Re: vDSP and iOS render callback
Re: vDSP and iOS render callback
- Subject: Re: vDSP and iOS render callback
- From: Rick Mann <email@hidden>
- Date: Wed, 25 Apr 2012 13:55:08 -0700
I would suggest vDSP, unless it actually causes a problem today. You'll get automatic improvements over time as Apple releases faster hardware and updates vDSP.
On Apr 25, 2012, at 13:42 , Kevin Dixon wrote:
> Ok great.
> So let's say I want to perform an FFT in my Render callback.
>
> Which option would be more performant?
>
> 1. Set my client type to Float32, and use vDSP
> 2. use the canonical type 8.24, and use a NEON-optimized FFT
> implementation (which I already have on hand)
>
> Thanks!
>
> -Kevin
>
> On Tue, Apr 24, 2012 at 8:23 PM, Brian Willoughby <email@hidden> wrote:
>>
>> On Apr 24, 2012, at 18:25, Paul Davis wrote:
>>>
>>> On Mon, Apr 23, 2012 at 4:16 PM, Kevin Dixon <email@hidden>
>>> wrote:
>>>>
>>>> Maybe I'm missing something, I was looking at this question stack
>>>> overflow
>>>> (http://stackoverflow.com/questions/3398753/using-the-apple-fft-and-accelerate-framework/3534926#3534926).
>>>> I was under the impression that the render callback supplied sample
>>>> data in 8.24 fixed format, yet I don't see any sort of type conversion
>>>> going on in the example autocorrelation implementation when they call
>>>> the FFT.
>>>> Can anyone shed some light on this?
>>>
>>>
>>> this is wrong. CoreAudio (like most audio SDKs) uses 32 bit floating
>>> point as its only "native" data type. this absolutely includes the
>>> render callback for an audio unit. unless iOS is totally different, in
>>> which case forgive my emphasis.
>>
>>
>>
>> Hey, Paul, the second thing you said is true: iOS is totally different.
>>
>> CoreAudio for OSX does default to Float32 unless the code specifically
>> requests otherwise.
>>
>> CoreAudio for iOS defaults to Q8.24, a fixed-point format that is more
>> efficient on the older ARM processors without fast float support. In that
>> world, I/O devices are 16-bit integer, and DSP is run at 8.24 for speed.
>>
>> The confusing thing is that the ARM chips are now getting to where they
>> support float as well as int, and so CoreAudio/iOS is moving towards
>> Float32.
>>
>> Brian Willoughby
>> Sound Consulting
>>
>
> _______________________________________________
> Do not post admin requests to the list. They will be ignored.
> Coreaudio-api mailing list (email@hidden)
> Help/Unsubscribe/Update your Subscription:
>
> This email sent to email@hidden
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Coreaudio-api mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden