[LOCKED] cpuminer-opt v3.12.3, open source optimized multi-algo CPU miner

[joblo]

I've just uploaded cpuminer-opt 3.4.7 to git

https://github.com/JayDDee/cpuminer-opt

Yay!   
Now all my Linux miners will get faster updates!

Thanks joblo

I'm only uploading releases for now, still doing all dev offline.

[1714024923]

1714024923

[1714024923]

1714024923

[felixbrucker]

I also noticed it displayed "cpuminer-multi" at startup. This is a leftover artifact that I haven't seen before.
It might be in the way you built the load. Did you modify any options from the default?

i get the same for my compiles on linux (windows compiles generate correct string), i fixed it by doing the following:

Code:

sed -i -- 's/\[cpuminer-multi\]/\[cpuminer-opt\]/g' configure.ac
sed -i -- "s/, \[1.2-dev\]/, \[${minerVer}\]/g" configure.ac

its in the first lines of that file

cheers

great to see an official git repo

[joblo]

I also noticed it displayed "cpuminer-multi" at startup. This is a leftover artifact that I haven't seen before.
It might be in the way you built the load. Did you modify any options from the default?

i get the same for my compiles on linux (windows compiles generate correct string), i fixed it by doing the following:

Code:

sed -i -- 's/\[cpuminer-multi\]/\[cpuminer-opt\]/g' configure.ac
sed -i -- "s/, \[1.2-dev\]/, \[${minerVer}\]/g" configure.ac

its in the first lines of that file

cheers

great to see an official git repo

Thanks, Already made the change for next release.

[t2yax]

I think you should add wild keccak to the miner,its the algo of boolberry

https://bitcointalk.org/index.php?topic=577267.0

http://coinmarketcap.com/currencies/boolberry/

[joblo]

I think you should add wild keccak to the miner,its the algo of boolberry

https://bitcointalk.org/index.php?topic=577267.0

http://coinmarketcap.com/currencies/boolberry/

I don't see much point, cpuminer-opt only works with stratum and I didn't find any working pools from the list.
Also the existing miner is already optimized so it's unlikely I could improve it.

[Epsylon3]

it was added to suprnova last week...

[joblo]

it was added to suprnova last week...

It's going to be more difficult to integrate because it has a few unique features such as the disk based
scratchpad that needs to be downloaded and requiring a new command arg. Finding all the diffferences
will be hard because the existing miner is only for wildkeccak so there are no hooks in the code
showing where it does things differently.

There is little to be gained so I'm in no rush. Lately I've been buried in yaamp code trying to find the
missing 20%. It's hard when the pool op isn't interested in helping find it.

[Epsylon3]

yep... and you didnt see this problem on yiimp right ? its much easier to check when there are "empty" pools

also, zpool can use rental services, unsure how that works...

[joblo]

yep... and you didnt see this problem on yiimp right ? its much easier to check when there are "empty" pools

also, zpool can use rental services, unsure how that works...

I haven't done any specific testing on yiimp but presume it would be more obvious if there was a problem
because you have smaller pools with fewer coins and no exchange. It would be easy to see if the
total unpaid dropped when a coin matured.

That was actually a good clue. the problem is probably in code you don't run. Other than estimates and exchange
related code are there any other sections you skip?

I'm not sure how much you are following it but a user demonstrated the problem when mining the same coin as his payout.
This seems odd as I would expect most of the estimate and exchange code to be skipped in such cases. Eevrything added
up until after it was cleared in the last 50 list. The value was still corect after it was cleared and that amount was deducted
from the pending and a lower amount credited to the balance, with a net loss of around 25% in his case.

[Epsylon3]

the captures ive seen today doesnt really show anything... the block reward % is missing (Last 50 Earnings)

[joblo]

the captures ive seen today doesnt really show anything... the block reward % is missing (Last 50 Earnings)

The image in this post shows it clearly.

https://bitcointalk.org/index.php?topic=1260863.msg16311116#msg16311116

Actually that example is not with the same payout coin but you can clearly see the cleared value
for the block but a decrease in the total unpaid.

[Epsylon3]

yep lol, very "clear"... kidding

no, seriously, this kind of use of the multi-pool... well now i remember why i dont want to touch it

Actually, we made a few tests on the empty pools (x11evo and c11) and ... well its a fully different behavior than yiimp for the % of reward.

Yiimp do that per block, a stopped miner can only get a part of one block/round... Not on zpool, we seen at least 10 blocks with a reward percent decreasing..

[joblo]

yep lol, very "clear"... kidding

no, seriously, this kind of use of the multi-pool... well now i remember why i dont want to touch it

Actually, we made a few tests on the empty pools (x11evo and c11) and ... well its a fully different behavior than yiimp for the % of reward.

