| Message ID | 20170301115108.14187-3-thibault.saunier@osg.samsung.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On 01.03.2017 12:51, Thibault Saunier wrote: > It is required by the standard that the field order is set by the > driver, default to NONE in case any is provided, but we can basically > accept any value provided by the userspace as we will anyway not > be able to do any deinterlacing. > > In this patch we also make sure to pass the interlacing mode provided > by userspace from the output to the capture side of the device so > that the information is given back to userspace. This way it can > handle it and potentially deinterlace afterward. As I wrote previously: - on output side you have encoded bytestream - you cannot say about interlacing in such case, so the only valid value is NONE, - on capture side you have decoded frames, and in this case it depends on the device and driver capabilities, if the driver/device does not support (de-)interlacing (I suppose this is MFC case), interlace type field should be filled according to decoded bytestream header (on output side), but no direct copying from output side!!! Regards Andrzej > > Signed-off-by: Thibault Saunier <thibault.saunier@osg.samsung.com> > > --- > > Changes in v6: > - Pass user output field value to the capture as the device is not > doing any deinterlacing and thus decoded content will still be > interlaced on the output. > > Changes in v5: > - Just adapt the field and never error out. > > Changes in v4: None > Changes in v3: > - Do not check values in the g_fmt functions as Andrzej explained in previous review > > Changes in v2: > - Fix a silly build error that slipped in while rebasing the patches > > drivers/media/platform/s5p-mfc/s5p_mfc_common.h | 2 ++ > drivers/media/platform/s5p-mfc/s5p_mfc_dec.c | 6 +++++- > 2 files changed, 7 insertions(+), 1 deletion(-) > > diff --git a/drivers/media/platform/s5p-mfc/s5p_mfc_common.h b/drivers/media/platform/s5p-mfc/s5p_mfc_common.h > index ab23236aa942..3816a37de4bc 100644 > --- a/drivers/media/platform/s5p-mfc/s5p_mfc_common.h > +++ b/drivers/media/platform/s5p-mfc/s5p_mfc_common.h > @@ -652,6 +652,8 @@ struct s5p_mfc_ctx { > size_t me_buffer_size; > size_t tmv_buffer_size; > > + enum v4l2_field field; > + > enum v4l2_mpeg_mfc51_video_force_frame_type force_frame_type; > > struct list_head ref_queue; > diff --git a/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c b/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c > index 367ef8e8dbf0..6e5ca86fb331 100644 > --- a/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c > +++ b/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c > @@ -345,7 +345,7 @@ static int vidioc_g_fmt(struct file *file, void *priv, struct v4l2_format *f) > rectangle. */ > pix_mp->width = ctx->buf_width; > pix_mp->height = ctx->buf_height; > - pix_mp->field = V4L2_FIELD_NONE; > + pix_mp->field = ctx->field; > pix_mp->num_planes = 2; > /* Set pixelformat to the format in which MFC > outputs the decoded frame */ > @@ -380,6 +380,9 @@ static int vidioc_try_fmt(struct file *file, void *priv, struct v4l2_format *f) > struct s5p_mfc_dev *dev = video_drvdata(file); > struct s5p_mfc_fmt *fmt; > > + if (f->fmt.pix.field == V4L2_FIELD_ANY) > + f->fmt.pix.field = V4L2_FIELD_NONE; > + > mfc_debug(2, "Type is %d\n", f->type); > if (f->type == V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE) { > fmt = find_format(f, MFC_FMT_DEC); > @@ -436,6 +439,7 @@ static int vidioc_s_fmt(struct file *file, void *priv, struct v4l2_format *f) > goto out; > } else if (f->type == V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE) { > /* src_fmt is validated by call to vidioc_try_fmt */ > + ctx->field = f->fmt.pix.field; > ctx->src_fmt = find_format(f, MFC_FMT_DEC); > ctx->codec_mode = ctx->src_fmt->codec_mode; > mfc_debug(2, "The codec number is: %d\n", ctx->codec_mode);
