On Wed, Jun 10, 2020 at 7:01 PM Dafna Hirschfeld
<dafna.hirschfeld@xxxxxxxxxxxxx> wrote:
>
>
>
> On 27.05.20 00:44, Tomasz Figa wrote:
> > Hi Dafna,
> >
> > On Fri, May 15, 2020 at 4:30 PM Dafna Hirschfeld
> > <dafna.hirschfeld@xxxxxxxxxxxxx> wrote:
> >>
> >> The resize block of rkisp1 always get the input as yuv422.
> >> Instead of defining the default hdiv, vdiv as macros, the
> >> code is more clear if it takes the (hv)div from the YUV422P
> >> info. This makes it clear where those values come from.
> >> The patch also adds documentation to explain that.
> >>
> >> Signed-off-by: Dafna Hirschfeld <dafna.hirschfeld@xxxxxxxxxxxxx>
> >> ---
> >> drivers/staging/media/rkisp1/rkisp1-resizer.c | 25 +++++++++----------
> >> 1 file changed, 12 insertions(+), 13 deletions(-)
> >>
> >
> > Thank you for the effort on improving this driver! Please see my comments below.
> >
> >> diff --git a/drivers/staging/media/rkisp1/rkisp1-resizer.c b/drivers/staging/media/rkisp1/rkisp1-resizer.c
> >> index d049374413dc..04a29af8cc92 100644
> >> --- a/drivers/staging/media/rkisp1/rkisp1-resizer.c
> >> +++ b/drivers/staging/media/rkisp1/rkisp1-resizer.c
> >> @@ -16,9 +16,6 @@
> >> #define RKISP1_DEF_FMT MEDIA_BUS_FMT_YUYV8_2X8
> >> #define RKISP1_DEF_PIXEL_ENC V4L2_PIXEL_ENC_YUV
> >>
> >> -#define RKISP1_MBUS_FMT_HDIV 2
> >> -#define RKISP1_MBUS_FMT_VDIV 1
> >> -
> >> enum rkisp1_shadow_regs_when {
> >> RKISP1_SHADOW_REGS_SYNC,
> >> RKISP1_SHADOW_REGS_ASYNC,
> >> @@ -361,11 +358,12 @@ static void rkisp1_rsz_config_regs(struct rkisp1_resizer *rsz,
> >> static void rkisp1_rsz_config(struct rkisp1_resizer *rsz,
> >> enum rkisp1_shadow_regs_when when)
> >> {
> >> - u8 hdiv = RKISP1_MBUS_FMT_HDIV, vdiv = RKISP1_MBUS_FMT_VDIV;
> >> struct v4l2_rect sink_y, sink_c, src_y, src_c;
> >> struct v4l2_mbus_framefmt *src_fmt;
> >> struct v4l2_rect *sink_crop;
> >> struct rkisp1_capture *cap = &rsz->rkisp1->capture_devs[rsz->id];
> >> + const struct v4l2_format_info *yuv422_info =
> >> + v4l2_format_info(V4L2_PIX_FMT_YUV422P);
> >
> > As I mentioned in another reply, this is a memory format, but we're
> > using it to configure a local (i.e. non-DMA) input.
> >
> >>
> >> sink_crop = rkisp1_rsz_get_pad_crop(rsz, NULL, RKISP1_RSZ_PAD_SINK,
> >> V4L2_SUBDEV_FORMAT_ACTIVE);
> >> @@ -386,8 +384,9 @@ static void rkisp1_rsz_config(struct rkisp1_resizer *rsz,
> >> src_y.width = src_fmt->width;
> >> src_y.height = src_fmt->height;
> >>
> >> - sink_c.width = sink_y.width / RKISP1_MBUS_FMT_HDIV;
> >> - sink_c.height = sink_y.height / RKISP1_MBUS_FMT_VDIV;
> >> + /* The input format of the resizer is always yuv422 */
> >> + sink_c.width = sink_y.width / yuv422_info->hdiv;
> >> + sink_c.height = sink_y.height / yuv422_info->vdiv;
> >
> > I'd honestly go the opposite way and just make it:
> >
> > /* The resizer input is always YUV 4:2:2 and so horizontally subsampled by 2. */
> > sink_c.width = sink_y.width / 2;
> > sink_c.height = sink_y.height;
> >
> >>
> >> /*
> >> * The resizer is used not only to change the dimensions of the frame
> >> @@ -395,17 +394,17 @@ static void rkisp1_rsz_config(struct rkisp1_resizer *rsz,
> >> * (4:2:2 -> 4:2:0 for example). So the width/height of the CbCr
> >> * streams should be set according to the pixel format in the capture.
> >> * The resizer always gets the input as YUV422. If the capture format
> >> - * is RGB then the memory input should be YUV422 so we don't change the
> >> - * default hdiv, vdiv in that case.
> >> + * is RGB then the memory input (the resizer output) should be YUV422
> >
> > It could be just me, but "memory input" sounds like there was an input
> > DMA involved, which is not the case. How about "Memory Interface
> > input"?
> >
> > Also, if already amending this, I would also fix the overuse of
> > "should". How about the following?
> >
> > "According to the hardware pipeline, the resizer input is always YUV
> > 4:2:2. For YUV formats, the Memory Interface requires its input to be
> > already properly subsampled. We can achieve subsampling factors other
> > than YUV 4:2:2 by scaling the planes appropriately. For RGB formats,
> > the Memory Interface requires YUV 4:2:2 input, so no additional
> > scaling is needed."
> >
> >> + * so we use the hdiv, vdiv of the YUV422 info in this case.
> >> */
> >> if (v4l2_is_format_yuv(cap->pix.info)) {
> >> - hdiv = cap->pix.info->hdiv;
> >> - vdiv = cap->pix.info->vdiv;
> >> + src_c.width = cap->pix.info->hdiv;
> >> + src_c.height = cap->pix.info->vdiv;
> >> + } else {
> >
> > How about making it an explicit else if (v4l2_is_format_rgb(...))?
> Actually the right way to deal with YUV downscaling is to support
> MEDIA_BUS_FMT_YUYV8_1_5X8 (which is YUV420) on the src format
> and not to look on what is configured on the capture.
Good point. Since we're dealing with the resizer configuration, it
should be contained within the resizer subdev. My mind was still stuck
with our downstream topology.
Best regards,
Tomasz
