whisper.rn 0.4.0-rc.1 → 0.4.0-rc.11
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +6 -6
- package/android/build.gradle +4 -0
- package/android/src/main/CMakeLists.txt +21 -1
- package/android/src/main/java/com/rnwhisper/AudioUtils.java +27 -92
- package/android/src/main/java/com/rnwhisper/RNWhisper.java +86 -40
- package/android/src/main/java/com/rnwhisper/WhisperContext.java +85 -131
- package/android/src/main/jni-utils.h +76 -0
- package/android/src/main/jni.cpp +226 -109
- package/android/src/newarch/java/com/rnwhisper/RNWhisperModule.java +10 -0
- package/android/src/oldarch/java/com/rnwhisper/RNWhisperModule.java +10 -0
- package/cpp/coreml/whisper-encoder-impl.h +1 -1
- package/cpp/coreml/whisper-encoder.h +4 -0
- package/cpp/coreml/whisper-encoder.mm +5 -3
- package/cpp/ggml-alloc.c +797 -400
- package/cpp/ggml-alloc.h +60 -10
- package/cpp/ggml-backend-impl.h +255 -0
- package/cpp/ggml-backend-reg.cpp +582 -0
- package/cpp/ggml-backend.cpp +2002 -0
- package/cpp/ggml-backend.h +354 -0
- package/cpp/ggml-common.h +1851 -0
- package/cpp/ggml-cpp.h +39 -0
- package/cpp/ggml-cpu-aarch64.cpp +4247 -0
- package/cpp/ggml-cpu-aarch64.h +8 -0
- package/cpp/ggml-cpu-impl.h +531 -0
- package/cpp/ggml-cpu-quants.c +12245 -0
- package/cpp/ggml-cpu-quants.h +63 -0
- package/cpp/ggml-cpu-traits.cpp +36 -0
- package/cpp/ggml-cpu-traits.h +38 -0
- package/cpp/ggml-cpu.c +14792 -0
- package/cpp/ggml-cpu.cpp +653 -0
- package/cpp/ggml-cpu.h +137 -0
- package/cpp/ggml-impl.h +567 -0
- package/cpp/ggml-metal-impl.h +288 -0
- package/cpp/ggml-metal.h +24 -43
- package/cpp/ggml-metal.m +4867 -1080
- package/cpp/ggml-opt.cpp +854 -0
- package/cpp/ggml-opt.h +216 -0
- package/cpp/ggml-quants.c +5238 -0
- package/cpp/ggml-quants.h +100 -0
- package/cpp/ggml-threading.cpp +12 -0
- package/cpp/ggml-threading.h +14 -0
- package/cpp/ggml-whisper.metallib +0 -0
- package/cpp/ggml.c +5106 -19431
- package/cpp/ggml.h +847 -669
- package/cpp/gguf.cpp +1329 -0
- package/cpp/gguf.h +202 -0
- package/cpp/rn-audioutils.cpp +68 -0
- package/cpp/rn-audioutils.h +14 -0
- package/cpp/rn-whisper-log.h +11 -0
- package/cpp/rn-whisper.cpp +221 -52
- package/cpp/rn-whisper.h +50 -15
- package/cpp/whisper.cpp +3174 -1533
- package/cpp/whisper.h +176 -44
- package/ios/RNWhisper.mm +139 -46
- package/ios/RNWhisperAudioUtils.h +1 -2
- package/ios/RNWhisperAudioUtils.m +18 -67
- package/ios/RNWhisperContext.h +11 -8
- package/ios/RNWhisperContext.mm +195 -150
- package/jest/mock.js +15 -2
- package/lib/commonjs/NativeRNWhisper.js.map +1 -1
- package/lib/commonjs/index.js +76 -28
- package/lib/commonjs/index.js.map +1 -1
- package/lib/commonjs/version.json +1 -1
- package/lib/module/NativeRNWhisper.js.map +1 -1
- package/lib/module/index.js +76 -28
- package/lib/module/index.js.map +1 -1
- package/lib/module/version.json +1 -1
- package/lib/typescript/NativeRNWhisper.d.ts +13 -4
- package/lib/typescript/NativeRNWhisper.d.ts.map +1 -1
- package/lib/typescript/index.d.ts +37 -5
- package/lib/typescript/index.d.ts.map +1 -1
- package/package.json +9 -7
- package/src/NativeRNWhisper.ts +20 -4
- package/src/index.ts +98 -42
