From: Hendrik Leppkes <h.leppkes@gmail.com>
Date: Thu, 17 Dec 2015 12:41:29 +0000 (+0100)
Subject: Merge commit '1520c6ff05d835da4b793318fc88bbbc129c86a1'
X-Git-Url: https://git.sesse.net/?a=commitdiff_plain;h=5fc17edc7dcec5ea2989a0d764f3cfb666a7e663;p=ffmpeg

Merge commit '1520c6ff05d835da4b793318fc88bbbc129c86a1'

* commit '1520c6ff05d835da4b793318fc88bbbc129c86a1':
  nvenc: export CPB props side data

Merged-by: Hendrik Leppkes <h.leppkes@gmail.com>
---

5fc17edc7dcec5ea2989a0d764f3cfb666a7e663
diff --cc libavcodec/nvenc.c
index 31f2dfdd334,92686c32864..a3b02fa99f4
--- a/libavcodec/nvenc.c
+++ b/libavcodec/nvenc.c
@@@ -458,749 -515,539 +458,757 @@@ error
      return 0;
  }
  
 -static int nvenc_setup_hevc_config(AVCodecContext *avctx)
 +static av_cold int nvenc_dyload_nvenc(AVCodecContext *avctx)
  {
 -    NVENCContext *ctx                      = avctx->priv_data;
 -    NV_ENC_CONFIG *cc                      = &ctx->config;
 -    NV_ENC_CONFIG_HEVC *hevc               = &cc->encodeCodecConfig.hevcConfig;
 +    PNVENCODEAPICREATEINSTANCE nvEncodeAPICreateInstance = 0;
 +    NVENCSTATUS nvstatus;
  
 -    hevc->disableSPSPPS = (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) ? 1 : 0;
 -    hevc->repeatSPSPPS  = (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) ? 0 : 1;
 +    NvencContext *ctx = avctx->priv_data;
 +    NvencDynLoadFunctions *dl_fn = &ctx->nvenc_dload_funcs;
  
 -    hevc->maxNumRefFramesInDPB = avctx->refs;
 -    hevc->idrPeriod            = cc->gopLength;
 +    if (!nvenc_check_cuda(avctx))
 +        return 0;
  
 -    /* No other profile is supported in the current SDK version 5 */
 -    cc->profileGUID = NV_ENC_HEVC_PROFILE_MAIN_GUID;
 -    avctx->profile  = FF_PROFILE_HEVC_MAIN;
 +    if (dl_fn->nvenc_lib)
 +        return 1;
  
 -    if (ctx->level) {
 -        hevc->level = ctx->level;
 +#if defined(_WIN32)
 +    if (sizeof(void*) == 8) {
 +        dl_fn->nvenc_lib = LoadLibrary(TEXT("nvEncodeAPI64.dll"));
      } else {
 -        hevc->level = NV_ENC_LEVEL_AUTOSELECT;
 +        dl_fn->nvenc_lib = LoadLibrary(TEXT("nvEncodeAPI.dll"));
      }
 +#else
 +    dl_fn->nvenc_lib = dlopen("libnvidia-encode.so.1", RTLD_LAZY);
 +#endif
  
 -    if (ctx->tier) {
 -        hevc->tier = ctx->tier;
 +    if (!dl_fn->nvenc_lib) {
 +        av_log(avctx, AV_LOG_FATAL, "Failed loading the nvenc library\n");
 +        goto error;
      }
  
 -    return 0;
 -}
 -static int nvenc_setup_codec_config(AVCodecContext *avctx)
 -{
 -    switch (avctx->codec->id) {
 -    case AV_CODEC_ID_H264:
 -        return nvenc_setup_h264_config(avctx);
 -    case AV_CODEC_ID_HEVC:
 -        return nvenc_setup_hevc_config(avctx);
 -    }
 -    return 0;
 -}
 +    nvEncodeAPICreateInstance = (PNVENCODEAPICREATEINSTANCE)LOAD_FUNC(dl_fn->nvenc_lib, "NvEncodeAPICreateInstance");
  
 -static int nvenc_setup_encoder(AVCodecContext *avctx)
 -{
 -    NVENCContext *ctx               = avctx->priv_data;
 -    NV_ENCODE_API_FUNCTION_LIST *nv = &ctx->nvel.nvenc_funcs;
 -    NV_ENC_PRESET_CONFIG preset_cfg = { 0 };
 -    AVCPBProperties *cpb_props;
 -    int ret;
 +    if (!nvEncodeAPICreateInstance) {
 +        av_log(avctx, AV_LOG_FATAL, "Failed to load nvenc entrypoint\n");
 +        goto error;
 +    }
  
 -    ctx->params.version = NV_ENC_INITIALIZE_PARAMS_VER;
 +    dl_fn->nvenc_funcs.version = NV_ENCODE_API_FUNCTION_LIST_VER;
  
 -    ctx->params.encodeHeight = avctx->height;
 -    ctx->params.encodeWidth  = avctx->width;
 +    nvstatus = nvEncodeAPICreateInstance(&dl_fn->nvenc_funcs);
  
 -    if (avctx->sample_aspect_ratio.num &&
 -        avctx->sample_aspect_ratio.den &&
 -        (avctx->sample_aspect_ratio.num != 1 ||
 -         avctx->sample_aspect_ratio.den != 1)) {
 -        av_reduce(&ctx->params.darWidth,
 -                  &ctx->params.darHeight,
 -                  avctx->width * avctx->sample_aspect_ratio.num,
 -                  avctx->height * avctx->sample_aspect_ratio.den,
 -                  INT_MAX / 8);
 -    } else {
 -        ctx->params.darHeight = avctx->height;
 -        ctx->params.darWidth  = avctx->width;
 +    if (nvstatus != NV_ENC_SUCCESS) {
 +        av_log(avctx, AV_LOG_FATAL, "Failed to create nvenc instance\n");
 +        goto error;
      }
  
 -    ctx->params.frameRateNum = avctx->time_base.den;
 -    ctx->params.frameRateDen = avctx->time_base.num * avctx->ticks_per_frame;
 +    av_log(avctx, AV_LOG_VERBOSE, "Nvenc initialized successfully\n");
  
 -    ctx->params.enableEncodeAsync = 0;
 -    ctx->params.enablePTD         = 1;
 +    return 1;
  
 -    ctx->params.encodeConfig = &ctx->config;
 +error:
 +    if (dl_fn->nvenc_lib)
 +        DL_CLOSE_FUNC(dl_fn->nvenc_lib);
  
 -    nvec_map_preset(ctx);
 +    dl_fn->nvenc_lib = NULL;
  
 -    preset_cfg.version           = NV_ENC_PRESET_CONFIG_VER;
 -    preset_cfg.presetCfg.version = NV_ENC_CONFIG_VER;
 +    return 0;
 +}
  
 -    ret = nv->nvEncGetEncodePresetConfig(ctx->nvenc_ctx,
 -                                         ctx->params.encodeGUID,
 -                                         ctx->params.presetGUID,
 -                                         &preset_cfg);
 -    if (ret != NV_ENC_SUCCESS) {
 -        av_log(avctx, AV_LOG_ERROR,
 -               "Cannot get the preset configuration\n");
 -        return AVERROR_UNKNOWN;
 -    }
 +static av_cold void nvenc_unload_nvenc(AVCodecContext *avctx)
 +{
 +    NvencContext *ctx = avctx->priv_data;
 +    NvencDynLoadFunctions *dl_fn = &ctx->nvenc_dload_funcs;
  
 -    memcpy(&ctx->config, &preset_cfg.presetCfg, sizeof(ctx->config));
 +    DL_CLOSE_FUNC(dl_fn->nvenc_lib);
 +    dl_fn->nvenc_lib = NULL;
  
