0ct0pu5/wesnoth
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Spritesheet Generator: implemented multi-thread image loading

Browse source 
This results in a rough performance increase of anywhere from 30% - 50%.
This commit is contained in:
Charles Dang 2018-04-10 17:00:23 +11:00
parent f9e44ea282
commit 325691107a
1 changed files with 34 additions and 7 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

41
src/spritesheet_generator.cpp
View file

@ -25,6 +25,8 @@
#include <algorithm>
#include <iostream>
#include <map>
#include <mutex>
#include <thread>
namespace image
{
@ -35,7 +37,7 @@ struct sheet_elment_data
/**
* The spritesheet.
*
* Do note The texture class is really a "texture reference" class; it only
* Do note the texture class is really a "texture reference" class; it only
* owns a shared pointer to the actual texture in memory. Therefor, having
* this object does not mean we're keeping multiple copies of each sheet.
*/
@ -63,13 +65,37 @@ void build_sheet_from_images(const std::vector<std::string>& file_paths)
std::vector<sheet_build_helper> surfs;
surfs.reserve(file_paths.size());
// Load all the images.
for(const auto& f : file_paths) {
surface temp = image::get_image(f);
std::vector<std::thread> loader_threads;
loader_threads.reserve(surfs.size());
if(!temp.null()) {
surfs.push_back({ std::move(temp), f});
}
std::mutex surf_list_lock;
// Load each image in a separate thread.
for(const auto& f : file_paths) {
// TODO: image::load_from_disk isn't fully thread-safe since it creates a
// temportary locator object. locator::init_index modifies a global state map,
// which forces the thread into an endless loop. We should cache management
// from the locator class, but for now we need to create the locator in the
// main thread. The captured f parameter should replace this once the treading
// issues are sorted.
const image::locator loc(f);
loader_threads.emplace_back([&surf_list_lock, &surfs, f, loc]() {
// Load image directly. We don't need it added to the surface cache.
surface temp = image::load_from_disk(loc);
// Lock AFTER image loading. That can happen concurrently but the vector access can't.
std::lock_guard<std::mutex> lock(surf_list_lock);
if(!temp.null()) {
surfs.push_back({ std::move(temp), f});
}
});
}
// Wait for all loader theads to finish.
for(auto& thread : loader_threads) {
thread.join();
}
if(surfs.empty()) {
@ -195,6 +221,7 @@ void build_spritesheet_from_impl(const std::string& dir, const std::string& subd
}
} // end anon namespace
void build_spritesheet_from(const std::string& subdir)
{
#ifdef DEBUG_SPRITESHEET_OUTPUT