apply new cpuminer (yespower)

[cryptozeny]

https://github.com/cryptozeny/cpuminer-mc-yespower

hi i released a new cpuminer using yespower. it boosts 2~15% hashpower. it has an improve on smartphones! 😃 it based on macchky's one.

ryzen + 2%
intel + 3%~7%
arm-aarch64 (smartphones) + 15%
arm-v7l (neon): not supported yet by algorithm author, and it can run as fallback mode (no boost). hopefully he said it will come with next release.

[cryptozeny]

today i pushed yespower to official cpuminer what you currently using. you can apply with

-a yespower option 😉

[OttyLab]

Thanks, I'll check that.

[OttyLab]

@cryptozeny

support-yespower branch supports yespower. Could you try this out? (iOS also supports)

[cryptozeny]

@OttyLab i lost my develop environments for android, so it takes some time... 😭 could you give me a test release?

[OttyLab]

@cryptozeny

Here's the build. (Sorry, the signature is changed because I lost the password 😅 )
build.zip

In my environment, yespower is worse than yescrypt. Is any build parameter required for optimization?

[cryptozeny]

it wont install on my note8 😅 i will try an another device

[cryptozeny]

@OttyLab please use this build option? it makes faster

./configure CFLAGS="-O2 -fomit-frame-pointer"

[OttyLab]

@cryptozeny

it wont install on my note8

What kind of error did you get? If the device has previous version, the new one fails due to the signature mismatch. In that case, uninstalling the previous one would work.

-fomit-frame-pointer

It improves the hash power, but looks not enough. I'll see to it.

[OttyLab]

Hmm, it's weird...

Nexus 7:

.	yescript	yespower
(1)	0.066 kh/s	0.071 kh/s
(2)	0.12 kh/s	0.1 kh/s

(1): run minerd on adb shell
(2): run libmier on the app

The hash rate is way different on yescript between (1) and (2)

[cryptozeny]

wired...🤔

[OttyLab]

I've checked the code but I could not find out any problem.

libmienr.c (used by Android app) is created based on cpu-miner.c to support start/stop the threads. That means the differences in libminer.c should not affect the hash rate in theory.

I performed additional tests with cpuminer2 on Nexus7

1. on adb shell

• minerd: 0.10kHash/s
• libminer: 0.10kHash/s

2. on Android app

• minerd: cannot test (does not support start/stop etc)
• libminer: 0.13kHash/s

So, it seems that the hash rate improves when libminer is called from Android app for some reason, but I do not know what causes this result.

[OttyLab]

Anyway, yespower itself runs. So I'd like to merge the code that supports yespower