Am 08.05.2018 um 17:49 schrieb wm4:
On Tue, 8 May 2018 17:43:49 +0200
Timo Rothenpieler <t...@rothenpieler.org> wrote:

-    frame->buf[0] = av_buffer_pool_get(ctx->pool);
+    if (frctx->flags & AV_CUDA_HWFRAMES_DUMMY_MODE)
+        frame->buf[0] = av_buffer_create(NULL, 0, NULL, NULL, 0);
+    else
+        frame->buf[0] = av_buffer_pool_get(ctx->pool);
+

Is this really needed? Because at least videotoolbox also lets the
decoder allocate frames, and allocates the "dummy" buffers outside of
the hwcontext. (I don't quite remember how it works.)

You mean compared to just leaving buf[0] empty?


No, compared to how the videotoolbox code does things.

videotoolbox seems to use an externally supplied AVHWFramesContext::pool.
Which should be possible to do for cuda as well, i.e. a custom buffer pool that returns mostly empty data, so the new API would be unneeded. I'll have a look if it works out.

Attachment: smime.p7s
Description: S/MIME Cryptographic Signature

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
http://ffmpeg.org/mailman/listinfo/ffmpeg-devel

Reply via email to