[gmx-developers] Branches w/working OpenCL support

Mirco Wahab mirco.wahab at chemie.tu-freiberg.de
Fri Jun 19 10:18:18 CEST 2015

On 19.06.2015 09:16, Roland Schulz wrote:
>     Yes, Update 4. Your response suprised me. What I found out after
>     further tests:
>     cmake with  -DGMX_SIMD=AVX_256 compiles fine  (only an <algorithm> for
>     min/max missing somewhere)
>     cmake with  -DGMX_SIMD=SSE2 fails on expansion of this said macro
>     (in bonded.cpp, clincs.cpp)
> Thanks. I didn't correctly test with SSE2 and thus thought it was OK
> with 2013. https://gerrit.gromacs.org/#/c/4744/ fixes it.

OK, if fetched the OpenCL version
   git fetch https://gerrit.gromacs.org/gromacs refs/changes/14/4314/26 
&& git checkout FETCH_HEAD
and pulled your change against it:
   git pull https://gerrit.gromacs.org/gromacs refs/changes/44/4744/2

Yes, now it compiles w/SSE2. There's only one single line to add to
gmx_wham.cpp (see appendix) and it compiles "out of the box" w/MSVC.

But the runs crash as before after reading the parameters
and immediately before, I'll guess here, the domain composition?
The next line in the log file would be:

Initializing Domain Decomposition on nnn ranks
Dynamic load balancing: auto

which doesn't appear. The mdrun segfaults before this (after
printing run parameters to md.log).




diff --git a/src/gromacs/gmxana/gmx_wham.cpp 
index 9e71686..c40cfd6 100644
--- a/src/gromacs/gmxana/gmx_wham.cpp
+++ b/src/gromacs/gmxana/gmx_wham.cpp
@@ -48,7 +48,7 @@
  #include <stdio.h>
  #include <stdlib.h>
  #include <string.h>
+#include <algorithm>
  #include <sstream>

  #include "gromacs/commandline/pargs.h"

More information about the gromacs.org_gmx-developers mailing list