Dear AMBER Development Team,
I am encountering a segmentation fault when running a TI simulation with lambda scheduling for restraints on AMBER 24 built with HIP support (ROCm 6.2.4) for AMD MI250X GPUs. The same input works correctly with the CUDA build. Normal TI simulations can run with HIP and produce reasonable results. The failure only occurs when lambda scheduling for restraints is enabled (i.e. gti_bat_sc is set and nmropt=1).
pmemd.hip -O -p unisc.parm7 -c 0.00000000_preTI.rst7 -i 0.00000000_ti.mdin -o t1/0.00000000_ti.mdout -r t1/0.00000000_ti.rst7 -x t1/0.00000000_ti.nc<
http://0.00000000_ti.nc/> -ref 0.00000000_preTI.rst7
> Lambda Scheduling: 6 2 0 1.0000 0.0000
> [1] 1349133 segmentation fault (core dumped) pmemd.hip -O -p unisc.parm7 -c 0.00000000_preTI.rst7 -i 0.00000000_ti.mdin -o
I have traced the crash to the `gti_restraint_setup_` routine but have not yet identified the root cause.
I have uploaded the relevant files at
https://drive.google.com/file/d/1U5n8iFw0-h0WGoVENMZ9vWmRmlw4iz_r/view?usp=sharing and can provide any further information you need.
Thanks!
Best regards,
Yuxuan Zhuang
Dror Lab
Stanford University
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Jun 09 2025 - 12:30:03 PDT
