media: bcm2835-isp: fix bytes per line calculations for some image fo… #3645
Conversation
| } | ||
| else | ||
| bpl = (bpl + align_minus_1) & ~align_minus_1; | ||
| return bpl; | ||
| } |
There was a problem hiding this comment.
Not entirely sure, but I would guess the upstream folks might object to using align_minus_1 variable over an explicit (align - 1).
There was a problem hiding this comment.
No problem. Given that I was using it several times I thought I'd make it explicit that I expected the value to be re-used, but if we think (a) the compiler would do it anyway and (b) upstream maintainers might object, then I'll happily change it. Is that the general feeling?
There was a problem hiding this comment.
They would probably prefer you to use the existing ALIGN and IS_ALIGNED macros.
There was a problem hiding this comment.
Personally I don't think we need the complication with manually working out if it's a bitmask or not.
24bpp is the only format that has quirks, and you have a near match with https://github.com/raspberrypi/linux/blob/rpi-5.4.y/drivers/staging/vc04_services/bcm2835-camera/bcm2835-camera.c#L994
if (fmt->depth == 24)
return <whatever the correct align is here, and it doesn't need to be 96>
else
return ((width * fmt->depth) >> 3, fmt->bytesperline_align);
Admittedly it's not a performance critical path, but manually computing whether we need to do the modulo seems nuts.
MMAL sets the stride via a multiple of width * bpp. 32pixels * 24bpp is 96 bytes, which fulfils the ISP requirement for a multiple of 32 bytes. So I think we can do
return ((ALIGN(width, fmt->bytesperline_align) * fmt->depth) >> 3);
and leave the bytesperline_align at 32 (or hardcode it here).
(edited - it helps to actually call the ALIGN macro)
…rmats The bytes per line numbers calculated by get_bytesperline was not matching the equivalent calculation being performed by the VideoCore (mostly by the calculate_pitch function there), resulting in failures to set the image format with some image width values. This patches up the RGB24 and YUYV type formats to match the VideoCore calculation. Signed-off-by: David Plowman <david.plowman@raspberrypi.com>
davidplowman
force-pushed
the
isp_align_fix
branch
from
4d178db
to
62facef
Compare
|
New version that hardcodes the special case. |
|
I'm happy with that. |
kernel: vc4: Set driver_name for card See: raspberrypi/linux#3656 kernel: configs: Add SND_SOC_MAX98357A=m See: https://www.raspberrypi.org/forums/viewtopic.php?f=107&t=275919 kernel: Add Micro Crystal RV-1805 to i2c-rtc overlays See: raspberrypi/linux#3651 kernel: media: bcm2835-isp: fix bytes per line calculations for some image formats See: raspberrypi/linux#3645 kernel: media: bcm2835-unicam: change minimum number of vb2_queue buffers to 1 See: raspberrypi/linux#3638 kernel: media: i2c: imx477: Return correct result on sensor id verification See: raspberrypi/linux#3630 kernel: Clean up the VCHIQ 2711 DMA support See: raspberrypi/linux#3629 kernel: overlays: i2c-rtc-gpio: Fix trickle-resistor-ohms param See: raspberrypi/linux#3642 kernel: Enable hwmon for thermal zones See: raspberrypi/linux#3307
kernel: vc4: Set driver_name for card See: raspberrypi/linux#3656 kernel: configs: Add SND_SOC_MAX98357A=m See: https://www.raspberrypi.org/forums/viewtopic.php?f=107&t=275919 kernel: Add Micro Crystal RV-1805 to i2c-rtc overlays See: raspberrypi/linux#3651 kernel: media: bcm2835-isp: fix bytes per line calculations for some image formats See: raspberrypi/linux#3645 kernel: media: bcm2835-unicam: change minimum number of vb2_queue buffers to 1 See: raspberrypi/linux#3638 kernel: media: i2c: imx477: Return correct result on sensor id verification See: raspberrypi/linux#3630 kernel: Clean up the VCHIQ 2711 DMA support See: raspberrypi/linux#3629 kernel: overlays: i2c-rtc-gpio: Fix trickle-resistor-ohms param See: raspberrypi/linux#3642 kernel: Enable hwmon for thermal zones See: raspberrypi/linux#3307
…rmats
The bytes per line numbers calculated by get_bytesperline was not
matching the equivalent calculation being performed by the VideoCore
(mostly by the calculate_pitch function there), resulting in failures
to set the image format with some image width values. This patches up
the RGB24 and YUYV type formats to match the VideoCore calculation.
Signed-off-by: David Plowman david.plowman@raspberrypi.com