-
Notifications
You must be signed in to change notification settings - Fork 18
/
Copy pathcudla_context_standalone.cpp
516 lines (464 loc) · 29.2 KB
/
cudla_context_standalone.cpp
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
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
/*
* SPDX-FileCopyrightText: Copyright (c) 2023 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
* SPDX-License-Identifier: MIT
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
* DEALINGS IN THE SOFTWARE.
*/
#include <cstdio>
#include <cstdlib>
#include <cuda.h>
#include <fstream>
#include <sstream>
#include <sys/stat.h>
#include <unistd.h>
#include "cudla_context_standalone.h"
#define DPRINTF(...) \
do \
{ \
printf("[standalone mode] "); \
printf(__VA_ARGS__); \
} while (0)
#define CHECK_CUDA_ERR(err, msg) \
do \
{ \
if (err != cudaSuccess) \
{ \
DPRINTF("%s FAILED in %s:%d, CUDA ERR: %d\n", msg, __FILE__, __LINE__, err); \
exit(1); \
} \
else \
{ \
DPRINTF("%s SUCCESS\n", msg); \
} \
} while (0)
#define CHECK_CUDLA_ERR(err, msg) \
do \
{ \
if (err != cudlaSuccess) \
{ \
DPRINTF("%s FAILED in %s:%d, CUDLA ERR: %d\n", msg, __FILE__, __LINE__, err); \
exit(1); \
} \
else \
{ \
DPRINTF("%s SUCCESS\n", msg); \
} \
} while (0)
#define CHECK_NVSCI_ERR(err, msg) \
do \
{ \
if (err != NvSciError_Success) \
{ \
DPRINTF("%s FAILED in %s:%d, NVSCI ERR: %d\n", msg, __FILE__, __LINE__, err); \
exit(1); \
} \
else \
{ \
DPRINTF("%s SUCCESS\n", msg); \
} \
} while (0)
static void printTensorDesc(cudlaModuleTensorDescriptor *tensorDesc)
{
DPRINTF("TENSOR NAME : %s\n", tensorDesc->name);
DPRINTF("size: %lu\n", tensorDesc->size);
DPRINTF("dims: [%lu, %lu, %lu, %lu]\n", tensorDesc->n, tensorDesc->c, tensorDesc->h, tensorDesc->w);
DPRINTF("data fmt: %d\n", tensorDesc->dataFormat);
DPRINTF("data type: %d\n", tensorDesc->dataType);
DPRINTF("data category: %d\n", tensorDesc->dataCategory);
DPRINTF("pixel fmt: %d\n", tensorDesc->pixelFormat);
DPRINTF("pixel mapping: %d\n", tensorDesc->pixelMapping);
DPRINTF("stride[0]: %d\n", tensorDesc->stride[0]);
DPRINTF("stride[1]: %d\n", tensorDesc->stride[1]);
DPRINTF("stride[2]: %d\n", tensorDesc->stride[2]);
DPRINTF("stride[3]: %d\n", tensorDesc->stride[3]);
}
void createAndSetAttrList(NvSciBufModule module, uint64_t bufSize, NvSciBufAttrList *attrList)
{
NvSciError sciStatus;
sciStatus = NvSciBufAttrListCreate(module, attrList);
CHECK_NVSCI_ERR(sciStatus, "create NvSci buffer attr list");
NvSciBufType bufType = NvSciBufType_RawBuffer;
bool cpu_access_flag = false;
NvSciBufAttrValAccessPerm perm = NvSciBufAccessPerm_ReadWrite;
CUuuid uuid;
CUresult res = cuDeviceGetUuid(&uuid, 0);
if (res != CUDA_SUCCESS)
{
fprintf(stderr, "Driver API error = %04d \n", res);
}
// Fill in values
NvSciBufAttrKeyValuePair raw_buf_attrs[] = {
{NvSciBufGeneralAttrKey_Types, &bufType, sizeof(bufType)},
{NvSciBufRawBufferAttrKey_Size, &bufSize, sizeof(bufSize)},
{NvSciBufGeneralAttrKey_NeedCpuAccess, &cpu_access_flag, sizeof(cpu_access_flag)},
{NvSciBufGeneralAttrKey_RequiredPerm, &perm, sizeof(perm)},
{NvSciBufGeneralAttrKey_GpuId, &uuid, sizeof(uuid)},
};
size_t length = sizeof(raw_buf_attrs) / sizeof(NvSciBufAttrKeyValuePair);
sciStatus = NvSciBufAttrListSetAttrs(*attrList, raw_buf_attrs, length);
CHECK_NVSCI_ERR(sciStatus, "set NvSci buffer attr list");
}
cuDLAContextStandalone::cuDLAContextStandalone(const char *loadableFilePath)
{
readDLALoadable(loadableFilePath);
initialize();
m_Has_initialized = true;
}
// Read loadable in to memory
void cuDLAContextStandalone::readDLALoadable(const char *loadableFilePath)
{
FILE * fp = nullptr;
struct stat st;
size_t actually_read = 0;
fp = fopen(loadableFilePath, "rb");
if (fp == nullptr)
{
DPRINTF("Cannot open file %s", loadableFilePath);
exit(1);
}
if (stat(loadableFilePath, &st) != 0)
{
DPRINTF("Cannot open file %s", loadableFilePath);
exit(1);
}
m_File_size = st.st_size;
m_LoadableData = (unsigned char *)malloc(m_File_size);
if (m_LoadableData == nullptr)
{
DPRINTF("Cannot Allocate memory for loadable");
exit(1);
}
actually_read = fread(m_LoadableData, 1, m_File_size, fp);
if (actually_read != m_File_size)
{
free(m_LoadableData);
DPRINTF("Read wrong size");
exit(1);
}
fclose(fp);
}
void cuDLAContextStandalone::initialize()
{
// Create CUDLA device
m_cudla_err = cudlaCreateDevice(0, &m_DevHandle, CUDLA_STANDALONE);
CHECK_CUDLA_ERR(m_cudla_err, "create CUDLA device");
// Load CUDLA module
m_cudla_err = cudlaModuleLoadFromMemory(m_DevHandle, m_LoadableData, m_File_size, &m_ModuleHandle, 0);
CHECK_CUDLA_ERR(m_cudla_err, "load cuDLA module from memory");
// Get number of input tensors
m_cudla_err = cudlaModuleGetAttributes(m_ModuleHandle, CUDLA_NUM_INPUT_TENSORS, &m_cudla_module_attribute);
CHECK_CUDLA_ERR(m_cudla_err, "get number of input tensors");
m_NumInputTensors = m_cudla_module_attribute.numInputTensors;
DPRINTF("numInputTensors = %d\n", m_NumInputTensors);
// Get number of output tensors
m_cudla_err = cudlaModuleGetAttributes(m_ModuleHandle, CUDLA_NUM_OUTPUT_TENSORS, &m_cudla_module_attribute);
CHECK_CUDLA_ERR(m_cudla_err, "get number of output tensors");
m_NumOutputTensors = m_cudla_module_attribute.numOutputTensors;
DPRINTF("numOutputTensors = %d\n", m_NumOutputTensors);
m_InputsBufContext.resize(m_NumInputTensors);
m_OutputsBufContext.resize(m_NumOutputTensors);
m_Input_Tensor_Descs.resize(m_NumInputTensors);
m_Output_Tensor_Descs.resize(m_NumOutputTensors);
m_InputBufRegPtrs.resize(m_NumInputTensors);
m_OutputBufRegPtrs.resize(m_NumOutputTensors);
m_cudla_module_attribute.inputTensorDesc = m_Input_Tensor_Descs.data();
// Get IO Tensor Descriptor
m_cudla_err = cudlaModuleGetAttributes(m_ModuleHandle, CUDLA_INPUT_TENSOR_DESCRIPTORS, &m_cudla_module_attribute);
CHECK_CUDLA_ERR(m_cudla_err, "get input tensor descriptor");
