Merge patch_2Sept2026 into oxdrh

.github/copilot-instructions.md

@@ -92,7 +92,9 @@ platform, style, and regression workflows.
 `src/` and the corresponding `make fix-*`; build failures -> check for `-S cmake`,
 package dependencies, and VLA usage; unit tests -> rerun the single test with
 `ctest -V -R <name>`; regression tests -> verify the Python environment and whether
-example inputs were modified.
+example inputs were modified.  A unit-test failure on only ONE Linux CI job is
+usually the `LAMMPS_SIZES=bigbig` configuration (64-bit `tagint`) -- reproduce with
+a minimal bigbig build first (see the testing guide) before suspecting anything else.
 
 ## Repository Structure
 
@@ -244,6 +246,7 @@ automatically: read them before starting the corresponding kind of work.
 |---|---|
 | `src/KOKKOS/` styles (rules, policies) | `.github/instructions/kokkos.instructions.md` (auto) |
 | porting a style to KOKKOS | `.github/dev-docs/kokkos-porting-guide.md` + `kokkos-porting-backlog.md` |
+| porting/auditing `src/OPENMP/` (`/omp`) styles | `.github/dev-docs/openmp-porting.md` |
 | granular/DEM code or tests | `.github/instructions/granular-tests.instructions.md` (auto) |
 | documentation (`doc/`) | `.github/instructions/documentation.instructions.md` (auto) |
 | force-style YAML tests (`unittest/`) | `.github/instructions/force-style-tests.instructions.md` (auto) |

.github/dev-docs/kokkos-porting-guide.md

@@ -609,6 +609,13 @@ capturer can exist at a time" failure -- the drivers echo each captured section
 completes, so the last echoed line pinpoints where the throw escaped. A style with no
 `/kk` variant that cannot run under VerletKokkos needs `skip_tests: kokkos_gpu` in YAML.
 
+Mixed precision is a separate validation target per backend: a style that passes the
+mixed-precision fixtures in a HOST-only KOKKOS build can still produce NaNs on a GPU
+build when intermediate float-range limits are hit (class example: `exp()` of a large
+positive argument overflows in `float` -- reaxff/kk).  Run the `kokkos_gpu_mixed`
+fixture on a real device before declaring mixed-precision support clean; if a style is
+numerically unfit for single/mixed, add the corresponding `skip_tests` entries instead.
+
 ## KOKKOS test build
 
 ```bash

.github/dev-docs/kokkos-tier1-verification/.gitignore

@@ -0,0 +1 @@
+out/

.github/dev-docs/kokkos-tier1-verification/README.md

@@ -0,0 +1,81 @@
+# KOKKOS Tier-1 verification decks
+
+Hand-written input decks that exercise every style ported to KOKKOS in the
+Tier-1 pass, for styles whose behavior the `unittest/force-styles` YAML
+fixtures do not cover (fixes, computes, regions, minimizers, granular pair
+styles) and, as a second opinion, for those they do.
+
+The point of running them is that the `package kokkos` defaults are not the
+settings a GPU run gets.  Without a GPU, LAMMPS defaults to `comm no`,
+`sort no`, `atom/map no`, `neigh half`, `newton on` and `gpu/aware off`,
+which routes around the device communication, sorting and atom-map paths.
+`run_checks.sh` therefore drives every KOKKOS run with
+
+    -pk kokkos neigh full newton off comm device sort device atom/map device gpu/aware on
+
+which is what a GPU build would pick, and falls back to `neigh half newton on`
+(third field of a `CASES` line) for the styles that require a half neighbor
+list.
+
+## Building
+
+`tier1-packages.cmake` turns on the packages these decks need and nothing
+else.  The two executables the passes below want, from the top of the
+repository:
+
+    cmake -S cmake -B build-sync -G Ninja \
+      -C .github/dev-docs/kokkos-tier1-verification/tier1-packages.cmake \
+      -D PKG_KOKKOS=on -D Kokkos_ENABLE_SERIAL=on -D BUILD_MPI=on \
+      -D CMAKE_BUILD_TYPE=Release -D KOKKOS_DEBUG_SYNC=on \
+      -D BUILD_SHARED_LIBS=on -D DOWNLOAD_POTENTIALS=off -D ENABLE_TESTING=on
+
+    cmake -S cmake -B build-poison -G Ninja \
+      -C .github/dev-docs/kokkos-tier1-verification/tier1-packages.cmake \
+      -D PKG_KOKKOS=on -D Kokkos_ENABLE_SERIAL=on -D BUILD_MPI=on \
+      -D CMAKE_BUILD_TYPE=RelWithDebInfo \
+      -D KOKKOS_DEBUG_SYNC=on -D KOKKOS_DEBUG_SYNC_ASAN=on \
+      -D BUILD_SHARED_LIBS=on -D DOWNLOAD_POTENTIALS=off
+
+`step0` and `mpi` only compare styles against each other and run in any
+build; the detector passes need `build-sync`, and `poison` needs
+`build-poison`.
