X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Fnvdec.c;h=251be039a833c30f5a6769febc4377b14340631c;hb=ef6a9e5e311f09fa8032974fa4d0c1e166a959bb;hp=b60da24301dffbf2849a885f2dc01a713c4a0c7c;hpb=493240a522fca34882601fbeeda4e17aa40a0303;p=ffmpeg diff --git a/libavcodec/nvdec.c b/libavcodec/nvdec.c index b60da24301d..251be039a83 100644 --- a/libavcodec/nvdec.c +++ b/libavcodec/nvdec.c @@ -44,6 +44,7 @@ typedef struct NVDECDecoder { CUvideodecoder decoder; AVBufferRef *hw_device_ref; + AVBufferRef *real_hw_frames_ref; CUcontext cuda_ctx; CUstream stream; @@ -61,6 +62,9 @@ typedef struct NVDECFramePool { static int map_avcodec_id(enum AVCodecID id) { switch (id) { +#if CONFIG_AV1_NVDEC_HWACCEL + case AV_CODEC_ID_AV1: return cudaVideoCodec_AV1; +#endif case AV_CODEC_ID_H264: return cudaVideoCodec_H264; case AV_CODEC_ID_HEVC: return cudaVideoCodec_HEVC; case AV_CODEC_ID_MJPEG: return cudaVideoCodec_JPEG; @@ -79,6 +83,9 @@ static int map_chroma_format(enum AVPixelFormat pix_fmt) { int shift_h = 0, shift_v = 0; + if (av_pix_fmt_count_planes(pix_fmt) == 1) + return cudaVideoChromaFormat_Monochrome; + av_pix_fmt_get_chroma_sub_sample(pix_fmt, &shift_h, &shift_v); if (shift_h == 1 && shift_v == 1) @@ -163,6 +170,7 @@ static void nvdec_decoder_free(void *opaque, uint8_t *data) CHECK_CU(decoder->cudl->cuCtxPopCurrent(&dummy)); } + av_buffer_unref(&decoder->real_hw_frames_ref); av_buffer_unref(&decoder->hw_device_ref); cuvid_free_functions(&decoder->cvdl); @@ -234,7 +242,7 @@ fail: return ret; } -static AVBufferRef *nvdec_decoder_frame_alloc(void *opaque, int size) +static AVBufferRef *nvdec_decoder_frame_alloc(void *opaque, size_t size) { NVDECFramePool *pool = opaque; AVBufferRef *ret; @@ -256,6 +264,7 @@ int ff_nvdec_decode_uninit(AVCodecContext *avctx) NVDECContext *ctx = avctx->internal->hwaccel_priv_data; av_freep(&ctx->bitstream); + av_freep(&ctx->bitstream_internal); ctx->bitstream_len = 0; ctx->bitstream_allocated = 0; @@ -269,10 +278,61 @@ int ff_nvdec_decode_uninit(AVCodecContext *avctx) return 0; } +static void nvdec_free_dummy(struct AVHWFramesContext *ctx) +{ + av_buffer_pool_uninit(&ctx->pool); +} + +static AVBufferRef *nvdec_alloc_dummy(size_t size) +{ + return av_buffer_create(NULL, 0, NULL, NULL, 0); +} + +static int nvdec_init_hwframes(AVCodecContext *avctx, AVBufferRef **out_frames_ref, int dummy) +{ + AVHWFramesContext *frames_ctx; + int ret; + + ret = avcodec_get_hw_frames_parameters(avctx, + avctx->hw_device_ctx, + avctx->hwaccel->pix_fmt, + out_frames_ref); + if (ret < 0) + return ret; + + frames_ctx = (AVHWFramesContext*)(*out_frames_ref)->data; + + if (dummy) { + // Copied from ff_decode_get_hw_frames_ctx for compatibility + frames_ctx->initial_pool_size += 3; + + frames_ctx->free = nvdec_free_dummy; + frames_ctx->pool = av_buffer_pool_init(0, nvdec_alloc_dummy); + + if (!frames_ctx->pool) { + av_buffer_unref(out_frames_ref); + return AVERROR(ENOMEM); + } + } else { + // This is normally not used to actually allocate frames from + frames_ctx->initial_pool_size = 0; + } + + ret = av_hwframe_ctx_init(*out_frames_ref); + if (ret < 0) { + av_buffer_unref(out_frames_ref); + return ret; + } + + return 0; +} + int ff_nvdec_decode_init(AVCodecContext *avctx) { NVDECContext *ctx = avctx->internal->hwaccel_priv_data; + NVDECDecoder *decoder; + AVBufferRef *real_hw_frames_ref; NVDECFramePool *pool; AVHWFramesContext *frames_ctx; const AVPixFmtDescriptor *sw_desc; @@ -301,9 +361,17 @@ int ff_nvdec_decode_init(AVCodecContext *avctx) chroma_444 = ctx->supports_444 && cuvid_chroma_format == cudaVideoChromaFormat_444; if (!avctx->hw_frames_ctx) { - ret = ff_decode_get_hw_frames_ctx(avctx, AV_HWDEVICE_TYPE_CUDA); + ret = nvdec_init_hwframes(avctx, &avctx->hw_frames_ctx, 1); if (ret < 0) return ret; + + ret = nvdec_init_hwframes(avctx, &real_hw_frames_ref, 0); + if (ret < 0) + return ret; + } else { + real_hw_frames_ref = av_buffer_ref(avctx->hw_frames_ctx); + if (!real_hw_frames_ref) + return AVERROR(ENOMEM); } switch (sw_desc->comp[0].depth) { @@ -318,6 +386,7 @@ int ff_nvdec_decode_init(AVCodecContext *avctx) break; default: av_log(avctx, AV_LOG_ERROR, "Unsupported bit depth\n"); + av_buffer_unref(&real_hw_frames_ref); return AVERROR(ENOSYS); } @@ -342,9 +411,14 @@ int ff_nvdec_decode_init(AVCodecContext *avctx) av_log(avctx, AV_LOG_WARNING, "Try lowering the amount of threads. Using %d right now.