Too bad it targets ARM and not rv64gc, but it's a start.
I'm curious why it isn't based on qemu-user.
Too bad it targets ARM and not rv64gc, but it's a start.
I'm curious why it isn't based on qemu-user.
It’s bit like arm and x86 15 years ago. There is not even a contest between them. Only in the later years with the new 64bit arm cores it’s getting there, heck the latest Apple chips actually beat Intel ones in IPC.
EDIT: The emulator doesn’t have anything arm specific. It will work on anything 32bit and little-endian.
I've also wondered what it would take to do multi-architecture dynamic linking in qemu-user. Such a capability would avoid one shortcoming of Box86, the current lack of dynamic recompilation.
I suspect that modifying QEMU in this way is not trivial.
Is this referring specifically to their in-order cores (Cortex-A53, A35), out-of-order cores (Cortex-A72, A73), or big.LITTLE configurations in general? *
Thinking more broadly, the next era of power-efficient computing may depend more on heterogeneous architectures than the CPU alone. The Arm ML Processor IP and corresponding offerings from competitors play a large role in this.
* Can be generalized to custom cores designed by NVIDIA, Qualcomm, Samsung, Apple, and other vendors.