2 * H.264 hardware encoding using nvidia nvenc
3 * Copyright (c) 2014 Timo Rothenpieler <timo@rothenpieler.org>
5 * This file is part of FFmpeg.
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.
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.
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
28 #include <nvEncodeAPI.h>
30 #include "libavutil/internal.h"
31 #include "libavutil/imgutils.h"
32 #include "libavutil/avassert.h"
33 #include "libavutil/opt.h"
34 #include "libavutil/mem.h"
40 #define CUDAAPI __stdcall
46 #define LOAD_FUNC(l, s) GetProcAddress(l, s)
47 #define DL_CLOSE_FUNC(l) FreeLibrary(l)
49 #define LOAD_FUNC(l, s) dlsym(l, s)
50 #define DL_CLOSE_FUNC(l) dlclose(l)
53 typedef enum cudaError_enum {
57 typedef void* CUcontext;
59 typedef CUresult(CUDAAPI *PCUINIT)(unsigned int Flags);
60 typedef CUresult(CUDAAPI *PCUDEVICEGETCOUNT)(int *count);
61 typedef CUresult(CUDAAPI *PCUDEVICEGET)(CUdevice *device, int ordinal);
62 typedef CUresult(CUDAAPI *PCUDEVICEGETNAME)(char *name, int len, CUdevice dev);
63 typedef CUresult(CUDAAPI *PCUDEVICECOMPUTECAPABILITY)(int *major, int *minor, CUdevice dev);
64 typedef CUresult(CUDAAPI *PCUCTXCREATE)(CUcontext *pctx, unsigned int flags, CUdevice dev);
65 typedef CUresult(CUDAAPI *PCUCTXPOPCURRENT)(CUcontext *pctx);
66 typedef CUresult(CUDAAPI *PCUCTXDESTROY)(CUcontext ctx);
68 typedef NVENCSTATUS (NVENCAPI* PNVENCODEAPICREATEINSTANCE)(NV_ENCODE_API_FUNCTION_LIST *functionList);
70 typedef struct NvencInputSurface
72 NV_ENC_INPUT_PTR input_surface;
78 NV_ENC_BUFFER_FORMAT format;
81 typedef struct NvencOutputSurface
83 NV_ENC_OUTPUT_PTR output_surface;
86 NvencInputSurface* input_surface;
91 typedef struct NvencData
95 NvencOutputSurface *surface;
99 typedef struct NvencDataList
108 typedef struct NvencDynLoadFunctions
111 PCUDEVICEGETCOUNT cu_device_get_count;
112 PCUDEVICEGET cu_device_get;
113 PCUDEVICEGETNAME cu_device_get_name;
114 PCUDEVICECOMPUTECAPABILITY cu_device_compute_capability;
115 PCUCTXCREATE cu_ctx_create;
116 PCUCTXPOPCURRENT cu_ctx_pop_current;
117 PCUCTXDESTROY cu_ctx_destroy;
119 NV_ENCODE_API_FUNCTION_LIST nvenc_funcs;
120 int nvenc_device_count;
121 CUdevice nvenc_devices[16];
130 } NvencDynLoadFunctions;
132 typedef struct NvencValuePair
138 typedef struct NvencContext
142 NvencDynLoadFunctions nvenc_dload_funcs;
144 NV_ENC_INITIALIZE_PARAMS init_encode_params;
145 NV_ENC_CONFIG encode_config;
146 CUcontext cu_context;
148 int max_surface_count;
149 NvencInputSurface *input_surfaces;
150 NvencOutputSurface *output_surfaces;
152 NvencDataList output_surface_queue;
153 NvencDataList output_surface_ready_queue;
154 NvencDataList timestamp_list;
169 static const NvencValuePair nvenc_h264_level_pairs[] = {
170 { "auto", NV_ENC_LEVEL_AUTOSELECT },
171 { "1" , NV_ENC_LEVEL_H264_1 },
172 { "1.0" , NV_ENC_LEVEL_H264_1 },
173 { "1b" , NV_ENC_LEVEL_H264_1b },
174 { "1.0b", NV_ENC_LEVEL_H264_1b },
175 { "1.1" , NV_ENC_LEVEL_H264_11 },
176 { "1.2" , NV_ENC_LEVEL_H264_12 },
177 { "1.3" , NV_ENC_LEVEL_H264_13 },
178 { "2" , NV_ENC_LEVEL_H264_2 },
179 { "2.0" , NV_ENC_LEVEL_H264_2 },
180 { "2.1" , NV_ENC_LEVEL_H264_21 },
181 { "2.2" , NV_ENC_LEVEL_H264_22 },
182 { "3" , NV_ENC_LEVEL_H264_3 },
183 { "3.0" , NV_ENC_LEVEL_H264_3 },
184 { "3.1" , NV_ENC_LEVEL_H264_31 },
185 { "3.2" , NV_ENC_LEVEL_H264_32 },
186 { "4" , NV_ENC_LEVEL_H264_4 },
187 { "4.0" , NV_ENC_LEVEL_H264_4 },
188 { "4.1" , NV_ENC_LEVEL_H264_41 },
189 { "4.2" , NV_ENC_LEVEL_H264_42 },
190 { "5" , NV_ENC_LEVEL_H264_5 },
191 { "5.0" , NV_ENC_LEVEL_H264_5 },
192 { "5.1" , NV_ENC_LEVEL_H264_51 },
196 static const NvencValuePair nvenc_hevc_level_pairs[] = {
197 { "auto", NV_ENC_LEVEL_AUTOSELECT },
198 { "1" , NV_ENC_LEVEL_HEVC_1 },
199 { "1.0" , NV_ENC_LEVEL_HEVC_1 },
200 { "2" , NV_ENC_LEVEL_HEVC_2 },
201 { "2.0" , NV_ENC_LEVEL_HEVC_2 },
202 { "2.1" , NV_ENC_LEVEL_HEVC_21 },
203 { "3" , NV_ENC_LEVEL_HEVC_3 },
204 { "3.0" , NV_ENC_LEVEL_HEVC_3 },
205 { "3.1" , NV_ENC_LEVEL_HEVC_31 },
206 { "4" , NV_ENC_LEVEL_HEVC_4 },
207 { "4.0" , NV_ENC_LEVEL_HEVC_4 },
208 { "4.1" , NV_ENC_LEVEL_HEVC_41 },
209 { "5" , NV_ENC_LEVEL_HEVC_5 },
210 { "5.0" , NV_ENC_LEVEL_HEVC_5 },
211 { "5.1" , NV_ENC_LEVEL_HEVC_51 },
212 { "5.2" , NV_ENC_LEVEL_HEVC_52 },
213 { "6" , NV_ENC_LEVEL_HEVC_6 },
214 { "6.0" , NV_ENC_LEVEL_HEVC_6 },
215 { "6.1" , NV_ENC_LEVEL_HEVC_61 },
216 { "6.2" , NV_ENC_LEVEL_HEVC_62 },
220 static int input_string_to_uint32(AVCodecContext *avctx, const NvencValuePair *pair, const char *input, uint32_t *output)
222 for (; pair->str; ++pair) {
223 if (!strcmp(input, pair->str)) {
229 return AVERROR(EINVAL);
232 static NvencData* data_queue_dequeue(NvencDataList* queue)
238 av_assert0(queue->size);
239 av_assert0(queue->data);
244 /* Size always is a multiple of two */
245 mask = queue->size - 1;
246 read_pos = (queue->pos - queue->count) & mask;
249 return &queue->data[read_pos];
252 static int data_queue_enqueue(NvencDataList* queue, NvencData *data)
254 NvencDataList new_queue;
259 /* size always has to be a multiple of two */
264 queue->data = av_malloc(queue->size * sizeof(*(queue->data)));
268 return AVERROR(ENOMEM);
272 if (queue->count == queue->size) {
273 new_queue.size = queue->size << 1;
276 new_queue.data = av_malloc(new_queue.size * sizeof(*(queue->data)));
279 return AVERROR(ENOMEM);
281 while (tmp_data = data_queue_dequeue(queue))