- package/src/version.json +1 -1
- package/whisper-rn.podspec +13 -20
- package/cpp/README.md +0 -4
- package/cpp/ggml-metal.metal +0 -2353
|
@@ -0,0 +1,582 @@
|
|
|
1
|
+
#include "ggml-backend-impl.h"
|
|
2
|
+
#include "ggml-backend.h"
|
|
3
|
+
#include "ggml-impl.h"
|
|
4
|
+
#include <algorithm>
|
|
5
|
+
#include <codecvt>
|
|
6
|
+
#include <cstring>
|
|
7
|
+
#include <filesystem>
|
|
8
|
+
#include <locale>
|
|
9
|
+
#include <memory>
|
|
10
|
+
#include <string>
|
|
11
|
+
#include <type_traits>
|
|
12
|
+
#include <vector>
|
|
13
|
+
|
|
14
|
+
#ifdef _WIN32
|
|
15
|
+
# define WIN32_LEAN_AND_MEAN
|
|
16
|
+
# ifndef NOMINMAX
|
|
17
|
+
# define NOMINMAX
|
|
18
|
+
# endif
|
|
19
|
+
# include <windows.h>
|
|
20
|
+
#elif defined(__APPLE__)
|
|
21
|
+
# include <mach-o/dyld.h>
|
|
22
|
+
# include <dlfcn.h>
|
|
23
|
+
#else
|
|
24
|
+
# include <dlfcn.h>
|
|
25
|
+
# include <unistd.h>
|
|
26
|
+
#endif
|
|
27
|
+
|
|
28
|
+
// Backend registry
|
|
29
|
+
#ifdef WSP_GGML_USE_CPU
|
|
30
|
+
#include "ggml-cpu.h"
|
|
31
|
+
#endif
|
|
32
|
+
|
|
33
|
+
#ifdef WSP_GGML_USE_CUDA
|
|
34
|
+
#include "ggml-cuda.h"
|
|
35
|
+
#endif
|
|
36
|
+
|
|
37
|
+
#ifdef WSP_GGML_USE_METAL
|
|
38
|
+
#include "ggml-metal.h"
|
|
39
|
+
#endif
|
|
40
|
+
|
|
41
|
+
#ifdef WSP_GGML_USE_SYCL
|
|
42
|
+
#include "ggml-sycl.h"
|
|
43
|
+
#endif
|
|
44
|
+
|
|
45
|
+
#ifdef WSP_GGML_USE_VULKAN
|
|
46
|
+
#include "ggml-vulkan.h"
|
|
47
|
+
#endif
|
|
48
|
+
|
|
49
|
+
#ifdef WSP_GGML_USE_OPENCL
|
|
50
|
+
#include "ggml-opencl.h"
|
|
51
|
+
#endif
|
|
52
|
+
|
|
53
|
+
#ifdef WSP_GGML_USE_BLAS
|
|
54
|
+
#include "ggml-blas.h"
|
|
55
|
+
#endif
|
|
56
|
+
|
|
57
|
+
#ifdef WSP_GGML_USE_RPC
|
|
58
|
+
#include "ggml-rpc.h"
|
|
59
|
+
#endif
|
|
60
|
+
|
|
61
|
+
#ifdef WSP_GGML_USE_CANN
|
|
62
|
+
#include "ggml-cann.h"
|
|
63
|
+
#endif
|
|
64
|
+
|
|
65
|
+
#ifdef WSP_GGML_USE_KOMPUTE
|
|
66
|
+
#include "ggml-kompute.h"
|
|
67
|
+
#endif
|
|
68
|
+
|
|
69
|
+
// disable C++17 deprecation warning for std::codecvt_utf8
|
|
70
|
+
#if defined(__clang__)
|
|
71
|
+
# pragma clang diagnostic push
|
|
72
|
+
# pragma clang diagnostic ignored "-Wdeprecated-declarations"
|
|
73
|
+
#endif
|
|
74
|
+
|
|
75
|
+
static std::wstring utf8_to_utf16(const std::string & str) {
|
|
76
|
+
std::wstring_convert<std::codecvt_utf8_utf16<wchar_t>> converter;
|
|
77
|
+
return converter.from_bytes(str);
|
|
78
|
+
}
|
|
79
|
+
|
|
80
|
+
static std::string utf16_to_utf8(const std::wstring & str) {
|
|
81
|
+
std::wstring_convert<std::codecvt_utf8_utf16<wchar_t>> converter;
|
|
82
|
+
return converter.to_bytes(str);
|
|
83
|
+
}
|
|
84
|
+
|
|
85
|
+
#if defined(__clang__)
|
|
86
|
+
# pragma clang diagnostic pop
|
|
87
|
+
#endif
|
|
88
|
+
|
|
89
|
+
#ifdef _WIN32
|
|
90
|
+
|
|
91
|
+
using dl_handle = std::remove_pointer_t<HMODULE>;
|
|
92
|
+
|
|
93
|
+
struct dl_handle_deleter {
|
|
94
|
+
void operator()(HMODULE handle) {
|
|
95
|
+
FreeLibrary(handle);
|
|
96
|
+
}
|
|
97
|
+
};
|
|
98
|
+
|
|
99
|
+
static dl_handle * dl_load_library(const std::wstring & path) {
|
|
100
|
+
// suppress error dialogs for missing DLLs
|
|
101
|
+
