[PATCH v5 0/3] Fixes for colorspace logic in exynos-gsc and s5p-mfc drivers

[PATCH v5 0/3] Fixes for colorspace logic in exynos-gsc and s5p-mfc drivers

[Thibault Saunier]

Hello,

This patchset fixes a few issues on the colorspace logic for the exynos-gsc
and s5p-mfc drivers.

We now handle the colorspace in those drivers, and make sure to respect user setting if
possible.

We also now set the 'v4l2_pix_format:field' if userspace passed ANY, avoiding GStreamer
spamming error at us about the driver not following the standard.

This is the fifth version of the patch serie.

Best regards,

Thibault Saunier

Changes in v5:
- Squash commit to always use output colorspace on the capture side
  inside this one
- Fix typo in commit message
- Just adapt the field and never error out.

Changes in v4:
- Reword commit message to better back our assumptions on specifications
- Set the colorspace only if the user passed V4L2_COLORSPACE_DEFAULT, in
  all other cases just use what userspace provided.

Changes in v3:
- Do not check values in the g_fmt functions as Andrzej explained in previous review
- Added 'Reviewed-by: Andrzej Hajda <a.hajda@samsung.com>'
- Do not check values in the g_fmt functions as Andrzej explained in previous review
- Set colorspace if user passed V4L2_COLORSPACE_DEFAULT in
- 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

Thibault Saunier (3):
  [media] exynos-gsc: Use user configured colorspace if provided
  [media] s5p-mfc: Set colorspace in VIDIO_{G,TRY}_FMT if DEFAULT
    provided
  [media] s5p-mfc: Check and set 'v4l2_pix_format:field' field in
    try_fmt

 drivers/media/platform/exynos-gsc/gsc-core.c | 20 +++++++++++++++-----
 drivers/media/platform/exynos-gsc/gsc-core.h |  1 +
 drivers/media/platform/s5p-mfc/s5p_mfc_dec.c | 17 +++++++++++++++++
 3 files changed, 33 insertions(+), 5 deletions(-)

-- 
2.11.1

[PATCH v5 1/3] [media] exynos-gsc: Use user configured colorspace if provided

[Thibault Saunier]

Use colorspace provided by the user as we are only doing scaling and
color encoding conversion, we won't be able to transform the colorspace
itself and the colorspace won't mater in that operation.

Also always use output colorspace on the capture side.

Start using 576p as a threashold to compute the colorspace.
The media documentation says that the V4L2_COLORSPACE_SMPTE170M colorspace
should be used for SDTV and V4L2_COLORSPACE_REC709 for HDTV. But drivers
don't agree on the display resolution that should be used as a threshold.

>From EIA CEA 861B about colorimetry for various resolutions:

  - 5.1 480p, 480i, 576p, 576i, 240p, and 288p
    The color space used by the 480-line, 576-line, 240-line, and 288-line
    formats will likely be based on SMPTE 170M [1].
  - 5.2 1080i, 1080p, and 720p
    The color space used by the high definition formats will likely be
    based on ITU-R BT.709-4

This indicates that in the case that userspace does not specify what
colorspace should be used, we should use 576p  as a threshold to set
V4L2_COLORSPACE_REC709 instead of V4L2_COLORSPACE_SMPTE170M. Even if it is
only 'likely' and not a requirement it is the best guess we can make.

The stream should have been encoded with the information and userspace
has to pass it to the driver if it is not the case, otherwise we won't be
able to handle it properly anyhow.

Also, check for the resolution in G_FMT instead unconditionally setting
the V4L2_COLORSPACE_REC709 colorspace.

Signed-off-by: Javier Martinez Canillas <javier@osg.samsung.com>
Signed-off-by: Thibault Saunier <thibault.saunier@osg.samsung.com>
Reviewed-by: Andrzej Hajda <a.hajda@samsung.com>

---

Changes in v5:
- Squash commit to always use output colorspace on the capture side
  inside this one
- Fix typo in commit message

Changes in v4:
- Reword commit message to better back our assumptions on specifications

Changes in v3:
- Do not check values in the g_fmt functions as Andrzej explained in previous review
- Added 'Reviewed-by: Andrzej Hajda <a.hajda@samsung.com>'

Changes in v2: None

 drivers/media/platform/exynos-gsc/gsc-core.c | 20 +++++++++++++++-----
 drivers/media/platform/exynos-gsc/gsc-core.h |  1 +
 2 files changed, 16 insertions(+), 5 deletions(-)

diff --git a/drivers/media/platform/exynos-gsc/gsc-core.c b/drivers/media/platform/exynos-gsc/gsc-core.c
index 59a634201830..772599de8c13 100644
--- a/drivers/media/platform/exynos-gsc/gsc-core.c
+++ b/drivers/media/platform/exynos-gsc/gsc-core.c
@@ -454,6 +454,7 @@ int gsc_try_fmt_mplane(struct gsc_ctx *ctx, struct v4l2_format *f)
 	} else {
 		min_w = variant->pix_min->target_rot_dis_w;
 		min_h = variant->pix_min->target_rot_dis_h;
+		pix_mp->colorspace = ctx->out_colorspace;
 	}
 
 	pr_debug("mod_x: %d, mod_y: %d, max_w: %d, max_h = %d",
@@ -472,10 +473,15 @@ int gsc_try_fmt_mplane(struct gsc_ctx *ctx, struct v4l2_format *f)
 
 	pix_mp->num_planes = fmt->num_planes;
 
-	if (pix_mp->width >= 1280) /* HD */
-		pix_mp->colorspace = V4L2_COLORSPACE_REC709;
-	else /* SD */
-		pix_mp->colorspace = V4L2_COLORSPACE_SMPTE170M;
+	if (pix_mp->colorspace == V4L2_COLORSPACE_DEFAULT) {
+		if (pix_mp->width > 720 && pix_mp->height > 576) /* HD */
+			pix_mp->colorspace = V4L2_COLORSPACE_REC709;
+		else /* SD */
+			pix_mp->colorspace = V4L2_COLORSPACE_SMPTE170M;
+	}
+
+	if (V4L2_TYPE_IS_OUTPUT(f->type))
+		ctx->out_colorspace = pix_mp->colorspace;
 
 	for (i = 0; i < pix_mp->num_planes; ++i) {
 		struct v4l2_plane_pix_format *plane_fmt = &pix_mp->plane_fmt[i];
@@ -519,9 +525,13 @@ int gsc_g_fmt_mplane(struct gsc_ctx *ctx, struct v4l2_format *f)
 	pix_mp->height		= frame->f_height;
 	pix_mp->field		= V4L2_FIELD_NONE;
 	pix_mp->pixelformat	= frame->fmt->pixelformat;
-	pix_mp->colorspace	= V4L2_COLORSPACE_REC709;
 	pix_mp->num_planes	= frame->fmt->num_planes;
 
+	if (pix_mp->width > 720 && pix_mp->height > 576) /* HD */
+		pix_mp->colorspace = V4L2_COLORSPACE_REC709;
+	else /* SD */
+		pix_mp->colorspace = V4L2_COLORSPACE_SMPTE170M;
+
 	for (i = 0; i < pix_mp->num_planes; ++i) {
 		pix_mp->plane_fmt[i].bytesperline = (frame->f_width *
 			frame->fmt->depth[i]) / 8;
diff --git a/drivers/media/platform/exynos-gsc/gsc-core.h b/drivers/media/platform/exynos-gsc/gsc-core.h
index 696217e9af66..715d9c9d8d30 100644
--- a/drivers/media/platform/exynos-gsc/gsc-core.h
+++ b/drivers/media/platform/exynos-gsc/gsc-core.h
@@ -376,6 +376,7 @@ struct gsc_ctx {
 	struct v4l2_ctrl_handler ctrl_handler;
 	struct gsc_ctrls	gsc_ctrls;
 	bool			ctrls_rdy;
+	enum v4l2_colorspace out_colorspace;
 };
 
 void gsc_set_prefbuf(struct gsc_dev *gsc, struct gsc_frame *frm);
-- 
2.11.1

[PATCH v5 2/3] [media] s5p-mfc: Set colorspace in VIDIO_{G,TRY}_FMT if DEFAULT provided

[Thibault Saunier]

The media documentation says that the V4L2_COLORSPACE_SMPTE170M colorspace
should be used for SDTV and V4L2_COLORSPACE_REC709 for HDTV but the driver
didn't set the colorimetry when userspace provided
V4L2_COLORSPACE_DEFAULT.

Use 576p display resolution as a threshold to set this as suggested by
EIA CEA 861B.

Signed-off-by: Thibault Saunier <thibault.saunier@osg.samsung.com>

---

Changes in v5: None
Changes in v4:
- Set the colorspace only if the user passed V4L2_COLORSPACE_DEFAULT, in
  all other cases just use what userspace provided.

