And what about the DSP ?

[SladeCraven]

As i received my pandora 3 weeks ago, i started to develop/port some stuff, just to bring me in the toolchain and the device (see my ongoing attempts around fMSX - - and the Thomson TO8D emulator - -).


Up to now, all the software delivered in the repo/boards/... seems to only use the ARM and never the integrated DSP.


So my question is : is there someone around that have played with the DSP and really compiled software on it ? If yes, is there a tutorial/guide on the way ?

 

[doragasu]

The last time I saw someone asking about this, the general answer was something like: "Well, if you want stuff using the DSP, do it yourself".


AFAIK, the problem the devs argue is that using the DSP requires to reserve a block of memory when the machine boots. This block of memory can only be used by the DSP, and is useless for any other task non DSP related. I think the benefits of having the DSP resources are worth enough to sacrifice for example 8 or 16 MiB RAM, but it looks like the devs with the required capabilities to make the DSP work don't think the same and aren't motivated to compile the necessary modules.

 

[mcobit]

There was a module for using the dsp posted a while ago, can't find it though.

 

[gunrock]

http://boards.openpandora.org/index.php?/topic/5285-dsplink/


No news is good news. I'm sure Ari64, Notaz or WizardStan will have something soon-ish.

 

[milkshake]

yes the module does exist and has been compiled, there is a thread somewhere about but i cannot find it also.

edit: ninjad well done gunrock also I belive it eats 32mb or memory

 

[fusion_power]

I doubt TI has designed the OMAP in a way that a DSP eats this memory all the time. I hope they were clever enough to adress the Memory to the DSP only when it is needed. I mean it would be really stupid otherwise. :huh:

 

[milkshake]

Fusion_power unfortunately they were not that smart.

 

[fusion_power]

Fusion_power unfortunately they were not that smart.

:blink:


How do other OMAP devices with DSP solve this problem? DO they solve it at all?

 

[Exophase]

http://boards.openpa...c/5285-dsplink/


No news is good news. I'm sure Ari64, Notaz or WizardStan will have something soon-ish.

No news is good if you were expecting bad news ;p Or maybe you're saying that it's good no one has come forward and vowed to never use the DSP.


I want to see something too but emulators are hard enough to make good use of threading them without all of the major hassles a heterogeneous system with incoherent memory provides.

I doubt TI has designed the OMAP in a way that a DSP eats this memory all the time. I hope they were clever enough to adress the Memory to the DSP only when it is needed. I mean it would be really stupid otherwise. :huh:

Fusion_power unfortunately they were not that smart.

It's not a hardware issue at all, it's 100% OS/software related. I believe that the module notaz provided has dynamic memory allocation.

 

[Darkknight512]

What are the things that are suited to be run on the DSP? I know video decoding would benefit nicely, but what else? We have to have something substantial to make up for 32 mb less ram. The only thing I think is craving more power trying to run on the Pandora is N64 emulation, will the DSP help with that if Ari recodes the emulator to use it?

 

[SladeCraven]

In addition of video decoding, I would say : color conversions, blending, audio decoding, resizing/rotation, special effects (?) ...

 

[mcobit]

Could help with n64 if audio is offloaded to the dsp maybe.


But the thing, that needs most optimisation there might be the renderingplugi.

 

[calc84maniac]

Didn't I hear discussion a few years ago that mp3 playback using only the DSP could theoretically run for 100 hours?

 

[Linux-SWAT]

Didn't I hear discussion a few years ago that mp3 playback using only the DSP could theoretically run for 100 hours?

Great if it's possible !

 

[SladeCraven]

Usually, 1 data processing DSP instruction is equivalent to X RISC32 instructions. So as far as you are doing digital signal algorithm, the DSP is more efficient, leading to powersaving.

 

[milkshake]

will the dsp help us with web browser HW acceloration? or at least with html5 video or flash ?

 

[fusion_power]

Didn't I hear discussion a few years ago that mp3 playback using only the DSP could theoretically run for 100 hours?

Craig mentioned this but I guess it was more or less the usual marketing speach. It seems more and more unlikeley that we ever get DSP or the Powersavre Functions of the OMAP to run like it was designed to run. For ~100hours of mp3 there would be needed some serious deep digging into the OMAP functions I guess.

 

[SladeCraven]

In the past, i was using the BeagleBoard in some projects, and we made some tests with the codec packages from TI. I just crawl again in the TI website (which is more like a digital maze of death) and i get those pages :


http://www.ti.com/lsds/ti/dsp/support/asp_overview.page


http://www.ti.com/tool/omap35xcodecs


Someone already tried it ?

 

[gunrock]

In the past, i was using the BeagleBoard in some projects, and we made some tests with the codec packages from TI. I just crawl again in the TI website (which is more like a digital maze of death) and i get those pages : http://www.ti.com/lsds/ti/dsp/support/asp_overview.page http://www.ti.com/tool/omap35xcodecs Someone already tried it ?

The list of codecs is impressive.


http://www.ti.com/lsds/ti/dsp/support/asp_overview.page


@Exophase: I know what you mean, but I have some faith that the prospect of it being "impossible" will spur on the developers to produce something that proves or disproves the usefulness of the DSP in the Pandora. So I don't necessarily expect pcsx-rearmed or Mupen with DSP support, but perhaps a proof of concept command-line mp3 player. Alternatively they may prove it won't work until a later kernel or without some kernel hacking, but at least the argument will be settled.

 

[notaz]

Someone already tried it ?

I'm pretty sure nobody did, or could not get usable results at least.