m_cudla_module_attribute.outputTensorDesc = m_Output_Tensor_Descs.data();
m_cudla_err = cudlaModuleGetAttributes(m_ModuleHandle, CUDLA_OUTPUT_TENSOR_DESCRIPTORS, &m_cudla_module_attribute);
CHECK_CUDLA_ERR(m_cudla_err, "get output tensor descriptor");
DPRINTF("Printing inputs tensor descriptor\n");
for (uint32_t i = 0; i < m_NumInputTensors; i++)
{
m_InputsBufContext[i].cudla_tensor_desc = &m_Input_Tensor_Descs[i];
}
DPRINTF("Printing output tensor descriptor\n");
for (uint32_t i = 0; i < m_NumOutputTensors; i++)
{
m_OutputsBufContext[i].cudla_tensor_desc = &m_Output_Tensor_Descs[i];
}
m_nvsci_err = NvSciBufModuleOpen(&m_NvSciBufModule);
CHECK_NVSCI_ERR(m_nvsci_err, "open NvSci buffer module");
// Create, set and reconcile NvSci attribute list.
// Import memory and get cuda and cudla registered pointer.
auto processNvSciBuf = [&](std::vector<NvSciBufferContext> &contexts) -> void {
for (auto &context : contexts)
{
DPRINTF("-------------------------------------------\n");
printTensorDesc(context.cudla_tensor_desc);
createAndSetAttrList(m_NvSciBufModule, context.cudla_tensor_desc->size, &(context.unreconciled_attr_list));
m_nvsci_err = NvSciBufAttrListReconcile(&(context.unreconciled_attr_list), 1,
&(context.reconciled_attr_list), &(context.conflict_list));
CHECK_NVSCI_ERR(m_nvsci_err, "reconcile NvSciBuf attribute list");
m_nvsci_err = NvSciBufObjAlloc(context.reconciled_attr_list, &(context.buf_obj));
CHECK_NVSCI_ERR(m_nvsci_err, "alloc NvSciBuf Obj");
memset(&(context.cudla_ext_mem_desc), 0, sizeof(context.cudla_ext_mem_desc));
context.cudla_ext_mem_desc.extBufObject = (void *)context.buf_obj;
context.cudla_ext_mem_desc.size = context.cudla_tensor_desc->size;
m_cudla_err = cudlaImportExternalMemory(m_DevHandle, &(context.cudla_ext_mem_desc),
&(context.buf_registered_ptr_cudla), 0);
CHECK_CUDLA_ERR(m_cudla_err, "import memory to cudla");
memset(&(context.cuda_mem_handle_desc), 0, sizeof(context.cuda_mem_handle_desc));
context.cuda_mem_handle_desc.type = cudaExternalMemoryHandleTypeNvSciBuf;
context.cuda_mem_handle_desc.handle.nvSciBufObject = context.buf_obj;
context.cuda_mem_handle_desc.size = context.cudla_tensor_desc->size;
m_cuda_err = cudaImportExternalMemory(&(context.ext_mem_raw_buf), &(context.cuda_mem_handle_desc));
CHECK_CUDA_ERR(m_cuda_err, "import external memory to cuda");
memset(&(context.cuda_ext_buffer_desc), 0, sizeof(context.cuda_ext_buffer_desc));
context.cuda_ext_buffer_desc.offset = 0;
context.cuda_ext_buffer_desc.size = context.cudla_tensor_desc->size;
m_cuda_err = cudaExternalMemoryGetMappedBuffer(&(context.buf_gpu), context.ext_mem_raw_buf,
&(context.cuda_ext_buffer_desc));
CHECK_CUDA_ERR(m_cuda_err, "map external memory to cuda buffer");
DPRINTF("-------------------------------------------\n");
}
};
processNvSciBuf(m_InputsBufContext);
processNvSciBuf(m_OutputsBufContext);
// Create vector for registered cudla buffer pointer, this is for cudlaSubmitTask.
auto createRegPtrVec = [&](std::vector<uint64_t *> &ptrs, const std::vector<NvSciBufferContext> &contexts) {
for (size_t i = 0; i < contexts.size(); i++)
{
ptrs[i] = contexts[i].buf_registered_ptr_cudla;
}
};
createRegPtrVec(m_InputBufRegPtrs, m_InputsBufContext);
createRegPtrVec(m_OutputBufRegPtrs, m_OutputsBufContext);
// Create NvSci sync module
m_nvsci_err = NvSciSyncModuleOpen(&m_NvSciSyncModule);
CHECK_NVSCI_ERR(m_nvsci_err, "create NvSci sync module");
// Create -> Setup -> Reconcile -> Alloc sync object for waiter
m_nvsci_err = NvSciSyncAttrListCreate(m_NvSciSyncModule, &(m_WaitEventContext.waiter_attr_list));
CHECK_NVSCI_ERR(m_nvsci_err, "create NvSci waiter attr list");
m_nvsci_err = NvSciSyncAttrListCreate(m_NvSciSyncModule, &(m_WaitEventContext.signaler_attr_list));
CHECK_NVSCI_ERR(m_nvsci_err, "create NvSci signal attr list");
m_cudla_err = cudlaGetNvSciSyncAttributes(reinterpret_cast<uint64_t *>(m_WaitEventContext.waiter_attr_list),
CUDLA_NVSCISYNC_ATTR_WAIT);
CHECK_CUDLA_ERR(m_cudla_err, "get NvSci waiter sync attributes");
m_cuda_err = cudaDeviceGetNvSciSyncAttributes(m_WaitEventContext.signaler_attr_list, 0, cudaNvSciSyncAttrSignal);
CHECK_CUDA_ERR(m_cuda_err, "cuda get NvSci signal list");
#ifdef USE_DETERMINISTIC_SEMAPHORE
memset(mKvPair, 0, sizeof(mKvPair));
mKvPair[0].attrKey = NvSciSyncAttrKey_RequireDeterministicFences;
mKvPair[0].value = (const void*)&m_RequireDeterFences;
mKvPair[0].len = sizeof(m_RequireDeterFences);
CHECK_NVSCI_ERR(NvSciSyncAttrListSetAttrs(m_WaitEventContext.waiter_attr_list, mKvPair, 1),
"USE_DETERMINISTIC_SEMAPHORE: set waiter's waiter attr list");
#endif // #ifdef USE_DETERMINISTIC_SEMAPHORE
NvSciSyncAttrList wait_event_attrs[2] = {m_WaitEventContext.signaler_attr_list,
m_WaitEventContext.waiter_attr_list};
m_nvsci_err = NvSciSyncAttrListReconcile(wait_event_attrs, 2, &(m_WaitEventContext.reconciled_attr_list),
&(m_WaitEventContext.conflict_list));
CHECK_NVSCI_ERR(m_nvsci_err, "reconciled NvSci sync attr list");
m_nvsci_err = NvSciSyncObjAlloc(m_WaitEventContext.reconciled_attr_list, &(m_WaitEventContext.sync_obj));
CHECK_NVSCI_ERR(m_nvsci_err, "allocate NvSci sync object");
// Import semaphore to cudla
memset(&m_WaitEventContext.cudla_ext_sema_mem_desc, 0, sizeof(m_WaitEventContext.cudla_ext_sema_mem_desc));
m_WaitEventContext.cudla_ext_sema_mem_desc.extSyncObject = m_WaitEventContext.sync_obj;
m_cudla_err = cudlaImportExternalSemaphore(m_DevHandle, &(m_WaitEventContext.cudla_ext_sema_mem_desc),
&m_WaitEventContext.nvsci_sync_obj_reg_ptr, 0);
CHECK_CUDLA_ERR(m_cudla_err, "cudla import external semaphore");
// Create wait event, this is for cudlaSubmitTask
m_WaitEvents = (cudlaWaitEvents *)malloc(sizeof(cudlaWaitEvents));