282 data_queue_enqueue(&new_queue, tmp_data);
284 av_free(queue->data);
288 mask = queue->size - 1;
290 queue->data[queue->pos] = *data;
291 queue->pos = (queue->pos + 1) & mask;
297 static int out_surf_queue_enqueue(NvencDataList* queue, NvencOutputSurface* surface)
300 data.u.surface = surface;
302 return data_queue_enqueue(queue, &data);
305 static NvencOutputSurface* out_surf_queue_dequeue(NvencDataList* queue)
307 NvencData* res = data_queue_dequeue(queue);
312 return res->u.surface;
315 static int timestamp_queue_enqueue(NvencDataList* queue, int64_t timestamp)
318 data.u.timestamp = timestamp;
320 return data_queue_enqueue(queue, &data);
323 static int64_t timestamp_queue_dequeue(NvencDataList* queue)
325 NvencData* res = data_queue_dequeue(queue);
328 return AV_NOPTS_VALUE;
330 return res->u.timestamp;
333 #define CHECK_LOAD_FUNC(t, f, s) \
335 (f) = (t)LOAD_FUNC(dl_fn->cuda_lib, s); \
337 av_log(avctx, AV_LOG_FATAL, "Failed loading %s from CUDA library\n", s); \
342 static av_cold int nvenc_dyload_cuda(AVCodecContext *avctx)
344 NvencContext *ctx = avctx->priv_data;
345 NvencDynLoadFunctions *dl_fn = &ctx->nvenc_dload_funcs;
351 dl_fn->cuda_lib = LoadLibrary(TEXT("nvcuda.dll"));
353 dl_fn->cuda_lib = dlopen("libcuda.so", RTLD_LAZY);
356 if (!dl_fn->cuda_lib) {
357 av_log(avctx, AV_LOG_FATAL, "Failed loading CUDA library\n");
361 CHECK_LOAD_FUNC(PCUINIT, dl_fn->cu_init, "cuInit");
362 CHECK_LOAD_FUNC(PCUDEVICEGETCOUNT, dl_fn->cu_device_get_count, "cuDeviceGetCount");
363 CHECK_LOAD_FUNC(PCUDEVICEGET, dl_fn->cu_device_get, "cuDeviceGet");
364 CHECK_LOAD_FUNC(PCUDEVICEGETNAME, dl_fn->cu_device_get_name, "cuDeviceGetName");
365 CHECK_LOAD_FUNC(PCUDEVICECOMPUTECAPABILITY, dl_fn->cu_device_compute_capability, "cuDeviceComputeCapability");
366 CHECK_LOAD_FUNC(PCUCTXCREATE, dl_fn->cu_ctx_create, "cuCtxCreate_v2");
367 CHECK_LOAD_FUNC(PCUCTXPOPCURRENT, dl_fn->cu_ctx_pop_current, "cuCtxPopCurrent_v2");
368 CHECK_LOAD_FUNC(PCUCTXDESTROY, dl_fn->cu_ctx_destroy, "cuCtxDestroy_v2");
375 DL_CLOSE_FUNC(dl_fn->cuda_lib);
377 dl_fn->cuda_lib = NULL;
382 static av_cold int check_cuda_errors(AVCodecContext *avctx, CUresult err, const char *func)
384 if (err != CUDA_SUCCESS) {
385 av_log(avctx, AV_LOG_FATAL, ">> %s - failed with error code 0x%x\n", func, err);
390 #define check_cuda_errors(f) if (!check_cuda_errors(avctx, f, #f)) goto error
392 static av_cold int nvenc_check_cuda(AVCodecContext *avctx)
394 int device_count = 0;
395 CUdevice cu_device = 0;
397 int smminor = 0, smmajor = 0;
398 int i, smver, target_smver;
400 NvencContext *ctx = avctx->priv_data;
401 NvencDynLoadFunctions *dl_fn = &ctx->nvenc_dload_funcs;
403 switch (avctx->codec->id) {
404 case AV_CODEC_ID_H264:
405 target_smver = avctx->pix_fmt == AV_PIX_FMT_YUV444P ? 0x52 : 0x30;
407 case AV_CODEC_ID_H265:
411 av_log(avctx, AV_LOG_FATAL, "Unknown codec name\n");
415 if (!nvenc_dyload_cuda(avctx))
418 if (dl_fn->nvenc_device_count > 0)
421 check_cuda_errors(dl_fn->cu_init(0));
423 check_cuda_errors(dl_fn->cu_device_get_count(&device_count));
426 av_log(avctx, AV_LOG_FATAL, "No CUDA capable devices found\n");
430 av_log(avctx, AV_LOG_VERBOSE, "%d CUDA capable devices found\n", device_count);
432 dl_fn->nvenc_device_count = 0;
434 for (i = 0; i < device_count; ++i) {
435 check_cuda_errors(dl_fn->cu_device_get(&cu_device, i));
436 check_cuda_errors(dl_fn->cu_device_get_name(gpu_name, sizeof(gpu_name), cu_device));
437 check_cuda_errors(dl_fn->cu_device_compute_capability(&smmajor, &smminor, cu_device));
439 smver = (smmajor << 4) | smminor;
441 av_log(avctx, AV_LOG_VERBOSE, "[ GPU #%d - < %s > has Compute SM %d.%d, NVENC %s ]\n", i, gpu_name, smmajor, smminor, (smver >= target_smver) ? "Available" : "Not Available");
443 if (smver >= target_smver)
444 dl_fn->nvenc_devices[dl_fn->nvenc_device_count++] = cu_device;
447 if (!dl_fn->nvenc_device_count) {
448 av_log(avctx, AV_LOG_FATAL, "No NVENC capable devices found\n");
456 dl_fn->nvenc_device_count = 0;
461 static av_cold int nvenc_dyload_nvenc(AVCodecContext *avctx)
463 PNVENCODEAPICREATEINSTANCE nvEncodeAPICreateInstance = 0;
464 NVENCSTATUS nvstatus;
466 NvencContext *ctx = avctx->priv_data;
467 NvencDynLoadFunctions *dl_fn = &ctx->nvenc_dload_funcs;
469 if (!nvenc_check_cuda(avctx))
472 if (dl_fn->nvenc_lib)
476 if (sizeof(void*) == 8) {
477 dl_fn->nvenc_lib = LoadLibrary(TEXT("nvEncodeAPI64.dll"));
479 dl_fn->nvenc_lib = LoadLibrary(TEXT("nvEncodeAPI.dll"));
482 dl_fn->nvenc_lib = dlopen("libnvidia-encode.so.1", RTLD_LAZY);
485 if (!dl_fn->nvenc_lib) {
486 av_log(avctx, AV_LOG_FATAL, "Failed loading the nvenc library\n");
490 nvEncodeAPICreateInstance = (PNVENCODEAPICREATEINSTANCE)LOAD_FUNC(dl_fn->nvenc_lib, "NvEncodeAPICreateInstance");
492 if (!nvEncodeAPICreateInstance) {
493 av_log(avctx, AV_LOG_FATAL, "Failed to load nvenc entrypoint\n");
497 dl_fn->nvenc_funcs.version = NV_ENCODE_API_FUNCTION_LIST_VER;
499 nvstatus = nvEncodeAPICreateInstance(&dl_fn->nvenc_funcs);
501 if (nvstatus != NV_ENC_SUCCESS) {
502 av_log(avctx, AV_LOG_FATAL, "Failed to create nvenc instance\n");
506 av_log(avctx, AV_LOG_VERBOSE, "Nvenc initialized successfully\n");
511 if (dl_fn->nvenc_lib)
512 DL_CLOSE_FUNC(dl_fn->nvenc_lib);
514 dl_fn->nvenc_lib = NULL;
519 static av_cold void nvenc_unload_nvenc(AVCodecContext *avctx)
