Actually I have a "better miner" but only a couple percent more and as far as my own card (series 58xx) is concerned. Thing is, I don't know if its better for everyone.
I gave replication instructions earlier, don't know if anyone tried it though:
on sph-sgminer, /kernel/darkcoin.cl, between line 263 and 264 you add #pragma unroll 1
=>
...you then recompile the program, del all .bin files and rerun the program. Report the hashrate and your card.
You can also try this a bit higher (between line 243 and 244) where it says:
and do it like this:
...but try them one by one, not together and see what happens. Always recompile the app + del the .bins.
This is actually "de-optimizing" as unrolled loops are usually faster, but it just so happens to run faster for me, for whatever reason. #pragma unroll 1 declares no loop unrolling (theoretically slower, faster in practice). I'm curious on what it does for others.
I gave replication instructions earlier, don't know if anyone tried it though:
on sph-sgminer, /kernel/darkcoin.cl, between line 263 and 264 you add #pragma unroll 1
=>
Code:
for (unsigned int u = 0; u < 16; u ++)
H[u] ^= xH[u];
#pragma unroll 1
for (unsigned int u = 0; u < 8; u ++)
hash.h8[u] = DEC64E(H[u + 8]);
...you then recompile the program, del all .bin files and rerun the program. Report the hashrate and your card.
You can also try this a bit higher (between line 243 and 244) where it says:
Code:
for (unsigned int u = 0; u < 16; u ++)
g[u] = m[u] ^ H[u];
and do it like this:
Code:
#pragma unroll 1
for (unsigned int u = 0; u < 16; u ++)
g[u] = m[u] ^ H[u];
...but try them one by one, not together and see what happens. Always recompile the app + del the .bins.
This is actually "de-optimizing" as unrolled loops are usually faster, but it just so happens to run faster for me, for whatever reason. #pragma unroll 1 declares no loop unrolling (theoretically slower, faster in practice). I'm curious on what it does for others.
BUMP:
Has anyone tried the above tweak in a series 7xxx / r9 card to check how much faster or slower it might go?
Just follow my message history 