m_WaitEvents->numEvents = 1;
m_WaitEventContext.nvsci_fence_ptr = (NvSciSyncFence *)calloc(1, sizeof(NvSciSyncFence));
m_WaitEventContext.cudla_fence_ptr = (CudlaFence *)malloc(sizeof(CudlaFence));
m_WaitEventContext.cudla_fence_ptr->fence = m_WaitEventContext.nvsci_fence_ptr;
m_WaitEventContext.cudla_fence_ptr->type = CUDLA_NVSCISYNC_FENCE;
m_WaitEvents->preFences = m_WaitEventContext.cudla_fence_ptr;
#ifdef USE_DETERMINISTIC_SEMAPHORE
// updated code: binding fence with syncobj with fence value
m_nvsci_err = NvSciSyncFenceUpdateFence(m_WaitEventContext.sync_obj, m_WaiterID, m_WaiterValue, m_WaitEventContext.nvsci_fence_ptr);
CHECK_NVSCI_ERR(m_nvsci_err, "USE_DETERMINISTIC_SEMAPHORE: update waiter fence");
m_nvsci_err = NvSciSyncFenceExtractFence(m_WaitEventContext.nvsci_fence_ptr,&m_WaiterID,&m_WaiterValue);
CHECK_NVSCI_ERR(m_nvsci_err, "USE_DETERMINISTIC_SEMAPHORE: extract waiter fence");
m_nvsci_err = NvSciSyncFenceUpdateFence(m_WaitEventContext.sync_obj, m_WaiterID, ++m_WaiterValue, m_WaitEventContext.nvsci_fence_ptr);
CHECK_NVSCI_ERR(m_nvsci_err, "USE_DETERMINISTIC_SEMAPHORE: update waiter fence");
#endif // #ifdef USE_DETERMINISTIC_SEMAPHORE
// Import semaphore to cuda
cudaExternalSemaphoreHandleDesc signalExtSemDesc;
memset(&signalExtSemDesc, 0, sizeof(signalExtSemDesc));
signalExtSemDesc.type = cudaExternalSemaphoreHandleTypeNvSciSync;
signalExtSemDesc.handle.nvSciSyncObj = (void *)(m_WaitEventContext.sync_obj);
m_cuda_err = cudaImportExternalSemaphore(&m_SignalSem, &signalExtSemDesc);
memset(&m_SignalParams, 0, sizeof(m_SignalParams));
m_SignalParams.params.nvSciSync.fence = (void *)(m_WaitEventContext.nvsci_fence_ptr);
m_SignalParams.flags = 0;
// Create -> Setup -> Reconcile -> Alloc sync object for signaler
m_nvsci_err = NvSciSyncAttrListCreate(m_NvSciSyncModule, &(m_SignalEventContext.signaler_attr_list));
CHECK_NVSCI_ERR(m_nvsci_err, "create NvSci signal attr list");
m_nvsci_err = NvSciSyncAttrListCreate(m_NvSciSyncModule, &(m_SignalEventContext.waiter_attr_list));
CHECK_NVSCI_ERR(m_nvsci_err, "create NvSci waiter attr list");
m_cudla_err = cudlaGetNvSciSyncAttributes(reinterpret_cast<uint64_t *>(m_SignalEventContext.signaler_attr_list),
CUDLA_NVSCISYNC_ATTR_SIGNAL);
CHECK_CUDLA_ERR(m_cudla_err, "get NvSci sync attributes");
m_cuda_err = cudaDeviceGetNvSciSyncAttributes(m_SignalEventContext.waiter_attr_list, 0, cudaNvSciSyncAttrWait);
CHECK_CUDA_ERR(m_cuda_err, "cuda get NvSci wait attribute list");
#ifdef USE_DETERMINISTIC_SEMAPHORE
memset(mKvPair, 0, sizeof(mKvPair));
mKvPair[0].attrKey = NvSciSyncAttrKey_RequireDeterministicFences;
mKvPair[0].value = (const void*)&m_RequireDeterFences;
mKvPair[0].len = sizeof(m_RequireDeterFences);
