1 // Copyright Contributors to the Open Shading Language project.
2 // SPDX-License-Identifier: BSD-3-Clause
3 // https://github.com/AcademySoftwareFoundation/OpenShadingLanguage
4
5 #include "optix_stringtable.h"
6
7 using OIIO::ustring;
8
9 OSL_NAMESPACE_ENTER
10
11
OptiXStringTable(optix::Context ctx)12 OptiXStringTable::OptiXStringTable(optix::Context ctx)
13 : m_ptr (nullptr),
14 m_size (1 << 16),
15 m_offset (0),
16 m_optix_ctx (ctx)
17 {
18 }
19
20
21
~OptiXStringTable()22 OptiXStringTable::~OptiXStringTable()
23 {
24 freetable();
25 }
26
27
28
29 void
freetable()30 OptiXStringTable::freetable()
31 {
32 if (m_ptr)
33 OSL::cudaFree (m_ptr);
34 m_ptr = nullptr;
35 }
36
37
38
init(optix::Context ctx OSL_MAYBE_UNUSED)39 void OptiXStringTable::init (optix::Context ctx OSL_MAYBE_UNUSED)
40 {
41 #ifdef OSL_USE_OPTIX
42 OSL_ASSERT (! m_ptr && "StringTable should only be initialized once");
43 m_optix_ctx = ctx;
44
45 #if (OPTIX_VERSION < 70000)
46 OSL_ASSERT ((m_optix_ctx->getEnabledDeviceCount() == 1) &&
47 "Only one CUDA device is currently supported");
48 #endif
49
50 OSL::cudaMalloc (reinterpret_cast<void**>(&m_ptr), (m_size));
51
52 // Add the statically-declared strings to the table, and create OptiX
53 // variables for them in the OSL::DeviceStrings namespace.
54 //
55 // The names of the variables created here must match the extern variables
56 // declared in OSL/device_string.h for OptiX's variable scoping mechanisms
57 // to work.
58
59 #define STRDECL(str,var_name) \
60 addString (ustring(str), ustring(OSL_NAMESPACE_STRING "::DeviceStrings::" #var_name));
61 #include <OSL/strdecls.h>
62 #undef STRDECL
63 #endif
64 }
65
66
addString(ustring str OSL_MAYBE_UNUSED,ustring var_name OSL_MAYBE_UNUSED)67 uint64_t OptiXStringTable::addString (ustring str OSL_MAYBE_UNUSED,
68 ustring var_name OSL_MAYBE_UNUSED)
69 {
70 #ifdef OSL_USE_OPTIX
71 OSL_ASSERT (m_ptr && "StringTable has not been initialized");
72
73 // The strings are laid out in the table as a struct:
74 //
75 // struct TableRep {
76 // size_t len;
77 // size_t hash;
78 // char str[len+1];
79 // };
80
81 // Compute the size of the entry before adding it to the table
82 size_t size = sizeof(size_t) + sizeof(size_t) + str.size() + 1;
83 if (((m_offset + size) >= m_size)) {
84 reallocTable();
85 }
86
87 // It should be hard to trigger this assert, unless the table size is
88 // very small and the string is very large.
89 OSL_ASSERT (m_offset + size <= m_size && "String table allocation error");
90
91 int offset = getOffset(str.string());
92 if (offset < 0) {
93 // Place the hash and length of the string before the characters
94 size_t hash = str.hash();
95 cudaMemcpy (m_ptr + m_offset, (void*)&hash, sizeof(size_t), cudaMemcpyHostToDevice);
96 m_offset += sizeof(size_t);
97
98 size_t len = str.length();
99 cudaMemcpy (m_ptr + m_offset, (void*)&len, sizeof(size_t), cudaMemcpyHostToDevice);
100 m_offset += sizeof(size_t);
101
102 offset = m_offset;
103 m_offset_map [str] = offset;
104 m_name_map [str] = var_name;
105
106 // Copy the raw characters to the table
107 cudaMemcpy (m_ptr + m_offset, str.c_str(), str.size() + 1, cudaMemcpyHostToDevice);
108 m_offset += str.size() + 1;
109
110 // Align the offset for the next entry to 8-byte boundaries
111 m_offset = (m_offset + 0x7u) & ~0x7u;
112 }
113 else if (!var_name.empty()) {
114 // update what str points to
115 m_name_map[str] = var_name;
116 }
117
118 uint64_t addr = reinterpret_cast<uint64_t>(m_ptr + offset);
119
120 // Optionally create an OptiX variable for the string. It's not necessary to
121 // create a variable for strings that do not appear by name in compiled code
122 // (in either the OSL library functions or in the renderer).
123 #if (OPTIX_VERSION < 70000)
124 if (! var_name.empty())
125 m_optix_ctx [var_name.string()]->setUserData (8, &addr);
126 #else
127 if (! var_name.empty())
128 m_addr_table [var_name] = addr;
129 #endif
130 return addr;
131 #else
132 return 0;
133 #endif
134 }
135
136
getOffset(const std::string & str) const137 int OptiXStringTable::getOffset (const std::string& str) const
138 {
139 auto it = m_offset_map.find (ustring(str));
140 return (it != m_offset_map.end()) ? it->second : -1;
141 }
142
143
reallocTable()144 void OptiXStringTable::reallocTable()
145 {
146 #ifdef OSL_USE_OPTIX
147
148 #if (OPTIX_VERSION < 70000)
149 OSL_ASSERT ((m_optix_ctx->getEnabledDeviceCount() == 1) &&
150 "Only one CUDA device is currently supported");
151 #endif
152
153 m_size *= 2;
154 OSL::cudaFree (m_ptr);
155 OSL::cudaMalloc (reinterpret_cast<void**>(&m_ptr), (m_size));
156
157 // The offsets need to be recomputed
158 m_offset = 0;
159 m_offset_map.clear();
160
161 // Add the string collection to the newly-allocated memory
162 for (auto& entry : m_name_map) {
163 addString (entry.first, entry.second);
164 }
165 #endif
166 }
167
168 OSL_NAMESPACE_EXIT
169