1 // REQUIRES: amdgpu-registered-target
2 // RUN: %clang_cc1 -triple amdgcn-amd-amdhsa -fcuda-is-device -x hip %s \
3 // RUN: -std=c++17 -O3 -mllvm -amdgpu-internalize-symbols -emit-llvm -o - \
4 // RUN: | FileCheck -check-prefix=DEV %s
5 // RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -x hip %s \
6 // RUN: -std=c++17 -O3 -emit-llvm -o - | FileCheck -check-prefix=HOST %s
7
8 // Negative tests.
9
10 // RUN: %clang_cc1 -triple amdgcn-amd-amdhsa -fcuda-is-device -x hip %s \
11 // RUN: -std=c++17 -O3 -mllvm -amdgpu-internalize-symbols -emit-llvm -o - \
12 // RUN: | FileCheck -check-prefix=DEV-NEG %s
13 // RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -x hip %s \
14 // RUN: -std=c++17 -O3 -emit-llvm -o - | FileCheck -check-prefix=HOST-NEG %s
15
16 #include "Inputs/cuda.h"
17
18 // DEV-DAG: @v1
19 __device__ int v1;
20
21 // DEV-DAG: @v2
22 __constant__ int v2;
23
24 // Check device variables used by neither host nor device functioins are not kept.
25
26 // DEV-NEG-NOT: @_ZL2v3
27 static __device__ int v3;
28
29 // Check device variables used by host functions are kept.
30
31 // DEV-DAG: @u1
32 __device__ int u1;
33
34 // DEV-DAG: @u2
35 __constant__ int u2;
36
37 // Check host-used static device var is in llvm.compiler.used.
38 // DEV-DAG: @_ZL2u3
39 static __device__ int u3;
40
41 // Check device-used static device var is emitted but is not in llvm.compiler.used.
42 // DEV-DAG: @_ZL2u4
43 static __device__ int u4;
44
45 // Check device variables with used attribute are always kept.
46 // DEV-DAG: @u5
47 __device__ __attribute__((used)) int u5;
48
49 // Test external device variable ODR-used by host code is not emitted or registered.
50 // DEV-NEG-NOT: @ext_var
51 extern __device__ int ext_var;
52
53 // DEV-DAG: @inline_var = linkonce_odr addrspace(1) externally_initialized global i32 0
54 __device__ inline int inline_var;
55
56 template<typename T>
57 using func_t = T (*) (T, T);
58
59 template <typename T>
add_func(T x,T y)60 __device__ T add_func (T x, T y)
61 {
62 return x + y;
63 }
64
65 // DEV-DAG: @_Z10p_add_funcIiE = linkonce_odr addrspace(1) externally_initialized global i32 (i32, i32)* @_Z8add_funcIiET_S0_S0_
66 template <typename T>
67 __device__ func_t<T> p_add_func = add_func<T>;
68
69 // Check non-constant constexpr variables ODR-used by host code only is not emitted.
70 // DEV-NEG-NOT: constexpr_var1a
71 // DEV-NEG-NOT: constexpr_var1b
72 constexpr int constexpr_var1a = 1;
73 inline constexpr int constexpr_var1b = 1;
74
75 // Check constant constexpr variables ODR-used by host code only.
76 // Non-inline constexpr variable has internal linkage, therefore it is not accessible by host and not kept.
77 // Inline constexpr variable has linkonce_ord linkage, therefore it can be accessed by host and kept.
78 // DEV-NEG-NOT: constexpr_var2a
79 // DEV-DAG: @constexpr_var2b = linkonce_odr addrspace(4) externally_initialized constant i32 2
80 __constant__ constexpr int constexpr_var2a = 2;
81 inline __constant__ constexpr int constexpr_var2b = 2;
82
83 void use(func_t<int> p);
84 __host__ __device__ void use(const int *p);
85
86 // Check static device variable in host function.
87 // DEV-DAG: @_ZZ4fun1vE11static_var1 = addrspace(1) externally_initialized global i32 3
fun1()88 void fun1() {
89 static __device__ int static_var1 = 3;
90 use(&u1);
91 use(&u2);
92 use(&u3);
93 use(&ext_var);
94 use(&inline_var);
95 use(p_add_func<int>);
96 use(&constexpr_var1a);
97 use(&constexpr_var1b);
98 use(&constexpr_var2a);
99 use(&constexpr_var2b);
100 use(&static_var1);
101 }
102
103 // Check static variable in host device function.
104 // DEV-DAG: @_ZZ4fun2vE11static_var2 = internal addrspace(1) global i32 4
105 // DEV-DAG: @_ZZ4fun2vE11static_var3 = addrspace(1) global i32 4
fun2()106 __host__ __device__ void fun2() {
107 static int static_var2 = 4;
108 static __device__ int static_var3 = 4;
109 use(&static_var2);
110 use(&static_var3);
111 }
112
kern1(int ** x)113 __global__ void kern1(int **x) {
114 *x = &u4;
115 fun2();
116 }
117
118 // Check static variables of lambda functions.
119
120 // Lambda functions are implicit host device functions.
121 // Default static variables in lambda functions should be treated
122 // as host variables on host side, therefore should not be forced
123 // to be emitted on device.
124
125 // DEV-DAG: @_ZZZN21TestStaticVarInLambda3funEvENKUlPcE_clES0_E4var2 = addrspace(1) externally_initialized global i32 5
126 // DEV-NEG-NOT: @_ZZZN21TestStaticVarInLambda3funEvENKUlPcE_clES0_E4var1
127 namespace TestStaticVarInLambda {
128 class A {
129 public:
130 A(char *);
131 };
fun()132 void fun() {
133 (void) [](char *c) {
134 static A var1(c);
135 static __device__ int var2 = 5;
136 (void) var1;
137 (void) var2;
138 };
139 }
140 }
141
142 // Check implicit constant variable ODR-used by host code is not emitted.
143
144 // AST contains instantiation of al<ar>, which triggers AST instantiation
145 // of x::al<ar>::am, which triggers AST instatiation of x::ap<ar>,
146 // which triggers AST instantiation of aw<ar>::c, which has type
147 // ar. ar has base class x which has member ah. x::ah is initialized
148 // with function pointer pointing to ar:as, which returns an object
149 // of type ou. The constexpr aw<ar>::c is an implicit constant variable
150 // which is ODR-used by host function x::ap<ar>. An incorrect implementation
151 // will force aw<ar>::c to be emitted on device side, which will trigger
152 // emit of x::as and further more ctor of ou and variable o.
153 // The ODR-use of aw<ar>::c in x::ap<ar> should be treated as a host variable
154 // instead of device variable.
155
156 // DEV-NEG-NOT: _ZN16TestConstexprVar1oE
157 namespace TestConstexprVar {
158 char o;
159 class ou {
160 public:
ou(char)161 ou(char) { __builtin_strlen(&o); }
162 };
163 template < typename ao > struct aw { static constexpr ao c; };
164 class x {
165 protected:
166 typedef ou (*y)(const x *);
x(y ag)167 constexpr x(y ag) : ah(ag) {}
168 template < bool * > struct ak;
169 template < typename > struct al {
170 static bool am;
171 static ak< &am > an;
172 };
ap()173 template < typename ao > static x ap() { (void)aw< ao >::c; return x(nullptr); }
174 y ah;
175 };
176 template < typename ao > bool x::al< ao >::am(&ap< ao >);
177 class ar : x {
178 public:
ar()179 constexpr ar() : x(as) {}
as(const x *)180 static ou as(const x *) { return 0; }
181 al< ar > av;
182 };
183 }
184
185 // Check the exact list of variables to ensure @_ZL2u4 is not among them.
186 // DEV: @llvm.compiler.used = {{[^@]*}} @_Z10p_add_funcIiE
187 // DEV-SAME: {{^[^@]*}} @_ZL2u3
188 // DEV-SAME: {{^[^@]*}} @_ZZ4fun1vE11static_var1
189 // DEV-SAME: {{^[^@]*}} @_ZZZN21TestStaticVarInLambda3funEvENKUlPcE_clES0_E4var2
190 // DEV-SAME: {{^[^@]*}} @constexpr_var2b
191 // DEV-SAME: {{^[^@]*}} @inline_var
192 // DEV-SAME: {{^[^@]*}} @u1
193 // DEV-SAME: {{^[^@]*}} @u2
194 // DEV-SAME: {{^[^@]*}} @u5
195 // DEV-SAME: {{^[^@]*$}}
196
197 // HOST-DAG: hipRegisterVar{{.*}}@u1
198 // HOST-DAG: hipRegisterVar{{.*}}@u2
199 // HOST-DAG: hipRegisterVar{{.*}}@_ZL2u3
200 // HOST-DAG: hipRegisterVar{{.*}}@constexpr_var2b
201 // HOST-DAG: hipRegisterVar{{.*}}@u5
202 // HOST-DAG: hipRegisterVar{{.*}}@inline_var
203 // HOST-DAG: hipRegisterVar{{.*}}@_Z10p_add_funcIiE
204 // HOST-NEG-NOT: hipRegisterVar{{.*}}@_ZZ4fun1vE11static_var1
205 // HOST-NEG-NOT: hipRegisterVar{{.*}}@_ZZ4fun2vE11static_var2
206 // HOST-NEG-NOT: hipRegisterVar{{.*}}@_ZZ4fun2vE11static_var3
207 // HOST-NEG-NOT: hipRegisterVar{{.*}}@_ZZZN21TestStaticVarInLambda3funEvENKUlPcE_clES0_E4var2
208 // HOST-NEG-NOT: hipRegisterVar{{.*}}@_ZZZN21TestStaticVarInLambda3funEvENKUlPcE_clES0_E4var1
209 // HOST-NEG-NOT: hipRegisterVar{{.*}}@ext_var
210 // HOST-NEG-NOT: hipRegisterVar{{.*}}@_ZL2u4
211 // HOST-NEG-NOT: hipRegisterVar{{.*}}@constexpr_var1a
212 // HOST-NEG-NOT: hipRegisterVar{{.*}}@constexpr_var1b
213 // HOST-NEG-NOT: hipRegisterVar{{.*}}@constexpr_var2a
214