Hello, On 03/01/2017 10:12 AM, Andrzej Hajda wrote: > On 01.03.2017 12:51, Thibault Saunier wrote: >> It is required by the standard that the field order is set by the >> driver, default to NONE in case any is provided, but we can basically >> accept any value provided by the userspace as we will anyway not >> be able to do any deinterlacing. >> >> In this patch we also make sure to pass the interlacing mode provided >> by userspace from the output to the capture side of the device so >> that the information is given back to userspace. This way it can >> handle it and potentially deinterlace afterward. > As I wrote previously: > - on output side you have encoded bytestream - you cannot say about > interlacing in such case, so the only valid value is NONE, Well, the encoded stream contains the info about interlacing and the most logical thing to do from my point of view is to keep that info passe it along to the capture side, which is what I am doing. What makes you think this is not the right way of handling that? > - on capture side you have decoded frames, and in this case it depends > on the device and driver capabilities, if the driver/device does not > support (de-)interlacing (I suppose this is MFC case), interlace type > field should be filled according to decoded bytestream header (on output > side), but no direct copying from output side!!! Well, why? If the userspace has already parsed the headers and passed the info to the decoder, there is no reason we should pass along that info to the capture side. Currently the bitstream parser in the decoder does not seem to take of the interlacing properly so if userspace specified it because it already parsed the stream or the info was in the container not in the codec bitstream, then using it is very sensible I think!!! Regards, Thibault > > Regards > Andrzej > >> Signed-off-by: Thibault Saunier <thibault.saunier@osg.samsung.com> >> >> --- >> >> Changes in v6: >> - Pass user output field value to the capture as the device is not >> doing any deinterlacing and thus decoded content will still be >> interlaced on the output. >> >> Changes in v5: >> - Just adapt the field and never error out. >> >> Changes in v4: None >> Changes in v3: >> - Do not check values in the g_fmt functions as Andrzej explained in previous review >> >> Changes in v2: >> - Fix a silly build error that slipped in while rebasing the patches >> >> drivers/media/platform/s5p-mfc/s5p_mfc_common.h | 2 ++ >> drivers/media/platform/s5p-mfc/s5p_mfc_dec.c | 6 +++++- >> 2 files changed, 7 insertions(+), 1 deletion(-) >> >> diff --git a/drivers/media/platform/s5p-mfc/s5p_mfc_common.h b/drivers/media/platform/s5p-mfc/s5p_mfc_common.h >> index ab23236aa942..3816a37de4bc 100644 >> --- a/drivers/media/platform/s5p-mfc/s5p_mfc_common.h >> +++ b/drivers/media/platform/s5p-mfc/s5p_mfc_common.h >> @@ -652,6 +652,8 @@ struct s5p_mfc_ctx { >> size_t me_buffer_size; >> size_t tmv_buffer_size; >> >> + enum v4l2_field field; >> + >> enum v4l2_mpeg_mfc51_video_force_frame_type force_frame_type; >> >> struct list_head ref_queue; >> diff --git a/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c b/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c >> index 367ef8e8dbf0..6e5ca86fb331 100644 >> --- a/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c >> +++ b/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c >> @@ -345,7 +345,7 @@ static int vidioc_g_fmt(struct file *file, void *priv, struct v4l2_format *f) >> rectangle. */ >> pix_mp->width = ctx->buf_width; >> pix_mp->height = ctx->buf_height; >> - pix_mp->field = V4L2_FIELD_NONE; >> + pix_mp->field = ctx->field; >> pix_mp->num_planes = 2; >> /* Set pixelformat to the format in which MFC >> outputs the decoded frame */ >> @@ -380,6 +380,9 @@ static int vidioc_try_fmt(struct file *file, void *priv, struct v4l2_format *f) >> struct s5p_mfc_dev *dev = video_drvdata(file); >> struct s5p_mfc_fmt *fmt; >> >> + if (f->fmt.pix.field == V4L2_FIELD_ANY) >> + f->fmt.pix.field = V4L2_FIELD_NONE; >> + >> mfc_debug(2, "Type is %d\n", f->type); >> if (f->type == V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE) { >> fmt = find_format(f, MFC_FMT_DEC); >> @@ -436,6 +439,7 @@ static int vidioc_s_fmt(struct file *file, void *priv, struct v4l2_format *f) >> goto out; >> } else if (f->type == V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE) { >> /* src_fmt is validated by call to vidioc_try_fmt */ >> + ctx->field = f->fmt.pix.field; >> ctx->src_fmt = find_format(f, MFC_FMT_DEC); >> ctx->codec_mode = ctx->src_fmt->codec_mode; >> mfc_debug(2, "The codec number is: %d\n", ctx->codec_mode); >
