FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
nvenc.c
Go to the documentation of this file.
1 /*
2  * H.264 hardware encoding using nvidia nvenc
3  * Copyright (c) 2014 Timo Rothenpieler <timo@rothenpieler.org>
4  *
5  * This file is part of FFmpeg.
6  *
7  * FFmpeg is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * FFmpeg is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with FFmpeg; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
22 #include "config.h"
23 
24 #if defined(_WIN32)
25 #include <windows.h>
26 
27 #define CUDA_LIBNAME TEXT("nvcuda.dll")
28 #if ARCH_X86_64
29 #define NVENC_LIBNAME TEXT("nvEncodeAPI64.dll")
30 #else
31 #define NVENC_LIBNAME TEXT("nvEncodeAPI.dll")
32 #endif
33 
34 #define dlopen(filename, flags) LoadLibrary((filename))
35 #define dlsym(handle, symbol) GetProcAddress(handle, symbol)
36 #define dlclose(handle) FreeLibrary(handle)
37 #else
38 #include <dlfcn.h>
39 
40 #define CUDA_LIBNAME "libcuda.so"
41 #define NVENC_LIBNAME "libnvidia-encode.so"
42 #endif
43 
44 #include "libavutil/hwcontext.h"
45 #include "libavutil/imgutils.h"
46 #include "libavutil/avassert.h"
47 #include "libavutil/mem.h"
48 #include "internal.h"
49 #include "nvenc.h"
50 
51 #define NVENC_CAP 0x30
52 #define IS_CBR(rc) (rc == NV_ENC_PARAMS_RC_CBR || \
53  rc == NV_ENC_PARAMS_RC_2_PASS_QUALITY || \
54  rc == NV_ENC_PARAMS_RC_2_PASS_FRAMESIZE_CAP)
55 
56 #define LOAD_LIBRARY(l, path) \
57  do { \
58  if (!((l) = dlopen(path, RTLD_LAZY))) { \
59  av_log(avctx, AV_LOG_ERROR, \
60  "Cannot load %s\n", \
61  path); \
62  return AVERROR_UNKNOWN; \
63  } \
64  } while (0)
65 
66 #define LOAD_SYMBOL(fun, lib, symbol) \
67  do { \
68  if (!((fun) = dlsym(lib, symbol))) { \
69  av_log(avctx, AV_LOG_ERROR, \
70  "Cannot load %s\n", \
71  symbol); \
72  return AVERROR_UNKNOWN; \
73  } \
74  } while (0)
75 
80 #if CONFIG_CUDA
82 #endif
84 };
85 
86 static const struct {
87  NVENCSTATUS nverr;
88  int averr;
89  const char *desc;
90 } nvenc_errors[] = {
91  { NV_ENC_SUCCESS, 0, "success" },
92  { NV_ENC_ERR_NO_ENCODE_DEVICE, AVERROR(ENOENT), "no encode device" },
93  { NV_ENC_ERR_UNSUPPORTED_DEVICE, AVERROR(ENOSYS), "unsupported device" },
94  { NV_ENC_ERR_INVALID_ENCODERDEVICE, AVERROR(EINVAL), "invalid encoder device" },
95  { NV_ENC_ERR_INVALID_DEVICE, AVERROR(EINVAL), "invalid device" },
96  { NV_ENC_ERR_DEVICE_NOT_EXIST, AVERROR(EIO), "device does not exist" },
97  { NV_ENC_ERR_INVALID_PTR, AVERROR(EFAULT), "invalid ptr" },
98  { NV_ENC_ERR_INVALID_EVENT, AVERROR(EINVAL), "invalid event" },
99  { NV_ENC_ERR_INVALID_PARAM, AVERROR(EINVAL), "invalid param" },
100  { NV_ENC_ERR_INVALID_CALL, AVERROR(EINVAL), "invalid call" },
101  { NV_ENC_ERR_OUT_OF_MEMORY, AVERROR(ENOMEM), "out of memory" },
102  { NV_ENC_ERR_ENCODER_NOT_INITIALIZED, AVERROR(EINVAL), "encoder not initialized" },
103  { NV_ENC_ERR_UNSUPPORTED_PARAM, AVERROR(ENOSYS), "unsupported param" },
104  { NV_ENC_ERR_LOCK_BUSY, AVERROR(EAGAIN), "lock busy" },
105  { NV_ENC_ERR_NOT_ENOUGH_BUFFER, AVERROR(ENOBUFS), "not enough buffer" },
106  { NV_ENC_ERR_INVALID_VERSION, AVERROR(EINVAL), "invalid version" },
107  { NV_ENC_ERR_MAP_FAILED, AVERROR(EIO), "map failed" },
108  { NV_ENC_ERR_NEED_MORE_INPUT, AVERROR(EAGAIN), "need more input" },
109  { NV_ENC_ERR_ENCODER_BUSY, AVERROR(EAGAIN), "encoder busy" },
110  { NV_ENC_ERR_EVENT_NOT_REGISTERD, AVERROR(EBADF), "event not registered" },
111  { NV_ENC_ERR_GENERIC, AVERROR_UNKNOWN, "generic error" },
112  { NV_ENC_ERR_INCOMPATIBLE_CLIENT_KEY, AVERROR(EINVAL), "incompatible client key" },
113  { NV_ENC_ERR_UNIMPLEMENTED, AVERROR(ENOSYS), "unimplemented" },
114  { NV_ENC_ERR_RESOURCE_REGISTER_FAILED, AVERROR(EIO), "resource register failed" },
115  { NV_ENC_ERR_RESOURCE_NOT_REGISTERED, AVERROR(EBADF), "resource not registered" },
116  { NV_ENC_ERR_RESOURCE_NOT_MAPPED, AVERROR(EBADF), "resource not mapped" },
117 };
118 
119 static int nvenc_map_error(NVENCSTATUS err, const char **desc)
120 {
121  int i;
122  for (i = 0; i < FF_ARRAY_ELEMS(nvenc_errors); i++) {
123  if (nvenc_errors[i].nverr == err) {
124  if (desc)
125  *desc = nvenc_errors[i].desc;
126  return nvenc_errors[i].averr;
127  }
128  }
129  if (desc)
130  *desc = "unknown error";
131  return AVERROR_UNKNOWN;
132 }
133 
134 static int nvenc_print_error(void *log_ctx, NVENCSTATUS err,
135  const char *error_string)
136 {
137  const char *desc;
138  int ret;
139  ret = nvenc_map_error(err, &desc);
140  av_log(log_ctx, AV_LOG_ERROR, "%s: %s (%d)\n", error_string, desc, err);
141  return ret;
142 }
143 
145 {
146  NvencContext *ctx = avctx->priv_data;
148  PNVENCODEAPICREATEINSTANCE nvenc_create_instance;
149  NVENCSTATUS err;
150 
151 #if CONFIG_CUDA
152  dl_fn->cu_init = cuInit;
153  dl_fn->cu_device_get_count = cuDeviceGetCount;
154  dl_fn->cu_device_get = cuDeviceGet;
155  dl_fn->cu_device_get_name = cuDeviceGetName;
156  dl_fn->cu_device_compute_capability = cuDeviceComputeCapability;
157  dl_fn->cu_ctx_create = cuCtxCreate_v2;
158  dl_fn->cu_ctx_pop_current = cuCtxPopCurrent_v2;
159  dl_fn->cu_ctx_destroy = cuCtxDestroy_v2;
160 #else
161  LOAD_LIBRARY(dl_fn->cuda, CUDA_LIBNAME);
162 
163  LOAD_SYMBOL(dl_fn->cu_init, dl_fn->cuda, "cuInit");
164  LOAD_SYMBOL(dl_fn->cu_device_get_count, dl_fn->cuda, "cuDeviceGetCount");
165  LOAD_SYMBOL(dl_fn->cu_device_get, dl_fn->cuda, "cuDeviceGet");
166  LOAD_SYMBOL(dl_fn->cu_device_get_name, dl_fn->cuda, "cuDeviceGetName");
168  "cuDeviceComputeCapability");
169  LOAD_SYMBOL(dl_fn->cu_ctx_create, dl_fn->cuda, "cuCtxCreate_v2");
170  LOAD_SYMBOL(dl_fn->cu_ctx_pop_current, dl_fn->cuda, "cuCtxPopCurrent_v2");
171  LOAD_SYMBOL(dl_fn->cu_ctx_destroy, dl_fn->cuda, "cuCtxDestroy_v2");
172 #endif
173 
175 
176  LOAD_SYMBOL(nvenc_create_instance, dl_fn->nvenc,
177  "NvEncodeAPICreateInstance");
178 
179  dl_fn->nvenc_funcs.version = NV_ENCODE_API_FUNCTION_LIST_VER;
180 
181  err = nvenc_create_instance(&dl_fn->nvenc_funcs);
182  if (err != NV_ENC_SUCCESS)
183  return nvenc_print_error(avctx, err, "Failed to create nvenc instance");
184 
185  av_log(avctx, AV_LOG_VERBOSE, "Nvenc initialized successfully\n");
186 
187  return 0;
188 }
189 
191 {
192  NV_ENC_OPEN_ENCODE_SESSION_EX_PARAMS params = { 0 };
193  NvencContext *ctx = avctx->priv_data;
194  NV_ENCODE_API_FUNCTION_LIST *p_nvenc = &ctx->nvenc_dload_funcs.nvenc_funcs;
195  NVENCSTATUS ret;
196 
197  params.version = NV_ENC_OPEN_ENCODE_SESSION_EX_PARAMS_VER;
198  params.apiVersion = NVENCAPI_VERSION;
199  params.device = ctx->cu_context;
200  params.deviceType = NV_ENC_DEVICE_TYPE_CUDA;
201 
202  ret = p_nvenc->nvEncOpenEncodeSessionEx(&params, &ctx->nvencoder);
203  if (ret != NV_ENC_SUCCESS) {
204  ctx->nvencoder = NULL;
205  return nvenc_print_error(avctx, ret, "OpenEncodeSessionEx failed");
206  }
207 
208  return 0;
209 }
210 
212 {
213  NvencContext *ctx = avctx->priv_data;
214  NV_ENCODE_API_FUNCTION_LIST *p_nvenc = &ctx->nvenc_dload_funcs.nvenc_funcs;
215  int i, ret, count = 0;
216  GUID *guids = NULL;
217 
218  ret = p_nvenc->nvEncGetEncodeGUIDCount(ctx->nvencoder, &count);
219 
220  if (ret != NV_ENC_SUCCESS || !count)
221  return AVERROR(ENOSYS);
222 
223  guids = av_malloc(count * sizeof(GUID));
224  if (!guids)
225  return AVERROR(ENOMEM);
226 
227  ret = p_nvenc->nvEncGetEncodeGUIDs(ctx->nvencoder, guids, count, &count);
228  if (ret != NV_ENC_SUCCESS) {
229  ret = AVERROR(ENOSYS);
230  goto fail;
231  }
232 
233  ret = AVERROR(ENOSYS);
234  for (i = 0; i < count; i++) {
235  if (!memcmp(&guids[i], &ctx->init_encode_params.encodeGUID, sizeof(*guids))) {
236  ret = 0;
237  break;
238  }
239  }
240 
241 fail:
242  av_free(guids);
243 
244  return ret;
245 }
246 
247 static int nvenc_check_cap(AVCodecContext *avctx, NV_ENC_CAPS cap)
248 {
249  NvencContext *ctx = avctx->priv_data;
250  NV_ENCODE_API_FUNCTION_LIST *p_nvenc = &ctx->nvenc_dload_funcs.nvenc_funcs;
251  NV_ENC_CAPS_PARAM params = { 0 };
252  int ret, val = 0;
253 
254  params.version = NV_ENC_CAPS_PARAM_VER;
255  params.capsToQuery = cap;
256 
257  ret = p_nvenc->nvEncGetEncodeCaps(ctx->nvencoder, ctx->init_encode_params.encodeGUID, &params, &val);
258 
259  if (ret == NV_ENC_SUCCESS)
260  return val;
261  return 0;
262 }
263 
265 {
266  NvencContext *ctx = avctx->priv_data;
267  int ret;
268 
269  ret = nvenc_check_codec_support(avctx);
270  if (ret < 0) {
271  av_log(avctx, AV_LOG_VERBOSE, "Codec not supported\n");
272  return ret;
273  }
274 
275  ret = nvenc_check_cap(avctx, NV_ENC_CAPS_SUPPORT_YUV444_ENCODE);
276  if (ctx->data_pix_fmt == AV_PIX_FMT_YUV444P && ret <= 0) {
277  av_log(avctx, AV_LOG_VERBOSE, "YUV444P not supported\n");
278  return AVERROR(ENOSYS);
279  }
280 
281  ret = nvenc_check_cap(avctx, NV_ENC_CAPS_SUPPORT_LOSSLESS_ENCODE);
282  if (ctx->preset >= PRESET_LOSSLESS_DEFAULT && ret <= 0) {
283  av_log(avctx, AV_LOG_VERBOSE, "Lossless encoding not supported\n");
284  return AVERROR(ENOSYS);
285  }
286 
287  ret = nvenc_check_cap(avctx, NV_ENC_CAPS_WIDTH_MAX);
288  if (ret < avctx->width) {
289  av_log(avctx, AV_LOG_VERBOSE, "Width %d exceeds %d\n",
290  avctx->width, ret);
291  return AVERROR(ENOSYS);
292  }
293 
294  ret = nvenc_check_cap(avctx, NV_ENC_CAPS_HEIGHT_MAX);
295  if (ret < avctx->height) {
296  av_log(avctx, AV_LOG_VERBOSE, "Height %d exceeds %d\n",
297  avctx->height, ret);
298  return AVERROR(ENOSYS);
299  }
300 
301  ret = nvenc_check_cap(avctx, NV_ENC_CAPS_NUM_MAX_BFRAMES);
302  if (ret < avctx->max_b_frames) {
303  av_log(avctx, AV_LOG_VERBOSE, "Max B-frames %d exceed %d\n",
304  avctx->max_b_frames, ret);
305 
306  return AVERROR(ENOSYS);
307  }
308 
309  ret = nvenc_check_cap(avctx, NV_ENC_CAPS_SUPPORT_FIELD_ENCODING);
310  if (ret < 1 && avctx->flags & AV_CODEC_FLAG_INTERLACED_DCT) {
311  av_log(avctx, AV_LOG_VERBOSE,
312  "Interlaced encoding is not supported. Supported level: %d\n",
313  ret);
314  return AVERROR(ENOSYS);
315  }
316 
317  return 0;
318 }
319 
320 static av_cold int nvenc_check_device(AVCodecContext *avctx, int idx)
321 {
322  NvencContext *ctx = avctx->priv_data;
324  NV_ENCODE_API_FUNCTION_LIST *p_nvenc = &dl_fn->nvenc_funcs;
325  char name[128] = { 0};
326  int major, minor, ret;
327  CUresult cu_res;
328  CUdevice cu_device;
330  int loglevel = AV_LOG_VERBOSE;
331 
332  if (ctx->device == LIST_DEVICES)
333  loglevel = AV_LOG_INFO;
334 
335  cu_res = dl_fn->cu_device_get(&cu_device, idx);
336  if (cu_res != CUDA_SUCCESS) {
337  av_log(avctx, AV_LOG_ERROR,
338  "Cannot access the CUDA device %d\n",
339  idx);
340  return -1;
341  }
342 
343  cu_res = dl_fn->cu_device_get_name(name, sizeof(name), cu_device);
344  if (cu_res != CUDA_SUCCESS)
345  return -1;
346 
347  cu_res = dl_fn->cu_device_compute_capability(&major, &minor, cu_device);
348  if (cu_res != CUDA_SUCCESS)
349  return -1;
350 
351  av_log(avctx, loglevel, "[ GPU #%d - < %s > has Compute SM %d.%d ]\n", idx, name, major, minor);
352  if (((major << 4) | minor) < NVENC_CAP) {
353  av_log(avctx, loglevel, "does not support NVENC\n");
354  goto fail;
355  }
356 
357  cu_res = dl_fn->cu_ctx_create(&ctx->cu_context_internal, 0, cu_device);
358  if (cu_res != CUDA_SUCCESS) {
359  av_log(avctx, AV_LOG_FATAL, "Failed creating CUDA context for NVENC: 0x%x\n", (int)cu_res);
360  goto fail;
361  }
362 
363  ctx->cu_context = ctx->cu_context_internal;
364 
365  cu_res = dl_fn->cu_ctx_pop_current(&dummy);
366  if (cu_res != CUDA_SUCCESS) {
367  av_log(avctx, AV_LOG_FATAL, "Failed popping CUDA context: 0x%x\n", (int)cu_res);
368  goto fail2;
369  }
370 
371  if ((ret = nvenc_open_session(avctx)) < 0)
372  goto fail2;
373 
374  if ((ret = nvenc_check_capabilities(avctx)) < 0)
375  goto fail3;
376 
377  av_log(avctx, loglevel, "supports NVENC\n");
378 
379  dl_fn->nvenc_device_count++;
380 
381  if (ctx->device == dl_fn->nvenc_device_count - 1 || ctx->device == ANY_DEVICE)
382  return 0;
383 
384 fail3:
385  p_nvenc->nvEncDestroyEncoder(ctx->nvencoder);
386  ctx->nvencoder = NULL;
387 
388 fail2:
389  dl_fn->cu_ctx_destroy(ctx->cu_context_internal);
390  ctx->cu_context_internal = NULL;
391 
392 fail:
393  return AVERROR(ENOSYS);
394 }
395 
397 {
398  NvencContext *ctx = avctx->priv_data;
400 
401  switch (avctx->codec->id) {
402  case AV_CODEC_ID_H264:
403  ctx->init_encode_params.encodeGUID = NV_ENC_CODEC_H264_GUID;
404  break;
405  case AV_CODEC_ID_HEVC:
406  ctx->init_encode_params.encodeGUID = NV_ENC_CODEC_HEVC_GUID;
407  break;
408  default:
409  return AVERROR_BUG;
410  }
411 
412  if (avctx->pix_fmt == AV_PIX_FMT_CUDA) {
413 #if CONFIG_CUDA
414  AVHWFramesContext *frames_ctx;
415  AVCUDADeviceContext *device_hwctx;
416  int ret;
417 
418  if (!avctx->hw_frames_ctx)
419  return AVERROR(EINVAL);
420 
421  frames_ctx = (AVHWFramesContext*)avctx->hw_frames_ctx->data;
422  device_hwctx = frames_ctx->device_ctx->hwctx;
423 
424  ctx->cu_context = device_hwctx->cuda_ctx;
425 
426  ret = nvenc_open_session(avctx);
427  if (ret < 0)
428  return ret;
429 
430  ret = nvenc_check_capabilities(avctx);
431  if (ret < 0) {
432  av_log(avctx, AV_LOG_FATAL, "Provided device doesn't support required NVENC features\n");
433  return ret;
434  }
435 #else
436  return AVERROR_BUG;
437 #endif
438  } else {
439  int i, nb_devices = 0;
440 
441  if ((dl_fn->cu_init(0)) != CUDA_SUCCESS) {
442  av_log(avctx, AV_LOG_ERROR,
443  "Cannot init CUDA\n");
444  return AVERROR_UNKNOWN;
445  }
446 
447  if ((dl_fn->cu_device_get_count(&nb_devices)) != CUDA_SUCCESS) {
448  av_log(avctx, AV_LOG_ERROR,
449  "Cannot enumerate the CUDA devices\n");
450  return AVERROR_UNKNOWN;
451  }
452 
453  if (!nb_devices) {
454  av_log(avctx, AV_LOG_FATAL, "No CUDA capable devices found\n");
455  return AVERROR_EXTERNAL;
456  }
457 
458  av_log(avctx, AV_LOG_VERBOSE, "%d CUDA capable devices found\n", nb_devices);
459 
460  dl_fn->nvenc_device_count = 0;
461  for (i = 0; i < nb_devices; ++i) {
462  if ((nvenc_check_device(avctx, i)) >= 0 && ctx->device != LIST_DEVICES)
463  return 0;
464  }
465 
466  if (ctx->device == LIST_DEVICES)
467  return AVERROR_EXIT;
468 
469  if (!dl_fn->nvenc_device_count) {
470  av_log(avctx, AV_LOG_FATAL, "No NVENC capable devices found\n");
471  return AVERROR_EXTERNAL;
472  }
473 
474  av_log(avctx, AV_LOG_FATAL, "Requested GPU %d, but only %d GPUs are available!\n", ctx->device, dl_fn->nvenc_device_count);
475  return AVERROR(EINVAL);
476  }
477 
478  return 0;
479 }
480 
481 typedef struct GUIDTuple {
482  const GUID guid;
483  int flags;
484 } GUIDTuple;
485 
487 {
488  GUIDTuple presets[] = {
489  { NV_ENC_PRESET_DEFAULT_GUID },
490  { NV_ENC_PRESET_HQ_GUID, NVENC_TWO_PASSES }, /* slow */
491  { NV_ENC_PRESET_HQ_GUID, NVENC_ONE_PASS }, /* medium */
492  { NV_ENC_PRESET_HP_GUID, NVENC_ONE_PASS }, /* fast */
493  { NV_ENC_PRESET_HP_GUID },
494  { NV_ENC_PRESET_HQ_GUID },
495  { NV_ENC_PRESET_BD_GUID },
496  { NV_ENC_PRESET_LOW_LATENCY_DEFAULT_GUID, NVENC_LOWLATENCY },
497  { NV_ENC_PRESET_LOW_LATENCY_HQ_GUID, NVENC_LOWLATENCY },
498  { NV_ENC_PRESET_LOW_LATENCY_HP_GUID, NVENC_LOWLATENCY },
499  { NV_ENC_PRESET_LOSSLESS_DEFAULT_GUID, NVENC_LOSSLESS },
500  { NV_ENC_PRESET_LOSSLESS_HP_GUID, NVENC_LOSSLESS },
501  };
502 
503  GUIDTuple *t = &presets[ctx->preset];
504 
505  ctx->init_encode_params.presetGUID = t->guid;
506  ctx->flags = t->flags;
507 }
508 
509 static av_cold void set_constqp(AVCodecContext *avctx)
510 {
511  NvencContext *ctx = avctx->priv_data;
512  NV_ENC_RC_PARAMS *rc = &ctx->encode_config.rcParams;
513 
514  rc->rateControlMode = NV_ENC_PARAMS_RC_CONSTQP;
515  rc->constQP.qpInterB = avctx->global_quality;
516  rc->constQP.qpInterP = avctx->global_quality;
517  rc->constQP.qpIntra = avctx->global_quality;
518 
519  avctx->qmin = -1;
520  avctx->qmax = -1;
521 }
522 
523 static av_cold void set_vbr(AVCodecContext *avctx)
524 {
525  NvencContext *ctx = avctx->priv_data;
526  NV_ENC_RC_PARAMS *rc = &ctx->encode_config.rcParams;
527  int qp_inter_p;
528 
529  if (avctx->qmin >= 0 && avctx->qmax >= 0) {
530  rc->enableMinQP = 1;
531  rc->enableMaxQP = 1;
532 
533  rc->minQP.qpInterB = avctx->qmin;
534  rc->minQP.qpInterP = avctx->qmin;
535  rc->minQP.qpIntra = avctx->qmin;
536 
537  rc->maxQP.qpInterB = avctx->qmax;
538  rc->maxQP.qpInterP = avctx->qmax;
539  rc->maxQP.qpIntra = avctx->qmax;
540 
541  qp_inter_p = (avctx->qmax + 3 * avctx->qmin) / 4; // biased towards Qmin
542  } else if (avctx->qmin >= 0) {
543  rc->enableMinQP = 1;
544 
545  rc->minQP.qpInterB = avctx->qmin;
546  rc->minQP.qpInterP = avctx->qmin;
547  rc->minQP.qpIntra = avctx->qmin;
548 
549  qp_inter_p = avctx->qmin;
550  } else {
551  qp_inter_p = 26; // default to 26
552  }
553 
554  rc->enableInitialRCQP = 1;
555  rc->initialRCQP.qpInterP = qp_inter_p;
556 
557  if (avctx->i_quant_factor != 0.0 && avctx->b_quant_factor != 0.0) {
558  rc->initialRCQP.qpIntra = av_clip(
559  qp_inter_p * fabs(avctx->i_quant_factor) + avctx->i_quant_offset, 0, 51);
560  rc->initialRCQP.qpInterB = av_clip(
561  qp_inter_p * fabs(avctx->b_quant_factor) + avctx->b_quant_offset, 0, 51);
562  } else {
563  rc->initialRCQP.qpIntra = qp_inter_p;
564  rc->initialRCQP.qpInterB = qp_inter_p;
565  }
566 }
567 
569 {
570  NvencContext *ctx = avctx->priv_data;
571  NV_ENC_RC_PARAMS *rc = &ctx->encode_config.rcParams;
572 
573  rc->rateControlMode = NV_ENC_PARAMS_RC_CONSTQP;
574  rc->constQP.qpInterB = 0;
575  rc->constQP.qpInterP = 0;
576  rc->constQP.qpIntra = 0;
577 
578  avctx->qmin = -1;
579  avctx->qmax = -1;
580 }
581 
583 {
584  NvencContext *ctx = avctx->priv_data;
585  NV_ENC_RC_PARAMS *rc = &ctx->encode_config.rcParams;
586 
587  switch (ctx->rc) {
588  case NV_ENC_PARAMS_RC_CONSTQP:
589  if (avctx->global_quality <= 0) {
590  av_log(avctx, AV_LOG_WARNING,
591  "The constant quality rate-control requires "
