Re: [ANN] sgminer v5 - new unified multi-algorithm on-the-fly kernel switching miner

Quote from: Zuikkis on November 12, 2014, 05:31:56 AM

Quote from: badman74 on November 12, 2014, 05:21:35 AM

for anyone that is still having issues with HW errors on various algo's try my latest binary
i think i may have found the problem but would like others to test and see if it is fixed
also, arebyp kernel can now be used for nscrypt although it seems to put out less hashrate than zuikkis, though it produces less errors for me

Hi,

I also fixed it in my local build, is this the same you found?

This fixed HW errors 100% for me.

sgminer.c, function test_nonce, near line 6873:

Code:

// for Neoscrypt, the diff1targe value is in work->target
  if ((work->pool->algorithm.name, "neoscrypt")) {
   diff1targ = ((uint32_t *)work->target)[7];
  } else {
   diff1targ = work->pool->algorithm.diff1targ;
  }

Should be like this, just add !safe_cmp:

Code:

// for Neoscrypt, the diff1targe value is in work->target
  if (!safe_cmp(work->pool->algorithm.name, "neoscrypt")) {
   diff1targ = ((uint32_t *)work->target)[7];
  } else {
   diff1targ = work->pool->algorithm.diff1targ;
  }

The original version is really a no-op, always true. So neoscrypt code was used for all algorithms.

Thank you so much for pointing this out. (Although it would have been nice to get a pull request or a notice on github lol...)

I must have pushed the wrong file up originally and when debugging the issue I was using corrected code. No wonder I didn't get HW errors. Chasing a bug that isn't there is can be quite a headache...

Anyway, I pushed the fix to the develop branch. Can anybody try it and confirm that the HW errors are gone?

- ystarnaud