https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119351
Bug ID: 119351
Summary: [15 Regression] Wrong code in GROMACS for AArch64
generic SVE VLS target
Product: gcc
Version: 15.0
Status: UNCONFIRMED
Keywords: wrong-code
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: ktkachov at gcc dot gnu.org
CC: acoplan at gcc dot gnu.org, tnfchris at gcc dot gnu.org
Target Milestone: ---
Target: aarch64
Tamar and I have been discussing this offline but now that we have a reproducer
with all public sources here's a bug report.
We see GROMACS giving an internal error when built for -march=armv9-a.
To reproduce you need to get GROMACS from:
https://gitlab.com/gromacs/gromacs.git
and use branch v2025.0-1
You may need to patch one of the header files trivially to make it build:
--- a/src/gromacs/mdtypes/energyhistory.h
+++ b/src/gromacs/mdtypes/energyhistory.h
@@ -51,6 +51,7 @@
#include <memory>
#include <vector>
+#include <cstdint>
Build it with cmake with the following options:
CMAKE_OPTIONS=" -DGMX_OPENMP=ON -DGMX_CYCLE_SUBCOUNTERS=ON \
-DGMX_BUILD_OWN_FFTW=ON \
-DGMX_GPU=OFF
-DCMAKE_BUILD_TYPE=Release -DGMX_DOUBLE=OFF -DGMX_CYCLE_SUBCOUNTERS=ON \
-DGMX_PREFER_STATIC_LIBS=OFF
-DGMX_INSTALL_NBLIB_API=OFF -DGMXAPI=OFF \
-DCMAKE_C_FLAGS="-march=armv9-a"
-DCMAKE_CXX_FLAGS="-march=armv9-a" \
-DCMAKE_C_COMPILER=$COMPILERBIN
-DCMAKE_CXX_COMPILER=$COMPILERXXBIN \
-DGMX_SIMD_ARM_SVE_LENGTH=128 \
-DGMX_SIMD=ARM_SVE -DGMX_USE_NVTX=ON \
-DGMX_MPI=OFF"
This builds GROMACS for -march=armv9-a and 128-bit VLS SVE.
The run command can be:
$PATH_TO_GMX_BUILD/gmx mdrun -v -resethway -noconfout -pin on -ntmpi 1 -ntomp
24 -nsteps 4000 -nb cpu -s benchmark
where "benchmark" is a benchmark.tpr input file that you can get from various
sources, for example https://www.hecbiosim.ac.uk/benchmark-files/gromacs.tar.gz
(you can use the 20k-atoms one)
The error at the bottom looks like:
Source file: src/gromacs/mdlib/sim_util.cpp (line 574)
Function: void gmx::checkPotentialEnergyValidity(int64_t, const
gmx_enerdata_t&, const t_inputrec&)
Internal error (bug):
Step 100: The total potential energy is -nan, which is not finite. The LJ and
electrostatic contributions to the energy are 0 and 0, respectively. A
non-finite potential energy can be caused by overlapping interactions in
bonded interactions or very large or Nan coordinate values. Usually this is
caused by a badly- or non-equilibrated initial configuration, incorrect
interactions or parameters in the topology.
For more information and tips for troubleshooting, please check the GROMACS
website at https://manual.gromacs.org/current/user-guide/run-time-errors.html
I've bisected this to g:68326d5d1a593dc0bf098c03aac25916168bc5a9
Before that commit the above command runs successfully.
Note, this triggers with -march=armv9-a in -DCMAKE_C_FLAGS and
-DCMAKE_CXX_FLAGS
Using values like -mcpu=neoverse-v2 doesn't trigger this