592  "the 'global_quality' option set.\n");
593  return;
594  }
595  set_constqp(avctx);
596  return;
597  case NV_ENC_PARAMS_RC_2_PASS_VBR:
598  case NV_ENC_PARAMS_RC_VBR:
599  if (avctx->qmin < 0 && avctx->qmax < 0) {
600  av_log(avctx, AV_LOG_WARNING,
601  "The variable bitrate rate-control requires "
602  "the 'qmin' and/or 'qmax' option set.\n");
603  set_vbr(avctx);
604  return;
605  }
606  case NV_ENC_PARAMS_RC_VBR_MINQP:
607  if (avctx->qmin < 0) {
608  av_log(avctx, AV_LOG_WARNING,
609  "The variable bitrate rate-control requires "
610  "the 'qmin' option set.\n");
611  set_vbr(avctx);
612  return;
613  }
614  set_vbr(avctx);
615  break;
616  case NV_ENC_PARAMS_RC_CBR:
617  case NV_ENC_PARAMS_RC_2_PASS_QUALITY:
618  case NV_ENC_PARAMS_RC_2_PASS_FRAMESIZE_CAP:
619  break;
620  }
621 
622  rc->rateControlMode = ctx->rc;
623 }
624 
626 {
627  NvencContext *ctx = avctx->priv_data;
628 
629  if (avctx->bit_rate > 0) {
630  ctx->encode_config.rcParams.averageBitRate = avctx->bit_rate;
631  } else if (ctx->encode_config.rcParams.averageBitRate > 0) {
632  ctx->encode_config.rcParams.maxBitRate = ctx->encode_config.rcParams.averageBitRate;
633  }
634 
635  if (avctx->rc_max_rate > 0)
636  ctx->encode_config.rcParams.maxBitRate = avctx->rc_max_rate;
637 
638  if (ctx->rc < 0) {
639  if (ctx->flags & NVENC_ONE_PASS)
640  ctx->twopass = 0;
641  if (ctx->flags & NVENC_TWO_PASSES)
642  ctx->twopass = 1;
643 
644  if (ctx->twopass < 0)
645  ctx->twopass = (ctx->flags & NVENC_LOWLATENCY) != 0;
646 
647  if (ctx->cbr) {
648  if (ctx->twopass) {
649  ctx->rc = NV_ENC_PARAMS_RC_2_PASS_QUALITY;
650  } else {
651  ctx->rc = NV_ENC_PARAMS_RC_CBR;
652  }
653  } else if (avctx->global_quality > 0) {
654  ctx->rc = NV_ENC_PARAMS_RC_CONSTQP;
655  } else if (ctx->twopass) {
656  ctx->rc = NV_ENC_PARAMS_RC_2_PASS_VBR;
657  } else if (avctx->qmin >= 0 && avctx->qmax >= 0) {
658  ctx->rc = NV_ENC_PARAMS_RC_VBR_MINQP;
659  }
660  }
661 
662  if (ctx->flags & NVENC_LOSSLESS) {
663  set_lossless(avctx);
664  } else if (ctx->rc >= 0) {
666  } else {
667  ctx->encode_config.rcParams.rateControlMode = NV_ENC_PARAMS_RC_VBR;
668  set_vbr(avctx);
669  }
670 
671  if (avctx->rc_buffer_size > 0) {
672  ctx->encode_config.rcParams.vbvBufferSize = avctx->rc_buffer_size;
673  } else if (ctx->encode_config.rcParams.averageBitRate > 0) {
674  ctx->encode_config.rcParams.vbvBufferSize = 2 * ctx->encode_config.rcParams.averageBitRate;
675  }
676 }
677 
679 {
680  NvencContext *ctx = avctx->priv_data;
681  NV_ENC_CONFIG *cc = &ctx->encode_config;
682  NV_ENC_CONFIG_H264 *h264 = &cc->encodeCodecConfig.h264Config;
683  NV_ENC_CONFIG_H264_VUI_PARAMETERS *vui = &h264->h264VUIParameters;
684 
685  vui->colourMatrix = avctx->colorspace;
686  vui->colourPrimaries = avctx->color_primaries;
687  vui->transferCharacteristics = avctx->color_trc;
688  vui->videoFullRangeFlag = (avctx->color_range == AVCOL_RANGE_JPEG
690 
691  vui->colourDescriptionPresentFlag =
692  (avctx->colorspace != 2 || avctx->color_primaries != 2 || avctx->color_trc != 2);
693 
694  vui->videoSignalTypePresentFlag =
695  (vui->colourDescriptionPresentFlag
696  || vui->videoFormat != 5
697  || vui->videoFullRangeFlag != 0);
698 
699  h264->sliceMode = 3;
700  h264->sliceModeData = 1;
701 
702  h264->disableSPSPPS = (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) ? 1 : 0;
703  h264->repeatSPSPPS = (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) ? 0 : 1;
704  h264->outputAUD = 1;
705 
706  if (avctx->refs >= 0) {
707  /* 0 means "let the hardware decide" */
708  h264->maxNumRefFrames = avctx->refs;
709  }
710  if (avctx->gop_size >= 0) {
711  h264->idrPeriod = cc->gopLength;
712  }
713 
714  if (IS_CBR(cc->rcParams.rateControlMode)) {
715  h264->outputBufferingPeriodSEI = 1;
716  h264->outputPictureTimingSEI = 1;
717  }
718 
719  if (cc->rcParams.rateControlMode == NV_ENC_PARAMS_RC_2_PASS_QUALITY ||
720  cc->rcParams.rateControlMode == NV_ENC_PARAMS_RC_2_PASS_FRAMESIZE_CAP ||
721  cc->rcParams.rateControlMode == NV_ENC_PARAMS_RC_2_PASS_VBR) {
722  h264->adaptiveTransformMode = NV_ENC_H264_ADAPTIVE_TRANSFORM_ENABLE;
723  h264->fmoMode = NV_ENC_H264_FMO_DISABLE;
724  }
725 
726  if (ctx->flags & NVENC_LOSSLESS) {
727  h264->qpPrimeYZeroTransformBypassFlag = 1;
728  } else {
729  switch(ctx->profile) {
731  cc->profileGUID = NV_ENC_H264_PROFILE_BASELINE_GUID;
733  break;
735  cc->profileGUID = NV_ENC_H264_PROFILE_MAIN_GUID;
736  avctx->profile = FF_PROFILE_H264_MAIN;
737  break;
739  cc->profileGUID = NV_ENC_H264_PROFILE_HIGH_GUID;
740  avctx->profile = FF_PROFILE_H264_HIGH;
741  break;
743  cc->profileGUID = NV_ENC_H264_PROFILE_HIGH_444_GUID;
745  break;
746  }
747  }
748 
749  // force setting profile as high444p if input is AV_PIX_FMT_YUV444P
750  if (ctx->data_pix_fmt == AV_PIX_FMT_YUV444P) {
751  cc->profileGUID = NV_ENC_H264_PROFILE_HIGH_444_GUID;
753  }
754 
755  h264->chromaFormatIDC = avctx->profile == FF_PROFILE_H264_HIGH_444_PREDICTIVE ? 3 : 1;
756 
757  h264->level = ctx->level;
758 
759  return 0;
760 }
761 
763 {
764  NvencContext *ctx = avctx->priv_data;
765  NV_ENC_CONFIG *cc = &ctx->encode_config;
766  NV_ENC_CONFIG_HEVC *hevc = &cc->encodeCodecConfig.hevcConfig;
767  NV_ENC_CONFIG_HEVC_VUI_PARAMETERS *vui = &hevc->hevcVUIParameters;
768 
769  vui->colourMatrix = avctx->colorspace;
770  vui->colourPrimaries = avctx->color_primaries;
771  vui->transferCharacteristics = avctx->color_trc;
772  vui->videoFullRangeFlag = (avctx->color_range == AVCOL_RANGE_JPEG
774 
775  vui->colourDescriptionPresentFlag =
776  (avctx->colorspace != 2 || avctx->color_primaries != 2 || avctx->color_trc != 2);
777 
778  vui->videoSignalTypePresentFlag =
779  (vui->colourDescriptionPresentFlag
780  || vui->videoFormat != 5
781  || vui->videoFullRangeFlag != 0);
782 
783  hevc->sliceMode = 3;
784  hevc->sliceModeData = 1;
785 
786  hevc->disableSPSPPS = (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) ? 1 : 0;
787  hevc->repeatSPSPPS = (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) ? 0 : 1;
788  hevc->outputAUD = 1;
789 
790  if (avctx->refs >= 0) {
791  /* 0 means "let the hardware decide" */
792  hevc->maxNumRefFramesInDPB = avctx->refs;
793  }
794  if (avctx->gop_size >= 0) {
795  hevc->idrPeriod = cc->gopLength;
796  }
797 
798  if (IS_CBR(cc->rcParams.rateControlMode)) {
799  hevc->outputBufferingPeriodSEI = 1;
800  hevc->outputPictureTimingSEI = 1;
801  }
802 
803  /* No other profile is supported in the current SDK version 5 */
804  cc->profileGUID = NV_ENC_HEVC_PROFILE_MAIN_GUID;
805  avctx->profile = FF_PROFILE_HEVC_MAIN;
806 
807  hevc->level = ctx->level;
808 
809  hevc->tier = ctx->tier;
810 
811  return 0;
812 }
813 
815 {
816  switch (avctx->codec->id) {
817  case AV_CODEC_ID_H264:
818  return nvenc_setup_h264_config(avctx);
819  case AV_CODEC_ID_HEVC:
820  return nvenc_setup_hevc_config(avctx);
821  /* Earlier switch/case will return if unknown codec is passed. */
822  }
823 
824  return 0;
825 }
826 
828 {
829  NvencContext *ctx = avctx->priv_data;
831  NV_ENCODE_API_FUNCTION_LIST *p_nvenc = &dl_fn->nvenc_funcs;
832 
833  NV_ENC_PRESET_CONFIG preset_config = { 0 };
834  NVENCSTATUS nv_status = NV_ENC_SUCCESS;
835  AVCPBProperties *cpb_props;
836  int res = 0;
837  int dw, dh;
838 
839  ctx->encode_config.version = NV_ENC_CONFIG_VER;
840  ctx->init_encode_params.version = NV_ENC_INITIALIZE_PARAMS_VER;
841 
842  ctx->init_encode_params.encodeHeight = avctx->height;
843  ctx->init_encode_params.encodeWidth = avctx->width;
844 
845  ctx->init_encode_params.encodeConfig = &ctx->encode_config;
846 
847  nvenc_map_preset(ctx);
848 
849  preset_config.version = NV_ENC_PRESET_CONFIG_VER;
850  preset_config.presetCfg.version = NV_ENC_CONFIG_VER;
851 
852  nv_status = p_nvenc->nvEncGetEncodePresetConfig(ctx->nvencoder,
853  ctx->init_encode_params.encodeGUID,
854  ctx->init_encode_params.presetGUID,
855  &preset_config);
856  if (nv_status != NV_ENC_SUCCESS)
857  return nvenc_print_error(avctx, nv_status, "Cannot get the preset configuration");
858 
859  memcpy(&ctx->encode_config, &preset_config.presetCfg, sizeof(ctx->encode_config));
860 
861  ctx->encode_config.version = NV_ENC_CONFIG_VER;
862 
863  if (avctx->sample_aspect_ratio.num && avctx->sample_aspect_ratio.den &&
864  (avctx->sample_aspect_ratio.num != 1 || avctx->sample_aspect_ratio.num != 1)) {
865  av_reduce(&dw, &dh,
866  avctx->width * avctx->sample_aspect_ratio.num,
867  avctx->height * avctx->sample_aspect_ratio.den,
868  1024 * 1024);
869  ctx->init_encode_params.darHeight = dh;
870  ctx->init_encode_params.darWidth = dw;
871  } else {
872  ctx->init_encode_params.darHeight = avctx->height;
873  ctx->init_encode_params.darWidth = avctx->width;
874  }
875 
876  // De-compensate for hardware, dubiously, trying to compensate for
877  // playback at 704 pixel width.
878  if (avctx->width == 720 &&
879  (avctx->height == 480 || avctx->height == 576)) {
880  av_reduce(&dw, &dh,
881  ctx->init_encode_params.darWidth * 44,
882  ctx->init_encode_params.darHeight * 45,
883  1024 * 1024);
884  ctx->init_encode_params.darHeight = dh;
885  ctx->init_encode_params.darWidth = dw;
886  }
887 
888  ctx->init_encode_params.frameRateNum = avctx->time_base.den;
889  ctx->init_encode_params.frameRateDen = avctx->time_base.num * avctx->ticks_per_frame;
890 
891  ctx->init_encode_params.enableEncodeAsync = 0;
892  ctx->init_encode_params.enablePTD = 1;
893 
894  if (avctx->gop_size > 0) {
895  if (avctx->max_b_frames >= 0) {
896  /* 0 is intra-only, 1 is I/P only, 2 is one B-Frame, 3 two B-frames, and so on. */
897  ctx->encode_config.frameIntervalP = avctx->max_b_frames + 1;
898  }
899 
900  ctx->encode_config.gopLength = avctx->gop_size;
901  } else if (avctx->gop_size == 0) {
902  ctx->encode_config.frameIntervalP = 0;
903  ctx->encode_config.gopLength = 1;
904  }
905 
906  ctx->initial_pts[0] = AV_NOPTS_VALUE;
907  ctx->initial_pts[1] = AV_NOPTS_VALUE;
908 
910 
911  if (avctx->flags & AV_CODEC_FLAG_INTERLACED_DCT) {
912  ctx->encode_config.frameFieldMode = NV_ENC_PARAMS_FRAME_FIELD_MODE_FIELD;
913  } else {
914  ctx->encode_config.frameFieldMode = NV_ENC_PARAMS_FRAME_FIELD_MODE_FRAME;
915  }
916 
917  res = nvenc_setup_codec_config(avctx);
918  if (res)
919  return res;
920 
921  nv_status = p_nvenc->nvEncInitializeEncoder(ctx->nvencoder, &ctx->init_encode_params);
922  if (nv_status != NV_ENC_SUCCESS) {
923  return nvenc_print_error(avctx, nv_status, "InitializeEncoder failed");
924  }
925 
926  if (ctx->encode_config.frameIntervalP > 1)
927  avctx->has_b_frames = 2;
928 
929  if (ctx->encode_config.rcParams.averageBitRate > 0)
930  avctx->bit_rate = ctx->encode_config.rcParams.averageBitRate;
931 
932  cpb_props = ff_add_cpb_side_data(avctx);
933  if (!cpb_props)
934  return AVERROR(ENOMEM);
935  cpb_props->max_bitrate = ctx->encode_config.rcParams.maxBitRate;
936  cpb_props->avg_bitrate = avctx->bit_rate;
937  cpb_props->buffer_size = ctx->encode_config.rcParams.vbvBufferSize;
938 
939  return 0;
940 }
941 
942 static av_cold int nvenc_alloc_surface(AVCodecContext *avctx, int idx)
943 {
944  NvencContext *ctx = avctx->priv_data;
946  NV_ENCODE_API_FUNCTION_LIST *p_nvenc = &dl_fn->nvenc_funcs;
947 
948  NVENCSTATUS nv_status;
949  NV_ENC_CREATE_BITSTREAM_BUFFER allocOut = { 0 };
950  allocOut.version = NV_ENC_CREATE_BITSTREAM_BUFFER_VER;
951 
952  switch (ctx->data_pix_fmt) {
953  case AV_PIX_FMT_YUV420P:
954  ctx->surfaces[idx].format = NV_ENC_BUFFER_FORMAT_YV12_PL;
955  break;
956 
957  case AV_PIX_FMT_NV12:
958  ctx->surfaces[idx].format = NV_ENC_BUFFER_FORMAT_NV12_PL;
959  break;
960 
961  case AV_PIX_FMT_YUV444P:
962  ctx->surfaces[idx].format = NV_ENC_BUFFER_FORMAT_YUV444_PL;
963  break;
964 
965  default:
966  av_log(avctx, AV_LOG_FATAL, "Invalid input pixel format\n");
967  return AVERROR(EINVAL);
968  }
969 
970  if (avctx->pix_fmt == AV_PIX_FMT_CUDA) {
971  ctx->surfaces[idx].in_ref = av_frame_alloc();
972  if (!ctx->surfaces[idx].in_ref)
973  return AVERROR(ENOMEM);
974  } else {
975  NV_ENC_CREATE_INPUT_BUFFER allocSurf = { 0 };
976  allocSurf.version = NV_ENC_CREATE_INPUT_BUFFER_VER;
977  allocSurf.width = (avctx->width + 31) & ~31;
978  allocSurf.height = (avctx->height + 31) & ~31;
979  allocSurf.memoryHeap = NV_ENC_MEMORY_HEAP_SYSMEM_CACHED;
980  allocSurf.bufferFmt = ctx->surfaces[idx].format;
981 
982  nv_status = p_nvenc->nvEncCreateInputBuffer(ctx->nvencoder, &allocSurf);
983  if (nv_status != NV_ENC_SUCCESS) {
984  return nvenc_print_error(avctx, nv_status, "CreateInputBuffer failed");
985  }
986 
987  ctx->surfaces[idx].input_surface = allocSurf.inputBuffer;
988  ctx->surfaces[idx].width = allocSurf.width;
989  ctx->surfaces[idx].height = allocSurf.height;
990  }
991 
992  ctx->surfaces[idx].lockCount = 0;
993 
994  /* 1MB is large enough to hold most output frames.