CHECK_NVSCI_ERR(NvSciSyncAttrListSetAttrs(m_SignalEventContext.waiter_attr_list, mKvPair, 1),
"USE_DETERMINISTIC_SEMAPHORE: set signaler's waiter attr list");
#endif // #ifdef USE_DETERMINISTIC_SEMAPHORE
NvSciSyncAttrList signal_event_attrs[2] = {m_SignalEventContext.signaler_attr_list,
m_SignalEventContext.waiter_attr_list};
m_nvsci_err = NvSciSyncAttrListReconcile(signal_event_attrs, 2, &(m_SignalEventContext.reconciled_attr_list),
&(m_SignalEventContext.conflict_list));
CHECK_NVSCI_ERR(m_nvsci_err, "reconciled NvSci sync attr list");
m_nvsci_err = NvSciSyncObjAlloc(m_SignalEventContext.reconciled_attr_list, &(m_SignalEventContext.sync_obj));
CHECK_NVSCI_ERR(m_nvsci_err, "allocate NvSci sync object");
// Import semaphore to cudla
memset(&m_SignalEventContext.cudla_ext_sema_mem_desc, 0, sizeof(m_SignalEventContext.cudla_ext_sema_mem_desc));
m_SignalEventContext.cudla_ext_sema_mem_desc.extSyncObject = m_SignalEventContext.sync_obj;
m_cudla_err = cudlaImportExternalSemaphore(m_DevHandle, &(m_SignalEventContext.cudla_ext_sema_mem_desc),
&m_SignalEventContext.nvsci_sync_obj_reg_ptr, 0);
CHECK_CUDLA_ERR(m_cudla_err, "cudla import external semaphore");
// Create signal event, this is for cudlaSubmitTask
m_SignalEvents = (cudlaSignalEvents *)malloc(sizeof(cudlaSignalEvents));
m_SignalEvents->numEvents = 1;
m_SignalEvents->devPtrs = &(m_SignalEventContext.nvsci_sync_obj_reg_ptr);
m_SignalEventContext.nvsci_fence_ptr = (NvSciSyncFence *)calloc(1, sizeof(NvSciSyncFence));
m_SignalEventContext.cudla_fence_ptr = (CudlaFence *)malloc(sizeof(CudlaFence));
#ifdef USE_DETERMINISTIC_SEMAPHORE
m_SignalEventContext.cudla_fence_ptr->fence = nullptr;
#else
m_SignalEventContext.cudla_fence_ptr->fence = m_SignalEventContext.nvsci_fence_ptr;
#endif // #ifdef USE_DETERMINISTIC_SEMAPHORE
m_SignalEventContext.cudla_fence_ptr->type = CUDLA_NVSCISYNC_FENCE;
m_SignalEvents->eofFences = m_SignalEventContext.cudla_fence_ptr;
#ifdef USE_DETERMINISTIC_SEMAPHORE
// updated code: binding fence with syncobj with expected fence value after task completion
m_nvsci_err = NvSciSyncFenceUpdateFence(m_SignalEventContext.sync_obj, m_SignalerID, m_SignalerValue, m_SignalEventContext.nvsci_fence_ptr);
CHECK_NVSCI_ERR(m_nvsci_err, "USE_DETERMINISTIC_SEMAPHORE: update signaler fence");
m_nvsci_err = NvSciSyncFenceExtractFence(m_SignalEventContext.nvsci_fence_ptr, &m_SignalerID, &m_SignalerValue);
CHECK_NVSCI_ERR(m_nvsci_err, "USE_DETERMINISTIC_SEMAPHORE: extract signaler fence");
m_nvsci_err = NvSciSyncFenceUpdateFence(m_SignalEventContext.sync_obj, m_SignalerID, ++m_SignalerValue, m_SignalEventContext.nvsci_fence_ptr);
CHECK_NVSCI_ERR(m_nvsci_err, "USE_DETERMINISTIC_SEMAPHORE: update signaler fence");
#endif // #ifdef USE_DETERMINISTIC_SEMAPHORE
