect store into original localint, not its privatized version.
538 // CHECK-NOT: store i32 {{.+}}, i32* [[LOCALINT]]
539   localint = (int)j;
540 // CHECK: ret void
541 }
542 
543 // CHECK-LABEL: define {{.*void}} @{{.*}}linear{{.*}}(float* {{.+}})
544 void linear(float *a) {
545   // CHECK: [[VAL_ADDR:%.+]] = alloca i64,
546   // CHECK: [[K_ADDR:%.+]] = alloca i64*,
547   long long val = 0;
548   long long &k = val;
549 
550   #pragma omp simd linear(k : 3)
551 // CHECK: store i64* [[VAL_ADDR]], i64** [[K_ADDR]],
552 // CHECK: [[VAL_REF:%.+]] = load i64*, i64** [[K_ADDR]],
553 // CHECK: store i64* [[VAL_REF]], i64** [[K_ADDR_REF:%.+]],
554 // CHECK: store i32 0, i32* [[OMP_IV:%[^,]+]]
555 // CHECK: [[K_REF:%.+]] = load i64*, i64** [[K_ADDR_REF]],
556 // CHECK: [[K0LOAD:%.+]] = load i64, i64* [[K_REF]]
557 // CHECK-NEXT: store i64 [[K0LOAD]], i64* [[LIN0:%[^,]+]]
558 
559 // CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
560 // CHECK-NEXT: [[CMP2:%.+]] = icmp slt i32 [[IV]], 9
561 // CHECK-NEXT: br i1 [[CMP2]], label %[[SIMPLE_LOOP_BODY:.+]], label %[[SIMPLE_LOOP_END:[^,]+]]
562   for (int i = 10; i > 1; i--) {
563 // CHECK: [[SIMPLE_LOOP_BODY]]:
564 // Start of body: calculate i from IV:
565 // CHECK: [[IV_0:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
566 // FIXME: It is interesting, why the following "mul 1" was not constant folded?
567 // CHECK-NEXT: [[IV_1:%.+]] = mul nsw i32 [[IV_0]], 1
568 // CHECK-NEXT: [[LC_I_1:%.+]] = sub nsw i32 10, [[IV_1]]
569 // CHECK-NEXT: store i32 [[LC_I_1]], i32* {{.+}}, !llvm.access.group
570 //
571 // CHECK-NEXT: [[LIN0_1:%.+]] = load i64, i64* [[LIN0]]{{.*}}!llvm.access.group
572 // CHECK-NEXT: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
573 // CHECK-NEXT: [[LIN_MUL1:%.+]] = mul nsw i32 [[IV_2]], 3
574 // CHECK-NEXT: [[LIN_EXT1:%.+]] = sext i32 [[LIN_MUL1]] to i64
575 // CHECK-NEXT: [[LIN_ADD1:%.+]] = add nsw i64 [[LIN0_1]], [[LIN_EXT1]]
576 // Update of the privatized version of linear variable!
577 // CHECK-NEXT: store i64 [[LIN_ADD1]], i64* [[K_PRIVATIZED:%[^,]+]]
578     a[k]++;
579     k = k + 3;
580 // CHECK: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
581 // CHECK-NEXT: [[ADD2_2:%.+]] = add nsw i32 [[IV_2]], 1
582 // CHECK-NEXT: store i32 [[ADD2_2]], i32* [[OMP_IV]]{{.*}}!llvm.access.group
583 // br label {{.+}}, !llvm.loop ![[SIMPLE_LOOP_ID]]
584   }
585 // CHECK: [[SIMPLE_LOOP_END]]:
586 //
587 // Update linear vars after loop, as the loop was operating on a private version.
588 // CHECK: [[K_REF:%.+]] = load i64*, i64** [[K_ADDR_REF]],
589 // CHECK: store i64* [[K_REF]], i64** [[K_PRIV_REF:%.+]],
590 // CHECK: [[LIN0_2:%.+]] = load i64, i64* [[LIN0]]
591 // CHECK-NEXT: [[LIN_ADD2:%.+]] = add nsw i64 [[LIN0_2]], 27
592 // CHECK-NEXT: [[K_REF:%.+]] = load i64*, i64** [[K_PRIV_REF]],
593 // CHECK-NEXT: store i64 [[LIN_ADD2]], i64* [[K_REF]]
594 //
595 
596   #pragma omp simd linear(val(k) : 3)
597 // CHECK: [[VAL_REF:%.+]] = load i64*, i64** [[K_ADDR]],
598 // CHECK: store i64* [[VAL_REF]], i64** [[K_ADDR_REF:%.+]],
599 // CHECK: store i32 0, i32* [[OMP_IV:%[^,]+]]
600 // CHECK: [[K_REF:%.+]] = load i64*, i64** [[K_ADDR_REF]],
601 // CHECK: [[K0LOAD:%.+]] = load i64, i64* [[K_REF]]
602 // CHECK-NEXT: store i64 [[K0LOAD]], i64* [[LIN0:%[^,]+]]
603 
604 // CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
605 // CHECK-NEXT: [[CMP2:%.+]] = icmp slt i32 [[IV]], 9
606 // CHECK-NEXT: br i1 [[CMP2]], label %[[SIMPLE_LOOP_BODY:.+]], label %[[SIMPLE_LOOP_END:[^,]+]]
607   for (int i = 10; i > 1; i--) {
608 // CHECK: [[SIMPLE_LOOP_BODY]]:
609 // Start of body: calculate i from IV:
610 // CHECK: [[IV_0:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
611 // FIXME: It is interesting, why the following "mul 1" was not constant folded?
