lavc/videotoolbox: fix threaded decoding

AVHWAccel.end_frame can run on a worker thread. The assumption of the
frame threading code is that the worker thread will change the AVFrame
image data, not the AVFrame fields. So the AVFrame fields are not synced
back to the main thread. But this breaks videotoolbox due to its special
requirements (everything else is fine). It actually wants to update
AVFrame fields.

The actual videotoolbox frame is now stored in the dummy AVBufferRef, so
it mimics what happens in non-videotoolbox cases. (Changing the
AVBufferRef contents is a bit like changing the image data.) The
post_process callback copies that reference to the proper AVFrame field.

Based on a patch by wm4.

Signed-off-by: Aman Gupta <aman@tmm1.net>

diff --git a/libavcodec/h264dec.c b/libavcodec/h264dec.c
index 8c9c6d9f3bfb2f13ff9530f2f92453f3de48872f..7494c7a8f2cedddf94c72c9f35b97df475780741 100644 (file)
--- a/libavcodec/h264dec.c
+++ b/libavcodec/h264dec.c
@@ -838,9 +838,6 @@ static int output_frame(H264Context *h, AVFrame *dst, H264Picture *srcp)
     AVFrame *src = srcp->f;
     int ret;
 
-    if (src->format == AV_PIX_FMT_VIDEOTOOLBOX && src->buf[0]->size == 1)
-        return AVERROR_INVALIDDATA;
-
     ret = av_frame_ref(dst, src);
     if (ret < 0)
         return ret;

diff --git a/libavcodec/videotoolbox.c b/libavcodec/videotoolbox.c
index afec1edf3f6a7be574bdbcfac823add9ba31ed2c..f82c31c5df653c401bb9e48baf0b1b1d986f2c8f 100644 (file)
--- a/libavcodec/videotoolbox.c
+++ b/libavcodec/videotoolbox.c
@@ -45,8 +45,10 @@ enum { kCMVideoCodecType_HEVC = 'hvc1' };
 
 static void videotoolbox_buffer_release(void *opaque, uint8_t *data)
 {
-    CVPixelBufferRef cv_buffer = (CVImageBufferRef)data;
+    CVPixelBufferRef cv_buffer = *(CVPixelBufferRef *)data;
     CVPixelBufferRelease(cv_buffer);
+
+    av_free(data);
 }
 
 static int videotoolbox_buffer_copy(VTContext *vtctx,
@@ -69,19 +71,47 @@ static int videotoolbox_buffer_copy(VTContext *vtctx,
     return 0;
 }
 
+static int videotoolbox_postproc_frame(void *avctx, AVFrame *frame)
+{
+    CVPixelBufferRef ref = *(CVPixelBufferRef *)frame->buf[0]->data;
+
+    if (!ref) {
+        av_log(avctx, AV_LOG_ERROR, "No frame decoded?\n");
+        av_frame_unref(frame);
+        return AVERROR_EXTERNAL;
+    }
+
+    frame->data[3] = (uint8_t*)ref;
+
+    return 0;
+}
+
 int ff_videotoolbox_alloc_frame(AVCodecContext *avctx, AVFrame *frame)
 {
+    size_t      size = sizeof(CVPixelBufferRef);
+    uint8_t    *data = NULL;
+    AVBufferRef *buf = NULL;
     int ret = ff_attach_decode_data(frame);
+    FrameDecodeData *fdd;
     if (ret < 0)
         return ret;
 
+    data = av_mallocz(size);
+    if (!data)
+        return AVERROR(ENOMEM);
+    buf = av_buffer_create(data, size, videotoolbox_buffer_release, NULL, 0);
+    if (!buf) {
+        av_freep(&data);
+        return AVERROR(ENOMEM);
+    }
+    frame->buf[0] = buf;
+
+    fdd = (FrameDecodeData*)frame->private_ref->data;
+    fdd->post_process = videotoolbox_postproc_frame;
+
     frame->width  = avctx->width;
     frame->height = avctx->height;
     frame->format = avctx->pix_fmt;
-    frame->buf[0] = av_buffer_alloc(1);
-
-    if (!frame->buf[0])
-        return AVERROR(ENOMEM);
 
     return 0;
 }
@@ -285,20 +315,24 @@ CFDataRef ff_videotoolbox_hvcc_extradata_create(AVCodecContext *avctx)
     return data;
 }
 
-int ff_videotoolbox_buffer_create(VTContext *vtctx, AVFrame *frame)
+static int videotoolbox_set_frame(AVCodecContext *avctx, AVFrame *frame)
 {
-    av_buffer_unref(&frame->buf[0]);
-
-    frame->buf[0] = av_buffer_create((uint8_t*)vtctx->frame,
-                                     sizeof(vtctx->frame),
-                                     videotoolbox_buffer_release,
-                                     NULL,
-                                     AV_BUFFER_FLAG_READONLY);
-    if (!frame->buf[0]) {
-        return AVERROR(ENOMEM);
+    VTContext *vtctx = avctx->internal->hwaccel_priv_data;
+    if (!frame->buf[0] || frame->data[3]) {
+        av_log(avctx, AV_LOG_ERROR, "videotoolbox: invalid state\n");
+        av_frame_unref(frame);
+        return AVERROR_EXTERNAL;
+    }
+
+    CVPixelBufferRef *ref = (CVPixelBufferRef *)frame->buf[0]->data;
+
+    if (*ref) {
+        av_log(avctx, AV_LOG_ERROR, "videotoolbox: frame already set?\n");
+        av_frame_unref(frame);
+        return AVERROR_EXTERNAL;
     }
 
-    frame->data[3] = (uint8_t*)vtctx->frame;
+    *ref = vtctx->frame;
     vtctx->frame = NULL;
 
     return 0;
@@ -406,7 +440,7 @@ static int videotoolbox_buffer_create(AVCodecContext *avctx, AVFrame *frame)
     AVHWFramesContext *cached_frames;
     int ret;
 
-    ret = ff_videotoolbox_buffer_create(vtctx, frame);
+    ret = videotoolbox_set_frame(avctx, frame);
     if (ret < 0)
         return ret;
 

diff --git a/libavcodec/vt_internal.h b/libavcodec/vt_internal.h
index 929fe423fcde525b0976ead694053f2b3e55a05d..fb64735b8cd3515f653597fd53290f97a1d2c650 100644 (file)
--- a/libavcodec/vt_internal.h
+++ b/libavcodec/vt_internal.h
@@ -46,7 +46,6 @@ typedef struct VTContext {
 
 int ff_videotoolbox_alloc_frame(AVCodecContext *avctx, AVFrame *frame);
 int ff_videotoolbox_uninit(AVCodecContext *avctx);
-int ff_videotoolbox_buffer_create(VTContext *vtctx, AVFrame *frame);
 int ff_videotoolbox_h264_start_frame(AVCodecContext *avctx,
                                      const uint8_t *buffer,
                                      uint32_t size);