DWORD old_mode = SetErrorMode(SEM_FAILCRITICALERRORS);
|
|
102
|
+
SetErrorMode(old_mode | SEM_FAILCRITICALERRORS);
|
|
103
|
+
|
|
104
|
+
HMODULE handle = LoadLibraryW(path.c_str());
|
|
105
|
+
|
|
106
|
+
SetErrorMode(old_mode);
|
|
107
|
+
|
|
108
|
+
return handle;
|
|
109
|
+
}
|
|
110
|
+
|
|
111
|
+
static void * dl_get_sym(dl_handle * handle, const char * name) {
|
|
112
|
+
DWORD old_mode = SetErrorMode(SEM_FAILCRITICALERRORS);
|
|
113
|
+
SetErrorMode(old_mode | SEM_FAILCRITICALERRORS);
|
|
114
|
+
|
|
115
|
+
void * p = (void *) GetProcAddress(handle, name);
|
|
116
|
+
|
|
117
|
+
SetErrorMode(old_mode);
|
|
118
|
+
|
|
119
|
+
return p;
|
|
120
|
+
}
|
|
121
|
+
|
|
122
|
+
#else
|
|
123
|
+
|
|
124
|
+
using dl_handle = void;
|
|
125
|
+
|
|
126
|
+
struct dl_handle_deleter {
|
|
127
|
+
void operator()(void * handle) {
|
|
128
|
+
dlclose(handle);
|
|
129
|
+
}
|
|
130
|
+
};
|
|
131
|
+
|
|
132
|
+
static void * dl_load_library(const std::wstring & path) {
|
|
133
|
+
dl_handle * handle = dlopen(utf16_to_utf8(path).c_str(), RTLD_NOW | RTLD_LOCAL);
|
|
134
|
+
|
|
135
|
+
return handle;
|
|
136
|
+
}
|
|
137
|
+
|
|
138
|
+
static void * dl_get_sym(dl_handle * handle, const char * name) {
|
|
139
|
+
return dlsym(handle, name);
|
|
140
|
+
}
|
|
141
|
+
|
|
142
|
+
#endif
|
|
143
|
+
|
|
144
|
+
using dl_handle_ptr = std::unique_ptr<dl_handle, dl_handle_deleter>;
|
|
145
|
+
|
|
146
|
+
struct wsp_ggml_backend_reg_entry {
|
|
147
|
+
wsp_ggml_backend_reg_t reg;
|
|
148
|
+
dl_handle_ptr handle;
|
|
149
|
+
};
|
|
150
|
+
|
|
151
|
+
struct wsp_ggml_backend_registry {
|
|
152
|
+
std::vector<wsp_ggml_backend_reg_entry> backends;
|
|
153
|
+
std::vector<wsp_ggml_backend_dev_t> devices;
|
|
154
|
+
|
|
155
|
+
wsp_ggml_backend_registry() {
|
|
156
|
+
#ifdef WSP_GGML_USE_CUDA
|
|
157
|
+
register_backend(wsp_ggml_backend_cuda_reg());
|
|
158
|
+
#endif
|
|
159
|
+
#ifdef WSP_GGML_USE_METAL
|
|
160
|
+
register_backend(wsp_ggml_backend_metal_reg());
|
|
161
|
+
#endif
|
|
162
|
+
#ifdef WSP_GGML_USE_SYCL
|
|
163
|
+
register_backend(wsp_ggml_backend_sycl_reg());
|
|
164
|
+
#endif
|
|
165
|
+
#ifdef WSP_GGML_USE_VULKAN
|
|
166
|
+
register_backend(wsp_ggml_backend_vk_reg());
|
|
167
|
+
#endif
|
|
168
|
+
#ifdef WSP_GGML_USE_OPENCL
|
|
169
|
+
register_backend(wsp_ggml_backend_opencl_reg());
|
|
170
|
+
#endif
|
|
171
|
+
#ifdef WSP_GGML_USE_CANN
|
|
172
|
+
register_backend(wsp_ggml_backend_cann_reg());
|
|
173
|
+
#endif
|
|
174
|
+
#ifdef WSP_GGML_USE_BLAS
|
|
175
|
+
register_backend(wsp_ggml_backend_blas_reg());
|
|
176
|
+
#endif
|
|
177
|
+
#ifdef WSP_GGML_USE_RPC
|
|
178
|
+
register_backend(wsp_ggml_backend_rpc_reg());
|
|
179
|
+
#endif
|
|
180
|
+
#ifdef WSP_GGML_USE_KOMPUTE
|
|
181
|
+
register_backend(wsp_ggml_backend_kompute_reg());
|
|
182
|
+
#endif
|
|
183
|
+
#ifdef WSP_GGML_USE_CPU
|
|
184
|
+
register_backend(wsp_ggml_backend_cpu_reg());
|
|
185
|
+
#endif
|
|
186
|
+
}
|
|
187
|
+
|
|
188
|
+
~wsp_ggml_backend_registry() {
|
|
189
|
+
// FIXME: backends cannot be safely unloaded without a function to destroy all the backend resources,
|
|
190
|
+
