GROMACS version:2021.2
GROMACS modification: Yes
Here post your question
My environment:
Windows 10 Version: 1607
Visual studio version: 2019
FFTW: 3.3.9
cmake: 3.21.2
CUDA: 11.2
Visual Studio has build options like “Debug” and “Release” options when building the C++ project. The “Debug” option doesn’t optimize the code, so its output binary is very slow. My PC has GPU but binary file by “Debug” mode can calculate only 23s per day.
However, I encountered following error only when I run the Release build of Visual Studio. Debug build didn’t return such errors.
You may be able to work around it by compiling with AVX2 SIMD instead of AVX512, that is cmake -DGMX_SIMD=AVX2_256
With this you may give up some performance in CPU-only runs (likely little to none in GPU runs), but nowhere near as much as you the performance you sacrifice with Debug mode.
Thank you so much, it avoided the error and succeed in compile.
I rebuilt the code in release mode and saw significant improvement in performance 23ns/day to 148ns/day for GROMACS Tutorial “Lysozyme in water” (all atomic simulation of 33876 atoms and dt=2fs).
Meanwhile, it remains unclear why Visual Studio cannot compile with the option of utilizing AVX512 instruction set. My CPU is Intel® Xeon® Gold 6134 so it should support it.
As before, this looks like a compiler bug (a new one). Try downgrading Visual Studio to workaround the issue and/or upvote the MS bug report to get it fixed: