Dean Howarth

@kostrzewa Third data points are always welcome! :) I will look over this code for sure. In the meantime (and I understand if you don't have the time) a third...

@weinbe2 Let me see how wildly it varies from double -> single precision...

There is also the fact that MILC performs a reunitarisation before any computation. This might have an effect, but if it did, I would expect it to show up in...

@kostrzewa That would be awesome, thank you! There are some 32c64 located here `/scratch/ETMC_gauge_configurations/tmclover_nf2/cA2.60.32` Is that too large?

@weinbe2 double precision: `Computed topological charge gauge precise is -4.3976775762481104e+00` single precision: `Computed topological charge gauge precise is -4.3976789550765041e+00` A noticeable difference, but not on the scale of the MILC-QUDA...

@kostrzewa Can please you use `/scratch/ETMC_gauge_configurations/tmclover_nf2/cA2.60.32/conf.2960` Thank you. The data from the Wilson flow issue comes from a monster L=72, but I saw equally poor agreement on an L=24 test...

This good? ``` cpviolator@nvsocal2:/scratch/ETMC_gauge_configurations/tmclover_nf2/cA2.60.32$ head -19 conf.2960 Eg????xlf-infoplaquette = 0.603483301982 trajectory nr = 3798 beta = 2.100000, kappa = 0.137300, mu = 0.006000, c2_rec = -0.331000 time = 1420235063 hmcversion...

Excellent, thank you @kostrzewa. These are QUDA's results: ``` t, Et, Es, (Et+Es), Q 0.000000e+00 +1.0744528619619671e+00 +1.0743300497637591e+00 +2.1487829117257262e+00 -1.1986946054939605e+01 3.000000e-02 +1.0064045485004334e+00 +1.0062787721926125e+00 +2.0126833206930459e+00 -1.1382100367471125e+01 6.000000e-02 +9.2211203079526127e-01 +9.2197263868241142e-01 +1.8440846694776727e+00 -1.0555243689364456e+01 9.000000e-02...

Yeah, the minus sign I can live with, but the Q data is again quite different. 6th DP on Esym is acceptable. I'll run this lattice through MILC and see...

Data after addressing https://github.com/lattice/quda/pull/942/files#r369848547 ``` Computed topological charge gauge precise is -1.1986946054939605e+01 Done in 0.012059 secs Computed topological charge gauge precise from density function is -1.1986946054939605e+01 GPU value -1.198695e+01 and...