It'd be good if people at least waited until the DSP on Pandora was used for something substantial before hyping C64x on Pandora 2..
There already is a sprite/blitter engine for the Pandora DSP that (now) is (almost exactly) twice as fast as the GPU and faster than any ARM/NEON blitter but apparently there is little interest in using it (
a tweaked version of SDL could help with that. Any volunteers?)
(..I'm also working on some things which look really promising so far..)
The DSP on the OMAP5, according to specs, does not seem to be an improvement over the one in the Pandora. It _may_ even be slower since TI removed some performance-critical features
(only 4 execution units instead of 8, no software pipelined loops, and the L1DSRAM can not be memory mapped any longer (but L2 still can..and there's this L1 cache maintenance/lockdown feature..)). Last but not least, it's not binary compatible to the OMAP3 DSP.
However, if the "P2" will be based on OMAP5, I'm definitely on board. The DSP may not be the best choice for 2D graphics, anymore, but that does not mean that it will be useless. Plus there are two A15 processors, one of them could always take on the role of the Pandora DSP (
maybe even at approximately the same power consumption level).
The main benefits of a "P2" will most likely be GPU speed(!), increased resolution, and improved connectivity (
HDMI/USB3_or_SATA/..).
Sticking to OMAP5 definitely makes sense since the scarce low level/kernel developer resources in this community (
*cough*, Notaz, *cough*) are already familiar with the OMAP family, and it retains important HW features (
e.g. display controller overlays / video scaler).
p.s.: regarding "hard to use DSP and so on": Funny you should say that because in reality things are quite different: ARM/NEON is hard to use (read what Exophase wrote about this topic! i.e. NEON is poorly documented and manual cycle-counting and assembly programming is mandatory) while, in my experience, compiling standard ANSI-C sources with the TI DSP compiler already results in very good performance that is mostly limited by the available memory bandwidth.