 -    ctx->config.version = NV_ENC_CONFIG_VER;
 +    dl_fn->nvenc_device_count = 0;
  
 -    if (avctx->gop_size > 0) {
 -        if (avctx->max_b_frames > 0) {
 -            ctx->last_dts = -2;
 -            /* 0 is intra-only,
 -             * 1 is I/P only,
 -             * 2 is one B Frame,
 -             * 3 two B frames, and so on. */
 -            ctx->config.frameIntervalP = avctx->max_b_frames + 1;
 -        } else if (avctx->max_b_frames == 0) {
 -            ctx->config.frameIntervalP = 1;
 -        }
 -        ctx->config.gopLength = avctx->gop_size;
 -    } else if (avctx->gop_size == 0) {
 -        ctx->config.frameIntervalP = 0;
 -        ctx->config.gopLength      = 1;
 -    }
 +    DL_CLOSE_FUNC(dl_fn->cuda_lib);
 +    dl_fn->cuda_lib = NULL;
  
 -    if (ctx->config.frameIntervalP > 1)
 -        avctx->max_b_frames = ctx->config.frameIntervalP - 1;
 +    dl_fn->cu_init = NULL;
 +    dl_fn->cu_device_get_count = NULL;
 +    dl_fn->cu_device_get = NULL;
 +    dl_fn->cu_device_get_name = NULL;
 +    dl_fn->cu_device_compute_capability = NULL;
 +    dl_fn->cu_ctx_create = NULL;
 +    dl_fn->cu_ctx_pop_current = NULL;
 +    dl_fn->cu_ctx_destroy = NULL;
  
 -    nvenc_setup_rate_control(avctx);
 +    av_log(avctx, AV_LOG_VERBOSE, "Nvenc unloaded\n");
 +}
  
 -    if (avctx->flags & AV_CODEC_FLAG_INTERLACED_DCT) {
 -        ctx->config.frameFieldMode = NV_ENC_PARAMS_FRAME_FIELD_MODE_FIELD;
 -    } else {
 -        ctx->config.frameFieldMode = NV_ENC_PARAMS_FRAME_FIELD_MODE_FRAME;
 +static av_cold int nvenc_encode_init(AVCodecContext *avctx)
 +{
 +    NV_ENC_OPEN_ENCODE_SESSION_EX_PARAMS encode_session_params = { 0 };
 +    NV_ENC_PRESET_CONFIG preset_config = { 0 };
 +    CUcontext cu_context_curr;
 +    CUresult cu_res;
 +    GUID encoder_preset = NV_ENC_PRESET_HQ_GUID;
 +    GUID codec;
 +    NVENCSTATUS nv_status = NV_ENC_SUCCESS;
++    AVCPBProperties *cpb_props;
 +    int surfaceCount = 0;
 +    int i, num_mbs;
 +    int isLL = 0;
 +    int lossless = 0;
 +    int res = 0;
 +    int dw, dh;
 +    int qp_inter_p;
 +
 +    NvencContext *ctx = avctx->priv_data;
 +    NvencDynLoadFunctions *dl_fn = &ctx->nvenc_dload_funcs;
 +    NV_ENCODE_API_FUNCTION_LIST *p_nvenc = &dl_fn->nvenc_funcs;
 +
 +    if (!nvenc_dyload_nvenc(avctx))
 +        return AVERROR_EXTERNAL;
 +
 +    ctx->last_dts = AV_NOPTS_VALUE;
 +
 +    ctx->encode_config.version = NV_ENC_CONFIG_VER;
 +    ctx->init_encode_params.version = NV_ENC_INITIALIZE_PARAMS_VER;
 +    preset_config.version = NV_ENC_PRESET_CONFIG_VER;
 +    preset_config.presetCfg.version = NV_ENC_CONFIG_VER;
 +    encode_session_params.version = NV_ENC_OPEN_ENCODE_SESSION_EX_PARAMS_VER;
 +    encode_session_params.apiVersion = NVENCAPI_VERSION;
 +
 +    if (ctx->gpu >= dl_fn->nvenc_device_count) {
 +        av_log(avctx, AV_LOG_FATAL, "Requested GPU %d, but only %d GPUs are available!\n", ctx->gpu, dl_fn->nvenc_device_count);
 +        res = AVERROR(EINVAL);
 +        goto error;
      }
  
 -    if ((ret = nvenc_setup_codec_config(avctx)) < 0)
 -        return ret;
 +    ctx->cu_context = NULL;
 +    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
  
 -    ret = nv->nvEncInitializeEncoder(ctx->nvenc_ctx, &ctx->params);
 -    if (ret != NV_ENC_SUCCESS) {
 -        av_log(avctx, AV_LOG_ERROR, "Cannot initialize the decoder");
 -        return AVERROR_UNKNOWN;
 +    if (cu_res != CUDA_SUCCESS) {
 +        av_log(avctx, AV_LOG_FATAL, "Failed creating CUDA context for NVENC: 0x%x\n", (int)cu_res);
 +        res = AVERROR_EXTERNAL;
 +        goto error;
      }
  
 -    cpb_props = ff_add_cpb_side_data(avctx);
 -    if (!cpb_props)
 -        return AVERROR(ENOMEM);
 -    cpb_props->max_bitrate = avctx->rc_max_rate;
 -    cpb_props->min_bitrate = avctx->rc_min_rate;
 -    cpb_props->avg_bitrate = avctx->bit_rate;
 -    cpb_props->buffer_size = avctx->rc_buffer_size;
 +    cu_res = dl_fn->cu_ctx_pop_current(&cu_context_curr);
  
 -    return 0;
 -}
 +    if (cu_res != CUDA_SUCCESS) {
 +        av_log(avctx, AV_LOG_FATAL, "Failed popping CUDA context: 0x%x\n", (int)cu_res);
 +        res = AVERROR_EXTERNAL;
 +        goto error;
 +    }
  
 -static int nvenc_alloc_surface(AVCodecContext *avctx, int idx)
 -{
 -    NVENCContext *ctx               = avctx->priv_data;
 -    NV_ENCODE_API_FUNCTION_LIST *nv = &ctx->nvel.nvenc_funcs;
 -    int ret;
 -    NV_ENC_CREATE_INPUT_BUFFER in_buffer      = { 0 };
 -    NV_ENC_CREATE_BITSTREAM_BUFFER out_buffer = { 0 };
 +    encode_session_params.device = ctx->cu_context;
 +    encode_session_params.deviceType = NV_ENC_DEVICE_TYPE_CUDA;
  
 -    in_buffer.version  = NV_ENC_CREATE_INPUT_BUFFER_VER;
 -    out_buffer.version = NV_ENC_CREATE_BITSTREAM_BUFFER_VER;
 +    nv_status = p_nvenc->nvEncOpenEncodeSessionEx(&encode_session_params, &ctx->nvencoder);
 +    if (nv_status != NV_ENC_SUCCESS) {
 +        ctx->nvencoder = NULL;
 +        av_log(avctx, AV_LOG_FATAL, "OpenEncodeSessionEx failed: 0x%x\n", (int)nv_status);
 +        res = AVERROR_EXTERNAL;
 +        goto error;
 +    }
  
