1 // Test host codegen.
2 // RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64
3 // RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
4 // RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64
5 // RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32
6 // RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
7 // RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32
8
9 // RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
10 // RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
11 // RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
12 // RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
13 // RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
14 // RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
15 // SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
16
17 // Test target codegen - host bc file has to be created first.
18 // RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
19 // RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64
20 // RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
21 // RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64
22 // RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
23 // RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32
24 // RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
25 // RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32
26
27 // RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
28 // RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s
29 // RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
30 // RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
31 // RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
32 // RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s
33 // RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
34 // RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
35 // SIMD-ONLY1-NOT: {{__kmpc|__tgt}}
36
37 // expected-no-diagnostics
38 #ifndef HEADER
39 #define HEADER
40
41 template <typename tx, typename ty>
42 struct TT {
43 tx X;
44 ty Y;
45 };
46 #pragma omp declare target
47 int ga = 5;
48 #pragma omp end declare target
49
50 // CHECK-DAG: [[TT:%.+]] = type { i64, i8 }
51 // CHECK-DAG: [[TTII:%.+]] = type { i32, i32 }
52 // CHECK-DAG: [[S1:%.+]] = type { double }
53
54 // TCHECK-DAG: [[TT:%.+]] = type { i64, i8 }
55 // TCHECK-DAG: [[TTII:%.+]] = type { i32, i32 }
56 // TCHECK-DAG: [[S1:%.+]] = type { double }
57
58 // CHECK-DAG: [[SIZET:@.+]] = private unnamed_addr constant [3 x i{{32|64}}] [i[[SZ:32|64]] 4, i{{64|32}} {{8|4}}, i[[SZ:32|64]] 4]
59 // CHECK-DAG: [[MAPT:@.+]] = private unnamed_addr constant [3 x i64] [i64 288, i64 49, i64 288]
60 // CHECK-DAG: [[SIZET2:@.+]] = private unnamed_addr constant [9 x i64] [i64 2, i64 40, i64 {{4|8}}, i64 0, i64 400, i64 {{4|8}}, i64 {{4|8}}, i64 0, i64 {{12|16}}]
61 // CHECK-DAG: [[MAPT2:@.+]] = private unnamed_addr constant [9 x i64] [i64 288, i64 161, i64 800, i64 161, i64 161, i64 800, i64 800, i64 161, i64 161]
62 // CHECK-DAG: [[SIZET3:@.+]] = private unnamed_addr constant [2 x i{{32|64}}] [i{{32|64}} 0, i{{32|64}} 8]
63 // CHECK-DAG: [[MAPT3:@.+]] = private unnamed_addr constant [2 x i64] [i64 32, i64 161]
64 // CHECK-DAG: [[SIZET4:@.+]] = private unnamed_addr constant [5 x i64] [i64 8, i64 4, i64 {{4|8}}, i64 {{4|8}}, i64 0]
65 // CHECK-DAG: [[MAPT4:@.+]] = private unnamed_addr constant [5 x i64] [i64 547, i64 288, i64 800, i64 800, i64 161]
66 // CHECK-DAG: [[SIZET5:@.+]] = private unnamed_addr constant [3 x i{{32|64}}] [i[[SZ]] 4, i[[SZ]] 1, i[[SZ]] 40]
67 // CHECK-DAG: [[MAPT5:@.+]] = private unnamed_addr constant [3 x i64] [i64 288, i64 288, i64 161]
68 // CHECK-DAG: [[SIZET6:@.+]] = private unnamed_addr constant [2 x i{{32|64}}] [i[[SZ]] 4, i[[SZ]] 40]
69 // CHECK-DAG: [[MAPT6:@.+]] = private unnamed_addr constant [2 x i64] [i64 288, i64 161]
70
71 // CHECK: define {{.*}}[[FOO:@.+]](
foo(int n,double * ptr)72 int foo(int n, double *ptr) {
73 int a = 0;
74 short aa = 0;
75 float b[10];
76 float bn[n];
77 double c[5][10];
78 double cn[5][n];
79 TT<long long, char> d;
80 const TT<int, int> e = {n, n};
81 int *p __attribute__ ((aligned (64))) = &a;
82
83 #pragma omp target firstprivate(a, p, ga)
84 {
85 }
86
87 // a is passed by value to tgt_target
88 // CHECK: [[N_ADDR:%.+]] = alloca i{{[0-9]+}},
89 // CHECK: [[PTR_ADDR:%.+]] = alloca double*,
90 // CHECK: [[A:%.+]] = alloca i{{[0-9]+}},
91 // CHECK: [[A2:%.+]] = alloca i{{[0-9]+}},
92 // CHECK: [[B:%.+]] = alloca [10 x float],
93 // CHECK: [[SSTACK:%.+]] = alloca i8*,
