C++11 + SDL2 + Windows: Multithreaded program hangs after any input event - windows

Question

I am working on a screen capture program using C++11, MinGW, and the Windows API. I am trying to use SDL2 to watch how my screen capture program works in real time.
The window opens fine, and the program seems to run well so long as I do nothing more than move the mouse cursor. But iff I click in the window, its menu bar, outside the window, or press any keys, the SDL window freezes.
I have set up some logging for the events to figure out what is happening. I never receive any events other than SDL_WINDOW_FOCUS_GAINED, SDL_TEXTEDITING, and SDL_WINDOWEVENT_SHOWN in that order. All of these are received at the start.
I have tried to find tutorials on SDL event handling since that's my best guess as to the source of the problem. I have found nothing more than basic event handling to watch for SDL_QUIT, basic mouse and keyboard events, and one on SDL_WINDOWEVENTs that does not seem to help. I have found nothing in-depth on what the events mean and best practices for handling them. That may not matter, because that might not be the source of the problem. For all I know, SDL is throwing a fit because there are other threads running.
Can anyone see any cause for this hanging in my code and provide an explanation as to how to fix it?
A quick explanation for the structure of my program is in order to cover the code I have omitted. The Captor class starts and runs a thread to grab a screenshot to pass to the Encoder. The Encoder starts a variable number of threads that receive a screenshot from the Captor, encode the screenshot, then passes the encoding to the Screen. The passing mechanism is the SynchronousQueue<T> class that provides paired methods put(const T&) and T get() to allow a producer and a consumer to synchronize using a resource; these methods time out to allow the the system to be responsive to kill messages.
Now for the source files (hopefully without too much bloat). While I would appreciate any comments on how to improve the performance of the application, my focus is on making the program responsive.
main.cpp
#include "RTSC.hpp"
int main(int argc, char** argv) {
RTSC rtsc {
(uint32_t) stoi(argv[1]),
(uint32_t) stoi(argv[2]),
(uint32_t) stoi(argv[3]),
(uint32_t) stoi(argv[4]),
(uint32_t) stoi(argv[5]),
(uint32_t) stoi(argv[6])
};
while (rtsc.isRunning()) {
SwitchToThread();
}
return 0;
}
RTSC.hpp
#ifndef RTSC_HPP
#define RTSC_HPP
#include "Captor.hpp"
#include "Encoder.hpp"
#include "Screen.hpp"
#include <iostream>
using namespace std;
class RTSC {
private:
Captor *captor;
Encoder *encoder;
SynchronousQueue<uint8_t*> imageQueue {1};
SynchronousQueue<RegionList> regionQueue {1};
Screen *screen;
public:
RTSC(
uint32_t width,
uint32_t height,
uint32_t maxRegionCount,
uint32_t threadCount,
uint32_t divisionsAlongThreadWidth,
uint32_t divisionsAlongThreadHeight
) {
captor = new Captor(width, height, imageQueue);
encoder = new Encoder(
width,
height,
maxRegionCount,
threadCount,
divisionsAlongThreadWidth,
divisionsAlongThreadHeight,
imageQueue,
regionQueue
);
screen = new Screen(
width,
height,
width >> 1,
height >> 1,
regionQueue
);
captor->start();
}
~RTSC() {
delete screen;
delete encoder;
delete captor;
}
bool isRunning() const {
return screen->isRunning();
}
};
#endif
Screen.hpp
#ifndef SCREEN_HPP
#define SCREEN_HPP
#include <atomic>
#include <SDL.h>
#include <windows.h>
#include "Region.hpp"
#include "SynchronousQueue.hpp"
using namespace std;
class Screen {
private:
atomic_bool running {false};
HANDLE thread;
SynchronousQueue<RegionList>* inputQueue;
uint32_t inputHeight;
uint32_t inputWidth;
uint32_t screenHeight;
uint32_t screenWidth;
SDL_Renderer* renderer;
SDL_Surface* surface;
SDL_Texture* texture;
SDL_Window* window;
void run() {
SDL_Event event;
while (running) {
while (SDL_PollEvent(&event)) {
switch (event.type) {
case SDL_QUIT:
running = false;
break;
case SDL_WINDOWEVENT:
switch (event.window.event) {
case SDL_WINDOWEVENT_CLOSE:
running = false;
break;
default:
break;
}
}
try {
RegionList rl = inputQueue->get();
SDL_RenderClear(renderer);
SDL_LockSurface(surface);
SDL_FillRect(surface, nullptr, 0);
for (uint32_t i = 0; i < rl.count; ++i) {
Region &r = rl.regions[i];
SDL_Rect rect {
(int) r.getX(),
(int) r.getY(),
(int) r.getWidth(),
(int) r.getHeight()
};
uint32_t color =
(r.getRed() << 16) +
(r.getGreen() << 8) +
r.getBlue();
SDL_FillRect(surface, &rect, color);
}
SDL_UnlockSurface(surface);
SDL_UpdateTexture(
texture,
nullptr,
surface->pixels,
surface->pitch
);
SDL_RenderCopyEx(
renderer,
texture,
nullptr,
nullptr,
0,
nullptr,
SDL_FLIP_VERTICAL
);
} catch (exception &e) {}
SDL_RenderPresent(renderer);
SwitchToThread();
}
}
static DWORD startThread(LPVOID self) {
((Screen*) self)->run();
return (DWORD) 0;
}
public:
Screen(
uint32_t inputWidth,
uint32_t inputHeight,
uint32_t windowWidth,
uint32_t windowHeight,
SynchronousQueue<RegionList> &inputQueue
): inputQueue {&inputQueue}, inputHeight {inputHeight} {
SDL_Init(SDL_INIT_VIDEO);
window = SDL_CreateWindow(
"RTSC",
SDL_WINDOWPOS_CENTERED,
SDL_WINDOWPOS_CENTERED,
windowWidth,
windowHeight,
SDL_WINDOW_SHOWN | SDL_WINDOW_RESIZABLE |
SDL_WINDOW_INPUT_FOCUS | SDL_WINDOW_MOUSE_FOCUS
);
renderer = SDL_CreateRenderer(window, -1, 0);
surface = SDL_CreateRGBSurface(
0,
inputWidth,
inputHeight,
24,
0xFF << 16,
0xFF << 8,
0xFF,
0
);
texture = SDL_CreateTexture(
renderer,
surface->format->format,
SDL_TEXTUREACCESS_STREAMING,
inputWidth,
inputHeight
);
running = true;
thread = CreateThread(nullptr, 0, startThread, this, 0, nullptr);
}
~Screen() {
running = false;
WaitForSingleObject(thread, INFINITE);
CloseHandle(thread);
SDL_FreeSurface(surface);
SDL_DestroyRenderer(renderer);
SDL_DestroyWindow(window);
SDL_Quit();
}
bool isRunning() const {
return running;
}
};
#endif

