Re: [PATCH v4 1/2] drm/format-helper: Add conversion from XRGB8888 to BGR888

["andriy.shevchenko@linux.intel.com" <andriy.shevchenko@linux.intel.com>]

On Mon, Feb 24, 2025 at 02:54:07PM +0000, Aditya Garg wrote:
> This conversion helper mimics the existing drm_fb_xrgb8888_to_rgb888 helper

Not really. See below.

> > On 24 Feb 2025, at 7:59 PM, andriy.shevchenko@linux.intel.com wrote:
> > On Mon, Feb 24, 2025 at 01:38:32PM +0000, Aditya Garg wrote:

...

> >> +static void drm_fb_xrgb8888_to_bgr888_line(void *dbuf, const void *sbuf, unsigned int pixels)
> > 
> > Okay the xrgb8888 is the actual pixel format independently on
> > the CPU endianess.
> > 
> >> +{
> >> + u8 *dbuf8 = dbuf;
> >> + const __le32 *sbuf32 = sbuf;
> > 
> > But here we assume that sbuf is __le32.
> > And I think we may benefit from the __be32 there.
> 
> So, like drm_fb_xrgb8888_to_rgb888, we are using __le32
> 
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/gpu/drm/drm_format_helper.c?h=v6.14-rc4#n657

The rgb888 != bgr888, that's where the byte swapping happens. So, one should
use __be32 if the other has already been using __le32.

> >> + unsigned int x;
> >> + u32 pix;
> >> +
> >> + for (x = 0; x < pixels; x++) {
> >> + pix = le32_to_cpu(sbuf32[x]);
> >> + /* write red-green-blue to output in little endianness */
> >> + *dbuf8++ = (pix & 0x00ff0000) >> 16;
> >> + *dbuf8++ = (pix & 0x0000ff00) >> 8;
> >> + *dbuf8++ = (pix & 0x000000ff) >> 0;
> > 
> > pix = be32_to_cpu(sbuf[4 * x]) >> 8;
> > put_unaligned_le24(pix, &dbuf[3 * x]);
> 
> Again, 
> 
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/gpu/drm/drm_format_helper.c?h=v6.14-rc4#n664

As per above.

> >> + }
> > 
> > Or, after all, this __le32 magic might be not needed at all. Wouldn't the below
> > be the equivalent
> > 
> > static void drm_fb_xrgb8888_to_bgr888_line(void *dbuf, const void *sbuf, unsigned int pixels)
> > {
> > unsigned int x;
> > u32 pix;
> > 
> > for (x = 0; x < pixels; x++) {
> > /* Read red-green-blue from input in big endianess and... */
> > pix = get_unaligned_be24(sbuf + x * 4 + 1);
> > /* ...write it to output in little endianness. */
> > put_unaligned_le24(pix, dbuf + x * 3);
> > }
> > }
> > 
> > The comments can even be dropped as the code quite clear about what's going on.
> 
> These comments are literally rewritten :
> 
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/gpu/drm/drm_format_helper.c?h=v6.14-rc4#n663
> 
> >> +}
> > 
> > But it's up to you. I don't know which solution gives better code generation
> > either.
> 
> I don't really mind any code change tbh, but I’d prefer that as an
> improvement to existing code, and not a part of this patchset.

Right, but see my argumentation above.

-- 
With Best Regards,
Andy Shevchenko