Changes in v3:
- Do not check values in the g_fmt functions as Andrzej explained in previous review
- Set colorspace if user passed V4L2_COLORSPACE_DEFAULT in

Changes in v2: None

 drivers/media/platform/s5p-mfc/s5p_mfc_dec.c | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

diff --git a/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c b/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c
index 367ef8e8dbf0..0976c3e0a5ce 100644
--- a/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c
+++ b/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c
@@ -354,6 +354,11 @@ static int vidioc_g_fmt(struct file *file, void *priv, struct v4l2_format *f)
 		pix_mp->plane_fmt[0].sizeimage = ctx->luma_size;
 		pix_mp->plane_fmt[1].bytesperline = ctx->buf_width;
 		pix_mp->plane_fmt[1].sizeimage = ctx->chroma_size;
+
+		if (pix_mp->width > 720 && pix_mp->height > 576) /* HD */
+			pix_mp->colorspace = V4L2_COLORSPACE_REC709;
+		else /* SD */
+			pix_mp->colorspace = V4L2_COLORSPACE_SMPTE170M;
 	} else if (f->type == V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE) {
 		/* This is run on OUTPUT
 		   The buffer contains compressed image
@@ -378,6 +383,7 @@ static int vidioc_g_fmt(struct file *file, void *priv, struct v4l2_format *f)
 static int vidioc_try_fmt(struct file *file, void *priv, struct v4l2_format *f)
 {
 	struct s5p_mfc_dev *dev = video_drvdata(file);
+	struct v4l2_pix_format_mplane *pix_mp = &f->fmt.pix_mp;
 	struct s5p_mfc_fmt *fmt;
 
 	mfc_debug(2, "Type is %d\n", f->type);
@@ -405,6 +411,14 @@ static int vidioc_try_fmt(struct file *file, void *priv, struct v4l2_format *f)
 			mfc_err("Unsupported format by this MFC version.\n");
 			return -EINVAL;
 		}
+
+		if (pix_mp->colorspace == V4L2_COLORSPACE_DEFAULT) {
+			if (pix_mp->width > 720 &&
+					pix_mp->height > 576) /* HD */
+				pix_mp->colorspace = V4L2_COLORSPACE_REC709;
+			else /* SD */
+				pix_mp->colorspace = V4L2_COLORSPACE_SMPTE170M;
+		}
 	}
 
 	return 0;
-- 
2.11.1

[PATCH v5 3/3] [media] s5p-mfc: Check and set 'v4l2_pix_format:field' field in try_fmt

[Thibault Saunier]

It is required by the standard that the field order is set by the
driver.

Signed-off-by: Thibault Saunier <thibault.saunier@osg.samsung.com>

---

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_dec.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c b/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c
index 0976c3e0a5ce..44ed2afe0780 100644
--- a/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c
+++ b/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c
@@ -386,6 +386,9 @@ static int vidioc_try_fmt(struct file *file, void *priv, struct v4l2_format *f)
 	struct v4l2_pix_format_mplane *pix_mp = &f->fmt.pix_mp;
 	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);
-- 
2.11.1

Re: [PATCH v5 1/3] [media] exynos-gsc: Use user configured colorspace if provided

[Hans Verkuil]

On 02/21/2017 11:20 AM, Thibault Saunier wrote:
> Use colorspace provided by the user as we are only doing scaling and
> color encoding conversion, we won't be able to transform the colorspace
> itself and the colorspace won't mater in that operation.
> 
> Also always use output colorspace on the capture side.
> 
> Start using 576p as a threashold to compute the colorspace.
> The media documentation says that the V4L2_COLORSPACE_SMPTE170M colorspace
> should be used for SDTV and V4L2_COLORSPACE_REC709 for HDTV. But drivers
> don't agree on the display resolution that should be used as a threshold.
> 
> From EIA CEA 861B about colorimetry for various resolutions:
> 
>   - 5.1 480p, 480i, 576p, 576i, 240p, and 288p
>     The color space used by the 480-line, 576-line, 240-line, and 288-line
>     formats will likely be based on SMPTE 170M [1].
>   - 5.2 1080i, 1080p, and 720p
>     The color space used by the high definition formats will likely be
>     based on ITU-R BT.709-4
> 
> This indicates that in the case that userspace does not specify what
> colorspace should be used, we should use 576p  as a threshold to set
> V4L2_COLORSPACE_REC709 instead of V4L2_COLORSPACE_SMPTE170M. Even if it is
> only 'likely' and not a requirement it is the best guess we can make.
> 
> The stream should have been encoded with the information and userspace
> has to pass it to the driver if it is not the case, otherwise we won't be
> able to handle it properly anyhow.
> 
> Also, check for the resolution in G_FMT instead unconditionally setting
> the V4L2_COLORSPACE_REC709 colorspace.
> 
> Signed-off-by: Javier Martinez Canillas <javier@osg.samsung.com>
> Signed-off-by: Thibault Saunier <thibault.saunier@osg.samsung.com>
> Reviewed-by: Andrzej Hajda <a.hajda@samsung.com>
> 
> ---
> 
> Changes in v5:
> - Squash commit to always use output colorspace on the capture side
>   inside this one
> - Fix typo in commit message
> 
> Changes in v4:
> - Reword commit message to better back our assumptions on specifications
> 
> Changes in v3:
> - Do not check values in the g_fmt functions as Andrzej explained in previous review
> - Added 'Reviewed-by: Andrzej Hajda <a.hajda@samsung.com>'
> 
> Changes in v2: None
> 
>  drivers/media/platform/exynos-gsc/gsc-core.c | 20 +++++++++++++++-----
>  drivers/media/platform/exynos-gsc/gsc-core.h |  1 +
>  2 files changed, 16 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/media/platform/exynos-gsc/gsc-core.c b/drivers/media/platform/exynos-gsc/gsc-core.c
> index 59a634201830..772599de8c13 100644
> --- a/drivers/media/platform/exynos-gsc/gsc-core.c
> +++ b/drivers/media/platform/exynos-gsc/gsc-core.c
> @@ -454,6 +454,7 @@ int gsc_try_fmt_mplane(struct gsc_ctx *ctx, struct v4l2_format *f)
>  	} else {
>  		min_w = variant->pix_min->target_rot_dis_w;
>  		min_h = variant->pix_min->target_rot_dis_h;
> +		pix_mp->colorspace = ctx->out_colorspace;
>  	}
>  
>  	pr_debug("mod_x: %d, mod_y: %d, max_w: %d, max_h = %d",
> @@ -472,10 +473,15 @@ int gsc_try_fmt_mplane(struct gsc_ctx *ctx, struct v4l2_format *f)
>  
>  	pix_mp->num_planes = fmt->num_planes;
>  
> -	if (pix_mp->width >= 1280) /* HD */
> -		pix_mp->colorspace = V4L2_COLORSPACE_REC709;
> -	else /* SD */
> -		pix_mp->colorspace = V4L2_COLORSPACE_SMPTE170M;
> +	if (pix_mp->colorspace == V4L2_COLORSPACE_DEFAULT) {
> +		if (pix_mp->width > 720 && pix_mp->height > 576) /* HD */

I'd use || instead of && here. Ditto for the next patch.

> +			pix_mp->colorspace = V4L2_COLORSPACE_REC709;
> +		else /* SD */
> +			pix_mp->colorspace = V4L2_COLORSPACE_SMPTE170M;
> +	}

Are you sure this is in fact how it is used? If the source of the video
is the sensor (camera), then the colorspace is typically SRGB. I'm not
really sure you should guess here. All a mem2mem driver should do is to
pass the colorspace etc. information from the output to the capture side,
and not invent things. That simply makes no sense.

We may have to update the documentation or v4l2-compliance test if this
is an issue. The more I think about this, the more I think we shouldn't
do this guessing game.

Regards,

	Hans

> +
> +	if (V4L2_TYPE_IS_OUTPUT(f->type))
> +		ctx->out_colorspace = pix_mp->colorspace;
>  
>  	for (i = 0; i < pix_mp->num_planes; ++i) {
>  		struct v4l2_plane_pix_format *plane_fmt = &pix_mp->plane_fmt[i];
> @@ -519,9 +525,13 @@ int gsc_g_fmt_mplane(struct gsc_ctx *ctx, struct v4l2_format *f)
>  	pix_mp->height		= frame->f_height;
>  	pix_mp->field		= V4L2_FIELD_NONE;
>  	pix_mp->pixelformat	= frame->fmt->pixelformat;
> -	pix_mp->colorspace	= V4L2_COLORSPACE_REC709;
>  	pix_mp->num_planes	= frame->fmt->num_planes;
>  
> +	if (pix_mp->width > 720 && pix_mp->height > 576) /* HD */
> +		pix_mp->colorspace = V4L2_COLORSPACE_REC709;
> +	else /* SD */
> +		pix_mp->colorspace = V4L2_COLORSPACE_SMPTE170M;
> +
>  	for (i = 0; i < pix_mp->num_planes; ++i) {
>  		pix_mp->plane_fmt[i].bytesperline = (frame->f_width *
>  			frame->fmt->depth[i]) / 8;
> diff --git a/drivers/media/platform/exynos-gsc/gsc-core.h b/drivers/media/platform/exynos-gsc/gsc-core.h
> index 696217e9af66..715d9c9d8d30 100644
> --- a/drivers/media/platform/exynos-gsc/gsc-core.h
> +++ b/drivers/media/platform/exynos-gsc/gsc-core.h
> @@ -376,6 +376,7 @@ struct gsc_ctx {
>  	struct v4l2_ctrl_handler ctrl_handler;
>  	struct gsc_ctrls	gsc_ctrls;
>  	bool			ctrls_rdy;
> +	enum v4l2_colorspace out_colorspace;
>  };
>  
>  void gsc_set_prefbuf(struct gsc_dev *gsc, struct gsc_frame *frm);
> 

Re: [PATCH v5 3/3] [media] s5p-mfc: Check and set 'v4l2_pix_format:field' field in try_fmt

[Andrzej Hajda]

On 21.02.2017 20:20, Thibault Saunier wrote:
> It is required by the standard that the field order is set by the
> driver.
>
> Signed-off-by: Thibault Saunier <thibault.saunier@osg.samsung.com>
>
> ---
>
> 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_dec.c | 3 +++
>  1 file changed, 3 insertions(+)
>
> diff --git a/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c b/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c
> index 0976c3e0a5ce..44ed2afe0780 100644
> --- a/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c
> +++ b/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c
> @@ -386,6 +386,9 @@ static int vidioc_try_fmt(struct file *file, void *priv, struct v4l2_format *f)
>  	struct v4l2_pix_format_mplane *pix_mp = &f->fmt.pix_mp;
>  	struct s5p_mfc_fmt *fmt;
>  
> +	if (f->fmt.pix.field == V4L2_FIELD_ANY)
> +		f->fmt.pix.field = V4L2_FIELD_NONE;
> +

In previous version the only supported field type was NONE, here you
support everything.
If the only supported format is none you should set 'field'
unconditionally to NONE, nothing more.

