BASILISK II FULL
BASILISK II CODE

For comparison, see this alternate port I put together without the use of shared memory, which has to regularly yield to the browser event loop. Performanceīy using a worker thread with shared memory, this port manages to achieve a great framerate, with the mouse movement feeling relatively fluid.

The audio read in the UI thread here and output using the Web Audio API. The audio buffer is passed to JS here and copied into a SharedArrayBuffer here. This could be improved by using Emscripten's pthreads support to spawn a separate audio thread. When the emulator is built natively, audio has a dedicated thread to avoid slowness on the emulator thread affecting playback, but to make things simpler I've just hacked in a call to output audio directly into the emulator's main loop here. The emulator thread will ensure that there are several buffers of audio queued up, to avoid gaps in playback in the case of a slowdown.

BASILISK II FULL

The implementation of audio uses a queue of buffers in shared memory, which each have a flag signifying whether they are full and ready to be consumed by the UI thread, or empty and ready to be written by the emulator thread. I also had to add a mapping to convert JS keycodes to ADB (Apple) keycodes.

BASILISK II CODE

Mouse and keyboard input are communicated via another SharedArrayBuffer here in the UI thread JS, then read in the worker thread JS here, when requested from the C code here. We could improve on this by using a circular queue of framebuffers for alternating frames (multiple buffering), which would allow locking (with less contention), but in practice this is okay. Even if the UI thread reads a frame from the video framebuffer while it is currently in the middle of being written to from the emulator worker, it's not really noticable, because in the shared memory the new frame contents is just being written directly over the old one at the same position, so visually it just presents as a bit of 'tearing' (showing part of the old frame, part of the new frame). I experimented with locking the video surface when reading and writing, but this hurt performance due to lock contention. This shared memory is read from the main thread here and output onto a Canvas. Where the EM_ASM_ macro is used to call this JS function to copy the contents of the video framebuffer to a SharedArrayBuffer. Emscripten provides an implementation of SDL for the browser, but this implementation isn't designed to run in a web worker, so I've hacked the emulator's video output code to write to the SharedArrayBuffer instead of calling SDL APIs. The original emulator codebase makes use of SDL (a cross-platform set of video/audio/input APIs). This allows the simulation to be more smooth, as the emulator thread can just focus on simulating the Macintosh computer, and leave the tasks of displaying the contents of the video framebuffer and playing audio up to the main (browser) thread. Communication with the main thread happens by reading and writing data in SharedArrayBuffers which are shared between the main browser thread and the worker, allowing the emulator's main loop to run continuously without ever yielding to the event loop.

The emulator code is compiled with Emscripten and run in a Web Worker. Anyway, let's get into the gory technical details of how it works. The Basilisk II port is the culmination of both of these things. Since then browser technologies have advanced, and I've learned a lot more about emulation (in part by making a simulated computer of my own). I've been hoping to get it running for some time, and previously made an attempt back in 2013, before switching to focus on the PCE emulator. I recently ported the Basilisk II Classic Macintosh emulator to run in a web browser. Basilisk II Classic Macintosh emulator in the browser.

Basilisk ii