521 NvencContext *ctx = avctx->priv_data;
522 NvencDynLoadFunctions *dl_fn = &ctx->nvenc_dload_funcs;
524 DL_CLOSE_FUNC(dl_fn->nvenc_lib);
525 dl_fn->nvenc_lib = NULL;
527 dl_fn->nvenc_device_count = 0;
529 DL_CLOSE_FUNC(dl_fn->cuda_lib);
530 dl_fn->cuda_lib = NULL;
532 dl_fn->cu_init = NULL;
533 dl_fn->cu_device_get_count = NULL;
534 dl_fn->cu_device_get = NULL;
535 dl_fn->cu_device_get_name = NULL;
536 dl_fn->cu_device_compute_capability = NULL;
537 dl_fn->cu_ctx_create = NULL;
538 dl_fn->cu_ctx_pop_current = NULL;
539 dl_fn->cu_ctx_destroy = NULL;
541 av_log(avctx, AV_LOG_VERBOSE, "Nvenc unloaded\n");
544 static av_cold int nvenc_encode_init(AVCodecContext *avctx)
546 NV_ENC_OPEN_ENCODE_SESSION_EX_PARAMS encode_session_params = { 0 };
547 NV_ENC_PRESET_CONFIG preset_config = { 0 };
548 CUcontext cu_context_curr;
550 GUID encoder_preset = NV_ENC_PRESET_HQ_GUID;
552 NVENCSTATUS nv_status = NV_ENC_SUCCESS;
553 AVCPBProperties *cpb_props;
554 int surfaceCount = 0;
562 NvencContext *ctx = avctx->priv_data;
563 NvencDynLoadFunctions *dl_fn = &ctx->nvenc_dload_funcs;
564 NV_ENCODE_API_FUNCTION_LIST *p_nvenc = &dl_fn->nvenc_funcs;
566 if (!nvenc_dyload_nvenc(avctx))
567 return AVERROR_EXTERNAL;
569 ctx->last_dts = AV_NOPTS_VALUE;
571 ctx->encode_config.version = NV_ENC_CONFIG_VER;
572 ctx->init_encode_params.version = NV_ENC_INITIALIZE_PARAMS_VER;
573 preset_config.version = NV_ENC_PRESET_CONFIG_VER;
574 preset_config.presetCfg.version = NV_ENC_CONFIG_VER;
575 encode_session_params.version = NV_ENC_OPEN_ENCODE_SESSION_EX_PARAMS_VER;
576 encode_session_params.apiVersion = NVENCAPI_VERSION;
578 if (ctx->gpu >= dl_fn->nvenc_device_count) {
579 av_log(avctx, AV_LOG_FATAL, "Requested GPU %d, but only %d GPUs are available!\n", ctx->gpu, dl_fn->nvenc_device_count);
580 res = AVERROR(EINVAL);
584 ctx->cu_context = NULL;
585 cu_res = dl_fn->cu_ctx_create(&ctx->cu_context, 4, dl_fn->nvenc_devices[ctx->gpu]); // CU_CTX_SCHED_BLOCKING_SYNC=4, avoid CPU spins
587 if (cu_res != CUDA_SUCCESS) {
588 av_log(avctx, AV_LOG_FATAL, "Failed creating CUDA context for NVENC: 0x%x\n", (int)cu_res);
589 res = AVERROR_EXTERNAL;
593 cu_res = dl_fn->cu_ctx_pop_current(&cu_context_curr);
595 if (cu_res != CUDA_SUCCESS) {
596 av_log(avctx, AV_LOG_FATAL, "Failed popping CUDA context: 0x%x\n", (int)cu_res);
597 res = AVERROR_EXTERNAL;
601 encode_session_params.device = ctx->cu_context;
602 encode_session_params.deviceType = NV_ENC_DEVICE_TYPE_CUDA;
604 nv_status = p_nvenc->nvEncOpenEncodeSessionEx(&encode_session_params, &ctx->nvencoder);
605 if (nv_status != NV_ENC_SUCCESS) {
606 ctx->nvencoder = NULL;
607 av_log(avctx, AV_LOG_FATAL, "OpenEncodeSessionEx failed: 0x%x\n", (int)nv_status);
608 res = AVERROR_EXTERNAL;
613 if (!strcmp(ctx->preset, "slow")) {
614 encoder_preset = NV_ENC_PRESET_HQ_GUID;
616 } else if (!strcmp(ctx->preset, "medium")) {
617 encoder_preset = NV_ENC_PRESET_HQ_GUID;
619 } else if (!strcmp(ctx->preset, "fast")) {
620 encoder_preset = NV_ENC_PRESET_HP_GUID;
622 } else if (!strcmp(ctx->preset, "hq")) {
623 encoder_preset = NV_ENC_PRESET_HQ_GUID;
624 } else if (!strcmp(ctx->preset, "hp")) {
625 encoder_preset = NV_ENC_PRESET_HP_GUID;
626 } else if (!strcmp(ctx->preset, "bd")) {
627 encoder_preset = NV_ENC_PRESET_BD_GUID;
628 } else if (!strcmp(ctx->preset, "ll")) {
629 encoder_preset = NV_ENC_PRESET_LOW_LATENCY_DEFAULT_GUID;
631 } else if (!strcmp(ctx->preset, "llhp")) {
632 encoder_preset = NV_ENC_PRESET_LOW_LATENCY_HP_GUID;
634 } else if (!strcmp(ctx->preset, "llhq")) {
635 encoder_preset = NV_ENC_PRESET_LOW_LATENCY_HQ_GUID;
637 } else if (!strcmp(ctx->preset, "lossless")) {
638 encoder_preset = NV_ENC_PRESET_LOSSLESS_DEFAULT_GUID;
640 } else if (!strcmp(ctx->preset, "losslesshp")) {
641 encoder_preset = NV_ENC_PRESET_LOSSLESS_HP_GUID;
643 } else if (!strcmp(ctx->preset, "default")) {
644 encoder_preset = NV_ENC_PRESET_DEFAULT_GUID;
646 av_log(avctx, AV_LOG_FATAL, "Preset \"%s\" is unknown! Supported presets: slow, medium, fast, hp, hq, bd, ll, llhp, llhq, lossless, losslesshp, default\n", ctx->preset);
647 res = AVERROR(EINVAL);
652 if (ctx->twopass < 0) {
656 switch (avctx->codec->id) {
657 case AV_CODEC_ID_H264:
658 codec = NV_ENC_CODEC_H264_GUID;
660 case AV_CODEC_ID_H265:
661 codec = NV_ENC_CODEC_HEVC_GUID;
664 av_log(avctx, AV_LOG_ERROR, "Unknown codec name\n");
665 res = AVERROR(EINVAL);
669 nv_status = p_nvenc->nvEncGetEncodePresetConfig(ctx->nvencoder, codec, encoder_preset, &preset_config);
670 if (nv_status != NV_ENC_SUCCESS) {
671 av_log(avctx, AV_LOG_FATAL, "GetEncodePresetConfig failed: 0x%x\n", (int)nv_status);
672 res = AVERROR_EXTERNAL;
676 ctx->init_encode_params.encodeGUID = codec;
677 ctx->init_encode_params.encodeHeight = avctx->height;
678 ctx->init_encode_params.encodeWidth = avctx->width;
680 if (avctx->sample_aspect_ratio.num && avctx->sample_aspect_ratio.den &&
681 (avctx->sample_aspect_ratio.num != 1 || avctx->sample_aspect_ratio.num != 1)) {
683 avctx->width * avctx->sample_aspect_ratio.num,
684 avctx->height * avctx->sample_aspect_ratio.den,
686 ctx->init_encode_params.darHeight = dh;
687 ctx->init_encode_params.darWidth = dw;
689 ctx->init_encode_params.darHeight = avctx->height;
690 ctx->init_encode_params.darWidth = avctx->width;
693 // De-compensate for hardware, dubiously, trying to compensate for
694 // playback at 704 pixel width.
