guacamole-spice-protocol/src/common/surface.c

2010 lines
60 KiB
C

/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing,
* software distributed under the License is distributed on an
* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
* KIND, either express or implied. See the License for the
* specific language governing permissions and limitations
* under the License.
*/
#include "config.h"
#include "common/rect.h"
#include "common/surface.h"
#include <cairo/cairo.h>
#include <guacamole/client.h>
#include <guacamole/layer.h>
#include <guacamole/protocol.h>
#include <guacamole/socket.h>
#include <guacamole/timestamp.h>
#include <guacamole/user.h>
#include <pthread.h>
#include <stdlib.h>
#include <stdint.h>
#include <string.h>
/**
* The width of an update which should be considered negible and thus
* trivial overhead compared ot the cost of two updates.
*/
#define GUAC_SURFACE_NEGLIGIBLE_WIDTH 64
/**
* The height of an update which should be considered negible and thus
* trivial overhead compared ot the cost of two updates.
*/
#define GUAC_SURFACE_NEGLIGIBLE_HEIGHT 64
/**
* The proportional increase in cost contributed by transfer and processing of
* image data, compared to processing an equivalent amount of client-side
* data.
*/
#define GUAC_SURFACE_DATA_FACTOR 16
/**
* The base cost of every update. Each update should be considered to have
* this starting cost, plus any additional cost estimated from its
* content.
*/
#define GUAC_SURFACE_BASE_COST 4096
/**
* An increase in cost is negligible if it is less than
* 1/GUAC_SURFACE_NEGLIGIBLE_INCREASE of the old cost.
*/
#define GUAC_SURFACE_NEGLIGIBLE_INCREASE 4
/**
* If combining an update because it appears to be follow a fill pattern,
* the combined cost must not exceed
* GUAC_SURFACE_FILL_PATTERN_FACTOR * (total uncombined cost).
*/
#define GUAC_SURFACE_FILL_PATTERN_FACTOR 3
/* Define cairo_format_stride_for_width() if missing */
#ifndef HAVE_CAIRO_FORMAT_STRIDE_FOR_WIDTH
#define cairo_format_stride_for_width(format, width) (width*4)
#endif
/**
* The framerate which, if exceeded, indicates that JPEG is preferred.
*/
#define GUAC_COMMON_SURFACE_JPEG_FRAMERATE 3
/**
* Minimum JPEG bitmap size (area). If the bitmap is smaller than this threshold,
* it should be compressed as a PNG image to avoid the JPEG compression tax.
*/
#define GUAC_SURFACE_JPEG_MIN_BITMAP_SIZE 4096
/**
* The JPEG compression min block size. This defines the optimal rectangle block
* size factor for JPEG compression. Usually 8x8 would suffice, but use 16 to
* reduce the occurrence of ringing artifacts further.
*/
#define GUAC_SURFACE_JPEG_BLOCK_SIZE 16
/**
* The WebP compression min block size. This defines the optimal rectangle block
* size factor for WebP compression. WebP does utilize variable block size, but
* ensuring a block size factor reduces any noise on the image edges.
*/
#define GUAC_SURFACE_WEBP_BLOCK_SIZE 8
void guac_common_surface_move(guac_common_surface* surface, int x, int y) {
pthread_mutex_lock(&surface->_lock);
surface->x = x;
surface->y = y;
surface->location_dirty = 1;
pthread_mutex_unlock(&surface->_lock);
}
void guac_common_surface_stack(guac_common_surface* surface, int z) {
pthread_mutex_lock(&surface->_lock);
surface->z = z;
surface->location_dirty = 1;
pthread_mutex_unlock(&surface->_lock);
}
void guac_common_surface_set_parent(guac_common_surface* surface,
const guac_layer* parent) {
pthread_mutex_lock(&surface->_lock);
surface->parent = parent;
surface->location_dirty = 1;
pthread_mutex_unlock(&surface->_lock);
}
void guac_common_surface_set_opacity(guac_common_surface* surface,
int opacity) {
pthread_mutex_lock(&surface->_lock);
surface->opacity = opacity;
surface->opacity_dirty = 1;
pthread_mutex_unlock(&surface->_lock);
}
/**
* Updates the coordinates of the given rectangle to be within the bounds of
* the given surface.
*
* @param surface The surface to use for clipping.
* @param rect The rectangle to clip.
* @param sx The X coordinate of the source rectangle, if any.
* @param sy The Y coordinate of the source rectangle, if any.
*/
static void __guac_common_bound_rect(guac_common_surface* surface,
guac_common_rect* rect, int* sx, int* sy) {
guac_common_rect bounds_rect = {
.x = 0,
.y = 0,
.width = surface->width,
.height = surface->height
};
int orig_x = rect->x;
int orig_y = rect->y;
guac_common_rect_constrain(rect, &bounds_rect);
/* Update source X/Y if given */
if (sx != NULL) *sx += rect->x - orig_x;
if (sy != NULL) *sy += rect->y - orig_y;
}
/**
* Updates the coordinates of the given rectangle to be within the clipping
* rectangle of the given surface, which must always be within the bounding
* rectangle of the given surface.
*
* @param surface The surface to use for clipping.
* @param rect The rectangle to clip.
* @param sx The X coordinate of the source rectangle, if any.
* @param sy The Y coordinate of the source rectangle, if any.
*/
static void __guac_common_clip_rect(guac_common_surface* surface,
guac_common_rect* rect, int* sx, int* sy) {
int orig_x = rect->x;
int orig_y = rect->y;
/* Just bound within surface if no clipping rectangle applied */
if (!surface->clipped) {
__guac_common_bound_rect(surface, rect, sx, sy);
return;
}
guac_common_rect_constrain(rect, &surface->clip_rect);
/* Update source X/Y if given */
if (sx != NULL) *sx += rect->x - orig_x;
if (sy != NULL) *sy += rect->y - orig_y;
}
/**
* Returns whether a rectangle within the given surface contains only fully
* opaque pixels.
*
* @param surface
* The surface to check.
*
* @param rect
* The rectangle to check.
*
* @return
* Non-zero if the rectangle contains only fully opaque pixels, zero
* otherwise.
*/
static int __guac_common_surface_is_opaque(guac_common_surface* surface,
guac_common_rect* rect) {
int x, y;
int stride = surface ->stride;
unsigned char* buffer =
surface->buffer + (stride * rect->y) + (4 * rect->x);
/* For each row */
for (y = 0; y < rect->height; y++) {
/* Search for a non-opaque pixel */
uint32_t* current = (uint32_t*) buffer;
for (x=0; x < rect->width; x++) {
/* Rectangle is non-opaque if a single non-opaque pixel is found */
uint32_t color = *(current++);
if ((color & 0xFF000000) != 0xFF000000)
return 0;
}
/* Next row */
buffer += stride;
}
/* Rectangle is opaque */
return 1;
}
/**
* Returns whether the given rectangle should be combined into the existing
* dirty rectangle, to be eventually flushed as image data, or would be best
* kept independent of the current rectangle.
*
* @param surface
* The surface being updated.
*
* @param rect
* The bounding rectangle of the updating being made to the surface.
*
* @param rect_only
* Non-zero if this update, by its nature, contains only metainformation
* about the update's bounding rectangle, zero if the update also contains
* image data.
*
* @return
* Non-zero if the update should be combined with any existing update, zero
* otherwise.
*/
static int __guac_common_should_combine(guac_common_surface* surface, const guac_common_rect* rect, int rect_only) {
/* Always favor combining updates if surface is currently a purely
* server-side scratch area */
if (!surface->realized)
return 1;
if (surface->dirty) {
int combined_cost, dirty_cost, update_cost;
/* Simulate combination */
guac_common_rect combined = surface->dirty_rect;
guac_common_rect_extend(&combined, rect);
/* Combine if result is still small */
if (combined.width <= GUAC_SURFACE_NEGLIGIBLE_WIDTH && combined.height <= GUAC_SURFACE_NEGLIGIBLE_HEIGHT)
return 1;
/* Estimate costs of the existing update, new update, and both combined */
combined_cost = GUAC_SURFACE_BASE_COST + combined.width * combined.height;
dirty_cost = GUAC_SURFACE_BASE_COST + surface->dirty_rect.width * surface->dirty_rect.height;
update_cost = GUAC_SURFACE_BASE_COST + rect->width * rect->height;
/* Reduce cost if no image data */
if (rect_only)
update_cost /= GUAC_SURFACE_DATA_FACTOR;
/* Combine if cost estimate shows benefit */
if (combined_cost <= update_cost + dirty_cost)
return 1;
/* Combine if increase in cost is negligible */
if (combined_cost - dirty_cost <= dirty_cost / GUAC_SURFACE_NEGLIGIBLE_INCREASE)
return 1;
if (combined_cost - update_cost <= update_cost / GUAC_SURFACE_NEGLIGIBLE_INCREASE)
return 1;
/* Combine if we anticipate further updates, as this update follows a common fill pattern */
if (rect->x == surface->dirty_rect.x && rect->y == surface->dirty_rect.y + surface->dirty_rect.height) {
if (combined_cost <= (dirty_cost + update_cost) * GUAC_SURFACE_FILL_PATTERN_FACTOR)
return 1;
}
}
/* Otherwise, do not combine */
return 0;
}
/**
* Expands the dirty rect of the given surface to contain the rect described by the given
* coordinates.
*
* @param surface The surface to mark as dirty.
* @param rect The rectangle of the update which is dirtying the surface.
*/
static void __guac_common_mark_dirty(guac_common_surface* surface, const guac_common_rect* rect) {
/* Ignore empty rects */
if (rect->width <= 0 || rect->height <= 0)
return;
/* If already dirty, update existing rect */
if (surface->dirty)
guac_common_rect_extend(&surface->dirty_rect, rect);
/* Otherwise init dirty rect */
else {
surface->dirty_rect = *rect;
surface->dirty = 1;
}
}
/**
* Calculate the current average framerate for a given area on the surface.
*
* @param surface
* The surface on which the framerate will be calculated.
*
* @param rect
* The rect containing the area for which the average framerate will be
* calculated.
*
* @return
* The average framerate of the given area, in frames per second.
*/
static unsigned int __guac_common_surface_calculate_framerate(
guac_common_surface* surface, const guac_common_rect* rect) {
int x, y;
/* Calculate heat map dimensions */
int heat_width = GUAC_COMMON_SURFACE_HEAT_DIMENSION(surface->width);
/* Calculate minimum X/Y coordinates intersecting given rect */
int min_x = rect->x / GUAC_COMMON_SURFACE_HEAT_CELL_SIZE;
int min_y = rect->y / GUAC_COMMON_SURFACE_HEAT_CELL_SIZE;
/* Calculate maximum X/Y coordinates intersecting given rect */
int max_x = min_x + (rect->width - 1) / GUAC_COMMON_SURFACE_HEAT_CELL_SIZE;
int max_y = min_y + (rect->height - 1) / GUAC_COMMON_SURFACE_HEAT_CELL_SIZE;
unsigned int sum_framerate = 0;
unsigned int count = 0;
/* Get start of buffer at given coordinates */
const guac_common_surface_heat_cell* heat_row =
surface->heat_map + min_y * heat_width + min_x;
/* Iterate over all the heat map cells for the area
* and calculate the average framerate */
for (y = min_y; y < max_y; y++) {
/* Get current row of heat map */
const guac_common_surface_heat_cell* heat_cell = heat_row;
/* For each cell in subset of row */
for (x = min_x; x < max_x; x++) {
/* Calculate indicies for latest and oldest history entries */
int oldest_entry = heat_cell->oldest_entry;
int latest_entry = oldest_entry - 1;
if (latest_entry < 0)
latest_entry = GUAC_COMMON_SURFACE_HEAT_CELL_HISTORY_SIZE - 1;
/* Calculate elapsed time covering entire history for this cell */
int elapsed_time = heat_cell->history[latest_entry]
- heat_cell->history[oldest_entry];
/* Calculate and add framerate */
if (elapsed_time)
sum_framerate += GUAC_COMMON_SURFACE_HEAT_CELL_HISTORY_SIZE
* 1000 / elapsed_time;
/* Next heat map cell */
heat_cell++;
count++;
}
/* Next heat map row */
heat_row += heat_width;
}
/* Calculate the average framerate over entire rect */
if (count)
return sum_framerate / count;
return 0;
}
/**
* Guesses whether a rectangle within a particular surface would be better
* compressed as PNG or using a lossy format like JPEG. Positive values
* indicate PNG is likely to be superior, while negative values indicate the
* opposite.
*
* @param surface
* The surface containing the image data to check.
*
* @param rect
* The rect to check within the given surface.
*
* @return
* Positive values if PNG compression is likely to perform better than
* lossy alternatives, or negative values if PNG is likely to perform
* worse.
*/
static int __guac_common_surface_png_optimality(guac_common_surface* surface,
const guac_common_rect* rect) {
int x, y;
int num_same = 0;
int num_different = 1;
/* Get image/buffer metrics */
int width = rect->width;
int height = rect->height;
int stride = surface->stride;
/* Get buffer from surface */
unsigned char* buffer = surface->buffer + rect->y * stride + rect->x * 4;
/* Image must be at least 1x1 */
if (width < 1 || height < 1)
return 0;
/* For each row */
for (y = 0; y < height; y++) {
uint32_t* row = (uint32_t*) buffer;
uint32_t last_pixel = *(row++) | 0xFF000000;
/* For each pixel in current row */
for (x = 1; x < width; x++) {
/* Get next pixel */
uint32_t current_pixel = *(row++) | 0xFF000000;
/* Update same/different counts according to pixel value */
if (current_pixel == last_pixel)
num_same++;
else
num_different++;
last_pixel = current_pixel;
}
/* Advance to next row */
buffer += stride;
}
/* Return rough approximation of optimality for PNG compression */
return 0x100 * num_same / num_different - 0x400;
}
/**
* Returns whether the given rectangle would be optimally encoded as JPEG
* rather than PNG.
*
* @param surface
* The surface to be queried.
*
* @param rect
* The rectangle to check.
*
* @return
* Non-zero if the rectangle would be optimally encoded as JPEG, zero
* otherwise.