Answer 1

I have no experience in using SDL API in a multithreaded environment but this isn't a big problem as you will see later. I've checked your code and there is at least one thing you should change in my opinion.
Generally, in case of GUI systems (and partly SDL is also a gui system) you should always access the gui only from the main thread and expect the gui events to come from the main thread. Most GUI APIs are single threaded and I wouldn't be surprised if this would apply to SDL too. Note that many gui systems run on the main thread of your process by default and you can't choose your own thread. Don't run the code of your Screen class on a worker thread, run it on your main thread and make EVERY SDL API call from the main thread.
If you are writing the game or a similar software then (first) write it as if it was single threaded. The subsystems of your engine (physics simulation, this-and-that-system, game logic, rendering) should be executed serially one-after-the-other on your main thread (from your main loop). If you want to make use of multithreading that do that in "another dimension": Convert some of the subsystems or a smaller unit of work (like merge sort) to multithreaded, for example a physics system tasks can often be split into several small tasks so when the physics system is updated by the main thread then the physics system can burn all of your cores...
Doing most of your tasks on the main thread has another advantage: It makes your code much more easy to port to any platform. Optionally if you write your code so that it can execute in single threaded mode then it can make debugging easier in many cases and then you also have a "reference build" to compare the multithreaded build to performancewise.