695 if (avctx->width == 720 &&
696 (avctx->height == 480 || avctx->height == 576)) {
698 ctx->init_encode_params.darWidth * 44,
699 ctx->init_encode_params.darHeight * 45,
701 ctx->init_encode_params.darHeight = dh;
702 ctx->init_encode_params.darWidth = dw;
705 ctx->init_encode_params.frameRateNum = avctx->time_base.den;
706 ctx->init_encode_params.frameRateDen = avctx->time_base.num * avctx->ticks_per_frame;
708 num_mbs = ((avctx->width + 15) >> 4) * ((avctx->height + 15) >> 4);
709 ctx->max_surface_count = (num_mbs >= 8160) ? 32 : 48;
711 if (ctx->buffer_delay >= ctx->max_surface_count)
712 ctx->buffer_delay = ctx->max_surface_count - 1;
714 ctx->init_encode_params.enableEncodeAsync = 0;
715 ctx->init_encode_params.enablePTD = 1;
717 ctx->init_encode_params.presetGUID = encoder_preset;
719 ctx->init_encode_params.encodeConfig = &ctx->encode_config;
720 memcpy(&ctx->encode_config, &preset_config.presetCfg, sizeof(ctx->encode_config));
721 ctx->encode_config.version = NV_ENC_CONFIG_VER;
723 if (avctx->refs >= 0) {
724 /* 0 means "let the hardware decide" */
725 switch (avctx->codec->id) {
726 case AV_CODEC_ID_H264:
727 ctx->encode_config.encodeCodecConfig.h264Config.maxNumRefFrames = avctx->refs;
729 case AV_CODEC_ID_H265:
730 ctx->encode_config.encodeCodecConfig.hevcConfig.maxNumRefFramesInDPB = avctx->refs;
732 /* Earlier switch/case will return if unknown codec is passed. */
736 if (avctx->gop_size > 0) {
737 if (avctx->max_b_frames >= 0) {
738 /* 0 is intra-only, 1 is I/P only, 2 is one B Frame, 3 two B frames, and so on. */
739 ctx->encode_config.frameIntervalP = avctx->max_b_frames + 1;
742 ctx->encode_config.gopLength = avctx->gop_size;
743 switch (avctx->codec->id) {
744 case AV_CODEC_ID_H264:
745 ctx->encode_config.encodeCodecConfig.h264Config.idrPeriod = avctx->gop_size;
747 case AV_CODEC_ID_H265:
748 ctx->encode_config.encodeCodecConfig.hevcConfig.idrPeriod = avctx->gop_size;
750 /* Earlier switch/case will return if unknown codec is passed. */
752 } else if (avctx->gop_size == 0) {
753 ctx->encode_config.frameIntervalP = 0;
754 ctx->encode_config.gopLength = 1;
755 switch (avctx->codec->id) {
756 case AV_CODEC_ID_H264:
757 ctx->encode_config.encodeCodecConfig.h264Config.idrPeriod = 1;
759 case AV_CODEC_ID_H265:
760 ctx->encode_config.encodeCodecConfig.hevcConfig.idrPeriod = 1;
762 /* Earlier switch/case will return if unknown codec is passed. */
766 /* when there're b frames, set dts offset */
767 if (ctx->encode_config.frameIntervalP >= 2)
770 if (avctx->bit_rate > 0) {
771 ctx->encode_config.rcParams.averageBitRate = avctx->bit_rate;
772 } else if (ctx->encode_config.rcParams.averageBitRate > 0) {
773 ctx->encode_config.rcParams.maxBitRate = ctx->encode_config.rcParams.averageBitRate;
776 if (avctx->rc_max_rate > 0)
777 ctx->encode_config.rcParams.maxBitRate = avctx->rc_max_rate;
780 if (avctx->codec->id == AV_CODEC_ID_H264)
781 ctx->encode_config.encodeCodecConfig.h264Config.qpPrimeYZeroTransformBypassFlag = 1;
783 ctx->encode_config.rcParams.rateControlMode = NV_ENC_PARAMS_RC_CONSTQP;
784 ctx->encode_config.rcParams.constQP.qpInterB = 0;
785 ctx->encode_config.rcParams.constQP.qpInterP = 0;
786 ctx->encode_config.rcParams.constQP.qpIntra = 0;
790 } else if (ctx->cbr) {
792 ctx->encode_config.rcParams.rateControlMode = NV_ENC_PARAMS_RC_CBR;
794 ctx->encode_config.rcParams.rateControlMode = NV_ENC_PARAMS_RC_2_PASS_QUALITY;
796 if (avctx->codec->id == AV_CODEC_ID_H264) {
797 ctx->encode_config.encodeCodecConfig.h264Config.adaptiveTransformMode = NV_ENC_H264_ADAPTIVE_TRANSFORM_ENABLE;
798 ctx->encode_config.encodeCodecConfig.h264Config.fmoMode = NV_ENC_H264_FMO_DISABLE;
801 } else if (avctx->global_quality > 0) {
802 ctx->encode_config.rcParams.rateControlMode = NV_ENC_PARAMS_RC_CONSTQP;
803 ctx->encode_config.rcParams.constQP.qpInterB = avctx->global_quality;
804 ctx->encode_config.rcParams.constQP.qpInterP = avctx->global_quality;
805 ctx->encode_config.rcParams.constQP.qpIntra = avctx->global_quality;
810 if (avctx->qmin >= 0 && avctx->qmax >= 0) {
811 ctx->encode_config.rcParams.enableMinQP = 1;
812 ctx->encode_config.rcParams.enableMaxQP = 1;
814 ctx->encode_config.rcParams.minQP.qpInterB = avctx->qmin;
815 ctx->encode_config.rcParams.minQP.qpInterP = avctx->qmin;
816 ctx->encode_config.rcParams.minQP.qpIntra = avctx->qmin;
818 ctx->encode_config.rcParams.maxQP.qpInterB = avctx->qmax;
819 ctx->encode_config.rcParams.maxQP.qpInterP = avctx->qmax;
820 ctx->encode_config.rcParams.maxQP.qpIntra = avctx->qmax;
822 qp_inter_p = (avctx->qmax + 3 * avctx->qmin) / 4; // biased towards Qmin
825 ctx->encode_config.rcParams.rateControlMode = NV_ENC_PARAMS_RC_2_PASS_VBR;
826 if (avctx->codec->id == AV_CODEC_ID_H264) {
827 ctx->encode_config.encodeCodecConfig.h264Config.adaptiveTransformMode = NV_ENC_H264_ADAPTIVE_TRANSFORM_ENABLE;
828 ctx->encode_config.encodeCodecConfig.h264Config.fmoMode = NV_ENC_H264_FMO_DISABLE;
831 ctx->encode_config.rcParams.rateControlMode = NV_ENC_PARAMS_RC_VBR_MINQP;
834 qp_inter_p = 26; // default to 26
837 ctx->encode_config.rcParams.rateControlMode = NV_ENC_PARAMS_RC_2_PASS_VBR;
839 ctx->encode_config.rcParams.rateControlMode = NV_ENC_PARAMS_RC_VBR;
843 ctx->encode_config.rcParams.enableInitialRCQP = 1;
844 ctx->encode_config.rcParams.initialRCQP.qpInterP = qp_inter_p;
846 if(avctx->i_quant_factor != 0.0 && avctx->b_quant_factor != 0.0) {
847 ctx->encode_config.rcParams.initialRCQP.qpIntra = av_clip(
848 qp_inter_p * fabs(avctx->i_quant_factor) + avctx->i_quant_offset, 0, 51);
849 ctx->encode_config.rcParams.initialRCQP.qpInterB = av_clip(
850 qp_inter_p * fabs(avctx->b_quant_factor) + avctx->b_quant_offset, 0, 51);
852 ctx->encode_config.rcParams.initialRCQP.qpIntra = qp_inter_p;
853 ctx->encode_config.rcParams.initialRCQP.qpInterB = qp_inter_p;
857 if (avctx->rc_buffer_size > 0) {
858 ctx->encode_config.rcParams.vbvBufferSize = avctx->rc_buffer_size;
859 } else if (ctx->encode_config.rcParams.averageBitRate > 0) {
860 ctx->encode_config.rcParams.vbvBufferSize = 2 * ctx->encode_config.rcParams.averageBitRate;
863 if (avctx->flags & AV_CODEC_FLAG_INTERLACED_DCT) {
864 ctx->encode_config.frameFieldMode = NV_ENC_PARAMS_FRAME_FIELD_MODE_FIELD;
866 ctx->encode_config.frameFieldMode = NV_ENC_PARAMS_FRAME_FIELD_MODE_FRAME;
869 switch (avctx->codec->id) {
870 case AV_CODEC_ID_H264:
871 ctx->encode_config.encodeCodecConfig.h264Config.h264VUIParameters.colourMatrix = avctx->colorspace;
872 ctx->encode_config.encodeCodecConfig.h264Config.h264VUIParameters.colourPrimaries = avctx->color_primaries;
873 ctx->encode_config.encodeCodecConfig.h264Config.h264VUIParameters.transferCharacteristics = avctx->color_trc;
874 ctx->encode_config.encodeCodecConfig.h264Config.h264VUIParameters.videoFullRangeFlag = (avctx->color_range == AVCOL_RANGE_JPEG
875 || avctx->pix_fmt == AV_PIX_FMT_YUVJ420P || avctx->pix_fmt == AV_PIX_FMT_YUVJ422P || avctx->pix_fmt == AV_PIX_FMT_YUVJ444P);
877 ctx->encode_config.encodeCodecConfig.h264Config.h264VUIParameters.colourDescriptionPresentFlag =
878 (avctx->colorspace != 2 || avctx->color_primaries != 2 || avctx->color_trc != 2);
880 ctx->encode_config.encodeCodecConfig.h264Config.h264VUIParameters.videoSignalTypePresentFlag =
881 (ctx->encode_config.encodeCodecConfig.h264Config.h264VUIParameters.colourDescriptionPresentFlag
882 || ctx->encode_config.encodeCodecConfig.h264Config.h264VUIParameters.videoFormat != 5
883 || ctx->encode_config.encodeCodecConfig.h264Config.h264VUIParameters.videoFullRangeFlag != 0);
885 ctx->encode_config.encodeCodecConfig.h264Config.sliceMode = 3;
886 ctx->encode_config.encodeCodecConfig.h264Config.sliceModeData = 1;
888 ctx->encode_config.encodeCodecConfig.h264Config.disableSPSPPS = (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) ? 1 : 0;
889 ctx->encode_config.encodeCodecConfig.h264Config.repeatSPSPPS = (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) ? 0 : 1;
891 ctx->encode_config.encodeCodecConfig.h264Config.outputAUD = 1;
894 switch (avctx->profile) {
895 case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
896 ctx->encode_config.profileGUID = NV_ENC_H264_PROFILE_HIGH_444_GUID;
898 case FF_PROFILE_H264_BASELINE:
899 ctx->encode_config.profileGUID = NV_ENC_H264_PROFILE_BASELINE_GUID;
901 case FF_PROFILE_H264_MAIN:
902 ctx->encode_config.profileGUID = NV_ENC_H264_PROFILE_MAIN_GUID;
904 case FF_PROFILE_H264_HIGH:
905 case FF_PROFILE_UNKNOWN:
906 ctx->encode_config.profileGUID = NV_ENC_H264_PROFILE_HIGH_GUID;
909 av_log(avctx, AV_LOG_WARNING, "Unsupported profile requested, falling back to high\n");
910 ctx->encode_config.profileGUID = NV_ENC_H264_PROFILE_HIGH_GUID;
914 if (!strcmp(ctx->profile, "high")) {
915 ctx->encode_config.profileGUID = NV_ENC_H264_PROFILE_HIGH_GUID;
916 avctx->profile = FF_PROFILE_H264_HIGH;
917 } else if (!strcmp(ctx->profile, "main")) {
918 ctx->encode_config.profileGUID = NV_ENC_H264_PROFILE_MAIN_GUID;
919 avctx->profile = FF_PROFILE_H264_MAIN;
920 } else if (!strcmp(ctx->profile, "baseline")) {
921 ctx->encode_config.profileGUID = NV_ENC_H264_PROFILE_BASELINE_GUID;
922 avctx->profile = FF_PROFILE_H264_BASELINE;
923 } else if (!strcmp(ctx->profile, "high444p")) {
924 ctx->encode_config.profileGUID = NV_ENC_H264_PROFILE_HIGH_444_GUID;
925 avctx->profile = FF_PROFILE_H264_HIGH_444_PREDICTIVE;
927 av_log(avctx, AV_LOG_FATAL, "Profile \"%s\" is unknown! Supported profiles: high, main, baseline\n", ctx->profile);
928 res = AVERROR(EINVAL);
933 // force setting profile as high444p if input is AV_PIX_FMT_YUV444P
934 if (avctx->pix_fmt == AV_PIX_FMT_YUV444P) {
935 ctx->encode_config.profileGUID = NV_ENC_H264_PROFILE_HIGH_444_GUID;
936 avctx->profile = FF_PROFILE_H264_HIGH_444_PREDICTIVE;
939 ctx->encode_config.encodeCodecConfig.h264Config.chromaFormatIDC = avctx->profile == FF_PROFILE_H264_HIGH_444_PREDICTIVE ? 3 : 1;
942 res = input_string_to_uint32(avctx, nvenc_h264_level_pairs, ctx->level, &ctx->encode_config.encodeCodecConfig.h264Config.level);
945 av_log(avctx, AV_LOG_FATAL, "Level \"%s\" is unknown! Supported levels: auto, 1, 1b, 1.1, 1.2, 1.3, 2, 2.1, 2.2, 3, 3.1, 3.2, 4, 4.1, 4.2, 5, 5.1\n", ctx->level);
949 ctx->encode_config.encodeCodecConfig.h264Config.level = NV_ENC_LEVEL_AUTOSELECT;
953 case AV_CODEC_ID_H265:
954 ctx->encode_config.encodeCodecConfig.hevcConfig.hevcVUIParameters.colourMatrix = avctx->colorspace;
955 ctx->encode_config.encodeCodecConfig.hevcConfig.hevcVUIParameters.colourPrimaries = avctx->color_primaries;
956 ctx->encode_config.encodeCodecConfig.hevcConfig.hevcVUIParameters.transferCharacteristics = avctx->color_trc;
957 ctx->encode_config.encodeCodecConfig.hevcConfig.hevcVUIParameters.videoFullRangeFlag = (avctx->color_range == AVCOL_RANGE_JPEG
958 || avctx->pix_fmt == AV_PIX_FMT_YUVJ420P || avctx->pix_fmt == AV_PIX_FMT_YUVJ422P || avctx->pix_fmt == AV_PIX_FMT_YUVJ444P);
960 ctx->encode_config.encodeCodecConfig.hevcConfig.hevcVUIParameters.colourDescriptionPresentFlag =
961 (avctx->colorspace != 2 || avctx->color_primaries != 2 || avctx->color_trc != 2);
963 ctx->encode_config.encodeCodecConfig.hevcConfig.hevcVUIParameters.videoSignalTypePresentFlag =
964 (ctx->encode_config.encodeCodecConfig.hevcConfig.hevcVUIParameters.colourDescriptionPresentFlag
965 || ctx->encode_config.encodeCodecConfig.hevcConfig.hevcVUIParameters.videoFormat != 5
966 || ctx->encode_config.encodeCodecConfig.hevcConfig.hevcVUIParameters.videoFullRangeFlag != 0);
968 ctx->encode_config.encodeCodecConfig.hevcConfig.sliceMode = 3;
969 ctx->encode_config.encodeCodecConfig.hevcConfig.sliceModeData = 1;
971 ctx->encode_config.encodeCodecConfig.hevcConfig.disableSPSPPS = (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) ? 1 : 0;
972 ctx->encode_config.encodeCodecConfig.hevcConfig.repeatSPSPPS = (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) ? 0 : 1;
974 ctx->encode_config.encodeCodecConfig.hevcConfig.outputAUD = 1;
976 /* No other profile is supported in the current SDK version 5 */
977 ctx->encode_config.profileGUID = NV_ENC_HEVC_PROFILE_MAIN_GUID;
978 avctx->profile = FF_PROFILE_HEVC_MAIN;
981 res = input_string_to_uint32(avctx, nvenc_hevc_level_pairs, ctx->level, &ctx->encode_config.encodeCodecConfig.hevcConfig.level);
984 av_log(avctx, AV_LOG_FATAL, "Level \"%s\" is unknown! Supported levels: auto, 1, 2, 2.1, 3, 3.1, 4, 4.1, 5, 5.1, 5.2, 6, 6.1, 6.2\n", ctx->level);
988 ctx->encode_config.encodeCodecConfig.hevcConfig.level = NV_ENC_LEVEL_AUTOSELECT;
992 if (!strcmp(ctx->tier, "main")) {
993 ctx->encode_config.encodeCodecConfig.hevcConfig.tier = NV_ENC_TIER_HEVC_MAIN;
994 } else if (!strcmp(ctx->tier, "high")) {
995 ctx->encode_config.encodeCodecConfig.hevcConfig.tier = NV_ENC_TIER_HEVC_HIGH;
997 av_log(avctx, AV_LOG_FATAL, "Tier \"%s\" is unknown! Supported tiers: main, high\n", ctx->tier);
998 res = AVERROR(EINVAL);
1004 /* Earlier switch/case will return if unknown codec is passed. */
1007 nv_status = p_nvenc->nvEncInitializeEncoder(ctx->nvencoder, &ctx->init_encode_params);
1008 if (nv_status != NV_ENC_SUCCESS) {
1009 av_log(avctx, AV_LOG_FATAL, "InitializeEncoder failed: 0x%x\n", (int)nv_status);
1010 res = AVERROR_EXTERNAL;
1014 ctx->input_surfaces = av_malloc(ctx->max_surface_count * sizeof(*ctx->input_surfaces));
1016 if (!ctx->input_surfaces) {
1017 res = AVERROR(ENOMEM);
1021 ctx->output_surfaces = av_malloc(ctx->max_surface_count * sizeof(*ctx->output_surfaces));
1023 if (!ctx->output_surfaces) {
1024 res = AVERROR(ENOMEM);
1028 for (surfaceCount = 0; surfaceCount < ctx->max_surface_count; ++surfaceCount) {
1029 NV_ENC_CREATE_INPUT_BUFFER allocSurf = { 0 };
1030 NV_ENC_CREATE_BITSTREAM_BUFFER allocOut = { 0 };
1031 allocSurf.version = NV_ENC_CREATE_INPUT_BUFFER_VER;
1032 allocOut.version = NV_ENC_CREATE_BITSTREAM_BUFFER_VER;
1034 allocSurf.width = (avctx->width + 31) & ~31;
1035 allocSurf.height = (avctx->height + 31) & ~31;
1037 allocSurf.memoryHeap = NV_ENC_MEMORY_HEAP_SYSMEM_CACHED;
1039 switch (avctx->pix_fmt) {
1040 case AV_PIX_FMT_YUV420P:
1041 allocSurf.bufferFmt = NV_ENC_BUFFER_FORMAT_YV12_PL;
1044 case AV_PIX_FMT_NV12:
1045 allocSurf.bufferFmt = NV_ENC_BUFFER_FORMAT_NV12_PL;
1048 case AV_PIX_FMT_YUV444P:
1049 allocSurf.bufferFmt = NV_ENC_BUFFER_FORMAT_YUV444_PL;
1053 av_log(avctx, AV_LOG_FATAL, "Invalid input pixel format\n");
1054 res = AVERROR(EINVAL);
1058 nv_status = p_nvenc->nvEncCreateInputBuffer(ctx->nvencoder, &allocSurf);
1059 if (nv_status != NV_ENC_SUCCESS) {
1060 av_log(avctx, AV_LOG_FATAL, "CreateInputBuffer failed\n");
1061 res = AVERROR_EXTERNAL;
1065 ctx->input_surfaces[surfaceCount].lockCount = 0;
1066 ctx->input_surfaces[surfaceCount].input_surface = allocSurf.inputBuffer;
1067 ctx->input_surfaces[surfaceCount].format = allocSurf.bufferFmt;
1068 ctx->input_surfaces[surfaceCount].width = allocSurf.width;
1069 ctx->input_surfaces[surfaceCount].height = allocSurf.height;
1071 /* 1MB is large enough to hold most output frames. NVENC increases this automaticaly if it's not enough. */
1072 allocOut.size = 1024 * 1024;
1074 allocOut.memoryHeap = NV_ENC_MEMORY_HEAP_SYSMEM_CACHED;
1076 nv_status = p_nvenc->nvEncCreateBitstreamBuffer(ctx->nvencoder, &allocOut);
1077 if (nv_status != NV_ENC_SUCCESS) {
1078 av_log(avctx, AV_LOG_FATAL, "CreateBitstreamBuffer failed\n");
1079 ctx->output_surfaces[surfaceCount++].output_surface = NULL;
1080 res = AVERROR_EXTERNAL;
1084 ctx->output_surfaces[surfaceCount].output_surface = allocOut.bitstreamBuffer;
1085 ctx->output_surfaces[surfaceCount].size = allocOut.size;
1086 ctx->output_surfaces[surfaceCount].busy = 0;
1089 if (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) {
1090 uint32_t outSize = 0;
1091 char tmpHeader[256];
1092 NV_ENC_SEQUENCE_PARAM_PAYLOAD payload = { 0 };
1093 payload.version = NV_ENC_SEQUENCE_PARAM_PAYLOAD_VER;
1095 payload.spsppsBuffer = tmpHeader;
1096 payload.inBufferSize = sizeof(tmpHeader);
1097 payload.outSPSPPSPayloadSize = &outSize;
1099 nv_status = p_nvenc->nvEncGetSequenceParams(ctx->nvencoder, &payload);
1100 if (nv_status != NV_ENC_SUCCESS) {
1101 av_log(avctx, AV_LOG_FATAL, "GetSequenceParams failed\n");
1105 avctx->extradata_size = outSize;
1106 avctx->extradata = av_mallocz(outSize + AV_INPUT_BUFFER_PADDING_SIZE);
1108 if (!avctx->extradata) {
1109 res = AVERROR(ENOMEM);
1113 memcpy(avctx->extradata, tmpHeader, outSize);
1116 if (ctx->encode_config.frameIntervalP > 1)
1117 avctx->has_b_frames = 2;
1119 if (ctx->encode_config.rcParams.averageBitRate > 0)
1120 avctx->bit_rate = ctx->encode_config.rcParams.averageBitRate;
1122 cpb_props = ff_add_cpb_side_data(avctx);
1124 return AVERROR(ENOMEM);
1125 cpb_props->max_bitrate = ctx->encode_config.rcParams.maxBitRate;
1126 cpb_props->avg_bitrate = avctx->bit_rate;
1127 cpb_props->buffer_size = ctx->encode_config.rcParams.vbvBufferSize;
1133 for (i = 0; i < surfaceCount; ++i) {
1134 p_nvenc->nvEncDestroyInputBuffer(ctx->nvencoder, ctx->input_surfaces[i].input_surface);
1135 if (ctx->output_surfaces[i].output_surface)
1136 p_nvenc->nvEncDestroyBitstreamBuffer(ctx->nvencoder, ctx->output_surfaces[i].output_surface);
1140 p_nvenc->nvEncDestroyEncoder(ctx->nvencoder);
1142 if (ctx->cu_context)
1143 dl_fn->cu_ctx_destroy(ctx->cu_context);
1145 nvenc_unload_nvenc(avctx);
1147 ctx->nvencoder = NULL;
1148 ctx->cu_context = NULL;
1153 static av_cold int nvenc_encode_close(AVCodecContext *avctx)
1155 NvencContext *ctx = avctx->priv_data;
1156 NvencDynLoadFunctions *dl_fn = &ctx->nvenc_dload_funcs;
1157 NV_ENCODE_API_FUNCTION_LIST *p_nvenc = &dl_fn->nvenc_funcs;
1160 av_freep(&ctx->timestamp_list.data);
1161 av_freep(&ctx->output_surface_ready_queue.data);
1162 av_freep(&ctx->output_surface_queue.data);
1164 for (i = 0; i < ctx->max_surface_count; ++i) {
1165 p_nvenc->nvEncDestroyInputBuffer(ctx->nvencoder, ctx->input_surfaces[i].input_surface);
1166 p_nvenc->nvEncDestroyBitstreamBuffer(ctx->nvencoder, ctx->output_surfaces[i].output_surface);
1168 ctx->max_surface_count = 0;
1170 p_nvenc->nvEncDestroyEncoder(ctx->nvencoder);
1171 ctx->nvencoder = NULL;
1173 dl_fn->cu_ctx_destroy(ctx->cu_context);
1174 ctx->cu_context = NULL;
1176 nvenc_unload_nvenc(avctx);
1181 static int process_output_surface(AVCodecContext *avctx, AVPacket *pkt, NvencOutputSurface *tmpoutsurf)
1183 NvencContext *ctx = avctx->priv_data;
1184 NvencDynLoadFunctions *dl_fn = &ctx->nvenc_dload_funcs;
1185 NV_ENCODE_API_FUNCTION_LIST *p_nvenc = &dl_fn->nvenc_funcs;
1187 uint32_t slice_mode_data;
1188 uint32_t *slice_offsets;
1189 NV_ENC_LOCK_BITSTREAM lock_params = { 0 };
1190 NVENCSTATUS nv_status;