 -    in_buffer.width  = avctx->width;
 -    in_buffer.height = avctx->height;
 +    if (ctx->preset) {
 +        if (!strcmp(ctx->preset, "slow")) {
 +            encoder_preset = NV_ENC_PRESET_HQ_GUID;
 +            ctx->twopass = 1;
 +        } else if (!strcmp(ctx->preset, "medium")) {
 +            encoder_preset = NV_ENC_PRESET_HQ_GUID;
 +            ctx->twopass = 0;
 +        } else if (!strcmp(ctx->preset, "fast")) {
 +            encoder_preset = NV_ENC_PRESET_HP_GUID;
 +            ctx->twopass = 0;
 +        } else if (!strcmp(ctx->preset, "hq")) {
 +            encoder_preset = NV_ENC_PRESET_HQ_GUID;
 +        } else if (!strcmp(ctx->preset, "hp")) {
 +            encoder_preset = NV_ENC_PRESET_HP_GUID;
 +        } else if (!strcmp(ctx->preset, "bd")) {
 +            encoder_preset = NV_ENC_PRESET_BD_GUID;
 +        } else if (!strcmp(ctx->preset, "ll")) {
 +            encoder_preset = NV_ENC_PRESET_LOW_LATENCY_DEFAULT_GUID;
 +            isLL = 1;
 +        } else if (!strcmp(ctx->preset, "llhp")) {
 +            encoder_preset = NV_ENC_PRESET_LOW_LATENCY_HP_GUID;
 +            isLL = 1;
 +        } else if (!strcmp(ctx->preset, "llhq")) {
 +            encoder_preset = NV_ENC_PRESET_LOW_LATENCY_HQ_GUID;
 +            isLL = 1;
 +        } else if (!strcmp(ctx->preset, "lossless")) {
 +            encoder_preset = NV_ENC_PRESET_LOSSLESS_DEFAULT_GUID;
 +            lossless = 1;
 +        } else if (!strcmp(ctx->preset, "losslesshp")) {
 +            encoder_preset = NV_ENC_PRESET_LOSSLESS_HP_GUID;
 +            lossless = 1;
 +        } else if (!strcmp(ctx->preset, "default")) {
 +            encoder_preset = NV_ENC_PRESET_DEFAULT_GUID;
 +        } else {
 +            av_log(avctx, AV_LOG_FATAL, "Preset \"%s\" is unknown! Supported presets: slow, medium, high, hp, hq, bd, ll, llhp, llhq, lossless, losslesshp, default\n", ctx->preset);
 +            res = AVERROR(EINVAL);
 +            goto error;
 +        }
 +    }
  
 -    in_buffer.memoryHeap = NV_ENC_MEMORY_HEAP_SYSMEM_UNCACHED;
 +    if (ctx->twopass < 0) {
 +        ctx->twopass = isLL;
 +    }
  
 -    switch (avctx->pix_fmt) {
 -    case AV_PIX_FMT_YUV420P:
 -        in_buffer.bufferFmt = NV_ENC_BUFFER_FORMAT_YV12_PL;
 -        break;
 -    case AV_PIX_FMT_NV12:
 -        in_buffer.bufferFmt = NV_ENC_BUFFER_FORMAT_NV12_PL;
 +    switch (avctx->codec->id) {
 +    case AV_CODEC_ID_H264:
 +        codec = NV_ENC_CODEC_H264_GUID;
          break;
 -    case AV_PIX_FMT_YUV444P:
 -        in_buffer.bufferFmt = NV_ENC_BUFFER_FORMAT_YUV444_PL;
 +    case AV_CODEC_ID_H265:
 +        codec = NV_ENC_CODEC_HEVC_GUID;
          break;
      default:
 -        return AVERROR_BUG;
 +        av_log(avctx, AV_LOG_ERROR, "Unknown codec name\n");
 +        res = AVERROR(EINVAL);
 +        goto error;
      }
  
 -    ret = nv->nvEncCreateInputBuffer(ctx->nvenc_ctx, &in_buffer);
 -    if (ret != NV_ENC_SUCCESS) {
 -        av_log(avctx, AV_LOG_ERROR, "CreateInputBuffer failed\n");
 -        return AVERROR_UNKNOWN;
 +    nv_status = p_nvenc->nvEncGetEncodePresetConfig(ctx->nvencoder, codec, encoder_preset, &preset_config);
 +    if (nv_status != NV_ENC_SUCCESS) {
 +        av_log(avctx, AV_LOG_FATAL, "GetEncodePresetConfig failed: 0x%x\n", (int)nv_status);
 +        res = AVERROR_EXTERNAL;
 +        goto error;
      }
  
 -    ctx->in[idx].in        = in_buffer.inputBuffer;
 -    ctx->in[idx].format    = in_buffer.bufferFmt;
 -
 -    /* 1MB is large enough to hold most output frames.
 -     * NVENC increases this automaticaly if it's not enough. */
 -    out_buffer.size = BITSTREAM_BUFFER_SIZE;
 -
 -    out_buffer.memoryHeap = NV_ENC_MEMORY_HEAP_SYSMEM_UNCACHED;
 +    ctx->init_encode_params.encodeGUID = codec;
 +    ctx->init_encode_params.encodeHeight = avctx->height;
 +    ctx->init_encode_params.encodeWidth = avctx->width;
  
 -    ret = nv->nvEncCreateBitstreamBuffer(ctx->nvenc_ctx, &out_buffer);
 -    if (ret != NV_ENC_SUCCESS) {
 -        av_log(avctx, AV_LOG_ERROR, "CreateBitstreamBuffer failed\n");
 -        return AVERROR_UNKNOWN;
 +    if (avctx->sample_aspect_ratio.num && avctx->sample_aspect_ratio.den &&
 +        (avctx->sample_aspect_ratio.num != 1 || avctx->sample_aspect_ratio.num != 1)) {
 +        av_reduce(&dw, &dh,
 +                  avctx->width * avctx->sample_aspect_ratio.num,
 +                  avctx->height * avctx->sample_aspect_ratio.den,
 +                  1024 * 1024);
 +        ctx->init_encode_params.darHeight = dh;
 +        ctx->init_encode_params.darWidth = dw;
 +    } else {
 +        ctx->init_encode_params.darHeight = avctx->height;
 +        ctx->init_encode_params.darWidth = avctx->width;
      }
  
 -    ctx->out[idx].out  = out_buffer.bitstreamBuffer;
 -    ctx->out[idx].busy = 0;
 +    // De-compensate for hardware, dubiously, trying to compensate for
 +    // playback at 704 pixel width.
 +    if (avctx->width == 720 &&
 +        (avctx->height == 480 || avctx->height == 576)) {
 +        av_reduce(&dw, &dh,
 +                  ctx->init_encode_params.darWidth * 44,
 +                  ctx->init_encode_params.darHeight * 45,
 +                  1024 * 1024);
 +        ctx->init_encode_params.darHeight = dh;
 +        ctx->init_encode_params.darWidth = dw;
 +    }
  
 -    return 0;
 -}
 +    ctx->init_encode_params.frameRateNum = avctx->time_base.den;
 +    ctx->init_encode_params.frameRateDen = avctx->time_base.num * avctx->ticks_per_frame;
  
 -static int nvenc_setup_surfaces(AVCodecContext *avctx)
 -{
 -    NVENCContext *ctx = avctx->priv_data;
 -    int i, ret;
 +    num_mbs = ((avctx->width + 15) >> 4) * ((avctx->height + 15) >> 4);
 +    ctx->max_surface_count = (num_mbs >= 8160) ? 32 : 48;
  
 -    ctx->nb_surfaces = FFMAX(4 + avctx->max_b_frames,
 -                             ctx->nb_surfaces);
 +    if (ctx->buffer_delay >= ctx->max_surface_count)
 +        ctx->buffer_delay = ctx->max_surface_count - 1;
  
 -    ctx->in = av_mallocz(ctx->nb_surfaces * sizeof(*ctx->in));
 -    if (!ctx->in)
 -        return AVERROR(ENOMEM);
 +    ctx->init_encode_params.enableEncodeAsync = 0;
 +    ctx->init_encode_params.enablePTD = 1;
  
 -    ctx->out = av_mallocz(ctx->nb_surfaces * sizeof(*ctx->out));
 -    if (!ctx->out)
 -        return AVERROR(ENOMEM);
 +    ctx->init_encode_params.presetGUID = encoder_preset;
  
