Have someone implemented watershed algorithm on the DSP?

Hi,
Before my job is to implement the video codec on the DSP.
Nowadays I was required to do image processing on DSP.
So I am trying to implement watershed algorithm on C64x+.
After reading some paper, I summarize the following:
there are popular three implementations.
1. Vincent and Soille's implementation based on immersion
2. Meyer's implementation based on topographic distance
3. watershed based on rainfalling

Watershed algorithm is a sequential algorithm. In the
implementation some data structures such queue or
ordered queue are used, but prevent the optimization
of DSP. On C64x+, memory operation needs
more cycles, so I doubt memory is the bottleneck.
From my limited image processing experience, image
processing needs little arithmetic computation, but
needs much memory access, especially access
pattern not raster pattern.

For a image of 1280 by 1024, it needs more than 3
seconds on a DSP of 600M. Does it make sense?

Thanks in advance.
Jogging Start a New Thread