Changes for the Monero v1 PoW

[vtnerd]

Monero v1 PoW changes. There was a warning about merging into master. I can update the patch for dev, but I'm not sure if that is what you want in this case. Please see monero PR to discuss the PoW changes.

Some implementation notes:

• I didn't "catch" how the build system was injecting an ITERATION flag until after I tested this patch. It should be easy to add a MONERO or similar macro define to remove branches in the AEON OpenCL inner loop.
• sgminer has implemented a macro'ed version for the OpenCL implementation. This is interesting because it detects the version in the CPU, then swaps out the kernel. Its not immediately clear this is the best approach because its going to vary based on what the pool provides. So it might be lots of book keeping during the fork. However, it won't branch in the inner loop. This is a little harder to implement but not terrible.
• The cuda version uses a boolean template for monero to remove branches from the aeon version.
• The cuda version could alternatively pass the input and state instead of adding another global variable, however that second process function now has to do two reads. So the alternative would remove a global write in the expansion phase but add another global read in the function containing the process loop.
• The CPU version uses a booleam template for monero to remove branches from the aeon version.
• The first tweak in the CPU version accesses the scratchpad directly after the simd write. I thought about putting that type (the 128-bit type) into a union, but it should mess with registers a bit since the tweak cannot be applied to multiple integers. So it would be messing around with simd registers vs L1 access. Perhaps you can twiddle something better.
• The second tweak modifies the stack variable in the 1 and 2 at a time implementations. For 3, 4, 5 at a time, the second tweak modifies the scratchpad directly. Alternatively, you could setup a 128-bit value where the "first" 64-bits are zero. Since zero is the identity the CPU could apply the result without smashing up the simd registers. This would require an additional 128-bit value, so I wasn't sure because it will mess with register allocations.
• It should be easy to add a template variable to remove branching from the monero version in the CPU and cuda versions, with the risk of more code generated. Possibly worth it.

Also see the PR linked above for "reference" hashes. Make sure you get my last commit if you want to verify.

[psychocrypt]

thank you for providing the pow patches. In that case it is not problematic that the PR is against the master. we plan some futher changes inside thw miner and merge everything by hand together.

[vtnerd]

I implemented a macro define for xmrig-amd that removes branches from the aeon build. Its not difficult to add to this implementation, but I will leave that to the team unless otherwise requested.

[fireice-uk]

@vtnerd Thanks for your work. Can you tell us when the PoW change is going get merged into monerod?

[GabrielKesler]

As developers, in your opinion, how long will this PoW change keep the botnets out ? Is it a quick fix for them ?

[psychocrypt]

I think this POW change will not help to kick of the bots. If the bot network owner can update there slaves than we will see in the best case a few days with less bots.

[Molokai]

POW change to take effect on March 28, block 1539500:

monero-project/monero@84decbe

https://www.reddit.com/r/Monero/comments/82pulx/network_upgrade_scheduled_for_block_1544555_on_28/
(Note, discrepancy in the title of the reddit and github pullrequest - and the actual code)

[nbourbaki909]

Will there be a new switch in the config for monero, aeon, and a third for CN coins that haven't changed their POW?

[stempelo]

Output from input "This is a test"

[stempelo]

Is it correct \xa0\x84\xf0\x1d\x14\x37\xa0\x9c\x69\x85\x40\x1b\x60\xd4\x35\x54\xae\x10\x58\x02\xc5\xf5\xd8\xa9\xb3\x25\x36\x49\xc0\xbe\x66\x05
for the input "This is a test" ??

[fireice-uk]

Yes, for the original cryptonote hash function.

[fireice-uk]

@nbourbaki909

Will there be a new switch in the config for monero, aeon, and a third for CN coins that haven't changed their POW?

Yes, we are not dropping support for any coins. As such you will need to use monero2 to actually use the new monero algo.

[plavirudar]

@fireice-uk Will the monero2 algo be backwards compatible with older pools? From my understanding, this is a change in the block template and it should be possible for the miner to dynamically adjust algos.

[fireice-uk]

@plavirudar you will simply need to enter the coin name that you want to mine, with exception of Monero, which will be called monero2 in the config. Algorithms will be mapped internally.

[fireice-uk]

@vtnerd Some of your code went into 1e7911e I added you as a co-author on that commit. Thanks for the contribution.