aco_validate error

[Titaniumtown]

When trying to mine I get this error:

2021-11-28T19:17:32.296059Z  INFO acominer::stratum: Pool set difficulty difficulty=0.83G
2021-11-28T19:17:32.298152Z  INFO acominer: device=AMD RADV POLARIS10 type=DiscreteGpu
2021-11-28T19:17:32.326010Z  INFO acominer: subgroup_size=64
2021-11-28T19:17:32.326051Z  INFO acominer: Generating light cache (CPU)...
2021-11-28T19:17:40.484549Z  INFO acominer: Light cache generation done.
2021-11-28T19:17:40.739476Z  INFO acominer: Light cache transferred to GPU.
ACO ERROR:
    In file ../src/amd/compiler/aco_validate.cpp:97
    Too many SGPRs/literals:  v1: %19 = v_alignbit_b32 %1948, %1949, %18
ACO ERROR:
    In file ../src/amd/compiler/aco_validate.cpp:97
    Too many SGPRs/literals:  v1: %20 = v_alignbit_b32 %1949, %1948, %18
ACO ERROR:
    In file ../src/amd/compiler/aco_validate.cpp:97
    Too many SGPRs/literals:  v1: %27 = v_alignbit_b32 %1950, %26, %18
ACO ERROR:
    In file ../src/amd/compiler/aco_validate.cpp:97
    Too many SGPRs/literals:  v1: %28 = v_alignbit_b32 %26, %1950, %18
ACO ERROR:
    In file ../src/amd/compiler/aco_validate.cpp:97
    Too many SGPRs/literals:  v1: %35 = v_alignbit_b32 %17, %1944, %18
ACO ERROR:
    In file ../src/amd/compiler/aco_validate.cpp:97
    Too many SGPRs/literals:  v1: %36 = v_alignbit_b32 %1944, %17, %18
ACO ERROR:
    In file ../src/amd/compiler/aco_validate.cpp:97
    Too many SGPRs/literals:  v1: %39 = v_alignbit_b32 %1945, %1946, %18
ACO ERROR:
    In file ../src/amd/compiler/aco_validate.cpp:97
    Too many SGPRs/literals:  v1: %40 = v_alignbit_b32 %1946, %1945, %18
acominer: ../src/amd/compiler/aco_interface.cpp:73: void validate(aco::Program*): Assertion `is_valid' failed.
fish: Job 1, './target/release/acominer str...' terminated by signal SIGABRT (Abort)

[ishitatsuyuki]

Are you running an up-to-date version of my mesa fork? You can also use the bundled version in the GitHub Actions binaries as well.

I tried to emulate your GPU generation but I couldn't reproduce the abort.

[ishitatsuyuki]

Nevermind, now I can reproduce this. Fix is incoming.

[ishitatsuyuki]

Fixed with https://gitlab.freedesktop.org/ishitatsuyuki/mesa/-/commit/2c81abbb85ff038d91cc25fd18951b7e78bc8fa6. Build will be available here.

[Titaniumtown]

Thank you so much! I'll test right now and I'll close the issue if the problem is gone.

[Titaniumtown]

Ok, it works now. but I'm now experiencing some performance regression. it went from 25 mh/s (last time I tested) to now 19.26 mh/s on my rx 480 :(

[ishitatsuyuki]

I checked past changes and I pretty much believe that none of them should result in a significant performance regression. Are you comparing the results to a proprietary miner? Which version was the last version working well for you?

[ishitatsuyuki]

Please try this build and let me know if it's better.

[Titaniumtown]

I was comparing results to a previous version of acominer (I don't remember which), but now I'm getting 1 mh/s with both the latest commit of acominer, and the build you sent me. Whereas with nsfminer (an opensource eth miner) I'm still getting the normal 27 mh/s. This is very strange! I'll test again later, I have to get ready for school.

[ishitatsuyuki]

A new build with optimizations targeting GCN is available here. You might want to try it.

[Titaniumtown]

I'm on it! I'll report back with results.

[Titaniumtown]

Hm, very strange. Same result. This is strange as I was getting normal hashrate on this miner before. I'll try on another machine when I get home later.

[Titaniumtown]

I cannot reproduce on my other rx 480. very strange. It's prob an issue on my end. Thanks for your help @ishitatsuyuki keep up the good work <3