Content deleted Content added
Quuxplusone (talk | contribs) m →Implementations: oops Arvindn |
→Implementation in assembly: new section |
||
Line 95:
:I think you did the right thing. I just restored the C and ARM implementations after [[User:Arvindn]] blanked them, for these reasons: IMHO, C isn't important for showing bitwise operations; it should be obvious that you'd use bitwise ops where appropriate. But the C implementation is easier to read than the English algorithm at the moment, so it needs to stay at least until there's an appropriate substitute ([[pseudocode]]? Knuth-style algorithm description?). IMHO, the ARM implementation is really enlightening, because it shows the real size and speed benefit of the algorithm in a way that no higher-level language's compiler even approaches. (In particular, it totally stomps the C implementation.) Without ''some'' hand-optimized assembly implementation, the advantage of binary GCD over Euclid is not obvious, IMHO. --[[User:Quuxplusone|Quuxplusone]] 21:18, 1 January 2007 (UTC)
== Implementation in assembly ==
Thanks [[User:Balabiot|Balabiot]] for pointing out the incorrect result (always zero) when u=0 or v=0. I have fixed the code to return the sum of the inputs when either or both are zero. The code now matches the output of the Euclidean algorithm.
I have reverted movne/movne to movs/movnes to ensure that r0 and r1 are both nonzero on entry to check_two_r0. (The previous code hung e.g. with u=0, v=3.)
I didn't understand why the case where u and v are powers of two needed fixing, could you please elaborate? -- [[User:Regregex|Regregex]] ([[User talk:Regregex|talk]]) 17:02, 10 October 2008 (UTC)
| |||