612 // CHECK-NEXT: [[IV_1:%.+]] = mul nsw i32 [[IV_0]], 1
613 // CHECK-NEXT: [[LC_I_1:%.+]] = sub nsw i32 10, [[IV_1]]
614 // CHECK-NEXT: store i32 [[LC_I_1]], i32* {{.+}}, !llvm.access.group
615 //
616 // CHECK-NEXT: [[LIN0_1:%.+]] = load i64, i64* [[LIN0]]{{.*}}!llvm.access.group
617 // CHECK-NEXT: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
618 // CHECK-NEXT: [[LIN_MUL1:%.+]] = mul nsw i32 [[IV_2]], 3
619 // CHECK-NEXT: [[LIN_EXT1:%.+]] = sext i32 [[LIN_MUL1]] to i64
620 // CHECK-NEXT: [[LIN_ADD1:%.+]] = add nsw i64 [[LIN0_1]], [[LIN_EXT1]]
621 // Update of the privatized version of linear variable!
622 // CHECK-NEXT: store i64 [[LIN_ADD1]], i64* [[K_PRIVATIZED:%[^,]+]]
623     a[k]++;
624     k = k + 3;
625 // CHECK: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
626 // CHECK-NEXT: [[ADD2_2:%.+]] = add nsw i32 [[IV_2]], 1
627 // CHECK-NEXT: store i32 [[ADD2_2]], i32* [[OMP_IV]]{{.*}}!llvm.access.group
628 // br label {{.+}}, !llvm.loop ![[SIMPLE_LOOP_ID]]
629   }
630 // CHECK: [[SIMPLE_LOOP_END]]:
631 //
632 // Update linear vars after loop, as the loop was operating on a private version.
633 // CHECK: [[K_REF:%.+]] = load i64*, i64** [[K_ADDR_REF]],
634 // CHECK: store i64* [[K_REF]], i64** [[K_PRIV_REF:%.+]],
635 // CHECK: [[LIN0_2:%.+]] = load i64, i64* [[LIN0]]
636 // CHECK-NEXT: [[LIN_ADD2:%.+]] = add nsw i64 [[LIN0_2]], 27
637 // CHECK-NEXT: [[K_REF:%.+]] = load i64*, i64** [[K_PRIV_REF]],
638 // CHECK-NEXT: store i64 [[LIN_ADD2]], i64* [[K_REF]]
639 //
640   #pragma omp simd linear(uval(k) : 3)
641 // CHECK: store i32 0, i32* [[OMP_IV:%[^,]+]]
642 // CHECK: [[K0LOAD:%.+]] = load i64, i64* [[VAL_ADDR]]
643 // CHECK-NEXT: store i64 [[K0LOAD]], i64* [[LIN0:%[^,]+]]
644 
645 // CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
646 // CHECK-NEXT: [[CMP2:%.+]] = icmp slt i32 [[IV]], 9
647 // CHECK-NEXT: br i1 [[CMP2]], label %[[SIMPLE_LOOP_BODY:.+]], label %[[SIMPLE_LOOP_END:[^,]+]]
648   for (int i = 10; i > 1; i--) {
649 // CHECK: [[SIMPLE_LOOP_BODY]]:
650 // Start of body: calculate i from IV:
651 // CHECK: [[IV_0:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
652 // FIXME: It is interesting, why the following "mul 1" was not constant folded?
653 // CHECK-NEXT: [[IV_1:%.+]] = mul nsw i32 [[IV_0]], 1
654 // CHECK-NEXT: [[LC_I_1:%.+]] = sub nsw i32 10, [[IV_1]]
655 // CHECK-NEXT: store i32 [[LC_I_1]], i32* {{.+}}, !llvm.access.group
656 //
657 // CHECK-NEXT: [[LIN0_1:%.+]] = load i64, i64* [[LIN0]]{{.*}}!llvm.access.group
658 // CHECK-NEXT: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
659 // CHECK-NEXT: [[LIN_MUL1:%.+]] = mul nsw i32 [[IV_2]], 3
660 // CHECK-NEXT: [[LIN_EXT1:%.+]] = sext i32 [[LIN_MUL1]] to i64
661 // CHECK-NEXT: [[LIN_ADD1:%.+]] = add nsw i64 [[LIN0_1]], [[LIN_EXT1]]
662 // Update of the privatized version of linear variable!
