> If simply comparing two algorithms by the number of operations needed to > compute a sample, would you include delays in filters as an operation? I'm > just wondering as some papers about FFT only include real multiplications > and additions as operations. >
It depends whether you are conducting an academic study or a real-world engineering. In the very past you had to know how much time a MUL operations required compared to a SUM, nowadays they are very similar as they are hardware implemented, even on microcontrollers such as Arduino's Atmel. Anyway, for academic stuff what I usually do (based on what currently people do) is summing together all real sums and multiplications into one number. If you have complex data operations, then 1 complex SUM = 2 real SUM, 1 complex MUL = 4 real MUL + 2 real SUM. As for divisions you can either indicate them as another count, or put them together with SUM and MUL in the global flops count. This way, however you can't be too abstract and have to take some reference values. I take the instruction manual of the target architecture and give an estimate of the number of clock cycles, then compare this to SUM or MUL clock cycles. Finally, for the delays, sometimes I indicate the size of the memory to allocate. Differently, if you are working on an engineering project, you should really take care of all pointer arithmetics, value copying, memory swapping. These can usually go in parallel with the arithmetic operations (if your code is nice) but you never know. Memory operations have a severe effect on embedded systems as the system can stall for many cycles waiting for the data to come. Definitely divisions are a nightmare on embedded processors and you should just try to avoid them at all costs (unless you are using a whole DSP for a digital effect). On x86 forget all the above and write your code as ugly as possible, you'll never incur in bottlenecks :P :P :P (joking, but really you have no big issues with divisions and memory ops there). Hope this helps. _______________________________________________ dupswapdrop: music-dsp mailing list music-dsp@music.columbia.edu https://lists.columbia.edu/mailman/listinfo/music-dsp