// since backend threads may still be running and accessing resources from the dynamic library
|
|
191
|
+
for (auto & entry : backends) {
|
|
192
|
+
if (entry.handle) {
|
|
193
|
+
entry.handle.release(); // NOLINT
|
|
194
|
+
}
|
|
195
|
+
}
|
|
196
|
+
}
|
|
197
|
+
|
|
198
|
+
void register_backend(wsp_ggml_backend_reg_t reg, dl_handle_ptr handle = nullptr) {
|
|
199
|
+
if (!reg) {
|
|
200
|
+
return;
|
|
201
|
+
}
|
|
202
|
+
|
|
203
|
+
#ifndef NDEBUG
|
|
204
|
+
WSP_GGML_LOG_DEBUG("%s: registered backend %s (%zu devices)\n",
|
|
205
|
+
__func__, wsp_ggml_backend_reg_name(reg), wsp_ggml_backend_reg_dev_count(reg));
|
|
206
|
+
#endif
|
|
207
|
+
backends.push_back({ reg, std::move(handle) });
|
|
208
|
+
for (size_t i = 0; i < wsp_ggml_backend_reg_dev_count(reg); i++) {
|
|
209
|
+
register_device(wsp_ggml_backend_reg_dev_get(reg, i));
|
|
210
|
+
}
|
|
211
|
+
}
|
|
212
|
+
|
|
213
|
+
void register_device(wsp_ggml_backend_dev_t device) {
|
|
214
|
+
#ifndef NDEBUG
|
|
215
|
+
WSP_GGML_LOG_DEBUG("%s: registered device %s (%s)\n", __func__, wsp_ggml_backend_dev_name(device), wsp_ggml_backend_dev_description(device));
|
|
216
|
+
#endif
|
|
217
|
+
devices.push_back(device);
|
|
218
|
+
}
|
|
219
|
+
|
|
220
|
+
wsp_ggml_backend_reg_t load_backend(const std::wstring & path, bool silent) {
|
|
221
|
+
dl_handle_ptr handle { dl_load_library(path) };
|
|
222
|
+
if (!handle) {
|
|
223
|
+
if (!silent) {
|
|
224
|
+
WSP_GGML_LOG_ERROR("%s: failed to load %s\n", __func__, utf16_to_utf8(path).c_str());
|
|
225
|
+
}
|
|
226
|
+
return nullptr;
|
|
227
|
+
}
|
|
228
|
+
|
|
229
|
+
auto score_fn = (wsp_ggml_backend_score_t) dl_get_sym(handle.get(), "wsp_ggml_backend_score");
|
|
230
|
+
if (score_fn && score_fn() == 0) {
|
|
231
|
+
if (!silent) {
|
|
232
|
+
WSP_GGML_LOG_INFO("%s: backend %s is not supported on this system\n", __func__, utf16_to_utf8(path).c_str());
|
|
233
|
+
}
|
|
234
|
+
return nullptr;
|
|
235
|
+
}
|
|
236
|
+
|
|
237
|
+
auto backend_init_fn = (wsp_ggml_backend_init_t) dl_get_sym(handle.get(), "wsp_ggml_backend_init");
|
|
238
|
+
if (!backend_init_fn) {
|
|
239
|
+
if (!silent) {
|
|
240
|
+
WSP_GGML_LOG_ERROR("%s: failed to find wsp_ggml_backend_init in %s\n", __func__, utf16_to_utf8(path).c_str());
|
|
241
|
+
}
|
|
242
|
+
return nullptr;
|
|
243
|
+
}
|
|
244
|
+
|
|
245
|
+
wsp_ggml_backend_reg_t reg = backend_init_fn();
|
|
246
|
+
if (!reg || reg->api_version != WSP_GGML_BACKEND_API_VERSION) {
|
|
247
|
+
if (!silent) {
|
|
248
|
+
if (!reg) {
|
|
249
|
+
WSP_GGML_LOG_ERROR("%s: failed to initialize backend from %s: wsp_ggml_backend_init returned NULL\n", __func__, utf16_to_utf8(path).c_str());
|
|
250
|
+
} else {
|
|
251
|
+
WSP_GGML_LOG_ERROR("%s: failed to initialize backend from %s: incompatible API version (backend: %d, current: %d)\n",
|
|
252
|
+
__func__, utf16_to_utf8(path).c_str(), reg->api_version, WSP_GGML_BACKEND_API_VERSION);
|
|
253
|
+
}
|
|
254
|
+
}
|
|
255
|
+
return nullptr;
|
|
256
|
+
}
|
|
257
|
+
|
|
258
|
+
WSP_GGML_LOG_INFO("%s: loaded %s backend from %s\n", __func__, wsp_ggml_backend_reg_name(reg), utf16_to_utf8(path).c_str());
|
|
259
|
+
|
|
260
|
+
