Land offthread rendering

[taisel]

We already run the emulator in a webworker if your browser supports sharedarraybuffer. We'll just spawn another webworker controlled by the emulator core inside first webworker, to render graphics. This really needs to land for performance reasons, we can always pull out if it doesn't give an improvement.

Basically if we land this chunk of code, it'll purposefully start rendering at the start of a scanline, and run in parallel to the core emulation. We're exploiting the undefined-ness of mid-scanline rendering to offthread this, even for scanline-by-scanline interrupts.

[taisel]

I already have this feature planned out, so I'll code monkey it.

[taisel]

I already have this feature planned out, so I'll code monkey it.

[taisel]

The renderer the graphics proxy holds a reference to will be replaced by a shim that contains two fifo ring buffers. They will proxy commands to the webworker and the framebuffer result back to the core thread. The graphics renderer webworker will only thread block/futexWait if the outbound framebuffer ring buffer is full, while the core thread will block/futexWait while the command buffer is full. I/O and VRAM writes will be pushed onto the fifo command buffer unprocessed, with a correspond ID sequence prior to the (optional) data.

[taisel]

The renderer the graphics proxy holds a reference to will be replaced by a shim that contains two fifo ring buffers. They will proxy commands to the webworker and the framebuffer result back to the core thread. The graphics renderer webworker will only thread block/futexWait if the outbound framebuffer ring buffer is full, while the core thread will block/futexWait while the command buffer is full. I/O and VRAM writes will be pushed onto the fifo command buffer unprocessed, with a correspond ID sequence prior to the (optional) data.

[taisel]

There will be a separate VRAM copy inside the core emulation thread, so it doesn't have to block on R/W of VRAM.

[taisel]

There will be a separate VRAM copy inside the core emulation thread, so it doesn't have to block on R/W of VRAM.

[taisel]

Starting work for this in https://github.com/taisel/IodineGBA/tree/offthread-gfx

[taisel]

Starting work for this in https://github.com/taisel/IodineGBA/tree/offthread-gfx

[taisel]

#22

[taisel]

#22