HIP: Heterogenous-computing Interface for Portability
device_library_decls.h
Go to the documentation of this file.
1 /*
2 Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
3 
4 Permission is hereby granted, free of charge, to any person obtaining a copy
5 of this software and associated documentation files (the "Software"), to deal
6 in the Software without restriction, including without limitation the rights
7 to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
8 copies of the Software, and to permit persons to whom the Software is
9 furnished to do so, subject to the following conditions:
10 
11 The above copyright notice and this permission notice shall be included in
12 all copies or substantial portions of the Software.
13 
14 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
17 AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
18 LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
19 OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
20 THE SOFTWARE.
21 */
22 
28 #ifndef HIP_INCLUDE_HIP_HCC_DETAIL_DEVICE_LIBRARY_DECLS_H
29 #define HIP_INCLUDE_HIP_HCC_DETAIL_DEVICE_LIBRARY_DECLS_H
30 
32 
33 typedef unsigned char uchar;
34 typedef unsigned short ushort;
35 typedef unsigned int uint;
36 typedef unsigned long ulong;
37 typedef unsigned long long ullong;
38 
39 extern "C" __device__ __attribute__((const)) bool __ockl_wfany_i32(int);
40 extern "C" __device__ __attribute__((const)) bool __ockl_wfall_i32(int);
41 extern "C" __device__ uint __ockl_activelane_u32(void);
42 
43 extern "C" __device__ __attribute__((const)) uint __ockl_mul24_u32(uint, uint);
44 extern "C" __device__ __attribute__((const)) int __ockl_mul24_i32(int, int);
45 extern "C" __device__ __attribute__((const)) uint __ockl_mul_hi_u32(uint, uint);
46 extern "C" __device__ __attribute__((const)) int __ockl_mul_hi_i32(int, int);
47 extern "C" __device__ __attribute__((const)) uint __ockl_sad_u32(uint, uint, uint);
48 
49 extern "C" __device__ __attribute__((const)) uchar __ockl_clz_u8(uchar);
50 extern "C" __device__ __attribute__((const)) ushort __ockl_clz_u16(ushort);
51 extern "C" __device__ __attribute__((const)) uint __ockl_clz_u32(uint);
52 extern "C" __device__ __attribute__((const)) ullong __ockl_clz_u64(ullong);
53 
54 extern "C" __device__ __attribute__((const)) float __ocml_floor_f32(float);
55 extern "C" __device__ __attribute__((const)) float __ocml_rint_f32(float);
56 extern "C" __device__ __attribute__((const)) float __ocml_ceil_f32(float);
57 extern "C" __device__ __attribute__((const)) float __ocml_trunc_f32(float);
58 
59 extern "C" __device__ __attribute__((const)) float __ocml_fmin_f32(float, float);
60 extern "C" __device__ __attribute__((const)) float __ocml_fmax_f32(float, float);
61 
62 extern "C" __device__ __attribute__((convergent)) void __ockl_gws_init(uint nwm1, uint rid);
63 extern "C" __device__ __attribute__((convergent)) void __ockl_gws_barrier(uint nwm1, uint rid);
64 
65 
66 // Introduce local address space
67 #define __local __attribute__((address_space(3)))
68 
69 #ifdef __HIP_DEVICE_COMPILE__
70 __device__ inline static __local void* __to_local(unsigned x) { return (__local void*)x; }
71 #endif //__HIP_DEVICE_COMPILE__
72 
73 #if defined(__HCC__) && (__hcc_minor__ < 3)
74 // __llvm_fence* functions from device-libs/irif/src/fence.ll
75 extern "C" __device__ void __llvm_fence_acq_sg(void);
76 extern "C" __device__ void __llvm_fence_acq_wg(void);
77 extern "C" __device__ void __llvm_fence_acq_dev(void);
78 extern "C" __device__ void __llvm_fence_acq_sys(void);
79 
80 extern "C" __device__ void __llvm_fence_rel_sg(void);
81 extern "C" __device__ void __llvm_fence_rel_wg(void);
82 extern "C" __device__ void __llvm_fence_rel_dev(void);
83 extern "C" __device__ void __llvm_fence_rel_sys(void);
84 
85 extern "C" __device__ void __llvm_fence_ar_sg(void);
86 extern "C" __device__ void __llvm_fence_ar_wg(void);
87 extern "C" __device__ void __llvm_fence_ar_dev(void);
88 extern "C" __device__ void __llvm_fence_ar_sys(void);
89 
90 
91 extern "C" __device__ void __llvm_fence_sc_sg(void);
92 extern "C" __device__ void __llvm_fence_sc_wg(void);
93 extern "C" __device__ void __llvm_fence_sc_dev(void);
94 extern "C" __device__ void __llvm_fence_sc_sys(void);
95 #else
96 // Using hip.amdgcn.bc - sync threads
97 #define __CLK_LOCAL_MEM_FENCE 0x01
98 typedef unsigned __cl_mem_fence_flags;
99 
100 typedef enum __memory_scope {
101  __memory_scope_work_item = __OPENCL_MEMORY_SCOPE_WORK_ITEM,
102  __memory_scope_work_group = __OPENCL_MEMORY_SCOPE_WORK_GROUP,
103  __memory_scope_device = __OPENCL_MEMORY_SCOPE_DEVICE,
104  __memory_scope_all_svm_devices = __OPENCL_MEMORY_SCOPE_ALL_SVM_DEVICES,
105  __memory_scope_sub_group = __OPENCL_MEMORY_SCOPE_SUB_GROUP
106 } __memory_scope;
107 
108 // enum values aligned with what clang uses in EmitAtomicExpr()
109 typedef enum __memory_order
110 {
111  __memory_order_relaxed = __ATOMIC_RELAXED,
112  __memory_order_acquire = __ATOMIC_ACQUIRE,
113  __memory_order_release = __ATOMIC_RELEASE,
114  __memory_order_acq_rel = __ATOMIC_ACQ_REL,
115  __memory_order_seq_cst = __ATOMIC_SEQ_CST
116 } __memory_order;
117 
118 // Linked from hip.amdgcn.bc
119 extern "C" __device__ void
120 __atomic_work_item_fence(__cl_mem_fence_flags, __memory_order, __memory_scope);
121 #endif
122 
123 #endif
TODO-doc.