examples/sprites: Performance improvement on Metal

[hajimehoshi]

FPS should keep 60 FPS even when the number of sprites exceeds 10000.

[hajimehoshi]

FPS reaches 60 with 10000 sprites, but does not with 20000 sprites.

BTW, it is hard to compare FPS with OpenGL era, since vsync didn't work at that time 🤔

[hajimehoshi]

As I expected, Device's MakeBuffer is very slow when 9000> sprites are shown.

[hajimehoshi]

https://developer.apple.com/documentation/metal/advanced_command_setup/cpu_and_gpu_synchronization?language=objc

This sample renders hundreds of small quads, also known as sprites. To animate the sprites, the sample updates their positions at the start of each frame and writes them into a vertex buffer. When a frame is complete, the CPU and GPU no longer need the vertex buffer for that frame. Discarding a used vertex buffer and creating a new one for each frame is wasteful. Instead, a more sustainable model can be implemented with a FIFO queue of reusable vertex buffers.

[hajimehoshi]

From my very simple experiment, the cost difference between copying bytes to a buffer and creating a new buffer was not so big.

[hajimehoshi]

Now we can keep 60 FPS with <20000 sprites. Can we keep it with more sprites by reusing buffers?

[hajimehoshi]

Reduce 'if' from the shader programs 81101e0

Remove 'while' from the shader programs 7a76a5b

Thanks to them, now we can keep 60FPS with 40000 sprites on my MacBook Pro 2018 😄

[hajimehoshi]

Now I think Metal version is fast enough.