+
+## Usage
+
+    ./run_checks.sh step0       # thermo: plain CPU styles vs KOKKOS styles, same binary
+    ./run_checks.sh detect      # LMP_KOKKOS_WATCH / LMP_KOKKOS_STALE detectors
+    ./run_checks.sh mpi         # step0 again under two MPI ranks
+    ./run_checks.sh detect_mpi  # detect again under two MPI ranks
+    ./run_checks.sh poison      # LMP_KOKKOS_POISON, needs the ASan build
+
+`step0` runs each deck twice from the same executable, once with the plain
+styles and once with `-sf kk`, and compares every thermo column with
+`cmp.py`.  A column is equal when `|a-b| <= 1e-8 + 1e-6*max(|a|,|b|)`; the
+reported number is the ratio to that tolerance, so anything above 1.0 fails.
+
+The `mpi` and `detect_mpi` passes repeat the same cases under
+`mpirun -np 2`, which is what exercises the exchange and border
+communication; `MPIRUN` can be overridden in the environment.
+
+`detect` needs an executable built with `-D KOKKOS_DEBUG_SYNC=on`; see
+`../kokkos-sync-debugging.md` for what the reports mean and for the
+poison-mode build that pinpoints a stale access.  `poison` runs that build
+(`build-poison/lmp`, or `POISON_LMP`) with `LMP_KOKKOS_POISON=1`, which
+makes the non-current side of every view unreadable so AddressSanitizer
+stops at the offending instruction; it is the only detector that sees a
+read through a plain pointer.  Judge its output by
+diffing the reported labels (array plus the routine that touched it)
+against a run of a style of the same shape that is known to be clean, not
+by their absolute number: the KOKKOS infrastructure reports a fixed set of
+labels for every input.  `LMP` and `OUT` can be set in the environment;
+`LMP` defaults to `build-sync/lmp` at the top of the repository.
+
+## Layout
+
+`common_lj.mod` (LJ melt), `common_sphere.mod` (sphere-style melt) and
+`mol_head.mod` (`data.fourmol`, read from `unittest/force-styles/tests`) are
+the shared setups; each `in.*` deck includes one of them and cycles through a
+family of styles with an index variable, so one deck covers a whole work
+package.  `common_lj.mod` leaves the boundary in an index variable `bnd` so a
+deck that needs a non-periodic or shrink-wrapped z can set it before the
+include.

.github/dev-docs/kokkos-tier1-verification/cmp.py

@@ -0,0 +1,40 @@
+#!/usr/bin/env python3
+"""Compare the thermo rows of two LAMMPS screen dumps.
+
+A column is considered equal when |a-b| <= ATOL + RTOL*max(|a|,|b|).  The
+absolute term matters: several decks print quantities that sit at round-off
+around zero (a constraint force that should cancel, the fmax of a converged
+minimization), where a pure relative test reports a huge difference for an
+absolute difference of 1e-16.  The reported number is the ratio to that
+tolerance, so 1.0 is exactly at the limit and anything above it fails.
