Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751806AbdFZKHV (ORCPT ); Mon, 26 Jun 2017 06:07:21 -0400 Received: from lb1-smtp-cloud2.xs4all.net ([194.109.24.21]:42481 "EHLO lb1-smtp-cloud2.xs4all.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751387AbdFZKHP (ORCPT ); Mon, 26 Jun 2017 06:07:15 -0400 Subject: Re: [PATCH v1 3/5] [media] stm32-dcmi: crop sensor image to match user resolution To: Hugues FRUCHET , Maxime Coquelin , Alexandre TORGUE , Mauro Carvalho Chehab References: <1498144371-13310-1-git-send-email-hugues.fruchet@st.com> <1498144371-13310-4-git-send-email-hugues.fruchet@st.com> <8ae4c160-3137-0fa8-3d78-e4e1284521a4@st.com> Cc: "devicetree@vger.kernel.org" , "linux-arm-kernel@lists.infradead.org" , "linux-kernel@vger.kernel.org" , "linux-media@vger.kernel.org" , Benjamin Gaignard , Yannick FERTRE From: Hans Verkuil Message-ID: <2844ed1a-d310-4381-c4a7-88ab4633f32f@xs4all.nl> Date: Mon, 26 Jun 2017 12:07:07 +0200 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:45.0) Gecko/20100101 Thunderbird/45.7.1 MIME-Version: 1.0 In-Reply-To: <8ae4c160-3137-0fa8-3d78-e4e1284521a4@st.com> Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 6210 Lines: 176 On 26/06/17 11:53, Hugues FRUCHET wrote: > Hi Hans, thanks for review. > > Reply below. > > BR > Hugues. > > On 06/22/2017 05:19 PM, Hans Verkuil wrote: >> On 06/22/2017 05:12 PM, Hugues Fruchet wrote: >>> Add flexibility on supported resolutions by cropping sensor >>> image to fit user resolution format request. >>> >>> Signed-off-by: Hugues Fruchet >>> --- >>> drivers/media/platform/stm32/stm32-dcmi.c | 54 ++++++++++++++++++++++++++++++- >>> 1 file changed, 53 insertions(+), 1 deletion(-) >>> >>> diff --git a/drivers/media/platform/stm32/stm32-dcmi.c b/drivers/media/platform/stm32/stm32-dcmi.c >>> index 75d53aa..bc5e052 100644 >>> --- a/drivers/media/platform/stm32/stm32-dcmi.c >>> +++ b/drivers/media/platform/stm32/stm32-dcmi.c >>> @@ -131,6 +131,8 @@ struct stm32_dcmi { >>> struct v4l2_async_notifier notifier; >>> struct dcmi_graph_entity entity; >>> struct v4l2_format fmt; >>> + struct v4l2_rect crop; >>> + bool do_crop; >>> >>> const struct dcmi_format **user_formats; >>> unsigned int num_user_formats; >>> @@ -538,6 +540,27 @@ static int dcmi_start_streaming(struct vb2_queue *vq, unsigned int count) >>> if (dcmi->bus.flags & V4L2_MBUS_PCLK_SAMPLE_RISING) >>> val |= CR_PCKPOL; >>> >>> + if (dcmi->do_crop) { >>> + u32 size, start; >>> + >>> + /* Crop resolution */ >>> + size = ((dcmi->crop.height - 1) << 16) | >>> + ((dcmi->crop.width << 1) - 1); >>> + reg_write(dcmi->regs, DCMI_CWSIZE, size); >>> + >>> + /* Crop start point */ >>> + start = ((dcmi->crop.top) << 16) | >>> + ((dcmi->crop.left << 1)); >>> + reg_write(dcmi->regs, DCMI_CWSTRT, start); >>> + >>> + dev_dbg(dcmi->dev, "Cropping to %ux%u@%u:%u\n", >>> + dcmi->crop.width, dcmi->crop.height, >>> + dcmi->crop.left, dcmi->crop.top); >>> + >>> + /* Enable crop */ >>> + val |= CR_CROP; >>> + }; >>> + >>> reg_write(dcmi->regs, DCMI_CR, val); >>> >>> /* Enable dcmi */ >>> @@ -707,6 +730,8 @@ static int dcmi_try_fmt(struct stm32_dcmi *dcmi, struct v4l2_format *f, >>> .which = V4L2_SUBDEV_FORMAT_TRY, >>> }; >>> int ret; >>> + __u32 width, height; >>> + struct v4l2_mbus_framefmt *mf = &format.format; >>> >>> dcmi_fmt = find_format_by_fourcc(dcmi, pixfmt->pixelformat); >>> if (!dcmi_fmt) { >>> @@ -724,8 +749,18 @@ static int dcmi_try_fmt(struct stm32_dcmi *dcmi, struct v4l2_format *f, >>> if (ret < 0) >>> return ret; >>> >>> + /* Align format on what sensor can do */ >>> + width = pixfmt->width; >>> + height = pixfmt->height; >>> v4l2_fill_pix_format(pixfmt, &format.format); >>> >>> + /* We can do any resolution thanks to crop */ >>> + if ((mf->width > width) || (mf->height > height)) { >>> + /* Restore width/height */ >>> + pixfmt->width = width; >>> + pixfmt->height = height; >>> + }; >>> + >>> pixfmt->field = V4L2_FIELD_NONE; >>> pixfmt->bytesperline = pixfmt->width * dcmi_fmt->bpp; >>> pixfmt->sizeimage = pixfmt->bytesperline * pixfmt->height; >>> @@ -741,6 +776,8 @@ static int dcmi_set_fmt(struct stm32_dcmi *dcmi, struct v4l2_format *f) >>> struct v4l2_subdev_format format = { >>> .which = V4L2_SUBDEV_FORMAT_ACTIVE, >>> }; >>> + struct v4l2_mbus_framefmt *mf = &format.format; >>> + struct v4l2_pix_format *pixfmt = &f->fmt.pix; >>> const struct dcmi_format *current_fmt; >>> int ret; >>> >>> @@ -748,13 +785,28 @@ static int dcmi_set_fmt(struct stm32_dcmi *dcmi, struct v4l2_format *f) >>> if (ret) >>> return ret; >>> >>> - v4l2_fill_mbus_format(&format.format, &f->fmt.pix, >>> + v4l2_fill_mbus_format(&format.format, pixfmt, >>> current_fmt->mbus_code); >>> ret = v4l2_subdev_call(dcmi->entity.subdev, pad, >>> set_fmt, NULL, &format); >>> if (ret < 0) >>> return ret; >>> >>> + /* Enable crop if sensor resolution is larger than request */ >>> + dcmi->do_crop = false; >>> + if ((mf->width > pixfmt->width) || (mf->height > pixfmt->height)) { >>> + dcmi->crop.width = pixfmt->width; >>> + dcmi->crop.height = pixfmt->height; >>> + dcmi->crop.left = (mf->width - pixfmt->width) / 2; >>> + dcmi->crop.top = (mf->height - pixfmt->height) / 2; >>> + dcmi->do_crop = true; >> >> Why not implement the selection API instead? I assume that you can crop from any >> region of the sensor, not just the center part. > > The point here was to add some flexibility for user in term of > resolution and also less memory consumption. > For example here I want to make a 480x272 preview: > - without this change: S_FMT(480x272) returns VGA (the OV9655 larger > discrete resolution), then app has to capture VGA frames then crop to > fit 480x272 frame buffer. > - with this change: S_FMT(480x272) returns 480x272 (crop done by > hardware), app can directly capture 480x272 then copy to framebuffer > without any conversion. > > Implementation of V4L2 crop using SELECTION API could also be used, > but I need to change app. > > More generally, with a given couple ISP+sensor, will S_FMT() > return the sensor only supported resolutions ? or the supported > resolutions of the couple ISP+sensor (ISP will downscale/upscale/crop > the sensor discrete resolution to fit user request) ? > > Hans, what are your recommendations ? This is not the way the V4L2 API works. Sensors report their supported resolutions through the ENUM_FRAMESIZES (and the related ENUM_FRAMEINTERVALS) ioctls. With S_FMT you pick the resolution you want. If you then want to crop the driver should implement the selection API (this will also automatically enable the G/S_CROP/CROPCAP ioctls) so the application can select which part of the image should be cropped. Assuming there is no scaler then after cropping the format size will be reduced to the size of the cropped image. That's in a nutshell how these ioctls relate to one another. Regards, Hans > >> >> Regards, >> >> Hans >> >>> + >>> + dev_dbg(dcmi->dev, "%ux%u cropped to %ux%u@(%u,%u)\n", >>> + mf->width, mf->height, >>> + dcmi->crop.width, dcmi->crop.height, >>> + dcmi->crop.left, dcmi->crop.top); >>> + }; >>> + >>> dcmi->fmt = *f; >>> dcmi->current_fmt = current_fmt; >>> >>>