 -    ctx->timestamps = av_fifo_alloc(ctx->nb_surfaces * sizeof(int64_t));
 -    if (!ctx->timestamps)
 -        return AVERROR(ENOMEM);
 -    ctx->pending = av_fifo_alloc(ctx->nb_surfaces * sizeof(ctx->out));
 -    if (!ctx->pending)
 -        return AVERROR(ENOMEM);
 -    ctx->ready = av_fifo_alloc(ctx->nb_surfaces * sizeof(ctx->out));
 -    if (!ctx->ready)
 -        return AVERROR(ENOMEM);
 +    ctx->init_encode_params.encodeConfig = &ctx->encode_config;
 +    memcpy(&ctx->encode_config, &preset_config.presetCfg, sizeof(ctx->encode_config));
 +    ctx->encode_config.version = NV_ENC_CONFIG_VER;
  
 -    for (i = 0; i < ctx->nb_surfaces; i++) {
 -        if ((ret = nvenc_alloc_surface(avctx, i)) < 0)
 -            return ret;
 +    if (avctx->refs >= 0) {
 +        /* 0 means "let the hardware decide" */
 +        switch (avctx->codec->id) {
 +        case AV_CODEC_ID_H264:
 +            ctx->encode_config.encodeCodecConfig.h264Config.maxNumRefFrames = avctx->refs;
 +            break;
 +        case AV_CODEC_ID_H265:
 +            ctx->encode_config.encodeCodecConfig.hevcConfig.maxNumRefFramesInDPB = avctx->refs;
 +            break;
 +        /* Earlier switch/case will return if unknown codec is passed. */
 +        }
      }
  
 -    return 0;
 -}
 +    if (avctx->gop_size > 0) {
 +        if (avctx->max_b_frames >= 0) {
 +            /* 0 is intra-only, 1 is I/P only, 2 is one B Frame, 3 two B frames, and so on. */
 +            ctx->encode_config.frameIntervalP = avctx->max_b_frames + 1;
 +        }
  
 -#define EXTRADATA_SIZE 512
 +        ctx->encode_config.gopLength = avctx->gop_size;
 +        switch (avctx->codec->id) {
 +        case AV_CODEC_ID_H264:
 +            ctx->encode_config.encodeCodecConfig.h264Config.idrPeriod = avctx->gop_size;
 +            break;
 +        case AV_CODEC_ID_H265:
 +            ctx->encode_config.encodeCodecConfig.hevcConfig.idrPeriod = avctx->gop_size;
 +            break;
 +        /* Earlier switch/case will return if unknown codec is passed. */
 +        }
 +    } else if (avctx->gop_size == 0) {
 +        ctx->encode_config.frameIntervalP = 0;
 +        ctx->encode_config.gopLength = 1;
 +        switch (avctx->codec->id) {
 +        case AV_CODEC_ID_H264:
 +            ctx->encode_config.encodeCodecConfig.h264Config.idrPeriod = 1;
 +            break;
 +        case AV_CODEC_ID_H265:
 +            ctx->encode_config.encodeCodecConfig.hevcConfig.idrPeriod = 1;
 +            break;
 +        /* Earlier switch/case will return if unknown codec is passed. */
 +        }
 +    }
  
 -static int nvenc_setup_extradata(AVCodecContext *avctx)
 -{
 -    NVENCContext *ctx                     = avctx->priv_data;
 -    NV_ENCODE_API_FUNCTION_LIST *nv       = &ctx->nvel.nvenc_funcs;
 -    NV_ENC_SEQUENCE_PARAM_PAYLOAD payload = { 0 };
 -    int ret;
 +    /* when there're b frames, set dts offset */
 +    if (ctx->encode_config.frameIntervalP >= 2)
 +        ctx->last_dts = -2;
  
 -    avctx->extradata = av_mallocz(EXTRADATA_SIZE + AV_INPUT_BUFFER_PADDING_SIZE);
 -    if (!avctx->extradata)
 -        return AVERROR(ENOMEM);
 +    if (avctx->bit_rate > 0) {
 +        ctx->encode_config.rcParams.averageBitRate = avctx->bit_rate;
 +    } else if (ctx->encode_config.rcParams.averageBitRate > 0) {
 +        ctx->encode_config.rcParams.maxBitRate = ctx->encode_config.rcParams.averageBitRate;
 +    }
  
 -    payload.version              = NV_ENC_SEQUENCE_PARAM_PAYLOAD_VER;
 -    payload.spsppsBuffer         = avctx->extradata;
 -    payload.inBufferSize         = EXTRADATA_SIZE;
 -    payload.outSPSPPSPayloadSize = &avctx->extradata_size;
 +    if (avctx->rc_max_rate > 0)
 +        ctx->encode_config.rcParams.maxBitRate = avctx->rc_max_rate;
 +
 +    if (lossless) {
 +        if (avctx->codec->id == AV_CODEC_ID_H264)
 +            ctx->encode_config.encodeCodecConfig.h264Config.qpPrimeYZeroTransformBypassFlag = 1;
 +
 +        ctx->encode_config.rcParams.rateControlMode = NV_ENC_PARAMS_RC_CONSTQP;
 +        ctx->encode_config.rcParams.constQP.qpInterB = 0;
 +        ctx->encode_config.rcParams.constQP.qpInterP = 0;
 +        ctx->encode_config.rcParams.constQP.qpIntra = 0;
 +
 +        avctx->qmin = -1;
 +        avctx->qmax = -1;
 +    } else if (ctx->cbr) {
 +        if (!ctx->twopass) {
 +            ctx->encode_config.rcParams.rateControlMode = NV_ENC_PARAMS_RC_CBR;
 +        } else {
 +            ctx->encode_config.rcParams.rateControlMode = NV_ENC_PARAMS_RC_2_PASS_QUALITY;
  
 -    ret = nv->nvEncGetSequenceParams(ctx->nvenc_ctx, &payload);
 -    if (ret != NV_ENC_SUCCESS) {
 -        av_log(avctx, AV_LOG_ERROR, "Cannot get the extradata\n");
 -        return AVERROR_UNKNOWN;
 -    }
 +            if (avctx->codec->id == AV_CODEC_ID_H264) {
 +                ctx->encode_config.encodeCodecConfig.h264Config.adaptiveTransformMode = NV_ENC_H264_ADAPTIVE_TRANSFORM_ENABLE;
 +                ctx->encode_config.encodeCodecConfig.h264Config.fmoMode = NV_ENC_H264_FMO_DISABLE;
 +            }
 +        }
 +    } else if (avctx->global_quality > 0) {
 +        ctx->encode_config.rcParams.rateControlMode = NV_ENC_PARAMS_RC_CONSTQP;
 +        ctx->encode_config.rcParams.constQP.qpInterB = avctx->global_quality;
 +        ctx->encode_config.rcParams.constQP.qpInterP = avctx->global_quality;
 +        ctx->encode_config.rcParams.constQP.qpIntra = avctx->global_quality;
  
