Add Level 3 multi-GPU applications and validation safeguards - #4

Draft

bowencui123 wants to merge 25 commits into

mainfrom

level3/second-batch-bringup

Commits on Sep 5, 2026

Level 3 first batch: audit, build strategy, LAMMPS/SPARTA/WarpX/SPECFEM3D/nekRS bring-up
SWE-bench
committed
da35285
Level 3 first batch: correctness/reproducibility fixes and nekRS coarse-solver decision
SWE-bench
committed
366b72f

Commits on Sep 6, 2026

Level 3 tools: per-profile paths and static-cudart backend check for the second batch
SWE-bench
committed
9327c26
Level 3 second batch: Nyx 26.09 (CUDA sm_100) built, validated at 1/2/4 GPUs
SWE-bench
committed
769482f
Level 3 tools: clear conda build variables for system-toolchain builds; real grep; robust static-cudart check
SWE-bench
committed
0b9daaf
Nyx: heating/cooling variant (SUNDIALS 7.2.1 CUDA) validated at 1/2/4 GPUs; synthetic strong deck
SWE-bench
committed
1f7e14f
Level 3 second batch: CP2K v2026.2 (CUDA sm_100, DBCSR 2.10.0) built, validated at 1/2/4 GPUs
SWE-bench
committed
b04db93
Level 3 tools: negative/positive tests for the second-batch application checkers
SWE-bench
committed
352cbb3
Level 3 second batch: QMCPACK 4.4.0 (OpenMP offload + CUDA, sm_100) with a private LLVM 23.1.0 toolchain
SWE-bench
committed
166657e
Level 3 tools: clear the login shell's foreign include/library search paths for system-toolchain builds
SWE-bench
committed
32e25e9
Level 3 second batch: DFT-FE 1.2.0 (CUDA sm_100) with a private deal.II 9.6.2 / ELPA 2026.02.001 stack
SWE-bench
committed
62f48e5
CP2K: link and run against the toolchain OpenBLAS (not the conda pthreads one); BLAS-resolution guard
SWE-bench
committed
7408614
Nyx: heat/cool strong-scaling timings and 8/40/80 dry-run record
SWE-bench
committed
c7b90f9
QMCPACK: walker-population guard, 256-walker strong/weak series, results
SWE-bench
committed
1e5d00d
GEOS: develop b7a0f133 + TPL 361-1070 on CUDA 13.2/sm_100, beam-bending validation 1/2/4 GPUs
SWE-bench
committed
e354870
DFT-FE: ELPA GPU probe parses ELPA's analytic-test errors; validation 1/2/4 PASS with the probe
SWE-bench
committed
e91119c
GEOS: record the unit-test probe (254/261; five flow/well physics tests fail on this GPU build)
SWE-bench
committed
fbaa962
Level 3 second batch: status, audit and build-strategy documentation
SWE-bench
committed
feaf8a3
docs: add CLAUDE.md (working notes for Claude Code sessions in this repository)
SWE-bench
committed
0bf3eb2

Commits on Sep 7, 2026

Level 3 tools: allow-listed environment wrapper and HPCPERF_L3_RUN_SUBDIR for regression campaigns
SWE-bench
committed
6711980
Nyx: strict fcompare wrapper, I_R_CHECK_PENDING verdict, tolerance provenance corrected, negative tests
SWE-bench
committed
f231217
Nyx run.sh: run directory through $L3_RUN_SUBDIR (left out of 6711980)
SWE-bench
committed
fc4d2a1
CLAUDE.md: regression run trees (HPCPERF_L3_RUN_SUBDIR) and the allow-listed environment wrapper
SWE-bench
committed
74c1d4d

Commits on Sep 8, 2026

Level 3 tools/Nyx: box-layout classification, verdict classes, credential deny-rule
SWE-bench
committed
1a6a60e
Level 3 status: joint-HEAD regression recorded, Nyx heat/cool PENDING, second-batch table
SWE-bench
committed
0dd99a1