Yiimp do that per block, a stopped miner can only get a part of one block/round... Not on zpool, we seen at least 10 blocks with a reward percent decreasing..

Hard to avoid that pun.

I've never seen that behaviour in a pool with only one coin. I assumed all the hash allocated to a specific coin is consumed,
the leftovers were for another coin.

The hash allocation is another mystery. It seems the accumulated hash can be transferred to another coin. I recall trying to time
mining dash when the diff was low and all pool hash was going to it. On more than one occasion I stopped mining when the dash diff
jumped and the pool shifted to ther coins and all my hash was consumed on a few shit coins before a dash block was found. But I digresss.

I'm still hoping the zpool op will step up and review his logs.

[felixbrucker]

i noticed a small bug (not sure which version introduced it though):
when mining cryptonight the DIFF available via API is zero, the result of "summary" is:

Code:

{
  NAME: 'cpuminer-opt',
  VER: '3.4.7',
  API: '1.0',
  ALGO: 'cryptonight',
  CPUS: '4',
  KHS: '0.09',
  ACC: '43',
  REJ: '0',
  ACCMN: '0.788',
  DIFF: '0.000000',
  TEMP: '0.0',
  FAN: '0',
  FREQ: '0',
  UPTIME: '3276',
  TS: '1474630263|'
}

[joblo]

i noticed a small bug (not sure which version introduced it though):
when mining cryptonight the DIFF available via API is zero, the result of "summary" is:

Code:

{
  NAME: 'cpuminer-opt',
  VER: '3.4.7',
  API: '1.0',
  ALGO: 'cryptonight',
  CPUS: '4',
  KHS: '0.09',
  ACC: '43',
  REJ: '0',
  ACCMN: '0.788',
  DIFF: '0.000000',
  TEMP: '0.0',
  FAN: '0',
  FREQ: '0',
  UPTIME: '3276',
  TS: '1474630263|'
}

I haven't made any changes to the API code but I looked through it and didn't see anything obvious. It should display
either net_diff if it's non-zero else stratum_diff. net-diff isn't used by cryptonight so it should display stratum_diff.
Don't know why it isn't.

[felixbrucker]

I haven't made any changes to the API code but I looked through it and didn't see anything obvious. It should display
either net_diff if it's non-zero else stratum_diff. net-diff isn't used by cryptonight so it should display stratum_diff.
Don't know why it isn't.

i just confirmed its broken since 3.4.5 or 3.4.4b, <=3.4.4b was working fine (though the stratum diff displayed was not 10k but 10000.016985)

[Epsylon3]

nice to see users using the api

[felixbrucker]

nice to see users using the api

i like apis

[joblo]

I haven't made any changes to the API code but I looked through it and didn't see anything obvious. It should display
either net_diff if it's non-zero else stratum_diff. net-diff isn't used by cryptonight so it should display stratum_diff.
Don't know why it isn't.

i just confirmed its broken since 3.4.5 or 3.4.4b, 3.4.4 was working fine (though the stratum diff displayed was not 10k but 10000.016985)

The change in v3.4.5 was suggested by Nicehash to resolve issues with new work on cryptonight. I don't see an obvious link
to broken statum_diff. It will take some digging.

[joblo]

I haven't made any changes to the API code but I looked through it and didn't see anything obvious. It should display
either net_diff if it's non-zero else stratum_diff. net-diff isn't used by cryptonight so it should display stratum_diff.
Don't know why it isn't.

i just confirmed its broken since 3.4.5 or 3.4.4b, 3.4.4 was working fine (though the stratum diff displayed was not 10k but 10000.016985)

The change in v3.4.5 was suggested by Nicehash to resolve issues with new work on cryptonight. I don't see an obvious link
to broken statum_diff. It will take some digging.

I have a fix. Although it's simple from a functional point of view the proper implementation involves code changes in 4 files.
However, there is a simple workaround you can use if you like.

In file cpu-miner.c line 2106

Code:

static void stratum_gen_work( struct stratum_ctx *sctx, struct work *work,
                              int thr_id )
{
    algo_gate.stratum_get_g_work( sctx, work, thr_id );
    if ( jsonrpc_2 ) return;                          // add this line
    if ( stratum_diff != sctx->job.diff )
    {
        char sdiff[32] = { 0 };
        // store for api stats
        stratum_diff = sctx->job.diff;
        if ( opt_showdiff && work->targetdiff != stratum_diff )
            snprintf( sdiff, 32, " (%.5f)", work->targetdiff );
        applog( LOG_WARNING, "Stratum difficulty set to %g%s", stratum_diff,
                            sdiff );
     }
}

Edit: corrected references for v3.4.7

Edit2: the display format is in the API code and is generic for all algos so it needs to support fractions.