Re: [PATCH v6 08/14] drm/mediatek: Support "Pre-multiplied" alpha blending in OVL

["CK Hu (胡俊光)" <ck.hu@mediatek.com>]

Hi, Shawn:

On Fri, 2024-03-22 at 13:28 +0800, Shawn Sung wrote:
> From: Hsiao Chien Sung <shawn.sung@mediatek.com>
> 
> Support "Pre-multiplied" blend mode on MediaTek's chips.
> Before this patch, only the "coverage" mode is supported.
> 
> Please refer to the description of the commit
> "drm/mediatek: Support alpha blending in display driver"
> for more information.
> 
> Signed-off-by: Hsiao Chien Sung <shawn.sung@mediatek.com>
> ---
>  drivers/gpu/drm/mediatek/mtk_disp_ovl.c | 84 +++++++++++++++++++++
> ----
>  1 file changed, 73 insertions(+), 11 deletions(-)
> 
> diff --git a/drivers/gpu/drm/mediatek/mtk_disp_ovl.c
> b/drivers/gpu/drm/mediatek/mtk_disp_ovl.c
> index 2a767a823c83a..b1e5d453316cc 100644
> --- a/drivers/gpu/drm/mediatek/mtk_disp_ovl.c
> +++ b/drivers/gpu/drm/mediatek/mtk_disp_ovl.c
> @@ -38,6 +38,7 @@
>  #define DISP_REG_OVL_PITCH_MSB(n)		(0x0040 + 0x20 * (n))
>  #define OVL_PITCH_MSB_2ND_SUBBUF			BIT(16)
>  #define DISP_REG_OVL_PITCH(n)			(0x0044 + 0x20
> * (n))
> +#define OVL_CONST_BLEND					BIT(28)
>  #define DISP_REG_OVL_RDMA_CTRL(n)		(0x00c0 + 0x20 * (n))
>  #define DISP_REG_OVL_RDMA_GMC(n)		(0x00c8 + 0x20 * (n))
>  #define DISP_REG_OVL_ADDR_MT2701		0x0040
> @@ -51,13 +52,16 @@
>  #define GMC_THRESHOLD_HIGH	((1 << GMC_THRESHOLD_BITS) / 4)
>  #define GMC_THRESHOLD_LOW	((1 << GMC_THRESHOLD_BITS) / 8)
>  
> +#define OVL_CON_CLRFMT_MAN	BIT(23)
>  #define OVL_CON_BYTE_SWAP	BIT(24)
> -#define OVL_CON_MTX_YUV_TO_RGB	(6 << 16)
> +#define OVL_CON_RGB_SWAP	BIT(25)
>  #define OVL_CON_CLRFMT_RGB	(1 << 12)
>  #define OVL_CON_CLRFMT_RGBA8888	(2 << 12)
>  #define OVL_CON_CLRFMT_ARGB8888	(3 << 12)
>  #define OVL_CON_CLRFMT_UYVY	(4 << 12)
>  #define OVL_CON_CLRFMT_YUYV	(5 << 12)
> +#define OVL_CON_MTX_YUV_TO_RGB	(6 << 16)
> +#define OVL_CON_CLRFMT_PARGB8888	(OVL_CON_CLRFMT_ARGB8888 |
> OVL_CON_CLRFMT_MAN)
>  #define OVL_CON_CLRFMT_RGB565(ovl)	((ovl)->data->fmt_rgb565_is_0 ?
> \
>  					0 : OVL_CON_CLRFMT_RGB)
>  #define OVL_CON_CLRFMT_RGB888(ovl)	((ovl)->data->fmt_rgb565_is_0 ?
> \
> @@ -71,6 +75,24 @@
>  #define	OVL_CON_VIRT_FLIP	BIT(9)
>  #define	OVL_CON_HORZ_FLIP	BIT(10)
>  
> +#define OVL_COLOR_ALPHA		GENMASK(31, 24)
> +
> +static inline bool is_10bit_rgb(u32 fmt)
> +{
> +	switch (fmt) {
> +	case DRM_FORMAT_XRGB2101010:
> +	case DRM_FORMAT_ARGB2101010:
> +	case DRM_FORMAT_RGBX1010102:
> +	case DRM_FORMAT_RGBA1010102:
> +	case DRM_FORMAT_XBGR2101010:
> +	case DRM_FORMAT_ABGR2101010:
> +	case DRM_FORMAT_BGRX1010102:
> +	case DRM_FORMAT_BGRA1010102:
> +		return true;
> +	}
> +	return false;
> +}
> +
>  static const u32 mt8173_formats[] = {
>  	DRM_FORMAT_XRGB8888,
>  	DRM_FORMAT_ARGB8888,
> @@ -88,12 +110,20 @@ static const u32 mt8173_formats[] = {
>  static const u32 mt8195_formats[] = {
>  	DRM_FORMAT_XRGB8888,
>  	DRM_FORMAT_ARGB8888,
> +	DRM_FORMAT_XRGB2101010,

As I know, the 'X' in XRGB means garbage data. It's not alpha. And pre-
multiplied alpha formula is:

result = source RGB + (destination RGB * (1 - source alpha))

So pre-multiplied alpha need source alpha. I think XRGB is not related
to pre-multiplied alpha. So separate X-related format to another patch.

>  	DRM_FORMAT_ARGB2101010,
>  	DRM_FORMAT_BGRX8888,
>  	DRM_FORMAT_BGRA8888,
> +	DRM_FORMAT_BGRX1010102,
>  	DRM_FORMAT_BGRA1010102,
>  	DRM_FORMAT_ABGR8888,
>  	DRM_FORMAT_XBGR8888,
> +	DRM_FORMAT_XBGR2101010,
> +	DRM_FORMAT_ABGR2101010,

It seems 10 bits format has no pre-multiplied alpha function. So
separate 10 bits format modification to another patch.

> +	DRM_FORMAT_RGBX8888,
> +	DRM_FORMAT_RGBA8888,

For RGBA8888, it both has coverage mode and pre-multiplied mode, so
it's not strongly related to pre-multiplied alpha. Separate this to
another patch which add support this new format.

> +	DRM_FORMAT_RGBX1010102,
> +	DRM_FORMAT_RGBA1010102,
>  	DRM_FORMAT_RGB888,
>  	DRM_FORMAT_BGR888,
>  	DRM_FORMAT_RGB565,
> @@ -253,9 +283,7 @@ static void mtk_ovl_set_bit_depth(struct device
> *dev, int idx, u32 format,
>  	reg = readl(ovl->regs + DISP_REG_OVL_CLRFMT_EXT);
>  	reg &= ~OVL_CON_CLRFMT_BIT_DEPTH_MASK(idx);
>  
> -	if (format == DRM_FORMAT_RGBA1010102 ||
> -	    format == DRM_FORMAT_BGRA1010102 ||
> -	    format == DRM_FORMAT_ARGB2101010)
> +	if (is_10bit_rgb(format))
>  		bit_depth = OVL_CON_CLRFMT_10_BIT;
>  
>  	reg |= OVL_CON_CLRFMT_BIT_DEPTH(bit_depth, idx);
> @@ -273,7 +301,13 @@ void mtk_ovl_config(struct device *dev, unsigned
> int w,
>  	if (w != 0 && h != 0)
>  		mtk_ddp_write_relaxed(cmdq_pkt, h << 16 | w, &ovl-
> >cmdq_reg, ovl->regs,
>  				      DISP_REG_OVL_ROI_SIZE);
> -	mtk_ddp_write_relaxed(cmdq_pkt, 0x0, &ovl->cmdq_reg, ovl->regs, 
> DISP_REG_OVL_ROI_BGCLR);
> +
> +	/*
> +	 * The background color should be opaque black (ARGB),
> +	 * otherwise there will be no effect with alpha blend
> +	 */
> +	mtk_ddp_write_relaxed(cmdq_pkt, OVL_COLOR_ALPHA, &ovl-
> >cmdq_reg,
> +			      ovl->regs, DISP_REG_OVL_ROI_BGCLR);

The pre-multiplied alpha formula is:

result = source RGB + (destination RGB * (1 - source alpha))

It does not care destination alpha. Background color is the
destination, so describe more detail why need this modification. For
example, how does hardware calculate with back ground color?

