I've ported the code to Alpha NT and discovered one thing I thought
worth sharing. I'm using VCC++ 5.0. My first attempt failed on
execution with a "floating point invalid operation" exception. It turned out
to be caused by an uninitialized data structure which contained some
non-numbers (NaN), which triggered the exception. Simply initializing the
data to zero fixed the problem, and the results on Alpha are quite good.
I'll post some comparitive results later.
Anyway in l3subband.c I changed the definition of l3_sb_sample to:
double l3_sb_sample[2][3][18][SBLIMIT] = { (2*3*18*SBLIMIT)*0 };
/maurice (Compaq)
