Slow AVX XOR reported

The AVX XOR performance reported during boot seems wrong.

raid6: using algorithm sse2x4 (10035 MB/s)
xor: automatically using best checksumming function:
avx : 3468.000 MB/sec

I would expect the value to be well north of 10000MB/s. I do not get a generic_sse performance number which I do on other non-AVX hosts.

The system is a dual AMD Opteron 6378 (16 core, 2.4GHz, piledriver). The kernel is 3.11.8-200.fc19.x86_64.

I did read that AMD's AVX-256 implementation performs poorly compared to AVX-128. Perhaps there are other AMD specific "optimizations" that need to be considered.

Is there any standalone code I ran run to benchmark the various XOR implementations?

--Larkin