[vlc-devel] [PATCH] avcodec: allow the va to request extra pictures for decoding

[Steve Lhomme]

On Mon, Nov 28, 2016 at 3:31 PM, Rémi Denis-Courmont <remi at remlab.net> wrote:
> On November 28, 2016 4:13:58 PM GMT+02:00, Steve Lhomme <robux4 at gmail.com> wrote:
>>On Mon, Nov 28, 2016 at 3:09 PM, Rémi Denis-Courmont <remi at remlab.net>
>>wrote:
>>> On November 28, 2016 3:08:48 PM GMT+02:00, Steve Lhomme
>><robux4 at videolabs.io> wrote:
>>>>Since direct3d11 requires more frames to extract to with some HEVC
>>>>sources.
>>>>
>>>>--
>>>>replaces https://patches.videolan.org/patch/15141/ by setting the
>>>>contraint
>>>>closer to where it's needed.
>>>>---
>>>> modules/codec/avcodec/va.c    | 14 ++++++++++++++
>>>> modules/codec/avcodec/va.h    |  7 +++++++
>>>> modules/codec/avcodec/video.c |  3 +++
>>>> 3 files changed, 24 insertions(+)
>>>>
>>>>diff --git a/modules/codec/avcodec/va.c b/modules/codec/avcodec/va.c
>>>>index 091c35a..b2c0e53 100644
>>>>--- a/modules/codec/avcodec/va.c
>>>>+++ b/modules/codec/avcodec/va.c
>>>>@@ -88,6 +88,20 @@ vlc_fourcc_t vlc_va_GetChroma(enum PixelFormat
>>>>hwfmt, enum PixelFormat swfmt)
>>>>     }
>>>> }
>>>>
>>>>+int vlc_va_GetExtraPictures(enum PixelFormat hwfmt)
>>>>+{
>>>>+    switch (hwfmt)
>>>>+    {
>>>>+#if LIBAVUTIL_VERSION_CHECK(54, 13, 1, 24, 100)
>>>>+        case AV_PIX_FMT_D3D11VA_VLD:
>>>>+            return 2;
>>>>+#endif
>>>>+        default:
>>>>+            break;
>>>>+    }
>>>>+    return 0;
>>>>+}
>>>>+
>>>> static int vlc_va_Start(void *func, va_list ap)
>>>> {
>>>>     vlc_va_t *va = va_arg(ap, vlc_va_t *);
>>>>diff --git a/modules/codec/avcodec/va.h b/modules/codec/avcodec/va.h
>>>>index 4f7df66..e863bd7 100644
>>>>--- a/modules/codec/avcodec/va.h
>>>>+++ b/modules/codec/avcodec/va.h
>>>>@@ -55,6 +55,13 @@ struct vlc_va_t {
>>>>vlc_fourcc_t vlc_va_GetChroma(enum PixelFormat hwfmt, enum
>>PixelFormat
>>>>swfmt);
>>>>
>>>> /**
>>>>+ * Determine the number of extra pictures the hardware decoder will
>>>>need.
>>>>+ * @param hwfmt the hardware acceleration pixel format
>>>>+ * @return the number of extra buffers to allocate for the decoder.
>>>>+ */
>>>>+int vlc_va_GetExtraPictures(enum PixelFormat hwfmt);
>>>>+
>>>>+/**
>>>>  * Creates an accelerated video decoding back-end for libavcodec.
>>>>  * @param obj parent VLC object
>>>>  * @param fmt VLC format of the content to decode
>>>>diff --git a/modules/codec/avcodec/video.c
>>>>b/modules/codec/avcodec/video.c
>>>>index 07db5ae..421e24c 100644
>>>>--- a/modules/codec/avcodec/video.c
>>>>+++ b/modules/codec/avcodec/video.c
>>>>@@ -153,6 +153,9 @@ static int lavc_GetVideoFormat(decoder_t *dec,
>>>>video_format_t *restrict fmt,
>>>>         return -1; /* invalid display size */
>>>>     }
>>>>
>>>>+    if (pix_fmt != sw_pix_fmt)
>>>>+        dec->i_extra_picture_buffers +=
>>>>vlc_va_GetExtraPictures(pix_fmt);
>>>>+
>>>>     fmt->i_width = width;
>>>>     fmt->i_height = height;
>>>>     fmt->i_visible_width = ctx->width;
>>>>--
>>>>2.10.1.windows.1
>>>>
>>>>_______________________________________________
>>>>vlc-devel mailing list
>>>>To unsubscribe or modify your subscription options:
>>>>https://mailman.videolan.org/listinfo/vlc-devel
>>>
>>> This looks completely out of place. I don't see any reason why it
>>would need to be hard-coded *there*, nor any reason why it should be
>>there.
>>
>>The reason why has been explained in other threads. The decoder
>>Extract() from the decoder to other pictures from the decoder pool.
>>Just the dbp size is not enough.
>>
>>The reason to do it there is because it needs to be done before the
>>vout is created (obviously) and since the va is not created before the
>>vout is created, it can only be in the generic va code.
>>
>>> --
>>> Rémi Denis-Courmont
>>> _______________________________________________
>>> vlc-devel mailing list
>>> To unsubscribe or modify your subscription options:
>>> https://mailman.videolan.org/listinfo/vlc-devel
>>_______________________________________________
>>vlc-devel mailing list
>>To unsubscribe or modify your subscription options:
>>https://mailman.videolan.org/listinfo/vlc-devel
>
> I don't see any plausible explanation anywhere and the reasoning from that looks flawed to me as well.

https://mailman.videolan.org/pipermail/vlc-devel/2016-November/110633.html

> So nack.
> --
> Rémi Denis-Courmont
> _______________________________________________
> vlc-devel mailing list
> To unsubscribe or modify your subscription options:
> https://mailman.videolan.org/listinfo/vlc-devel