Re: nfloat on x86-64

[Aleksey Demakov]

Hi Jakob,

On Mon, Jan 13, 2020 at 4:34 PM Jakob Löw <address@hidden> wrote:

> Btw math on nfloats isnt implemented either and relies on intrinsic
> calls, so a third option maybe would be to set nfloat == float64 and
> remove all support for the FPU. The i686 instruction set supports SSE
> too, which would allow us to remove all FPU related code. Maybe thats
> just me, but i never liked the stack based register of the FPU, as they
> add more complexity to a compiler. I guess that would mean we loose
> support for native i386, which afaik gcc did a while ago?
>

I think that removing FPU support and doing all fp with SSE is fine for
the x86-64 backend. But I'd also keep i386 support for a while. I think
that one possible way to do it is to have the new register allocator
optional for a while as well. That is we need an option to switch it on/off.

This is also important form another point of view. The new allocator is
too big a change to make it all at once. I'm afraid it might destabilize
libjit for indefinite period of time.

I'd suggest, if you have the time, to start submitting small PRs that
add new functionality in this order:

1. Bug fixes that apply both to master and your reg alloc branch.
2. Support of an option that switches between the old and new allocators.
3. The bulk of the new allocator.
4. Then probably for a long time extensions and bug fixes to it as
needed until it stabilizes.
5. At some point later we could get rid of the old allocator.

Regards,
Aleksey