Running GROMACS with OpenCL on accelerator

User discussions
1

GROMACS version:2023.3
GROMACS modification: Yes
Here post your question
I am trying to run GROMACS on an accelerator connected to PCI bus.
I made the following modifications to the code:

  1. changed CL_DEVICE_TYPE_GPU to CL_DEVICE_TYPE_ALL
  2. Added my accelerator vendor
    checkGpu returns value TRUE, so the dummy kernel is able to compile on my accelerator.

However, I am getting the following errors on Regression Tests. I have pasted some of the errors here.

If someone has tried to run the OpenCL build of GROMACS on an accelerator, or have any advice, please do let me know. Also lemme know if more details are needed.
Thanks for the help.
Aastha

Regression Test Errors:

Start 84: regressiontests/complex
84/87 Test #84: regressiontests/complex ...................***Failed  132.49 sec

Abnormal return value for ' gmx mdrun        -notunepme >mdrun.out 2>&1' was -1
FAILED. Check mdrun.out, md.log file(s) in aminoacids for aminoacids
Re-running aminoacids using only CPU-based non-bonded kernels

Abnormal return value for ' gmx mdrun        -notunepme -cpi /home/morphing/redefine_sdk/redefine/REDEFINE/GROMACS/gromacs_clean2/gromacs-2023.3/build/tests/regressiontests-2023.3/complex/awh_multibias/continue -noappend >mdrun.out 2>&1' was -1
FAILED. Check mdrun.out, md.log file(s) in awh_multibias for awh_multibias

Abnormal return value for ' gmx mdrun        -notunepme >mdrun.out 2>&1' was -1
FAILED. Check mdrun.out, md.log file(s) in awh_multidim for awh_multidim
Re-running awh_multidim using only CPU-based non-bonded kernels

Abnormal return value for ' gmx mdrun        -notunepme >mdrun.out 2>&1' was -1
FAILED. Check mdrun.out, md.log file(s) in butane for butane
Re-running butane using only CPU-based non-bonded kernels
Source file: src/gromacs/gpu_utils/ocl_compiler.cpp (line 499)
Program:     mdrun-vsites-test, version 2023.3
Function:    _cl_program* gmx::ocl::compileProgram(FILE*, const string&, const string&, const string&, cl_context, cl_device_id, DeviceVendor)
Source file: src/gromacs/gpu_utils/ocl_compiler.cpp (line 499)
MPI rank:    0 (out of 2)

Internal error (bug):
Function:    _cl_program* gmx::ocl::compileProgram(FILE*, const string&, const string&, const string&, cl_context, cl_device_id, DeviceVendor)
MPI rank:    1 (out of 2)

Internal error (bug):
Failed to compile/load nbnxm kernels for GPU #0 fsim
Failed to compile/load nbnxm kernels for GPU #0 fsim
 Could not build OpenCL program, error was CL_BUILD_PROGRAM_FAILURE
 Could not build OpenCL program, error was CL_BUILD_PROGRAM_FAILURE

Program:     mdrun-pull-test, version 2023.3
Source file: src/gromacs/gpu_utils/ocl_compiler.cpp (line 499)
Function:    _cl_program* gmx::ocl::compileProgram(FILE*, const string&, const string&, const string&, cl_context, cl_device_id, DeviceVendor)

Internal error (bug):
Failed to compile/load nbnxm kernels for GPU #0 fsim

Could not build OpenCL program, error was CL_BUILD_PROGRAM_FAILURE
2

Hi Aastha, I am facing the similar problem with 2023.3 with Test #74:regressiontests/complex only, 99% passed in my case. Were you able to fix the error? Please let me know. Thanks