register_backend(reg, std::move(handle));
|
|
261
|
+
|
|
262
|
+
return reg;
|
|
263
|
+
}
|
|
264
|
+
|
|
265
|
+
void unload_backend(wsp_ggml_backend_reg_t reg, bool silent) {
|
|
266
|
+
auto it = std::find_if(backends.begin(), backends.end(),
|
|
267
|
+
[reg](const wsp_ggml_backend_reg_entry & entry) { return entry.reg == reg; });
|
|
268
|
+
|
|
269
|
+
if (it == backends.end()) {
|
|
270
|
+
if (!silent) {
|
|
271
|
+
WSP_GGML_LOG_ERROR("%s: backend not found\n", __func__);
|
|
272
|
+
}
|
|
273
|
+
return;
|
|
274
|
+
}
|
|
275
|
+
|
|
276
|
+
if (!silent) {
|
|
277
|
+
WSP_GGML_LOG_DEBUG("%s: unloading %s backend\n", __func__, wsp_ggml_backend_reg_name(reg));
|
|
278
|
+
}
|
|
279
|
+
|
|
280
|
+
// remove devices
|
|
281
|
+
devices.erase(
|
|
282
|
+
std::remove_if(devices.begin(), devices.end(),
|
|
283
|
+
[reg](wsp_ggml_backend_dev_t dev) { return wsp_ggml_backend_dev_backend_reg(dev) == reg; }),
|
|
284
|
+
devices.end());
|
|
285
|
+
|
|
286
|
+
// remove backend
|
|
287
|
+
backends.erase(it);
|
|
288
|
+
}
|
|
289
|
+
};
|
|
290
|
+
|
|
291
|
+
static wsp_ggml_backend_registry & get_reg() {
|
|
292
|
+
static wsp_ggml_backend_registry reg;
|
|
293
|
+
return reg;
|
|
294
|
+
}
|
|
295
|
+
|
|
296
|
+
// Internal API
|
|
297
|
+
void wsp_ggml_backend_register(wsp_ggml_backend_reg_t reg) {
|
|
298
|
+
get_reg().register_backend(reg);
|
|
299
|
+
}
|
|
300
|
+
|
|
301
|
+
void wsp_ggml_backend_device_register(wsp_ggml_backend_dev_t device) {
|
|
302
|
+
get_reg().register_device(device);
|
|
303
|
+
}
|
|
304
|
+
|
|
305
|
+
// Backend (reg) enumeration
|
|
306
|
+
static bool striequals(const char * a, const char * b) {
|
|
307
|
+
for (; *a && *b; a++, b++) {
|
|
308
|
+
if (std::tolower(*a) != std::tolower(*b)) {
|
|
309
|
+
return false;
|
|
310
|
+
}
|
|
311
|
+
}
|
|
312
|
+
return *a == *b;
|
|
313
|
+
}
|
|
314
|
+
|
|
315
|
+
size_t wsp_ggml_backend_reg_count() {
|
|
316
|
+
return get_reg().backends.size();
|
|
317
|
+
}
|
|
318
|
+
|
|
319
|
+
wsp_ggml_backend_reg_t wsp_ggml_backend_reg_get(size_t index) {
|
|
320
|
+
WSP_GGML_ASSERT(index < wsp_ggml_backend_reg_count());
|
|
321
|
+
return get_reg().backends[index].reg;
|
|
322
|
+
}
|
|
323
|
+
|
|
324
|
+
wsp_ggml_backend_reg_t wsp_ggml_backend_reg_by_name(const char * name) {
|
|
325
|
+
for (size_t i = 0; i < wsp_ggml_backend_reg_count(); i++) {
|
|
326
|
+
wsp_ggml_backend_reg_t reg = wsp_ggml_backend_reg_get(i);
|
|
327
|
+
if (striequals(wsp_ggml_backend_reg_name(reg), name)) {
|
|
328
|
+
return reg;
|
|
329
|
+
}
|
|
330
|
+
}
|
|
331
|
+
return nullptr;
|
|
332
|
+
}
|
|
333
|
+
|
|
334
|
+
// Device enumeration
|
|
335
|
+
size_t wsp_ggml_backend_dev_count() {
|
|
336
|
+
return get_reg().devices.size();
|
|
337
|
+
}
|
|
338
|
+
|
|
339
|
+
wsp_ggml_backend_dev_t wsp_ggml_backend_dev_get(size_t index) {
|
|
340
|
+
WSP_GGML_ASSERT(index < wsp_ggml_backend_dev_count());
|
|
341
|
+
return get_reg().devices[index];
|
|
342
|
+
}
|
|
343
|
+
|
|
344
|
+
wsp_ggml_backend_dev_t wsp_ggml_backend_dev_by_name(const char * name) {
|
|
345
|
+
for (size_t i = 0; i < wsp_ggml_backend_dev_count(); i++) {
|
|
346
|
+
