linux_dsm_epyc7002/drivers/gpu/drm/udl/udl_transfer.c
Haixia Shi 865844448c drm/udl: optimize udl_compress_hline16 (v2)
The run-length encoding algorithm should compare 16-bit encoded pixel
values instead of comparing raw pixel values. It allows pixels
with similar but different colors to be encoded as repeat pixels, and
thus potentially save USB bandwidth.

Signed-off-by: Haixia Shi <hshi@chromium.org>
Reviewed-by: Daniel Kurtz <djkurtz@chromium.org>
Tested-by: Haixia Shi <hshi@chromium.org>
Signed-off-by: Dave Airlie <airlied@gmail.com>
2015-01-31 10:08:47 +10:00

264 lines
7.6 KiB
C

/*
* Copyright (C) 2012 Red Hat
* based in parts on udlfb.c:
* Copyright (C) 2009 Roberto De Ioris <roberto@unbit.it>
* Copyright (C) 2009 Jaya Kumar <jayakumar.lkml@gmail.com>
* Copyright (C) 2009 Bernie Thompson <bernie@plugable.com>
*
* This file is subject to the terms and conditions of the GNU General Public
* License v2. See the file COPYING in the main directory of this archive for
* more details.
*/
#include <linux/module.h>
#include <linux/slab.h>
#include <linux/fb.h>
#include <linux/prefetch.h>
#include <drm/drmP.h>
#include "udl_drv.h"
#define MAX_CMD_PIXELS 255
#define RLX_HEADER_BYTES 7
#define MIN_RLX_PIX_BYTES 4
#define MIN_RLX_CMD_BYTES (RLX_HEADER_BYTES + MIN_RLX_PIX_BYTES)
#define RLE_HEADER_BYTES 6
#define MIN_RLE_PIX_BYTES 3
#define MIN_RLE_CMD_BYTES (RLE_HEADER_BYTES + MIN_RLE_PIX_BYTES)
#define RAW_HEADER_BYTES 6
#define MIN_RAW_PIX_BYTES 2
#define MIN_RAW_CMD_BYTES (RAW_HEADER_BYTES + MIN_RAW_PIX_BYTES)
/*
* Trims identical data from front and back of line
* Sets new front buffer address and width
* And returns byte count of identical pixels
* Assumes CPU natural alignment (unsigned long)
* for back and front buffer ptrs and width
*/
#if 0
static int udl_trim_hline(const u8 *bback, const u8 **bfront, int *width_bytes)
{
int j, k;
const unsigned long *back = (const unsigned long *) bback;
const unsigned long *front = (const unsigned long *) *bfront;
const int width = *width_bytes / sizeof(unsigned long);
int identical = width;
int start = width;
int end = width;
prefetch((void *) front);
prefetch((void *) back);
for (j = 0; j < width; j++) {
if (back[j] != front[j]) {
start = j;
break;
}
}
for (k = width - 1; k > j; k--) {
if (back[k] != front[k]) {
end = k+1;
break;
}
}
identical = start + (width - end);
*bfront = (u8 *) &front[start];
*width_bytes = (end - start) * sizeof(unsigned long);
return identical * sizeof(unsigned long);
}
#endif
static inline u16 pixel32_to_be16(const uint32_t pixel)
{
return (((pixel >> 3) & 0x001f) |
((pixel >> 5) & 0x07e0) |
((pixel >> 8) & 0xf800));
}
static inline u16 get_pixel_val16(const uint8_t *pixel, int bpp)
{
u16 pixel_val16 = 0;
if (bpp == 2)
pixel_val16 = *(const uint16_t *)pixel;
else if (bpp == 4)
pixel_val16 = pixel32_to_be16(*(const uint32_t *)pixel);
return pixel_val16;
}
/*
* Render a command stream for an encoded horizontal line segment of pixels.
*
* A command buffer holds several commands.
* It always begins with a fresh command header
* (the protocol doesn't require this, but we enforce it to allow
* multiple buffers to be potentially encoded and sent in parallel).
* A single command encodes one contiguous horizontal line of pixels
*
* The function relies on the client to do all allocation, so that
* rendering can be done directly to output buffers (e.g. USB URBs).
* The function fills the supplied command buffer, providing information
* on where it left off, so the client may call in again with additional
* buffers if the line will take several buffers to complete.
*
* A single command can transmit a maximum of 256 pixels,
* regardless of the compression ratio (protocol design limit).
* To the hardware, 0 for a size byte means 256
*
* Rather than 256 pixel commands which are either rl or raw encoded,
* the rlx command simply assumes alternating raw and rl spans within one cmd.
* This has a slightly larger header overhead, but produces more even results.