995  * NVENC increases this automaticaly if it is not enough. */
996  allocOut.size = 1024 * 1024;
997 
998  allocOut.memoryHeap = NV_ENC_MEMORY_HEAP_SYSMEM_CACHED;
999 
1000  nv_status = p_nvenc->nvEncCreateBitstreamBuffer(ctx->nvencoder, &allocOut);
1001  if (nv_status != NV_ENC_SUCCESS) {
1002  int err = nvenc_print_error(avctx, nv_status, "CreateBitstreamBuffer failed");
1003  if (avctx->pix_fmt != AV_PIX_FMT_CUDA)
1004  p_nvenc->nvEncDestroyInputBuffer(ctx->nvencoder, ctx->surfaces[idx].input_surface);
1005  av_frame_free(&ctx->surfaces[idx].in_ref);
1006  return err;
1007  }
1008 
1009  ctx->surfaces[idx].output_surface = allocOut.bitstreamBuffer;
1010  ctx->surfaces[idx].size = allocOut.size;
1011 
1012  return 0;
1013 }
1014 
1016 {
1017  NvencContext *ctx = avctx->priv_data;
1018  int i, res;
1019  int num_mbs = ((avctx->width + 15) >> 4) * ((avctx->height + 15) >> 4);
1020  ctx->nb_surfaces = FFMAX((num_mbs >= 8160) ? 32 : 48,
1021  ctx->nb_surfaces);
1022  ctx->async_depth = FFMIN(ctx->async_depth, ctx->nb_surfaces - 1);
1023 
1024 
1025  ctx->surfaces = av_mallocz_array(ctx->nb_surfaces, sizeof(*ctx->surfaces));
1026  if (!ctx->surfaces)
1027  return AVERROR(ENOMEM);
1028 
1029  ctx->timestamp_list = av_fifo_alloc(ctx->nb_surfaces * sizeof(int64_t));
1030  if (!ctx->timestamp_list)
1031  return AVERROR(ENOMEM);
1033  if (!ctx->output_surface_queue)
1034  return AVERROR(ENOMEM);
1036  if (!ctx->output_surface_ready_queue)
1037  return AVERROR(ENOMEM);
1038 
1039  for (i = 0; i < ctx->nb_surfaces; i++) {
1040  if ((res = nvenc_alloc_surface(avctx, i)) < 0)
1041  return res;
1042  }
1043 
1044  return 0;
1045 }
1046 
1048 {
1049  NvencContext *ctx = avctx->priv_data;
1051  NV_ENCODE_API_FUNCTION_LIST *p_nvenc = &dl_fn->nvenc_funcs;
1052 
1053  NVENCSTATUS nv_status;
1054  uint32_t outSize = 0;
1055  char tmpHeader[256];
1056  NV_ENC_SEQUENCE_PARAM_PAYLOAD payload = { 0 };
1057  payload.version = NV_ENC_SEQUENCE_PARAM_PAYLOAD_VER;
1058 
1059  payload.spsppsBuffer = tmpHeader;
1060  payload.inBufferSize = sizeof(tmpHeader);
1061  payload.outSPSPPSPayloadSize = &outSize;
1062 
1063  nv_status = p_nvenc->nvEncGetSequenceParams(ctx->nvencoder, &payload);
1064  if (nv_status != NV_ENC_SUCCESS) {
1065  return nvenc_print_error(avctx, nv_status, "GetSequenceParams failed");
1066  }
1067 
1068  avctx->extradata_size = outSize;
1070 
1071  if (!avctx->extradata) {
1072  return AVERROR(ENOMEM);
1073  }
1074 
1075  memcpy(avctx->extradata, tmpHeader, outSize);
1076 
1077  return 0;
1078 }
1079 
1081 {
1082  NvencContext *ctx = avctx->priv_data;
1084  NV_ENCODE_API_FUNCTION_LIST *p_nvenc = &dl_fn->nvenc_funcs;
1085  int i;
1086 
1087  /* the encoder has to be flushed before it can be closed */
1088  if (ctx->nvencoder) {
1089  NV_ENC_PIC_PARAMS params = { .version = NV_ENC_PIC_PARAMS_VER,
1090  .encodePicFlags = NV_ENC_PIC_FLAG_EOS };
1091 
1092  p_nvenc->nvEncEncodePicture(ctx->nvencoder, &params);
1093  }
1094 
1098 
1099  if (ctx->surfaces && avctx->pix_fmt == AV_PIX_FMT_CUDA) {
1100  for (i = 0; i < ctx->nb_surfaces; ++i) {
1101  if (ctx->surfaces[i].input_surface) {
1102  p_nvenc->nvEncUnmapInputResource(ctx->nvencoder, ctx->surfaces[i].in_map.mappedResource);
1103  }
1104  }
1105  for (i = 0; i < ctx->nb_registered_frames; i++) {
1106  if (ctx->registered_frames[i].regptr)
1107  p_nvenc->nvEncUnregisterResource(ctx->nvencoder, ctx->registered_frames[i].regptr);
1108  }
1109  ctx->nb_registered_frames = 0;
1110  }
1111 
1112  if (ctx->surfaces) {
1113  for (i = 0; i < ctx->nb_surfaces; ++i) {
1114  if (avctx->pix_fmt != AV_PIX_FMT_CUDA)
1115  p_nvenc->nvEncDestroyInputBuffer(ctx->nvencoder, ctx->surfaces[i].input_surface);
1116  av_frame_free(&ctx->surfaces[i].in_ref);
1117  p_nvenc->nvEncDestroyBitstreamBuffer(ctx->nvencoder, ctx->surfaces[i].output_surface);
1118  }
1119  }
1120  av_freep(&ctx->surfaces);
1121  ctx->nb_surfaces = 0;
1122 
1123  if (ctx->nvencoder)
1124  p_nvenc->nvEncDestroyEncoder(ctx->nvencoder);
1125  ctx->nvencoder = NULL;
1126 
1127  if (ctx->cu_context_internal)
1128  dl_fn->cu_ctx_destroy(ctx->cu_context_internal);
1129  ctx->cu_context = ctx->cu_context_internal = NULL;
1130 
1131  if (dl_fn->nvenc)
1132  dlclose(dl_fn->nvenc);
1133  dl_fn->nvenc = NULL;
1134 
1135  dl_fn->nvenc_device_count = 0;
1136 
1137 #if !CONFIG_CUDA
1138  if (dl_fn->cuda)
1139  dlclose(dl_fn->cuda);
1140  dl_fn->cuda = NULL;
1141 #endif
1142 
1143  dl_fn->cu_init = NULL;
1144  dl_fn->cu_device_get_count = NULL;
1145  dl_fn->cu_device_get = NULL;
1146  dl_fn->cu_device_get_name = NULL;
1148  dl_fn->cu_ctx_create = NULL;
1149  dl_fn->cu_ctx_pop_current = NULL;
1150  dl_fn->cu_ctx_destroy = NULL;
1151 
1152  av_log(avctx, AV_LOG_VERBOSE, "Nvenc unloaded\n");
1153 
1154  return 0;
1155 }
1156 
1158 {
1159  NvencContext *ctx = avctx->priv_data;
1160  int ret;
1161 
1162  if (avctx->pix_fmt == AV_PIX_FMT_CUDA) {
1163  AVHWFramesContext *frames_ctx;
1164  if (!avctx->hw_frames_ctx) {
1165  av_log(avctx, AV_LOG_ERROR,
1166  "hw_frames_ctx must be set when using GPU frames as input\n");
1167  return AVERROR(EINVAL);
1168  }
1169  frames_ctx = (AVHWFramesContext*)avctx->hw_frames_ctx->data;
1170  ctx->data_pix_fmt = frames_ctx->sw_format;
1171  } else {
1172  ctx->data_pix_fmt = avctx->pix_fmt;
1173  }
1174 
1175  if ((ret = nvenc_load_libraries(avctx)) < 0)
1176  return ret;
1177 
1178  if ((ret = nvenc_setup_device(avctx)) < 0)
1179  return ret;
1180 
1181  if ((ret = nvenc_setup_encoder(avctx)) < 0)
1182  return ret;
1183 
1184  if ((ret = nvenc_setup_surfaces(avctx)) < 0)
1185  return ret;
1186 
1187  if (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) {
1188  if ((ret = nvenc_setup_extradata(avctx)) < 0)
1189  return ret;
1190  }
1191 
1192  return 0;
1193 }
1194 
1196 {
1197  int i;
1198 
1199  for (i = 0; i < ctx->nb_surfaces; ++i) {
1200  if (!ctx->surfaces[i].lockCount) {
1201  ctx->surfaces[i].lockCount = 1;
1202  return &ctx->surfaces[i];
1203  }
1204  }
1205 
1206  return NULL;
1207 }
1208 
1209 static int nvenc_copy_frame(AVCodecContext *avctx, NvencSurface *inSurf,
1210  NV_ENC_LOCK_INPUT_BUFFER *lockBufferParams, const AVFrame *frame)
1211 {
1212  uint8_t *buf = lockBufferParams->bufferDataPtr;
1213  int off = inSurf->height * lockBufferParams->pitch;
1214 
1215  if (frame->format == AV_PIX_FMT_YUV420P) {
1216  av_image_copy_plane(buf, lockBufferParams->pitch,
1217  frame->data[0], frame->linesize[0],
1218  avctx->width, avctx->height);
1219 
1220  buf += off;
1221 
1222  av_image_copy_plane(buf, lockBufferParams->pitch >> 1,
1223  frame->data[2], frame->linesize[2],
1224  avctx->width >> 1, avctx->height >> 1);
1225 
1226  buf += off >> 2;
1227 
1228  av_image_copy_plane(buf, lockBufferParams->pitch >> 1,
1229  frame->data[1], frame->linesize[1],
1230  avctx->width >> 1, avctx->height >> 1);
1231  } else if (frame->format == AV_PIX_FMT_NV12) {
1232  av_image_copy_plane(buf, lockBufferParams->pitch,
1233  frame->data[0], frame->linesize[0],
1234  avctx->width, avctx->height);
1235 
1236  buf += off;
1237 
1238  av_image_copy_plane(buf, lockBufferParams->pitch,
1239  frame->data[1], frame->linesize[1],
1240  avctx->width, avctx->height >> 1);
1241  } else if (frame->format == AV_PIX_FMT_YUV444P) {
1242  av_image_copy_plane(buf, lockBufferParams->pitch,
1243  frame->data[0], frame->linesize[0],
1244  avctx->width, avctx->height);
1245 
1246  buf += off;
1247 
1248  av_image_copy_plane(buf, lockBufferParams->pitch,
1249  frame->data[1], frame->linesize[1],
1250  avctx->width, avctx->height);
1251 
1252  buf += off;
1253 
1254  av_image_copy_plane(buf, lockBufferParams->pitch,
1255  frame->data[2], frame->linesize[2],
1256  avctx->width, avctx->height);
1257  } else {
1258  av_log(avctx, AV_LOG_FATAL, "Invalid pixel format!\n");
1259  return AVERROR(EINVAL);
1260  }
1261 
1262  return 0;
1263 }
1264 
1266 {
1267  NvencContext *ctx = avctx->priv_data;
1269  NV_ENCODE_API_FUNCTION_LIST *p_nvenc = &dl_fn->nvenc_funcs;
1270 
1271  int i;
1272 
1274  for (i = 0; i < ctx->nb_registered_frames; i++) {
1275  if (!ctx->registered_frames[i].mapped) {
1276  if (ctx->registered_frames[i].regptr) {
1277  p_nvenc->nvEncUnregisterResource(ctx->nvencoder,
1278  ctx->registered_frames[i].regptr);
1279  ctx->registered_frames[i].regptr = NULL;
1280  }
1281  return i;
1282  }
1283  }
1284  } else {
1285  return ctx->nb_registered_frames++;
1286  }
1287 
1288  av_log(avctx, AV_LOG_ERROR, "Too many registered CUDA frames\n");
1289  return AVERROR(ENOMEM);
1290 }
1291 
1293 {
1294  NvencContext *ctx = avctx->priv_data;
1296  NV_ENCODE_API_FUNCTION_LIST *p_nvenc = &dl_fn->nvenc_funcs;
1297 
1298  AVHWFramesContext *frames_ctx = (AVHWFramesContext*)avctx->hw_frames_ctx->data;
1299  NV_ENC_REGISTER_RESOURCE reg;
1300  int i, idx, ret;
1301 
1302  for (i = 0; i < ctx->nb_registered_frames; i++) {
1303  if (ctx->registered_frames[i].ptr == (CUdeviceptr)frame->data[0])
1304  return i;
1305  }
1306 
1307  idx = nvenc_find_free_reg_resource(avctx);
1308  if (idx < 0)
1309  return idx;
1310 
1311  reg.version = NV_ENC_REGISTER_RESOURCE_VER;
1312  reg.resourceType = NV_ENC_INPUT_RESOURCE_TYPE_CUDADEVICEPTR;
1313  reg.width = frames_ctx->width;
1314  reg.height = frames_ctx->height;
1315  reg.bufferFormat = ctx->surfaces[0].format;
1316  reg.pitch = frame->linesize[0];
1317  reg.resourceToRegister = frame->data[0];
1318 
1319  ret = p_nvenc->nvEncRegisterResource(ctx->nvencoder, &reg);
1320  if (ret != NV_ENC_SUCCESS) {
1321  nvenc_print_error(avctx, ret, "Error registering an input resource");
1322  return AVERROR_UNKNOWN;
1323  }
1324 
1325  ctx->registered_frames[idx].ptr = (CUdeviceptr)frame->data[0];
1326  ctx->registered_frames[idx].regptr = reg.registeredResource;
1327  return idx;
1328 }
1329 
1331  NvencSurface *nvenc_frame)
1332 {
1333  NvencContext *ctx = avctx->priv_data;
1335  NV_ENCODE_API_FUNCTION_LIST *p_nvenc = &dl_fn->nvenc_funcs;
1336 
1337  int res;
1338  NVENCSTATUS nv_status;
1339 
1340  if (avctx->pix_fmt == AV_PIX_FMT_CUDA) {
1341  int reg_idx = nvenc_register_frame(avctx, frame);
1342  if (reg_idx < 0) {
1343  av_log(avctx, AV_LOG_ERROR, "Could not register an input CUDA frame\n");
1344  return reg_idx;
1345  }
1346 
1347  res = av_frame_ref(nvenc_frame->in_ref, frame);
1348  if (res < 0)
1349  return res;
1350 
1351  nvenc_frame->in_map.version = NV_ENC_MAP_INPUT_RESOURCE_VER;
1352  nvenc_frame->in_map.registeredResource = ctx->registered_frames[reg_idx].regptr;