+"""
+import sys, re, math
+
+ATOL = 1.0e-8
+RTOL = 1.0e-6
+
+num = re.compile(r'^\s*\d+((\s+-?(\d+\.?\d*|nan|inf)([eE][-+]?\d+)?){2,})\s*$')
+
+def rows(f):
+    out = []
+    try: lines = open(f, errors="replace")
+    except OSError: return out
+    for line in lines:
+        if "%" in line or "CPU" in line: continue
+        if num.match(line): out.append([float(x) for x in line.split()])
+    return out
+
+a, b, label = rows(sys.argv[1]), rows(sys.argv[2]), sys.argv[3]
+if not a or not b:
+    print("%-46s %-12s %s" % (label, "NO-OUTPUT", "cpu=%d kk=%d rows" % (len(a), len(b)))); sys.exit()
+if len(a) != len(b):
+    print("%-46s %-12s %s" % (label, "ROWMISMATCH", "cpu=%d kk=%d" % (len(a), len(b)))); sys.exit()
+m = 0.0; nan = 0
+for ra, rb in zip(a, b):
+    for u, v in zip(ra, rb):
+        if math.isnan(u) or math.isnan(v):
+            if not (math.isnan(u) and math.isnan(v)): nan += 1
+            continue
+        m = max(m, abs(u - v) / (ATOL + RTOL * max(abs(u), abs(v))))
+verdict = "OK" if (m <= 1.0 and nan == 0) else "DIVERGED"
+print("%-46s %-12s %.3e%s" % (label, verdict, m, "  nan-mismatch=%d" % nan if nan else ""))

.github/dev-docs/kokkos-tier1-verification/common_lj.mod

@@ -0,0 +1,17 @@
+# shared LJ melt setup used by the fix/compute/region decks
+units           lj
+atom_style      atomic
+atom_modify     map array
+variable        bnd index "p p p"
+boundary        ${bnd}
+lattice         fcc 0.8442
+region          box block 0 8 0 8 0 8
+create_box      2 box
+create_atoms    1 box
+mass            * 1.0
+set             group all type/fraction 2 0.35 5738
+velocity        all create 1.2 87287 loop geom
+pair_style      lj/cut 2.5
+pair_coeff      * * 1.0 1.0 2.5
+neighbor        0.3 bin
+neigh_modify    delay 0 every 1 check yes

.github/dev-docs/kokkos-tier1-verification/common_sphere.mod

@@ -0,0 +1,15 @@
+units           lj
+atom_style      sphere
+atom_modify     map array
+lattice         fcc 0.8442
+region          box block 0 8 0 8 0 8
+create_box      1 box
+create_atoms    1 box
+variable        dia index 1.0
+variable        den index 1.0
+set             group all diameter ${dia} density ${den}
+velocity        all create 1.2 87287 loop geom
+pair_style      lj/cut 2.5
+pair_coeff      * * 1.0 1.0 2.5
+neighbor        0.3 bin
+neigh_modify    delay 0 every 1 check yes

.github/dev-docs/kokkos-tier1-verification/in.asphere

@@ -0,0 +1,24 @@
+units           lj
+atom_style      hybrid ellipsoid dipole
+atom_modify     map array
+lattice         sc 0.4
+region          box block 0 6 0 6 0 6
+create_box      1 box
+create_atoms    1 box
+set             group all mass 1.0
+set             group all shape 1.0 0.7 0.5
+set             group all quat/random 18238
+set             group all dipole/random 98347 1.0
+velocity        all create 1.0 87287 loop geom
+pair_style      lj/cut 2.5
+pair_coeff      * * 1.0 1.0 2.5
+neighbor        0.3 bin
+neigh_modify    delay 0 every 1 check yes
+compute         q all property/atom quatw quati quatj quatk
+fix             1 all nve/asphere/noforce
+fix             at all addtorque/atom 0.01 0.02 0.03
+fix             st all settorque/atom 0.0 0.0 0.05
+fix             ps all propel/self dipole 0.05
+thermo_style    custom step temp pe ke etotal press
+thermo          5
+run             20 post no

.github/dev-docs/kokkos-tier1-verification/in.bonded

@@ -0,0 +1,36 @@
+variable        tdir index ../../../unittest/force-styles/tests
+units           real
+atom_style      full
+atom_modify     map array
+pair_style      zero 10.0
+bond_style      zero
+angle_style     zero
+dihedral_style  zero
+improper_style  zero
