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

[joblo]

Hey guys, I'm a little pissed right now at doktor83 for using my code for zcoin (it was mine, not optiminer as he claimed)
and not releasing the source for his claimed 40% improvements to the timetravel algo.

So please take this elsewhere.

It's how open source works.

Only when it stays open.

[joblo]

Hey guys, I'm a little pissed right now at doktor83 for using my code for zcoin (it was mine, not optiminer as he claimed)
and not releasing the source for his claimed 40% improvements to the timetravel algo.

So please take this elsewhere.

It's how open source works.

Only when it stays open.

WRONG. You open source code with a given license, and people can do what they want within the terms of said license. Often this includes not releasing modifications.

That was not my point. If the source goes closed it's no longer "open source".

It's not about him using my code, it's more about a first release that is binary only making ourageous
performance claims. I'm skeptical.

In the past year I've learned a lot about what can be optimized, how it can be optimized and what improvements
can be expected. This is way off the chart.

[joblo]

I'm working hard on the next release with support for the new zcoin algo but I don't think it will
be ready for the cutover.

In still in the middle of long testing another algo with 5 of 8 test cases complete. I don't have control
over the test cases so I have to wait for the right conditions to occur in the algo. When that's
done I have to retest all the lyra2 algos in multiple architectures to make sure nothing broke.

I'm looking at 7-10% faster for all Lyra2.

[maxfunky]

Thanks for your great work to all of you btw!

[doktor83]

joblo, you are all wrong.
I said i implemented matrix usage like optiminer did in his miner a long time ago, and i used it on your code, yes.
Never said it was optiminers code, lol.
Anyways i like reworking your miner because i like it the most.. coulda work on tpruvot version or any other.. but yours is the nicest

And i don't get it why are you pissed on me for the speedup i achieved with timetravel algo.
I told you, test it for yourself if you don't believe in that speedup.

[joblo]

joblo, you are all wrong.
I said i implemented matrix usage like optiminer did in his miner a long time ago, and i used it on your code, yes.
Never said it was optiminers code, lol.
Anyways i like reworking your miner because i like it the most.. coulda work on tpruvot version or any other.. but yours is the nicest

And i don't get it why are you pissed on me for the speedup i achieved with timetravel algo.
I told you, test it for yourself if you don't believe in that speedup.

Yes it was the matrix implementation in lyra2 that you claimed from optiminer are you are correct.
That solved the misaligment issues with AVX2 and provided some speed improvement.

I only care about source code. Your claims mean nothing to me without it. I haven't seen anyone else
praising your accomplishment, 40% faster than the previous fastest on timetravel is certainly praise worthy.

Your explanation of optimizing JH doesn't add up. I need proof and the binary is only slick advertising.

[doktor83]

affraid of viruses, gotta run it in a vm 

[joblo]

affraid of viruses, gotta run it in a vm 

It isn't that I'm afraid of a virus from you, it's that I'm just not interested in closed source, whether
it's from Claymore, SP, or anyone else.

If you just want me to confirm your results, I won't do that, it doesn't interest me either.

[integrale]

affraid of viruses, gotta run it in a vm  

It isn't that I'm afraid of a virus from you, it's that I'm just not interested in closed source, whether
it's from Claymore, SP, or anyone else.

If you just want me to confirm your results, I won't do that, it doesn't interest me either.

First it was that you're mad because he made a claim you found ridiculous - now you can verify it, so it's back to hating it because he didn't release the source... why didn't you just say that in the first place?

if he need confirmation about,  he should ask ordinary miner-people , if he ask a coder , developer he should deliver code too , thats my sight of this case

[doktor83]

I am sorry but i did not ask for any confirmations from anyone, i just released an optimization for an algo, that's all.
Anyone can use it, that's why i shared it.
I don't get it why you act like this, so if i ask wolf to give me the source of his kernel because i don't believe its faster than mine, and he won't do it (for obvious reasons), i should get mad at him or what?

[joblo]

affraid of viruses, gotta run it in a vm  

It isn't that I'm afraid of a virus from you, it's that I'm just not interested in closed source, whether
it's from Claymore, SP, or anyone else.

If you just want me to confirm your results, I won't do that, it doesn't interest me either.

First it was that you're mad because he made a claim you found ridiculous - now you can verify it, so it's back to hating it because he didn't release the source... why didn't you just say that in the first place?

You're blowing it out of proportion. I only resented discussing problems with his fork in this thread.

My skepticism of the claims is just an opinion, based on factors I previously mentioned.

My disagreement with witholding the source code is all on me. Without it I'm just not interested in
what he is doing. I was a little provocative trying to get him to release it, and I may be dismissive,
but there's no hate.

[joblo]

Yet another weird problem, this time with zcoin.

I'm getting stack smashing with AVX on Zcoin. Other lyra algos ok, zcoin ok on AVX2.

There is no difference between the AVX targetted code for zcoin vs other lyra algos. There is no
difference in any of the portable code bwtween AVX and AVX2. In other words all the code
works, either for a different algo, or a different architecture, but it's all covered.

About the problem, stack smashing is iften called by infinitely recursive function calls or a
function with an extremely large local variable. In either case I would expect the overflow
to occur on the first pass.

But it doesn't. It completes several hashes calling all functions down to the Intel intrinsics
and returns before doing it again. The code path is the same on every path, there are no deviations
that lead down a rathole.

This is my signal I need a break.

Other test almost complete 2 of 8 cases failed, so need to back out these 2 changes.

[joblo]

Another delay. The problem with zcoin AVX also affects zcoin on Windows with AVX2.

[Epsylon3]

they changed again the algo, and its only temporary (again) :p https://zcoin.io/language/en/update-zcoin-development-16-january-2017/

this one is close to the first lyra2, but different :p anyway, the old one will never be reused, can be replaced so

[oldDIN]

Joblo do not rush. Work carefully and reliably. We are not in a hurry.

[integrale]

Hey Guy´s

another old crap is digging on timetravel , Thanks joblo´s great work

Intel Pentium Dual-Core E6600   SSE2   6:170a:2   102.776 kH   2   3060   -   -   cpuminer-opt/3.5.5   linux   GCC 4.8.4 SSE3

Ubuntu 14.04 LTS
if someone wondering why im using old crap i´ll explain

Motherboard Asus P5Q/epu = present from friend
other stuff like cpu,ram,powersupply laying around , Mix them togehter no Invest only the cost of electricity , Headless ! no screen only the essential stuffmaybe i add later an gpu

cheers

[joblo]

Pointer arithmetic strikes again. It caused the stack smashing when I was up scaling a pointer
offset to AVX size and messed it up badly, thought I was converting from uint64_ when it was actually bytes.
I inspected that function many times carefully but everytime I was thinking uint64_t to __m256i.

Lyra2z now works in AVX and I assume on Windows too.

One more tweak to do, then cleanup and full test. Should be released tomorrow.

This will likely be the end of any Lyra2 optimizing. AVX2 is fully done and AVX has only one function left
to convert. It's a difficult one and it's only AVX so it would only be for the learning experience.
It's at the point now that being too aggressive with manual optimizing is having a decreasing benefit
and increasing risk.

[integrale]

im sorry but all you wrote is spanish village (nur Bahnhof) for me , even i dont understand this , i believe you makes the right decision and deliver us a good and stable release as usually
thanks a lot for putting so much time and headache into this 

[joblo]

im sorry but all you wrote is spanish village (nur Bahnhof) for me , even i dont understand this , i believe you makes the right decision and deliver us a good and stable release as usually
thanks a lot for putting so much time and headache into this 

Haha. Yes that happens some times, not much interest to miners, mostly other devs.
I'm an experienced programmer, but not an experienced "C" programmer. As a result
I get hung up a lot on some of the C quirks, such as pointer arithmetic. Bugs in pointer
arithmetic always take me the longest to find and I seem to make them a lot.
I think it's evil. Just venting.

Things are much better now. Performance boost is better than expected. just testing left to do.
I managed to squeeze a little more juice out those old oranges.

[melea]

hi
need help for run only 4 miner threads , not 8 miners

thanks for your help and the work here.


./cpuminer -a hodl -o stratum+tcp://erc.suprnova.cc:7674 -u worker.worker -p ppppp -q

         **********  cpuminer-opt 3.5.5  ***********
     A CPU miner with multi algo support and optimized for CPUs
     with AES_NI and AVX extensions.
     BTC donation address: 12tdvfF7KmAsihBXQXynT6E6th2c2pByTT
     Forked from TPruvot's cpuminer-multi with credits
     to Lucas Jones, elmad, palmd, djm34, pooler, ig0tik3d,
     Wolf0, Jeff Garzik and Optiminer.

CPU: Intel(R) Core(TM) i7-4710HQ CPU @ 2.50GHz
CPU features: SSE2 AES AVX AVX2
SW built on Feb 12 2017 with GCC 5.4.0
SW features: SSE2
Algo features: SSE2 AES AVX AVX2
Start mining with SSE2

[2017-02-12 06:49:29] Starting Stratum on stratum+tcp://erc.suprnova.cc:7674
[2017-02-12 06:49:29] 8 miner threads started, using 'hodl' algorithm.
[2017-02-12 06:49:32] Stratum difficulty set to 8 (0.00000)
[2017-02-12 06:49:36] Accepted 1/1 (100%), 925 H, 397.17 H/s, 76C
[2017-02-12 06:49:38] Accepted 2/2 (100%), 824 H, 377.71 H/s, 79C
[2017-02-12 06:49:57] Accepted 3/3 (100%), 923 H, 412.94 H/s, 88C
[2017-02-12 06:50:09] Accepted 4/4 (100%), 964 H, 412.68 H/s, 88C