Regards
Andrzej

>  	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);

Re: [PATCH v5 1/3] [media] exynos-gsc: Use user configured colorspace if provided

[Thibault Saunier]

Hello,


On 02/21/2017 11:19 PM, Hans Verkuil wrote:
>
> On 02/21/2017 11:20 AM, Thibault Saunier wrote:
>> Use colorspace provided by the user as we are only doing scaling and
>> color encoding conversion, we won't be able to transform the colorspace
>> itself and the colorspace won't mater in that operation.
>>
>> Also always use output colorspace on the capture side.
>>
>> Start using 576p as a threashold to compute the colorspace.
>> The media documentation says that the V4L2_COLORSPACE_SMPTE170M colorspace
>> should be used for SDTV and V4L2_COLORSPACE_REC709 for HDTV. But drivers
>> don't agree on the display resolution that should be used as a threshold.
>>
>>  From EIA CEA 861B about colorimetry for various resolutions:
>>
>>    - 5.1 480p, 480i, 576p, 576i, 240p, and 288p
>>      The color space used by the 480-line, 576-line, 240-line, and 288-line
>>      formats will likely be based on SMPTE 170M [1].
>>    - 5.2 1080i, 1080p, and 720p
>>      The color space used by the high definition formats will likely be
>>      based on ITU-R BT.709-4
>>
>> This indicates that in the case that userspace does not specify what
>> colorspace should be used, we should use 576p  as a threshold to set
>> V4L2_COLORSPACE_REC709 instead of V4L2_COLORSPACE_SMPTE170M. Even if it is
>> only 'likely' and not a requirement it is the best guess we can make.
>>
>> The stream should have been encoded with the information and userspace
>> has to pass it to the driver if it is not the case, otherwise we won't be
>> able to handle it properly anyhow.
>>
>> Also, check for the resolution in G_FMT instead unconditionally setting
>> the V4L2_COLORSPACE_REC709 colorspace.
>>
>> Signed-off-by: Javier Martinez Canillas <javier@osg.samsung.com>
>> Signed-off-by: Thibault Saunier <thibault.saunier@osg.samsung.com>
>> Reviewed-by: Andrzej Hajda <a.hajda@samsung.com>
>>
>> ---
>>
>> Changes in v5:
>> - Squash commit to always use output colorspace on the capture side
>>    inside this one
>> - Fix typo in commit message
>>
>> Changes in v4:
>> - Reword commit message to better back our assumptions on specifications
>>
>> Changes in v3:
>> - Do not check values in the g_fmt functions as Andrzej explained in previous review
>> - Added 'Reviewed-by: Andrzej Hajda <a.hajda@samsung.com>'
>>
>> Changes in v2: None
>>
>>   drivers/media/platform/exynos-gsc/gsc-core.c | 20 +++++++++++++++-----
>>   drivers/media/platform/exynos-gsc/gsc-core.h |  1 +
>>   2 files changed, 16 insertions(+), 5 deletions(-)
>>
>> diff --git a/drivers/media/platform/exynos-gsc/gsc-core.c b/drivers/media/platform/exynos-gsc/gsc-core.c
>> index 59a634201830..772599de8c13 100644
>> --- a/drivers/media/platform/exynos-gsc/gsc-core.c
>> +++ b/drivers/media/platform/exynos-gsc/gsc-core.c
>> @@ -454,6 +454,7 @@ int gsc_try_fmt_mplane(struct gsc_ctx *ctx, struct v4l2_format *f)
>>   	} else {
>>   		min_w = variant->pix_min->target_rot_dis_w;
>>   		min_h = variant->pix_min->target_rot_dis_h;
>> +		pix_mp->colorspace = ctx->out_colorspace;
>>   	}
>>   
>>   	pr_debug("mod_x: %d, mod_y: %d, max_w: %d, max_h = %d",
>> @@ -472,10 +473,15 @@ int gsc_try_fmt_mplane(struct gsc_ctx *ctx, struct v4l2_format *f)
>>   
>>   	pix_mp->num_planes = fmt->num_planes;
>>   
>> -	if (pix_mp->width >= 1280) /* HD */
>> -		pix_mp->colorspace = V4L2_COLORSPACE_REC709;
>> -	else /* SD */
>> -		pix_mp->colorspace = V4L2_COLORSPACE_SMPTE170M;
>> +	if (pix_mp->colorspace == V4L2_COLORSPACE_DEFAULT) {
>> +		if (pix_mp->width > 720 && pix_mp->height > 576) /* HD */
> I'd use || instead of && here. Ditto for the next patch.
>
>> +			pix_mp->colorspace = V4L2_COLORSPACE_REC709;
>> +		else /* SD */
>> +			pix_mp->colorspace = V4L2_COLORSPACE_SMPTE170M;
>> +	}
> Are you sure this is in fact how it is used? If the source of the video
> is the sensor (camera), then the colorspace is typically SRGB. I'm not
> really sure you should guess here. All a mem2mem driver should do is to
> pass the colorspace etc. information from the output to the capture side,
> and not invent things. That simply makes no sense.

This is not a camera device but a colorspace conversion device, in many
cases the info will not be passed by the userspace, basically because 
the info
is not encoded in the media stream (this often happens). I am not 
inventing here,
just making sure we use the most likely value in case none was provided 
(and if none
was provided it should always be correct given that the encoded stream 
was not broken).

In the case the source is a camera and then we use the colorspace 
converter then the info
should copied from the camera to the transcoding node (by userspace) so 
there should be
no problem.

What I am doing here is what is done in other drivers.

Regards,

Thibault
>
> We may have to update the documentation or v4l2-compliance test if this
> is an issue. The more I think about this, the more I think we shouldn't
> do this guessing game.
>
> Regards,
>
> 	Hans
>
>> +
>> +	if (V4L2_TYPE_IS_OUTPUT(f->type))
>> +		ctx->out_colorspace = pix_mp->colorspace;
>>   
>>   	for (i = 0; i < pix_mp->num_planes; ++i) {
>>   		struct v4l2_plane_pix_format *plane_fmt = &pix_mp->plane_fmt[i];
>> @@ -519,9 +525,13 @@ int gsc_g_fmt_mplane(struct gsc_ctx *ctx, struct v4l2_format *f)
>>   	pix_mp->height		= frame->f_height;
>>   	pix_mp->field		= V4L2_FIELD_NONE;
>>   	pix_mp->pixelformat	= frame->fmt->pixelformat;
>> -	pix_mp->colorspace	= V4L2_COLORSPACE_REC709;
>>   	pix_mp->num_planes	= frame->fmt->num_planes;
>>   
>> +	if (pix_mp->width > 720 && pix_mp->height > 576) /* HD */
>> +		pix_mp->colorspace = V4L2_COLORSPACE_REC709;
>> +	else /* SD */
>> +		pix_mp->colorspace = V4L2_COLORSPACE_SMPTE170M;
>> +
>>   	for (i = 0; i < pix_mp->num_planes; ++i) {
>>   		pix_mp->plane_fmt[i].bytesperline = (frame->f_width *
>>   			frame->fmt->depth[i]) / 8;
>> diff --git a/drivers/media/platform/exynos-gsc/gsc-core.h b/drivers/media/platform/exynos-gsc/gsc-core.h
>> index 696217e9af66..715d9c9d8d30 100644
>> --- a/drivers/media/platform/exynos-gsc/gsc-core.h
>> +++ b/drivers/media/platform/exynos-gsc/gsc-core.h
>> @@ -376,6 +376,7 @@ struct gsc_ctx {
>>   	struct v4l2_ctrl_handler ctrl_handler;
>>   	struct gsc_ctrls	gsc_ctrls;
>>   	bool			ctrls_rdy;
>> +	enum v4l2_colorspace out_colorspace;
>>   };
>>   
>>   void gsc_set_prefbuf(struct gsc_dev *gsc, struct gsc_frame *frm);
>>

Re: [PATCH v5 3/3] [media] s5p-mfc: Check and set 'v4l2_pix_format:field' field in try_fmt

[Thibault Saunier]

Hello,

On 02/22/2017 06:29 AM, Andrzej Hajda wrote:
> On 21.02.2017 20:20, Thibault Saunier wrote:
>> It is required by the standard that the field order is set by the
>> driver.
>>
>> Signed-off-by: Thibault Saunier <thibault.saunier@osg.samsung.com>
>>
>> ---
>>
>> 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_dec.c | 3 +++
>>   1 file changed, 3 insertions(+)
>>
>> diff --git a/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c b/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c
>> index 0976c3e0a5ce..44ed2afe0780 100644
>> --- a/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c
>> +++ b/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c
>> @@ -386,6 +386,9 @@ static int vidioc_try_fmt(struct file *file, void *priv, struct v4l2_format *f)
>>   	struct v4l2_pix_format_mplane *pix_mp = &f->fmt.pix_mp;
>>   	struct s5p_mfc_fmt *fmt;
>>   
>> +	if (f->fmt.pix.field == V4L2_FIELD_ANY)
>> +		f->fmt.pix.field = V4L2_FIELD_NONE;
>> +
> In previous version the only supported field type was NONE, here you
> support everything.
> If the only supported format is none you should set 'field'
> unconditionally to NONE, nothing more.
Afaict we  support 2 things:

   1. NONE
   2. INTERLACE

Until now we were not checking what was supported or not and basically 
ignoring that info, this patch
keeps the old behaviour making sure to be compliant.

