Benchmark_dwf* fails in branch develop about grid

[aportelli]

Hi Guido from what I can see the develop branch is fine (https://travis-ci.org/paboyle/Grid/builds/141859994) the failed build is on your merge branch so it might up to you to correct this?

[coppolachan]

Hi Antonin, I switched to the develop branch to test exactly this, because it was failing on my merge.

The build is ok, is the executable that fails.

I'm running with 4 threads, this could be the source of the discrepancy.

[paboyle]

The executable Benchmark_dwf IS tested in the build on travis, and I agree that Guido appears to have
broken it by removing the patch that I put in that fixed it.

Travis Log prior to Guido's commit:

https://travis-ci.org/paboyle/Grid/builds/141859994

$ ./benchmarks/Benchmark_dwf --threads 1
Grid : Warning : 0 ms : '--threads' option used but Grid was not compiled with thread support
||||||||||||||__
||||||||||||||__
|_ | | | | | | | | | | | | _|
|_ _|
|_ GGGG RRRR III DDDD _|
|_ G R R I D D _|
|_ G R R I D D _|
|_ G GG RRRR I D D _|
|_ G G R R I D D _|
|_ GGGG R R III DDDD _|
|_ _|
||||||||||||||__
||||||||||||||__
| | | | | | | | | | | | | |
Copyright (C) 2015 Peter Boyle, Azusa Yamaguchi, Guido Cossu, Antonin Portelli and other authors
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
Grid : Message : 25 ms : Grid is setup to use 1 threads
Grid : Message : 25 ms : Making s innermost grids
Grid : Message : 25 ms : Making s innermost rb grids
Grid : Message : 2449 ms : Calling Dw
Grid : Message : 2659 ms : Called Dw 10 times in 208498 us
Grid : Message : 2662 ms : norm result 1.95449e+07
Grid : Message : 2666 ms : norm ref 1.95449e+07
Grid : Message : 2670 ms : mflop/s = 4224.52
Grid : Message : 2672 ms : mflop/s per node = 4224.52
Grid : Message : 2685 ms : norm diff 6.68005e-25
Grid : Message : 3107 ms : Called Dw sinner 10 times in 230982 us
Grid : Message : 3109 ms : mflop/s = 3813.3
Grid : Message : 3109 ms : mflop/s per node = 3813.3
difference between normal and simd is 0
Grid : Message : 3352 ms : EO src norm diff 0
Grid : Message : 3472 ms : sDeo mflop/s = 4013.47
Grid : Message : 3474 ms : sDeo mflop/s per node 4013.47
Grid : Message : 3560 ms : sE norm diff 0
Grid : Message : 3566 ms : sO norm diff 0
Grid : Message : 3744 ms : Called DwDag
Grid : Message : 3746 ms : norm result 1.95238e+07
Grid : Message : 3750 ms : norm ref 1.95238e+07
Grid : Message : 3759 ms : norm diff 4.81804e-25
Grid : Message : 3767 ms : Calling Deo and Doe
Grid : Message : 3809 ms : src_e261804
Grid : Message : 3812 ms : src_o262391
Grid : Message : 3917 ms : Deo mflop/s = 4306.86
Grid : Message : 3919 ms : Deo mflop/s per node 4306.86
Grid : Message : 3968 ms : r_e9.78579e+06
Grid : Message : 3969 ms : r_o9.75908e+06
Grid : Message : 3970 ms : res1.95449e+07
Grid : Message : 4063 ms : norm diff 0
Grid : Message : 4107 ms : norm diff even 0
Grid : Message : 4110 ms : norm diff odd 0

for example.

This fails in Guido's commit:

https://travis-ci.org/paboyle/Grid/jobs/142254863

[paboyle]

For what it is worth, the patch was in

lib/qcd/actions/fermion/WilsonKernels.cc

to disable the call to assembler unless AVX512 is set.

It was already in develop for some days.

[coppolachan]

Fails also with only 1 thread.

I have no commits in the branch develop.

[paboyle]

Can you post the precise nature of the fail & also try with SSE4 ?

I don't understand how this is consistent with the travis compile of develop passing (which uses SSE4 since the VM's dont guarantee to have AVX and beyond - possible an AVX only issue though)

[coppolachan]

My bad, the branch I checked out was updated to June.
After the full update, it is working.
Sorry for taking you time.