1353  nv_status = p_nvenc->nvEncMapInputResource(ctx->nvencoder, &nvenc_frame->in_map);
1354  if (nv_status != NV_ENC_SUCCESS) {
1355  av_frame_unref(nvenc_frame->in_ref);
1356  return nvenc_print_error(avctx, nv_status, "Error mapping an input resource");
1357  }
1358 
1359  ctx->registered_frames[reg_idx].mapped = 1;
1360  nvenc_frame->reg_idx = reg_idx;
1361  nvenc_frame->input_surface = nvenc_frame->in_map.mappedResource;
1362  return 0;
1363  } else {
1364  NV_ENC_LOCK_INPUT_BUFFER lockBufferParams = { 0 };
1365 
1366  lockBufferParams.version = NV_ENC_LOCK_INPUT_BUFFER_VER;
1367  lockBufferParams.inputBuffer = nvenc_frame->input_surface;
1368 
1369  nv_status = p_nvenc->nvEncLockInputBuffer(ctx->nvencoder, &lockBufferParams);
1370  if (nv_status != NV_ENC_SUCCESS) {
1371  return nvenc_print_error(avctx, nv_status, "Failed locking nvenc input buffer");
1372  }
1373 
1374  res = nvenc_copy_frame(avctx, nvenc_frame, &lockBufferParams, frame);
1375 
1376  nv_status = p_nvenc->nvEncUnlockInputBuffer(ctx->nvencoder, nvenc_frame->input_surface);
1377  if (nv_status != NV_ENC_SUCCESS) {
1378  return nvenc_print_error(avctx, nv_status, "Failed unlocking input buffer!");
1379  }
1380 
1381  return res;
1382  }
1383 }
1384 
1386  NV_ENC_PIC_PARAMS *params)
1387 {
1388  NvencContext *ctx = avctx->priv_data;
1389 
1390  switch (avctx->codec->id) {
1391  case AV_CODEC_ID_H264:
1392  params->codecPicParams.h264PicParams.sliceMode =
1393  ctx->encode_config.encodeCodecConfig.h264Config.sliceMode;
1394  params->codecPicParams.h264PicParams.sliceModeData =
1395  ctx->encode_config.encodeCodecConfig.h264Config.sliceModeData;
1396  break;
1397  case AV_CODEC_ID_HEVC:
1398  params->codecPicParams.hevcPicParams.sliceMode =
1399  ctx->encode_config.encodeCodecConfig.hevcConfig.sliceMode;
1400  params->codecPicParams.hevcPicParams.sliceModeData =
1401  ctx->encode_config.encodeCodecConfig.hevcConfig.sliceModeData;
1402  break;
1403  }
1404 }
1405 
1406 static inline void timestamp_queue_enqueue(AVFifoBuffer* queue, int64_t timestamp)
1407 {
1408  av_fifo_generic_write(queue, &timestamp, sizeof(timestamp), NULL);
1409 }
1410 
1411 static inline int64_t timestamp_queue_dequeue(AVFifoBuffer* queue)
1412 {
1413  int64_t timestamp = AV_NOPTS_VALUE;
1414  if (av_fifo_size(queue) > 0)
1415  av_fifo_generic_read(queue, &timestamp, sizeof(timestamp), NULL);
1416 
1417  return timestamp;
1418 }
1419 
1421  NV_ENC_LOCK_BITSTREAM *params,
1422  AVPacket *pkt)
1423 {
1424  NvencContext *ctx = avctx->priv_data;
1425 
1426  pkt->pts = params->outputTimeStamp;
1427 
1428  /* generate the first dts by linearly extrapolating the
1429  * first two pts values to the past */
1430  if (avctx->max_b_frames > 0 && !ctx->first_packet_output &&
1431  ctx->initial_pts[1] != AV_NOPTS_VALUE) {
1432  int64_t ts0 = ctx->initial_pts[0], ts1 = ctx->initial_pts[1];
1433  int64_t delta;
1434 
1435  if ((ts0 < 0 && ts1 > INT64_MAX + ts0) ||
1436  (ts0 > 0 && ts1 < INT64_MIN + ts0))
1437  return AVERROR(ERANGE);
1438  delta = ts1 - ts0;
1439 
1440  if ((delta < 0 && ts0 > INT64_MAX + delta) ||
1441  (delta > 0 && ts0 < INT64_MIN + delta))
1442  return AVERROR(ERANGE);
1443  pkt->dts = ts0 - delta;
1444 
1445  ctx->first_packet_output = 1;
1446  return 0;
1447  }
1448 
1450 
1451  return 0;
1452 }
1453 
1455 {
1456  NvencContext *ctx = avctx->priv_data;
1458  NV_ENCODE_API_FUNCTION_LIST *p_nvenc = &dl_fn->nvenc_funcs;
1459 
1460  uint32_t slice_mode_data;
1461  uint32_t *slice_offsets;
1462  NV_ENC_LOCK_BITSTREAM lock_params = { 0 };
1463  NVENCSTATUS nv_status;
1464  int res = 0;
1465 
1466  enum AVPictureType pict_type;
1467 
1468  switch (avctx->codec->id) {
1469  case AV_CODEC_ID_H264:
1470  slice_mode_data = ctx->encode_config.encodeCodecConfig.h264Config.sliceModeData;
1471  break;
1472  case AV_CODEC_ID_H265:
1473  slice_mode_data = ctx->encode_config.encodeCodecConfig.hevcConfig.sliceModeData;
1474  break;
1475  default:
1476  av_log(avctx, AV_LOG_ERROR, "Unknown codec name\n");
1477  res = AVERROR(EINVAL);
1478  goto error;
1479  }
1480  slice_offsets = av_mallocz(slice_mode_data * sizeof(*slice_offsets));
1481 
1482  if (!slice_offsets)
1483  goto error;
1484 
1485  lock_params.version = NV_ENC_LOCK_BITSTREAM_VER;
1486 
1487  lock_params.doNotWait = 0;
1488  lock_params.outputBitstream = tmpoutsurf->output_surface;
1489  lock_params.sliceOffsets = slice_offsets;
1490 
1491  nv_status = p_nvenc->nvEncLockBitstream(ctx->nvencoder, &lock_params);
1492  if (nv_status != NV_ENC_SUCCESS) {
1493  res = nvenc_print_error(avctx, nv_status, "Failed locking bitstream buffer");
1494  goto error;
1495  }
1496 
1497  if (res = ff_alloc_packet2(avctx, pkt, lock_params.bitstreamSizeInBytes,0)) {
1498  p_nvenc->nvEncUnlockBitstream(ctx->nvencoder, tmpoutsurf->output_surface);
1499  goto error;
1500  }
1501 
1502  memcpy(pkt->data, lock_params.bitstreamBufferPtr, lock_params.bitstreamSizeInBytes);
1503 
1504  nv_status = p_nvenc->nvEncUnlockBitstream(ctx->nvencoder, tmpoutsurf->output_surface);
1505  if (nv_status != NV_ENC_SUCCESS)
1506  nvenc_print_error(avctx, nv_status, "Failed unlocking bitstream buffer, expect the gates of mordor to open");
1507 
1508 
1509  if (avctx->pix_fmt == AV_PIX_FMT_CUDA) {
1510  p_nvenc->nvEncUnmapInputResource(ctx->nvencoder, tmpoutsurf->in_map.mappedResource);
1511  av_frame_unref(tmpoutsurf->in_ref);
1512  ctx->registered_frames[tmpoutsurf->reg_idx].mapped = 0;
1513 
1514  tmpoutsurf->input_surface = NULL;
1515  }
1516 
1517  switch (lock_params.pictureType) {
1518  case NV_ENC_PIC_TYPE_IDR:
1519  pkt->flags |= AV_PKT_FLAG_KEY;
1520  case NV_ENC_PIC_TYPE_I:
1521  pict_type = AV_PICTURE_TYPE_I;
1522  break;
1523  case NV_ENC_PIC_TYPE_P:
1524  pict_type = AV_PICTURE_TYPE_P;
1525  break;
1526  case NV_ENC_PIC_TYPE_B:
1527  pict_type = AV_PICTURE_TYPE_B;
1528  break;
1529  case NV_ENC_PIC_TYPE_BI:
1530  pict_type = AV_PICTURE_TYPE_BI;
1531  break;
1532  default:
1533  av_log(avctx, AV_LOG_ERROR, "Unknown picture type encountered, expect the output to be broken.\n");
1534  av_log(avctx, AV_LOG_ERROR, "Please report this error and include as much information on how to reproduce it as possible.\n");
1535  res = AVERROR_EXTERNAL;
1536  goto error;
1537  }
1538 
1539 #if FF_API_CODED_FRAME
1541  avctx->coded_frame->pict_type = pict_type;
1543 #endif
1544 
1546  (lock_params.frameAvgQP - 1) * FF_QP2LAMBDA, NULL, 0, pict_type);
1547 
1548  res = nvenc_set_timestamp(avctx, &lock_params, pkt);
1549  if (res < 0)
1550  goto error2;
1551 
1552  av_free(slice_offsets);
1553 
1554  return 0;
1555 
1556 error:
1558 
1559 error2:
1560  av_free(slice_offsets);
1561 
1562  return res;
1563 }
1564 
1565 static int output_ready(AVCodecContext *avctx, int flush)
1566 {
1567  NvencContext *ctx = avctx->priv_data;
1568  int nb_ready, nb_pending;
1569 
1570  /* when B-frames are enabled, we wait for two initial timestamps to
1571  * calculate the first dts */
1572  if (!flush && avctx->max_b_frames > 0 &&
1573  (ctx->initial_pts[0] == AV_NOPTS_VALUE || ctx->initial_pts[1] == AV_NOPTS_VALUE))
1574  return 0;
1575 
1576  nb_ready = av_fifo_size(ctx->output_surface_ready_queue) / sizeof(NvencSurface*);
1577  nb_pending = av_fifo_size(ctx->output_surface_queue) / sizeof(NvencSurface*);
1578  if (flush)
1579  return nb_ready > 0;
1580  return (nb_ready > 0) && (nb_ready + nb_pending >= ctx->async_depth);
1581 }
1582 
1584  const AVFrame *frame, int *got_packet)
1585 {
1586  NVENCSTATUS nv_status;
1587  NvencSurface *tmpoutsurf, *inSurf;
1588  int res;
1589 
1590  NvencContext *ctx = avctx->priv_data;
1592  NV_ENCODE_API_FUNCTION_LIST *p_nvenc = &dl_fn->nvenc_funcs;
1593 
1594  NV_ENC_PIC_PARAMS pic_params = { 0 };
1595  pic_params.version = NV_ENC_PIC_PARAMS_VER;
1596 
1597  if (frame) {
1598  inSurf = get_free_frame(ctx);
1599  if (!inSurf) {
1600  av_log(avctx, AV_LOG_ERROR, "No free surfaces\n");
1601  return AVERROR_BUG;
1602  }
1603 
1604  res = nvenc_upload_frame(avctx, frame, inSurf);
1605  if (res) {
1606  inSurf->lockCount = 0;
1607  return res;
1608  }
1609 
1610  pic_params.inputBuffer = inSurf->input_surface;
1611  pic_params.bufferFmt = inSurf->format;
1612  pic_params.inputWidth = avctx->width;
1613  pic_params.inputHeight = avctx->height;
1614  pic_params.outputBitstream = inSurf->output_surface;
1615 
1616  if (avctx->flags & AV_CODEC_FLAG_INTERLACED_DCT) {
1617  if (frame->top_field_first)
1618  pic_params.pictureStruct = NV_ENC_PIC_STRUCT_FIELD_TOP_BOTTOM;
1619  else
1620  pic_params.pictureStruct = NV_ENC_PIC_STRUCT_FIELD_BOTTOM_TOP;
1621  } else {
1622  pic_params.pictureStruct = NV_ENC_PIC_STRUCT_FRAME;
1623  }
1624 
1625  pic_params.encodePicFlags = 0;
1626  pic_params.inputTimeStamp = frame->pts;
1627 
1628  nvenc_codec_specific_pic_params(avctx, &pic_params);
1629  } else {
1630  pic_params.encodePicFlags = NV_ENC_PIC_FLAG_EOS;
1631  }
1632 
1633  nv_status = p_nvenc->nvEncEncodePicture(ctx->nvencoder, &pic_params);
1634  if (nv_status != NV_ENC_SUCCESS &&
1635  nv_status != NV_ENC_ERR_NEED_MORE_INPUT)
1636  return nvenc_print_error(avctx, nv_status, "EncodePicture failed!");
1637 
1638  if (frame) {
1639  av_fifo_generic_write(ctx->output_surface_queue, &inSurf, sizeof(inSurf), NULL);
1641 
1642  if (ctx->initial_pts[0] == AV_NOPTS_VALUE)
1643  ctx->initial_pts[0] = frame->pts;
1644  else if (ctx->initial_pts[1] == AV_NOPTS_VALUE)
1645  ctx->initial_pts[1] = frame->pts;
1646  }
1647 
1648  /* all the pending buffers are now ready for output */
1649  if (nv_status == NV_ENC_SUCCESS) {
1650  while (av_fifo_size(ctx->output_surface_queue) > 0) {
1651  av_fifo_generic_read(ctx->output_surface_queue, &tmpoutsurf, sizeof(tmpoutsurf), NULL);
1652  av_fifo_generic_write(ctx->output_surface_ready_queue, &tmpoutsurf, sizeof(tmpoutsurf), NULL);
1653  }
1654  }
1655 
1656  if (output_ready(avctx, !frame)) {
1657  av_fifo_generic_read(ctx->output_surface_ready_queue, &tmpoutsurf, sizeof(tmpoutsurf), NULL);
1658 
1659  res = process_output_surface(avctx, pkt, tmpoutsurf);
1660 
1661  if (res)
1662  return res;
1663 
1664  av_assert0(tmpoutsurf->lockCount);
1665  tmpoutsurf->lockCount--;
1666 
1667  *got_packet = 1;
1668  } else {
1669  *got_packet = 0;
1670  }
1671 
1672  return 0;
1673 }
const GUID guid
Definition: nvenc.c:482
#define FF_PROFILE_H264_MAIN
Definition: avcodec.h:3187
#define NULL
Definition: coverity.c:32
const struct AVCodec * codec
Definition: avcodec.h:1658
const char const char void * val
Definition: avisynth_c.h:634
BI type.