*/
static int __guac_common_surface_should_use_jpeg(guac_common_surface* surface,
const guac_common_rect* rect) {
/* Calculate the average framerate for the given rect */
int framerate = __guac_common_surface_calculate_framerate(surface, rect);
int rect_size = rect->width * rect->height;
/* JPEG is preferred if:
* - frame rate is high enough
* - image size is large enough
* - PNG is not more optimal based on image contents */
return framerate >= GUAC_COMMON_SURFACE_JPEG_FRAMERATE
&& rect_size > GUAC_SURFACE_JPEG_MIN_BITMAP_SIZE
&& __guac_common_surface_png_optimality(surface, rect) < 0;
}
/**
* Returns whether the given rectangle would be optimally encoded as WebP
* rather than PNG.
*
* @param surface
* The surface to be queried.
*
* @param rect
* The rectangle to check.
*
* @return
* Non-zero if the rectangle would be optimally encoded as WebP, zero
* otherwise.
*/
static int __guac_common_surface_should_use_webp(guac_common_surface* surface,
const guac_common_rect* rect) {
/* Do not use WebP if not supported */
if (!guac_client_supports_webp(surface->client))
return 0;
/* Calculate the average framerate for the given rect */
int framerate = __guac_common_surface_calculate_framerate(surface, rect);
/* WebP is preferred if:
* - frame rate is high enough
* - PNG is not more optimal based on image contents */
return framerate >= GUAC_COMMON_SURFACE_JPEG_FRAMERATE
&& __guac_common_surface_png_optimality(surface, rect) < 0;
}
/**
* Updates the heat map cells which intersect the given rectangle using the
* given timestamp. This timestamp, along with timestamps from past updates,
* is used to calculate the framerate of each heat cell.
*
* @param surface
* The surface containing the heat map cells to be updated.
*
* @param rect
* The rectangle containing the heat map cells to be updated.
*
* @param time
* The timestamp to use when updating the heat map cells which intersect
* the given rectangle.
*/
static void __guac_common_surface_touch_rect(guac_common_surface* surface,
guac_common_rect* rect, guac_timestamp time) {
int x, y;
/* Calculate heat map dimensions */
int heat_width = GUAC_COMMON_SURFACE_HEAT_DIMENSION(surface->width);
/* Calculate minimum X/Y coordinates intersecting given rect */
int min_x = rect->x / GUAC_COMMON_SURFACE_HEAT_CELL_SIZE;
int min_y = rect->y / GUAC_COMMON_SURFACE_HEAT_CELL_SIZE;
/* Calculate maximum X/Y coordinates intersecting given rect */
int max_x = min_x + (rect->width - 1) / GUAC_COMMON_SURFACE_HEAT_CELL_SIZE;
int max_y = min_y + (rect->height - 1) / GUAC_COMMON_SURFACE_HEAT_CELL_SIZE;
/* Get start of buffer at given coordinates */
guac_common_surface_heat_cell* heat_row =
surface->heat_map + min_y * heat_width + min_x;
/* Update all heat map cells which intersect with rectangle */
for (y = min_y; y <= max_y; y++) {
/* Get current row of heat map */
guac_common_surface_heat_cell* heat_cell = heat_row;
/* For each cell in subset of row */
for (x = min_x; x <= max_x; x++) {
/* Replace oldest entry with new timestamp */
heat_cell->history[heat_cell->oldest_entry] = time;
/* Update to next oldest entry */
heat_cell->oldest_entry++;
if (heat_cell->oldest_entry >=
GUAC_COMMON_SURFACE_HEAT_CELL_HISTORY_SIZE)
heat_cell->oldest_entry = 0;
/* Advance to next heat map cell */
heat_cell++;
}
/* Next heat map row */
heat_row += heat_width;
}
}
/**
* Flushes the bitmap update currently described by the dirty rectangle within the
* given surface to that surface's bitmap queue. There MUST be space within the
* queue.
*
* @param surface The surface to flush.
*/
static void __guac_common_surface_flush_to_queue(guac_common_surface* surface) {
guac_common_surface_bitmap_rect* rect;
/* Do not flush if not dirty */
if (!surface->dirty)
return;
/* Add new rect to queue */
rect = &(surface->bitmap_queue[surface->bitmap_queue_length++]);
rect->rect = surface->dirty_rect;
rect->flushed = 0;
/* Surface now flushed */
surface->dirty = 0;
}
/**
* Flushes the given surface, drawing any pending operations on the remote
* display. Surface properties are not flushed.
*
* @param surface
* The surface to flush.
*/
static void __guac_common_surface_flush(guac_common_surface* surface);
/**
* Schedules a deferred flush of the given surface. This will not immediately
* flush the surface to the client. Instead, the result of the flush is
* added to a queue which is reinspected and combined (if possible) with other
* deferred flushes during the call to guac_common_surface_flush().
*
* @param surface The surface to flush.
*/
static void __guac_common_surface_flush_deferred(guac_common_surface* surface) {
/* Do not flush if not dirty */
if (!surface->dirty)
return;
/* Flush if queue size has reached maximum (space is reserved for the final
* dirty rect, as __guac_common_surface_flush() MAY add an additional rect
* to the queue */
if (surface->bitmap_queue_length == GUAC_COMMON_SURFACE_QUEUE_SIZE-1)
__guac_common_surface_flush(surface);
/* Append dirty rect to queue */
__guac_common_surface_flush_to_queue(surface);
}
/**
* Transfers a single uint32_t using the given transfer function.
*
* @param op The transfer function to use.
* @param src The source of the uint32_t value.
* @param dst THe destination which will hold the result of the transfer.
* @return Non-zero if the destination value was changed, zero otherwise.
*/
static int __guac_common_surface_transfer_int(guac_transfer_function op, uint32_t* src, uint32_t* dst) {
uint32_t orig = *dst;
switch (op) {
case GUAC_TRANSFER_BINARY_BLACK:
*dst = 0xFF000000;
break;
case GUAC_TRANSFER_BINARY_WHITE:
*dst = 0xFFFFFFFF;
break;
case GUAC_TRANSFER_BINARY_SRC:
*dst = *src;
break;
case GUAC_TRANSFER_BINARY_DEST:
/* NOP */
break;
case GUAC_TRANSFER_BINARY_NSRC:
*dst = *src ^ 0x00FFFFFF;
break;
case GUAC_TRANSFER_BINARY_NDEST:
*dst = *dst ^ 0x00FFFFFF;
break;
case GUAC_TRANSFER_BINARY_AND:
*dst = ((*dst) & (0xFF000000 | *src));
break;
case GUAC_TRANSFER_BINARY_NAND:
*dst = ((*dst) & (0xFF000000 | *src)) ^ 0x00FFFFFF;
break;
case GUAC_TRANSFER_BINARY_OR:
*dst = ((*dst) | (0x00FFFFFF & *src));
break;
case GUAC_TRANSFER_BINARY_NOR:
*dst = ((*dst) | (0x00FFFFFF & *src)) ^ 0x00FFFFFF;
break;
case GUAC_TRANSFER_BINARY_XOR:
*dst = ((*dst) ^ (0x00FFFFFF & *src));
break;
case GUAC_TRANSFER_BINARY_XNOR:
*dst = ((*dst) ^ (0x00FFFFFF & *src)) ^ 0x00FFFFFF;
break;
case GUAC_TRANSFER_BINARY_NSRC_AND:
*dst = ((*dst) & (0xFF000000 | (*src ^ 0x00FFFFFF)));
break;
case GUAC_TRANSFER_BINARY_NSRC_NAND:
*dst = ((*dst) & (0xFF000000 | (*src ^ 0x00FFFFFF))) ^ 0x00FFFFFF;
break;
case GUAC_TRANSFER_BINARY_NSRC_OR:
*dst = ((*dst) | (0x00FFFFFF & (*src ^ 0x00FFFFFF)));
break;
case GUAC_TRANSFER_BINARY_NSRC_NOR:
*dst = ((*dst) | (0x00FFFFFF & (*src ^ 0x00FFFFFF))) ^ 0x00FFFFFF;
break;
}
return *dst != orig;
}
/**
* Assigns the given value to all pixels within a rectangle of the backing
* surface of the given destination surface. The color of all pixels within the
* rectangle, including the alpha component, is entirely replaced.