I had a doubt and was pondering doing:

``` diff

+	if (f->fmt.pix.field != V4L2_FIELD_INTERLACED)
+		f->fmt.pix.field = V4L2_FIELD_NONE;
+

```

instead, it is probably more correct, do you think it is what should be 
done here?

Regards,

Thibault

>
> Regards
> Andrzej
>
>>   	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);
>

Re: [PATCH v5 3/3] [media] s5p-mfc: Check and set 'v4l2_pix_format:field' field in try_fmt

[Nicolas Dufresne]

[-- Attachment #1: Type: text/plain, Size: 2492 bytes --]

Le mercredi 22 février 2017 à 10:10 -0300, Thibault Saunier a écrit :
> Hello,
> 
> On 02/22/2017 06:29 AM, Andrzej Hajda wrote:
> > On 21.02.2017 20:20, Thibault Saunier wrote:
> > > It is required by the standard that the field order is set by the
> > > driver.
> > > 
> > > Signed-off-by: Thibault Saunier <thibault.saunier@osg.samsung.com
> > > >
> > > 
> > > ---
> > > 
> > > 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_dec.c | 3 +++
> > >   1 file changed, 3 insertions(+)
> > > 
> > > diff --git a/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c
> > > b/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c
> > > index 0976c3e0a5ce..44ed2afe0780 100644
> > > --- a/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c
> > > +++ b/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c
> > > @@ -386,6 +386,9 @@ static int vidioc_try_fmt(struct file *file,
> > > void *priv, struct v4l2_format *f)
> > >   	struct v4l2_pix_format_mplane *pix_mp = &f->fmt.pix_mp;
> > >   	struct s5p_mfc_fmt *fmt;
> > >   
> > > +	if (f->fmt.pix.field == V4L2_FIELD_ANY)
> > > +		f->fmt.pix.field = V4L2_FIELD_NONE;
> > > +
> > 
> > In previous version the only supported field type was NONE, here
> > you
> > support everything.
> > If the only supported format is none you should set 'field'
> > unconditionally to NONE, nothing more.
> 
> Afaict we  support 2 things:
> 
>    1. NONE
>    2. INTERLACE
> 
> Until now we were not checking what was supported or not and
> basically 
> ignoring that info, this patch
> keeps the old behaviour making sure to be compliant.
> 
> I had a doubt and was pondering doing:
> 
> ``` diff
> 
> +	if (f->fmt.pix.field != V4L2_FIELD_INTERLACED)
> +		f->fmt.pix.field = V4L2_FIELD_NONE;
> +
> 

This looks better to me.

> ```
> 
> instead, it is probably more correct, do you think it is what should
> be 
> done here?
> 
> Regards,
> 
> Thibault
> 
> > 
> > Regards
> > Andrzej
> > 
> > >   	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);
> 
> 

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 181 bytes --]

Re: [PATCH v5 1/3] [media] exynos-gsc: Use user configured colorspace if provided

[Hans Verkuil]

On 02/22/2017 05:05 AM, Thibault Saunier wrote:
> Hello,
> 
> 
> On 02/21/2017 11:19 PM, Hans Verkuil wrote:
>>
>> On 02/21/2017 11:20 AM, Thibault Saunier wrote:
>>> Use colorspace provided by the user as we are only doing scaling and
>>> color encoding conversion, we won't be able to transform the colorspace
>>> itself and the colorspace won't mater in that operation.
>>>
>>> Also always use output colorspace on the capture side.
>>>
>>> Start using 576p as a threashold to compute the colorspace.
>>> The media documentation says that the V4L2_COLORSPACE_SMPTE170M colorspace
>>> should be used for SDTV and V4L2_COLORSPACE_REC709 for HDTV. But drivers
>>> don't agree on the display resolution that should be used as a threshold.
>>>
>>>  From EIA CEA 861B about colorimetry for various resolutions:
>>>
>>>    - 5.1 480p, 480i, 576p, 576i, 240p, and 288p
>>>      The color space used by the 480-line, 576-line, 240-line, and 288-line
>>>      formats will likely be based on SMPTE 170M [1].
>>>    - 5.2 1080i, 1080p, and 720p
>>>      The color space used by the high definition formats will likely be
>>>      based on ITU-R BT.709-4
>>>
>>> This indicates that in the case that userspace does not specify what
>>> colorspace should be used, we should use 576p  as a threshold to set
>>> V4L2_COLORSPACE_REC709 instead of V4L2_COLORSPACE_SMPTE170M. Even if it is
>>> only 'likely' and not a requirement it is the best guess we can make.
>>>
>>> The stream should have been encoded with the information and userspace
>>> has to pass it to the driver if it is not the case, otherwise we won't be
>>> able to handle it properly anyhow.
>>>
>>> Also, check for the resolution in G_FMT instead unconditionally setting
>>> the V4L2_COLORSPACE_REC709 colorspace.
>>>
>>> Signed-off-by: Javier Martinez Canillas <javier@osg.samsung.com>
>>> Signed-off-by: Thibault Saunier <thibault.saunier@osg.samsung.com>
>>> Reviewed-by: Andrzej Hajda <a.hajda@samsung.com>
>>>
>>> ---
>>>
>>> Changes in v5:
>>> - Squash commit to always use output colorspace on the capture side
>>>    inside this one
>>> - Fix typo in commit message
>>>
>>> Changes in v4:
>>> - Reword commit message to better back our assumptions on specifications
>>>
>>> Changes in v3:
>>> - Do not check values in the g_fmt functions as Andrzej explained in previous review
>>> - Added 'Reviewed-by: Andrzej Hajda <a.hajda@samsung.com>'
>>>
>>> Changes in v2: None
>>>
>>>   drivers/media/platform/exynos-gsc/gsc-core.c | 20 +++++++++++++++-----
>>>   drivers/media/platform/exynos-gsc/gsc-core.h |  1 +
>>>   2 files changed, 16 insertions(+), 5 deletions(-)
>>>
>>> diff --git a/drivers/media/platform/exynos-gsc/gsc-core.c b/drivers/media/platform/exynos-gsc/gsc-core.c
>>> index 59a634201830..772599de8c13 100644
>>> --- a/drivers/media/platform/exynos-gsc/gsc-core.c
>>> +++ b/drivers/media/platform/exynos-gsc/gsc-core.c
>>> @@ -454,6 +454,7 @@ int gsc_try_fmt_mplane(struct gsc_ctx *ctx, struct v4l2_format *f)
>>>       } else {
>>>           min_w = variant->pix_min->target_rot_dis_w;
>>>           min_h = variant->pix_min->target_rot_dis_h;
>>> +        pix_mp->colorspace = ctx->out_colorspace;
>>>       }
>>>         pr_debug("mod_x: %d, mod_y: %d, max_w: %d, max_h = %d",
>>> @@ -472,10 +473,15 @@ int gsc_try_fmt_mplane(struct gsc_ctx *ctx, struct v4l2_format *f)
>>>         pix_mp->num_planes = fmt->num_planes;
>>>   -    if (pix_mp->width >= 1280) /* HD */
>>> -        pix_mp->colorspace = V4L2_COLORSPACE_REC709;
>>> -    else /* SD */
>>> -        pix_mp->colorspace = V4L2_COLORSPACE_SMPTE170M;
>>> +    if (pix_mp->colorspace == V4L2_COLORSPACE_DEFAULT) {
>>> +        if (pix_mp->width > 720 && pix_mp->height > 576) /* HD */
>> I'd use || instead of && here. Ditto for the next patch.
>>
>>> +            pix_mp->colorspace = V4L2_COLORSPACE_REC709;
>>> +        else /* SD */
>>> +            pix_mp->colorspace = V4L2_COLORSPACE_SMPTE170M;
>>> +    }
>> Are you sure this is in fact how it is used? If the source of the video
>> is the sensor (camera), then the colorspace is typically SRGB. I'm not
>> really sure you should guess here. All a mem2mem driver should do is to
>> pass the colorspace etc. information from the output to the capture side,
>> and not invent things. That simply makes no sense.
> 
> This is not a camera device but a colorspace conversion device, in many

Not really, this is a color encoding conversion device. I.e. it only affects
the Y'CbCr encoding and quantization range. The colorspace (aka chromaticities)
and transfer function remain unchanged.