1193 enum AVPictureType pict_type;
1195 switch (avctx->codec->id) {
1196 case AV_CODEC_ID_H264:
1197 slice_mode_data = ctx->encode_config.encodeCodecConfig.h264Config.sliceModeData;
1199 case AV_CODEC_ID_H265:
1200 slice_mode_data = ctx->encode_config.encodeCodecConfig.hevcConfig.sliceModeData;
1203 av_log(avctx, AV_LOG_ERROR, "Unknown codec name\n");
1204 res = AVERROR(EINVAL);
1207 slice_offsets = av_mallocz(slice_mode_data * sizeof(*slice_offsets));
1210 return AVERROR(ENOMEM);
1212 lock_params.version = NV_ENC_LOCK_BITSTREAM_VER;
1214 lock_params.doNotWait = 0;
1215 lock_params.outputBitstream = tmpoutsurf->output_surface;
1216 lock_params.sliceOffsets = slice_offsets;
1218 nv_status = p_nvenc->nvEncLockBitstream(ctx->nvencoder, &lock_params);
1219 if (nv_status != NV_ENC_SUCCESS) {
1220 av_log(avctx, AV_LOG_ERROR, "Failed locking bitstream buffer\n");
1221 res = AVERROR_EXTERNAL;
1225 if (res = ff_alloc_packet2(avctx, pkt, lock_params.bitstreamSizeInBytes,0)) {
1226 p_nvenc->nvEncUnlockBitstream(ctx->nvencoder, tmpoutsurf->output_surface);
1230 memcpy(pkt->data, lock_params.bitstreamBufferPtr, lock_params.bitstreamSizeInBytes);
1232 nv_status = p_nvenc->nvEncUnlockBitstream(ctx->nvencoder, tmpoutsurf->output_surface);
1233 if (nv_status != NV_ENC_SUCCESS)
1234 av_log(avctx, AV_LOG_ERROR, "Failed unlocking bitstream buffer, expect the gates of mordor to open\n");
1236 switch (lock_params.pictureType) {
1237 case NV_ENC_PIC_TYPE_IDR:
1238 pkt->flags |= AV_PKT_FLAG_KEY;
1239 case NV_ENC_PIC_TYPE_I:
1240 pict_type = AV_PICTURE_TYPE_I;
1242 case NV_ENC_PIC_TYPE_P:
1243 pict_type = AV_PICTURE_TYPE_P;
1245 case NV_ENC_PIC_TYPE_B:
1246 pict_type = AV_PICTURE_TYPE_B;
1248 case NV_ENC_PIC_TYPE_BI:
1249 pict_type = AV_PICTURE_TYPE_BI;
1252 av_log(avctx, AV_LOG_ERROR, "Unknown picture type encountered, expect the output to be broken.\n");
1253 av_log(avctx, AV_LOG_ERROR, "Please report this error and include as much information on how to reproduce it as possible.\n");
1254 res = AVERROR_EXTERNAL;
1258 #if FF_API_CODED_FRAME
1259 FF_DISABLE_DEPRECATION_WARNINGS
1260 avctx->coded_frame->pict_type = pict_type;
1261 FF_ENABLE_DEPRECATION_WARNINGS
1264 ff_side_data_set_encoder_stats(pkt,
1265 (lock_params.frameAvgQP - 1) * FF_QP2LAMBDA, NULL, 0, pict_type);
1267 pkt->pts = lock_params.outputTimeStamp;
1268 pkt->dts = timestamp_queue_dequeue(&ctx->timestamp_list);
1270 /* when there're b frame(s), set dts offset */
1271 if (ctx->encode_config.frameIntervalP >= 2)
1274 if (pkt->dts > pkt->pts)
1275 pkt->dts = pkt->pts;
1277 if (ctx->last_dts != AV_NOPTS_VALUE && pkt->dts <= ctx->last_dts)
1278 pkt->dts = ctx->last_dts + 1;
1280 ctx->last_dts = pkt->dts;
1282 av_free(slice_offsets);
1288 av_free(slice_offsets);
1289 timestamp_queue_dequeue(&ctx->timestamp_list);
1294 static int nvenc_encode_frame(AVCodecContext *avctx, AVPacket *pkt,
1295 const AVFrame *frame, int *got_packet)
1297 NVENCSTATUS nv_status;
1298 NvencOutputSurface *tmpoutsurf;
1301 NvencContext *ctx = avctx->priv_data;
1302 NvencDynLoadFunctions *dl_fn = &ctx->nvenc_dload_funcs;
1303 NV_ENCODE_API_FUNCTION_LIST *p_nvenc = &dl_fn->nvenc_funcs;
1305 NV_ENC_PIC_PARAMS pic_params = { 0 };
1306 pic_params.version = NV_ENC_PIC_PARAMS_VER;
1309 NV_ENC_LOCK_INPUT_BUFFER lockBufferParams = { 0 };
1310 NvencInputSurface *inSurf = NULL;
1312 for (i = 0; i < ctx->max_surface_count; ++i) {
1313 if (!ctx->input_surfaces[i].lockCount) {
1314 inSurf = &ctx->input_surfaces[i];
1321 inSurf->lockCount = 1;
1323 lockBufferParams.version = NV_ENC_LOCK_INPUT_BUFFER_VER;
1324 lockBufferParams.inputBuffer = inSurf->input_surface;
1326 nv_status = p_nvenc->nvEncLockInputBuffer(ctx->nvencoder, &lockBufferParams);
1327 if (nv_status != NV_ENC_SUCCESS) {
1328 av_log(avctx, AV_LOG_ERROR, "Failed locking nvenc input buffer\n");
1332 if (avctx->pix_fmt == AV_PIX_FMT_YUV420P) {
1333 uint8_t *buf = lockBufferParams.bufferDataPtr;
1335 av_image_copy_plane(buf, lockBufferParams.pitch,
1336 frame->data[0], frame->linesize[0],
1337 avctx->width, avctx->height);
1339 buf += inSurf->height * lockBufferParams.pitch;
1341 av_image_copy_plane(buf, lockBufferParams.pitch >> 1,
1342 frame->data[2], frame->linesize[2],
1343 avctx->width >> 1, avctx->height >> 1);
1345 buf += (inSurf->height * lockBufferParams.pitch) >> 2;
1347 av_image_copy_plane(buf, lockBufferParams.pitch >> 1,
1348 frame->data[1], frame->linesize[1],
1349 avctx->width >> 1, avctx->height >> 1);
1350 } else if (avctx->pix_fmt == AV_PIX_FMT_NV12) {
1351 uint8_t *buf = lockBufferParams.bufferDataPtr;
1353 av_image_copy_plane(buf, lockBufferParams.pitch,
1354 frame->data[0], frame->linesize[0],
1355 avctx->width, avctx->height);
1357 buf += inSurf->height * lockBufferParams.pitch;
1359 av_image_copy_plane(buf, lockBufferParams.pitch,
1360 frame->data[1], frame->linesize[1],
1361 avctx->width, avctx->height >> 1);
1362 } else if (avctx->pix_fmt == AV_PIX_FMT_YUV444P) {
1363 uint8_t *buf = lockBufferParams.bufferDataPtr;
1365 av_image_copy_plane(buf, lockBufferParams.pitch,
1366 frame->data[0], frame->linesize[0],
1367 avctx->width, avctx->height);
1369 buf += inSurf->height * lockBufferParams.pitch;
1371 av_image_copy_plane(buf, lockBufferParams.pitch,
1372 frame->data[1], frame->linesize[1],
1373 avctx->width, avctx->height);
1375 buf += inSurf->height * lockBufferParams.pitch;
1377 av_image_copy_plane(buf, lockBufferParams.pitch,
1378 frame->data[2], frame->linesize[2],
1379 avctx->width, avctx->height);
1381 av_log(avctx, AV_LOG_FATAL, "Invalid pixel format!\n");
1382 return AVERROR(EINVAL);
1385 nv_status = p_nvenc->nvEncUnlockInputBuffer(ctx->nvencoder, inSurf->input_surface);
1386 if (nv_status != NV_ENC_SUCCESS) {
1387 av_log(avctx, AV_LOG_FATAL, "Failed unlocking input buffer!\n");
1388 return AVERROR_EXTERNAL;
1391 for (i = 0; i < ctx->max_surface_count; ++i)
1392 if (!ctx->output_surfaces[i].busy)
1395 if (i == ctx->max_surface_count) {
1396 inSurf->lockCount = 0;
1397 av_log(avctx, AV_LOG_FATAL, "No free output surface found!\n");
1398 return AVERROR_EXTERNAL;
1401 ctx->output_surfaces[i].input_surface = inSurf;
1403 pic_params.inputBuffer = inSurf->input_surface;
