Re: Proposal for fixing race conditions

Den 2013-06- 11:42:07 skrev Robert O'Callahan <robert@ocallahan.org>:

> On Fri, Jun 28, 2013 at 12:06 PM, Chris Rogers <crogers@google.com>  
> wrote:
>> Additionally, it's important to keep in mind the performance hit of any  
>> memory copies.  Performance is very very important.
>
> I don't think it's critical that the Web Audio content *in use today*  
> have optimal performance. If we introduce new APIs --- soon --- that  
> support minimal >memory copying, and steer authors towards them, then  
> that should be good enough. I have a strong desire to support legacy  
> content, but I don't think >it's very important for that content to be  
> maximally efficient.

I agree (with both of you ;) ). Performance is one of the key design  
principles of the Web Audio API, and should continue to be so. At the same  
time, I only think it's natural for new usage patterns & interfaces to be  
prioritized in terms of optimal performance. Take OpenGL for instance,  
which has a history full of design optimized for newer interfaces but  
still having [suboptimal] backwards compatibility (e.g. display lists from  
OpenGL 1.0 were really fast in the days, but superseded in performance by  
VBOs at some point in time).

In general I really like the concept of immutable buffers, because that  
opens up for so many ways to optimize - even in ways we wouldn't conceive  
right now (such as dedicated hardware & memory, e.g. as found in the Cell  
architecture in the PS3, etc).

-- 
Marcus Geelnard
Technical Lead, Mobile Infrastructure
Opera Software