94 // CHECK: [[C:%.+]] = alloca [5 x [10 x double]],
95 // CHECK: [[D:%.+]] = alloca [[TT]],
96 // CHECK: [[FP_E:%.+]] = alloca [[TTII]],
97 // CHECK: [[P:%.+]] = alloca i32*, align 64
98 // CHECK: [[ACAST:%.+]] = alloca i{{[0-9]+}},
99 // CHECK: [[BASE_PTR_ARR:%.+]] = alloca [3 x i8*],
100 // CHECK: [[PTR_ARR:%.+]] = alloca [3 x i8*],
101 // CHECK: [[A2CAST:%.+]] = alloca i{{[0-9]+}},
102 // CHECK: [[BASE_PTR_ARR2:%.+]] = alloca [9 x i8*],
103 // CHECK: [[PTR_ARR2:%.+]] = alloca [9 x i8*],
104 // CHECK: [[SIZET2:%.+]] = alloca [9 x i{{[0-9]+}}],
105 // CHECK: [[BASE_PTR_ARR3:%.+]] = alloca [2 x i8*],
106 // CHECK: [[PTR_ARR3:%.+]] = alloca [2 x i8*],
107 // CHECK: [[N_ADDR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[N_ADDR]],
108 // CHECK-64: [[N_EXT:%.+]] = zext i{{[0-9]+}} [[N_ADDR_VAL]] to i{{[0-9]+}}
109 // CHECK: [[SSAVE_RET:%.+]] = call i8* @llvm.stacksave()
110 // CHECK: store i8* [[SSAVE_RET]], i8** [[SSTACK]],
111 // CHECK-64: [[BN_VLA:%.+]] = alloca float, i{{[0-9]+}} [[N_EXT]],
112 // CHECK-32: [[BN_VLA:%.+]] = alloca float, i{{[0-9]+}} [[N_ADDR_VAL]],
113 // CHECK: [[N_ADDR_VAL2:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[N_ADDR]],
114 // CHECK-64: [[N_EXT2:%.+]] = zext i{{[0-9]+}} [[N_ADDR_VAL2]] to i{{[0-9]+}}
115 // CHECK-64: [[CN_SIZE:%.+]] = mul{{.+}} i{{[0-9]+}} 5, [[N_EXT2]]
116 // CHECK-32: [[CN_SIZE:%.+]] = mul{{.+}} i{{[0-9]+}} 5, [[N_ADDR_VAL2]]
117 // CHECK: [[CN_VLA:%.+]] = alloca double, i{{[0-9]+}} [[CN_SIZE]],
118 // CHECK: [[AVAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[A]],
119 // CHECK-64: [[CONV:%.+]] = bitcast i{{[0-9]+}}* [[ACAST]] to i{{[0-9]+}}*
120 // CHECK-64: store i{{[0-9]+}} [[AVAL]], i{{[0-9]+}}* [[CONV]],
121 // CHECK-32: store i{{[0-9]+}} [[AVAL]], i{{[0-9]+}}* [[ACAST]],
122 // CHECK: [[ACAST_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[ACAST]],
123 // CHECK: [[P_PTR:%.+]] = load i32*, i32** [[P]], align 64
124 // CHECK: [[BASE_PTR_GEP:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BASE_PTR_ARR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
125 // CHECK: [[ACAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP]] to i{{[0-9]+}}*
126 // CHECK: store i{{[0-9]+}} [[ACAST_VAL]], i{{[0-9]+}}* [[ACAST_TOPTR]],
127 // CHECK: [[PTR_GEP:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PTR_ARR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
128 // CHECK: [[ACAST_TOPTR2:%.+]] = bitcast i8** [[PTR_GEP]] to i{{[0-9]+}}*
129 // CHECK: store i{{[0-9]+}} [[ACAST_VAL]], i{{[0-9]+}}* [[ACAST_TOPTR2]],
130 // CHECK: [[BASE_PTR_GEP:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BASE_PTR_ARR]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
131 // CHECK: [[PCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP]] to i32***
132 // CHECK: store i32** [[P]], i32*** [[PCAST_TOPTR]],
133 // CHECK: [[PTR_GEP:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PTR_ARR]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
134 // CHECK: [[PCAST_TOPTR2:%.+]] = bitcast i8** [[PTR_GEP]] to i32**
135 // CHECK: store i32* [[P_PTR]], i32** [[PCAST_TOPTR2]],
136 // CHECK: [[BASE_PTR_GEP:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BASE_PTR_ARR]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
137 // CHECK: [[PCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP]] to i{{64|32}}*
138 // CHECK: store i{{64|32}} [[GA_VAL:%.*]], i{{64|32}}* [[PCAST_TOPTR]],
139 // CHECK: [[PTR_GEP:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PTR_ARR]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
140 // CHECK: [[PCAST_TOPTR2:%.+]] = bitcast i8** [[PTR_GEP]] to i{{64|32}}*
141 // CHECK: store i{{64|32}} [[GA_VAL]], i{{64|32}}* [[PCAST_TOPTR2]],
142 // CHECK: [[BASE_PTR_GEP_ARG:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BASE_PTR_ARR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
143 // CHECK: [[PTR_GEP_ARG:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PTR_ARR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
144 // CHECK: {{.+}} = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 -1, i32 -1, i32 0, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]])
145
146 // TCHECK: define weak_odr void @__omp_offloading_{{.+}}(i{{[0-9]+}} noundef [[A_IN:%.+]], i32** noundef nonnull align {{[0-9]+}} dereferenceable({{[0-9]+}}) [[P_IN:%.+]], i{{[0-9]+}} noundef [[GA_IN:%.+]])
147 // TCHECK: [[A_ADDR:%.+]] = alloca i{{[0-9]+}},
148 // TCHECK: [[P_ADDR:%.+]] = alloca i32**,
149 // TCHECK: [[GA_ADDR:%.+]] = alloca i{{64|32}},
150 // TCHECK: [[P_PRIV:%.+]] = alloca i32*,
151 // TCHECK-NOT: alloca i{{[0-9]+}}
152 // TCHECK: store i{{[0-9]+}} [[A_IN]], i{{[0-9]+}}* [[A_ADDR]],
153 // TCHECK: store i32** [[P_IN]], i32*** [[P_ADDR]],
154 // TCHECK: store i{{[0-9]+}} [[GA_IN]], i{{[0-9]+}}* [[GA_ADDR]],
155 // TCHECK-NOT: store i{{[0-9]+}} %
156 // TCHECK: ret void
157
158 #pragma omp target firstprivate(aa, b, bn, c, cn, d)
159 {
160 aa += 1;
161 b[2] = 1.0;
162 bn[3] = 1.0;
163 c[1][2] = 1.0;
164 cn[1][3] = 1.0;
165 d.X = 1;
166 d.Y = 1;
167 }
168
169 // CHECK: [[A2VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[A2]],
170 // CHECK: [[A2CASTCONV:%.+]] = bitcast i{{[0-9]+}}* [[A2CAST]] to i{{[0-9]+}}*
171 // CHECK: store i{{[0-9]+}} [[A2VAL]], i{{[0-9]+}}* [[A2CASTCONV]],
172 // CHECK: [[A2CAST_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[A2CAST]],
173 // CHECK-64: [[BN_SIZE:%.+]] = mul{{.+}} i{{[0-9]+}} [[N_EXT]], 4
174 // CHECK-32: [[BN_SZ_SIZE:%.+]] = mul{{.+}} i{{[0-9]+}} [[N_ADDR_VAL]], 4
175 // CHECK-32: [[BN_SIZE:%.+]] = sext i32 [[BN_SZ_SIZE]] to i64
176 // CHECK-64: [[CN_SIZE_1:%.+]] = mul{{.+}} i{{[0-9]+}} 5, [[N_EXT2]]
177 // CHECK-32: [[CN_SIZE_1:%.+]] = mul{{.+}} i{{[0-9]+}} 5, [[N_ADDR_VAL2]]
178 // CHECK-64: [[CN_SIZE_2:%.+]] = mul{{.+}} i{{[0-9]+}} [[CN_SIZE_1]], 8
179 // CHECK-32: [[CN_SZ_SIZE_2:%.+]] = mul{{.+}} i{{[0-9]+}} [[CN_SIZE_1]], 8
180 // CHECK-32: [[CN_SIZE_2:%.+]] = sext i32 [[CN_SZ_SIZE_2]] to i64
181
182 // firstprivate(aa) --> base_ptr = aa, ptr = aa, size = 2 (short)
183 // CHECK: [[BASE_PTR_GEP2_0:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
184 // CHECK: [[ACAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_0]] to i{{[0-9]+}}*
185 // CHECK: store i{{[0-9]+}} [[A2CAST_VAL]], i{{[0-9]+}}* [[ACAST_TOPTR]],
186 // CHECK: [[PTR_GEP2_0:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
187 // CHECK: [[ACAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_0]] to i{{[0-9]+}}*
188 // CHECK: store i{{[0-9]+}} [[A2CAST_VAL]], i{{[0-9]+}}* [[ACAST_TOPTR]],
189
190 // firstprivate(b): base_ptr = &b[0], ptr = &b[0], size = 40 (sizeof(float)*10)
191 // CHECK: [[BASE_PTR_GEP2_1:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
192 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_1]] to [10 x float]**
193 // CHECK: store [10 x float]* [[B]], [10 x float]** [[BCAST_TOPTR]],
194 // CHECK: [[PTR_GEP2_1:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
195 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_1]] to [10 x float]**
196 // CHECK: store [10 x float]* [[B]], [10 x float]** [[BCAST_TOPTR]],
197
198 // firstprivate(bn), 2 entries, n and bn: (1) base_ptr = n, ptr = n, size = 8 ; (2) base_ptr = &c[0], ptr = &c[0], size = n*sizeof(float)
199 // CHECK: [[BASE_PTR_GEP2_2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
200 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_2]] to i{{[0-9]+}}*
201 // CHECK-64: store i{{[0-9]+}} [[N_EXT]], i{{[0-9]+}}* [[BCAST_TOPTR]],
202 // CHECK-32: store i{{[0-9]+}} [[N_ADDR_VAL]], i{{[0-9]+}}* [[BCAST_TOPTR]],
203 // CHECK: [[PTR_GEP2_2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
204 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_2]] to i{{[0-9]+}}*
205 // CHECK-64: store i{{[0-9]+}} [[N_EXT]], i{{[0-9]+}}* [[BCAST_TOPTR]],
206 // CHECK-32: store i{{[0-9]+}} [[N_ADDR_VAL]], i{{[0-9]+}}* [[BCAST_TOPTR]],
207 // CHECK: [[BASE_PTR_GEP2_3:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 3
208 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_3]] to float**
209 // CHECK: store float* [[BN_VLA]], float** [[BCAST_TOPTR]],
210 // CHECK: [[SIZE_GEPBN_3:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 3
211 // CHECK: store i{{[0-9]+}} [[BN_SIZE]], i{{[0-9]+}}* [[SIZE_GEPBN_3]]
212
213 // firstprivate(c): base_ptr = &c[0], ptr = &c[0], size = 400 (5*10*sizeof(double))
214 // CHECK: [[BASE_PTR_GEP2_4:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 4
215 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_4]] to [5 x [10 x double]]**
216 // CHECK: store [5 x [10 x double]]* [[C]], [5 x [10 x double]]** [[BCAST_TOPTR]],
217 // CHECK: [[PTR_GEP2_4:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 4
218 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_4]] to [5 x [10 x double]]**
219 // CHECK: store [5 x [10 x double]]* [[C]], [5 x [10 x double]]** [[BCAST_TOPTR]],
220
221 // firstprivate(cn), 3 entries, 5, n, cn: (1) base_ptr = 5, ptr = 5, size = 8; (2) (1) base_ptr = n, ptr = n, size = 8; (3) base_ptr = &cn[0], ptr = &cn[0], size = 5*n*sizeof(double)
222 // CHECK: [[BASE_PTR_GEP2_5:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 5
223 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_5]] to i{{[0-9]+}}*
224 // CHECK: store i{{[0-9]+}} 5, i{{[0-9]+}}* [[BCAST_TOPTR]],
225 // CHECK: [[PTR_GEP2_5:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 5
226 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_5]] to i{{[0-9]+}}*
227 // CHECK: store i{{[0-9]+}} 5, i{{[0-9]+}}* [[BCAST_TOPTR]],
228 // CHECK: [[BASE_PTR_GEP2_6:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 6
229 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_6]] to i{{[0-9]+}}*
230 // CHECK-64: store i{{[0-9]+}} [[N_EXT2]], i{{[0-9]+}}* [[BCAST_TOPTR]],
231 // CHECK-32: store i{{[0-9]+}} [[N_ADDR_VAL2]], i{{[0-9]+}}* [[BCAST_TOPTR]],
232 // CHECK: [[PTR_GEP2_6:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 6
233 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_6]] to i{{[0-9]+}}*
234 // CHECK-64: store i{{[0-9]+}} [[N_EXT2]], i{{[0-9]+}}* [[BCAST_TOPTR]],
235 // CHECK-32: store i{{[0-9]+}} [[N_ADDR_VAL2]], i{{[0-9]+}}* [[BCAST_TOPTR]],
236 // CHECK: [[BASE_PTR_GEP2_7:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 7
237 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_7]] to double**
238 // CHECK: store double* [[CN_VLA]], double** [[BCAST_TOPTR]],
239 // CHECK: [[PTR_GEP2_7:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 7
240 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_7]] to double**
241 // CHECK: store double* [[CN_VLA]], double** [[BCAST_TOPTR]],
242 // CHECK: [[SIZE_GEPCN_7:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 7
243 // CHECK: store i{{[0-9]+}} [[CN_SIZE_2]], i{{[0-9]+}}* [[SIZE_GEPCN_7]],
244
245 // firstprivate(d): base_ptr = &d, ptr = &d, size = 16
246 // CHECK: [[BASE_PTR_GEP2_8:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 8
247 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_8]] to [[TT]]**
248 // CHECK: store [[TT]]* [[D]], [[TT]]** [[BCAST_TOPTR]],
249 // CHECK: [[PTR_GEP2_8:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 8
250 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_8]] to [[TT]]**
251 // CHECK: store [[TT]]* [[D]], [[TT]]** [[BCAST_TOPTR]],
252
253 // CHECK: [[BASE_PTR_GEP_ARG2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
254 // CHECK: [[PTR_GEP_ARG2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
255 // CHECK: [[SIZES_ARG2:%.+]] = getelementptr inbounds [9 x i[[SZ]]], [9 x i[[SZ]]]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
256 // CHECK: {{.+}} = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 -1, i32 -1, i32 0, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]])
257
258 // make sure that firstprivate variables are generated in all cases and that we use those instances for operations inside the
259 // target region
260 // TCHECK: define {{.*}}void @__omp_offloading_{{.+}}(i{{[0-9]+}} noundef [[A2_IN:%.+]], [10 x float]* {{.+}} [[B_IN:%.+]], i{{[0-9]+}} noundef [[BN_SZ:%.+]], float* {{.+}} [[BN_IN:%.+]], [5 x [10 x double]]* {{.+}} [[C_IN:%.+]], i{{[0-9]+}} noundef [[CN_SZ1:%.+]], i{{[0-9]+}} noundef [[CN_SZ2:%.+]], double* {{.+}} [[CN_IN:%.+]], [[TT]]* {{.+}} [[D_IN:%.+]])
261 // TCHECK: [[A2_ADDR:%.+]] = alloca i{{[0-9]+}},
262 // TCHECK: [[B_ADDR:%.+]] = alloca [10 x float]*,
263 // TCHECK: [[VLA_ADDR:%.+]] = alloca i{{[0-9]+}},
264 // TCHECK: [[BN_ADDR:%.+]] = alloca float*,
265 // TCHECK: [[C_ADDR:%.+]] = alloca [5 x [10 x double]]*,
266 // TCHECK: [[VLA_ADDR2:%.+]] = alloca i{{[0-9]+}},
267 // TCHECK: [[VLA_ADDR4:%.+]] = alloca i{{[0-9]+}},
268 // TCHECK: [[CN_ADDR:%.+]] = alloca double*,
269 // TCHECK: [[D_ADDR:%.+]] = alloca [[TT]]*,
270 // TCHECK-NOT: alloca i{{[0-9]+}},
271 // TCHECK: [[B_PRIV:%.+]] = alloca [10 x float],
272 // TCHECK: [[SSTACK:%.+]] = alloca i8*,
273 // TCHECK: [[C_PRIV:%.+]] = alloca [5 x [10 x double]],
274 // TCHECK: [[D_PRIV:%.+]] = alloca [[TT]],
275 // TCHECK: store i{{[0-9]+}} [[A2_IN]], i{{[0-9]+}}* [[A2_ADDR]],
276 // TCHECK: store [10 x float]* [[B_IN]], [10 x float]** [[B_ADDR]],
277 // TCHECK: store i{{[0-9]+}} [[BN_SZ]], i{{[0-9]+}}* [[VLA_ADDR]],
278 // TCHECK: store float* [[BN_IN]], float** [[BN_ADDR]],
279 // TCHECK: store [5 x [10 x double]]* [[C_IN]], [5 x [10 x double]]** [[C_ADDR]],
280 // TCHECK: store i{{[0-9]+}} [[CN_SZ1]], i{{[0-9]+}}* [[VLA_ADDR2]],
281 // TCHECK: store i{{[0-9]+}} [[CN_SZ2]], i{{[0-9]+}}* [[VLA_ADDR4]],
282 // TCHECK: store double* [[CN_IN]], double** [[CN_ADDR]],
283 // TCHECK: store [[TT]]* [[D_IN]], [[TT]]** [[D_ADDR]],
284 // TCHECK: [[CONV_A2ADDR:%.+]] = bitcast i{{[0-9]+}}* [[A2_ADDR]] to i{{[0-9]+}}*
285 // TCHECK: [[B_ADDR_REF:%.+]] = load [10 x float]*, [10 x float]** [[B_ADDR]],
286 // TCHECK: [[BN_SZ_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR]],
287 // TCHECK: [[BN_ADDR_REF:%.+]] = load float*, float** [[BN_ADDR]],
288 // TCHECK: [[C_ADDR_REF:%.+]] = load [5 x [10 x double]]*, [5 x [10 x double]]** [[C_ADDR]],
289 // TCHECK: [[CN_SZ1_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR2]],
290 // TCHECK: [[CN_SZ2_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR4]],
291 // TCHECK: [[CN_ADDR_REF:%.+]] = load double*, double** [[CN_ADDR]],
292 // TCHECK: [[D_ADDR_REF:%.+]] = load [[TT]]*, [[TT]]** [[D_ADDR]],
293
294 // firstprivate(aa): a_priv = a_in
295 // TCHECK-NOT: store i{{[0-9]+}} %
296
297 // firstprivate(b): memcpy(b_priv,b_in)
298 // TCHECK: [[B_PRIV_BCAST:%.+]] = bitcast [10 x float]* [[B_PRIV]] to i8*
299 // TCHECK: [[B_ADDR_REF_BCAST:%.+]] = bitcast [10 x float]* [[B_ADDR_REF]] to i8*
300 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[B_PRIV_BCAST]], i8* align {{[0-9]+}} [[B_ADDR_REF_BCAST]], {{.+}})
301
302 // TCHECK: [[RET_STACK:%.+]] = call i8* @llvm.stacksave()
303 // TCHECK: store i8* [[RET_STACK]], i8** [[SSTACK]],
304
305 // firstprivate(bn)
306 // TCHECK: [[BN_PRIV:%.+]] = alloca float, i{{[0-9]+}} [[BN_SZ_VAL]],
307 // TCHECK: [[BN_COPY_SZ:%.+]] = mul{{.+}} i{{[0-9]+}} [[BN_SZ_VAL]], 4
308 // TCHECK: [[BN_PRIV__BCAST:%.+]] = bitcast float* [[BN_PRIV]] to i8*
309 // TCHECK: [[BN_REF_IN_BCAST:%.+]] = bitcast float* [[BN_ADDR_REF]] to i8*
310 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BN_PRIV__BCAST]], i8* align {{[0-9]+}} [[BN_REF_IN_BCAST]], i{{[0-9]+}} [[BN_COPY_SZ]],{{.+}})
311
312 // firstprivate(c)
313 // TCHECK: [[C_PRIV_BCAST:%.+]] = bitcast [5 x [10 x double]]* [[C_PRIV]] to i8*
314 // TCHECK: [[C_IN_BCAST:%.+]] = bitcast [5 x [10 x double]]* [[C_ADDR_REF]] to i8*
315 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[C_PRIV_BCAST]], i8* align {{[0-9]+}} [[C_IN_BCAST]],{{.+}})
316
317 // firstprivate(cn)
318 // TCHECK: [[CN_SZ:%.+]] = mul{{.+}} i{{[0-9]+}} [[CN_SZ1_VAL]], [[CN_SZ2_VAL]]
319 // TCHECK: [[CN_PRIV:%.+]] = alloca double, i{{[0-9]+}} [[CN_SZ]],
320 // TCHECK: [[CN_SZ2:%.+]] = mul{{.+}} i{{[0-9]+}} [[CN_SZ1_VAL]], [[CN_SZ2_VAL]]
321 // TCHECK: [[CN_SZ2_CPY:%.+]] = mul{{.+}} i{{[0-9]+}} [[CN_SZ2]], 8
322 // TCHECK: [[CN_PRIV_BCAST:%.+]] = bitcast double* [[CN_PRIV]] to i8*
323 // TCHECK: [[CN_IN_BCAST:%.+]] = bitcast double* [[CN_ADDR_REF]] to i8*
324 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[CN_PRIV_BCAST]], i8* align {{[0-9]+}} [[CN_IN_BCAST]], i{{[0-9]+}} [[CN_SZ2_CPY]],{{.+}})
325
326 // firstprivate(d)
327 // TCHECK: [[D_PRIV_BCAST:%.+]] = bitcast [[TT]]* [[D_PRIV]] to i8*
328 // TCHECK: [[D_IN_BCAST:%.+]] = bitcast [[TT]]* [[D_ADDR_REF]] to i8*
329 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[D_PRIV_BCAST]], i8* align {{[0-9]+}} [[D_IN_BCAST]],{{.+}})
330
331 #pragma omp target firstprivate(ptr, e)
332 {
333 ptr[0] = e.X;
334 ptr[0]++;
335 }
336 // CHECK: [[PTR_ADDR_REF:%.+]] = load double*, double** [[PTR_ADDR]],
337
338 // CHECK: [[BASE_PTR_GEP3_0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BASE_PTR_ARR3]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
339 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP3_0]] to double**
340 // CHECK: store double* [[PTR_ADDR_REF]], double** [[BCAST_TOPTR]],
341 // CHECK: [[PTR_GEP3_0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[PTR_ARR3]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
342 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP3_0]] to double**
343 // CHECK: store double* [[PTR_ADDR_REF]], double** [[BCAST_TOPTR]],
344 // CHECK: [[BASE_PTR_GEP3_1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BASE_PTR_ARR3]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
345 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP3_1]] to [[TTII]]**
346 // CHECK: store [[TTII]]* [[FP_E]], [[TTII]]** [[BCAST_TOPTR]],
347 // CHECK: [[PTR_GEP3_1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[PTR_ARR3]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
348 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP3_1]] to [[TTII]]**
349 // CHECK: store [[TTII]]* [[FP_E]], [[TTII]]** [[BCAST_TOPTR]],
350
351 // CHECK: [[BASE_PTR_GEP_ARG3:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BASE_PTR_ARR3]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
352 // CHECK: [[PTR_GEP_ARG3:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[PTR_ARR3]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
353 // CHECK: {{.+}} = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 -1, i32 -1, i32 0, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]])
354
355 // TCHECK: define weak_odr void @__omp_offloading_{{.+}}(double* noundef [[PTR_IN:%.+]], [[TTII]]* noundef nonnull align {{[0-9]+}} dereferenceable({{[0-9]+}}) [[E:%.+]])
356 // TCHECK: [[PTR_ADDR:%.+]] = alloca double*,
357 // TCHECK-NOT: alloca [[TTII]],
358 // TCHECK-NOT: alloca double*,
359 // TCHECK: store double* [[PTR_IN]], double** [[PTR_ADDR]],
360 // TCHECK-NOT: store double* %
361
362 return a;
363 }
364
365 template <typename tx>
ftemplate(int n)366 tx ftemplate(int n) {
367 tx a = 0;
368 tx b[10];
369
370 #pragma omp target firstprivate(a, b)
371 {
372 a += 1;
373 b[2] += 1;
374 }
375
376 return a;
377 }
378
fstatic(int n)379 static int fstatic(int n) {
380 int a = 0;
381 char aaa = 0;
382 int b[10];
383
384 #pragma omp target firstprivate(a, aaa, b)
385 {
386 a += 1;
387 aaa += 1;
388 b[2] += 1;
389 }
390
391 return a;
392 }
393
394 // TCHECK: define weak_odr void @__omp_offloading_{{.+}}(i{{[0-9]+}} noundef [[A_IN:%.+]], i{{[0-9]+}} noundef [[A3_IN:%.+]], [10 x i{{[0-9]+}}]*{{.+}} [[B_IN:%.+]])
395 // TCHECK: [[A_ADDR:%.+]] = alloca i{{[0-9]+}},
396 // TCHECK: [[A3_ADDR:%.+]] = alloca i{{[0-9]+}},
397 // TCHECK: [[B_ADDR:%.+]] = alloca [10 x i{{[0-9]+}}]*,
398 // TCHECK-NOT: alloca i{{[0-9]+}},
399 // TCHECK: [[B_PRIV:%.+]] = alloca [10 x i{{[0-9]+}}],
400 // TCHECK: store i{{[0-9]+}} [[A_IN]], i{{[0-9]+}}* [[A_ADDR]],
401 // TCHECK: store i{{[0-9]+}} [[A3_IN]], i{{[0-9]+}}* [[A3_ADDR]],
402 // TCHECK: store [10 x i{{[0-9]+}}]* [[B_IN]], [10 x i{{[0-9]+}}]** [[B_ADDR]],
403 // TCHECK-64: [[A_CONV:%.+]] = bitcast i{{[0-9]+}}* [[A_ADDR]] to i{{[0-9]+}}*
404 // TCHECK: [[A3_CONV:%.+]] = bitcast i{{[0-9]+}}* [[A3_ADDR]] to i8*
405 // TCHECK: [[B_ADDR_REF:%.+]] = load [10 x i{{[0-9]+}}]*, [10 x i{{[0-9]+}}]** [[B_ADDR]],
406
407 // firstprivate(a): a_priv = a_in
408
409 // firstprivate(aaa)
410 // TCHECK-NOT: store i{{[0-9]+}} %
411
412 // firstprivate(b)
413 // TCHECK: [[B_PRIV_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_PRIV]] to i8*
414 // TCHECK: [[B_IN_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_ADDR_REF]] to i8*
415 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[B_PRIV_BCAST]], i8* align {{[0-9]+}} [[B_IN_BCAST]],{{.+}})
416
417 // TCHECK: ret void
418
419 struct S1 {
420 double a;
421
r1S1422 int r1(int n) {
423 int b = n + 1;
424 short int c[2][n];
425
426 #pragma omp target firstprivate(b, c)
427 {
428 this->a = (double)b + 1.5;
429 c[1][1] = ++a;
430 }
431
432 return c[1][1] + (int)b;
433 }
434
435 // on the host side, we first generate r1, then the static function and the template above
436 // CHECK: define{{.+}} i32 {{.+}}([[S1]]* {{.+}}, i{{[0-9]+}} {{.+}})
437 // CHECK: [[BASE_PTRS4:%.+]] = alloca [5 x i8*],
438 // CHECK: [[PTRS4:%.+]] = alloca [5 x i8*],
439 // CHECK: [[SIZET4:%.+]] = alloca [5 x i{{[0-9]+}}],
440
441 // map(this: this ptr is implicitly captured (not firstprivate matter)
442 // CHECK: [[BP0:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
443 // CHECK: [[CBP0:%.+]] = bitcast i8** [[BP0]] to %struct.S1**
444 // CHECK: store %struct.S1* [[THIS:%.+]], %struct.S1** [[CBP0]],
445 // CHECK: [[P0:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
446 // CHECK: [[CP0:%.+]] = bitcast i8** [[P0]] to double**
447 // CHECK: store double* [[A:%.+]], double** [[CP0]],
448
449 // firstprivate(b): base_ptr = b, ptr = b, size = 4 (pass by-value)
450 // CHECK: [[BASE_PTRS_GEP4_1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
451 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP4_1]] to i{{[0-9]+}}*
452 // CHECK: store i{{[0-9]+}} [[B_CAST:%.+]], i{{[0-9]+}}* [[BCAST_TOPTR]],
453 // CHECK: [[PTRS_GEP4_1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
454 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP4_1]] to i{{[0-9]+}}*
455 // CHECK: store i{{[0-9]+}} [[B_CAST]], i{{[0-9]+}}* [[BCAST_TOPTR]],
456
457 // firstprivate(c), 3 entries: 2, n, c
458 // CHECK: [[BASE_PTRS_GEP4_2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
459 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP4_2]] to i{{[0-9]+}}*
460 // CHECK: store i{{[0-9]+}} 2, i{{[0-9]+}}* [[BCAST_TOPTR]],
461 // CHECK: [[PTRS_GEP4_2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
