GPUs are fast, software emulation is slow - what happens if both get combined? I don't know either, this project is an attempt at finding it out.
Neither execution nor memory models fit, so this can be a challenge.
It probably turns out to be incredibly slow, but still, Linux running on a graphics card sounds amazing.
The pre-hackweek state is that the execution reaches systemd until it tries to execute not yet implemented floating point instructions.
Target emulated hardware is a basic Cortex-A53-compatible SMP system with UART, PL011 UART and GICv2 INTC.
Maybe there'll be a AArch64 -> OpenCL JIT in the host as well.
This project is one of its kind!