 -    return 0;
 -}
 +        avctx->qmin = -1;
 +        avctx->qmax = -1;
 +    } else {
 +        if (avctx->qmin >= 0 && avctx->qmax >= 0) {
 +            ctx->encode_config.rcParams.enableMinQP = 1;
 +            ctx->encode_config.rcParams.enableMaxQP = 1;
 +
 +            ctx->encode_config.rcParams.minQP.qpInterB = avctx->qmin;
 +            ctx->encode_config.rcParams.minQP.qpInterP = avctx->qmin;
 +            ctx->encode_config.rcParams.minQP.qpIntra = avctx->qmin;
 +
 +            ctx->encode_config.rcParams.maxQP.qpInterB = avctx->qmax;
 +            ctx->encode_config.rcParams.maxQP.qpInterP = avctx->qmax;
 +            ctx->encode_config.rcParams.maxQP.qpIntra = avctx->qmax;
 +
 +            qp_inter_p = (avctx->qmax + 3 * avctx->qmin) / 4; // biased towards Qmin
 +
 +            if (ctx->twopass) {
 +                ctx->encode_config.rcParams.rateControlMode = NV_ENC_PARAMS_RC_2_PASS_VBR;
 +                if (avctx->codec->id == AV_CODEC_ID_H264) {
 +                    ctx->encode_config.encodeCodecConfig.h264Config.adaptiveTransformMode = NV_ENC_H264_ADAPTIVE_TRANSFORM_ENABLE;
 +                    ctx->encode_config.encodeCodecConfig.h264Config.fmoMode = NV_ENC_H264_FMO_DISABLE;
 +                }
 +            } else {
 +                ctx->encode_config.rcParams.rateControlMode = NV_ENC_PARAMS_RC_VBR_MINQP;
 +            }
 +        } else {
 +            qp_inter_p = 26; // default to 26
  
 -av_cold int ff_nvenc_encode_close(AVCodecContext *avctx)
 -{
 -    NVENCContext *ctx               = avctx->priv_data;
 -    NV_ENCODE_API_FUNCTION_LIST *nv = &ctx->nvel.nvenc_funcs;
 -    int i;
 +            if (ctx->twopass) {
 +                ctx->encode_config.rcParams.rateControlMode = NV_ENC_PARAMS_RC_2_PASS_VBR;
 +            } else {
 +                ctx->encode_config.rcParams.rateControlMode = NV_ENC_PARAMS_RC_VBR;
 +            }
 +        }
  
 -    av_fifo_free(ctx->timestamps);
 -    av_fifo_free(ctx->pending);
 -    av_fifo_free(ctx->ready);
 +        ctx->encode_config.rcParams.enableInitialRCQP = 1;
 +        ctx->encode_config.rcParams.initialRCQP.qpInterP  = qp_inter_p;
  
 -    if (ctx->in) {
 -        for (i = 0; i < ctx->nb_surfaces; ++i) {
 -            nv->nvEncDestroyInputBuffer(ctx->nvenc_ctx, ctx->in[i].in);
 -            nv->nvEncDestroyBitstreamBuffer(ctx->nvenc_ctx, ctx->out[i].out);
 +        if(avctx->i_quant_factor != 0.0 && avctx->b_quant_factor != 0.0) {
 +            ctx->encode_config.rcParams.initialRCQP.qpIntra = av_clip(
 +                qp_inter_p * fabs(avctx->i_quant_factor) + avctx->i_quant_offset, 0, 51);
 +            ctx->encode_config.rcParams.initialRCQP.qpInterB = av_clip(
 +                qp_inter_p * fabs(avctx->b_quant_factor) + avctx->b_quant_offset, 0, 51);
 +        } else {
 +            ctx->encode_config.rcParams.initialRCQP.qpIntra = qp_inter_p;
 +            ctx->encode_config.rcParams.initialRCQP.qpInterB = qp_inter_p;
          }
      }
  
 -    av_freep(&ctx->in);
 -    av_freep(&ctx->out);
 +    if (avctx->rc_buffer_size > 0) {
 +        ctx->encode_config.rcParams.vbvBufferSize = avctx->rc_buffer_size;
 +    } else if (ctx->encode_config.rcParams.averageBitRate > 0) {
 +        ctx->encode_config.rcParams.vbvBufferSize = 2 * ctx->encode_config.rcParams.averageBitRate;
 +    }
  
 -    if (ctx->nvenc_ctx)
 -        nv->nvEncDestroyEncoder(ctx->nvenc_ctx);
 +    if (avctx->flags & AV_CODEC_FLAG_INTERLACED_DCT) {
 +        ctx->encode_config.frameFieldMode = NV_ENC_PARAMS_FRAME_FIELD_MODE_FIELD;
 +    } else {
 +        ctx->encode_config.frameFieldMode = NV_ENC_PARAMS_FRAME_FIELD_MODE_FRAME;
 +    }
  
 -    if (ctx->cu_context)
 -        ctx->nvel.cu_ctx_destroy(ctx->cu_context);
 +    switch (avctx->codec->id) {
 +    case AV_CODEC_ID_H264:
 +        ctx->encode_config.encodeCodecConfig.h264Config.h264VUIParameters.colourDescriptionPresentFlag = 1;
 +        ctx->encode_config.encodeCodecConfig.h264Config.h264VUIParameters.videoSignalTypePresentFlag = 1;
 +
 +        ctx->encode_config.encodeCodecConfig.h264Config.h264VUIParameters.colourMatrix = avctx->colorspace;
 +        ctx->encode_config.encodeCodecConfig.h264Config.h264VUIParameters.colourPrimaries = avctx->color_primaries;
 +        ctx->encode_config.encodeCodecConfig.h264Config.h264VUIParameters.transferCharacteristics = avctx->color_trc;
 +
 +        ctx->encode_config.encodeCodecConfig.h264Config.h264VUIParameters.videoFullRangeFlag = avctx->color_range == AVCOL_RANGE_JPEG;
 +
 +        ctx->encode_config.encodeCodecConfig.h264Config.sliceMode = 3;
 +        ctx->encode_config.encodeCodecConfig.h264Config.sliceModeData = 1;
 +
 +        ctx->encode_config.encodeCodecConfig.h264Config.disableSPSPPS = (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) ? 1 : 0;
 +        ctx->encode_config.encodeCodecConfig.h264Config.repeatSPSPPS = (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) ? 0 : 1;
 +
 +        if (!ctx->profile) {
 +            switch (avctx->profile) {
 +            case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
 +                ctx->encode_config.profileGUID = NV_ENC_H264_PROFILE_HIGH_444_GUID;
 +                break;
 +            case FF_PROFILE_H264_BASELINE:
 +                ctx->encode_config.profileGUID = NV_ENC_H264_PROFILE_BASELINE_GUID;
 +                break;
 +            case FF_PROFILE_H264_MAIN:
 +                ctx->encode_config.profileGUID = NV_ENC_H264_PROFILE_MAIN_GUID;
 +                break;
 +            case FF_PROFILE_H264_HIGH:
 +            case FF_PROFILE_UNKNOWN:
 +                ctx->encode_config.profileGUID = NV_ENC_H264_PROFILE_HIGH_GUID;
 +                break;
 +            default:
 +                av_log(avctx, AV_LOG_WARNING, "Unsupported profile requested, falling back to high\n");
 +                ctx->encode_config.profileGUID = NV_ENC_H264_PROFILE_HIGH_GUID;
 +                break;
 +            }
 +        } else {
 +            if (!strcmp(ctx->profile, "high")) {
 +                ctx->encode_config.profileGUID = NV_ENC_H264_PROFILE_HIGH_GUID;
 +                avctx->profile = FF_PROFILE_H264_HIGH;
 +            } else if (!strcmp(ctx->profile, "main")) {
 +                ctx->encode_config.profileGUID = NV_ENC_H264_PROFILE_MAIN_GUID;
 +                avctx->profile = FF_PROFILE_H264_MAIN;
 +            } else if (!strcmp(ctx->profile, "baseline")) {
 +                ctx->encode_config.profileGUID = NV_ENC_H264_PROFILE_BASELINE_GUID;
 +                avctx->profile = FF_PROFILE_H264_BASELINE;
 +            } else if (!strcmp(ctx->profile, "high444p")) {
 +                ctx->encode_config.profileGUID = NV_ENC_H264_PROFILE_HIGH_444_GUID;
 +                avctx->profile = FF_PROFILE_H264_HIGH_444_PREDICTIVE;
 +            } else {
 +                av_log(avctx, AV_LOG_FATAL, "Profile \"%s\" is unknown! Supported profiles: high, main, baseline\n", ctx->profile);
 +                res = AVERROR(EINVAL);
 +                goto error;
 +            }
 +        }
  