// Import semaphore to cuda
cudaExternalSemaphoreHandleDesc waiterExtSemDesc;
memset(&waiterExtSemDesc, 0, sizeof(waiterExtSemDesc));
waiterExtSemDesc.type = cudaExternalSemaphoreHandleTypeNvSciSync;
waiterExtSemDesc.handle.nvSciSyncObj = (void *)(m_SignalEventContext.sync_obj);
m_cuda_err = cudaImportExternalSemaphore(&m_WaitSem, &waiterExtSemDesc);
CHECK_CUDA_ERR(m_cuda_err, "import external semaphore to cuda");
memset(&m_WaitParams, 0, sizeof(m_WaitParams));
m_WaitParams.params.nvSciSync.fence = (void *)m_SignalEventContext.nvsci_fence_ptr;
m_WaitParams.flags = 0;
// Init cudla task
m_Task.moduleHandle = m_ModuleHandle;
m_Task.numInputTensors = m_NumInputTensors;
m_Task.numOutputTensors = m_NumOutputTensors;
m_Task.inputTensor = m_InputBufRegPtrs.data();
m_Task.outputTensor = m_OutputBufRegPtrs.data();
m_Task.waitEvents = m_WaitEvents;
m_Task.signalEvents = m_SignalEvents;
}
uint64_t cuDLAContextStandalone::getInputTensorSizeWithIndex(int32_t index) { return m_Input_Tensor_Descs[index].size; }
uint64_t cuDLAContextStandalone::getOutputTensorSizeWithIndex(int32_t index)
{
return m_Output_Tensor_Descs[index].size;
}
void *cuDLAContextStandalone::getInputCudaBufferPtr(int32_t index) { return m_InputsBufContext[index].buf_gpu; }
void *cuDLAContextStandalone::getOutputCudaBufferPtr(int32_t index) { return m_OutputsBufContext[index].buf_gpu; }
uint32_t cuDLAContextStandalone::getNumInputTensors() { return m_NumInputTensors; }
uint32_t cuDLAContextStandalone::getNumOutputTensors() { return m_NumOutputTensors; }
int cuDLAContextStandalone::submitDLATask(cudaStream_t streamToRun)
{
m_cuda_err = cudaSignalExternalSemaphoresAsync(&m_SignalSem, &m_SignalParams, 1, streamToRun);
CHECK_CUDA_ERR(m_cuda_err, "signal external semaphores on previous stream");
m_cudla_err = cudlaSubmitTask(m_DevHandle, &m_Task, 1, nullptr, 0);
CHECK_CUDLA_ERR(m_cudla_err, "submit cudla task");
m_cuda_err = cudaWaitExternalSemaphoresAsync(&m_WaitSem, &m_WaitParams, 1, streamToRun);
CHECK_CUDA_ERR(m_cuda_err, "wait external semaphores on previous stream");
#ifdef USE_DETERMINISTIC_SEMAPHORE
// update prefence for next submission
m_nvsci_err = NvSciSyncFenceExtractFence(m_WaitEventContext.nvsci_fence_ptr ,&m_WaiterID, &m_WaiterValue);
CHECK_NVSCI_ERR(m_nvsci_err, "USE_DETERMINISTIC_SEMAPHORE: extract waiter fence");
m_nvsci_err = NvSciSyncFenceUpdateFence(m_WaitEventContext.sync_obj,
m_WaiterID,++m_WaiterValue, m_WaitEventContext.nvsci_fence_ptr);
CHECK_NVSCI_ERR(m_nvsci_err, "USE_DETERMINISTIC_SEMAPHORE: update waiter fence");
// update Eoffence for next submission
m_nvsci_err = NvSciSyncFenceExtractFence(m_SignalEventContext.nvsci_fence_ptr, &m_SignalerID, &m_SignalerValue);
