1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
|
/* SPDX-License-Identifier: Apache-2.0
* Copyright 2021-2022 Intel Corporation */
#ifdef WITH_ONEAPI
# include <sycl/sycl.hpp>
# include "device/device.h"
# include "device/oneapi/device.h"
# include "device/oneapi/queue.h"
# include "kernel/device/oneapi/kernel.h"
# include "util/map.h"
CCL_NAMESPACE_BEGIN
class DeviceQueue;
typedef void (*OneAPIDeviceIteratorCallback)(const char *id,
const char *name,
int num,
void *user_ptr);
class OneapiDevice : public Device {
private:
SyclQueue *device_queue_;
using ConstMemMap = map<string, device_vector<uchar> *>;
ConstMemMap const_mem_map_;
device_vector<TextureInfo> texture_info_;
bool need_texture_info_;
void *kg_memory_;
void *kg_memory_device_;
size_t kg_memory_size_ = (size_t)0;
size_t max_memory_on_device_ = (size_t)0;
std::string oneapi_error_string_;
public:
virtual BVHLayoutMask get_bvh_layout_mask() const override;
OneapiDevice(const DeviceInfo &info, Stats &stats, Profiler &profiler);
virtual ~OneapiDevice();
bool check_peer_access(Device *peer_device) override;
bool load_kernels(const uint requested_features) override;
void load_texture_info();
void generic_alloc(device_memory &mem);
void generic_copy_to(device_memory &mem);
void generic_free(device_memory &mem);
string oneapi_error_message();
void *kernel_globals_device_pointer();
void mem_alloc(device_memory &mem) override;
void mem_copy_to(device_memory &mem) override;
void mem_copy_from(device_memory &mem, size_t y, size_t w, size_t h, size_t elem) override;
void mem_copy_from(device_memory &mem)
{
mem_copy_from(mem, 0, 0, 0, 0);
}
void mem_zero(device_memory &mem) override;
void mem_free(device_memory &mem) override;
device_ptr mem_alloc_sub_ptr(device_memory &mem, size_t offset, size_t /*size*/) override;
virtual void const_copy_to(const char *name, void *host, size_t size) override;
void global_alloc(device_memory &mem);
void global_free(device_memory &mem);
void tex_alloc(device_texture &mem);
void tex_free(device_texture &mem);
/* Graphics resources interoperability. */
virtual bool should_use_graphics_interop() override;
virtual unique_ptr<DeviceQueue> gpu_queue_create() override;
/* NOTE(@nsirgien): Create this methods to avoid some compilation problems on Windows with host
* side compilation (MSVC). */
void *usm_aligned_alloc_host(size_t memory_size, size_t alignment);
void usm_free(void *usm_ptr);
static std::vector<sycl::device> available_devices();
static char *device_capabilities();
static int parse_driver_build_version(const sycl::device &device);
static void iterate_devices(OneAPIDeviceIteratorCallback cb, void *user_ptr);
size_t get_memcapacity();
int get_num_multiprocessors();
int get_max_num_threads_per_multiprocessor();
bool queue_synchronize(SyclQueue *queue);
bool kernel_globals_size(size_t &kernel_global_size);
void set_global_memory(SyclQueue *queue,
void *kernel_globals,
const char *memory_name,
void *memory_device_pointer);
bool enqueue_kernel(KernelContext *kernel_context, int kernel, size_t global_size, void **args);
SyclQueue *sycl_queue();
protected:
void check_usm(SyclQueue *queue, const void *usm_ptr, bool allow_host);
bool create_queue(SyclQueue *&external_queue, int device_index);
void free_queue(SyclQueue *queue);
void *usm_aligned_alloc_host(SyclQueue *queue, size_t memory_size, size_t alignment);
void *usm_alloc_device(SyclQueue *queue, size_t memory_size);
void usm_free(SyclQueue *queue, void *usm_ptr);
bool usm_memcpy(SyclQueue *queue, void *dest, void *src, size_t num_bytes);
bool usm_memset(SyclQueue *queue, void *usm_ptr, unsigned char value, size_t num_bytes);
};
CCL_NAMESPACE_END
#endif
|