 -    if (ctx->nvel.nvenc)
 -        dlclose(ctx->nvel.nvenc);
 +        // force setting profile as high444p if input is AV_PIX_FMT_YUV444P
 +        if (avctx->pix_fmt == AV_PIX_FMT_YUV444P) {
 +            ctx->encode_config.profileGUID = NV_ENC_H264_PROFILE_HIGH_444_GUID;
 +            avctx->profile = FF_PROFILE_H264_HIGH_444_PREDICTIVE;
 +        }
  
 -    if (ctx->nvel.cuda)
 -        dlclose(ctx->nvel.cuda);
 +        ctx->encode_config.encodeCodecConfig.h264Config.chromaFormatIDC = avctx->profile == FF_PROFILE_H264_HIGH_444_PREDICTIVE ? 3 : 1;
  
 -    return 0;
 -}
 +        if (ctx->level) {
 +            res = input_string_to_uint32(avctx, nvenc_h264_level_pairs, ctx->level, &ctx->encode_config.encodeCodecConfig.h264Config.level);
  
 -av_cold int ff_nvenc_encode_init(AVCodecContext *avctx)
 -{
 -    int ret;
 +            if (res) {
 +                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);
 +                goto error;
 +            }
 +        } else {
 +            ctx->encode_config.encodeCodecConfig.h264Config.level = NV_ENC_LEVEL_AUTOSELECT;
 +        }
  
 -    if ((ret = nvenc_load_libraries(avctx)) < 0)
 -        return ret;
 +        break;
 +    case AV_CODEC_ID_H265:
 +        ctx->encode_config.encodeCodecConfig.hevcConfig.sliceMode = 3;
 +        ctx->encode_config.encodeCodecConfig.hevcConfig.sliceModeData = 1;
  
 -    if ((ret = nvenc_setup_device(avctx)) < 0)
 -        return ret;
 +        ctx->encode_config.encodeCodecConfig.hevcConfig.disableSPSPPS = (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) ? 1 : 0;
 +        ctx->encode_config.encodeCodecConfig.hevcConfig.repeatSPSPPS = (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) ? 0 : 1;
  
 -    if ((ret = nvenc_setup_encoder(avctx)) < 0)
 -        return ret;
 +        /* No other profile is supported in the current SDK version 5 */
 +        ctx->encode_config.profileGUID = NV_ENC_HEVC_PROFILE_MAIN_GUID;
 +        avctx->profile = FF_PROFILE_HEVC_MAIN;
  
 -    if ((ret = nvenc_setup_surfaces(avctx)) < 0)
 -        return ret;
 +        if (ctx->level) {
 +            res = input_string_to_uint32(avctx, nvenc_hevc_level_pairs, ctx->level, &ctx->encode_config.encodeCodecConfig.hevcConfig.level);
  
 -    if (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) {
 -        if ((ret = nvenc_setup_extradata(avctx)) < 0)
 -            return ret;
 -    }
 +            if (res) {
 +                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);
 +                goto error;
 +            }
 +        } else {
 +            ctx->encode_config.encodeCodecConfig.hevcConfig.level = NV_ENC_LEVEL_AUTOSELECT;
 +        }
  
 -    return 0;
 -}
 +        if (ctx->tier) {
 +            if (!strcmp(ctx->tier, "main")) {
 +                ctx->encode_config.encodeCodecConfig.hevcConfig.tier = NV_ENC_TIER_HEVC_MAIN;
 +            } else if (!strcmp(ctx->tier, "high")) {
 +                ctx->encode_config.encodeCodecConfig.hevcConfig.tier = NV_ENC_TIER_HEVC_HIGH;
 +            } else {
 +                av_log(avctx, AV_LOG_FATAL, "Tier \"%s\" is unknown! Supported tiers: main, high\n", ctx->tier);
 +                res = AVERROR(EINVAL);
 +                goto error;
 +            }
 +        }
  
 -static NVENCInputSurface *get_input_surface(NVENCContext *ctx)
 -{
 -    int i;
 +        break;
 +    /* Earlier switch/case will return if unknown codec is passed. */
 +    }
  
 -    for (i = 0; i < ctx->nb_surfaces; i++) {
 -        if (!ctx->in[i].locked) {
 -            ctx->in[i].locked = 1;
 -            return &ctx->in[i];
 -        }
 +    nv_status = p_nvenc->nvEncInitializeEncoder(ctx->nvencoder, &ctx->init_encode_params);
 +    if (nv_status != NV_ENC_SUCCESS) {
 +        av_log(avctx, AV_LOG_FATAL, "InitializeEncoder failed: 0x%x\n", (int)nv_status);
 +        res = AVERROR_EXTERNAL;
 +        goto error;
      }
  
 -    return NULL;
 -}
 +    ctx->input_surfaces = av_malloc(ctx->max_surface_count * sizeof(*ctx->input_surfaces));
  
 -static NVENCOutputSurface *get_output_surface(NVENCContext *ctx)
 -{
 -    int i;
 +    if (!ctx->input_surfaces) {
 +        res = AVERROR(ENOMEM);
 +        goto error;
 +    }
  
 -    for (i = 0; i < ctx->nb_surfaces; i++) {
 -        if (!ctx->out[i].busy) {
 -            return &ctx->out[i];
 -        }
 +    ctx->output_surfaces = av_malloc(ctx->max_surface_count * sizeof(*ctx->output_surfaces));
 +
 +    if (!ctx->output_surfaces) {
 +        res = AVERROR(ENOMEM);
 +        goto error;
      }
  
 -    return NULL;
 -}
 +    for (surfaceCount = 0; surfaceCount < ctx->max_surface_count; ++surfaceCount) {
 +        NV_ENC_CREATE_INPUT_BUFFER allocSurf = { 0 };
 +        NV_ENC_CREATE_BITSTREAM_BUFFER allocOut = { 0 };
 +        allocSurf.version = NV_ENC_CREATE_INPUT_BUFFER_VER;
 +        allocOut.version = NV_ENC_CREATE_BITSTREAM_BUFFER_VER;
  
 -static int nvenc_copy_frame(NV_ENC_LOCK_INPUT_BUFFER *in, const AVFrame *frame)
 -{
 -    uint8_t *buf = in->bufferDataPtr;
 -    int off      = frame->height * in->pitch;
 +        allocSurf.width = (avctx->width + 31) & ~31;
 +        allocSurf.height = (avctx->height + 31) & ~31;
  
 -    switch (frame->format) {
 -    case AV_PIX_FMT_YUV420P:
 -        av_image_copy_plane(buf, in->pitch,
 -                            frame->data[0], frame->linesize[0],
 -                            frame->width, frame->height);
 -        buf += off;
 +        allocSurf.memoryHeap = NV_ENC_MEMORY_HEAP_SYSMEM_CACHED;
  
 -        av_image_copy_plane(buf, in->pitch >> 1,
 -                            frame->data[2], frame->linesize[2],
 -                            frame->width >> 1, frame->height >> 1);
 +        switch (avctx->pix_fmt) {
 +        case AV_PIX_FMT_YUV420P:
 +            allocSurf.bufferFmt = NV_ENC_BUFFER_FORMAT_YV12_PL;
 +            break;
  
 -        buf += off >> 2;
 +        case AV_PIX_FMT_NV12:
 +            allocSurf.bufferFmt = NV_ENC_BUFFER_FORMAT_NV12_PL;
 +            break;
  
