| Message ID | 201301300901.22486.hverkuil@xs4all.nl (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Em Wed, 30 Jan 2013 09:01:22 +0100 Hans Verkuil <hverkuil@xs4all.nl> escreveu: > This was part of my original em28xx patch series. That particular patch > combined two things: this fix and the change where TRY_FMT would no > longer return -EINVAL for unsupported pixelformats. The latter change was > rejected (correctly), but we all forgot about the second part of the patch > which fixed a real bug. I'm reposting just that fix. > > Changes since v1: > > - v1 still miscalculated the bytesperline and imagesize values (they were > too large). > - G_FMT had the same calculation bug. > > Tested with my em28xx. > > Regards, > > Hans > > The bytesperline calculation was incorrect: it used the old width instead of > the provided width in the case of TRY_FMT, and it miscalculated the bytesperline > value for the depth == 12 (planar YUV 4:1:1) case. For planar formats the > bytesperline value should be the bytesperline of the widest plane, which is > the Y plane which has 8 bits per pixel, not 12. > > Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com> > --- > drivers/media/usb/em28xx/em28xx-video.c | 8 ++++---- > 1 file changed, 4 insertions(+), 4 deletions(-) > > diff --git a/drivers/media/usb/em28xx/em28xx-video.c b/drivers/media/usb/em28xx/em28xx-video.c > index 2eabf2a..6ced426 100644 > --- a/drivers/media/usb/em28xx/em28xx-video.c > +++ b/drivers/media/usb/em28xx/em28xx-video.c > @@ -837,8 +837,8 @@ static int vidioc_g_fmt_vid_cap(struct file *file, void *priv, > f->fmt.pix.width = dev->width; > f->fmt.pix.height = dev->height; > f->fmt.pix.pixelformat = dev->format->fourcc; > - f->fmt.pix.bytesperline = (dev->width * dev->format->depth + 7) >> 3; > - f->fmt.pix.sizeimage = f->fmt.pix.bytesperline * dev->height; > + f->fmt.pix.bytesperline = dev->width * (dev->format->depth >> 3); Why did you remove the round up here? > + f->fmt.pix.sizeimage = (dev->width * dev->height * dev->format->depth + 7) >> 3; > f->fmt.pix.colorspace = V4L2_COLORSPACE_SMPTE170M; > > /* FIXME: TOP? NONE? BOTTOM? ALTENATE? */ > @@ -906,8 +906,8 @@ static int vidioc_try_fmt_vid_cap(struct file *file, void *priv, > f->fmt.pix.width = width; > f->fmt.pix.height = height; > f->fmt.pix.pixelformat = fmt->fourcc; > - f->fmt.pix.bytesperline = (dev->width * fmt->depth + 7) >> 3; > - f->fmt.pix.sizeimage = f->fmt.pix.bytesperline * height; > + f->fmt.pix.bytesperline = width * (fmt->depth >> 3); Why did you remove the round up here? > + f->fmt.pix.sizeimage = (width * height * fmt->depth + 7) >> 3; > f->fmt.pix.colorspace = V4L2_COLORSPACE_SMPTE170M; > if (dev->progressive) > f->fmt.pix.field = V4L2_FIELD_NONE; Regards, Mauro -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Wed 30 January 2013 10:40:30 Mauro Carvalho Chehab wrote: > Em Wed, 30 Jan 2013 09:01:22 +0100 > Hans Verkuil <hverkuil@xs4all.nl> escreveu: > > > This was part of my original em28xx patch series. That particular patch > > combined two things: this fix and the change where TRY_FMT would no > > longer return -EINVAL for unsupported pixelformats. The latter change was > > rejected (correctly), but we all forgot about the second part of the patch > > which fixed a real bug. I'm reposting just that fix. > > > > Changes since v1: > > > > - v1 still miscalculated the bytesperline and imagesize values (they were > > too large). > > - G_FMT had the same calculation bug. > > > > Tested with my em28xx. > > > > Regards, > > > > Hans > > > > The bytesperline calculation was incorrect: it used the old width instead of > > the provided width in the case of TRY_FMT, and it miscalculated the bytesperline > > value for the depth == 12 (planar YUV 4:1:1) case. For planar formats the > > bytesperline value should be the bytesperline of the widest plane, which is > > the Y plane which has 8 bits per pixel, not 12. > > > > Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com> > > --- > > drivers/media/usb/em28xx/em28xx-video.c | 8 ++++---- > > 1 file changed, 4 insertions(+), 4 deletions(-) > > > > diff --git a/drivers/media/usb/em28xx/em28xx-video.c b/drivers/media/usb/em28xx/em28xx-video.c > > index 2eabf2a..6ced426 100644 > > --- a/drivers/media/usb/em28xx/em28xx-video.c > > +++ b/drivers/media/usb/em28xx/em28xx-video.c > > @@ -837,8 +837,8 @@ static int vidioc_g_fmt_vid_cap(struct file *file, void *priv, > > f->fmt.pix.width = dev->width; > > f->fmt.pix.height = dev->height; > > f->fmt.pix.pixelformat = dev->format->fourcc; > > - f->fmt.pix.bytesperline = (dev->width * dev->format->depth + 7) >> 3; > > - f->fmt.pix.sizeimage = f->fmt.pix.bytesperline * dev->height; > > + f->fmt.pix.bytesperline = dev->width * (dev->format->depth >> 3); > > Why did you remove the round up here? Because that would give the wrong result. Depth can be 8, 12 or 16. The