1404 pic_params.bufferFmt = inSurf->format;
1405 pic_params.inputWidth = avctx->width;
1406 pic_params.inputHeight = avctx->height;
1407 pic_params.outputBitstream = ctx->output_surfaces[i].output_surface;
1408 pic_params.completionEvent = 0;
1410 if (avctx->flags & AV_CODEC_FLAG_INTERLACED_DCT) {
1411 if (frame->top_field_first) {
1412 pic_params.pictureStruct = NV_ENC_PIC_STRUCT_FIELD_TOP_BOTTOM;
1414 pic_params.pictureStruct = NV_ENC_PIC_STRUCT_FIELD_BOTTOM_TOP;
1417 pic_params.pictureStruct = NV_ENC_PIC_STRUCT_FRAME;
1420 pic_params.encodePicFlags = 0;
1421 pic_params.inputTimeStamp = frame->pts;
1422 pic_params.inputDuration = 0;
1423 switch (avctx->codec->id) {
1424 case AV_CODEC_ID_H264:
1425 pic_params.codecPicParams.h264PicParams.sliceMode = ctx->encode_config.encodeCodecConfig.h264Config.sliceMode;
1426 pic_params.codecPicParams.h264PicParams.sliceModeData = ctx->encode_config.encodeCodecConfig.h264Config.sliceModeData;
1428 case AV_CODEC_ID_H265:
1429 pic_params.codecPicParams.hevcPicParams.sliceMode = ctx->encode_config.encodeCodecConfig.hevcConfig.sliceMode;
1430 pic_params.codecPicParams.hevcPicParams.sliceModeData = ctx->encode_config.encodeCodecConfig.hevcConfig.sliceModeData;
1433 av_log(avctx, AV_LOG_ERROR, "Unknown codec name\n");
1434 return AVERROR(EINVAL);
1437 res = timestamp_queue_enqueue(&ctx->timestamp_list, frame->pts);
1442 pic_params.encodePicFlags = NV_ENC_PIC_FLAG_EOS;
1445 nv_status = p_nvenc->nvEncEncodePicture(ctx->nvencoder, &pic_params);
1447 if (frame && nv_status == NV_ENC_ERR_NEED_MORE_INPUT) {
1448 res = out_surf_queue_enqueue(&ctx->output_surface_queue, &ctx->output_surfaces[i]);
1453 ctx->output_surfaces[i].busy = 1;
1456 if (nv_status != NV_ENC_SUCCESS && nv_status != NV_ENC_ERR_NEED_MORE_INPUT) {
1457 av_log(avctx, AV_LOG_ERROR, "EncodePicture failed!\n");
1458 return AVERROR_EXTERNAL;
1461 if (nv_status != NV_ENC_ERR_NEED_MORE_INPUT) {
1462 while (ctx->output_surface_queue.count) {
1463 tmpoutsurf = out_surf_queue_dequeue(&ctx->output_surface_queue);
1464 res = out_surf_queue_enqueue(&ctx->output_surface_ready_queue, tmpoutsurf);
1471 res = out_surf_queue_enqueue(&ctx->output_surface_ready_queue, &ctx->output_surfaces[i]);
1476 ctx->output_surfaces[i].busy = 1;
1480 if (ctx->output_surface_ready_queue.count && (!frame || ctx->output_surface_ready_queue.count + ctx->output_surface_queue.count >= ctx->buffer_delay)) {
1481 tmpoutsurf = out_surf_queue_dequeue(&ctx->output_surface_ready_queue);
1483 res = process_output_surface(avctx, pkt, tmpoutsurf);
1488 tmpoutsurf->busy = 0;
1489 av_assert0(tmpoutsurf->input_surface->lockCount);
1490 tmpoutsurf->input_surface->lockCount--;
1500 static const enum AVPixelFormat pix_fmts_nvenc[] = {
1507 #define OFFSET(x) offsetof(NvencContext, x)
1508 #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
1509 static const AVOption options[] = {
1510 { "preset", "Set the encoding preset (one of slow = hq 2pass, medium = hq, fast = hp, hq, hp, bd, ll, llhq, llhp, default)", OFFSET(preset), AV_OPT_TYPE_STRING, { .str = "medium" }, 0, 0, VE },
1511 { "profile", "Set the encoding profile (high, main, baseline or high444p)", OFFSET(profile), AV_OPT_TYPE_STRING, { .str = "main" }, 0, 0, VE },
1512 { "level", "Set the encoding level restriction (auto, 1.0, 1.0b, 1.1, 1.2, ..., 4.2, 5.0, 5.1)", OFFSET(level), AV_OPT_TYPE_STRING, { .str = "auto" }, 0, 0, VE },
1513 { "tier", "Set the encoding tier (main or high)", OFFSET(tier), AV_OPT_TYPE_STRING, { .str = "main" }, 0, 0, VE },
1514 { "cbr", "Use cbr encoding mode", OFFSET(cbr), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, VE },
1515 { "2pass", "Use 2pass encoding mode", OFFSET(twopass), AV_OPT_TYPE_BOOL, { .i64 = -1 }, -1, 1, VE },
1516 { "gpu", "Selects which NVENC capable GPU to use. First GPU is 0, second is 1, and so on.", OFFSET(gpu), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, INT_MAX, VE },
1517 { "delay", "Delays frame output by the given amount of frames.", OFFSET(buffer_delay), AV_OPT_TYPE_INT, { .i64 = INT_MAX }, 0, INT_MAX, VE },
1521 static const AVCodecDefault nvenc_defaults[] = {
1533 #if CONFIG_NVENC_ENCODER
1534 static const AVClass nvenc_class = {
1535 .class_name = "nvenc",
1536 .item_name = av_default_item_name,
1538 .version = LIBAVUTIL_VERSION_INT,
1541 AVCodec ff_nvenc_encoder = {
1543 .long_name = NULL_IF_CONFIG_SMALL("NVIDIA NVENC h264 encoder"),
1544 .type = AVMEDIA_TYPE_VIDEO,
1545 .id = AV_CODEC_ID_H264,
1546 .priv_data_size = sizeof(NvencContext),
1547 .init = nvenc_encode_init,
1548 .encode2 = nvenc_encode_frame,
1549 .close = nvenc_encode_close,
1550 .capabilities = AV_CODEC_CAP_DELAY,
1551 .priv_class = &nvenc_class,
1552 .defaults = nvenc_defaults,
1553 .pix_fmts = pix_fmts_nvenc,
1557 /* Add an alias for nvenc_h264 */
1558 #if CONFIG_NVENC_H264_ENCODER
1559 static const AVClass nvenc_h264_class = {
1560 .class_name = "nvenc_h264",
1561 .item_name = av_default_item_name,
1563 .version = LIBAVUTIL_VERSION_INT,
1566 AVCodec ff_nvenc_h264_encoder = {
1567 .name = "nvenc_h264",
1568 .long_name = NULL_IF_CONFIG_SMALL("NVIDIA NVENC h264 encoder"),
1569 .type = AVMEDIA_TYPE_VIDEO,
1570 .id = AV_CODEC_ID_H264,
1571 .priv_data_size = sizeof(NvencContext),
1572 .init = nvenc_encode_init,
1573 .encode2 = nvenc_encode_frame,
1574 .close = nvenc_encode_close,
1575 .capabilities = AV_CODEC_CAP_DELAY,
1576 .priv_class = &nvenc_h264_class,
1577 .defaults = nvenc_defaults,
1578 .pix_fmts = pix_fmts_nvenc,
1582 #if CONFIG_NVENC_HEVC_ENCODER
1583 static const AVClass nvenc_hevc_class = {
1584 .class_name = "nvenc_hevc",
1585 .item_name = av_default_item_name,
1587 .version = LIBAVUTIL_VERSION_INT,
1590 AVCodec ff_nvenc_hevc_encoder = {
1591 .name = "nvenc_hevc",
1592 .long_name = NULL_IF_CONFIG_SMALL("NVIDIA NVENC hevc encoder"),
1593 .type = AVMEDIA_TYPE_VIDEO,
1594 .id = AV_CODEC_ID_H265,
1595 .priv_data_size = sizeof(NvencContext),
1596 .init = nvenc_encode_init,
1597 .encode2 = nvenc_encode_frame,
1598 .close = nvenc_encode_close,
1599 .capabilities = AV_CODEC_CAP_DELAY,
1600 .priv_class = &nvenc_hevc_class,
1601 .defaults = nvenc_defaults,
1602 .pix_fmts = pix_fmts_nvenc,