*
* @param dst
* The destination surface.
*
* @param rect
* The rectangle to draw.
*
* @param red
* The red component of the color value to assign to all pixels within the
* rectangle.
*
* @param green
* The green component of the color value to assign to all pixels within
* the rectangle.
*
* @param blue
* The blue component of the color value to assign to all pixels within the
* rectangle.
*
* @param alpha
* The alpha component of the color value to assign to all pixels within
* the rectangle.
*/
static void __guac_common_surface_set(guac_common_surface* dst,
guac_common_rect* rect, int red, int green, int blue, int alpha) {
int x, y;
int dst_stride;
unsigned char* dst_buffer;
uint32_t color = (alpha << 24) | (red << 16) | (green << 8) | blue;
int min_x = rect->width - 1;
int min_y = rect->height - 1;
int max_x = 0;
int max_y = 0;
dst_stride = dst->stride;
dst_buffer = dst->buffer + (dst_stride * rect->y) + (4 * rect->x);
/* For each row */
for (y=0; y < rect->height; y++) {
uint32_t* dst_current = (uint32_t*) dst_buffer;
/* Set row */
for (x=0; x < rect->width; x++) {
uint32_t old_color = *dst_current;
if (old_color != color) {
if (x < min_x) min_x = x;
if (y < min_y) min_y = y;
if (x > max_x) max_x = x;
if (y > max_y) max_y = y;
*dst_current = color;
}
dst_current++;
}
/* Next row */
dst_buffer += dst_stride;
}
/* Restrict destination rect to only updated pixels */
if (max_x >= min_x && max_y >= min_y) {
rect->x += min_x;
rect->y += min_y;
rect->width = max_x - min_x + 1;
rect->height = max_y - min_y + 1;
}
else {
rect->width = 0;
rect->height = 0;
}
}
/**
* Applies the Porter-Duff "over" composite operator, blending the two given
* color components using the given alpha value.
*
* @param dst
* The destination color component.
*
* @param src
* The source color component.
*
* @param alpha
* The alpha value which applies to the blending operation.
*
* @return
* The result of applying the Porter-Duff "over" composite operator to the
* given source and destination components.
*/
static int guac_common_surface_blend_component(int dst, int src, int alpha) {
int blended = src + dst * (0xFF - alpha);
/* Do not exceed maximum component value */
if (blended > 0xFF)
return 0xFF;
return blended;
}
/**
* Applies the Porter-Duff "over" composite operator, blending each component
* of the two given ARGB colors.
*
* @param dst
* The destination ARGB color.
*
* @param src
* The source ARGB color.
*
* @return
* The result of applying the Porter-Duff "over" composite operator to the
* given source and destination colors.
*/
static uint32_t guac_common_surface_argb_blend(uint32_t dst, uint32_t src) {
/* Separate destination ARGB color into its components */
int dst_a = (dst >> 24) & 0xFF;
int dst_r = (dst >> 16) & 0xFF;
int dst_g = (dst >> 8) & 0xFF;
int dst_b = dst & 0xFF;
/* Separate source ARGB color into its components */
int src_a = (src >> 24) & 0xFF;
int src_r = (src >> 16) & 0xFF;
int src_g = (src >> 8) & 0xFF;
int src_b = src & 0xFF;
/* If source is fully opaque (or destination is fully transparent), the
* blended result is the source */
if (src_a == 0xFF || dst_a == 0x00)
return src;
/* If source is fully transparent, the blended result is the destination */
if (src_a == 0x00)
return dst;
/* Otherwise, blend each ARGB component, assuming pre-multiplied alpha */
int r = guac_common_surface_blend_component(dst_r, src_r, src_a);
int g = guac_common_surface_blend_component(dst_g, src_g, src_a);
int b = guac_common_surface_blend_component(dst_b, src_b, src_a);
int a = guac_common_surface_blend_component(dst_a, src_a, src_a);
/* Recombine blended components */
return (a << 24) | (r << 16) | (g << 8) | b;
}
/**
* Copies data from the given buffer to the surface at the given coordinates.
* The dimensions and location of the destination rectangle will be altered
* to remove as many unchanged pixels as possible.
*
* @param src_buffer The buffer to copy.
* @param src_stride The number of bytes in each row of the source buffer.
* @param sx The X coordinate of the source rectangle.
* @param sy The Y coordinate of the source rectangle.
* @param dst The destination surface.
* @param rect The destination rectangle.
* @param opaque Non-zero if the source surface is opaque (its alpha channel
* should be ignored), zero otherwise.
*/
static void __guac_common_surface_put(unsigned char* src_buffer, int src_stride,
int* sx, int* sy,
guac_common_surface* dst, guac_common_rect* rect,
int opaque) {
unsigned char* dst_buffer = dst->buffer;
int dst_stride = dst->stride;
int x, y;
int min_x = rect->width;
int min_y = rect->height;
int max_x = 0;
int max_y = 0;
int orig_x = rect->x;
int orig_y = rect->y;
src_buffer += src_stride * (*sy) + 4 * (*sx);
dst_buffer += (dst_stride * rect->y) + (4 * rect->x);
/* For each row */
for (y=0; y < rect->height; y++) {
uint32_t* src_current = (uint32_t*) src_buffer;
uint32_t* dst_current = (uint32_t*) dst_buffer;
/* Copy row */
for (x=0; x < rect->width; x++) {
uint32_t color;
/* Get source and destination color values */
uint32_t src_color = *src_current;
uint32_t dst_color = *dst_current;
/* Ignore alpha channel if opaque */
if (opaque)
color = src_color | 0xFF000000;
/* Otherwise, perform alpha blending operation */
else
color = guac_common_surface_argb_blend(dst_color, src_color);
/* If the destination color is changing, update rectangle bounds
* and store the new color */
if (dst_color != color) {
if (x < min_x) min_x = x;
if (y < min_y) min_y = y;
if (x > max_x) max_x = x;
if (y > max_y) max_y = y;
*dst_current = color;
}
/* Advance to next pixel */
src_current++;
dst_current++;
}
/* Next row */
src_buffer += src_stride;
dst_buffer += dst_stride;
}
/* Restrict destination rect to only updated pixels */
if (max_x >= min_x && max_y >= min_y) {
rect->x += min_x;
rect->y += min_y;
rect->width = max_x - min_x + 1;
rect->height = max_y - min_y + 1;
}
else {
rect->width = 0;
rect->height = 0;
}
/* Update source X/Y */
*sx += rect->x - orig_x;
*sy += rect->y - orig_y;
}
/**
* Fills the given surface with color, using the given buffer as a mask. Color
* will be added to the given surface iff the corresponding pixel within the
* buffer is opaque.