wsp_ggml_backend_dev_t dev = wsp_ggml_backend_dev_get(i);
|
|
347
|
+
if (striequals(wsp_ggml_backend_dev_name(dev), name)) {
|
|
348
|
+
return dev;
|
|
349
|
+
}
|
|
350
|
+
}
|
|
351
|
+
return nullptr;
|
|
352
|
+
}
|
|
353
|
+
|
|
354
|
+
wsp_ggml_backend_dev_t wsp_ggml_backend_dev_by_type(enum wsp_ggml_backend_dev_type type) {
|
|
355
|
+
for (size_t i = 0; i < wsp_ggml_backend_dev_count(); i++) {
|
|
356
|
+
wsp_ggml_backend_dev_t dev = wsp_ggml_backend_dev_get(i);
|
|
357
|
+
if (wsp_ggml_backend_dev_type(dev) == type) {
|
|
358
|
+
return dev;
|
|
359
|
+
}
|
|
360
|
+
}
|
|
361
|
+
return nullptr;
|
|
362
|
+
}
|
|
363
|
+
|
|
364
|
+
// Convenience functions
|
|
365
|
+
wsp_ggml_backend_t wsp_ggml_backend_init_by_name(const char * name, const char * params) {
|
|
366
|
+
wsp_ggml_backend_dev_t dev = wsp_ggml_backend_dev_by_name(name);
|
|
367
|
+
if (!dev) {
|
|
368
|
+
return nullptr;
|
|
369
|
+
}
|
|
370
|
+
return wsp_ggml_backend_dev_init(dev, params);
|
|
371
|
+
}
|
|
372
|
+
|
|
373
|
+
wsp_ggml_backend_t wsp_ggml_backend_init_by_type(enum wsp_ggml_backend_dev_type type, const char * params) {
|
|
374
|
+
wsp_ggml_backend_dev_t dev = wsp_ggml_backend_dev_by_type(type);
|
|
375
|
+
if (!dev) {
|
|
376
|
+
return nullptr;
|
|
377
|
+
}
|
|
378
|
+
return wsp_ggml_backend_dev_init(dev, params);
|
|
379
|
+
}
|
|
380
|
+
|
|
381
|
+
wsp_ggml_backend_t wsp_ggml_backend_init_best(void) {
|
|
382
|
+
wsp_ggml_backend_dev_t dev = wsp_ggml_backend_dev_by_type(WSP_GGML_BACKEND_DEVICE_TYPE_GPU);
|
|
383
|
+
if (!dev) {
|
|
384
|
+
dev = wsp_ggml_backend_dev_by_type(WSP_GGML_BACKEND_DEVICE_TYPE_CPU);
|
|
385
|
+
}
|
|
386
|
+
if (!dev) {
|
|
387
|
+
return nullptr;
|
|
388
|
+
}
|
|
389
|
+
return wsp_ggml_backend_dev_init(dev, nullptr);
|
|
390
|
+
}
|
|
391
|
+
|
|
392
|
+
// Dynamic loading
|
|
393
|
+
wsp_ggml_backend_reg_t wsp_ggml_backend_load(const char * path) {
|
|
394
|
+
return get_reg().load_backend(utf8_to_utf16(path), false);
|
|
395
|
+
}
|
|
396
|
+
|
|
397
|
+
void wsp_ggml_backend_unload(wsp_ggml_backend_reg_t reg) {
|
|
398
|
+
get_reg().unload_backend(reg, true);
|
|
399
|
+
}
|
|
400
|
+
|
|
401
|
+
static std::wstring get_executable_path() {
|
|
402
|
+
#if defined(__APPLE__)
|
|
403
|
+
// get executable path
|
|
404
|
+
std::vector<char> path;
|
|
405
|
+
uint32_t size;
|
|
406
|
+
while (true) {
|
|
407
|
+
size = path.size();
|
|
408
|
+
if (_NSGetExecutablePath(path.data(), &size) == 0) {
|
|
409
|
+
break;
|
|
410
|
+
}
|
|
411
|
+
path.resize(size);
|
|
412
|
+
}
|
|
413
|
+
std::string base_path(path.data(), size);
|
|
414
|
+
// remove executable name
|
|
415
|
+
auto last_slash = base_path.find_last_of('/');
|
|
416
|
+
if (last_slash != std::string::npos) {
|
|
417
|
+
base_path = base_path.substr(0, last_slash);
|
|
418
|
+
}
|
|
419
|
+
return utf8_to_utf16(base_path + "/");
|
|
420
|
+
#elif defined(__linux__) || defined(__FreeBSD__)
|
|
421
|
+
std::string base_path = ".";
|
|
422
|
+
std::vector<char> path(1024);
|
|
423
|
+
while (true) {
|
|
424
|
+
// get executable path
|
|
425
|
+
# if defined(__linux__)
|
|
426
|
+
ssize_t len = readlink("/proc/self/exe", path.data(), path.size());
