Question

I want to make my code parallel, but have some questions for the experienced.

1] Is it a good idea to do something like this?

#pragma omp parallel num_threads(2)
{
    const int threadID=omp_get_thread_num();
    if(threadID==0) while(WM_QUIT != msg.message){ // < Engine stuff
        if(PeekMessage(&msg, NULL, 0, 0, PM_REMOVE)){
            TranslateMessage(&msg);
            DispatchMessage(&msg);
        }else{
            engine.omp_renderPrepare();
        }
    }
    if(threadID==1) while(WM_QUIT != msg.message){ // < Render stuff
        engine.omp_renderApply();
    }
}

// Functions from the engine class:
inline void omp_renderPrepare(){
    if(gDevice.omp_isReady()){
#pragma omp critical(renderPrepare)
        if(!omp_rendering){
            omp_sceneReady=0;
            renderPrepare();
            omp_sceneReady=1;
        }
    }else{
        Sleep(42); // If the device is not ready, dont spam.
    }
}
inline void omp_renderApply(){
    if(gDevice.omp_isReady()){
#pragma omp critical(omp_renderApply)
        if(omp_sceneReady){
            omp_rendering=1;
            renderApply();
            omp_rendering=0;
        }
    }else{
        Sleep(42); // If the device is not ready, dont spam.
    }
}

I already applied it, but it is heavy on the CPU & my code slows to a crawl when I resize the window, since it must:

1. Set a bool “omp_deviceReady” to false so the rendering will stop.
2. Wait for the parallel sections to stop.
3. Resizing the graphics buffers.
4. Set “omp_deviceReady” to true so rendering can happen again.
5. Repeat as long as the user resizes the window.

2] Since my OMP code is heavy on the CPU [about 50% accord to task mgr], what is a good way to handle parallel rendering?

Answer 1

“OMP isn’t too well suited for MT rendering (OMP would work better for parallel processing of game tasks)” – Necrolis

“Deffered contexts is a good way to go, …” – catflier