*
* @param src_buffer The buffer to use as a mask.
* @param src_stride The number of bytes in each row of the source buffer.
* @param sx The X coordinate of the source rectangle.
* @param sy The Y coordinate of the source rectangle.
* @param dst The destination surface.
* @param rect The destination rectangle.
* @param red The red component of the color of the fill.
* @param green The green component of the color of the fill.
* @param blue The blue component of the color of the fill.
*/
static void __guac_common_surface_fill_mask(unsigned char* src_buffer, int src_stride,
int sx, int sy,
guac_common_surface* dst, guac_common_rect* rect,
int red, int green, int blue) {
unsigned char* dst_buffer = dst->buffer;
int dst_stride = dst->stride;
uint32_t color = 0xFF000000 | (red << 16) | (green << 8) | blue;
int x, y;
src_buffer += src_stride*sy + 4*sx;
dst_buffer += (dst_stride * rect->y) + (4 * rect->x);
/* For each row */
for (y=0; y < rect->height; y++) {
uint32_t* src_current = (uint32_t*) src_buffer;
uint32_t* dst_current = (uint32_t*) dst_buffer;
/* Stencil row */
for (x=0; x < rect->width; x++) {
/* Fill with color if opaque */
if (*src_current & 0xFF000000)
*dst_current = color;
src_current++;
dst_current++;
}
/* Next row */
src_buffer += src_stride;
dst_buffer += dst_stride;
}
}
/**
* Copies data from the given surface to the given destination surface using
* the specified transfer function.
*
* @param src_buffer The buffer to copy.
* @param src_stride The number of bytes in each row of the source buffer.
* @param sx The X coordinate of the source rectangle.
* @param sy The Y coordinate of the source rectangle.
* @param op The transfer function to use.
* @param dst The destination surface.
* @param rect The destination rectangle.
*/
static void __guac_common_surface_transfer(guac_common_surface* src, int* sx, int* sy,
guac_transfer_function op,
guac_common_surface* dst, guac_common_rect* rect) {
unsigned char* src_buffer = src->buffer;
unsigned char* dst_buffer = dst->buffer;
int x, y;
int src_stride, dst_stride;
int step = 1;
int min_x = rect->width - 1;
int min_y = rect->height - 1;
int max_x = 0;
int max_y = 0;
int orig_x = rect->x;
int orig_y = rect->y;
/* Copy forwards only if destination is in a different surface or is before source */
if (src != dst || rect->y < *sy || (rect->y == *sy && rect->x < *sx)) {
src_buffer += src->stride * (*sy) + 4 * (*sx);
dst_buffer += (dst->stride * rect->y) + (4 * rect->x);
src_stride = src->stride;
dst_stride = dst->stride;
step = 1;
}
/* Otherwise, copy backwards */
else {
src_buffer += src->stride * (*sy + rect->height - 1) + 4 * (*sx + rect->width - 1);
dst_buffer += dst->stride * (rect->y + rect->height - 1) + 4 * (rect->x + rect->width - 1);
src_stride = -src->stride;
dst_stride = -dst->stride;
step = -1;
}
/* For each row */
for (y=0; y < rect->height; y++) {
uint32_t* src_current = (uint32_t*) src_buffer;
uint32_t* dst_current = (uint32_t*) dst_buffer;
/* Transfer each pixel in row */
for (x=0; x < rect->width; x++) {
if (__guac_common_surface_transfer_int(op, src_current, dst_current)) {
if (x < min_x) min_x = x;
if (y < min_y) min_y = y;
if (x > max_x) max_x = x;
if (y > max_y) max_y = y;
}
src_current += step;
dst_current += step;
}
/* Next row */
src_buffer += src_stride;
dst_buffer += dst_stride;
}
/* Translate X coordinate space of moving backwards */
if (step < 0) {
int old_max_x = max_x;
max_x = rect->width - 1 - min_x;
min_x = rect->width - 1 - old_max_x;
}
/* Translate Y coordinate space of moving backwards */
if (dst_stride < 0) {
int old_max_y = max_y;
max_y = rect->height - 1 - min_y;
min_y = rect->height - 1 - old_max_y;
}
/* Restrict destination rect to only updated pixels */
if (max_x >= min_x && max_y >= min_y) {
rect->x += min_x;
rect->y += min_y;
rect->width = max_x - min_x + 1;
rect->height = max_y - min_y + 1;
}
else {
rect->width = 0;
rect->height = 0;
}
/* Update source X/Y */
*sx += rect->x - orig_x;
*sy += rect->y - orig_y;
}
guac_common_surface* guac_common_surface_alloc(guac_client* client,
guac_socket* socket, const guac_layer* layer, int w, int h) {
/* Calculate heat map dimensions */
int heat_width = GUAC_COMMON_SURFACE_HEAT_DIMENSION(w);
int heat_height = GUAC_COMMON_SURFACE_HEAT_DIMENSION(h);
/* Init surface */
guac_common_surface* surface = calloc(1, sizeof(guac_common_surface));
surface->client = client;
surface->socket = socket;
surface->layer = layer;
surface->parent = GUAC_DEFAULT_LAYER;
surface->opacity = 0xFF;
surface->width = w;
surface->height = h;
pthread_mutex_init(&surface->_lock, NULL);
/* Create corresponding Cairo surface */
surface->stride = cairo_format_stride_for_width(CAIRO_FORMAT_ARGB32, w);
surface->buffer = calloc(h, surface->stride);
/* Create corresponding heat map */
surface->heat_map = calloc(heat_width * heat_height,
sizeof(guac_common_surface_heat_cell));
/* Reset clipping rect */
guac_common_surface_reset_clip(surface);
/* Layers must initially exist */
if (layer->index >= 0) {
guac_protocol_send_size(socket, layer, w, h);
surface->realized = 1;
}
/* Defer creation of buffers */
else
surface->realized = 0;
return surface;
}
void guac_common_surface_free(guac_common_surface* surface) {
/* Only dispose of surface if it exists */
if (surface->realized)
guac_protocol_send_dispose(surface->socket, surface->layer);
pthread_mutex_destroy(&surface->_lock);
free(surface->heat_map);
free(surface->buffer);
free(surface);
}
void guac_common_surface_resize(guac_common_surface* surface, int w, int h) {
pthread_mutex_lock(&surface->_lock);
/* Ignore if resize will have no effect */
if (w == surface->width && h == surface->height)
goto complete;
guac_socket* socket = surface->socket;
const guac_layer* layer = surface->layer;
unsigned char* old_buffer;
int old_stride;
guac_common_rect old_rect;
int sx = 0;
int sy = 0;
/* Calculate heat map dimensions */
int heat_width = GUAC_COMMON_SURFACE_HEAT_DIMENSION(w);
int heat_height = GUAC_COMMON_SURFACE_HEAT_DIMENSION(h);
/* Copy old surface data */
old_buffer = surface->buffer;
old_stride = surface->stride;
guac_common_rect_init(&old_rect, 0, 0, surface->width, surface->height);
/* Re-initialize at new size */
surface->width = w;
surface->height = h;
surface->stride = cairo_format_stride_for_width(CAIRO_FORMAT_ARGB32, w);
surface->buffer = calloc(h, surface->stride);
__guac_common_bound_rect(surface, &surface->clip_rect, NULL, NULL);
/* Copy relevant old data */