In fact, I suspect (correct me if I am wrong) that it only converts between
RGB, Y'CbCr 601 and Y'CbCr 709 encodings, where RGB is full range and the Y'CbCr
formats are limited range.

If you pass in limited range RGB it will probably do the wrong thing as I don't
seen any quantization checks in the code.

So the colorspace and xfer_func fields remain unchanged by this driver.

If you want to do this really correctly, then you need to do more. I don't have
time right now to go into this in-depth, but I will try to get back to this on
Monday. I am thinking of documenting this as part of the V4L2 colorspace
documentation. This stuff is complex and if you don't know the rules then it
can be hard to implement correctly.

> cases the info will not be passed by the userspace, basically because the info
> is not encoded in the media stream (this often happens). I am not inventing here,
> just making sure we use the most likely value in case none was provided (and if none
> was provided it should always be correct given that the encoded stream was not broken).
> 
> In the case the source is a camera and then we use the colorspace converter then the info
> should copied from the camera to the transcoding node (by userspace) so there should be
> no problem.
> 
> What I am doing here is what is done in other drivers.

Most (all?) other mem2mem drivers do not do color encoding conversion, they just copy
all the colorspace info from output to capture.

If there are other m2m drivers that do this, then I doubt they do the right thing.
They are probably right for most cases, but not all.

Regards,

	Hans

Re: [PATCH v5 1/3] [media] exynos-gsc: Use user configured colorspace if provided

[Thibault Saunier]

On 02/22/2017 03:06 PM, Hans Verkuil wrote:
>
> On 02/22/2017 05:05 AM, Thibault Saunier wrote:
>> Hello,
>>
>>
>> On 02/21/2017 11:19 PM, Hans Verkuil wrote:
>>> On 02/21/2017 11:20 AM, Thibault Saunier wrote:
>>>> Use colorspace provided by the user as we are only doing scaling and
>>>> color encoding conversion, we won't be able to transform the colorspace
>>>> itself and the colorspace won't mater in that operation.
>>>>
>>>> Also always use output colorspace on the capture side.
>>>>
>>>> Start using 576p as a threashold to compute the colorspace.
>>>> The media documentation says that the V4L2_COLORSPACE_SMPTE170M colorspace
>>>> should be used for SDTV and V4L2_COLORSPACE_REC709 for HDTV. But drivers
>>>> don't agree on the display resolution that should be used as a threshold.
>>>>
>>>>   From EIA CEA 861B about colorimetry for various resolutions:
>>>>
>>>>     - 5.1 480p, 480i, 576p, 576i, 240p, and 288p
>>>>       The color space used by the 480-line, 576-line, 240-line, and 288-line
>>>>       formats will likely be based on SMPTE 170M [1].
>>>>     - 5.2 1080i, 1080p, and 720p
>>>>       The color space used by the high definition formats will likely be
>>>>       based on ITU-R BT.709-4
>>>>
>>>> This indicates that in the case that userspace does not specify what
>>>> colorspace should be used, we should use 576p  as a threshold to set
>>>> V4L2_COLORSPACE_REC709 instead of V4L2_COLORSPACE_SMPTE170M. Even if it is
>>>> only 'likely' and not a requirement it is the best guess we can make.
>>>>
>>>> The stream should have been encoded with the information and userspace
>>>> has to pass it to the driver if it is not the case, otherwise we won't be
>>>> able to handle it properly anyhow.
>>>>
>>>> Also, check for the resolution in G_FMT instead unconditionally setting
>>>> the V4L2_COLORSPACE_REC709 colorspace.
>>>>
>>>> Signed-off-by: Javier Martinez Canillas <javier@osg.samsung.com>
>>>> Signed-off-by: Thibault Saunier <thibault.saunier@osg.samsung.com>
>>>> Reviewed-by: Andrzej Hajda <a.hajda@samsung.com>
>>>>
>>>> ---
>>>>
>>>> Changes in v5:
>>>> - Squash commit to always use output colorspace on the capture side
>>>>     inside this one
>>>> - Fix typo in commit message
>>>>
>>>> Changes in v4:
>>>> - Reword commit message to better back our assumptions on specifications
>>>>
>>>> Changes in v3:
>>>> - Do not check values in the g_fmt functions as Andrzej explained in previous review
>>>> - Added 'Reviewed-by: Andrzej Hajda <a.hajda@samsung.com>'
>>>>
>>>> Changes in v2: None
>>>>
>>>>    drivers/media/platform/exynos-gsc/gsc-core.c | 20 +++++++++++++++-----
>>>>    drivers/media/platform/exynos-gsc/gsc-core.h |  1 +
>>>>    2 files changed, 16 insertions(+), 5 deletions(-)
>>>>
>>>> diff --git a/drivers/media/platform/exynos-gsc/gsc-core.c b/drivers/media/platform/exynos-gsc/gsc-core.c
>>>> index 59a634201830..772599de8c13 100644
>>>> --- a/drivers/media/platform/exynos-gsc/gsc-core.c
>>>> +++ b/drivers/media/platform/exynos-gsc/gsc-core.c
>>>> @@ -454,6 +454,7 @@ int gsc_try_fmt_mplane(struct gsc_ctx *ctx, struct v4l2_format *f)
>>>>        } else {
>>>>            min_w = variant->pix_min->target_rot_dis_w;
>>>>            min_h = variant->pix_min->target_rot_dis_h;
>>>> +        pix_mp->colorspace = ctx->out_colorspace;
>>>>        }
>>>>          pr_debug("mod_x: %d, mod_y: %d, max_w: %d, max_h = %d",
>>>> @@ -472,10 +473,15 @@ int gsc_try_fmt_mplane(struct gsc_ctx *ctx, struct v4l2_format *f)
>>>>          pix_mp->num_planes = fmt->num_planes;
>>>>    -    if (pix_mp->width >= 1280) /* HD */
>>>> -        pix_mp->colorspace = V4L2_COLORSPACE_REC709;
>>>> -    else /* SD */
>>>> -        pix_mp->colorspace = V4L2_COLORSPACE_SMPTE170M;
>>>> +    if (pix_mp->colorspace == V4L2_COLORSPACE_DEFAULT) {
>>>> +        if (pix_mp->width > 720 && pix_mp->height > 576) /* HD */
>>> I'd use || instead of && here. Ditto for the next patch.
>>>
>>>> +            pix_mp->colorspace = V4L2_COLORSPACE_REC709;
>>>> +        else /* SD */
>>>> +            pix_mp->colorspace = V4L2_COLORSPACE_SMPTE170M;
>>>> +    }
>>> Are you sure this is in fact how it is used? If the source of the video
>>> is the sensor (camera), then the colorspace is typically SRGB. I'm not
>>> really sure you should guess here. All a mem2mem driver should do is to
>>> pass the colorspace etc. information from the output to the capture side,
>>> and not invent things. That simply makes no sense.
>> This is not a camera device but a colorspace conversion device, in many
> Not really, this is a color encoding conversion device. I.e. it only affects
> the Y'CbCr encoding and quantization range. The colorspace (aka chromaticities)
> and transfer function remain unchanged.

Well, right, sorry I am talking in GStreamer terminlogy where what you call
colorspace is called colorimetry, and colorspace is what I am talking 
about here.

> In fact, I suspect (correct me if I am wrong) that it only converts between
> RGB, Y'CbCr 601 and Y'CbCr 709 encodings, where RGB is full range and the Y'CbCr
> formats are limited range.
That sounds correct.

> If you pass in limited range RGB it will probably do the wrong thing as I don't
> seen any quantization checks in the code.
>
> So the colorspace and xfer_func fields remain unchanged by this driver.
>
> If you want to do this really correctly, then you need to do more. I don't have
> time right now to go into this in-depth, but I will try to get back to this on
> Monday. I am thinking of documenting this as part of the V4L2 colorspace
> documentation. This stuff is complex and if you don't know the rules then it
> can be hard to implement correctly.

Here I am just making sure that we set the colorspace (v4l2 terminology)
if the case the user is passing DEFAULT (meaning he does not know what
it should be and lets the driver set it, this will happen in the case 
where the
information was not contain in the source, which means the value has to be
guessed, and here I am simply doing an educated guess relying on what the
previously named standard suggests.

I would be happy to read more information about that subject and will
try to fix remaining suggested points once I have a better understanding
of the whole concepts and problems in that driver, but I still think 
that this
patch is correct for what it is aiming at fixing.