|
|
427
|
+
# elif defined(__FreeBSD__)
|
|
428
|
+
ssize_t len = readlink("/proc/curproc/file", path.data(), path.size());
|
|
429
|
+
# endif
|
|
430
|
+
if (len == -1) {
|
|
431
|
+
break;
|
|
432
|
+
}
|
|
433
|
+
if (len < (ssize_t) path.size()) {
|
|
434
|
+
base_path = std::string(path.data(), len);
|
|
435
|
+
// remove executable name
|
|
436
|
+
auto last_slash = base_path.find_last_of('/');
|
|
437
|
+
if (last_slash != std::string::npos) {
|
|
438
|
+
base_path = base_path.substr(0, last_slash);
|
|
439
|
+
}
|
|
440
|
+
break;
|
|
441
|
+
}
|
|
442
|
+
path.resize(path.size() * 2);
|
|
443
|
+
}
|
|
444
|
+
|
|
445
|
+
return utf8_to_utf16(base_path + "/");
|
|
446
|
+
#elif defined(_WIN32)
|
|
447
|
+
std::vector<wchar_t> path(MAX_PATH);
|
|
448
|
+
DWORD len = GetModuleFileNameW(NULL, path.data(), path.size());
|
|
449
|
+
if (len == 0) {
|
|
450
|
+
return {};
|
|
451
|
+
}
|
|
452
|
+
std::wstring base_path(path.data(), len);
|
|
453
|
+
// remove executable name
|
|
454
|
+
auto last_slash = base_path.find_last_of('\\');
|
|
455
|
+
if (last_slash != std::string::npos) {
|
|
456
|
+
base_path = base_path.substr(0, last_slash);
|
|
457
|
+
}
|
|
458
|
+
return base_path + L"\\";
|
|
459
|
+
#else
|
|
460
|
+
return {};
|
|
461
|
+
#endif
|
|
462
|
+
}
|
|
463
|
+
|
|
464
|
+
static std::wstring backend_filename_prefix() {
|
|
465
|
+
#ifdef _WIN32
|
|
466
|
+
return L"ggml-";
|
|
467
|
+
#else
|
|
468
|
+
return L"libggml-";
|
|
469
|
+
#endif
|
|
470
|
+
}
|
|
471
|
+
|
|
472
|
+
static std::wstring backend_filename_suffix() {
|
|
473
|
+
#ifdef _WIN32
|
|
474
|
+
return L".dll";
|
|
475
|
+
#else
|
|
476
|
+
return L".so";
|
|
477
|
+
#endif
|
|
478
|
+
}
|
|
479
|
+
|
|
480
|
+
static std::wstring path_separator() {
|
|
481
|
+
#ifdef _WIN32
|
|
482
|
+
return L"\\";
|
|
483
|
+
#else
|
|
484
|
+
return L"/";
|
|
485
|
+
#endif
|
|
486
|
+
}
|
|
487
|
+
|
|
488
|
+
static wsp_ggml_backend_reg_t wsp_ggml_backend_load_best(const char * name, bool silent, const char * user_search_path) {
|
|
489
|
+
// enumerate all the files that match [lib]ggml-name-*.[so|dll] in the search paths
|
|
490
|
+
// TODO: search system paths
|
|
491
|
+
std::wstring file_prefix = backend_filename_prefix() + utf8_to_utf16(name) + L"-";
|
|
492
|
+
std::vector<std::wstring> search_paths;
|
|
493
|
+
if (user_search_path == nullptr) {
|
|
494
|
+
search_paths.push_back(L"." + path_separator());
|
|
495
|
+
search_paths.push_back(get_executable_path());
|
|
496
|
+
} else {
|
|
497
|
+
search_paths.push_back(utf8_to_utf16(user_search_path) + path_separator());
|
|
498
|
+
}
|
|
499
|
+
|
|
500
|
+
int best_score = 0;
|
|
501
|
+
std::wstring best_path;
|
|
502
|
+
|
|
503
|
+
namespace fs = std::filesystem;
|
|
504
|
+
for (const auto & search_path : search_paths) {
|
|
505
|
+
if (!fs::exists(search_path)) {
|
|
506
|
+
continue;
|
|
507
|
+
}
|
|
508
|
+
fs::directory_iterator dir_it(search_path, fs::directory_options::skip_permission_denied);
|
|
509
|
+
for (const auto & entry : dir_it) {
|
|
510
|
+
if (entry.is_regular_file()) {
|
|
511
|
+
std::wstring filename = entry.path().filename().wstring();
