avcodec/aacdec_fixed: use 64bit to avoid overflow in rounding in apply_dependent_coup...
[ffmpeg.git] / libavcodec / cuviddec.c
index 33e9140..4d3caf9 100644 (file)
@@ -74,6 +74,8 @@ typedef struct CuvidContext
     int internal_error;
     int decoder_flushing;
 
+    int *key_frame;
+
     cudaVideoCodec codec_type;
     cudaVideoChromaFormat chroma_format;
 
@@ -340,6 +342,8 @@ static int CUDAAPI cuvid_handle_picture_decode(void *opaque, CUVIDPICPARAMS* pic
 
     av_log(avctx, AV_LOG_TRACE, "pfnDecodePicture\n");
 
+    ctx->key_frame[picparams->CurrPicIdx] = picparams->intra_pic_flag;
+
     ctx->internal_error = CHECK_CU(ctx->cvdl->cuvidDecodePicture(ctx->cudecoder, picparams));
     if (ctx->internal_error < 0)
         return 0;
@@ -546,12 +550,16 @@ static int cuvid_output_frame(AVCodecContext *avctx, AVFrame *frame)
                     .Height        = avctx->height >> (i ? 1 : 0),
                 };
 
-                ret = CHECK_CU(ctx->cudl->cuMemcpy2D(&cpy));
+                ret = CHECK_CU(ctx->cudl->cuMemcpy2DAsync(&cpy, device_hwctx->stream));
                 if (ret < 0)
                     goto error;
 
                 offset += avctx->height;
             }
+
+            ret = CHECK_CU(ctx->cudl->cuStreamSynchronize(device_hwctx->stream));
+            if (ret < 0)
+                goto error;
         } else if (avctx->pix_fmt == AV_PIX_FMT_NV12 ||
                    avctx->pix_fmt == AV_PIX_FMT_P010 ||
                    avctx->pix_fmt == AV_PIX_FMT_P016) {
@@ -590,6 +598,7 @@ static int cuvid_output_frame(AVCodecContext *avctx, AVFrame *frame)
             goto error;
         }
 
+        frame->key_frame = ctx->key_frame[parsed_frame.dispinfo.picture_index];
         frame->width = avctx->width;
         frame->height = avctx->height;
         if (avctx->pkt_timebase.num && avctx->pkt_timebase.den)
@@ -693,6 +702,8 @@ static av_cold int cuvid_decode_end(AVCodecContext *avctx)
     av_buffer_unref(&ctx->hwframe);
     av_buffer_unref(&ctx->hwdevice);
 
+    av_freep(&ctx->key_frame);
+
     cuvid_free_functions(&ctx->cvdl);
 
     return 0;
@@ -977,6 +988,12 @@ static av_cold int cuvid_decode_init(AVCodecContext *avctx)
                FFMIN(sizeof(ctx->cuparse_ext.raw_seqhdr_data), avctx->extradata_size));
     }
 
+    ctx->key_frame = av_mallocz(ctx->nb_surfaces * sizeof(int));
+    if (!ctx->key_frame) {
+        ret = AVERROR(ENOMEM);
+        goto error;
+    }
+
     ctx->cuparseinfo.ulMaxNumDecodeSurfaces = ctx->nb_surfaces;
     ctx->cuparseinfo.ulMaxDisplayDelay = 4;
     ctx->cuparseinfo.pUserData = avctx;
@@ -1115,12 +1132,6 @@ static const AVCodecHWConfigInternal *cuvid_hw_configs[] = {
         .option = options, \
         .version = LIBAVUTIL_VERSION_INT, \
     }; \
-    AVHWAccel ff_##x##_cuvid_hwaccel = { \
-        .name           = #x "_cuvid", \
-        .type           = AVMEDIA_TYPE_VIDEO, \
-        .id             = AV_CODEC_ID_##X, \
-        .pix_fmt        = AV_PIX_FMT_CUDA, \
-    }; \
     AVCodec ff_##x##_cuvid_decoder = { \
         .name           = #x "_cuvid", \
         .long_name      = NULL_IF_CONFIG_SMALL("Nvidia CUVID " #X " decoder"), \
@@ -1133,13 +1144,14 @@ static const AVCodecHWConfigInternal *cuvid_hw_configs[] = {
         .decode         = cuvid_decode_frame, \
         .receive_frame  = cuvid_output_frame, \
         .flush          = cuvid_flush, \
-        .capabilities   = AV_CODEC_CAP_DELAY | AV_CODEC_CAP_AVOID_PROBING, \
+        .capabilities   = AV_CODEC_CAP_DELAY | AV_CODEC_CAP_AVOID_PROBING | AV_CODEC_CAP_HARDWARE, \
         .pix_fmts       = (const enum AVPixelFormat[]){ AV_PIX_FMT_CUDA, \
                                                         AV_PIX_FMT_NV12, \
                                                         AV_PIX_FMT_P010, \
                                                         AV_PIX_FMT_P016, \
                                                         AV_PIX_FMT_NONE }, \
         .hw_configs     = cuvid_hw_configs, \
+        .wrapper_name   = "cuvid", \
     };
 
 #if CONFIG_HEVC_CUVID_DECODER