Re: LuaJIT roadmap 2008 question

["Alex Davies" <alex.mania@...>]

Perhaps.  There's at least one other advantage - in a register poor 
architecture like x86, it allows a function to have full access to its 
callinfo without reserving a second register to point to the CallInfo stack 
(assuming a register is already reserved for the TValue stack, which makes 
sense).  And even on architectures with more registers they then have to 
worry then about keeping them valid through growths/shrinks.

It would also allow just one bounds test on entering a function, vs the 
current two.  In Jit every instruction counts... something as minor as that 
will show up on recursion tests.

Definitely looking forward to playing around with it when it comes out ;)

- Alex

----- Original Message ----- 
From: "tankxx" <tankxx@gmail.com>
To: "Lua list" <lua@bazar2.conectiva.com.br>
Sent: Thursday, October 29, 2009 12:35 PM
Subject: Re: LuaJIT roadmap 2008 question


> I cannot figure out why this speeds up call handling substantially.
> May it reduce cache misses?