Regards,

Thibault Saunier

>> cases the info will not be passed by the userspace, basically because the info
>> is not encoded in the media stream (this often happens). I am not inventing here,
>> just making sure we use the most likely value in case none was provided (and if none
>> was provided it should always be correct given that the encoded stream was not broken).
>>
>> In the case the source is a camera and then we use the colorspace converter then the info
>> should copied from the camera to the transcoding node (by userspace) so there should be
>> no problem.
>>
>> What I am doing here is what is done in other drivers.
> Most (all?) other mem2mem drivers do not do color encoding conversion, they just copy
> all the colorspace info from output to capture.
>
> If there are other m2m drivers that do this, then I doubt they do the right thing.
> They are probably right for most cases, but not all.
>
> Regards,
>
> 	Hans

Re: [PATCH v5 1/3] [media] exynos-gsc: Use user configured colorspace if provided

[Nicolas Dufresne]

[-- Attachment #1: Type: text/plain, Size: 9960 bytes --]

Le mercredi 22 février 2017 à 15:57 -0300, Thibault Saunier a écrit :
> On 02/22/2017 03:06 PM, Hans Verkuil wrote:
> > 
> > On 02/22/2017 05:05 AM, Thibault Saunier wrote:
> > > Hello,
> > > 
> > > 
> > > On 02/21/2017 11:19 PM, Hans Verkuil wrote:
> > > > On 02/21/2017 11:20 AM, Thibault Saunier wrote:
> > > > > Use colorspace provided by the user as we are only doing
> > > > > scaling and
> > > > > color encoding conversion, we won't be able to transform the
> > > > > colorspace
> > > > > itself and the colorspace won't mater in that operation.
> > > > > 
> > > > > Also always use output colorspace on the capture side.
> > > > > 
> > > > > Start using 576p as a threashold to compute the colorspace.
> > > > > The media documentation says that the
> > > > > V4L2_COLORSPACE_SMPTE170M colorspace
> > > > > should be used for SDTV and V4L2_COLORSPACE_REC709 for HDTV.
> > > > > But drivers
> > > > > don't agree on the display resolution that should be used as
> > > > > a threshold.
> > > > > 
> > > > >   From EIA CEA 861B about colorimetry for various
> > > > > resolutions:
> > > > > 
> > > > >     - 5.1 480p, 480i, 576p, 576i, 240p, and 288p
> > > > >       The color space used by the 480-line, 576-line, 240-
> > > > > line, and 288-line
> > > > >       formats will likely be based on SMPTE 170M [1].
> > > > >     - 5.2 1080i, 1080p, and 720p
> > > > >       The color space used by the high definition formats
> > > > > will likely be
> > > > >       based on ITU-R BT.709-4
> > > > > 
> > > > > This indicates that in the case that userspace does not
> > > > > specify what
> > > > > colorspace should be used, we should use 576p  as a threshold
> > > > > to set
> > > > > V4L2_COLORSPACE_REC709 instead of V4L2_COLORSPACE_SMPTE170M.
> > > > > Even if it is
> > > > > only 'likely' and not a requirement it is the best guess we
> > > > > can make.
> > > > > 
> > > > > The stream should have been encoded with the information and
> > > > > userspace
> > > > > has to pass it to the driver if it is not the case, otherwise
> > > > > we won't be
> > > > > able to handle it properly anyhow.
> > > > > 
> > > > > Also, check for the resolution in G_FMT instead
> > > > > unconditionally setting
> > > > > the V4L2_COLORSPACE_REC709 colorspace.
> > > > > 
> > > > > Signed-off-by: Javier Martinez Canillas <javier@osg.samsung.c
> > > > > om>
> > > > > Signed-off-by: Thibault Saunier <thibault.saunier@osg.samsung
> > > > > .com>
> > > > > Reviewed-by: Andrzej Hajda <a.hajda@samsung.com>
> > > > > 
> > > > > ---
> > > > > 
> > > > > Changes in v5:
> > > > > - Squash commit to always use output colorspace on the
> > > > > capture side
> > > > >     inside this one
> > > > > - Fix typo in commit message
> > > > > 
> > > > > Changes in v4:
> > > > > - Reword commit message to better back our assumptions on
> > > > > specifications
> > > > > 
> > > > > Changes in v3:
> > > > > - Do not check values in the g_fmt functions as Andrzej
> > > > > explained in previous review
> > > > > - Added 'Reviewed-by: Andrzej Hajda <a.hajda@samsung.com>'
> > > > > 
> > > > > Changes in v2: None
> > > > > 
> > > > >    drivers/media/platform/exynos-gsc/gsc-core.c | 20
> > > > > +++++++++++++++-----
> > > > >    drivers/media/platform/exynos-gsc/gsc-core.h |  1 +
> > > > >    2 files changed, 16 insertions(+), 5 deletions(-)
> > > > > 
> > > > > diff --git a/drivers/media/platform/exynos-gsc/gsc-core.c
> > > > > b/drivers/media/platform/exynos-gsc/gsc-core.c
> > > > > index 59a634201830..772599de8c13 100644
> > > > > --- a/drivers/media/platform/exynos-gsc/gsc-core.c
> > > > > +++ b/drivers/media/platform/exynos-gsc/gsc-core.c
> > > > > @@ -454,6 +454,7 @@ int gsc_try_fmt_mplane(struct gsc_ctx
> > > > > *ctx, struct v4l2_format *f)
> > > > >        } else {
> > > > >            min_w = variant->pix_min->target_rot_dis_w;
> > > > >            min_h = variant->pix_min->target_rot_dis_h;
> > > > > +        pix_mp->colorspace = ctx->out_colorspace;
> > > > >        }
> > > > >          pr_debug("mod_x: %d, mod_y: %d, max_w: %d, max_h =
> > > > > %d",
> > > > > @@ -472,10 +473,15 @@ int gsc_try_fmt_mplane(struct gsc_ctx
> > > > > *ctx, struct v4l2_format *f)
> > > > >          pix_mp->num_planes = fmt->num_planes;
> > > > >    -    if (pix_mp->width >= 1280) /* HD */
> > > > > -        pix_mp->colorspace = V4L2_COLORSPACE_REC709;
> > > > > -    else /* SD */
> > > > > -        pix_mp->colorspace = V4L2_COLORSPACE_SMPTE170M;
> > > > > +    if (pix_mp->colorspace == V4L2_COLORSPACE_DEFAULT) {
> > > > > +        if (pix_mp->width > 720 && pix_mp->height > 576) /*
> > > > > HD */
> > > > 
> > > > I'd use || instead of && here. Ditto for the next patch.
> > > > 
> > > > > +            pix_mp->colorspace = V4L2_COLORSPACE_REC709;
> > > > > +        else /* SD */
> > > > > +            pix_mp->colorspace = V4L2_COLORSPACE_SMPTE170M;
> > > > > +    }
> > > > 
> > > > Are you sure this is in fact how it is used? If the source of
> > > > the video
> > > > is the sensor (camera), then the colorspace is typically SRGB.
> > > > I'm not
> > > > really sure you should guess here. All a mem2mem driver should
> > > > do is to
> > > > pass the colorspace etc. information from the output to the
> > > > capture side,
> > > > and not invent things. That simply makes no sense.
> > > 
> > > This is not a camera device but a colorspace conversion device,
> > > in many
> > 
> > Not really, this is a color encoding conversion device. I.e. it
> > only affects
> > the Y'CbCr encoding and quantization range. The colorspace (aka
> > chromaticities)
> > and transfer function remain unchanged.
> 
> Well, right, sorry I am talking in GStreamer terminlogy where what
> you call
> colorspace is called colorimetry, and colorspace is what I am
> talking 
> about here.
> 
> > In fact, I suspect (correct me if I am wrong) that it only converts
> > between
> > RGB, Y'CbCr 601 and Y'CbCr 709 encodings, where RGB is full range
> > and the Y'CbCr
> > formats are limited range.
> 
> That sounds correct.
> 
> > If you pass in limited range RGB it will probably do the wrong
> > thing as I don't
> > seen any quantization checks in the code.
> > 
> > So the colorspace and xfer_func fields remain unchanged by this
> > driver.
> > 
> > If you want to do this really correctly, then you need to do more.
> > I don't have
> > time right now to go into this in-depth, but I will try to get back
> > to this on
> > Monday. I am thinking of documenting this as part of the V4L2
> > colorspace
> > documentation. This stuff is complex and if you don't know the
> > rules then it
> > can be hard to implement correctly.
> 
> Here I am just making sure that we set the colorspace (v4l2
> terminology)
> if the case the user is passing DEFAULT (meaning he does not know
> what
> it should be and lets the driver set it, this will happen in the
> case 
> where the
> information was not contain in the source, which means the value has
> to be
> guessed, and here I am simply doing an educated guess relying on what
> the
> previously named standard suggests.
> 
> I would be happy to read more information about that subject and will
> try to fix remaining suggested points once I have a better
> understanding
> of the whole concepts and problems in that driver, but I still think 
> that this
> patch is correct for what it is aiming at fixing.
> 
> Regards,
> 
> Thibault Saunier
> 
> > > cases the info will not be passed by the userspace, basically
> > > because the info
> > > is not encoded in the media stream (this often happens). I am not
> > > inventing here,
> > > just making sure we use the most likely value in case none was
> > > provided (and if none
> > > was provided it should always be correct given that the encoded
> > > stream was not broken).
> > > 
> > > In the case the source is a camera and then we use the colorspace
> > > converter then the info
> > > should copied from the camera to the transcoding node (by
> > > userspace) so there should be
> > > no problem.
> > > 
> > > What I am doing here is what is done in other drivers.
> > 
> > Most (all?) other mem2mem drivers do not do color encoding
> > conversion, they just copy
> > all the colorspace info from output to capture.
> > 
> > If there are other m2m drivers that do this, then I doubt they do
> > the right thing.
> > They are probably right for most cases, but not all.
> > 
> > Regards,
> > 
> > 	Hans

I think Hans is right about the colorspace here. Unlike the field for
interlace modes, there it is no mandatory requirement in the spec for
the driver to change DEFAULT into something else. If you don't know the
input colorspace, then you don't known the output. Even in GStreamer,
this would be best, since the only guessing will happen at the display,
or in some conversion to RGB. Though, if you don't support a specific
value, then the driver should change it.

Where I don't agree though, is that an m2m color converter will always
copy blindly that colorspace. Even though the support is limited,
sometime it may change, this need to be sorted case by case. I have
rarely seen such a driver keep the limited range when turning buffers
into RGB data.

But changing the colorspace is complicated, since it's slightly vague
concept. It will in fact imply a specific set of v4l2_xfer_func,
v4l2_ycbcr_encoding and v4l2_quantization (ignoring hsv, as it does not
apply). If you have a limited driver, that only target one of these 3
items, you will likely keep the colorspace as-is, and override the
specific functions.

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 181 bytes --]

Re: [PATCH v5 1/3] [media] exynos-gsc: Use user configured colorspace if provided

[Thibault Saunier]

On 02/22/2017 05:03 PM, Nicolas Dufresne wrote:
> Le mercredi 22 février 2017 à 15:57 -0300, Thibault Saunier a écrit :
>> On 02/22/2017 03:06 PM, Hans Verkuil wrote:
>>> On 02/22/2017 05:05 AM, Thibault Saunier wrote:
>>>> Hello,
>>>>
>>>>
>>>> On 02/21/2017 11:19 PM, Hans Verkuil wrote:
>>>>> On 02/21/2017 11:20 AM, Thibault Saunier wrote:
>>>>>> Use colorspace provided by the user as we are only doing
>>>>>> scaling and
>>>>>> color encoding conversion, we won't be able to transform the
>>>>>> colorspace
>>>>>> itself and the colorspace won't mater in that operation.
>>>>>>
>>>>>> Also always use output colorspace on the capture side.
>>>>>>
>>>>>> Start using 576p as a threashold to compute the colorspace.
>>>>>> The media documentation says that the
>>>>>> V4L2_COLORSPACE_SMPTE170M colorspace
>>>>>> should be used for SDTV and V4L2_COLORSPACE_REC709 for HDTV.
>>>>>> But drivers
>>>>>> don't agree on the display resolution that should be used as
>>>>>> a threshold.
>>>>>>
>>>>>>    From EIA CEA 861B about colorimetry for various
>>>>>> resolutions:
>>>>>>
>>>>>>      - 5.1 480p, 480i, 576p, 576i, 240p, and 288p
>>>>>>        The color space used by the 480-line, 576-line, 240-
>>>>>> line, and 288-line
>>>>>>        formats will likely be based on SMPTE 170M [1].
>>>>>>      - 5.2 1080i, 1080p, and 720p
>>>>>>        The color space used by the high definition formats
>>>>>> will likely be
>>>>>>        based on ITU-R BT.709-4
>>>>>>
>>>>>> This indicates that in the case that userspace does not
>>>>>> specify what
>>>>>> colorspace should be used, we should use 576p  as a threshold
>>>>>> to set
>>>>>> V4L2_COLORSPACE_REC709 instead of V4L2_COLORSPACE_SMPTE170M.
>>>>>> Even if it is
>>>>>> only 'likely' and not a requirement it is the best guess we
>>>>>> can make.
>>>>>>
>>>>>> The stream should have been encoded with the information and
>>>>>> userspace
>>>>>> has to pass it to the driver if it is not the case, otherwise
>>>>>> we won't be
>>>>>> able to handle it properly anyhow.
>>>>>>
>>>>>> Also, check for the resolution in G_FMT instead
>>>>>> unconditionally setting
>>>>>> the V4L2_COLORSPACE_REC709 colorspace.
>>>>>>
>>>>>> Signed-off-by: Javier Martinez Canillas <javier@osg.samsung.c
>>>>>> om>
>>>>>> Signed-off-by: Thibault Saunier <thibault.saunier@osg.samsung
>>>>>> .com>
>>>>>> Reviewed-by: Andrzej Hajda <a.hajda@samsung.com>
>>>>>>
>>>>>> ---
>>>>>>
>>>>>> Changes in v5:
>>>>>> - Squash commit to always use output colorspace on the
>>>>>> capture side
>>>>>>      inside this one
>>>>>> - Fix typo in commit message
>>>>>>
>>>>>> Changes in v4:
>>>>>> - Reword commit message to better back our assumptions on
>>>>>> specifications
>>>>>>
>>>>>> Changes in v3:
>>>>>> - Do not check values in the g_fmt functions as Andrzej
>>>>>> explained in previous review
>>>>>> - Added 'Reviewed-by: Andrzej Hajda <a.hajda@samsung.com>'
>>>>>>
>>>>>> Changes in v2: None
>>>>>>
>>>>>>     drivers/media/platform/exynos-gsc/gsc-core.c | 20
>>>>>> +++++++++++++++-----
>>>>>>     drivers/media/platform/exynos-gsc/gsc-core.h |  1 +
>>>>>>     2 files changed, 16 insertions(+), 5 deletions(-)
>>>>>>
>>>>>> diff --git a/drivers/media/platform/exynos-gsc/gsc-core.c
>>>>>> b/drivers/media/platform/exynos-gsc/gsc-core.c
>>>>>> index 59a634201830..772599de8c13 100644
>>>>>> --- a/drivers/media/platform/exynos-gsc/gsc-core.c
>>>>>> +++ b/drivers/media/platform/exynos-gsc/gsc-core.c
>>>>>> @@ -454,6 +454,7 @@ int gsc_try_fmt_mplane(struct gsc_ctx
>>>>>> *ctx, struct v4l2_format *f)
>>>>>>         } else {
>>>>>>             min_w = variant->pix_min->target_rot_dis_w;
>>>>>>             min_h = variant->pix_min->target_rot_dis_h;
>>>>>> +        pix_mp->colorspace = ctx->out_colorspace;
>>>>>>         }
>>>>>>           pr_debug("mod_x: %d, mod_y: %d, max_w: %d, max_h =
>>>>>> %d",
>>>>>> @@ -472,10 +473,15 @@ int gsc_try_fmt_mplane(struct gsc_ctx
>>>>>> *ctx, struct v4l2_format *f)
>>>>>>           pix_mp->num_planes = fmt->num_planes;
>>>>>>     -    if (pix_mp->width >= 1280) /* HD */
>>>>>> -        pix_mp->colorspace = V4L2_COLORSPACE_REC709;
>>>>>> -    else /* SD */
>>>>>> -        pix_mp->colorspace = V4L2_COLORSPACE_SMPTE170M;
>>>>>> +    if (pix_mp->colorspace == V4L2_COLORSPACE_DEFAULT) {
>>>>>> +        if (pix_mp->width > 720 && pix_mp->height > 576) /*
>>>>>> HD */
>>>>> I'd use || instead of && here. Ditto for the next patch.
>>>>>
>>>>>> +            pix_mp->colorspace = V4L2_COLORSPACE_REC709;
>>>>>> +        else /* SD */
>>>>>> +            pix_mp->colorspace = V4L2_COLORSPACE_SMPTE170M;
>>>>>> +    }
>>>>> Are you sure this is in fact how it is used? If the source of
>>>>> the video
>>>>> is the sensor (camera), then the colorspace is typically SRGB.
>>>>> I'm not
>>>>> really sure you should guess here. All a mem2mem driver should
>>>>> do is to
>>>>> pass the colorspace etc. information from the output to the
>>>>> capture side,
>>>>> and not invent things. That simply makes no sense.
>>>> This is not a camera device but a colorspace conversion device,
>>>> in many
>>> Not really, this is a color encoding conversion device. I.e. it
>>> only affects
>>> the Y'CbCr encoding and quantization range. The colorspace (aka
>>> chromaticities)
>>> and transfer function remain unchanged.
>> Well, right, sorry I am talking in GStreamer terminlogy where what
>> you call
>> colorspace is called colorimetry, and colorspace is what I am
>> talking
>> about here.
>>
>>> In fact, I suspect (correct me if I am wrong) that it only converts
>>> between
>>> RGB, Y'CbCr 601 and Y'CbCr 709 encodings, where RGB is full range
>>> and the Y'CbCr
>>> formats are limited range.
>> That sounds correct.
>>
>>> If you pass in limited range RGB it will probably do the wrong
>>> thing as I don't
>>> seen any quantization checks in the code.
>>>
>>> So the colorspace and xfer_func fields remain unchanged by this
>>> driver.
>>>
>>> If you want to do this really correctly, then you need to do more.
>>> I don't have
>>> time right now to go into this in-depth, but I will try to get back
>>> to this on
>>> Monday. I am thinking of documenting this as part of the V4L2
>>> colorspace
>>> documentation. This stuff is complex and if you don't know the
>>> rules then it
>>> can be hard to implement correctly.
>> Here I am just making sure that we set the colorspace (v4l2
>> terminology)
>> if the case the user is passing DEFAULT (meaning he does not know
>> what
>> it should be and lets the driver set it, this will happen in the
>> case
>> where the
>> information was not contain in the source, which means the value has
>> to be
>> guessed, and here I am simply doing an educated guess relying on what
>> the
>> previously named standard suggests.
>>
>> I would be happy to read more information about that subject and will
>> try to fix remaining suggested points once I have a better
>> understanding
>> of the whole concepts and problems in that driver, but I still think
>> that this
>> patch is correct for what it is aiming at fixing.
>>
>> Regards,
>>
>> Thibault Saunier
>>
>>>> cases the info will not be passed by the userspace, basically
>>>> because the info
>>>> is not encoded in the media stream (this often happens). I am not
>>>> inventing here,
>>>> just making sure we use the most likely value in case none was
>>>> provided (and if none
>>>> was provided it should always be correct given that the encoded
>>>> stream was not broken).
>>>>
>>>> In the case the source is a camera and then we use the colorspace
>>>> converter then the info
>>>> should copied from the camera to the transcoding node (by
>>>> userspace) so there should be
>>>> no problem.
>>>>
>>>> What I am doing here is what is done in other drivers.
>>> Most (all?) other mem2mem drivers do not do color encoding
>>> conversion, they just copy
>>> all the colorspace info from output to capture.
>>>
>>> If there are other m2m drivers that do this, then I doubt they do
>>> the right thing.
>>> They are probably right for most cases, but not all.
>>>
>>> Regards,
>>>
>>> 	Hans
> I think Hans is right about the colorspace here. Unlike the field for
> interlace modes, there it is no mandatory requirement in the spec for
> the driver to change DEFAULT into something else. If you don't know the
> input colorspace, then you don't known the output. Even in GStreamer,
> this would be best, since the only guessing will happen at the display,
> or in some conversion to RGB. Though, if you don't support a specific
> value, then the driver should change it.

