Quote from: ssvb on January 19, 2012, 07:13:34 PM

OK, thanks for the clarifications. Now we can see that it was just scaremongering and/or handwaving. So much for "do less calculations" and "sequential memory access" promises. In reality the demonstrated code snippet has more XOR operations and does three separate passes over 128K buffer instead of two (which means more L1 cache misses unless the hardware prefetcher saves the day). The pointless replacement of left rotations with right rotations also looks amusing.

Well... There are some explanations:

Code I posted here is C++ edition of the algo (for better reading). CPU runs assembler code which looks completely different and u should know this unless u r just a coin trader. Ur post looks like an attempt to defend ArtForz' position. But everyone sees that he is just upset that there are other miners which are better then his one (pooler's, mine and I know 2 other guys who developed much better miners).

My implementation uses less calculations coz CPU process 3 instructions at once using a trick to avoid register stalls. So extra cycle between 2 others is necessary for "less calculations" and "sequential memory access". Btw, a CPU with 128K L1 per core is quite... uncommon and ur statement about 3 passes is wrong.

Regarding replacement of left rotations with right ones. I suspect that u r NOT serious. Even a young programmer won't belive that this was made for optimizations or something like that. I made it coz there was already defined function for calculating SHA-256 that used right rotation.

PS: Ask more if u have questions