462 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP4_2]] to i{{[0-9]+}}*
463 // CHECK: store i{{[0-9]+}} 2, i{{[0-9]+}}* [[BCAST_TOPTR]],
464 // CHECK: [[BASE_PTRS_GEP4_3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 3
465 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP4_3]] to i{{[0-9]+}}*
466 // CHECK: store i{{[0-9]+}} [[N:%.+]], i{{[0-9]+}}* [[BCAST_TOPTR]],
467 // CHECK: [[PTRS_GEP4_3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 3
468 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP4_3]] to i{{[0-9]+}}*
469 // CHECK: store i{{[0-9]+}} [[N]], i{{[0-9]+}}* [[BCAST_TOPTR]],
470 // CHECK: [[BASE_PTRS_GEP4_4:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 4
471 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP4_4]] to i{{[0-9]+}}**
472 // CHECK: store i{{[0-9]+}}* [[B:%.+]], i{{[0-9]+}}** [[BCAST_TOPTR]],
473 // CHECK: [[PTRS_GEP4_4:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 4
474 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP4_4]] to i{{[0-9]+}}**
475 // CHECK: store i{{[0-9]+}}* [[B]], i{{[0-9]+}}** [[BCAST_TOPTR]],
476 // CHECK: [[SIZES_GEP4_4:%.+]] = getelementptr inbounds [5 x i{{[0-9]+}}], [5 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 4
477 // CHECK: store i{{[0-9]+}} [[B_SIZE:%.+]], i{{[0-9]+}}* [[SIZES_GEP4_4]],
478
479 // only check that we use the map types stored in the global variable
480 // CHECK: {{.+}} = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 -1, i32 -1, i32 0, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]])
481
482 // TCHECK: define weak_odr void @__omp_offloading_{{.+}}([[S1]]* noundef [[TH:%.+]], i{{[0-9]+}} noundef [[B_IN:%.+]], i{{[0-9]+}} noundef [[VLA:%.+]], i{{[0-9]+}} noundef [[VLA1:%.+]], i{{[0-9]+}}{{.+}} [[C_IN:%.+]])
483 // TCHECK: [[TH_ADDR:%.+]] = alloca [[S1]]*,
484 // TCHECK: [[B_ADDR:%.+]] = alloca i{{[0-9]+}},
485 // TCHECK: [[VLA_ADDR:%.+]] = alloca i{{[0-9]+}},
486 // TCHECK: [[VLA_ADDR2:%.+]] = alloca i{{[0-9]+}},
487 // TCHECK: [[C_ADDR:%.+]] = alloca i{{[0-9]+}}*,
488 // TCHECK-NOT: alloca i{{[0-9]+}},
489 // TCHECK: [[SSTACK:%.+]] = alloca i8*,
490
491 // TCHECK: store [[S1]]* [[TH]], [[S1]]** [[TH_ADDR]],
492 // TCHECK: store i{{[0-9]+}} [[B_IN]], i{{[0-9]+}}* [[B_ADDR]],
493 // TCHECK: store i{{[0-9]+}} [[VLA]], i{{[0-9]+}}* [[VLA_ADDR]],
494 // TCHECK: store i{{[0-9]+}} [[VLA1]], i{{[0-9]+}}* [[VLA_ADDR2]],
495 // TCHECK: store i{{[0-9]+}}* [[C_IN]], i{{[0-9]+}}** [[C_ADDR]],
496 // TCHECK: [[TH_ADDR_REF:%.+]] = load [[S1]]*, [[S1]]** [[TH_ADDR]],
497 // TCHECK-64: [[B_ADDR_CONV:%.+]] = bitcast i{{[0-9]+}}* [[B_ADDR]] to i{{[0-9]+}}*
498 // TCHECK: [[VLA_ADDR_REF:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR]],
499 // TCHECK: [[VLA_ADDR_REF2:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR2]],
500 // TCHECK: [[C_ADDR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[C_ADDR]],
501
502 // firstprivate(b)
503 // TCHECK-NOT: store i{{[0-9]+}} %
504
505 // TCHECK: [[RET_STACK:%.+]] = call i8* @llvm.stacksave()
506 // TCHECK: store i8* [[RET_STACK:%.+]], i8** [[SSTACK]],
507
508 // firstprivate(c)
509 // TCHECK: [[C_SZ:%.+]] = mul{{.+}} i{{[0-9]+}} [[VLA_ADDR_REF]], [[VLA_ADDR_REF2]]
510 // TCHECK: [[C_PRIV:%.+]] = alloca i{{[0-9]+}}, i{{[0-9]+}} [[C_SZ]],
511 // TCHECK: [[C_SZ2:%.+]] = mul{{.+}} i{{[0-9]+}} [[VLA_ADDR_REF]], [[VLA_ADDR_REF2]]
512 // TCHECK: [[C_SZ_CPY:%.+]] = mul{{.+}} i{{[0-9]+}} [[C_SZ2]], 2
513 // TCHECK: [[C_PRIV_BCAST:%.+]] = bitcast i{{[0-9]+}}* [[C_PRIV]] to i8*
514 // TCHECK: [[C_IN_BCAST:%.+]] = bitcast i{{[0-9]+}}* [[C_ADDR_REF]] to i8*
515 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[C_PRIV_BCAST]], i8* align {{[0-9]+}} [[C_IN_BCAST]],{{.+}})
516
517 // finish
518 // TCHECK: [[RELOAD_SSTACK:%.+]] = load i8*, i8** [[SSTACK]],
519 // TCHECK: call void @llvm.stackrestore(i8* [[RELOAD_SSTACK]])
520 // TCHECK: ret void
521
522 // static host function
523 // CHECK: define{{.+}} i32 {{.+}}(i{{[0-9]+}} {{.+}})
524 // CHECK: [[BASE_PTRS5:%.+]] = alloca [3 x i8*],
525 // CHECK: [[PTRS5:%.+]] = alloca [3 x i8*],
526
527 // firstprivate(a): by value
528 // CHECK: [[BASE_PTRS_GEP5_0:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BASE_PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
529 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP5_0]] to i{{[0-9]+}}*
530 // CHECK: store i{{[0-9]+}} [[A_CAST:%.+]], i{{[0-9]+}}* [[BCAST_TOPTR]],
531 // CHECK: [[PTRS_GEP5_0:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
532 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP5_0]] to i{{[0-9]+}}*
