Evan Weinberg

@cpviolator yep, I agree that's divergence, point noted.

> Could this be related to using `QUDA_TEX=ON`? No with 99.99% confidence

It'd be related to the way the reconstruct is done, not the read itself (which is abstracted at a lower level). The local volume isn't large enough for there to...

This depends on #1002 being merged.

This line looks a little weird: https://github.com/milc-qcd/milc_qcd/blob/ac033d6faf9edb86ecb89acabb5ae3108fd618d3/wilson_flow/fmunu.c#L56 But otherwise they look consistent.

I wonder if it's a difference in the order of reduction... `symmE` is per-site positive by construction and thus less susceptible to issues, while the topological charge fluctuates in sign...

@cpviolator gotcha, so it's likely not that...

@cpviolator one idea is running the Q compute on the CPU instead of the GPU in QUDA. I think you can just hack that in my manually copying the gauge...

Complimentary reference PR for MILC: https://github.com/milc-qcd/milc_qcd/pull/55

With the exception of `clang-format`, which I can do as a last step, this is ready for review.