+read_data       ${tdir}/data.fourmol
+pair_style      lj/cut 8.0
+pair_coeff      * * 0.02 3.0
+variable        ts index linear
+bond_style      table ${ts} 1000
+bond_coeff      1 ${tdir}/bond_table.txt harmonic_1
+bond_coeff      2 ${tdir}/bond_table.txt harmonic_2
+bond_coeff      3 ${tdir}/bond_table.txt harmonic_3
+bond_coeff      4 ${tdir}/bond_table.txt harmonic_4
+bond_coeff      5 ${tdir}/bond_table.txt harmonic_5
+angle_style     table ${ts} 1000
+angle_coeff     1 ${tdir}/angle_table.txt harmonic_1
+angle_coeff     2 ${tdir}/angle_table.txt harmonic_2
+angle_coeff     3 ${tdir}/angle_table.txt harmonic_3
+angle_coeff     4 ${tdir}/angle_table.txt harmonic_4
+dihedral_style  table ${ts} 3600
+dihedral_coeff  1 ${tdir}/harmonic_dihedral.txt HARMONIC_1
+dihedral_coeff  2 ${tdir}/harmonic_dihedral.txt HARMONIC_2
+dihedral_coeff  3 ${tdir}/harmonic_dihedral.txt HARMONIC_3
+dihedral_coeff  4 ${tdir}/harmonic_dihedral.txt HARMONIC_4
+dihedral_coeff  5 ${tdir}/harmonic_dihedral.txt HARMONIC_5
+neighbor        2.0 bin
+neigh_modify    delay 0 every 1 check yes
+fix             1 all nve
+thermo_style    custom step temp pe ebond eangle edihed press
+thermo          5
+run             20 post no

.github/dev-docs/kokkos-tier1-verification/in.bonded_cut

@@ -0,0 +1,25 @@
+variable        tdir index ../../../unittest/force-styles/tests
+units           real
+atom_style      full
+atom_modify     map array
+pair_style      zero 10.0
+bond_style      zero
+angle_style     zero
+dihedral_style  zero
+improper_style  zero
+read_data       ${tdir}/data.fourmol
+pair_style      lj/cut 8.0
+pair_coeff      * * 0.02 3.0
+variable        ts index linear
+dihedral_style  table/cut ${ts} 3600
+dihedral_coeff  1 aat 1.00 170 180 ${tdir}/harmonic_dihedral.txt HARMONIC_1
+dihedral_coeff  2 aat 1.00 170 180 ${tdir}/harmonic_dihedral.txt HARMONIC_2
+dihedral_coeff  3 aat 0.50 175 180 ${tdir}/harmonic_dihedral.txt HARMONIC_3
+dihedral_coeff  4 aat 1.00 170 180 ${tdir}/harmonic_dihedral.txt HARMONIC_4
+dihedral_coeff  5 aat 0.25 170 175 ${tdir}/harmonic_dihedral.txt HARMONIC_5
+neighbor        2.0 bin
+neigh_modify    delay 0 every 1 check yes
+fix             1 all nve
+thermo_style    custom step temp pe edihed press
+thermo          5
+run             20 post no

.github/dev-docs/kokkos-tier1-verification/in.computes

@@ -0,0 +1,21 @@
+include         common_lj.mod
+fix             1 all nve
+group           half type 1
+compute         ke1 all ke
+compute         ke2 half ke
+compute         com1 all com
+compute         com2 half com
+compute         in1 all inertia
+compute         gy1 all gyration
+compute         gyt half gyration
+compute         ce1 all centro/atom fcc
+compute         ce2 all centro/atom fcc axes yes
+compute         hx1 all hexorder/atom
+compute         en1 all entropy/atom 0.1 2.0
+compute         sce all reduce sum c_ce1
+compute         shx all reduce sum c_hx1[1]
+compute         sen all reduce sum c_en1
+thermo_style    custom step c_ke1 c_ke2 c_com1[1] c_com2[3] c_in1[1] c_in1[4] &
+                c_gy1 c_gyt[1] c_sce c_shx c_sen
+thermo          5
+run             20 post no

.github/dev-docs/kokkos-tier1-verification/in.cs

@@ -0,0 +1,29 @@
+include         mol_head.mod
+variable        ps index coul/long/cs
+if "${ps} == coul/long/cs" then "pair_style ${ps} 9.0" &
+elif "${ps} == coul/wolf/cs" "pair_style ${ps} 0.2 9.0" &
+elif "${ps} == born/coul/wolf/cs" "pair_style ${ps} 0.2 9.0 9.0" &
+elif "${ps} == born/coul/dsf/cs" "pair_style ${ps} 0.1 9.0 9.0" &
+elif "${ps} == born/coul/dsf" "pair_style ${ps} 0.1 9.0 9.0" &
+elif "${ps} == born/coul/long/cs" "pair_style ${ps} 9.0 9.0" &