Well the documentation, even if not really precise states:

    V4L2_COLORSPACE_DEFAULT: The default colorspace. This can be used by 
applications to let the driver fill in the colorspace.

Which is why I supposed the driver should fill it in, also this is what 
other drivers do too.

Regards,

Thibault
> Where I don't agree though, is that an m2m color converter will always
> copy blindly that colorspace. Even though the support is limited,
> sometime it may change, this need to be sorted case by case. I have
> rarely seen such a driver keep the limited range when turning buffers
> into RGB data.
>
> But changing the colorspace is complicated, since it's slightly vague
> concept. It will in fact imply a specific set of v4l2_xfer_func,
> v4l2_ycbcr_encoding and v4l2_quantization (ignoring hsv, as it does not
> apply). If you have a limited driver, that only target one of these 3
> items, you will likely keep the colorspace as-is, and override the
> specific functions.

Re: [PATCH v5 3/3] [media] s5p-mfc: Check and set 'v4l2_pix_format:field' field in try_fmt

[Andrzej Hajda]

On 22.02.2017 14:10, Thibault Saunier wrote:
> Hello,
>
> On 02/22/2017 06:29 AM, Andrzej Hajda wrote:
>> On 21.02.2017 20:20, Thibault Saunier wrote:
>>> It is required by the standard that the field order is set by the
>>> driver.
>>>
>>> Signed-off-by: Thibault Saunier <thibault.saunier@osg.samsung.com>
>>>
>>> ---
>>>
>>> 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_dec.c | 3 +++
>>>   1 file changed, 3 insertions(+)
>>>
>>> diff --git a/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c b/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c
>>> index 0976c3e0a5ce..44ed2afe0780 100644
>>> --- a/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c
>>> +++ b/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c
>>> @@ -386,6 +386,9 @@ static int vidioc_try_fmt(struct file *file, void *priv, struct v4l2_format *f)
>>>   	struct v4l2_pix_format_mplane *pix_mp = &f->fmt.pix_mp;
>>>   	struct s5p_mfc_fmt *fmt;
>>>   
>>> +	if (f->fmt.pix.field == V4L2_FIELD_ANY)
>>> +		f->fmt.pix.field = V4L2_FIELD_NONE;
>>> +
>> In previous version the only supported field type was NONE, here you
>> support everything.
>> If the only supported format is none you should set 'field'
>> unconditionally to NONE, nothing more.
> Afaict we  support 2 things:
>
>    1. NONE
>    2. INTERLACE
>
> Until now we were not checking what was supported or not and basically 
> ignoring that info, this patch
> keeps the old behaviour making sure to be compliant.
>
> I had a doubt and was pondering doing:
>
> ``` diff
>
> +	if (f->fmt.pix.field != V4L2_FIELD_INTERLACED)
> +		f->fmt.pix.field = V4L2_FIELD_NONE;
> +
>
> ```
>
> instead, it is probably more correct, do you think it is what should be 
> done here?

I have realized I have forgot that it is not simple mem2mem device, it
is decoder. It is feed with compressed data which is just byte stream,
the only valid field value on output side is V4L2_FIELD_NONE.
About possible interlacing we could only say in case of capture side.
And in this case there are multiple questions:
- how MFC decodes stream with interlaced content?
- is it possible to convince it to decode it as user asks?
- does it perform (de-)interlacing?
- what is implemented in the driver?

After answering above questions we will be able to say how fmt.pix.field
should be treated on capture side.
And similar question we can pose in case of encoder.

So simple patch and so many doubts.

Regards
Andrzej

>
> Regards,
>
> Thibault
>
>> Regards
>> Andrzej
>>
>>>   	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);
> --
> To unsubscribe from this list: send the line "unsubscribe linux-samsung-soc" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>
>

Re: [PATCH v5 2/3] [media] s5p-mfc: Set colorspace in VIDIO_{G,TRY}_FMT if DEFAULT provided

[Andrzej Hajda]

On 21.02.2017 20:20, Thibault Saunier wrote:
> The media documentation says that the V4L2_COLORSPACE_SMPTE170M colorspace
> should be used for SDTV and V4L2_COLORSPACE_REC709 for HDTV but the driver
> didn't set the colorimetry when userspace provided
> V4L2_COLORSPACE_DEFAULT.
>
> Use 576p display resolution as a threshold to set this as suggested by
> EIA CEA 861B.
>
> Signed-off-by: Thibault Saunier <thibault.saunier@osg.samsung.com>
>
> ---
>
> Changes in v5: None
> Changes in v4:
> - Set the colorspace only if the user passed V4L2_COLORSPACE_DEFAULT, in
>   all other cases just use what userspace provided.
>
> Changes in v3:
> - Do not check values in the g_fmt functions as Andrzej explained in previous review
> - Set colorspace if user passed V4L2_COLORSPACE_DEFAULT in
>
> Changes in v2: None
>
>  drivers/media/platform/s5p-mfc/s5p_mfc_dec.c | 14 ++++++++++++++
>  1 file changed, 14 insertions(+)
>
> diff --git a/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c b/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c
> index 367ef8e8dbf0..0976c3e0a5ce 100644
> --- a/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c
> +++ b/drivers/media/platform/s5p-mfc/s5p_mfc_dec.c
> @@ -354,6 +354,11 @@ static int vidioc_g_fmt(struct file *file, void *priv, struct v4l2_format *f)
>  		pix_mp->plane_fmt[0].sizeimage = ctx->luma_size;
>  		pix_mp->plane_fmt[1].bytesperline = ctx->buf_width;
>  		pix_mp->plane_fmt[1].sizeimage = ctx->chroma_size;
> +
> +		if (pix_mp->width > 720 && pix_mp->height > 576) /* HD */
> +			pix_mp->colorspace = V4L2_COLORSPACE_REC709;
> +		else /* SD */
> +			pix_mp->colorspace = V4L2_COLORSPACE_SMPTE170M;

Again, it seems much more complicated for decoder, I suppose colorspace
field should be filled according to decoded information from the header
of byte stream. It looks like MFC does not parse  stream header for such
info. So it should be done either by the driver, either by userspace, if
userspace is able to get such info. For example in H.264 this info is
encoded in VUI header [1], I do not know about other codecs. I guess the
best solution for now is to not touch this field unless you can retrieve
this info from MFC.

[1]:
https://github.com/copiousfreetime/mp4parser/blob/master/isoparser/src/main/java/com/googlecode/mp4parser/h264/model/SeqParameterSet.java#L227

Regards
Andrzej

>  	} else if (f->type == V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE) {
>  		/* This is run on OUTPUT
>  		   The buffer contains compressed image
> @@ -378,6 +383,7 @@ static int vidioc_g_fmt(struct file *file, void *priv, struct v4l2_format *f)
>  static int vidioc_try_fmt(struct file *file, void *priv, struct v4l2_format *f)
>  {
>  	struct s5p_mfc_dev *dev = video_drvdata(file);
> +	struct v4l2_pix_format_mplane *pix_mp = &f->fmt.pix_mp;
>  	struct s5p_mfc_fmt *fmt;
>  
>  	mfc_debug(2, "Type is %d\n", f->type);
> @@ -405,6 +411,14 @@ static int vidioc_try_fmt(struct file *file, void *priv, struct v4l2_format *f)
>  			mfc_err("Unsupported format by this MFC version.\n");
>  			return -EINVAL;
>  		}
> +
> +		if (pix_mp->colorspace == V4L2_COLORSPACE_DEFAULT) {
> +			if (pix_mp->width > 720 &&
> +					pix_mp->height > 576) /* HD */
> +				pix_mp->colorspace = V4L2_COLORSPACE_REC709;
> +			else /* SD */
> +				pix_mp->colorspace = V4L2_COLORSPACE_SMPTE170M;
> +		}
>  	}
>  
>  	return 0;