Message-ID: <Zx9ebNKW3YdIuA-D@fedora>
Date: Mon, 28 Oct 2024 10:50:36 +0100
From: Louis Chauvet <louis.chauvet@...tlin.com>
To: Maíra Canal <mairacanal@...eup.net>
Cc: Rodrigo Siqueira <rodrigosiqueiramelo@...il.com>,
	Melissa Wen <melissa.srw@...il.com>,
	Haneen Mohammed <hamohammed.sa@...il.com>,
	Daniel Vetter <daniel@...ll.ch>,
	Maarten Lankhorst <maarten.lankhorst@...ux.intel.com>,
	Maxime Ripard <mripard@...nel.org>,
	Thomas Zimmermann <tzimmermann@...e.de>,
	David Airlie <airlied@...il.com>, dri-devel@...ts.freedesktop.org,
	arthurgrillo@...eup.net, linux-kernel@...r.kernel.org,
	jeremie.dautheribes@...tlin.com, miquel.raynal@...tlin.com,
	thomas.petazzoni@...tlin.com, seanpaul@...gle.com,
	nicolejadeyee@...gle.com
Subject: Re: [PATCH v2 1/3] drm/vkms: Re-introduce line-by-line algorithm for
 writeback

On 26/10/24 - 12:05, Maíra Canal wrote:
> Hi Louis,
> 
> On 14/08/24 05:42, Louis Chauvet wrote:
> > Re-introduce a line-by-line writeback algorithm for each pixel format.
> > This allows more performance by not requiring an indirection per pixel
> > write.
> > 
> > Line-by-line writeback was introduced by [1] but rewritten back to
> > pixel-by-pixel algorithm in [2]. At this time, nobody noticed the impact
> > on performance, and it was merged.
> > 
> > This patch is almost a revert of [2], but with some effort to avoid code
> > duplication. Now only the loop is repeated, but it is required to have
> > good performances.
> > 
> > The performance gain is around 5 to 10%.
> > 
> > [1]: https://lore.kernel.org/all/20211005201637.58563-7-igormtorrente@gmail.com/
> > [2]: https://lore.kernel.org/all/20230515135204.115393-4-mcanal@igalia.com/
> > 
> > Signed-off-by: Louis Chauvet <louis.chauvet@...tlin.com>
> > ---
> >   drivers/gpu/drm/vkms/vkms_composer.c  |  17 +++++
> >   drivers/gpu/drm/vkms/vkms_drv.h       |  20 +++---
> >   drivers/gpu/drm/vkms/vkms_formats.c   | 117 +++++++++++++++++++++++++++-------
> >   drivers/gpu/drm/vkms/vkms_formats.h   |   2 +-
> >   drivers/gpu/drm/vkms/vkms_writeback.c |   2 +-
> >   5 files changed, 124 insertions(+), 34 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/vkms/vkms_composer.c b/drivers/gpu/drm/vkms/vkms_composer.c
> > index 76d4aa8a0ef6..f0cae142ac22 100644
> > --- a/drivers/gpu/drm/vkms/vkms_composer.c
> > +++ b/drivers/gpu/drm/vkms/vkms_composer.c
> > @@ -176,6 +176,23 @@ static enum pixel_read_direction direction_for_rotation(unsigned int rotation)
> >   	return READ_LEFT_TO_RIGHT;
> >   }
> > +/**
> > + * Write a line to the writeback buffer > + *
> > + * @wb: Job where to insert the final image
> > + * @src_buffer: Line to write
> > + * @y: Row to write in the writeback buffer
> > + */
> 
> Please, review the documentation using the kernel-doc format.
> 
> > +static void vkms_writeback_row(struct vkms_writeback_job *wb,
> > +			       const struct line_buffer *src_buffer, size_t y_start)
> > +{
> > +	struct vkms_frame_info *frame_info = &wb->wb_frame_info;
> > +	int x_start = frame_info->dst.x1;
> > +	int count = min_t(size_t, drm_rect_width(&frame_info->dst), src_buffer->n_pixels);
> > +
> > +	wb->pixel_write(wb, src_buffer->pixels, count, x_start, y_start);
> > +}
> > +
> >   /**
> >    * clamp_line_coordinates() - Compute and clamp the coordinate to read and write during the blend
> >    * process.
> > diff --git a/drivers/gpu/drm/vkms/vkms_drv.h b/drivers/gpu/drm/vkms/vkms_drv.h
> > index 3870e825da81..526bf5207524 100644
> > --- a/drivers/gpu/drm/vkms/vkms_drv.h
> > +++ b/drivers/gpu/drm/vkms/vkms_drv.h
> > @@ -52,20 +52,25 @@ struct line_buffer {
> >   	struct pixel_argb_u16 *pixels;
> >   };
> > +struct vkms_writeback_job;
> >   /**
> > - * typedef pixel_write_t - These functions are used to read a pixel from a
> > - * &struct pixel_argb_u16, convert it in a specific format and write it in the @dst_pixels
> > - * buffer.
> > + * typedef pixel_write_line_t - These functions are used to read a pixel line from a
> > + * struct pixel_argb_u16 buffer, convert it and write it in the @wb job.
> >    *
> > - * @out_pixel: destination address to write the pixel
> > - * @in_pixel: pixel to write
> > + * @wb: the writeback job to write the output of the conversion
> > + * @in_pixels: Source buffer containing the line to convert
> > + * @count: The width of a line
> > + * @x_start: The x (width) coordinate in the destination plane
> > + * @y_start: The y (height) coordinate in the destination plane
> >    */
> > -typedef void (*pixel_write_t)(u8 *out_pixel, const struct pixel_argb_u16 *in_pixel);
> > +typedef void (*pixel_write_line_t)(struct vkms_writeback_job *wb,
> > +			      struct pixel_argb_u16 *in_pixels, int count, int x_start,
> > +			      int y_start);
> >   struct vkms_writeback_job {
> >   	struct iosys_map data[DRM_FORMAT_MAX_PLANES];
> >   	struct vkms_frame_info wb_frame_info;
> > -	pixel_write_t pixel_write;
> > +	pixel_write_line_t pixel_write;
> >   };
> >   /**
> > @@ -232,7 +237,6 @@ int vkms_verify_crc_source(struct drm_crtc *crtc, const char *source_name,
> >   /* Composer Support */
> >   void vkms_composer_worker(struct work_struct *work);
> >   void vkms_set_composer(struct vkms_output *out, bool enabled);
> > -void vkms_writeback_row(struct vkms_writeback_job *wb, const struct line_buffer *src_buffer, int y);
> >   /* Writeback */
> >   int vkms_enable_writeback_connector(struct vkms_device *vkmsdev);
> > diff --git a/drivers/gpu/drm/vkms/vkms_formats.c b/drivers/gpu/drm/vkms/vkms_formats.c
> > index d1abfb1c3e3c..d1309f6d307f 100644
> > --- a/drivers/gpu/drm/vkms/vkms_formats.c
> > +++ b/drivers/gpu/drm/vkms/vkms_formats.c
> > @@ -587,7 +587,7 @@ static void planar_yuv_read_line(const struct vkms_plane_state *plane, int x_sta
> >    * The following functions take one &struct pixel_argb_u16 and convert it to a specific format.
> >    * The result is stored in @out_pixel.
> >    *
> > - * They are used in vkms_writeback_row() to convert and store a pixel from the src_buffer to
> > + * They are used in the `write_line` functions to convert and store a pixel from the src_buffer to
> >    * the writeback buffer.
> >    */
> >   static void argb_u16_to_ARGB8888(u8 *out_pixel, const struct pixel_argb_u16 *in_pixel)
> > @@ -654,28 +654,97 @@ static void argb_u16_to_RGB565(u8 *out_pixel, const struct pixel_argb_u16 *in_pi
> >   	*pixel = cpu_to_le16(r << 11 | g << 5 | b);
> >   }
> > -/**
> > - * vkms_writeback_row() - Generic loop for all supported writeback format. It is executed just
> > - * after the blending to write a line in the writeback buffer.
> > +/*
> > + * The following functions are write_line function for each pixel format supported by VKMS.
> > + *
> > + * They write a full line at index y. They must read data from the line src_pixels.
> > + *
> > + * The caller must ensure that count is not larger than the framebuffer and the src_pixels.
> > + *
> > + * Those function are very similar, but it is required for performance reason. In the past, some
> > + * experiment were done, and with a generic loop the performance are very reduced [1].
> >    *
> > - * @wb: Job where to insert the final image
> > - * @src_buffer: Line to write
> > - * @y: Row to write in the writeback buffer
> > + * [1]: https://lore.kernel.org/dri-devel/d258c8dc-78e9-4509-9037-a98f7f33b3a3@riseup.net/
> 
> I'm not sure if a link is approriate here.

It was to have some context about the comment. If someone wants to modify 
this in the future (for example, factorise again the code :)), he will 
have some informations.
 
> >    */
> > -void vkms_writeback_row(struct vkms_writeback_job *wb,
> > -			const struct line_buffer *src_buffer, int y)
> > +
> > +static void ARGB8888_write_line(struct vkms_writeback_job *wb,
> > +				struct pixel_argb_u16 *src_pixels, int count, int x_start,
> > +				int y_start)
> >   {
> > -	struct vkms_frame_info *frame_info = &wb->wb_frame_info;
> > -	int x_dst = frame_info->dst.x1;
> >   	u8 *dst_pixels;
> > -	int rem_x, rem_y;
> > -	packed_pixels_addr(frame_info, x_dst, y, 0, &dst_pixels, &rem_x, &rem_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);
> > +	packed_pixels_addr_1x1(&wb->wb_frame_info, x_start, y_start, 0, &dst_pixels);
> > -	for (size_t x = 0; x < x_limit; x++, dst_pixels += frame_info->fb->format->cpp[0])
> > -		wb->pixel_write(dst_pixels, &in_pixels[x]);
> > +	while (count) {
> > +		argb_u16_to_ARGB8888(dst_pixels, src_pixels);
> > +		dst_pixels += wb->wb_frame_info.fb->format->char_per_block[0];
> > +		src_pixels += 1;
> > +		count--;
> > +	}
> > +}
> > +
> > +static void XRGB8888_write_line(struct vkms_writeback_job *wb,
> > +				struct pixel_argb_u16 *src_pixels, int count, int x_start,
> > +				int y_start)
> > +{
> > +	u8 *dst_pixels;
> > +
> > +	packed_pixels_addr_1x1(&wb->wb_frame_info, x_start, y_start, 0, &dst_pixels);
> > +
> > +	while (count) {
> > +		argb_u16_to_XRGB8888(dst_pixels, src_pixels);
> > +		dst_pixels += wb->wb_frame_info.fb->format->char_per_block[0];
> > +		src_pixels += 1;
> > +		count--;
> > +	}
> > +}
> > +
> > +static void ARGB16161616_write_line(struct vkms_writeback_job *wb,
> > +				    struct pixel_argb_u16 *src_pixels, int count, int x_start,
> > +				    int y_start)
> > +{
> > +	u8 *dst_pixels;
> > +
> > +	packed_pixels_addr_1x1(&wb->wb_frame_info, x_start, y_start, 0, &dst_pixels);
> > +
> > +	while (count) {
> > +		argb_u16_to_ARGB16161616(dst_pixels, src_pixels);
> > +		dst_pixels += wb->wb_frame_info.fb->format->char_per_block[0];
> > +		src_pixels += 1;
> > +		count--;
> > +	}
> > +}
> > +
> > +static void XRGB16161616_write_line(struct vkms_writeback_job *wb,
> > +				    struct pixel_argb_u16 *src_pixels, int count, int x_start,
> > +				    int y_start)
> > +{
> > +	u8 *dst_pixels;
> > +
> > +	packed_pixels_addr_1x1(&wb->wb_frame_info, x_start, y_start, 0, &dst_pixels);
> > +
> > +	while (count) {
> > +		argb_u16_to_XRGB16161616(dst_pixels, src_pixels);
> > +		dst_pixels += wb->wb_frame_info.fb->format->char_per_block[0];
> > +		src_pixels += 1;
> > +		count--;
> > +	}
> > +}
> > +
> > +static void RGB565_write_line(struct vkms_writeback_job *wb,
> > +			      struct pixel_argb_u16 *src_pixels, int count, int x_start,
> > +			      int y_start)
> > +{
> > +	u8 *dst_pixels;
> > +
> > +	packed_pixels_addr_1x1(&wb->wb_frame_info, x_start, y_start, 0, &dst_pixels);
> > +
> > +	while (count) {
> > +		argb_u16_to_RGB565(dst_pixels, src_pixels);
> > +		dst_pixels += wb->wb_frame_info.fb->format->char_per_block[0];
> > +		src_pixels += 1;
> > +		count--;
> > +	}
> >   }
> >   /**
> > @@ -936,25 +1005,25 @@ void get_conversion_matrix_to_argb_u16(u32 format,
> >   }
> >   /**
> > - * get_pixel_write_function() - Retrieve the correct write_pixel function for a specific format.
> > + * get_pixel_write_function() - Retrieve the correct write_line function for a specific format.
> 
> Correct the docs.

Nice catch!

Thanks,
Louis Chauvet

> Best Regards,
> - Maíra
> 
> >    * The returned pointer is NULL for unsupported pixel formats. The caller must ensure that the
> >    * pointer is valid before using it in a vkms_writeback_job.
> >    *
> >    * @format: DRM_FORMAT_* value for which to obtain a conversion function (see [drm_fourcc.h])
> >    */
> > -pixel_write_t get_pixel_write_function(u32 format)
> > +pixel_write_line_t get_pixel_write_line_function(u32 format) >   {
> >   	switch (format) {
> >   	case DRM_FORMAT_ARGB8888:
> > -		return &argb_u16_to_ARGB8888;
> > +		return &ARGB8888_write_line;
> >   	case DRM_FORMAT_XRGB8888:
> > -		return &argb_u16_to_XRGB8888;
> > +		return &XRGB8888_write_line;
> >   	case DRM_FORMAT_ARGB16161616:
> > -		return &argb_u16_to_ARGB16161616;
> > +		return &ARGB16161616_write_line;
> >   	case DRM_FORMAT_XRGB16161616:
> > -		return &argb_u16_to_XRGB16161616;
> > +		return &XRGB16161616_write_line;
> >   	case DRM_FORMAT_RGB565:
> > -		return &argb_u16_to_RGB565;
> > +		return &RGB565_write_line;
> >   	default:
> >   		/*
> >   		 * This is a bug in vkms_writeback_atomic_check. All the supported
> > diff --git a/drivers/gpu/drm/vkms/vkms_formats.h b/drivers/gpu/drm/vkms/vkms_formats.h
> > index eeb208cdd6b1..852ab9a4cee5 100644
> > --- a/drivers/gpu/drm/vkms/vkms_formats.h
> > +++ b/drivers/gpu/drm/vkms/vkms_formats.h
> > @@ -7,7 +7,7 @@
> >   pixel_read_line_t get_pixel_read_line_function(u32 format);
> > -pixel_write_t get_pixel_write_function(u32 format);
> > +pixel_write_line_t get_pixel_write_line_function(u32 format);
> >   void get_conversion_matrix_to_argb_u16(u32 format, enum drm_color_encoding encoding,
> >   				       enum drm_color_range range,
> > diff --git a/drivers/gpu/drm/vkms/vkms_writeback.c b/drivers/gpu/drm/vkms/vkms_writeback.c
> > index c8582df1f739..f6ed3aa69af8 100644
> > --- a/drivers/gpu/drm/vkms/vkms_writeback.c
> > +++ b/drivers/gpu/drm/vkms/vkms_writeback.c
> > @@ -150,7 +150,7 @@ static void vkms_wb_atomic_commit(struct drm_connector *conn,
> >   	crtc_state->wb_pending = true;
> >   	spin_unlock_irq(&output->composer_lock);
> >   	drm_writeback_queue_job(wb_conn, connector_state);
> > -	active_wb->pixel_write = get_pixel_write_function(wb_format);
> > +	active_wb->pixel_write = get_pixel_write_line_function(wb_format);
> >   	drm_rect_init(&wb_frame_info->src, 0, 0, crtc_width, crtc_height);
> >   	drm_rect_init(&wb_frame_info->dst, 0, 0, crtc_width, crtc_height);
> >   }
> > 

Powered by blists - more mailing lists