Lack of hardware support [LWN.net]

Lack of hardware support

Posted Jul 6, 2009 21:41 UTC (Mon) by midg3t (guest, #30998)
Parent article: Ogg codecs dropped from HTML5

Can anyone expand on Apple's citation of "lack of hardware support". Is it impossible to port the Theora decoder to some architecture that Apple uses?

to post comments

Lack of hardware support

Posted Jul 6, 2009 22:01 UTC (Mon) by joedrew (guest, #828) [Link] (9 responses)

There is currently no hardware implementation of a Theora decoder, which means that the CPU has to be used, which drains battery life faster.

Lack of hardware support

Posted Jul 6, 2009 22:53 UTC (Mon) by timschmidt (guest, #38269) [Link]

That's a strawman argument... true hardware en/decoders rarely exist at all anymore. What _is_ common are general-purpose DSPs running codec-specific firmware. Supporting Vorbis or Theora is strictly a software problem, and Apple knows this.

Lack of hardware support

Posted Jul 6, 2009 22:56 UTC (Mon) by leoc (guest, #39773) [Link]

Apple has their own in house chip design capability.

Lack of hardware support

Posted Jul 6, 2009 23:14 UTC (Mon) by moxfyre (guest, #13847) [Link] (6 responses)

There is currently no hardware implementation of a Theora decoder, which means that the CPU has to be used, which drains battery life faster.

Indeed, but that's irrelevant.

Basically no one does hardware decoding of audio or video anymore. The iPod/iPhone? Software decoding on an ARM-core processor. TiVo? Software decoding on x86 I think. Sandisk MP3 players? Software decoding on ARM. Some DVRs? Software decoding on MIPS.

Software decoding is sooooo much more flexible (allowing bugfixes, new codecs, etc.) and faster-to-market that there's no reason to use hardware decoders anymore... now that it only costs a few $ to integrate a 100MHz ARM-core system-on-a-chip into a product.

It's true that hardware acceleration of mathematical operations is still important to efficient software decoding. Which is why all the major processor families offer SIMD instruction sets for fast, parallelizable math operations (e.g. SSE1/2/3/4 for x86, NEON for ARM, AltiVec for PPC, MDMX for MIPS).

The Theora encoders/decoders aren't (yet) totally optimized for all hardware, but that's cause it's a volunteer-run project with very little financial support. I personally patched Theora to use SIMD instruction sets on the x86_64 architecture, and it was about a day's work with no previous experience on the project. So Theora implementations can get a lot more efficient--quickly--if the adoption is there. The power of open source :-)

Lack of hardware support

Posted Jul 7, 2009 0:27 UTC (Tue) by pphaneuf (guest, #23480) [Link]

You need quite the solid machine to decode 1080p HD content. See the other recent article about VA API, for example, and the strong return of all sorts of video acceleration hardware to graphic cards.

Lack of hardware support

Posted Jul 7, 2009 0:39 UTC (Tue) by joedrew (guest, #828) [Link] (3 responses)

> Basically no one does hardware decoding of audio or video anymore. The iPod/iPhone? Software decoding on an ARM-core processor.

Wrong - the iPod has hardware for decoding AAC, MP3 and H.264. They *also* have a general-purpose ARM core.

Lack of hardware support

Posted Jul 7, 2009 4:58 UTC (Tue) by gmaxwell (guest, #30048) [Link]

An arm core which is significantly faster than is required to decode Theora at the resolution of the iphone screen

Okay, so perhaps it would have lower battery life on that device. And? The proposal is a baseline. If the user and the content provider prefer something else, they are free to use something else.

Lack of hardware support

Posted Jul 7, 2009 11:16 UTC (Tue) by robert_s (subscriber, #42402) [Link] (1 responses)

"Wrong - the iPod has hardware for decoding AAC, MP3 and H.264."

This 'hardware' will actually be a relatively general purpose DSP which is almost certainly reprogrammable to decode theora.

Lack of hardware support

Posted Jul 7, 2009 17:57 UTC (Tue) by joedrew (guest, #828) [Link]

> This 'hardware' will actually be a relatively general purpose DSP which is almost certainly reprogrammable to decode theora.

Nope - it's a specialized block of functionality added to Apple's Samsung-fabbed ASIC, the chip that also includes the ARM core.

Some iPods, the iPod video in particular, did include a DSP, but all recent ones are specialized hardware.

software support right here

Posted Jul 7, 2009 1:26 UTC (Tue) by xoddam (subscriber, #2322) [Link]

> I personally patched Theora to use SIMD instruction sets on the x86_64 architecture, and it was about a day's work with no previous experience on the project.

Thankyou; your work is very much appreciated.