\n", avctx->thread_count); } + av_buffer_unref(&real_hw_frames_ref); return ret; } + decoder = (NVDECDecoder*)ctx->decoder_ref->data; + decoder->real_hw_frames_ref = real_hw_frames_ref; + real_hw_frames_ref = NULL; + pool = av_mallocz(sizeof(*pool)); if (!pool) { ret = AVERROR(ENOMEM); @@ -374,6 +448,7 @@ static void nvdec_fdd_priv_free(void *priv) av_buffer_unref(&cf->idx_ref); av_buffer_unref(&cf->decoder_ref); + av_buffer_unref(&cf->ref_idx_ref); av_freep(&priv); } @@ -398,6 +473,7 @@ static void nvdec_unmap_mapped_frame(void *opaque, uint8_t *data) finish: av_buffer_unref(&unmap_data->idx_ref); av_buffer_unref(&unmap_data->decoder_ref); + av_buffer_unref(&unmap_data->ref_idx_ref); av_free(unmap_data); } @@ -447,6 +523,13 @@ static int nvdec_retrieve_data(void *logctx, AVFrame *frame) goto copy_fail; } + av_buffer_unref(&frame->hw_frames_ctx); + frame->hw_frames_ctx = av_buffer_ref(decoder->real_hw_frames_ref); + if (!frame->hw_frames_ctx) { + ret = AVERROR(ENOMEM); + goto copy_fail; + } + unmap_data->idx = cf->idx; unmap_data->idx_ref = av_buffer_ref(cf->idx_ref); unmap_data->decoder_ref = av_buffer_ref(cf->decoder_ref); @@ -502,7 +585,7 @@ int ff_nvdec_start_frame(AVCodecContext *avctx, AVFrame *frame) ret = AVERROR(ENOMEM); goto fail; } - cf->idx = *(unsigned int*)cf->idx_ref->data; + cf->ref_idx = cf->idx = *(unsigned int*)cf->idx_ref->data; fdd->hwaccel_priv = cf; fdd->hwaccel_priv_free = nvdec_fdd_priv_free; @@ -515,6 +598,40 @@ fail: } +int ff_nvdec_start_frame_sep_ref(AVCodecContext *avctx, AVFrame *frame, int has_sep_ref) +{ + NVDECContext *ctx = avctx->internal->hwaccel_priv_data; + FrameDecodeData *fdd = (FrameDecodeData*)frame->private_ref->data; + NVDECFrame *cf; + int ret; + + ret = ff_nvdec_start_frame(avctx, frame); + if (ret < 0) + return ret; + + cf = fdd->hwaccel_priv; + + if (has_sep_ref) { + if (!cf->ref_idx_ref) { + cf->ref_idx_ref = av_buffer_pool_get(ctx->decoder_pool); + if (!cf->ref_idx_ref) { + av_log(avctx, AV_LOG_ERROR, "No decoder surfaces left\n"); + ret = AVERROR(ENOMEM); + goto fail; + } + } + cf->ref_idx = *(unsigned int*)cf->ref_idx_ref->data; + } else { + av_buffer_unref(&cf->ref_idx_ref); + cf->ref_idx = cf->idx; + } + + return 0; +fail: + nvdec_fdd_priv_free(cf); + return ret; +} + int ff_nvdec_end_frame(AVCodecContext *avctx) { NVDECContext *ctx = avctx->internal->hwaccel_priv_data; @@ -575,16 +692,6 @@ int ff_nvdec_simple_decode_slice(AVCodecContext *avctx, const uint8_t *buffer, return 0; } -static void nvdec_free_dummy(struct AVHWFramesContext *ctx) -{ - av_buffer_pool_uninit(&ctx->pool); -} - -static AVBufferRef *nvdec_alloc_dummy(int size) -{ - return av_buffer_create(NULL, 0, NULL, NULL, 0); -} - int ff_nvdec_frame_params(AVCodecContext *avctx, AVBufferRef *hw_frames_ctx, int dpb_size, @@ -620,12 +727,6 @@ int ff_nvdec_frame_params(AVCodecContext *avctx, */ frames_ctx->initial_pool_size = dpb_size + 2; - frames_ctx->free = nvdec_free_dummy; - frames_ctx->pool = av_buffer_pool_init(0, nvdec_alloc_dummy); - - if (!frames_ctx->pool) - return AVERROR(ENOMEM); - switch (sw_desc->comp[0].depth) { case 8: frames_ctx->sw_format = chroma_444 ? AV_PIX_FMT_YUV444P : AV_PIX_FMT_NV12; @@ -656,5 +757,5 @@ int ff_nvdec_get_ref_idx(AVFrame *frame) if (!cf) return -1; - return cf->idx; + return cf->ref_idx; }