__guac_common_bound_rect(surface, &old_rect, NULL, NULL);
__guac_common_surface_put(old_buffer, old_stride, &sx, &sy, surface, &old_rect, 1);
/* Free old data */
free(old_buffer);
/* Allocate completely new heat map (can safely discard old stats) */
free(surface->heat_map);
surface->heat_map = calloc(heat_width * heat_height,
sizeof(guac_common_surface_heat_cell));
/* Resize dirty rect to fit new surface dimensions */
if (surface->dirty) {
__guac_common_bound_rect(surface, &surface->dirty_rect, NULL, NULL);
if (surface->dirty_rect.width <= 0 || surface->dirty_rect.height <= 0)
surface->dirty = 0;
}
/* Update Guacamole layer */
if (surface->realized)
guac_protocol_send_size(socket, layer, w, h);
complete:
pthread_mutex_unlock(&surface->_lock);
}
void guac_common_surface_draw(guac_common_surface* surface, int x, int y, cairo_surface_t* src) {
pthread_mutex_lock(&surface->_lock);
unsigned char* buffer = cairo_image_surface_get_data(src);
cairo_format_t format = cairo_image_surface_get_format(src);
int stride = cairo_image_surface_get_stride(src);
int w = cairo_image_surface_get_width(src);
int h = cairo_image_surface_get_height(src);
int sx = 0;
int sy = 0;
guac_common_rect rect;
guac_common_rect_init(&rect, x, y, w, h);
/* Clip operation */
__guac_common_clip_rect(surface, &rect, &sx, &sy);
if (rect.width <= 0 || rect.height <= 0)
goto complete;
/* Update backing surface */
__guac_common_surface_put(buffer, stride, &sx, &sy, surface, &rect, format != CAIRO_FORMAT_ARGB32);
if (rect.width <= 0 || rect.height <= 0)
goto complete;
/* Update the heat map for the update rectangle. */
guac_timestamp time = guac_timestamp_current();
__guac_common_surface_touch_rect(surface, &rect, time);
/* Flush if not combining */
if (!__guac_common_should_combine(surface, &rect, 0))
__guac_common_surface_flush_deferred(surface);
/* Always defer draws */
__guac_common_mark_dirty(surface, &rect);
complete:
pthread_mutex_unlock(&surface->_lock);
}
void guac_common_surface_paint(guac_common_surface* surface, int x, int y,
cairo_surface_t* src, int red, int green, int blue) {
pthread_mutex_lock(&surface->_lock);
unsigned char* buffer = cairo_image_surface_get_data(src);
int stride = cairo_image_surface_get_stride(src);
int w = cairo_image_surface_get_width(src);
int h = cairo_image_surface_get_height(src);
int sx = 0;
int sy = 0;
guac_common_rect rect;
guac_common_rect_init(&rect, x, y, w, h);
/* Clip operation */
__guac_common_clip_rect(surface, &rect, &sx, &sy);
if (rect.width <= 0 || rect.height <= 0)
goto complete;
/* Update backing surface */
__guac_common_surface_fill_mask(buffer, stride, sx, sy, surface, &rect, red, green, blue);
/* Flush if not combining */
if (!__guac_common_should_combine(surface, &rect, 0))
__guac_common_surface_flush_deferred(surface);
/* Always defer draws */
__guac_common_mark_dirty(surface, &rect);
complete:
pthread_mutex_unlock(&surface->_lock);
}
void guac_common_surface_copy(guac_common_surface* src, int sx, int sy,
int w, int h, guac_common_surface* dst, int dx, int dy) {
/* Lock both surfaces */
pthread_mutex_lock(&dst->_lock);
if (src != dst)
pthread_mutex_lock(&src->_lock);
guac_socket* socket = dst->socket;
const guac_layer* src_layer = src->layer;
const guac_layer* dst_layer = dst->layer;
guac_common_rect srect;
guac_common_rect_init(&srect, sx, sy, w, h);
/* Clip operation source rect to bounds */
__guac_common_bound_rect(src, &srect, &dx, &dy);
if (srect.width <= 0 || srect.height <= 0)
goto complete;
guac_common_rect drect;
guac_common_rect_init(&drect, dx, dy,
srect.width, srect.height);
/* Clip operation destination rect */
__guac_common_clip_rect(dst, &drect, &srect.x, &srect.y);
if (drect.width <= 0 || drect.height <= 0)
goto complete;
/* NOTE: Being the last rectangle to be adjusted, only the width/height of
* drect is now correct! */
/* Update backing surface first only if drect cannot intersect srect */
if (src != dst) {
__guac_common_surface_transfer(src, &srect.x, &srect.y,
GUAC_TRANSFER_BINARY_SRC, dst, &drect);
if (drect.width <= 0 || drect.height <= 0)
goto complete;
}
/* Defer if combining */
if (__guac_common_should_combine(dst, &drect, 1))
__guac_common_mark_dirty(dst, &drect);
/* Otherwise, flush and draw immediately */
else {
__guac_common_surface_flush(dst);
__guac_common_surface_flush(src);
guac_protocol_send_copy(socket, src_layer, srect.x, srect.y,
drect.width, drect.height, GUAC_COMP_OVER, dst_layer,
drect.x, drect.y);
dst->realized = 1;
}
/* Update backing surface last if drect can intersect srect */
if (src == dst)
__guac_common_surface_transfer(src, &srect.x, &srect.y,
GUAC_TRANSFER_BINARY_SRC, dst, &drect);
complete:
/* Unlock both surfaces */
pthread_mutex_unlock(&dst->_lock);
if (src != dst)
pthread_mutex_unlock(&src->_lock);
}
void guac_common_surface_transfer(guac_common_surface* src, int sx, int sy, int w, int h,
guac_transfer_function op, guac_common_surface* dst, int dx, int dy) {
/* Lock both surfaces */
pthread_mutex_lock(&dst->_lock);
if (src != dst)
pthread_mutex_lock(&src->_lock);
guac_socket* socket = dst->socket;
const guac_layer* src_layer = src->layer;
const guac_layer* dst_layer = dst->layer;
guac_common_rect srect;
guac_common_rect_init(&srect, sx, sy, w, h);
/* Clip operation source rect to bounds */
__guac_common_bound_rect(src, &srect, &dx, &dy);
if (srect.width <= 0 || srect.height <= 0)
goto complete;
guac_common_rect drect;
guac_common_rect_init(&drect, dx, dy,
srect.width, srect.height);
/* Clip operation destination rect */
__guac_common_clip_rect(dst, &drect, &srect.x, &srect.y);
if (drect.width <= 0 || drect.height <= 0)
goto complete;
/* NOTE: Being the last rectangle to be adjusted, only the width/height of
* drect is now correct! */
/* Update backing surface first only if drect cannot intersect srect */
if (src != dst) {
__guac_common_surface_transfer(src, &srect.x, &srect.y, op, dst, &drect);
if (drect.width <= 0 || drect.height <= 0)
goto complete;
}
/* Defer if combining */
if (__guac_common_should_combine(dst, &drect, 1))
__guac_common_mark_dirty(dst, &drect);
/* Otherwise, flush and draw immediately */
else {
__guac_common_surface_flush(dst);
__guac_common_surface_flush(src);
guac_protocol_send_transfer(socket, src_layer, srect.x, srect.y,
drect.width, drect.height, op, dst_layer, drect.x, drect.y);
dst->realized = 1;
}