 -        av_image_copy_plane(buf, in->pitch >> 1,
 -                            frame->data[1], frame->linesize[1],
 -                            frame->width >> 1, frame->height >> 1);
 -        break;
 -    case AV_PIX_FMT_NV12:
 -        av_image_copy_plane(buf, in->pitch,
 -                            frame->data[0], frame->linesize[0],
 -                            frame->width, frame->height);
 -        buf += off;
 -
 -        av_image_copy_plane(buf, in->pitch,
 -                            frame->data[1], frame->linesize[1],
 -                            frame->width, frame->height >> 1);
 -        break;
 -    case AV_PIX_FMT_YUV444P:
 -        av_image_copy_plane(buf, in->pitch,
 -                            frame->data[0], frame->linesize[0],
 -                            frame->width, frame->height);
 -        buf += off;
 -
 -        av_image_copy_plane(buf, in->pitch,
 -                            frame->data[1], frame->linesize[1],
 -                            frame->width, frame->height);
 -        buf += off;
 -
 -        av_image_copy_plane(buf, in->pitch,
 -                            frame->data[2], frame->linesize[2],
 -                            frame->width, frame->height);
 -        break;
 -    default:
 -        return AVERROR_BUG;
 -    }
 +        case AV_PIX_FMT_YUV444P:
 +            allocSurf.bufferFmt = NV_ENC_BUFFER_FORMAT_YUV444_PL;
 +            break;
  
 -    return 0;
 -}
 +        default:
 +            av_log(avctx, AV_LOG_FATAL, "Invalid input pixel format\n");
 +            res = AVERROR(EINVAL);
 +            goto error;
 +        }
  
 -static int nvenc_enqueue_frame(AVCodecContext *avctx, const AVFrame *frame,
 -                               NVENCInputSurface **in_surf)
 -{
 -    NVENCContext *ctx               = avctx->priv_data;
 -    NV_ENCODE_API_FUNCTION_LIST *nv = &ctx->nvel.nvenc_funcs;
 -    NV_ENC_LOCK_INPUT_BUFFER params = { 0 };
 -    NVENCInputSurface *in           = get_input_surface(ctx);
 -    int ret;
 +        nv_status = p_nvenc->nvEncCreateInputBuffer(ctx->nvencoder, &allocSurf);
 +        if (nv_status != NV_ENC_SUCCESS) {
 +            av_log(avctx, AV_LOG_FATAL, "CreateInputBuffer failed\n");
 +            res = AVERROR_EXTERNAL;
 +            goto error;
 +        }
 +
 +        ctx->input_surfaces[surfaceCount].lockCount = 0;
 +        ctx->input_surfaces[surfaceCount].input_surface = allocSurf.inputBuffer;
 +        ctx->input_surfaces[surfaceCount].format = allocSurf.bufferFmt;
 +        ctx->input_surfaces[surfaceCount].width = allocSurf.width;
 +        ctx->input_surfaces[surfaceCount].height = allocSurf.height;
  
 -    if (!in)
 -        return AVERROR_BUG;
 +        /* 1MB is large enough to hold most output frames. NVENC increases this automaticaly if it's not enough. */
 +        allocOut.size = 1024 * 1024;
  
 -    params.version     = NV_ENC_LOCK_INPUT_BUFFER_VER;
 -    params.inputBuffer = in->in;
 +        allocOut.memoryHeap = NV_ENC_MEMORY_HEAP_SYSMEM_CACHED;
  
 +        nv_status = p_nvenc->nvEncCreateBitstreamBuffer(ctx->nvencoder, &allocOut);
 +        if (nv_status != NV_ENC_SUCCESS) {
 +            av_log(avctx, AV_LOG_FATAL, "CreateBitstreamBuffer failed\n");
 +            ctx->output_surfaces[surfaceCount++].output_surface = NULL;
 +            res = AVERROR_EXTERNAL;
 +            goto error;
 +        }
  
 -    ret = nv->nvEncLockInputBuffer(ctx->nvenc_ctx, &params);
 -    if (ret != NV_ENC_SUCCESS) {
 -        av_log(avctx, AV_LOG_ERROR, "Cannot lock the buffer %p.\n",
 -               in);
 -        return AVERROR_UNKNOWN;
 +        ctx->output_surfaces[surfaceCount].output_surface = allocOut.bitstreamBuffer;
 +        ctx->output_surfaces[surfaceCount].size = allocOut.size;
 +        ctx->output_surfaces[surfaceCount].busy = 0;
      }
  
 -    ret = nvenc_copy_frame(&params, frame);
 -    if (ret < 0)
 -        goto fail;
 +    if (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) {
 +        uint32_t outSize = 0;
 +        char tmpHeader[256];
 +        NV_ENC_SEQUENCE_PARAM_PAYLOAD payload = { 0 };
 +        payload.version = NV_ENC_SEQUENCE_PARAM_PAYLOAD_VER;
 +
 +        payload.spsppsBuffer = tmpHeader;
 +        payload.inBufferSize = sizeof(tmpHeader);
 +        payload.outSPSPPSPayloadSize = &outSize;
 +
 +        nv_status = p_nvenc->nvEncGetSequenceParams(ctx->nvencoder, &payload);
 +        if (nv_status != NV_ENC_SUCCESS) {
 +            av_log(avctx, AV_LOG_FATAL, "GetSequenceParams failed\n");
 +            goto error;
 +        }
  
 -    ret = nv->nvEncUnlockInputBuffer(ctx->nvenc_ctx, in->in);
 -    if (ret != NV_ENC_SUCCESS) {
 -        av_log(avctx, AV_LOG_ERROR, "Cannot unlock the buffer %p.\n",
 -               in);
 -        return AVERROR_UNKNOWN;
 +        avctx->extradata_size = outSize;
 +        avctx->extradata = av_mallocz(outSize + AV_INPUT_BUFFER_PADDING_SIZE);
 +
 +        if (!avctx->extradata) {
 +            res = AVERROR(ENOMEM);
 +            goto error;
 +        }
 +
 +        memcpy(avctx->extradata, tmpHeader, outSize);
      }
  
 -    *in_surf = in;
 +    if (ctx->encode_config.frameIntervalP > 1)
 +        avctx->has_b_frames = 2;
  
 -    return 0;
 +    if (ctx->encode_config.rcParams.averageBitRate > 0)
 +        avctx->bit_rate = ctx->encode_config.rcParams.averageBitRate;
  
 -fail:
 -    nv->nvEncUnlockInputBuffer(ctx->nvenc_ctx, in->in);
++    cpb_props = ff_add_cpb_side_data(avctx);
++    if (!cpb_props)
++        return AVERROR(ENOMEM);
++    cpb_props->max_bitrate = ctx->encode_config.rcParams.maxBitRate;
++    cpb_props->avg_bitrate = avctx->bit_rate;
++    cpb_props->buffer_size = ctx->encode_config.rcParams.vbvBufferSize;
+ 
 -    return ret;
 -}
 +    return 0;
  
 -static void nvenc_codec_specific_pic_params(AVCodecContext *avctx,
 -                                            NV_ENC_PIC_PARAMS *params)
 -{
 -    NVENCContext *ctx = avctx->priv_data;
 +error:
  
 -    switch (avctx->codec->id) {
 -    case AV_CODEC_ID_H264:
 -        params->codecPicParams.h264PicParams.sliceMode =
 -            ctx->config.encodeCodecConfig.h264Config.sliceMode;
 -        params->codecPicParams.h264PicParams.sliceModeData =
 -            ctx->config.encodeCodecConfig.h264Config.sliceModeData;
 -        break;
 -    case AV_CODEC_ID_HEVC:
 -        params->codecPicParams.hevcPicParams.sliceMode =
 -            ctx->config.encodeCodecConfig.hevcConfig.sliceMode;
 -        params->codecPicParams.hevcPicParams.sliceModeData =
 -            ctx->config.encodeCodecConfig.hevcConfig.sliceModeData;
 -        break;
 +    for (i = 0; i < surfaceCount; ++i) {
 +        p_nvenc->nvEncDestroyInputBuffer(ctx->nvencoder, ctx->input_surfaces[i].input_surface);
 +        if (ctx->output_surfaces[i].output_surface)
 +            p_nvenc->nvEncDestroyBitstreamBuffer(ctx->nvencoder, ctx->output_surfaces[i].output_surface);
      }
 -}
  
