[Nauty-list] nauty performance: vectorization, parallel

[bootch at nc.rr.com]

I am interested in speeding up nauty.

Is there a bottleneck in nauty, a routine to concentrate speedup efforts?

Does the code vectorize for Intel i86 SSE vector instructions?

Does the code compile for Cell SPE processors (parallel engines on the PS3 platform)?

I would appreciate any references to the literature.

>From my limited understanding of the algorithm, nauty is NOT compute intensive in the usual scientific way (floating point formulas on large data sets) but is compute intensive in a business way (integer operations on sets and sorting.)

My exploration so far has been to use many copies of nauty in an MPI architecture.
I am brute-force generating graphs and checking for isomorphic duplicates.
I am wondering about other approaches: vectorizing or using the Sony/IBM Cell BPE.