|
|
512
|
+
std::wstring ext = entry.path().extension().wstring();
|
|
513
|
+
if (filename.find(file_prefix) == 0 && ext == backend_filename_suffix()) {
|
|
514
|
+
dl_handle_ptr handle { dl_load_library(entry.path().wstring()) };
|
|
515
|
+
if (!handle && !silent) {
|
|
516
|
+
WSP_GGML_LOG_ERROR("%s: failed to load %s\n", __func__, utf16_to_utf8(entry.path().wstring()).c_str());
|
|
517
|
+
}
|
|
518
|
+
if (handle) {
|
|
519
|
+
auto score_fn = (wsp_ggml_backend_score_t) dl_get_sym(handle.get(), "wsp_ggml_backend_score");
|
|
520
|
+
if (score_fn) {
|
|
521
|
+
int s = score_fn();
|
|
522
|
+
#ifndef NDEBUG
|
|
523
|
+
WSP_GGML_LOG_DEBUG("%s: %s score: %d\n", __func__, utf16_to_utf8(entry.path().wstring()).c_str(), s);
|
|
524
|
+
#endif
|
|
525
|
+
if (s > best_score) {
|
|
526
|
+
best_score = s;
|
|
527
|
+
best_path = entry.path().wstring();
|
|
528
|
+
}
|
|
529
|
+
} else {
|
|
530
|
+
if (!silent) {
|
|
531
|
+
WSP_GGML_LOG_INFO("%s: failed to find wsp_ggml_backend_score in %s\n", __func__, utf16_to_utf8(entry.path().wstring()).c_str());
|
|
532
|
+
}
|
|
533
|
+
}
|
|
534
|
+
}
|
|
535
|
+
}
|
|
536
|
+
}
|
|
537
|
+
}
|
|
538
|
+
}
|
|
539
|
+
|
|
540
|
+
if (best_score == 0) {
|
|
541
|
+
// try to load the base backend
|
|
542
|
+
for (const auto & search_path : search_paths) {
|
|
543
|
+
std::wstring path = search_path + backend_filename_prefix() + utf8_to_utf16(name) + backend_filename_suffix();
|
|
544
|
+
if (fs::exists(path)) {
|
|
545
|
+
return get_reg().load_backend(path, silent);
|
|
546
|
+
}
|
|
547
|
+
}
|
|
548
|
+
return nullptr;
|
|
549
|
+
}
|
|
550
|
+
|
|
551
|
+
return get_reg().load_backend(best_path, silent);
|
|
552
|
+
}
|
|
553
|
+
|
|
554
|
+
void wsp_ggml_backend_load_all() {
|
|
555
|
+
wsp_ggml_backend_load_all_from_path(nullptr);
|
|
556
|
+
}
|
|
557
|
+
|
|
558
|
+
void wsp_ggml_backend_load_all_from_path(const char * dir_path) {
|
|
559
|
+
#ifdef NDEBUG
|
|
560
|
+
bool silent = true;
|
|
561
|
+
#else
|
|
562
|
+
bool silent = false;
|
|
563
|
+
#endif
|
|
564
|
+
|
|
565
|
+
wsp_ggml_backend_load_best("blas", silent, dir_path);
|
|
566
|
+
wsp_ggml_backend_load_best("cann", silent, dir_path);
|
|
567
|
+
wsp_ggml_backend_load_best("cuda", silent, dir_path);
|
|
568
|
+
wsp_ggml_backend_load_best("hip", silent, dir_path);
|
|
569
|
+
wsp_ggml_backend_load_best("kompute", silent, dir_path);
|
|
570
|
+
wsp_ggml_backend_load_best("metal", silent, dir_path);
|
|
571
|
+
wsp_ggml_backend_load_best("rpc", silent, dir_path);
|
|
572
|
+
wsp_ggml_backend_load_best("sycl", silent, dir_path);
|
|
573
|
+
wsp_ggml_backend_load_best("vulkan", silent, dir_path);
|
|
574
|
+
wsp_ggml_backend_load_best("opencl", silent, dir_path);
|
|
575
|
+
wsp_ggml_backend_load_best("musa", silent, dir_path);
|
|
576
|
+
wsp_ggml_backend_load_best("cpu", silent, dir_path);
|
|
577
|
+
// check the environment variable WSP_GGML_BACKEND_PATH to load an out-of-tree backend
|
|
578
|
+
const char * backend_path = std::getenv("WSP_GGML_BACKEND_PATH");
|
|
579
|
+
if (backend_path) {
|
|
580
|
+
wsp_ggml_backend_load(backend_path);
|
|
581
|
+
}
|
|
582
|
+
}
|