Re: Patch to use colored NaNs on LP64

[Florian Weimer <fw@...>]

* Mike Pall:

> Florian Weimer wrote:
>> Here's an experimental patch to shrink the size of a Lua value to 64
>> bit on LP64 architectures, by encoding non-double values as special
>> NaNs.  I think this was initially suggested by Mike Pall, but I'm not
>> sure if it's been implemented yet.
>
> Yes, it will be in LuaJIT 2.x. But as I've mentioned in the past
> it doesn't pay off for the Lua interpreter:
>
>   http://lua-users.org/lists/lua-l/2008-02/msg00093.html
>
> Apart from the portability problem (the popular targets are ILP32,
> LP64 and LLP64) your patch is lacking the important tricks to make
> it fast. I've disclosed a couple of them in a bug report for Tamarin:

I guess most of the speed-up comes from improved locality due to
smaller object size anyway.  I can't use the 32/32 split because I
need 48 bits for pointers, and as a side-effect, I bypass the store
forwarding issues (I think, at the very least I always store and load
64 bit quantities).

Do you see a larger speed-up than 10% (with heap-intense loads)?