* It also processes all data (read and write) in a single pass.
* Performance benchmarks of common cases show it having just slightly better
* compression than 256 pixel raw or rle commands, with similar CPU consumpion.
* But for very rl friendly data, will compress not quite as well.
*/
static void udl_compress_hline16(
const u8 **pixel_start_ptr,
const u8 *const pixel_end,
uint32_t *device_address_ptr,
uint8_t **command_buffer_ptr,
const uint8_t *const cmd_buffer_end, int bpp)
{
const u8 *pixel = *pixel_start_ptr;
uint32_t dev_addr = *device_address_ptr;
uint8_t *cmd = *command_buffer_ptr;
while ((pixel_end > pixel) &&
(cmd_buffer_end - MIN_RLX_CMD_BYTES > cmd)) {
uint8_t *raw_pixels_count_byte = NULL;
uint8_t *cmd_pixels_count_byte = NULL;
const u8 *raw_pixel_start = NULL;
const u8 *cmd_pixel_start, *cmd_pixel_end = NULL;
uint16_t pixel_val16;
prefetchw((void *) cmd); /* pull in one cache line at least */
*cmd++ = 0xaf;
*cmd++ = 0x6b;
*cmd++ = (uint8_t) ((dev_addr >> 16) & 0xFF);
*cmd++ = (uint8_t) ((dev_addr >> 8) & 0xFF);
*cmd++ = (uint8_t) ((dev_addr) & 0xFF);
cmd_pixels_count_byte = cmd++; /* we'll know this later */
cmd_pixel_start = pixel;
raw_pixels_count_byte = cmd++; /* we'll know this later */
raw_pixel_start = pixel;
cmd_pixel_end = pixel + (min(MAX_CMD_PIXELS + 1,
min((int)(pixel_end - pixel) / bpp,
(int)(cmd_buffer_end - cmd) / 2))) * bpp;
prefetch_range((void *) pixel, (cmd_pixel_end - pixel) * bpp);
pixel_val16 = get_pixel_val16(pixel, bpp);
while (pixel < cmd_pixel_end) {
const u8 *const start = pixel;
const uint16_t repeating_pixel_val16 = pixel_val16;
*(uint16_t *)cmd = cpu_to_be16(pixel_val16);
cmd += 2;
pixel += bpp;
while (pixel < cmd_pixel_end) {
pixel_val16 = get_pixel_val16(pixel, bpp);
if (pixel_val16 != repeating_pixel_val16)
break;
pixel += bpp;
}
if (unlikely(pixel > start + bpp)) {
/* go back and fill in raw pixel count */
*raw_pixels_count_byte = (((start -
raw_pixel_start) / bpp) + 1) & 0xFF;
/* immediately after raw data is repeat byte */
*cmd++ = (((pixel - start) / bpp) - 1) & 0xFF;
/* Then start another raw pixel span */
raw_pixel_start = pixel;
raw_pixels_count_byte = cmd++;
}
}
if (pixel > raw_pixel_start) {
/* finalize last RAW span */
*raw_pixels_count_byte = ((pixel-raw_pixel_start) / bpp) & 0xFF;
}
*cmd_pixels_count_byte = ((pixel - cmd_pixel_start) / bpp) & 0xFF;
dev_addr += ((pixel - cmd_pixel_start) / bpp) * 2;
}
if (cmd_buffer_end <= MIN_RLX_CMD_BYTES + cmd) {
/* Fill leftover bytes with no-ops */
if (cmd_buffer_end > cmd)
memset(cmd, 0xAF, cmd_buffer_end - cmd);
cmd = (uint8_t *) cmd_buffer_end;
}
*command_buffer_ptr = cmd;
*pixel_start_ptr = pixel;
*device_address_ptr = dev_addr;
return;
}
/*
* There are 3 copies of every pixel: The front buffer that the fbdev
* client renders to, the actual framebuffer across the USB bus in hardware
* (that we can only write to, slowly, and can never read), and (optionally)
* our shadow copy that tracks what's been sent to that hardware buffer.
*/
int udl_render_hline(struct drm_device *dev, int bpp, struct urb **urb_ptr,
const char *front, char **urb_buf_ptr,
u32 byte_offset, u32 device_byte_offset,
u32 byte_width,
int *ident_ptr, int *sent_ptr)
{
const u8 *line_start, *line_end, *next_pixel;
u32 base16 = 0 + (device_byte_offset / bpp) * 2;
struct urb *urb = *urb_ptr;
u8 *cmd = *urb_buf_ptr;
u8 *cmd_end = (u8 *) urb->transfer_buffer + urb->transfer_buffer_length;
BUG_ON(!(bpp == 2 || bpp == 4));
line_start = (u8 *) (front + byte_offset);
next_pixel = line_start;
line_end = next_pixel + byte_width;
while (next_pixel < line_end) {
udl_compress_hline16(&next_pixel,
line_end, &base16,
(u8 **) &cmd, (u8 *) cmd_end, bpp);
if (cmd >= cmd_end) {
int len = cmd - (u8 *) urb->transfer_buffer;
if (udl_submit_urb(dev, urb, len))
return 1; /* lost pixels is set */
*sent_ptr += len;
urb = udl_get_urb(dev);
if (!urb)
return 1; /* lost_pixels is set */
*urb_ptr = urb;
cmd = urb->transfer_buffer;
cmd_end = &cmd[urb->transfer_buffer_length];
}
}
*urb_buf_ptr = cmd;
return 0;
}