[Libre-soc-dev] things we need for fast qemu

[Jacob Lifshay]

On Wed, Apr 6, 2022, 15:59 Jacob Lifshay <programmerjake at gmail.com> wrote:

> stuff we mentioned in today's meeting:
> tso memory model
> coherent icache because qemu rewrites branch targets a bunch
> fully programmable radix mmu -- avoiding the need to sw emulate the mmu
> qemu ir peephole optimizations
> emulating interrupts (probably not needed)
>
> stuff i just now thought of:
> fast trap into qemu on modification of guest's instructions aka.
> user-space interrupt handler skipping kernel mode -- page fault interrupt
> may be too slow
>

after some time digging through qemu's git repo, i discovered that its tcg
compiler (used for basically all emulation) has no support for
floating-point ops (though it apparently used to?), instead it translates
all fp ops in the frontend into calls to helper functions ... fixing this
should give major speedups.

Jacob

>