vDSP_biquadD crash in 32-bits?

[Michael Norris]

Hi there

(Hopefully this is the correct forum for my question — apologies if not.)

I'm authoring a plug-in that uses the vDSP_biquadD method from the Accelerate framework. It works fine when the host is running in 64-bit mode, but crashes when running in 32-bit. All my other vDSP routines run fine. Is this a known issue?

I’m running Mac OS X 10.12.1.

Cheers
Michael
_______________________________________________
Do not post admin requests to the list. They will be ignored.
PerfOptimization-dev mailing list (PerfOptimi...@lists.apple.com)
Help/Unsubscribe/Update your Subscription:
https://lists.apple.com/mailman/options/perfoptimization-dev/perfoptimization-dev-garchive-8409%40googlegroups.com

This email sent to perfoptimization-...@googlegroups.com

[Michael Norris]

Thanks — yes, I have extensively checked all of the expected pointer sizes, setup structure addresses and other initialization factors, and they all look as expected.

FYI, commenting out the vDSP_biquadD allows operation to proceed as normal in 32-bit (albeit without the biquad filter), even though the same pointer structures are used later by other vDSP routines.

Michael

> On 21/11/2016, at 1:39 PM, Eric Postpischil <PerfOptimization-d...@edp.org> wrote:

>
> On Nov 20, 2016, at 17:44, Michael Norris <michael...@xtra.co.nz> wrote:
>
>> I'm authoring a plug-in that uses the vDSP_biquadD method from the Accelerate framework. It works fine when the host is running in 64-bit mode, but crashes when running in 32-bit. All my other vDSP routines run fine. Is this a known issue?
>

> FYI, Apple is largely shut down this week for a holiday (US Thanksgiving). I do not know of such a crash offhand. However, diagnosing it de novo, my first thought would be that running in 32-bit mode instead of 64-bit naturally changes the memory layout in various ways, and a buffer overrun that might have been harmless (or caused only unobserved harm) in 64-bit mode might hit unmapped memory in 32-bit mode, thus causing a crash. I would carefully check the actual values passed to vDSP_biquadD and its setup routines.
>
> — edp (Eric Postpischil)
> http://edp.org
>
> “I would rather have questions that can’t be answered than answers that can’t be questioned.” — Richard Feynman, aprocyphal