Le mercredi 01 mars 2017 à 14:12 +0100, Andrzej Hajda a écrit : > On 01.03.2017 12:51, Thibault Saunier wrote: > > It is required by the standard that the field order is set by the > > driver, default to NONE in case any is provided, but we can > > basically > > accept any value provided by the userspace as we will anyway not > > be able to do any deinterlacing. > > > > In this patch we also make sure to pass the interlacing mode > > provided > > by userspace from the output to the capture side of the device so > > that the information is given back to userspace. This way it can > > handle it and potentially deinterlace afterward. > > As I wrote previously: > - on output side you have encoded bytestream - you cannot say about > interlacing in such case, so the only valid value is NONE, Userspace may know. It's important for the driver not to reset it back to NONE, it would tell the userspace that this encoded format is not supported when interlaced. Obviously, when userspace don't know (ANY), it does not matter, it will fail when we try to set the CAPTURE format. Though, it's quite late in the process for the userspace, which makes implementing software fallback difficult. > - on capture side you have decoded frames, and in this case it > depends > on the device and driver capabilities, if the driver/device does not > support (de-)interlacing (I suppose this is MFC case), interlace type > field should be filled according to decoded bytestream header (on > output > side), but no direct copying from output side!!! That is exact. > > Regards > Andrzej > > > > > Signed-off-by: Thibault Saunier <thibault.saunier@osg.samsung.com> > > > > --- > > > > Changes in v6: > > - Pass user output field value to the capture as the device is not > > doing any deinterlacing and thus decoded content will still be > > interlaced on the output. > > > > Changes in v5: > > - Just adapt the field and never error out. > > > > Changes in v4: None > > Changes in v3: > > - Do not check values in the g_fmt functions as Andrzej explained > > in previous review > > > > Changes in v2: > > - Fix a silly build error that slipped in while rebasing the > > patches > > > > drivers/media/platform/s5p-mfc/s5p_mfc_common.h | 2 ++ > > drivers/media/platform/s5p-mfc/s5p_mfc_dec.c | 6 +++++- > > 2 files changed, 7 insertions(+), 1 deletion(-) > > > > diff --git a/drivers/media/platform/s5p-mfc/s5p_mfc_common.h > > b/drivers/media/platform/s5p-mfc/s5p_mfc_common.h > > index ab23236aa942..3816a37de4bc 100644 > > --- a/drivers/media/platform/s5p-mfc/s5p_mfc_common.h > > +++ b/drivers/media/platform/s5p-mfc/s5p_mfc_common.h > > @@ -652,6 +652,8 @@ struct s5p_mfc_ctx { > > size_t me_buffer_size; > > size_t tmv_buffer_size; > > > > + enum v4l2_field field; > > + > > enum v4l2_mpeg_mfc51_video_force_frame_type > > force_frame_type; > > > > struct list_head ref_queue; > > diff --git a/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c > > b/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c > > index 367ef8e8dbf0..6e5ca86fb331 100644 > > --- a/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c > > +++ b/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c > > @@ -345,7 +345,7 @@ static int vidioc_g_fmt(struct file *file, void > > *priv, struct v4l2_format *f) > > rectangle. */ > > pix_mp->width = ctx->buf_width; > > pix_mp->height = ctx->buf_height; > > - pix_mp->field = V4L2_FIELD_NONE; > > + pix_mp->field = ctx->field; > > pix_mp->num_planes = 2; > > /* Set pixelformat to the format in which MFC > > outputs the decoded frame */ > > @@ -380,6 +380,9 @@ static int vidioc_try_fmt(struct file *file, > > void *priv, struct v4l2_format *f) > > struct s5p_mfc_dev *dev = video_drvdata(file); > > struct s5p_mfc_fmt *fmt; > > > > + if (f->fmt.pix.field == V4L2_FIELD_ANY) > > + f->fmt.pix.field = V4L2_FIELD_NONE; > > + > > mfc_debug(2, "Type is %d\n", f->type); > > if (f->type == V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE) { > > fmt = find_format(f, MFC_FMT_DEC); > > @@ -436,6 +439,7 @@ static int vidioc_s_fmt(struct file *file, void > > *priv, struct v4l2_format *f) > > goto out; > > } else if (f->type == V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE) { > > /* src_fmt is validated by call to vidioc_try_fmt > > */ > > + ctx->field = f->fmt.pix.field; > > ctx->src_fmt = find_format(f, MFC_FMT_DEC); > > ctx->codec_mode = ctx->src_fmt->codec_mode; > > mfc_debug(2, "The codec number is: %d\n", ctx- > > >codec_mode); > >
Hello, On 03/01/2017 11:35 AM, Nicolas Dufresne wrote: > Le mercredi 01 mars 2017 à 14:12 +0100, Andrzej Hajda a écrit : >> On 01.03.2017 12:51, Thibault Saunier wrote: >>> It is required by the standard that the field order is set by the >>> driver, default to NONE in case any is provided, but we can >>> basically >>> accept any value provided by the userspace as we will anyway not >>> be able to do any deinterlacing. >>> >>> In this patch we also make sure to pass the interlacing mode >>> provided >>> by userspace from the output to the capture side of the device so >>> that the information is given back to userspace. This way it can >>> handle it and potentially deinterlace afterward. >> As I wrote previously: >> - on output side you have encoded bytestream - you cannot say about >> interlacing in such case, so the only valid value is NONE, > Userspace may know. It's important for the driver not to reset it back > to NONE, it would tell the userspace that this encoded format is not > supported when interlaced. > > Obviously, when userspace don't know (ANY), it does not matter, it will > fail when we try to set the CAPTURE format. Though, it's quite late in > the process for the userspace, which makes implementing software > fallback difficult. > >> - on capture side you have decoded frames, and in this case it >> depends >> on the device and driver capabilities, if the driver/device does not >> support (de-)interlacing (I suppose this is MFC case), interlace type >> field should be filled according to decoded bytestream header (on >> output >> side), but no direct copying from output side!!! > That is exact. It is yes, in case the driver is determining it properly, which is not the case here. So the information we have is what the userspace figured. That part should be fixed in the driver, but meanwhile it is better to copy to the output side than just loosing the info as otherwise the userspace never tries to deinterlace after the decoding because the driver always says field=NONE now. Regards, Thibault >> Regards >> Andrzej >> >>> Signed-off-by: Thibault Saunier <thibault.saunier@osg.samsung.com> >>> >>> --- >>> >>> Changes in v6: >>> - Pass user output field value to the capture as the device is not >>> doing any deinterlacing and thus decoded content will still be >>> interlaced on the output. >>> >>> Changes in v5: >>> - Just adapt the field and never error out. >>> >>> Changes in v4: None >>> Changes in v3: >>> - Do not check values in the g_fmt functions as Andrzej explained >>> in previous review >>> >>> Changes in v2: >>> - Fix a silly build error that slipped in while rebasing the >>> patches >>> >>> drivers/media/platform/s5p-mfc/s5p_mfc_common.h | 2 ++ >>> drivers/media/platform/s5p-mfc/s5p_mfc_dec.c | 6 +++++- >>> 2 files changed, 7 insertions(+), 1 deletion(-) >>> >>> diff --git a/drivers/media/platform/s5p-mfc/s5p_mfc_common.h >>> b/drivers/media/platform/s5p-mfc/s5p_mfc_common.h >>> index ab23236aa942..3816a37de4bc 100644 >>> --- a/drivers/media/platform/s5p-mfc/s5p_mfc_common.h >>> +++ b/drivers/media/platform/s5p-mfc/s5p_mfc_common.h >>> @@ -652,6 +652,8 @@ struct s5p_mfc_ctx { >>> size_t me_buffer_size; >>> size_t tmv_buffer_size; >>> >>> + enum v4l2_field field; >>> + >>> enum v4l2_mpeg_mfc51_video_force_frame_type >>> force_frame_type; >>> >>> struct list_head ref_queue; >>> diff --git a/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c >>> b/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c >>> index 367ef8e8dbf0..6e5ca86fb331 100644 >>> --- a/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c >>> +++ b/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c >>> @@ -345,7 +345,7 @@ static int vidioc_g_fmt(struct file *file, void >>> *priv, struct v4l2_format *f) >>> rectangle. */ >>> pix_mp->width = ctx->buf_width; >>> pix_mp->height = ctx->buf_height; >>> - pix_mp->field = V4L2_FIELD_NONE; >>> + pix_mp->field = ctx->field; >>> pix_mp->num_planes = 2; >>> /* Set pixelformat to the format in which MFC >>> outputs the decoded frame */ >>> @@ -380,6 +380,9 @@ static int vidioc_try_fmt(struct file *file, >>> void *priv, struct v4l2_format *f) >>> struct s5p_mfc_dev *dev = video_drvdata(file); >>> struct s5p_mfc_fmt *fmt; >>> >>> + if (f->fmt.pix.field == V4L2_FIELD_ANY) >>> + f->fmt.pix.field = V4L2_FIELD_NONE; >>> + >>> mfc_debug(2, "Type is %d\n", f->type); >>> if (f->type == V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE) { >>> fmt = find_format(f, MFC_FMT_DEC); >>> @@ -436,6 +439,7 @@ static int vidioc_s_fmt(struct file *file, void >>> *priv, struct v4l2_format *f) >>> goto out; >>> } else if (f->type == V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE) { >>> /* src_fmt is validated by call to vidioc_try_fmt >>> */ >>> + ctx->field = f->fmt.pix.field; >>> ctx->src_fmt = find_format(f, MFC_FMT_DEC); >>> ctx->codec_mode = ctx->src_fmt->codec_mode; >>> mfc_debug(2, "The codec number is: %d\n", ctx- >>>> codec_mode);
Le mercredi 01 mars 2017 à 14:12 +0100, Andrzej Hajda a écrit : > - on output side you have encoded bytestream - you cannot say about > interlacing in such case, so the only valid value is NONE, > - on capture side you have decoded frames, and in this case it > depends > on the device and driver capabilities, if the driver/device does not > support (de-)interlacing (I suppose this is MFC case), interlace type > field should be filled according to decoded bytestream header (on > output > side), but no direct copying from output side!!! I think we need some nuance here for this to actually be usable. If the information is not provided by the driver (yes, hardware is limiting sometimes), it would make sense to copy over the information that userspace provided. Setting NONE is just the worst approximation in my opinion. About MFC, it will be worth trying to read the DISPLAY_STATUS after the headers has been processed. It's not clearly stated in the spec if this will be set or not. Nicolas
On 01.03.2017 16:21, Nicolas Dufresne wrote: > Le mercredi 01 mars 2017 à 14:12 +0100, Andrzej Hajda a écrit : >> - on output side you have encoded bytestream - you cannot say about >> interlacing in such case, so the only valid value is NONE, >> - on capture side you have decoded frames, and in this case it >> depends >> on the device and driver capabilities, if the driver/device does not >> support (de-)interlacing (I suppose this is MFC case), interlace type >> field should be filled according to decoded bytestream header (on >> output >> side), but no direct copying from output side!!! > I think we need some nuance here for this to actually be usable. If the > information is not provided by the driver (yes, hardware is limiting > sometimes), it would make sense to copy over the information that > userspace provided. Setting NONE is just the worst approximation in my > opinion. The whole point is that s_fmt on output side is to describe format of the stream passed to the device, and