Re: [Pearpc-devel] amd64 in cvs

[Cassy \SnowGirl\ <pue...@go...>]

2006/10/26, Sebastian Biallas <sb...@bi...>:
>
> Current status of AMD64:
>
> - - it compiles :)
> - - float and vector jit disabled
> - - no longer gcpu and gjitc (copied to the stack)
>
> - - And a problem:
>   Since near calls can only reach addresses in the +- 2 GiB range, the
> translation cache should be that near to the normal code. Unfortunately
> mmap reserve memory far far away from the normal code.
>
> And I don't know how to tell mmap where I want to alloc. The hint is
> only used when MAP_FIXED is used (it seems) and when using MAP_FIXED
> mmap destroys already mapped and important addresses.
>
> Any idea how to get around these mmap limitations?



Wouldn't it be possible to move an address into a register and jump to that
address?  So, rather than using a near call to get to the translation cache
use a form of far call.  Is there a reason why we need the near calls to
access the translation cache?

-- 
Cassy