On Aug 23, 2011, at 11:06 31PM, Nicolas Cellier wrote:

> 2011/8/23 Stéphane Ducasse <[email protected]>:
>> BTW igor and henrik are coding like nuts floating point support with 
>> nativeboost. No need to have a float plugin.
>> They seems to have fun like kids with a cool toy :).
>> 
>> Stef
>> 
> 
> Cool, they might add missing libm functions but also control float
> rounding modes, trap exceptions and so on...
> 
> Nicolas

That is definitely one direction that would be worth persuing.
While in the back of my head  when I started (due to reading the Float chapter, 
+ vmdev discussions about nonsense float-related compiler flags), the main 
motivations for it were:
 
- Showing a use of NativeBoost that isn't Just Another FFI (tm)
- Figuring out how to structure NB code with multiple paths in a good way 
(Floats were perfect for this due to SSE2/x87 available on the same platform)
- Providing an actual usecase where coordination between NB and Cog could be 
useful.

And that's basically the 10 min talk I signed up for summed up, so those who 
read it  and have no interest, feel free to leave early :)

Have had to implement missing assembling of x87 and SSE2 machine code in 
AsmJit, and write some useful ObjectFormat-related macros for checks, so gotten 
a bit  shorter than I'd like in two afternoons...
So so far we only have running x87-replacements for most of the standard 
primitives.

Cheers,
Henry

Reply via email to