Definition: avutil.h:272
void * nvencoder
Definition: nvenc.h:165
av_cold int ff_nvenc_encode_close(AVCodecContext *avctx)
Definition: nvenc.c:1080
int twopass
Definition: nvenc.h:173
NV_ENC_BUFFER_FORMAT format
Definition: nvenc.h:61
int height
Definition: nvenc.h:58
This structure describes decoded (raw) audio or video data.
Definition: frame.h:184
static av_cold int nvenc_setup_codec_config(AVCodecContext *avctx)
Definition: nvenc.c:814
AVFifoBuffer * timestamp_list
Definition: nvenc.h:147
int ff_side_data_set_encoder_stats(AVPacket *pkt, int quality, int64_t *error, int error_count, int pict_type)
Definition: avpacket.c:641
static void flush(AVCodecContext *avctx)
int mapped
Definition: nvenc.h:152
#define AV_CODEC_FLAG_INTERLACED_DCT
Use interlaced DCT.
Definition: avcodec.h:871
static av_cold void set_vbr(AVCodecContext *avctx)
Definition: nvenc.c:523
AVFrame * in_ref
Definition: nvenc.h:54
planar YUV 4:4:4, 24bpp, (1 Cr & Cb sample per 1x1 Y samples)
Definition: pixfmt.h:67
misc image utilities
#define AV_LOG_WARNING
Something somehow does not look correct.
Definition: log.h:182
int64_t bit_rate
the average bitrate
Definition: avcodec.h:1714
memory handling functions
static av_cold int nvenc_setup_device(AVCodecContext *avctx)
Definition: nvenc.c:396
const char * desc
Definition: nvenc.c:89
int max_bitrate
Maximum bitrate of the stream, in bits per second.
Definition: avcodec.h:1303
int max_b_frames
maximum number of B-frames between non-B-frames Note: The output will be delayed by max_b_frames+1 re...
Definition: avcodec.h:1935
NV_ENC_MAP_INPUT_RESOURCE in_map
Definition: nvenc.h:55
int CUdevice
Definition: nvenc.h:44
enum AVColorRange color_range
MPEG vs JPEG YUV range.
Definition: avcodec.h:2385
int num
numerator
Definition: rational.h:44
PCUCTXDESTROY cu_ctx_destroy
Definition: nvenc.h:91
static av_cold int nvenc_setup_surfaces(AVCodecContext *avctx)
Definition: nvenc.c:1015
NV_ENCODE_API_FUNCTION_LIST nvenc_funcs
Definition: nvenc.h:93
NvencDynLoadFunctions nvenc_dload_funcs
Definition: nvenc.h:135
AVRational sample_aspect_ratio
sample aspect ratio (0 if unknown) That is the width of a pixel divided by the height of the pixel...
Definition: avcodec.h:2060
int width
The allocated dimensions of the frames in this pool.
Definition: hwcontext.h:221
int first_packet_output
Definition: nvenc.h:163
PCUDEVICEGETNAME cu_device_get_name
Definition: nvenc.h:87
enum AVPixelFormat pix_fmt
Pixel format, see AV_PIX_FMT_xxx.
Definition: avcodec.h:1877
static AVPacket pkt
int profile
profile
Definition: avcodec.h:3153
int preset
Definition: nvenc.h:167
float i_quant_offset
qscale offset between P and I-frames
Definition: avcodec.h:1993
static void nvenc_override_rate_control(AVCodecContext *avctx)
Definition: nvenc.c:582
static NvencSurface * get_free_frame(NvencContext *ctx)
Definition: nvenc.c:1195
static av_cold void nvenc_setup_rate_control(AVCodecContext *avctx)
Definition: nvenc.c:625
int av_fifo_generic_write(AVFifoBuffer *f, void *src, int size, int(*func)(void *, void *, int))
Feed data from a user-supplied callback to an AVFifoBuffer.
Definition: fifo.c:122
int nvenc_device_count
Definition: nvenc.h:94
#define FF_PROFILE_H264_HIGH_444_PREDICTIVE
Definition: avcodec.h:3195
NV_ENC_INPUT_PTR input_surface
Definition: nvenc.h:53
AVRational time_base
This is the fundamental unit of time (in seconds) in terms of which frame timestamps are represented...
Definition: avcodec.h:1786
NVENCSTATUS nverr
Definition: nvenc.c:87
#define av_assert0(cond)
assert() equivalent, that is always enabled.
Definition: avassert.h:37
CUcontext cu_context
Definition: nvenc.h:139
#define FF_PROFILE_H264_BASELINE
Definition: avcodec.h:3185
uint8_t
#define av_cold
Definition: attributes.h:82
#define av_malloc(s)
AVFrame * av_frame_alloc(void)
Allocate an AVFrame and set its fields to default values.
Definition: frame.c:140
float delta
PCUDEVICEGETCOUNT cu_device_get_count
Definition: nvenc.h:85
enum AVPixelFormat ff_nvenc_pix_fmts[]
Definition: nvenc.c:76
float b_quant_factor
qscale factor between IP and B-frames If > 0 then the last P-frame quantizer will be used (q= lastp_q...
Definition: avcodec.h:1944
#define FF_PROFILE_HEVC_MAIN
Definition: avcodec.h:3232
PCUDEVICECOMPUTECAPABILITY cu_device_compute_capability
Definition: nvenc.h:88
int av_frame_ref(AVFrame *dst, const AVFrame *src)
Set up a new reference to the data described by the source frame.
Definition: frame.c:374
int64_t pts
Presentation timestamp in time_base units (time when frame should be shown to user).
Definition: frame.h:268
uint8_t * extradata
some codecs need / can use extradata like Huffman tables.
Definition: avcodec.h:1764
#define LOAD_LIBRARY(l, path)
Definition: nvenc.c:56
NV_ENC_INITIALIZE_PARAMS init_encode_params
Definition: nvenc.h:137
static AVFrame * frame
void * hwctx
The format-specific data, allocated and freed by libavutil along with this context.
Definition: hwcontext.h:84
static av_cold int nvenc_alloc_surface(AVCodecContext *avctx, int idx)
Definition: nvenc.c:942
#define height
uint8_t * data
Definition: avcodec.h:1580
static av_cold int nvenc_setup_extradata(AVCodecContext *avctx)
Definition: nvenc.c:1047
planar YUV 4:2:2, 16bpp, full scale (JPEG), deprecated in favor of AV_PIX_FMT_YUV422P and setting col...
Definition: pixfmt.h:75
#define AV_LOG_VERBOSE
Detailed information.
Definition: log.h:192
static int nvenc_check_capabilities(AVCodecContext *avctx)
Definition: nvenc.c:264
static int nvenc_register_frame(AVCodecContext *avctx, const AVFrame *frame)
Definition: nvenc.c:1292
int buffer_size
The size of the buffer to which the ratecontrol is applied, in bits.
Definition: avcodec.h:1319
AVFifoBuffer * output_surface_ready_queue
Definition: nvenc.h:146
#define av_log(a,...)
CUdeviceptr ptr
Definition: nvenc.h:150
CUcontext cu_context_internal
Definition: nvenc.h:140
#define AV_PKT_FLAG_KEY
The packet contains a keyframe.
Definition: avcodec.h:1612
int async_depth
Definition: nvenc.h:176
enum AVCodecID id
Definition: avcodec.h:3556
static av_cold int nvenc_open_session(AVCodecContext *avctx)
Definition: nvenc.c:190
#define AV_LOG_ERROR
Something went wrong and cannot losslessly be recovered.
Definition: log.h:176
int has_b_frames
Size of the frame reordering buffer in the decoder.
Definition: avcodec.h:1971
static void timestamp_queue_enqueue(AVFifoBuffer *queue, int64_t timestamp)
Definition: nvenc.c:1406
int rc
Definition: nvenc.h:171
NVENCSTATUS(NVENCAPI * PNVENCODEAPICREATEINSTANCE)(NV_ENCODE_API_FUNCTION_LIST *functionList)
Definition: nvenc.h:75
#define LOAD_SYMBOL(fun, lib, symbol)
Definition: nvenc.c:66
#define AVERROR(e)
Definition: error.h:43
int nb_registered_frames
Definition: nvenc.h:154
void av_frame_free(AVFrame **frame)
Free the frame and any dynamically allocated objects in it, e.g.
Definition: frame.c:153
int qmax
maximum quantizer
Definition: avcodec.h:2598
PCUDEVICEGET cu_device_get
Definition: nvenc.h:86
static int nvenc_map_error(NVENCSTATUS err, const char **desc)
Definition: nvenc.c:119
int av_fifo_generic_read(AVFifoBuffer *f, void *dest, int buf_size, void(*func)(void *, void *, int))
Feed data from an AVFifoBuffer to a user-supplied callback.
Definition: fifo.c:213
#define FF_PROFILE_H264_HIGH
Definition: avcodec.h:3189
int flags
AV_CODEC_FLAG_*.
Definition: avcodec.h:1744
GLenum GLint * params
Definition: opengl_enc.c:114
planar YUV 4:2:0, 12bpp, 1 plane for Y and 1 plane for the UV components, which are interleaved (firs...
Definition: pixfmt.h:90
simple assert() macros that are a bit more flexible than ISO C assert().
float i_quant_factor
qscale factor between P- and I-frames If > 0 then the last P-frame quantizer will be used (q = lastp_...
Definition: avcodec.h:1986
static av_cold int nvenc_setup_hevc_config(AVCodecContext *avctx)
Definition: nvenc.c:762
GLsizei count
Definition: opengl_enc.c:109
#define FFMAX(a, b)
Definition: common.h:94
static av_cold void set_constqp(AVCodecContext *avctx)
Definition: nvenc.c:509
#define fail()
Definition: checkasm.h:81
int level
Definition: nvenc.h:169
#define NVENC_LIBNAME
Definition: nvenc.c:41
int flags
A combination of AV_PKT_FLAG values.