>  
>  	mtk_ddp_write(cmdq_pkt, 0x1, &ovl->cmdq_reg, ovl->regs,
> DISP_REG_OVL_RST);
>  	mtk_ddp_write(cmdq_pkt, 0x0, &ovl->cmdq_reg, ovl->regs,
> DISP_REG_OVL_RST);
> @@ -353,7 +387,8 @@ void mtk_ovl_layer_off(struct device *dev,
> unsigned int idx,
>  		      DISP_REG_OVL_RDMA_CTRL(idx));
>  }
>  
> -static unsigned int ovl_fmt_convert(struct mtk_disp_ovl *ovl,
> unsigned int fmt)
> +static unsigned int ovl_fmt_convert(struct mtk_disp_ovl *ovl,
> unsigned int fmt,
> +				    unsigned int blend_mode)
>  {
>  	/* The return value in switch "MEM_MODE_INPUT_FORMAT_XXX"
>  	 * is defined in mediatek HW data sheet.
> @@ -372,17 +407,37 @@ static unsigned int ovl_fmt_convert(struct
> mtk_disp_ovl *ovl, unsigned int fmt)
>  		return OVL_CON_CLRFMT_RGB888(ovl) | OVL_CON_BYTE_SWAP;
>  	case DRM_FORMAT_RGBX8888:
>  	case DRM_FORMAT_RGBA8888:
> +		return blend_mode == DRM_MODE_BLEND_COVERAGE ?
> +		       OVL_CON_CLRFMT_ARGB8888 :
> +		       OVL_CON_CLRFMT_PARGB8888;
> +	case DRM_FORMAT_RGBX1010102:
> +	case DRM_FORMAT_RGBA1010102:
>  		return OVL_CON_CLRFMT_ARGB8888;
>  	case DRM_FORMAT_BGRX8888:
>  	case DRM_FORMAT_BGRA8888:
> +		return OVL_CON_BYTE_SWAP |
> +		       (blend_mode == DRM_MODE_BLEND_COVERAGE ?
> +		       OVL_CON_CLRFMT_ARGB8888 :
> +		       OVL_CON_CLRFMT_PARGB8888);
> +	case DRM_FORMAT_BGRX1010102:
>  	case DRM_FORMAT_BGRA1010102:
>  		return OVL_CON_CLRFMT_ARGB8888 | OVL_CON_BYTE_SWAP;
>  	case DRM_FORMAT_XRGB8888:
>  	case DRM_FORMAT_ARGB8888:
> +		return blend_mode == DRM_MODE_BLEND_COVERAGE ?
> +		       OVL_CON_CLRFMT_RGBA8888 :
> +		       OVL_CON_CLRFMT_PARGB8888;
> +	case DRM_FORMAT_XRGB2101010:
>  	case DRM_FORMAT_ARGB2101010:
>  		return OVL_CON_CLRFMT_RGBA8888;
>  	case DRM_FORMAT_XBGR8888:
>  	case DRM_FORMAT_ABGR8888:
> +		return OVL_CON_RGB_SWAP |
> +		       (blend_mode == DRM_MODE_BLEND_COVERAGE ?
> +		       OVL_CON_CLRFMT_RGBA8888 :
> +		       OVL_CON_CLRFMT_PARGB8888);
> +	case DRM_FORMAT_XBGR2101010:
> +	case DRM_FORMAT_ABGR2101010:
>  		return OVL_CON_CLRFMT_RGBA8888 | OVL_CON_BYTE_SWAP;
>  	case DRM_FORMAT_UYVY:
>  		return OVL_CON_CLRFMT_UYVY | OVL_CON_MTX_YUV_TO_RGB;
> @@ -404,6 +459,8 @@ void mtk_ovl_layer_config(struct device *dev,
> unsigned int idx,
>  	unsigned int fmt = pending->format;
>  	unsigned int offset = (pending->y << 16) | pending->x;
>  	unsigned int src_size = (pending->height << 16) | pending-
> >width;
> +	unsigned int blend_mode = state->base.pixel_blend_mode;
> +	unsigned int ignore_pixel_alpha = 0;
>  	unsigned int con;
>  	bool is_afbc = pending->modifier != DRM_FORMAT_MOD_LINEAR;
>  	union overlay_pitch {
> @@ -421,9 +478,14 @@ void mtk_ovl_layer_config(struct device *dev,
> unsigned int idx,
>  		return;
>  	}
>  
> -	con = ovl_fmt_convert(ovl, fmt);
> -	if (state->base.fb && state->base.fb->format->has_alpha)
> -		con |= OVL_CON_AEN | OVL_CON_ALPHA;
> +	con = ovl_fmt_convert(ovl, fmt, blend_mode);
> +	if (state->base.fb) {
> +		con |= OVL_CON_AEN;
> +		con |= state->base.alpha & OVL_CON_ALPHA;

Now support pixel alpha, and this seems to support plane alpha. So
separate this to another patch. If possible, describe how plane alpha
calculate with pixel alpha.

> +	}
> +
> +	if (state->base.fb && !state->base.fb->format->has_alpha)
> +		ignore_pixel_alpha = OVL_CONST_BLEND;

Ditto.

Regards,
CK

>  
>  	if (pending->rotation & DRM_MODE_REFLECT_Y) {
>  		con |= OVL_CON_VIRT_FLIP;
> @@ -440,8 +502,8 @@ void mtk_ovl_layer_config(struct device *dev,
> unsigned int idx,
>  
>  	mtk_ddp_write_relaxed(cmdq_pkt, con, &ovl->cmdq_reg, ovl->regs,
>  			      DISP_REG_OVL_CON(idx));
> -	mtk_ddp_write_relaxed(cmdq_pkt, overlay_pitch.split_pitch.lsb,
> &ovl->cmdq_reg, ovl->regs,
> -			      DISP_REG_OVL_PITCH(idx));
> +	mtk_ddp_write_relaxed(cmdq_pkt, overlay_pitch.split_pitch.lsb |
> ignore_pixel_alpha,
> +			      &ovl->cmdq_reg, ovl->regs,
> DISP_REG_OVL_PITCH(idx));
>  	mtk_ddp_write_relaxed(cmdq_pkt, src_size, &ovl->cmdq_reg, ovl-
> >regs,
>  			      DISP_REG_OVL_SRC_SIZE(idx));
>  	mtk_ddp_write_relaxed(cmdq_pkt, offset, &ovl->cmdq_reg, ovl-
> >regs,