/* Update backing surface last if drect can intersect srect */
if (src == dst)
__guac_common_surface_transfer(src, &srect.x, &srect.y, op, dst, &drect);
complete:
/* Unlock both surfaces */
pthread_mutex_unlock(&dst->_lock);
if (src != dst)
pthread_mutex_unlock(&src->_lock);
}
void guac_common_surface_set(guac_common_surface* surface,
int x, int y, int w, int h, int red, int green, int blue, int alpha) {
pthread_mutex_lock(&surface->_lock);
guac_socket* socket = surface->socket;
const guac_layer* layer = surface->layer;
guac_common_rect rect;
guac_common_rect_init(&rect, x, y, w, h);
/* Clip operation */
__guac_common_clip_rect(surface, &rect, NULL, NULL);
if (rect.width <= 0 || rect.height <= 0)
goto complete;
/* Update backing surface */
__guac_common_surface_set(surface, &rect, red, green, blue, alpha);
if (rect.width <= 0 || rect.height <= 0)
goto complete;
/* Handle as normal draw if non-opaque */
if (alpha != 0xFF) {
/* Flush if not combining */
if (!__guac_common_should_combine(surface, &rect, 0))
__guac_common_surface_flush_deferred(surface);
/* Always defer draws */
__guac_common_mark_dirty(surface, &rect);
}
/* Defer if combining */
else if (__guac_common_should_combine(surface, &rect, 1))
__guac_common_mark_dirty(surface, &rect);
/* Otherwise, flush and draw immediately */
else {
__guac_common_surface_flush(surface);
guac_protocol_send_rect(socket, layer, rect.x, rect.y, rect.width, rect.height);
guac_protocol_send_cfill(socket, GUAC_COMP_OVER, layer, red, green, blue, alpha);
surface->realized = 1;
}
complete:
pthread_mutex_unlock(&surface->_lock);
}
void guac_common_surface_clip(guac_common_surface* surface, int x, int y, int w, int h) {
pthread_mutex_lock(&surface->_lock);
guac_common_rect clip;
/* Init clipping rectangle if clipping not already applied */
if (!surface->clipped) {
guac_common_rect_init(&surface->clip_rect, 0, 0, surface->width, surface->height);
surface->clipped = 1;
}
guac_common_rect_init(&clip, x, y, w, h);
guac_common_rect_constrain(&surface->clip_rect, &clip);
pthread_mutex_unlock(&surface->_lock);
}
void guac_common_surface_reset_clip(guac_common_surface* surface) {
pthread_mutex_lock(&surface->_lock);
surface->clipped = 0;
pthread_mutex_unlock(&surface->_lock);
}
/**
* Flushes the bitmap update currently described by the dirty rectangle within
* the given surface directly via an "img" instruction as PNG data. The
* resulting instructions will be sent over the socket associated with the
* given surface.
*
* @param surface
* The surface to flush.
*
* @param opaque
* Whether the rectangle being flushed contains only fully-opaque pixels.
*/
static void __guac_common_surface_flush_to_png(guac_common_surface* surface,
int opaque) {
if (surface->dirty) {
guac_socket* socket = surface->socket;
const guac_layer* layer = surface->layer;
/* Get Cairo surface for specified rect */
unsigned char* buffer = surface->buffer
+ surface->dirty_rect.y * surface->stride
+ surface->dirty_rect.x * 4;
cairo_surface_t* rect;
/* Use RGB24 if the image is fully opaque */
if (opaque)
rect = cairo_image_surface_create_for_data(buffer,
CAIRO_FORMAT_RGB24, surface->dirty_rect.width,
surface->dirty_rect.height, surface->stride);
/* Otherwise ARGB32 is needed */
else {
rect = cairo_image_surface_create_for_data(buffer,
CAIRO_FORMAT_ARGB32, surface->dirty_rect.width,
surface->dirty_rect.height, surface->stride);
/* Clear destination rect first */
guac_protocol_send_rect(socket, layer,
surface->dirty_rect.x, surface->dirty_rect.y,
surface->dirty_rect.width, surface->dirty_rect.height);
guac_protocol_send_cfill(socket, GUAC_COMP_ROUT, layer,
0x00, 0x00, 0x00, 0xFF);
}
/* Send PNG for rect */
guac_client_stream_png(surface->client, socket, GUAC_COMP_OVER,
layer, surface->dirty_rect.x, surface->dirty_rect.y, rect);
cairo_surface_destroy(rect);
surface->realized = 1;
/* Surface is no longer dirty */
surface->dirty = 0;
}
}
/**
* Returns an appropriate quality between 0 and 100 for lossy encoding
* depending on the current processing lag calculated for the given client.
*
* @param client
* The client for which the lossy quality is being calculated.
*
* @return
* A value between 0 and 100 inclusive which seems appropriate for the
* client based on lag measurements.
*/
static int guac_common_surface_suggest_quality(guac_client* client) {
int lag = guac_client_get_processing_lag(client);
/* Scale quality linearly from 90 to 30 as lag varies from 20ms to 80ms */
int quality = 90 - (lag - 20);
/* Do not exceed 90 for quality */
if (quality > 90)
return 90;
/* Do not go below 30 for quality */
if (quality < 30)
return 30;
return quality;
}
/**
* Flushes the bitmap update currently described by the dirty rectangle within
* the given surface directly via an "img" instruction as JPEG data. The
* resulting instructions will be sent over the socket associated with the
* given surface.
*
* @param surface
* The surface to flush.
*/
static void __guac_common_surface_flush_to_jpeg(guac_common_surface* surface) {
if (surface->dirty) {
guac_socket* socket = surface->socket;
const guac_layer* layer = surface->layer;
guac_common_rect max;
guac_common_rect_init(&max, 0, 0, surface->width, surface->height);
/* Expand the dirty rect size to fit in a grid with cells equal to the
* minimum JPEG block size */
guac_common_rect_expand_to_grid(GUAC_SURFACE_JPEG_BLOCK_SIZE,
&surface->dirty_rect, &max);
/* Get Cairo surface for specified rect */
unsigned char* buffer = surface->buffer
+ surface->dirty_rect.y * surface->stride
+ surface->dirty_rect.x * 4;
cairo_surface_t* rect = cairo_image_surface_create_for_data(buffer,
CAIRO_FORMAT_RGB24, surface->dirty_rect.width,
surface->dirty_rect.height, surface->stride);
/* Send JPEG for rect */
guac_client_stream_jpeg(surface->client, socket, GUAC_COMP_OVER, layer,
surface->dirty_rect.x, surface->dirty_rect.y, rect,
guac_common_surface_suggest_quality(surface->client));
cairo_surface_destroy(rect);
surface->realized = 1;
/* Surface is no longer dirty */
surface->dirty = 0;
}
}
/**
* Flushes the bitmap update currently described by the dirty rectangle within
* the given surface directly via an "img" instruction as WebP data. The
* resulting instructions will be sent over the socket associated with the
* given surface.
*
* @param surface
* The surface to flush.
*
* @param opaque
* Whether the rectangle being flushed contains only fully-opaque pixels.