CHECK_NVSCI_ERR(m_nvsci_err, "USE_DETERMINISTIC_SEMAPHORE: extract signaler fence");
m_nvsci_err = NvSciSyncFenceUpdateFence(m_SignalEventContext.sync_obj, m_SignalerID,
++m_SignalerValue, m_SignalEventContext.nvsci_fence_ptr);
CHECK_NVSCI_ERR(m_nvsci_err, "USE_DETERMINISTIC_SEMAPHORE: update signaler fence");
#endif // #ifdef USE_DETERMINISTIC_SEMAPHORE
return 0;
}
void cuDLAContextStandalone::releaseNvSciBufferContexts(std::vector<NvSciBufferContext> &contexts)
{
// release resource in reverse construct order, good practice but may not always be necessary
for (auto &context : contexts)
{
m_cuda_err = cudaDestroyExternalMemory(context.ext_mem_raw_buf);
CHECK_CUDA_ERR(m_cuda_err, "destroy external cuda memory");
context.buf_gpu = nullptr;
m_cudla_err = cudlaMemUnregister(m_DevHandle, context.buf_registered_ptr_cudla);
CHECK_CUDLA_ERR(m_cudla_err, "unregister cudla buffer ptr");
NvSciBufObjFree(context.buf_obj);
NvSciBufAttrListFree(context.conflict_list);
NvSciBufAttrListFree(context.reconciled_attr_list);
NvSciBufAttrListFree(context.unreconciled_attr_list);
}
}
void cuDLAContextStandalone::releaseNvSciSyncContext(NvSciSyncContext &context)
{
// release resource in reverse construct order, good practice but may not always be necessary
free(context.cudla_fence_ptr);
context.cudla_fence_ptr = nullptr;
NvSciSyncFenceClear(context.nvsci_fence_ptr);
context.nvsci_fence_ptr = nullptr;
m_cudla_err = cudlaMemUnregister(m_DevHandle, context.nvsci_sync_obj_reg_ptr);
CHECK_CUDLA_ERR(m_cudla_err, "unregister cudla sync ptr");
NvSciSyncObjFree(context.sync_obj);
NvSciSyncAttrListFree(context.conflict_list);
NvSciSyncAttrListFree(context.reconciled_attr_list);
NvSciSyncAttrListFree(context.waiter_attr_list);
NvSciSyncAttrListFree(context.signaler_attr_list);
}
cuDLAContextStandalone::~cuDLAContextStandalone()
{
m_cuda_err = cudaDestroyExternalSemaphore(m_WaitSem);
CHECK_CUDA_ERR(m_cuda_err, "destroy external cuda wait semaphore");
m_cuda_err = cudaDestroyExternalSemaphore(m_SignalSem);
CHECK_CUDA_ERR(m_cuda_err, "destroy external cuda signal semaphore");
releaseNvSciBufferContexts(m_InputsBufContext);
releaseNvSciBufferContexts(m_OutputsBufContext);
releaseNvSciSyncContext(m_WaitEventContext);
releaseNvSciSyncContext(m_SignalEventContext);
free(m_WaitEvents);
m_WaitEvents = nullptr;
free(m_SignalEvents);
m_SignalEvents = nullptr;
NvSciBufModuleClose(m_NvSciBufModule);
NvSciSyncModuleClose(m_NvSciSyncModule);
m_cudla_err = cudlaModuleUnload(m_ModuleHandle, 0);
CHECK_CUDLA_ERR(m_cudla_err, "unload cudla module");
m_ModuleHandle = nullptr;
m_cudla_err = cudlaDestroyDevice(m_DevHandle);
CHECK_CUDLA_ERR(m_cudla_err, "unload cudla device handle");
m_DevHandle = nullptr;
free(m_LoadableData);
m_LoadableData = nullptr;
}