in case of decoder it is just mpeg/h.26x/... stream. It does not contain frames, fields, width, height - it is just raw stream of bytes. We cannot say in such case about field type, there is not such thing as interlaced byte stream. Using s_fmt on output to describe things on capture side look for me unnecessary and abuses V4L2 API IMO. > > About MFC, it will be worth trying to read the DISPLAY_STATUS after the > headers has been processed. It's not clearly stated in the spec if this > will be set or not. > Documentation for MFC6.5 states clearly: > Note: On SEQ_DONE, INTERLACE_PICTURE will return the picture type to > be decoded based on the > sequence header information. In case of MFC5.1 it is unclear, but I hope HW behaves the same way. Anyway I agree it will be good to fix it at least for MFC6.5+, any volunteer? Regards Andrzej
On 01/03/17 12:51, Thibault Saunier wrote: > It is required by the standard that the field order is set by the > driver, default to NONE in case any is provided, but we can basically > accept any value provided by the userspace as we will anyway not > be able to do any deinterlacing. > > In this patch we also make sure to pass the interlacing mode provided > by userspace from the output to the capture side of the device so > that the information is given back to userspace. This way it can > handle it and potentially deinterlace afterward. > > Signed-off-by: Thibault Saunier <thibault.saunier@osg.samsung.com> > > --- > > Changes in v6: > - Pass user output field value to the capture as the device is not > doing any deinterlacing and thus decoded content will still be > interlaced on the output. > > Changes in v5: > - Just adapt the field and never error out. > > Changes in v4: None > Changes in v3: > - Do not check values in the g_fmt functions as Andrzej explained in previous review > > Changes in v2: > - Fix a silly build error that slipped in while rebasing the patches > > drivers/media/platform/s5p-mfc/s5p_mfc_common.h | 2 ++ > drivers/media/platform/s5p-mfc/s5p_mfc_dec.c | 6 +++++- > 2 files changed, 7 insertions(+), 1 deletion(-) > > diff --git a/drivers/media/platform/s5p-mfc/s5p_mfc_common.h b/drivers/media/platform/s5p-mfc/s5p_mfc_common.h > index ab23236aa942..3816a37de4bc 100644 > --- a/drivers/media/platform/s5p-mfc/s5p_mfc_common.h > +++ b/drivers/media/platform/s5p-mfc/s5p_mfc_common.h > @@ -652,6 +652,8 @@ struct s5p_mfc_ctx { > size_t me_buffer_size; > size_t tmv_buffer_size; > > + enum v4l2_field field; > + > enum v4l2_mpeg_mfc51_video_force_frame_type force_frame_type; > > struct list_head ref_queue; > diff --git a/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c b/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c > index 367ef8e8dbf0..6e5ca86fb331 100644 > --- a/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c > +++ b/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c > @@ -345,7 +345,7 @@ static int vidioc_g_fmt(struct file *file, void *priv, struct v4l2_format *f) > rectangle. */ > pix_mp->width = ctx->buf_width; > pix_mp->height = ctx->buf_height; > - pix_mp->field = V4L2_FIELD_NONE; > + pix_mp->field = ctx->field; > pix_mp->num_planes = 2; > /* Set pixelformat to the format in which MFC > outputs the decoded frame */ > @@ -380,6 +380,9 @@ static int vidioc_try_fmt(struct file *file, void *priv, struct v4l2_format *f) > struct s5p_mfc_dev *dev = video_drvdata(file); > struct s5p_mfc_fmt *fmt; > > + if (f->fmt.pix.field == V4L2_FIELD_ANY) > + f->fmt.pix.field = V4L2_FIELD_NONE; > + > mfc_debug(2, "Type is %d\n", f->type); > if (f->type == V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE) { > fmt = find_format(f, MFC_FMT_DEC); > @@ -436,6 +439,7 @@ static int vidioc_s_fmt(struct file *file, void *priv, struct v4l2_format *f) > goto out; > } else if (f->type == V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE) { > /* src_fmt is validated by call to vidioc_try_fmt */ > + ctx->field = f->fmt.pix.field; Doing this means that you also