+elif "${ps} == buck/coul/long/cs" "pair_style ${ps} 9.0 9.0" &
+elif "${ps} == lj/cut/coul/long/cs" "pair_style ${ps} 9.0 9.0" &
+else "pair_style ${ps} 9.0 9.0"
+if "${ps} == coul/long/cs" then "pair_coeff * *" &
+elif "${ps} == coul/wolf/cs" "pair_coeff * *" &
+elif "${ps} == born/coul/wolf/cs" "pair_coeff * * 1.0 0.3 0.0 1.0 0.5" &
+elif "${ps} == born/coul/dsf/cs" "pair_coeff * * 1.0 0.3 0.0 1.0 0.5" &
+elif "${ps} == born/coul/dsf" "pair_coeff * * 1.0 0.3 0.0 1.0 0.5" &
+elif "${ps} == born/coul/long/cs" "pair_coeff * * 1.0 0.3 0.0 1.0 0.5" &
+elif "${ps} == buck/coul/long/cs" "pair_coeff * * 100.0 0.3 10.0" &
+elif "${ps} == lj/class2/coul/long/cs" "pair_coeff * * 0.02 3.0" &
+else "pair_coeff * * 0.02 3.0"
+if "${ps} == coul/long/cs" then "kspace_style ewald 1.0e-6" &
+elif "${ps} == born/coul/long/cs" "kspace_style ewald 1.0e-6" &
+elif "${ps} == buck/coul/long/cs" "kspace_style ewald 1.0e-6" &
+elif "${ps} == lj/cut/coul/long/cs" "kspace_style ewald 1.0e-6" &
+elif "${ps} == lj/class2/coul/long/cs" "kspace_style ewald 1.0e-6"
+fix             1 all nve
+thermo_style    custom step temp pe evdwl ecoul elong press
+thermo          5
+run             20 post no

.github/dev-docs/kokkos-tier1-verification/in.fep

@@ -0,0 +1,21 @@
+include         mol_head.mod
+variable        ps index lj/cut/soft
+if "${ps} == coul/cut/soft"  then "pair_style ${ps} 1 10.0 9.0"     &
+elif "${ps} == coul/long/soft"  "pair_style ${ps} 1 10.0 9.0"       &
+elif "${ps} == morse/soft"      "pair_style ${ps} 4 0.9 10.0"       &
+elif "${ps} == lj/class2/soft"  "pair_style ${ps} 2 0.5 10.0"       &
+elif "${ps} == lj/cut/coul/cut/soft" "pair_style ${ps} 2 0.5 10.0 9.0 9.0" &
+elif "${ps} == lj/cut/coul/long/soft" "pair_style ${ps} 2 0.5 10.0 9.0"    &
+elif "${ps} == lj/charmm/coul/long/soft" "pair_style ${ps} 2 0.5 10.0 8.0 9.0" &
+else "pair_style ${ps} 2 0.5 10.0"
+if "${ps} == morse/soft" then "pair_coeff * * 0.2 2.0 3.0 0.9" &
+elif "${ps} == coul/cut/soft" "pair_coeff * * 0.9" &
+elif "${ps} == coul/long/soft" "pair_coeff * * 0.9" &
+else "pair_coeff * * 0.02 3.0 0.9"
+if "${ps} == coul/long/soft" then "kspace_style ewald 1.0e-6" &
+elif "${ps} == lj/cut/coul/long/soft" "kspace_style ewald 1.0e-6" &
+elif "${ps} == lj/charmm/coul/long/soft" "kspace_style ewald 1.0e-6"
+fix             1 all nve
+thermo_style    custom step temp pe evdwl ecoul elong press
+thermo          5
+run             20 post no

.github/dev-docs/kokkos-tier1-verification/in.fixes

@@ -0,0 +1,21 @@
+variable        bnd index "p p f"
+include         common_lj.mod
+fix             cnf all wall/reflect zlo EDGE zhi EDGE
+fix             1 all nve
+fix             sf  all store/force
+fix             fg  all flow/gauss 1 0 0
+fix             who all wall/harmonic/outside zlo 0.5 1.0 1.0 2.0 units box
+thermo_style    custom step temp pe ke etotal press f_fg[1]
+thermo          5
+run             20 post no
+unfix           sf
+unfix           fg
+unfix           who
+unfix           1
+fix             2 all nvk
+thermo_style    custom step temp pe ke etotal press
+run             10 post no
+unfix           2
+fix             3 all nve
+fix             pv all propel/self velocity 0.02
+run             10 post no

.github/dev-docs/kokkos-tier1-verification/in.fixes_sphere

@@ -0,0 +1,9 @@
+include         common_sphere.mod
+fix             1 all nve/sphere
+fix             dc  all damping/cundall 0.1 0.1
+fix             vn  all viscous/nonlinear 0.05 1.0
+fix             at  all addtorque/atom 0.01 0.02 0.03
+fix             st  all settorque/atom 0.0 0.0 0.05
+thermo_style    custom step temp pe ke etotal press
+thermo          5
+run             20 post no