Re: True 64 bit port (was Re: Projects I may be willing to sponsor)
- From: zejian ju <juzejian@xxxxxxxxx>
- To: luajit@xxxxxxxxxxxxx
- Date: Thu, 11 Dec 2014 10:43:10 +0800
Great news.
The port of arm64 DynASM is finished. All arm64 instructions(including all
SIMD/fp instructions) are supported(although it may be unnecessary for
luajit). I am testing it these days. Port of arm64 interpreter will begin
next week.
2014-12-11 上午7:22于 "Mike Pall" <mike-1412@xxxxxxxxxx>写道：
> Yichun Zhang (agentzh) wrote:
> > Eventually we can further make use of the full 4GB memory once we (or
> > Mike Pall) remove the pointer sign extension limitation from LuaJIT.
>
> A new LJ_GC64 mode with 47 bit pointers will end up in the git
> repo in the next days. This is a requirement for the ARM64 port.
> But that won't benefit the existing ports without substantial
> rewrites.
>
> There's already a contributed port of the x64 interpreter for that
> mode. But it's currently stuck in legal limbo to get it open
> sourced. It needs to be substantially cleaned up and rewritten,
> too. And the x64 JIT compiler backend hasn't been ported, yet.
>
> There's another problem remaining: the bytecode that needs to be
> generated for that mode has a different frame size, which makes it
> incompatible with the existing 32 bit mode -- two 47 bit pointers
> simply don't fit into one 64 bit stack slot. I'm not sure how to
> handle the bytecode incompatibility, yet.
>
> --Mike
>
>
Other related posts:
