Hi, D. Hugh Redelmeier: > If you double the gfpow table's size you could elminate the % 255.
That was a substantial improvement. It now needs only 10 % more time than the old implementation. I.e. you cut off 80% of the initial performance deficit. One has to unroll the table gfpow[] up to 511 elements because the highest sum of two gflog[] elements is 510. My own optimization idea about Q index computation appears to have been anticipated by gcc -O2 already. But i was able to squeeze out the remaining 10 % by computing LSB and MSB of the 16 bit words in one loop rather than in two. Now it depends mainly on external influences which of both algorithms wins the race. Maybe the old one is by an epsilon faster. Have a nice day :) Thomas -- To UNSUBSCRIBE, email to [email protected] with a subject of "unsubscribe". Trouble? Contact [email protected]
