utils: disable object validity check in release builds

[mechakotik]

These checks seem to affect performance, particularly they take a huge part of SDL_RenderCopy time:

44.54% SDL_RenderTextureRotated
   - 44.48% SDL_RenderTextureRotated_REAL
      - 43.57% SDL_RenderTexture_REAL
         + 20.38% SDL_RenderTextureInternal
         + 19.42% SDL_ObjectValid
         + 1.65% SDL_GetRectIntersectionFloat_REAL

This perf report is taken in debug build, as in release everything gets optimized out and perf doesn't report anything meaningful. So I've also made a simple RenderCopy benchmark:

Source

#include <SDL3/SDL.h>
#include <SDL3_image/SDL_image.h>
#include <random>

int main() {
    SDL_Init(SDL_INIT_VIDEO);
    SDL_Window* window = SDL_CreateWindow("window", 1280, 720, SDL_WINDOW_RESIZABLE);
    SDL_Renderer* renderer = SDL_CreateRenderer(window, nullptr);
    SDL_Surface* surface = IMG_Load("image.png");
    SDL_Texture* texture = SDL_CreateTextureFromSurface(renderer, surface);
    SDL_SetRenderVSync(renderer, SDL_RENDERER_VSYNC_DISABLED);

    SDL_FRect src, dst;
    src.x = src.y = 0;
    src.w = src.h = 8;
    dst.w = 32;
    dst.h = 32;

    std::mt19937 gen(42);

    for(int it = 0; it < 500000; it++) {
        SDL_RenderClear(renderer);
        for(int i = 0; i < 500; i++) {
            dst.x = gen() % 1280;
            dst.y = gen() % 720;
            SDL_RenderTexture(renderer, texture, &src, &dst);
        }
        SDL_RenderPresent(renderer);
    }

    SDL_DestroyTexture(texture);
    SDL_DestroyRenderer(renderer);
    SDL_DestroyWindow(window);
    SDL_Quit();
}

image.png:

Ran on Ryzen 9 7940HS, Gentoo Linux, performance governor. There is a noticeable performance difference (and it becomes more noticeable when increasing number of draws in single frame).

# with validity checks
Executed in   39.40 secs    fish           external
   usr time   26.64 secs    0.21 millis   26.64 secs
   sys time   10.08 secs    2.90 millis   10.08 secs

# without validity checks
Executed in   33.71 secs    fish           external
   usr time   22.34 secs    0.12 millis   22.34 secs
   sys time    9.70 secs    1.02 millis    9.70 secs

I've added a CMake option to enable/disable validity checks. It is enabled by default in Debug builds and disabled otherwise. There's a problem that default value doesn't get updated when changing build type in existing build directory, I don't know how to fix it.

P.S. Are these checks needed at all when there's AddressSanitizer to detect use-after-free efficiently?

[mechakotik]

Okay, looks like ObjectValid is used not only to detect use-after-free and can't be just disabled everywhere. What's the actual purpose of it?