Bonus: MSVC version for Windows is faster than other miners by 3-4% on Ryzen CPUs.

Update: #2278 - GCC version is a bit faster than MSVC now. This code is now the fastest on Zen3 with big margin (7.8% faster than SRBMiner, 15.7% faster than XMRigCC on my Ryzen 5 5600X).

Interesting fact: on Zen 1+ (2700) the configuration ...

 "cn/upx2": [
        [1, 0],
        [1, 1],
        [1, 2],
        [1, 3],
        [1, 4],
        [1, 5],
        [1, 6],
        [1, 7],
        [1, 8],
        [1, 9],
        [1, 10],
        [1, 11],
        [1, 12],
        [1, 13],
        [1, 14],
        [1, 15]
    ],

...is faster than the default config...

 "cn/upx2": [
        [2, 0],
        [2, 1],
        [2, 2],
        [2, 3],
        [2, 4],
        [2, 5],
        [2, 6],
        [2, 7],
        [2, 8],
        [2, 9],
        [2, 10],
        [2, 11],
        [2, 12],
        [2, 13],
        [2, 14],
        [2, 15]
    ],

Not much, but reproducable 43 kH/s vs. 42 hH/s.

On Zen 3 instead, [2, x] is much fast than [1, x], on an 5800X 77kH/s vs. 71 kH/s.

still slower than srbminer on intel.

I've tested the MSVC version of 6.12.0 in UPX. It throws errors:

XHV and XMR are working fine. It's a prime-stable Ryzen 7 5800X. Perhaps a problem with the special asm for Zen 3?

Gonna countertest the GCC version now...

Yes, it's something with Zen3 asm code, I'll check it.

Yes, it's the same with GCC. Countertest with asm false.

Edit: strange. Even when I set asm: "intel" manually I get rejected shares. Is xmrig using the Zen 3 asm on a Zen 3 cpu even if I manually choose "intel"?

Edit 2: asm: false is working well, but -11K hashrate!

Fix: #2296