Pull in improved kernels about sgminer HOT 15 CLOSED

Looks like the Zuikkis modifications are only available in .bin format, which isn't very useful since it can't be inspected....

from sgminer.

Actually, it's detailed here
https://litecointalk.org/index.php?topic=6058.msg92691#msg92691
And here
http://www.reddit.com/r/dogecoin/comments/1ui3bx/increase_such_hashrate_1_to_5_scrypt_tweaking/

from sgminer.

This is the original optimized script file built by LANTIS:
https://litecointalk.org/index.php?topic=6020.0

Script included. However the way his works is values like TC and GAP LOOKUP are hard-coded in the script so its not universal.

from sgminer.

Yup, I've noticed the hard-coded stuff. Are there any other changes, or is that it? Including 2^N files for all possible configurations, especially binary kernels, is unacceptible.

EDIT: I'll only get to faithfully review it once some of the more pressing issues are resolved.

I'm not following the forums that much, so if there are any other kernels, open up new issues or pull requests, by all means.

from sgminer.

Here's another one https://bitcointalk.org/index.php?topic=369858.0

It's going to be tough sorting through versions.

I suspect that at least there may be a break in either the driver versions, or hardware versions.

from sgminer.

Most of these optimized kernels (including mine), have fixed lookup-gap=2. However, I have never seen anyone use any other value, for any GPU, so I think you could just remove the configurable value.

Or with some #if LOOKUP_GAP==2 magic it is of course possible to make such source that allows any value.

Lantis code is problematic. If you take his released sources and compile them, you end up with slower hashrate than stock. However binaries generated by him are much faster. I assume he had some compiler parameter tweaking which is not showing in the source?

Some users have reported slightly slower hashrate with my kernel as well, but this could be some misconfiguration also.. If scrypt kernel becomes faster, you may need to lower the GPU engine clock to get full speed. Same as if you increase GPU clock too high, you will get a drop in hash rate.

My source is free to use in sgminer. And if you diff to original you will see that the changes are not very big.

Removing of #pragma unrolls helps in any GPU, in my opinion.. Current compilers know better when unrolling helps.

from sgminer.

@Zuikkis: I've been looking over the different kernels, but just haven't had the time to organise them properly. Thanks for outlining your changes, and the reason why you made them.

The good thing is that we don't have to have just one kernel, we can easily have them all (as long as there are sources) and use the kernel switching functionality that has remained from cgminer.

from sgminer.

Removing #pragma unroll improved my 7850's by a few percent but reduced a 7550 and a 6870 by a few percent

from sgminer.

Yes, it seems to help a little with low end cards, but hurts on higher end.

Try my scrypt main loop change as well?

2014/1/20 Gabriel Devenyi [email protected]

Removing #pragma unroll improved my 7850's by a few percent but reduced a
7550 and a 6870 by a few percent

—
Reply to this email directly or view it on GitHubhttps://github.com/veox/sgminer/issues/4#issuecomment-32792055
.

Teemu Suikki
http://www.ksportfinland.com
http://www.z-power.fi/

from sgminer.

Technically, my 7850's are high end (~400kHash) compared to the 6870 (320kHash) and 7550 (~120kHash)

from sgminer.

The first two commits above have been rebased, so just use the third in branch kernels. Commit history.

from sgminer.

OTOH, we can have some hard-coded stuff, same way that vectors is now hard-coded to 1. Worksize and lookup gap are not as important as thread-concurrency. It can probably "be improved later".

from sgminer.

agreed, as far as I know, worksize and lookup gap arent really changed from config to config.

from sgminer.

Okay, I suggest we start splitting off this issue into one per proposed "major" kernel modification, to discuss their merits and drawbacks, as well as their best implementation.

from sgminer.

@gdevenyi OK, done.

The original scrypt kernel (that's now called ckolivas.cl, after the latest committer) should probably stay the default for now, to avoid unexpected behaviour.

Closing issue, reopen if something remained unanswered.

from sgminer.