YUV 4:1:1 planar format is the one with depth 12. But for the purposes of the bytesperline calculation only the depth of the largest plane counts, which is the luma plane with a depth of 8. So for a width of 720 the value of bytesperline should be: depth=8 -> bytesperline = 720 depth=12 -> bytesperline = 720 depth=16 -> bytesperline = 1440 By rounding the bytesperline for depth = 12 would become 1440, which is wrong. > > > + f->fmt.pix.sizeimage = (dev->width * dev->height * dev->format->depth + 7) >> 3; > > f->fmt.pix.colorspace = V4L2_COLORSPACE_SMPTE170M; > > > > /* FIXME: TOP? NONE? BOTTOM? ALTENATE? */ > > @@ -906,8 +906,8 @@ static int vidioc_try_fmt_vid_cap(struct file *file, void *priv, > > f->fmt.pix.width = width; > > f->fmt.pix.height = height; > > f->fmt.pix.pixelformat = fmt->fourcc; > > - f->fmt.pix.bytesperline = (dev->width * fmt->depth + 7) >> 3; > > - f->fmt.pix.sizeimage = f->fmt.pix.bytesperline * height; > > + f->fmt.pix.bytesperline = width * (fmt->depth >> 3); > > Why did you remove the round up here? See above. Regards, Hans > > > + f->fmt.pix.sizeimage = (width * height * fmt->depth + 7) >> 3; > > f->fmt.pix.colorspace = V4L2_COLORSPACE_SMPTE170M; > > if (dev->progressive) > > f->fmt.pix.field = V4L2_FIELD_NONE; > > > Regards, > Mauro > -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Am 30.01.2013 10:49, schrieb Hans Verkuil: > On Wed 30 January 2013 10:40:30 Mauro Carvalho Chehab wrote: >> Em Wed, 30 Jan 2013 09:01:22 +0100 >> Hans Verkuil <hverkuil@xs4all.nl> escreveu: >> >>> This was part of my original em28xx patch series. That particular patch >>> combined two things: this fix and the change where TRY_FMT would no >>> longer return -EINVAL for unsupported pixelformats. The latter change was >>> rejected (correctly), but we all forgot about the second part of the patch >>> which fixed a real bug. I'm reposting just that fix. >>> >>> Changes since v1: >>> >>> - v1 still miscalculated the bytesperline and imagesize values (they were >>> too large). >>> - G_FMT had the same calculation bug. >>> >>> Tested with my em28xx. >>> >>> Regards, >>> >>> Hans >>> >>> The bytesperline calculation was incorrect: it used the old width instead of >>> the provided width in the case of TRY_FMT, and it miscalculated the bytesperline >>> value for the depth == 12 (planar YUV 4:1:1) case. For planar formats the >>> bytesperline value should be the bytesperline of the widest plane, which is >>> the Y plane which has 8 bits per pixel, not 12. >>> >>> Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com> >>> --- >>> drivers/media/usb/em28xx/em28xx-video.c | 8 ++++---- >>> 1 file changed, 4 insertions(+), 4 deletions(-) >>> >>> diff --git a/drivers/media/usb/em28xx/em28xx-video.c b/drivers/media/usb/em28xx/em28xx-video.c >>> index 2eabf2a..6ced426 100644 >>> --- a/drivers/media/usb/em28xx/em28xx-video.c >>> +++ b/drivers/media/usb/em28xx/em28xx-video.c >>> @@ -837,8 +837,8 @@ static int vidioc_g_fmt_vid_cap(struct file *file, void *priv, >>> f->fmt.pix.width = dev->width; >>> f->fmt.pix.height = dev->height; >>> f->fmt.pix.pixelformat = dev->format->fourcc; >>> - f->fmt.pix.bytesperline = (dev->width * dev->format->depth + 7) >> 3; >>> - f->fmt.pix.sizeimage = f->fmt.pix.bytesperline * dev->height; >>> + f->fmt.pix.bytesperline = dev->width * (dev->format->depth >> 3); >> Why did you remove the round up here? > Because that would give the wrong result. Depth can be 8, 12 or 16. The YUV 4:1:1 > planar format is the one with depth 12. But for the purposes of the bytesperline > calculation only the depth of the largest plane counts, which is the luma plane > with a depth of 8. So for a width of 720 the value of bytesperline should be: > > depth=8 -> bytesperline = 720 > depth=12 -> bytesperline = 720 > depth=16 -> bytesperline = 1440 > > By rounding the bytesperline for depth = 12 would become 1440, which is wrong. Isn't the actual problem that the bytesperline value is calculated based on the depth instead of size of the largest plane ? While this formula gives us the right value for YUV411P (12bit), the calculated value for YUV422P (same plane size but 16bit depth) is wrong. I see two possibilities to solve this: 1) use switch/case(format_type) in vidioc_g_fmt_vid_cap() and vidioc_try_fmt_vid_cap() 2) add plane size to format data struct em28xx_fmt (either as total or relative value) I prefer 2) to keep things together and to make adding new format easier. Could you please also fix static void em28xx_copy_video() - int bytesperline = dev->width << 1; + int bytesperline = (dev->width * dev->format->depth) >> 8 AFAICS, bytesperline here is supposed to describe the actual number of bytes used per line (in opposition to g/try_fmt). Regards, Frank > >>> + f->fmt.pix.sizeimage = (dev->width * dev->height * dev->format->depth + 7) >> 3; >>> f->fmt.pix.colorspace = V4L2_COLORSPACE_SMPTE170M; >>> >>> /* FIXME: TOP? NONE? BOTTOM? ALTENATE? */ >>> @@ -906,8 +906,8 @@ static int vidioc_try_fmt_vid_cap(struct file *file, void *priv, >>> f->fmt.pix.width = width; >>> f->fmt.pix.height = height; >>> f->fmt.pix.pixelformat = fmt->fourcc; >>> - f->fmt.pix.bytesperline = (dev->width * fmt->depth + 7) >> 3; >>> - f->fmt.pix.sizeimage = f->fmt.pix.bytesperline * height; >>> + f->fmt.pix.bytesperline = width * (fmt->depth >> 3); >> Why did you remove the round up here? > See above. > > Regards, > > Hans > >>> + f->fmt.pix.sizeimage = (width * height * fmt->depth + 7) >> 3; >>> f->fmt.pix.colorspace = V4L2_COLORSPACE_SMPTE170M; >>> if (dev->progressive) >>> f->fmt.pix.field = V4L2_FIELD_NONE; >> >> Regards, >> Mauro >> -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Wed January 30 2013 17:32:38 Frank Schäfer wrote: > Am 30.01.2013 10:49, schrieb Hans Verkuil: > > On Wed 30 January 2013 10:40:30 Mauro Carvalho Chehab wrote: > >> Em Wed, 30 Jan 2013 09:01:22 +0100 > >> Hans Verkuil <hverkuil@xs4all.nl> escreveu: > >> > >>> This was part of my original em28xx patch series. That particular patch > >>> combined two things: this fix and the change where TRY_FMT would no > >>> longer return -EINVAL for unsupported pixelformats. The latter change was > >>> rejected (correctly), but we all forgot about the second part of the patch > >>> which fixed a real bug. I'm reposting just that fix. > >>> > >>> Changes since v1: > >>> > >>> - v1 still miscalculated the bytesperline and imagesize values (they were > >>> too large). > >>> - G_FMT had the same calculation bug. > >>> > >>> Tested with my em28xx. > >>> > >>> Regards, > >>> > >>> Hans > >>> > >>> The bytesperline calculation was incorrect: it used the old width instead of > >>> the provided width in the case of TRY_FMT, and it miscalculated the bytesperline > >>> value for the depth == 12 (planar YUV 4:1:1) case. For planar formats the > >>> bytesperline value should be the bytesperline of the widest plane, which is > >>> the Y plane which has 8 bits per pixel, not 12. > >>> > >>> Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com> > >>> --- > >>> drivers/media/usb/em28xx/em28xx-video.c | 8 ++++---- > >>> 1 file changed, 4 insertions(+), 4 deletions(-) > >>> > >>> diff --git a/drivers/media/usb/em28xx/em28xx-video.c b/drivers/media/usb/em28xx/em28xx-video.c > >>> index 2eabf2a..6ced426 100644 > >>> --- a/drivers/media/usb/em28xx/em28xx-video.c > >>> +++ b/drivers/media/usb/em28xx/em28xx-video.c > >>> @@ -837,8 +837,8 @@ static int vidioc_g_fmt_vid_cap(struct file *file, void *priv, > >>> f->fmt.pix.width = dev->width; > >>> f->fmt.pix.height = dev->height; > >>> f->fmt.pix.pixelformat = dev->format->fourcc; > >>> - f->fmt.pix.bytesperline = (dev->width * dev->format->depth + 7) >> 3; > >>> - f->fmt.pix.sizeimage = f->fmt.pix.bytesperline * dev->height; > >>> + f->fmt.pix.bytesperline = dev->width * (dev->format->depth >> 3); > >> Why did you remove the round up here? > > Because that would give the wrong result. Depth can be 8, 12 or 16. The YUV 4:1:1 > > planar format is the one with depth 12. But for the purposes of the bytesperline > > calculation only the depth of the largest plane counts, which is the luma plane > > with a depth of 8. So for a width of 720 the value of bytesperline should be: > > > > depth=8 -> bytesperline = 720 > > depth=12 -> bytesperline = 720 > > depth=16 -> bytesperline = 1440 > > > > By rounding the bytesperline for depth = 12 would become 1440, which is wrong. > > Isn't the actual problem that the bytesperline value is calculated based > on the depth instead of size of the largest plane ? Yes. > While this formula gives us the right value for YUV411P (12bit), the > calculated value for YUV422P (same plane size but 16bit depth) is wrong. But em28xx doesn't support YUV422P :-) Basically I just want to fix the bug without rewriting the whole thing... > > I see two possibilities to solve this: > 1) use switch/case(format_type) in vidioc_g_fmt_vid_cap() and > vidioc_try_fmt_vid_cap() > 2) add plane size to format data struct em28xx_fmt (either as total or > relative value) > > I prefer 2) to keep things together and to make adding new format easier. > > Could you please also fix > static void em28xx_copy_video() > > - int bytesperline = dev->width << 1; > + int bytesperline = (dev->width * dev->format->depth) >> 8 I'm hesitant to do this since I don't understand that code well enough. It seems bytesperline is set to the worst case bytesperline, and that seems to work. To be honest, I don't have the time to work on this much more. If you want to take this further, then feel free. Regards, Hans > AFAICS, bytesperline here is supposed to describe the actual number of > bytes used per line (in opposition to g/try_fmt). > > Regards, > Frank > > > > >>> + f->fmt.pix.sizeimage = (dev->width * dev->height * dev->format->depth + 7) >> 3; > >>> f->fmt.pix.colorspace = V4L2_COLORSPACE_SMPTE170M; > >>> > >>> /* FIXME: TOP? NONE? BOTTOM? ALTENATE? */ > >>> @@ -906,8 +906,8 @@ static int vidioc_try_fmt_vid_cap(struct file *file, void *priv, > >>> f->fmt.pix.width = width; > >>> f->fmt.pix.height = height; > >>> f->fmt.pix.pixelformat = fmt->fourcc; > >>> - f->fmt.pix.bytesperline = (dev->width * fmt->depth + 7) >> 3; > >>> - f->fmt.pix.sizeimage = f->fmt.pix.bytesperline * height; > >>> + f->fmt.pix.bytesperline = width * (fmt->depth >> 3); > >> Why did you remove the round up here? > > See above. > > > > Regards, > > > > Hans > > > >>> + f->fmt.pix.sizeimage = (width * height * fmt->depth + 7) >> 3; > >>> f->fmt.pix.colorspace = V4L2_COLORSPACE_SMPTE170M; > >>> if (dev->progressive) > >>> f->fmt.pix.field = V4L2_FIELD_NONE; > >> > >> Regards, > >> Mauro > >> > -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Em Wed, 30 Jan 2013 10:49:25 +0100 Hans Verkuil <hverkuil@xs4all.nl> escreveu: > On Wed 30 January 2013 10:40:30 Mauro Carvalho Chehab wrote: > > Em Wed, 30 Jan 2013 09:01:22 +0100 > > Hans Verkuil <hverkuil@xs4all.nl> escreveu: > > > > > This was part of my original em28xx patch series. That particular patch > > > combined two things: this fix and the change where TRY_FMT would no > > > longer return -EINVAL for unsupported pixelformats. The latter change was > > > rejected (correctly), but we all forgot about the second part of the patch > > > which fixed a real bug. I'm reposting just that fix. > > > > > > Changes since v1: > > > > > > - v1 still miscalculated the bytesperline and imagesize values (they were > > > too large). > > > - G_FMT had the same calculation bug. > > > > > > Tested with my em28xx. > > > > > > Regards, > > > > > > Hans > > > > > > The bytesperline calculation was incorrect: it used the old width instead of > > > the provided width in the case of TRY_FMT, and it miscalculated the bytesperline > > > value for the depth == 12 (planar YUV 4:1:1) case. For planar formats the > > > bytesperline value should be the bytesperline of the widest plane, which is > > > the Y plane which has 8 bits per pixel, not 12. > > > > > > Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com> > > > --- > > > drivers/media/usb/em28xx/em28xx-video.c | 8 ++++---- > > > 1 file changed, 4 insertions(+), 4 deletions(-) > > > > > > diff --git a/drivers/media/usb/em28xx/em28xx-video.c b/drivers/media/usb/em28xx/em28xx-video.c > > > index 2eabf2a..6ced426 100644 > > > --- a/drivers/media/usb/em28xx/em28xx-video.c > > > +++ b/drivers/media/usb/em28xx/em28xx-video.c > > > @@ -837,8 +837,8 @@ static int vidioc_g_fmt_vid_cap(struct file *file, void *priv, > > > f->fmt.pix.width = dev->width; > > > f->fmt.pix.height = dev->height; > > > f->fmt.pix.pixelformat = dev->format->fourcc; > > > - f->fmt.pix.bytesperline = (dev->width * dev->format->depth + 7) >> 3; > > > - f->fmt.pix.sizeimage = f->fmt.pix.bytesperline * dev->height; > > > + f->fmt.pix.bytesperline = dev->width * (dev->format->depth >> 3); > > > > Why did you remove the round up here? > > Because that would give the wrong result. Depth can be 8, 12 or 16. The YUV 4:1:1 > planar format is the one with depth 12. But for the purposes of the bytesperline > calculation only the depth of the largest plane counts, which is the luma plane > with a depth of 8. So for a width of 720 the value of bytesperline should be: > > depth=8 -> bytesperline = 720 > depth=12 -> bytesperline = 720 With depth=12, it should be, instead, 1080, as 2 pixels need 3 bytes. > depth=16 -> bytesperline = 1440 Well, depth=8 -> bytesperline = (720 * 8) + 7) / 8 = 720 depth=12 -> bytesperline = (720 * 12) + 7) / 8 = 1080 depth=16 -> bytesperline = (720 * 16) + 7) / 8 = 1440 So, this sounds perfectly OK on my eyes: f->fmt.pix.bytesperline = (dev->width * dev->format->depth + 7) >> 3; Regards, Mauro -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Wed January 30 2013 20:07:29 Mauro Carvalho Chehab wrote: > Em Wed, 30 Jan 2013 10:49:25 +0100 > Hans Verkuil <hverkuil@xs4all.nl> escreveu: > > > On Wed 30 January 2013 10:40:30 Mauro Carvalho Chehab wrote: > > > Em Wed, 30 Jan 2013 09:01:22 +0100 > > > Hans Verkuil <hverkuil@xs4all.nl> escreveu: > > > > > > > This was part of my original em28xx patch series. That particular patch > > > > combined two things: this fix and the change where TRY_FMT would no > > > > longer return -EINVAL for unsupported pixelformats. The latter change was > > > > rejected (correctly), but we all forgot about the second part of the patch > > > > which fixed a real bug. I'm reposting just that fix. > > > > > > > > Changes since v1: > > > > > > > > - v1 still miscalculated the bytesperline and imagesize values (they were > > > > too large). > > > > - G_FMT had the same calculation bug. > > > > > > > > Tested with my em28xx. > > > > > > > > Regards, > > > > > > > > Hans > > > > > > > > The bytesperline calculation was incorrect: it used the old width instead of > > > > the provided width in the case of TRY_FMT, and it miscalculated the bytesperline > > > > value for the depth == 12 (planar YUV 4:1:1) case. For planar formats the > > > > bytesperline value should be the bytesperline of the widest plane, which is > > > > the Y plane which has 8 bits per pixel, not 12. > > > > > > > > Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com> > > > > --- > > > > drivers/media/usb/em28xx/em28xx-video.c | 8 ++++---- > > > > 1 file changed, 4 insertions(+), 4 deletions(-) > > > > > > > > diff --git a/drivers/media/usb/em28xx/em28xx-video.c b/drivers/media/usb/em28xx/em28xx-video.c > > > > index 2eabf2a..6ced426 100644 > > > > --- a/drivers/media/usb/em28xx/em28xx-video.c > > > > +++ b/drivers/media/usb/em28xx/em28xx-video.c > > > > @@ -837,8 +837,8 @@ static int vidioc_g_fmt_vid_cap(struct file *file, void *priv, > > > > f->fmt.pix.width = dev->width; > > > > f->fmt.pix.height = dev->height; > > > > f->fmt.pix.pixelformat = dev->format->fourcc; > > > > - f->fmt.pix.bytesperline = (dev->width * dev->format->depth + 7) >> 3; > > > > - f->fmt.pix.sizeimage = f->fmt.pix.bytesperline * dev->height; > > > > + f->fmt.pix.bytesperline = dev->width * (dev->format->depth >> 3); > > > > > > Why did you remove the round up here? > > > > Because that would give the wrong result. Depth can be 8, 12 or 16. The YUV 4:1:1 > > planar format is the one with depth 12. But for the purposes of the bytesperline > > calculation only the depth of the largest plane counts, which is the luma plane > > with a depth of 8. So for a width of 720 the value of bytesperline should be: > > > > depth=8 -> bytesperline = 720 > > depth=12 -> bytesperline = 720 > > With depth=12, it should be, instead, 1080, as 2 pixels need 3 bytes. No, it's not. It's a *planar* format: first the Y plane, then the two smaller chroma planes. The spec says that bytesperline for planar formats refers to the largest plane. For this format the luma plane is one byte per pixel. Each of the two chroma planes have effectively two bits per pixel (actually one byte per four pixels), so you end up with 8+2+2=12 bits per pixel. Hence bytesperline should be 720 for this particular format. Regards, Hans > > > depth=16 -> bytesperline = 1440 > > Well, > > depth=8 -> bytesperline = (720 * 8) + 7) / 8 = 720 > depth=12 -> bytesperline = (720 * 12) + 7) / 8 = 1080 > depth=16 -> bytesperline = (720 * 16) + 7) / 8 = 1440 > > So, this sounds perfectly OK on my eyes: > f->fmt.pix.bytesperline = (dev->width * dev->format->depth + 7) >> 3; > > Regards, > Mauro > -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Em Thu, 31 Jan 2013 08:16:39 +0100 Hans Verkuil <hverkuil@xs4all.nl> escreveu: > On Wed January 30 2013 20:07:29 Mauro Carvalho Chehab wrote: > > Em Wed, 30 Jan 2013 10:49:25 +0100 > > Hans Verkuil <hverkuil@xs4all.nl> escreveu: > > > > > On Wed 30 January 2013 10:40:30 Mauro Carvalho Chehab wrote: > > > > Em Wed, 30 Jan 2013 09:01:22 +0100 > > > > Hans Verkuil <hverkuil@xs4all.nl> escreveu: > > > > > > > > > This was part of my original em28xx patch series. That particular patch > > > > > combined two things: this fix and the change where TRY_FMT would no > > > > > longer return -EINVAL for unsupported pixelformats. The latter change was > > > > > rejected (correctly), but we all forgot about the second part of the patch > > > > > which fixed a real bug. I'm reposting just that fix. > > > > > > > > > > Changes since v1: > > > > > > > > > > - v1 still miscalculated the bytesperline and imagesize values (they were > > > > > too large). > > > > > - G_FMT had the same calculation bug. > > > > > > > > > > Tested with my em28xx. > > > > > > > > > > Regards, > > > > > > > > > > Hans > > > > > > > > > > The bytesperline calculation was incorrect: it used the old width instead of > > > > > the provided width in the case of TRY_FMT, and it miscalculated the bytesperline > > > > > value for the depth == 12 (planar YUV 4:1:1) case. For planar formats the > > > > > bytesperline value should be the bytesperline of the widest plane, which is > > > > > the Y plane which has 8 bits per pixel, not 12. > > > > > > > > > > Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com> > > > > > --- > > > > > drivers/media/usb/em28xx/em28xx-video.c | 8 ++++---- > > > > > 1 file changed, 4 insertions(+), 4 deletions(-) > > > > > > > > > > diff --git a/drivers/media/usb/em28xx/em28xx-video.c b/drivers/media/usb/em28xx/em28xx-video.c > > > > > index 2eabf2a..6ced426 100644 > > > > > --- a/drivers/media/usb/em28xx/em28xx-video.c > > > > > +++ b/drivers/media/usb/em28xx/em28xx-video.c > > > > > @@ -837,8 +837,8 @@ static int vidioc_g_fmt_vid_cap(struct file *file, void *priv, > > > > > f->fmt.pix.width = dev->width; > > > > > f->fmt.pix.height = dev->height; > > > > > f->fmt.pix.pixelformat = dev->format->fourcc; > > > > > - f->fmt.pix.bytesperline = (dev->width * dev->format->depth + 7) >> 3; > > > > > - f->fmt.pix.sizeimage = f->fmt.pix.bytesperline * dev->height; > > > > > + f->fmt.pix.bytesperline = dev->width * (dev->format->depth >> 3); > > > > > > > > Why did you remove the round up here? > > > > > > Because that would give the wrong result. Depth can be 8, 12 or 16. The YUV 4:1:1 > > > planar format is the one with depth 12. But for the purposes of the bytesperline > > > calculation only the depth of the largest plane counts, which is the luma plane > > > with a depth of 8. So for a width of 720 the value of bytesperline should be: > > > > > > depth=8 -> bytesperline = 720 > > > depth=12 -> bytesperline = 720 > > > > With depth=12, it should be, instead, 1080, as 2 pixels need 3 bytes. > > No, it's not. It's a *planar* format: first the Y plane, then the two smaller > chroma planes. The spec says that bytesperline for planar formats refers to > the largest plane. > > For this format the luma plane is one byte per pixel. Each of the two chroma > planes have effectively two bits per pixel (actually one byte per four pixels), > so you end up with 8+2+2=12 bits per pixel. > > Hence bytesperline should be 720 for this particular format. If I understood what you just said, you're talking that the only format marked as depth=12 is actually depth=8, right? Then the fix would be to change depth in the table, and not here. Yet, I'm not sure if this is the proper fix. The only used I saw on userspace apps for this field is to allocate size for the memory buffer. Some userspace applications use to get bytesperline and multiply by the image height and get the image size, instead of relying on sizeimage, as some drivers didn't use to fill sizeimage properly. By using bytesperline equal to 1080 in this case warrants that the buffers on userspace will have enough space. Regards, Mauro -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Thu 31 January 2013 11:08:07 Mauro Carvalho Chehab wrote: > Em Thu, 31 Jan 2013 08:16:39 +0100 > Hans Verkuil <hverkuil@xs4all.nl> escreveu: > > > On Wed January 30 2013 20:07:29 Mauro Carvalho Chehab wrote: > > > Em Wed, 30 Jan 2013 10:49:25 +0100 > > > Hans Verkuil <hverkuil@xs4all.nl> escreveu: > > > > > > > On Wed 30 January 2013 10:40:30 Mauro Carvalho Chehab wrote: > > > > > Em Wed, 30 Jan 2013 09:01:22 +0100 > > > > > Hans Verkuil <hverkuil@xs4all.nl> escreveu: > > > > > > > > > > > This was part of my original em28xx patch series. That particular patch > > > > > > combined two things: this fix and the change where TRY_FMT would no > > > > > > longer return -EINVAL for unsupported pixelformats. The latter change was > > > > > > rejected (correctly), but we all forgot about the second part of the patch > > > > > > which fixed a real bug. I'm reposting just that fix. > > > > > > > > > > > > Changes since v1: > > > > > > > > > > > > - v1 still miscalculated the bytesperline and imagesize values (they were > > > > > > too large). > > > > > > - G_FMT had the same calculation bug. > > > > > > > > > > > > Tested with my em28xx. > > > > > > > > > > > > Regards, > > > > > > > > > > > > Hans > > > > > > > > > > > > The bytesperline calculation was incorrect: it used the old width instead of > > > > > > the provided width in the case of TRY_FMT, and it miscalculated the bytesperline > > > > > > value for the depth == 12 (planar YUV 4:1:1) case. For planar formats the > > > > > > bytesperline value should be the bytesperline of the widest plane, which is > > > > > > the Y plane which has 8 bits per pixel, not 12. > > > > > > > > > > > > Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com> > > > > > > --- > > > > > > drivers/media/usb/em28xx/em28xx-video.c | 8 ++++---- > > > > > > 1 file changed, 4 insertions(+), 4 deletions(-) > > > > > > > > > > > > diff --git a/drivers/media/usb/em28xx/em28xx-video.c b/drivers/media/usb/em28xx/em28xx-video.c > > > > > > index 2eabf2a..6ced426 100644 > > > > > > --- a/drivers/media/usb/em28xx/em28xx-video.c > > > > > > +++ b/drivers/media/usb/em28xx/em28xx-video.c > > > > > > @@ -837,8 +837,8 @@ static int vidioc_g_fmt_vid_cap(struct file *file, void *priv, > > > > > > f->fmt.pix.width = dev->width; > > > > > > f->fmt.pix.height = dev->height; > > > > > > f->fmt.pix.pixelformat = dev->format->fourcc; > > > > > > - f->fmt.pix.bytesperline = (dev->width * dev->format->depth + 7) >> 3; > > > > > > - f->fmt.pix.sizeimage = f->fmt.pix.bytesperline * dev->height; > > > > > > + f->fmt.pix.bytesperline = dev->width * (dev->format->depth >> 3); > > > > > > > > > > Why did you remove the round up here? > > > > > > > > Because that would give the wrong result. Depth can be 8, 12 or 16. The YUV 4:1:1 > > > > planar format is the