533 // CHECK: store i{{[0-9]+}} [[A_CAST]], i{{[0-9]+}}* [[BCAST_TOPTR]],
534
535 // firstprivate(aaa): by value
536 // CHECK: [[BASE_PTRS_GEP5_1:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BASE_PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
537 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP5_1]] to i{{[0-9]+}}*
538 // CHECK: store i{{[0-9]+}} [[A3_CAST:%.+]], i{{[0-9]+}}* [[BCAST_TOPTR]],
539 // CHECK: [[PTRS_GEP5_1:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
540 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP5_1]] to i{{[0-9]+}}*
541 // CHECK: store i{{[0-9]+}} [[A3_CAST]], i{{[0-9]+}}* [[BCAST_TOPTR]],
542
543 // firstprivate(b): base_ptr = &b[0], ptr= &b[0]
544 // CHECK: [[BASE_PTRS_GEP5_2:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BASE_PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
545 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP5_2]] to [10 x i{{[0-9]+}}]**
546 // CHECK: store [10 x i{{[0-9]+}}]* [[B:%.+]], [10 x i{{[0-9]+}}]** [[BCAST_TOPTR]],
547 // CHECK: [[PTRS_GEP5_2:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
548 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP5_2]] to [10 x i{{[0-9]+}}]**
549 // CHECK: store [10 x i{{[0-9]+}}]* [[B]], [10 x i{{[0-9]+}}]** [[BCAST_TOPTR]],
550
551 // only check that the right sizes and map types are used
552 // CHECK: {{.+}} = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 -1, i32 -1, i32 0, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]])
553 };
554
bar(int n,double * ptr)555 int bar(int n, double *ptr) {
556 int a = 0;
557 a += foo(n, ptr);
558 S1 S;
559 a += S.r1(n);
560 a += fstatic(n);
561 a += ftemplate<int>(n);
562
563 return a;
564 }
565
566 // template host and device
567
568 // CHECK: define{{.+}} i32 {{.+}}(i{{[0-9]+}} {{.+}})
569 // CHECK: [[BASE_PTRS6:%.+]] = alloca [2 x i8*],
570 // CHECK: [[PTRS6:%.+]] = alloca [2 x i8*],
571
572 // firstprivate(a): by value
573 // CHECK: [[BASE_PTRS_GEP6_0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BASE_PTRS6]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
574 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP6_0]] to i{{[0-9]+}}*
575 // CHECK: store i{{[0-9]+}} [[AT_CAST:%.+]], i{{[0-9]+}}* [[BCAST_TOPTR]],
576 // CHECK: [[PTRS_GEP6_0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[PTRS6]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
577 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP6_0]] to i{{[0-9]+}}*
578 // CHECK: store i{{[0-9]+}} [[AT_CAST]], i{{[0-9]+}}* [[BCAST_TOPTR]],
579
580 // firstprivate(b): pointer
581 // CHECK: [[BASE_PTRS_GEP6_1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BASE_PTRS6]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
582 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP6_1]] to [10 x i{{[0-9]+}}]**
583 // CHECK: store [10 x i{{[0-9]+}}]* [[B:%.+]], [10 x i{{[0-9]+}}]** [[BCAST_TOPTR]],
584 // CHECK: [[PTRS_GEP6_1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[PTRS6]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
585 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP6_1]] to [10 x i{{[0-9]+}}]**
586 // CHECK: store [10 x i{{[0-9]+}}]* [[B]], [10 x i{{[0-9]+}}]** [[BCAST_TOPTR]],
587
588 // CHECK: {{.+}} = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 -1, i32 -1, i32 0, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]])
589
590 // TCHECK: define weak_odr void @__omp_offloading_{{.+}}(i{{[0-9]+}} noundef [[A_IN:%.+]], [10 x i{{[0-9]+}}]*{{.+}} [[B_IN:%.+]])
591 // TCHECK: [[A_ADDR:%.+]] = alloca i{{[0-9]+}},
592 // TCHECK: [[B_ADDR:%.+]] = alloca [10 x i{{[0-9]+}}]*,
593 // TCHECK-NOT: alloca i{{[0-9]+}},
594 // TCHECK: [[B_PRIV:%.+]] = alloca [10 x i{{[0-9]+}}],
595 // TCHECK: store i{{[0-9]+}} [[A_IN]], i{{[0-9]+}}* [[A_ADDR]],
596 // TCHECK: store [10 x i{{[0-9]+}}]* [[B_IN]], [10 x i{{[0-9]+}}]** [[B_ADDR]],
597 // TCHECK-64: [[A_ADDR_CONV:%.+]] = bitcast i{{[0-9]+}}* [[A_ADDR]] to i{{[0-9]+}}*
598 // TCHECK: [[B_ADDR_REF:%.+]] = load [10 x i{{[0-9]+}}]*, [10 x i{{[0-9]+}}]** [[B_ADDR]],
599
600 // firstprivate(a)
601 // TCHECK-NOT: store i{{[0-9]+}} %
602
603 // firstprivate(b)
604 // TCHECK: [[B_PRIV_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_PRIV]] to i8*
605 // TCHECK: [[B_IN_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_ADDR_REF]] to i8*
606 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[B_PRIV_BCAST]], i8* align {{[0-9]+}} [[B_IN_BCAST]],{{.+}})
607
608 // TCHECK: ret void
609
610 #endif
611