663 // CHECK-NEXT: store i64 [[LIN_ADD1]], i64* [[K_PRIVATIZED:%[^,]+]]
664     a[k]++;
665     k = k + 3;
666 // CHECK: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
667 // CHECK-NEXT: [[ADD2_2:%.+]] = add nsw i32 [[IV_2]], 1
668 // CHECK-NEXT: store i32 [[ADD2_2]], i32* [[OMP_IV]]{{.*}}!llvm.access.group
669 // br label {{.+}}, !llvm.loop ![[SIMPLE_LOOP_ID]]
670   }
671 // CHECK: [[SIMPLE_LOOP_END]]:
672 //
673 // Update linear vars after loop, as the loop was operating on a private version.
674 // CHECK: [[LIN0_2:%.+]] = load i64, i64* [[LIN0]]
675 // CHECK-NEXT: [[LIN_ADD2:%.+]] = add nsw i64 [[LIN0_2]], 27
676 // CHECK-NEXT: store i64 [[LIN_ADD2]], i64* [[VAL_ADDR]]
677 //
678 }
679 
680 // TERM_DEBUG-LABEL: bar
681 int bar() {return 0;};
682 
683 // TERM_DEBUG-LABEL: parallel_simd
684 void parallel_simd(float *a) {
685 #pragma omp parallel
686 #pragma omp simd
687   // TERM_DEBUG-NOT: __kmpc_global_thread_num
688   // TERM_DEBUG:     invoke i32 {{.*}}bar{{.*}}()
689   // TERM_DEBUG:     unwind label %[[TERM_LPAD:[^,]+]],
690   // TERM_DEBUG-NOT: __kmpc_global_thread_num
691   // TERM_DEBUG:     [[TERM_LPAD]]
692   // TERM_DEBUG:     call void @__clang_call_terminate
693   // TERM_DEBUG:     unreachable
694   for (unsigned i = 131071; i <= 2147483647; i += 127)
695     a[i] += bar();
696 }
697 // TERM_DEBUG: !{{[0-9]+}} = !DILocation(line: [[@LINE-11]],
698 
699 // CHECK-LABEL: S8
700 // CHECK-DAG: ptrtoint [[SS_TY]]* %{{.+}} to i64
701 // CHECK-DAG: ptrtoint [[SS_TY]]* %{{.+}} to i64
702 // CHECK-DAG: ptrtoint [[SS_TY]]* %{{.+}} to i64
703 // CHECK-DAG: ptrtoint [[SS_TY]]* %{{.+}} to i64
704 
705 // CHECK-DAG: and i64 %{{.+}}, 15
706 // CHECK-DAG: icmp eq i64 %{{.+}}, 0
707 // CHECK-DAG: call void @llvm.assume(i1
708 
709 // CHECK-DAG: and i64 %{{.+}}, 7
710 // CHECK-DAG: icmp eq i64 %{{.+}}, 0
711 // CHECK-DAG: call void @llvm.assume(i1
712 
713 // CHECK-DAG: and i64 %{{.+}}, 15
714 // CHECK-DAG: icmp eq i64 %{{.+}}, 0
715 // CHECK-DAG: call void @llvm.assume(i1
716 
717 // CHECK-DAG: and i64 %{{.+}}, 3
718 // CHECK-DAG: icmp eq i64 %{{.+}}, 0
719 // CHECK-DAG: call void @llvm.assume(i1
720 struct SS {
721   SS(): a(0) {}
722   SS(int v) : a(v) {}
723   int a;
724   typedef int type;
725 };
726 
727 template <typename T>
728 class S7 : public T {
729 protected:
730   T *a;
731   T b[2];
732   S7() : a(0) {}
733 
734 public:
735   S7(typename T::type &v) : a((T*)&v) {
736 #pragma omp simd aligned(a)
737     for (int k = 0; k < a->a; ++k)
738       ++this->a->a;
739 #pragma omp simd aligned(this->b : 8)
740     for (int k = 0; k < a->a; ++k)
741       ++a->a;
742   }
743 };
744 
745 class S8 : private IterDouble, public S7<SS> {
746   S8() {}
747 
748 public:
749   S8(int v) : S7<SS>(v){
750 #pragma omp parallel private(a)
751 #pragma omp simd aligned(S7<SS>::a)
752     for (int k = 0; k < a->a; ++k)
753       ++this->a->a;
754 #pragma omp parallel shared(b)
755 #pragma omp simd aligned(this->b: 4)
756     for (int k = 0; k < a->a; ++k)
757       ++a->a;
758   }
759 };
760 S8 s8(0);
761 
762 // TERM_DEBUG-NOT: line: 0,
763 // TERM_DEBUG: distinct !DISubprogram(linkageName: "_GLOBAL__sub_I_simd_codegen.cpp",
764 
765 #endif // HEADER
766 
767