one with depth 12. But for the purposes of the bytesperline > > > > calculation only the depth of the largest plane counts, which is the luma plane > > > > with a depth of 8. So for a width of 720 the value of bytesperline should be: > > > > > > > > depth=8 -> bytesperline = 720 > > > > depth=12 -> bytesperline = 720 > > > > > > With depth=12, it should be, instead, 1080, as 2 pixels need 3 bytes. > > > > No, it's not. It's a *planar* format: first the Y plane, then the two smaller > > chroma planes. The spec says that bytesperline for planar formats refers to > > the largest plane. > > > > For this format the luma plane is one byte per pixel. Each of the two chroma > > planes have effectively two bits per pixel (actually one byte per four pixels), > > so you end up with 8+2+2=12 bits per pixel. > > > > Hence bytesperline should be 720 for this particular format. > > If I understood what you just said, you're talking that the only format marked > as depth=12 is actually depth=8, right? Then the fix would be to change depth > in the table, and not here. > > Yet, I'm not sure if this is the proper fix. > > The only used I saw on userspace apps for this field is to allocate size for > the memory buffer. Some userspace applications use to get bytesperline and > multiply by the image height and get the image size, instead of relying > on sizeimage, as some drivers didn't use to fill sizeimage properly. > > By using bytesperline equal to 1080 in this case warrants that the buffers > on userspace will have enough space. I did some research into planar formats in our drivers and (of course) it's a big mess. I looked at drivers that support V4L2_PIX_FMT_YUV422P, which is fairly common: s5p-fimc: follows the spec arv: follows the spec vpif_capture: follows the spec vpif_display: follows the spec pxa_camera: no idea, I can't figure this out s3c-camif: follows the spec exynos-gsc: uses depth s2255: uses depth usbvision: uses depth saa7146: uses depth saa7134: uses depth bttv: follows the spec I think we should follow the spec here. In practice, nobody uses planar formats for consumer-type hardware as it is a very awkward format, and libv4l doesn't support it either. Since there is no clear common practice in our drivers, I'd say we stick to the spec. Regards, Hans -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Am 31.01.2013 11:08, schrieb Mauro Carvalho Chehab: > Em Thu, 31 Jan 2013 08:16:39 +0100 > Hans Verkuil <hverkuil@xs4all.nl> escreveu: > >> On Wed January 30 2013 20:07:29 Mauro Carvalho Chehab wrote: >>> Em Wed, 30 Jan 2013 10:49:25 +0100 >>> Hans Verkuil <hverkuil@xs4all.nl> escreveu: >>> >>>> On Wed 30 January 2013 10:40:30 Mauro Carvalho Chehab wrote: >>>>> Em Wed, 30 Jan 2013 09:01:22 +0100 >>>>> Hans Verkuil <hverkuil@xs4all.nl> escreveu: >>>>> >>>>>> This was part of my original em28xx patch series. That particular patch >>>>>> combined two things: this fix and the change where TRY_FMT would no >>>>>> longer return -EINVAL for unsupported pixelformats. The latter change was >>>>>> rejected (correctly), but we all forgot about the second part of the patch >>>>>> which fixed a real bug. I'm reposting just that fix. >>>>>> >>>>>> Changes since v1: >>>>>> >>>>>> - v1 still miscalculated the bytesperline and imagesize values (they were >>>>>> too large). >>>>>> - G_FMT had the same calculation bug. >>>>>> >>>>>> Tested with my em28xx. >>>>>> >>>>>> Regards, >>>>>> >>>>>> Hans >>>>>> >>>>>> The bytesperline calculation was incorrect: it used the old width instead of >>>>>> the provided width in the case of TRY_FMT, and it miscalculated the bytesperline >>>>>> value for the depth == 12 (planar YUV 4:1:1) case. For planar formats the >>>>>> bytesperline value should be the bytesperline of the widest plane, which is >>>>>> the Y plane which has 8 bits per pixel, not 12. >>>>>> >>>>>> Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com> >>>>>> --- >>>>>> drivers/media/usb/em28xx/em28xx-video.c | 8 ++++---- >>>>>> 1 file changed, 4 insertions(+), 4 deletions(-) >>>>>> >>>>>> diff --git a/drivers/media/usb/em28xx/em28xx-video.c b/drivers/media/usb/em28xx/em28xx-video.c >>>>>> index 2eabf2a..6ced426 100644 >>>>>> --- a/drivers/media/usb/em28xx/em28xx-video.c >>>>>> +++ b/drivers/media/usb/em28xx/em28xx-video.c >>>>>> @@ -837,8 +837,8 @@ static int vidioc_g_fmt_vid_cap(struct file *file, void *priv, >>>>>> f->fmt.pix.width = dev->width; >>>>>> f->fmt.pix.height = dev->height; >>>>>> f->fmt.pix.pixelformat = dev->format->fourcc; >>>>>> - f->fmt.pix.bytesperline = (dev->width * dev->format->depth + 7) >> 3; >>>>>> - f->fmt.pix.sizeimage = f->fmt.pix.bytesperline * dev->height; >>>>>> + f->fmt.pix.bytesperline = dev->width * (dev->format->depth >> 3); >>>>> Why did you remove the round up here? >>>> Because that would give the wrong result. Depth can be 8, 12 or 16. The YUV 4:1:1 >>>> planar format is the one with depth 12. But for the purposes of the bytesperline >>>> calculation only the depth of the largest plane counts, which is the luma plane >>>> with a depth of 8. So for a width of 720 the value of bytesperline should be: >>>> >>>> depth=8 -> bytesperline = 720 >>>> depth=12 -> bytesperline = 720 >>> With depth=12, it should be, instead, 1080, as 2 pixels need 3 bytes. >> No, it's not. It's a *planar* format: first the Y plane, then the two smaller >> chroma planes. The spec says that bytesperline for planar formats refers to >> the largest plane. >> >> For this format the luma plane is one byte per pixel. Each of the two chroma >> planes have effectively two bits per pixel (actually one byte per four pixels), >> so you end up with 8+2+2=12 bits per pixel. >> >> Hence bytesperline should be 720 for this particular format. > If I understood what you just said, you're talking that the only format marked > as depth=12 is actually depth=8, right? Then the fix would be to change depth > in the table, and not here. No, the depth is correct. and it is needed for image size calculation. > Yet, I'm not sure if this is the proper fix. The problem here which causes confusion is, that bytesperline actually isn't bytesperline for planar formats (and hence size != bytesperline * height) when we follow the current spec. I don't understand the reason for handling this different for planar and non-planar formats, but I'm sure there is a good one. Hans ? Given that we can't and/or don't want to change the spec, the correct fix would be to consider the plane size in the calculation. This info can be derived either from the format type (implicit) or we can add it to the format struct (explicit). Example (using the ratio between the size of the largest plane and the size of all planes together): lp_ratio = 1 (for non-planar formats) lp_ratio = 4/(4+1+1) = 4/6 = 2/3 (for YUV411P) lp_ratio = 4/(4+2+2) = 4/8 = 1/2 (for YUV422P) ... => bytesperline = (width * depth * lp_ratio + 7) >> 3 Regards, Frank > The only used I saw on userspace apps for this field is to allocate size for > the memory buffer. Some userspace applications use to get bytesperline and > multiply by the image height and get the image size, instead of relying > on sizeimage, as some drivers didn't use to fill sizeimage properly. > > By using bytesperline equal to 1080 in this case warrants that the buffers > on userspace will have enough space. > > Regards, > Mauro -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/drivers/media/usb/em28xx/em28xx-video.c b/drivers/media/usb/em28xx/em28xx-video.c index 2eabf2a..6ced426 100644 --- a/drivers/media/usb/em28xx/em28xx-video.c +++ b/drivers/media/usb/em28xx/em28xx-video.c @@ -837,8 +837,8 @@ static int vidioc_g_fmt_vid_cap(struct file *file, void *priv, f->fmt.pix.width = dev->width; f->fmt.pix.height = dev->height; f->fmt.pix.pixelformat = dev->format->fourcc; - f->fmt.pix.bytesperline = (dev->width * dev->format->depth + 7) >> 3; - f->fmt.pix.sizeimage = f->fmt.pix.bytesperline * dev->height; + f->fmt.pix.bytesperline = dev->width * (dev->format->depth >> 3); + f->fmt.pix.sizeimage = (dev->width * dev->height * dev->format->depth + 7) >> 3; f->fmt.pix.colorspace = V4L2_COLORSPACE_SMPTE170M; /* FIXME: TOP? NONE? BOTTOM? ALTENATE? */ @@ -906,8 +906,8 @@ static int vidioc_try_fmt_vid_cap(struct file *file, void *priv, f->fmt.pix.width = width; f->fmt.pix.height = height; f->fmt.pix.pixelformat = fmt->fourcc; - f->fmt.pix.bytesperline = (dev->width * fmt->depth + 7) >> 3; - f->fmt.pix.sizeimage = f->fmt.pix.bytesperline * height; + f->fmt.pix.bytesperline = width * (fmt->depth >> 3); + f->fmt.pix.sizeimage = (width * height * fmt->depth + 7) >> 3; f->fmt.pix.colorspace = V4L2_COLORSPACE_SMPTE170M; if (dev->progressive) f->fmt.pix.field = V4L2_FIELD_NONE;
This was part of my original em28xx patch series. That particular patch combined two things: this fix and the change where TRY_FMT would no longer return -EINVAL for unsupported pixelformats. The latter change was rejected (correctly), but we all forgot about the second part of the patch which fixed a real bug. I'm reposting just that fix. Changes since v1: - v1 still miscalculated the bytesperline and imagesize values (they were too large). - G_FMT had the same calculation bug. Tested with my em28xx. Regards, Hans The bytesperline calculation was incorrect: it used the old width instead of the provided width in the case of TRY_FMT, and it miscalculated the bytesperline value for the depth == 12 (planar YUV 4:1:1) case. For planar formats the bytesperline value should be the bytesperline of the widest plane, which is the Y plane which has 8 bits per pixel, not 12. Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com> --- drivers/media/usb/em28xx/em28xx-video.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)