Create a better FPS Counter (Frame Time Graph)

[devshgraphicsprogramming]

Provide this via an irr::ext::perf_counter, get the times via GPU time elapsed query ( actually have two modes to measure elapsed time queries between start_lap/stop_lap, one for CPU and one for GPU time taken) .

Keep the last N frametimes (in uint32_t microseconds) in a pre-allocated uint32_t* buffer (use %N addressing). Doesn't matter if measuring GPU or CPU timestamp.

Provide an irr::ext::perf_graph extension to visualize graphs on screen (with anti-aliased lines with blending), with a user specified sample delay (show data from K frames behind).

Preferrrably irr::ext::perf_counter would output into a persistenly mapped IGPUBuffer which would be used by irr::ext::perf_graph as an UBO for the pixel shader that would draw the graph.

We want to measure multiple things and have multiple graphs.

Also irr::ext:perf_counter should have an option to serialize to a std::string .csv style.

[devshgraphicsprogramming]

The heuristics of this thing https://github.com/buildaworldnet/IrrlichtBAW/blob/master/source/Irrlicht/CFPSCounter.cpp are abysmal

[devshgraphicsprogramming]

Also make it possible to collect other statistics within the time period, such as:

• Number of calls to _IRR_MALLOC_ALIGNED in the time period (only if compiled with correct flag, dont want to add that overhead to release configs)
• Number of calls to IDriver::createBuffer* and Driver Memory Object Allocations
• Number of GPU flushes and invalidates
• Number of command submits to a graphics execution queue (when API gets made)
• Number of resource transitions in the submitted command buffers
• Number of pipeline barriers in the submitted command buffers
• Number of subpasses and renderpasses in the submitted command buffers
• Number of Compute and Graphics pipeline swaps
• Number of descriptor set rebinds in submitted command buffers
• Total number of rebound descriptors in the descriptor sets that were rebound
• Number of Dynamic State changes (viewport, scissor, etc.)
• Number of push constants calls and total data specified as push constants
• Number of dispatch commands in the submitted command buffers
• Number of draw commands in the submitted command buffers

[devshgraphicsprogramming]

Should probably integrate Tracy or some other CPU and GPU profiler