*/
static void __guac_common_surface_flush_to_webp(guac_common_surface* surface,
int opaque) {
if (surface->dirty) {
guac_socket* socket = surface->socket;
const guac_layer* layer = surface->layer;
guac_common_rect max;
guac_common_rect_init(&max, 0, 0, surface->width, surface->height);
/* Expand the dirty rect size to fit in a grid with cells equal to the
* minimum WebP block size */
guac_common_rect_expand_to_grid(GUAC_SURFACE_WEBP_BLOCK_SIZE,
&surface->dirty_rect, &max);
/* Get Cairo surface for specified rect */
unsigned char* buffer = surface->buffer
+ surface->dirty_rect.y * surface->stride
+ surface->dirty_rect.x * 4;
cairo_surface_t* rect;
/* Use RGB24 if the image is fully opaque */
if (opaque)
rect = cairo_image_surface_create_for_data(buffer,
CAIRO_FORMAT_RGB24, surface->dirty_rect.width,
surface->dirty_rect.height, surface->stride);
/* Otherwise ARGB32 is needed */
else
rect = cairo_image_surface_create_for_data(buffer,
CAIRO_FORMAT_ARGB32, surface->dirty_rect.width,
surface->dirty_rect.height, surface->stride);
/* Send WebP for rect */
guac_client_stream_webp(surface->client, socket, GUAC_COMP_OVER, layer,
surface->dirty_rect.x, surface->dirty_rect.y, rect,
guac_common_surface_suggest_quality(surface->client), 0);
cairo_surface_destroy(rect);
surface->realized = 1;
/* Surface is no longer dirty */
surface->dirty = 0;
}
}
/**
* Comparator for instances of guac_common_surface_bitmap_rect, the elements
* which make up a surface's bitmap buffer.
*
* @see qsort
*/
static int __guac_common_surface_bitmap_rect_compare(const void* a, const void* b) {
guac_common_surface_bitmap_rect* ra = (guac_common_surface_bitmap_rect*) a;
guac_common_surface_bitmap_rect* rb = (guac_common_surface_bitmap_rect*) b;
/* Order roughly top to bottom, left to right */
if (ra->rect.y != rb->rect.y) return ra->rect.y - rb->rect.y;
if (ra->rect.x != rb->rect.x) return ra->rect.x - rb->rect.x;
/* Wider updates should come first (more likely to intersect later) */
if (ra->rect.width != rb->rect.width) return rb->rect.width - ra->rect.width;
/* Shorter updates should come first (less likely to increase cost) */
return ra->rect.height - rb->rect.height;
}
/**
* Flushes only the properties of the given surface, such as layer location or
* opacity. Image state is not flushed. If the surface represents a buffer or
* the default layer, this function has no effect.
*
* @param surface
* The surface to flush.
*/
static void __guac_common_surface_flush_properties(
guac_common_surface* surface) {
guac_socket* socket = surface->socket;
/* Only applicable to non-default visible layers */
if (surface->layer->index <= 0)
return;
/* Flush opacity */
if (surface->opacity_dirty) {
guac_protocol_send_shade(socket, surface->layer, surface->opacity);
surface->opacity_dirty = 0;
}
/* Flush location and hierarchy */
if (surface->location_dirty) {
guac_protocol_send_move(socket, surface->layer,
surface->parent, surface->x, surface->y, surface->z);
surface->location_dirty = 0;
}
}
static void __guac_common_surface_flush(guac_common_surface* surface) {
/* Flush final dirty rectangle to queue. */
__guac_common_surface_flush_to_queue(surface);
guac_common_surface_bitmap_rect* current = surface->bitmap_queue;
int i, j;
int original_queue_length;
int flushed = 0;
original_queue_length = surface->bitmap_queue_length;
/* Sort updates to make combination less costly */
qsort(surface->bitmap_queue, surface->bitmap_queue_length, sizeof(guac_common_surface_bitmap_rect),
__guac_common_surface_bitmap_rect_compare);
/* Flush all rects in queue */
for (i=0; i < surface->bitmap_queue_length; i++) {
/* Get next unflushed candidate */
guac_common_surface_bitmap_rect* candidate = current;
if (!candidate->flushed) {
int combined = 0;
/* Build up rect as much as possible */
for (j=i; j < surface->bitmap_queue_length; j++) {
if (!candidate->flushed) {
/* Clip candidate within current bounds */
__guac_common_bound_rect(surface, &candidate->rect, NULL, NULL);
if (candidate->rect.width <= 0 || candidate->rect.height <= 0)
candidate->flushed = 1;
/* Combine if reasonable */
else if (__guac_common_should_combine(surface, &candidate->rect, 0) || !surface->dirty) {
__guac_common_mark_dirty(surface, &candidate->rect);
candidate->flushed = 1;
combined++;
}
}
candidate++;
}
/* Re-add to queue if there's room and this update was modified or we expect others might be */
if ((combined > 1 || i < original_queue_length)
&& surface->bitmap_queue_length < GUAC_COMMON_SURFACE_QUEUE_SIZE)
__guac_common_surface_flush_to_queue(surface);
/* Flush as bitmap otherwise */
else if (surface->dirty) {
flushed++;
int opaque = __guac_common_surface_is_opaque(surface,
&surface->dirty_rect);
/* Prefer WebP when reasonable */
if (__guac_common_surface_should_use_webp(surface,
&surface->dirty_rect))
__guac_common_surface_flush_to_webp(surface, opaque);
/* If not WebP, JPEG is the next best (lossy) choice */
else if (opaque && __guac_common_surface_should_use_jpeg(
surface, &surface->dirty_rect))
__guac_common_surface_flush_to_jpeg(surface);
/* Use PNG if no lossy formats are appropriate */
else
__guac_common_surface_flush_to_png(surface, opaque);
}
}
current++;
}
/* Flush complete */
surface->bitmap_queue_length = 0;
}
void guac_common_surface_flush(guac_common_surface* surface) {
pthread_mutex_lock(&surface->_lock);
/* Flush any applicable layer properties */
__guac_common_surface_flush_properties(surface);
/* Flush surface contents */
__guac_common_surface_flush(surface);
pthread_mutex_unlock(&surface->_lock);
}
void guac_common_surface_dup(guac_common_surface* surface, guac_user* user,
guac_socket* socket) {
pthread_mutex_lock(&surface->_lock);
/* Do nothing if not realized */
if (!surface->realized)
goto complete;
/* Synchronize layer-specific properties if applicable */
if (surface->layer->index > 0) {
/* Synchronize opacity */
guac_protocol_send_shade(socket, surface->layer, surface->opacity);
/* Synchronize location and hierarchy */
guac_protocol_send_move(socket, surface->layer,
surface->parent, surface->x, surface->y, surface->z);
}
/* Sync size to new socket */
guac_protocol_send_size(socket, surface->layer,
surface->width, surface->height);
/* Send contents of layer, if non-empty */
if (surface->width > 0 && surface->height > 0) {
/* Get entire surface */
cairo_surface_t* rect = cairo_image_surface_create_for_data(
surface->buffer, CAIRO_FORMAT_ARGB32,
surface->width, surface->height, surface->stride);
/* Send PNG for rect */
guac_user_stream_png(user, socket, GUAC_COMP_OVER, surface->layer,
0, 0, rect);
cairo_surface_destroy(rect);
}
complete:
pthread_mutex_unlock(&surface->_lock);
}