allow for V4L2_FIELD_ALTERNATE. If you do that, then there are additional requirements when you queue up buffers in that the 'field' has to be set to TOP or BOTTOM in struct v4l2_buffer. I am sure that is a requirement for the encoder, how this would work for a decoder I am not sure. Also, values like V4L2_FIELD_SEQ_TB/BT would not behave well if you pass them through a decoder. Frankly I think the only 'reasonable' values would be FIELD_NONE, FIELD_INTERLACED and possibly FIELD_ALTERNATE. I don't know enough about how codecs handle interlaced formats, so I can't tell how much sense FIELD_ALTERNATE would make. Properly supporting interlaced formats should only be done if you actually tested it and know how it works and what exactly is supported. Regards, Hans > ctx->src_fmt = find_format(f, MFC_FMT_DEC); > ctx->codec_mode = ctx->src_fmt->codec_mode; > mfc_debug(2, "The codec number is: %d\n", ctx->codec_mode); >
diff --git a/drivers/media/platform/s5p-mfc/s5p_mfc_common.h b/drivers/media/platform/s5p-mfc/s5p_mfc_common.h index ab23236aa942..3816a37de4bc 100644 --- a/drivers/media/platform/s5p-mfc/s5p_mfc_common.h +++ b/drivers/media/platform/s5p-mfc/s5p_mfc_common.h @@ -652,6 +652,8 @@ struct s5p_mfc_ctx { size_t me_buffer_size; size_t tmv_buffer_size; + enum v4l2_field field; + enum v4l2_mpeg_mfc51_video_force_frame_type force_frame_type; struct list_head ref_queue; diff --git a/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c b/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c index 367ef8e8dbf0..6e5ca86fb331 100644 --- a/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c +++ b/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c @@ -345,7 +345,7 @@ static int vidioc_g_fmt(struct file *file, void *priv, struct v4l2_format *f) rectangle. */ pix_mp->width = ctx->buf_width; pix_mp->height = ctx->buf_height; - pix_mp->field = V4L2_FIELD_NONE; + pix_mp->field = ctx->field; pix_mp->num_planes = 2; /* Set pixelformat to the format in which MFC outputs the decoded frame */ @@ -380,6 +380,9 @@ static int vidioc_try_fmt(struct file *file, void *priv, struct v4l2_format *f) struct s5p_mfc_dev *dev = video_drvdata(file); struct s5p_mfc_fmt *fmt; + if (f->fmt.pix.field == V4L2_FIELD_ANY) + f->fmt.pix.field = V4L2_FIELD_NONE; + mfc_debug(2, "Type is %d\n", f->type); if (f->type == V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE) { fmt = find_format(f, MFC_FMT_DEC); @@ -436,6 +439,7 @@ static int vidioc_s_fmt(struct file *file, void *priv, struct v4l2_format *f) goto out; } else if (f->type == V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE) { /* src_fmt is validated by call to vidioc_try_fmt */ + ctx->field = f->fmt.pix.field; ctx->src_fmt = find_format(f, MFC_FMT_DEC); ctx->codec_mode = ctx->src_fmt->codec_mode; mfc_debug(2, "The codec number is: %d\n", ctx->codec_mode);
It is required by the standard that the field order is set by the driver, default to NONE in case any is provided, but we can basically accept any value provided by the userspace as we will anyway not be able to do any deinterlacing. In this patch we also make sure to pass the interlacing mode provided by userspace from the output to the capture side of the device so that the information is given back to userspace. This way it can handle it and potentially deinterlace afterward. Signed-off-by: Thibault Saunier <thibault.saunier@osg.samsung.com> --- Changes in v6: - Pass user output field value to the capture as the device is not doing any deinterlacing and thus decoded content will still be interlaced on the output. Changes in v5: - Just adapt the field and never error out. Changes in v4: None Changes in v3: - Do not check values in the g_fmt functions as Andrzej explained in previous review Changes in v2: - Fix a silly build error that slipped in while rebasing the patches drivers/media/platform/s5p-mfc/s5p_mfc_common.h | 2 ++ drivers/media/platform/s5p-mfc/s5p_mfc_dec.c | 6 +++++- 2 files changed, 7 insertions(+), 1 deletion(-)