Definition: avcodec.h:1586
int rc_buffer_size
decoder bitstream buffer size
Definition: avcodec.h:2625
int av_reduce(int *dst_num, int *dst_den, int64_t num, int64_t den, int64_t max)
Reduce a fraction.
Definition: rational.c:35
static int nvenc_check_codec_support(AVCodecContext *avctx)
Definition: nvenc.c:211
PCUINIT cu_init
Definition: nvenc.h:84
int refs
number of reference frames
Definition: avcodec.h:2329
int flags
Definition: nvenc.h:175
enum AVPictureType pict_type
Picture type of the frame.
Definition: frame.h:258
static int nvenc_upload_frame(AVCodecContext *avctx, const AVFrame *frame, NvencSurface *nvenc_frame)
Definition: nvenc.c:1330
NV_ENC_REGISTERED_PTR regptr
Definition: nvenc.h:151
#define FFMIN(a, b)
Definition: common.h:96
static av_cold int nvenc_setup_encoder(AVCodecContext *avctx)
Definition: nvenc.c:827
AVHWDeviceContext * device_ctx
The parent AVHWDeviceContext.
Definition: hwcontext.h:141
int lockCount
Definition: nvenc.h:63
planar YUV 4:2:0, 12bpp, full scale (JPEG), deprecated in favor of AV_PIX_FMT_YUV420P and setting col...
Definition: pixfmt.h:74
av_cold int ff_nvenc_encode_init(AVCodecContext *avctx)
Definition: nvenc.c:1157
#define width
static int process_output_surface(AVCodecContext *avctx, AVPacket *pkt, NvencSurface *tmpoutsurf)
Definition: nvenc.c:1454
int ff_nvenc_encode_frame(AVCodecContext *avctx, AVPacket *pkt, const AVFrame *frame, int *got_packet)
Definition: nvenc.c:1583
int width
picture width / height.
Definition: avcodec.h:1836
AVBufferRef * hw_frames_ctx
Encoding only.
Definition: avcodec.h:3495
#define NVENC_CAP
Definition: nvenc.c:51
AVFormatContext * ctx
Definition: movenc.c:48
enum AVColorPrimaries color_primaries
Chromaticity coordinates of the source primaries.
Definition: avcodec.h:2364
CUresult
Definition: nvenc.h:41
int dummy
Definition: motion.c:64
int ticks_per_frame
For some codecs, the time base is closer to the field rate than the frame rate.
Definition: avcodec.h:1795
int profile
Definition: nvenc.h:168
AVFifoBuffer * output_surface_queue
Definition: nvenc.h:145
#define AVERROR_EXIT
Immediate exit was requested; the called function should not be restarted.
Definition: error.h:56
HW acceleration through CUDA.
Definition: pixfmt.h:248
static av_cold int nvenc_setup_h264_config(AVCodecContext *avctx)
Definition: nvenc.c:678
#define FF_ARRAY_ELEMS(a)
the normal 2^n-1 "JPEG" YUV ranges
Definition: pixfmt.h:457
static int nvenc_print_error(void *log_ctx, NVENCSTATUS err, const char *error_string)
Definition: nvenc.c:134
enum AVPixelFormat data_pix_fmt
Definition: nvenc.h:158
int format
format of the frame, -1 if unknown or unset Values correspond to enum AVPixelFormat for video frames...
Definition: frame.h:248
This structure describes the bitrate properties of an encoded bitstream.
Definition: avcodec.h:1298
PCUCTXCREATE cu_ctx_create
Definition: nvenc.h:89
#define AV_LOG_INFO
Standard information.
Definition: log.h:187
NV_ENC_CONFIG encode_config
Definition: nvenc.h:138
int av_fifo_size(const AVFifoBuffer *f)
Return the amount of data in bytes in the AVFifoBuffer, that is the amount of data you can read from ...
Definition: fifo.c:77
int linesize[AV_NUM_DATA_POINTERS]
For video, size in bytes of each picture line.
Definition: frame.h:215
int64_t initial_pts[2]
Definition: nvenc.h:162
main external API structure.
Definition: avcodec.h:1649
uint8_t * data
The data buffer.
Definition: buffer.h:89
int qmin
minimum quantizer
Definition: avcodec.h:2591
void * buf
Definition: avisynth_c.h:553
int extradata_size
Definition: avcodec.h:1765
#define AVERROR_BUG
Internal bug, also see AVERROR_BUG2.
Definition: error.h:50
void * CUdeviceptr
Definition: nvenc.h:46
This struct is allocated as AVHWDeviceContext.hwctx.
static int nvenc_check_cap(AVCodecContext *avctx, NV_ENC_CAPS cap)
Definition: nvenc.c:247
enum AVColorSpace colorspace
YUV colorspace type.
Definition: avcodec.h:2378
enum AVColorTransferCharacteristic color_trc
Color Transfer Characteristic.
Definition: avcodec.h:2371
struct NvencContext::@85 registered_frames[MAX_REGISTERED_FRAMES]
int width
Definition: nvenc.h:57
This struct describes a set or pool of "hardware" frames (i.e.
Definition: hwcontext.h:116
static void nvenc_codec_specific_pic_params(AVCodecContext *avctx, NV_ENC_PIC_PARAMS *params)
Definition: nvenc.c:1385
#define IS_CBR(rc)
Definition: nvenc.c:52
AVPictureType
Definition: avutil.h:264
int flags
Definition: nvenc.c:483
float b_quant_offset
qscale offset between IP and B-frames
Definition: avcodec.h:1963
int ff_alloc_packet2(AVCodecContext *avctx, AVPacket *avpkt, int64_t size, int64_t min_size)
Check AVPacket size and/or allocate data.
Definition: utils.c:1690
int cbr
Definition: nvenc.h:172
static int nvenc_find_free_reg_resource(AVCodecContext *avctx)
Definition: nvenc.c:1265
int averr
Definition: nvenc.c:88
void av_frame_unref(AVFrame *frame)
Unreference all the buffers referenced by frame and reset the frame fields.
Definition: frame.c:484
int global_quality
Global quality for codecs which cannot change it per frame.
Definition: avcodec.h:1730
static int flags
Definition: cpu.c:47
uint8_t * data[AV_NUM_DATA_POINTERS]
pointer to the picture/channel planes.
Definition: frame.h:198
int reg_idx
Definition: nvenc.h:56
#define AV_CODEC_FLAG_GLOBAL_HEADER
Place global headers in extradata instead of every keyframe.
Definition: avcodec.h:879
int gop_size
the number of pictures in a group of pictures, or 0 for intra_only
Definition: avcodec.h:1862
static int64_t timestamp_queue_dequeue(AVFifoBuffer *queue)
Definition: nvenc.c:1411
planar YUV 4:2:0, 12bpp, (1 Cr & Cb sample per 2x2 Y samples)
Definition: pixfmt.h:62
#define FF_DISABLE_DEPRECATION_WARNINGS
Definition: internal.h:80
common internal api header.
PCUCTXPOPCURRENT cu_ctx_pop_current
Definition: nvenc.h:90
static int output_ready(AVCodecContext *avctx, int flush)
Definition: nvenc.c:1565
Bi-dir predicted.
Definition: avutil.h:268
planar YUV 4:4:4, 24bpp, full scale (JPEG), deprecated in favor of AV_PIX_FMT_YUV444P and setting col...
Definition: pixfmt.h:76
attribute_deprecated AVFrame * coded_frame
the picture in the bitstream
Definition: avcodec.h:3070
int size
Definition: nvenc.h:62
NvencSurface * surfaces
Definition: nvenc.h:143
int den
denominator
Definition: rational.h:45
#define AVERROR_UNKNOWN
Unknown error, typically from an external library.
Definition: error.h:71
AVCPBProperties * ff_add_cpb_side_data(AVCodecContext *avctx)
Add a CPB properties side data to an encoding context.
Definition: utils.c:3992
#define AV_INPUT_BUFFER_PADDING_SIZE
Required number of additionally allocated bytes at the end of the input bitstream for decoding...
Definition: avcodec.h:731
void * priv_data
Definition: avcodec.h:1691
static int nvenc_set_timestamp(AVCodecContext *avctx, NV_ENC_LOCK_BITSTREAM *params, AVPacket *pkt)
Definition: nvenc.c:1420
#define av_free(p)
#define FF_ENABLE_DEPRECATION_WARNINGS
Definition: internal.h:81
AVFifoBuffer * av_fifo_alloc(unsigned int size)
Initialize an AVFifoBuffer.
Definition: fifo.c:43
int top_field_first
If the content is interlaced, is top field displayed first.
Definition: frame.h:323
static av_cold int nvenc_load_libraries(AVCodecContext *avctx)
Definition: nvenc.c:144
static av_cold int nvenc_check_device(AVCodecContext *avctx, int idx)
Definition: nvenc.c:320
int avg_bitrate
Average bitrate of the stream, in bits per second.
Definition: avcodec.h:1313
#define CUDA_LIBNAME
Definition: nvenc.c:40
int device
Definition: nvenc.h:174
#define FF_QP2LAMBDA
factor to convert from H.263 QP to lambda
Definition: avutil.h:219
int64_t dts
Decompression timestamp in AVStream->time_base units; the time at which the packet is decompressed...
Definition: avcodec.h:1579
static void * av_mallocz_array(size_t nmemb, size_t size)
Definition: mem.h:229
int nb_surfaces
Definition: nvenc.h:142
#define av_freep(p)
#define AV_CODEC_ID_H265
Definition: avcodec.h:370
static av_cold void set_lossless(AVCodecContext *avctx)
Definition: nvenc.c:568
void av_fifo_freep(AVFifoBuffer **f)
Free an AVFifoBuffer and reset pointer to NULL.
Definition: fifo.c:63
#define AV_LOG_FATAL
Something went wrong and recovery is not possible.
Definition: log.h:170
void av_image_copy_plane(uint8_t *dst, int dst_linesize, const uint8_t *src, int src_linesize, int bytewidth, int height)
Copy image plane from src to dst.
Definition: imgutils.c:287
int tier
Definition: nvenc.h:170
void * CUcontext
Definition: nvenc.h:45
static const struct @80 nvenc_errors[]
enum AVPixelFormat sw_format
The pixel format identifying the actual data layout of the hardware frames.
Definition: hwcontext.h:214
#define AVERROR_EXTERNAL
Generic error in an external library.
Definition: error.h:57
AVPixelFormat
Pixel format.
Definition: pixfmt.h:60
This structure stores compressed data.
Definition: avcodec.h:1557
void * av_mallocz(size_t size)
Allocate a block of size bytes with alignment suitable for all memory accesses (including vectors if ...
Definition: mem.c:252
static int nvenc_copy_frame(AVCodecContext *avctx, NvencSurface *inSurf, NV_ENC_LOCK_INPUT_BUFFER *lockBufferParams, const AVFrame *frame)
Definition: nvenc.c:1209
NV_ENC_OUTPUT_PTR output_surface
Definition: nvenc.h:60
int64_t pts
Presentation timestamp in AVStream->time_base units; the time at which the decompressed packet will b...
Definition: avcodec.h:1573
for(j=16;j >0;--j)
#define AV_NOPTS_VALUE
Undefined timestamp value.
Definition: avutil.h:240
Predicted.
Definition: avutil.h:267
static void nvenc_map_preset(NvencContext *ctx)
Definition: nvenc.c:486
int64_t rc_max_rate
maximum bitrate
Definition: avcodec.h:2648
const char * name
Definition: opengl_enc.c:103