Re: [PATCH v5 9/9] drm: vkms: Add support to the RGB565 format

[Pekka Paalanen <ppaalanen@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 7506 bytes --]

On Mon,  4 Apr 2022 17:45:15 -0300
Igor Torrente <igormtorrente@gmail.com> wrote:

> Adds this common format to vkms.
> 
> This commit also adds new helper macros to deal with fixed-point
> arithmetic.
> 
> It was done to improve the precision of the conversion to ARGB16161616
> since the "conversion ratio" is not an integer.
> 
> V3: Adapt the handlers to the new format introduced in patch 7 V3.
> V5: Minor improvements
> 
> Signed-off-by: Igor Torrente <igormtorrente@gmail.com>
> ---
>  drivers/gpu/drm/vkms/vkms_formats.c   | 70 +++++++++++++++++++++++++++
>  drivers/gpu/drm/vkms/vkms_plane.c     |  6 ++-
>  drivers/gpu/drm/vkms/vkms_writeback.c |  3 +-
>  3 files changed, 76 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/gpu/drm/vkms/vkms_formats.c b/drivers/gpu/drm/vkms/vkms_formats.c
> index 8d913fa7dbde..4af8b295f31e 100644
> --- a/drivers/gpu/drm/vkms/vkms_formats.c
> +++ b/drivers/gpu/drm/vkms/vkms_formats.c
> @@ -5,6 +5,23 @@
>  
>  #include "vkms_formats.h"
>  
> +/* The following macros help doing fixed point arithmetic. */
> +/*
> + * With Fixed-Point scale 15 we have 17 and 15 bits of integer and fractional
> + * parts respectively.
> + *  | 0000 0000 0000 0000 0.000 0000 0000 0000 |
> + * 31                                          0
> + */
> +#define FIXED_SCALE 15

I think this would usually be called a "shift" since it's used in
bit-shifts.

> +
> +#define INT_TO_FIXED(a) ((a) << FIXED_SCALE)
> +#define FIXED_MUL(a, b) ((s32)(((s64)(a) * (b)) >> FIXED_SCALE))
> +#define FIXED_DIV(a, b) ((s32)(((s64)(a) << FIXED_SCALE) / (b)))

A truncating div, ok.

> +/* This macro converts a fixed point number to int, and round half up it */
> +#define FIXED_TO_INT_ROUND(a) (((a) + (1 << (FIXED_SCALE - 1))) >> FIXED_SCALE)

Yes.

> +/* Convert divisor and dividend to Fixed-Point and performs the division */
> +#define INT_TO_FIXED_DIV(a, b) (FIXED_DIV(INT_TO_FIXED(a), INT_TO_FIXED(b)))

Ok, this is obvious to read, even though it's the same as FIXED_DIV()
alone. Not sure the compiler would optimize that extra bit-shift away...

If one wanted to, it would be possible to write type-safe functions for
these so that fixed and integer could not be mixed up.

> +
>  static int pixel_offset(const struct vkms_frame_info *frame_info, int x, int y)
>  {
>  	return frame_info->offset + (y * frame_info->pitch)
> @@ -112,6 +129,30 @@ static void XRGB16161616_to_argb_u16(struct line_buffer *stage_buffer,
>  	}
>  }
>  
> +static void RGB565_to_argb_u16(struct line_buffer *stage_buffer,
> +			       const struct vkms_frame_info *frame_info, int y)
> +{
> +	struct pixel_argb_u16 *out_pixels = stage_buffer->pixels;
> +	u16 *src_pixels = get_packed_src_addr(frame_info, y);
> +	int x, x_limit = min_t(size_t, drm_rect_width(&frame_info->dst),
> +			       stage_buffer->n_pixels);
> +
> +	for (x = 0; x < x_limit; x++, src_pixels++) {
> +		u16 rgb_565 = le16_to_cpu(*src_pixels);
> +		int fp_r = INT_TO_FIXED((rgb_565 >> 11) & 0x1f);
> +		int fp_g = INT_TO_FIXED((rgb_565 >> 5) & 0x3f);
> +		int fp_b = INT_TO_FIXED(rgb_565 & 0x1f);
> +
> +		int fp_rb_ratio = INT_TO_FIXED_DIV(65535, 31);
> +		int fp_g_ratio = INT_TO_FIXED_DIV(65535, 63);

These two should be outside of the loop since they are constants.
Likely no difference for performance because the compiler is probably
doing that already, but I think it would read better.

> +
> +		out_pixels[x].a = (u16)0xffff;
> +		out_pixels[x].r = FIXED_TO_INT_ROUND(FIXED_MUL(fp_r, fp_rb_ratio));
> +		out_pixels[x].g = FIXED_TO_INT_ROUND(FIXED_MUL(fp_g, fp_g_ratio));
> +		out_pixels[x].b = FIXED_TO_INT_ROUND(FIXED_MUL(fp_b, fp_rb_ratio));

Looks good.

> +	}
> +}
> +
>  
>  /*
>   * The following  functions take an line of argb_u16 pixels from the
> @@ -199,6 +240,31 @@ static void argb_u16_to_XRGB16161616(struct vkms_frame_info *frame_info,
>  	}
>  }
>  
> +static void argb_u16_to_RGB565(struct vkms_frame_info *frame_info,
> +			       const struct line_buffer *src_buffer, int y)
> +{
> +	int x, x_dst = frame_info->dst.x1;
> +	u16 *dst_pixels = packed_pixels_addr(frame_info, x_dst, y);
> +	struct pixel_argb_u16 *in_pixels = src_buffer->pixels;
> +	int x_limit = min_t(size_t, drm_rect_width(&frame_info->dst),
> +			    src_buffer->n_pixels);
> +
> +	for (x = 0; x < x_limit; x++, dst_pixels++) {
> +		int fp_r = INT_TO_FIXED(in_pixels[x].r);
> +		int fp_g = INT_TO_FIXED(in_pixels[x].g);
> +		int fp_b = INT_TO_FIXED(in_pixels[x].b);
> +
> +		int fp_rb_ratio = INT_TO_FIXED_DIV(65535, 31);
> +		int fp_g_ratio = INT_TO_FIXED_DIV(65535, 63);

Move these out of the loop.

> +
> +		u16 r = FIXED_TO_INT_ROUND(FIXED_DIV(fp_r, fp_rb_ratio));
> +		u16 g = FIXED_TO_INT_ROUND(FIXED_DIV(fp_g, fp_g_ratio));
> +		u16 b = FIXED_TO_INT_ROUND(FIXED_DIV(fp_b, fp_rb_ratio));
> +
> +		*dst_pixels = cpu_to_le16(r << 11 | g << 5 | b);

Looks good.

You are using signed variables (int, s64, s32) when negative values
should never occur. It doesn't seem wrong, just unexpected.

The use of int in code vs. s32 in the macros is a bit inconsistent as
well.

> +	}
> +}
> +
>  plane_format_transform_func get_plane_fmt_transform_function(u32 format)
>  {
>  	if (format == DRM_FORMAT_ARGB8888)
> @@ -209,6 +275,8 @@ plane_format_transform_func get_plane_fmt_transform_function(u32 format)
>  		return &ARGB16161616_to_argb_u16;
>  	else if (format == DRM_FORMAT_XRGB16161616)
>  		return &XRGB16161616_to_argb_u16;
> +	else if (format == DRM_FORMAT_RGB565)
> +		return &RGB565_to_argb_u16;
>  	else
>  		return NULL;
>  }
> @@ -223,6 +291,8 @@ wb_format_transform_func get_wb_fmt_transform_function(u32 format)
>  		return &argb_u16_to_ARGB16161616;
>  	else if (format == DRM_FORMAT_XRGB16161616)
>  		return &argb_u16_to_XRGB16161616;
> +	else if (format == DRM_FORMAT_RGB565)
> +		return &argb_u16_to_RGB565;

Now it's starting to become clear that a switch statement would be nice.

>  	else
>  		return NULL;
>  }
> diff --git a/drivers/gpu/drm/vkms/vkms_plane.c b/drivers/gpu/drm/vkms/vkms_plane.c
> index 60054a85204a..94a8e412886f 100644
> --- a/drivers/gpu/drm/vkms/vkms_plane.c
> +++ b/drivers/gpu/drm/vkms/vkms_plane.c
> @@ -14,14 +14,16 @@
>  
>  static const u32 vkms_formats[] = {
>  	DRM_FORMAT_XRGB8888,
> -	DRM_FORMAT_XRGB16161616
> +	DRM_FORMAT_XRGB16161616,
> +	DRM_FORMAT_RGB565
>  };
>  
>  static const u32 vkms_plane_formats[] = {
>  	DRM_FORMAT_ARGB8888,
>  	DRM_FORMAT_XRGB8888,
>  	DRM_FORMAT_XRGB16161616,
> -	DRM_FORMAT_ARGB16161616
> +	DRM_FORMAT_ARGB16161616,
> +	DRM_FORMAT_RGB565
>  };
>  
>  static struct drm_plane_state *
> diff --git a/drivers/gpu/drm/vkms/vkms_writeback.c b/drivers/gpu/drm/vkms/vkms_writeback.c
> index cb63a5da9af1..98da7bee0f4b 100644
> --- a/drivers/gpu/drm/vkms/vkms_writeback.c
> +++ b/drivers/gpu/drm/vkms/vkms_writeback.c
> @@ -16,7 +16,8 @@
>  static const u32 vkms_wb_formats[] = {
>  	DRM_FORMAT_XRGB8888,
>  	DRM_FORMAT_XRGB16161616,
> -	DRM_FORMAT_ARGB16161616
> +	DRM_FORMAT_ARGB16161616,
> +	DRM_FORMAT_RGB565
>  };
>  
>  static const struct drm_connector_funcs vkms_wb_connector_funcs = {

I wonder, would it be possible to add a unit test to make sure that
get_plane_fmt_transform_function() or get_wb_fmt_transform_function()
does not return NULL for any of the listed formats, respectively?
Or is that too paranoid?


Thanks,
pq

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]