 -static inline int nvenc_enqueue_timestamp(AVFifoBuffer *f, int64_t pts)
 -{
 -    return av_fifo_generic_write(f, &pts, sizeof(pts), NULL);
 -}
 +    if (ctx->nvencoder)
 +        p_nvenc->nvEncDestroyEncoder(ctx->nvencoder);
  
 -static inline int nvenc_dequeue_timestamp(AVFifoBuffer *f, int64_t *pts)
 -{
 -    return av_fifo_generic_read(f, pts, sizeof(*pts), NULL);
 -}
 +    if (ctx->cu_context)
 +        dl_fn->cu_ctx_destroy(ctx->cu_context);
  
 -static inline int nvenc_enqueue_surface(AVFifoBuffer *f,
 -                                        NVENCOutputSurface *surf)
 -{
 -    surf->busy = 1;
 -    return av_fifo_generic_write(f, &surf, sizeof(surf), NULL);
 -}
 +    nvenc_unload_nvenc(avctx);
  
 -static inline int nvenc_dequeue_surface(AVFifoBuffer *f,
 -                                        NVENCOutputSurface **surf)
 -{
 -    return av_fifo_generic_read(f, surf, sizeof(*surf), NULL);
 +    ctx->nvencoder = NULL;
 +    ctx->cu_context = NULL;
 +
 +    return res;
  }
  
 -static int nvenc_set_timestamp(NVENCContext *ctx,
 -                               NV_ENC_LOCK_BITSTREAM *params,
 -                               AVPacket *pkt)
 +static av_cold int nvenc_encode_close(AVCodecContext *avctx)
  {
 -    pkt->pts      = params->outputTimeStamp;
 -    pkt->duration = params->outputDuration;
 +    NvencContext *ctx = avctx->priv_data;
 +    NvencDynLoadFunctions *dl_fn = &ctx->nvenc_dload_funcs;
 +    NV_ENCODE_API_FUNCTION_LIST *p_nvenc = &dl_fn->nvenc_funcs;
 +    int i;
 +
 +    av_freep(&ctx->timestamp_list.data);
 +    av_freep(&ctx->output_surface_ready_queue.data);
 +    av_freep(&ctx->output_surface_queue.data);
  
 -    return nvenc_dequeue_timestamp(ctx->timestamps, &pkt->dts);
 +    for (i = 0; i < ctx->max_surface_count; ++i) {
 +        p_nvenc->nvEncDestroyInputBuffer(ctx->nvencoder, ctx->input_surfaces[i].input_surface);
 +        p_nvenc->nvEncDestroyBitstreamBuffer(ctx->nvencoder, ctx->output_surfaces[i].output_surface);
 +    }
 +    ctx->max_surface_count = 0;
 +
 +    p_nvenc->nvEncDestroyEncoder(ctx->nvencoder);
 +    ctx->nvencoder = NULL;
 +
 +    dl_fn->cu_ctx_destroy(ctx->cu_context);
 +    ctx->cu_context = NULL;
 +
 +    nvenc_unload_nvenc(avctx);
 +
 +    return 0;
  }
  
 -static int nvenc_get_frame(AVCodecContext *avctx, AVPacket *pkt)
 +static int process_output_surface(AVCodecContext *avctx, AVPacket *pkt, NvencOutputSurface *tmpoutsurf)
  {
 -    NVENCContext *ctx               = avctx->priv_data;
 -    NV_ENCODE_API_FUNCTION_LIST *nv = &ctx->nvel.nvenc_funcs;
 -    NV_ENC_LOCK_BITSTREAM params    = { 0 };
 -    NVENCOutputSurface *out         = NULL;
 -    int ret;
 +    NvencContext *ctx = avctx->priv_data;
 +    NvencDynLoadFunctions *dl_fn = &ctx->nvenc_dload_funcs;
 +    NV_ENCODE_API_FUNCTION_LIST *p_nvenc = &dl_fn->nvenc_funcs;
  
 -    ret = nvenc_dequeue_surface(ctx->pending, &out);
 -    if (ret)
 -        return ret;
 +    uint32_t slice_mode_data;
 +    uint32_t *slice_offsets;
 +    NV_ENC_LOCK_BITSTREAM lock_params = { 0 };
 +    NVENCSTATUS nv_status;
 +    int res = 0;
  
 -    params.version         = NV_ENC_LOCK_BITSTREAM_VER;
 -    params.outputBitstream = out->out;
 +    switch (avctx->codec->id) {
 +    case AV_CODEC_ID_H264:
 +      slice_mode_data = ctx->encode_config.encodeCodecConfig.h264Config.sliceModeData;
 +      break;
 +    case AV_CODEC_ID_H265:
 +      slice_mode_data = ctx->encode_config.encodeCodecConfig.hevcConfig.sliceModeData;
 +      break;
 +    default:
 +      av_log(avctx, AV_LOG_ERROR, "Unknown codec name\n");
 +      res = AVERROR(EINVAL);
 +      goto error;
 +    }
 +    slice_offsets = av_mallocz(slice_mode_data * sizeof(*slice_offsets));
  
 -    ret = nv->nvEncLockBitstream(ctx->nvenc_ctx, &params);
 -    if (ret < 0)
 -        return AVERROR_UNKNOWN;
 +    if (!slice_offsets)
 +        return AVERROR(ENOMEM);
 +
 +    lock_params.version = NV_ENC_LOCK_BITSTREAM_VER;
  
 -    ret = ff_alloc_packet(pkt, params.bitstreamSizeInBytes);
 -    if (ret < 0)
 -        return ret;
 +    lock_params.doNotWait = 0;
 +    lock_params.outputBitstream = tmpoutsurf->output_surface;
 +    lock_params.sliceOffsets = slice_offsets;
  
 -    memcpy(pkt->data, params.bitstreamBufferPtr, pkt->size);
 +    nv_status = p_nvenc->nvEncLockBitstream(ctx->nvencoder, &lock_params);
 +    if (nv_status != NV_ENC_SUCCESS) {
 +        av_log(avctx, AV_LOG_ERROR, "Failed locking bitstream buffer\n");
 +        res = AVERROR_EXTERNAL;
 +        goto error;
 +    }
  
 -    ret = nv->nvEncUnlockBitstream(ctx->nvenc_ctx, out->out);
 -    if (ret < 0)
 -        return AVERROR_UNKNOWN;
 +    if (res = ff_alloc_packet2(avctx, pkt, lock_params.bitstreamSizeInBytes,0)) {
 +        p_nvenc->nvEncUnlockBitstream(ctx->nvencoder, tmpoutsurf->output_surface);
 +        goto error;
 +    }
  
 -    out->busy = out->in->locked = 0;
 +    memcpy(pkt->data, lock_params.bitstreamBufferPtr, lock_params.bitstreamSizeInBytes);
  
 -    ret = nvenc_set_timestamp(ctx, &params, pkt);
 -    if (ret < 0)
 -        return ret;
 +    nv_status = p_nvenc->nvEncUnlockBitstream(ctx->nvencoder, tmpoutsurf->output_surface);
 +    if (nv_status != NV_ENC_SUCCESS)
 +        av_log(avctx, AV_LOG_ERROR, "Failed unlocking bitstream buffer, expect the gates of mordor to open\n");
  
 -    switch (params.pictureType) {
 +    switch (lock_params.pictureType) {
      case NV_ENC_PIC_TYPE_IDR:
          pkt->flags |= AV_PKT_FLAG_KEY;
  #if FF_API_CODED_FRAME