feat: add 3D reacting mixing-layer example (H2/N2-air, Mach 1.5) + hcid=371 - #1758
Open
ecisneros8 wants to merge 13 commits into
Open
feat: add 3D reacting mixing-layer example (H2/N2-air, Mach 1.5) + hcid=371#1758ecisneros8 wants to merge 13 commits into
ecisneros8 wants to merge 13 commits into
Conversation
mpirun is not the correct MPI launcher on NCSA DeltaAI; srun (matching delta.mako's working pattern) is required for batch submissions to succeed.
The dai-gpu module set never set LD_LIBRARY_PATH for libnvToolsExt.so.1, which nvfortran's OpenACC/CUDA runtime links against but the loaded cudatoolkit/25.5_12.9 module doesn't provide (only the SDK's bundled CUDA 11.8 tree ships it), causing GPU binaries to fail to launch with a missing shared library error.
…1.5) New hcid=371 (3dHardcodedIC.fpp): extends hcid=370's 3D extrusion of a 2D flamelet profile with a closed-form spanwise (z) modulation of the in-plane velocity perturbation, so the IC has genuine 3D content from step 0 without relying on mixlayer_perturb (whose fixed wavenumber range is meaningless at this case's millimeter length scale). New example examples/3D_reacting_mixing_layer/: a supersonic (Mach_c=1.5) temporally-evolving H2/N2-air mixing layer initialized from a 1-D flamelet solve (flamelet_ic.py), diluted to X_H2=0.5 (pure H2's high sound speed otherwise forces an unphysically extreme velocity split to reach Mach_c=1.5), with Unity-Lewis transport to match the flamelet solve's own diffusivity assumption. Validated over 165k+ steps with no numerical instability.
z_cc(p) - z_cc(0) is a per-rank quantity: s_mpi_decompose_computational_domain overwrites p and z_domain with this rank's slab before the grid is built, so the imposed wavenumber scaled with num_procs_z and the IC depended on the decomposition. It also fell one cell short of L_z, leaving the modulation discontinuous across the periodic z wrap even on one rank. Use p_glb cells of the uniform cell width instead, and correct the comment and case.md: only the cross-stream component is modulated, and the spanwise component is set from it, not the full in-plane perturbation.
The 3D extrusion read only bounds-checked its row indices, so a file that is merely too fine stays in range and silently supplies a corner of itself. Check spacing and alignment up front, as hcid=274 already does for its full-field read.
perturb_xy defaulted seed=None, so np.random.default_rng drew from OS entropy. IC/ is gitignored and regenerated on every checkout, which made the example irreproducible run to run. Make the seed a required argument, set it in case.py, and key the IC cache on it. Also bin the phase blocks relative to x_coord[0] (binning the raw coordinate clipped every x < 0 into block 0, leaving two of five blocks unused), write the IC with savetxt rather than ~1.6M per-element writes, and record the mechanism's provenance in sandiego.yaml.
The Example sweep caps 3D cases to 25^3 while case.py writes its IC at the declared 210x560. Offsets rounded to zero, so the run read file rows 1-26 in x and y: a pure-oxidizer sliver where the perturbation mollifier is identically zero. Both hcid=371 statements then operated on zeros, and the golden covered nothing. Skip it in the Example sweep like the 2D siblings and register it under Chemistry with --scale 0.05, which drives case.py's own grid so the IC matches. The new golden carries nonzero spanwise momentum at step 0.
The literal 25.5 path breaks silently when dai-all's nvidia module bumps. Prefer $NVHPC_ROOT, keep the validated path as the fallback, and drop the trailing colon when LD_LIBRARY_PATH is empty.
Covers the stream conditions, grid, hcid=371's spanwise modulation, the seeded IC and its cache, and the --scale/--hot flags.
web.eng.ucsd.edu serves an untrusted certificate, so lychee fails on the link the 3D_reacting_mixing_layer README cites. Same treatment as the HPE Cray docs entry.
# Conflicts: # .lychee.toml
Lines of Code
|
sbryngelson
approved these changes
Aug 25, 2026
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## master #1758 +/- ##
==========================================
- Coverage 61.67% 60.25% -1.43%
==========================================
Files 84 84
Lines 21619 21846 +227
Branches 3196 3218 +22
==========================================
- Hits 13334 13163 -171
- Misses 6093 6280 +187
- Partials 2192 2403 +211 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Description
Adds a new 3D temporally-evolving reacting mixing-layer example (H2/N2-air, convective
Mach 1.5) initialized from a 1D flamelet solve, plus a new hardcoded IC (
hcid=371) thatextends
hcid=370's 3D extrusion of a 2D profile with a closed-form spanwise (z)modulation of the in-plane velocity perturbation, giving the IC genuine 3D content from
step 0 instead of staying z-invariant. Also registers
hcid=371incase_validator.py's extrusion-hcid check and documents it incase.md.Two unrelated NCSA DeltaAI cluster fixes are included: the batch template was using
mpirun, which is not the correct launcher on DeltaAI (srunis, matchingdelta.mako's working pattern), and thedai-gpumodule set never setLD_LIBRARY_PATHforlibnvToolsExt.so.1, which nvfortran's OpenACC/CUDA runtimerequires but the loaded
cudatoolkitmodule doesn't provide.Type of change (delete unused ones)
Testing
./mfc.sh precheck -j 8passes (formatting, spelling, toolchain/source lint, docreferences, parameter docs, example case validation).
./mfc.sh build -j 8builds all three targets (pre_process,simulation,post_process) cleanly —hcid=371lives in the sharedsrc/common/include/3dHardcodedIC.fpp, included by bothpre_processandsimulation../mfc.sh test -j 8: full suite passes (681 tests, plus the new3D -> Example -> reacting_mixing_layergolden generated and added intests/2E1858B2/).(not part of the regression suite at that step count — the golden covers the
standard 50-step example smoke test). No CPU-vs-GPU comparison run was done.
hcid=371is only used viapre_process'sicpppatches in this example (noimmersed boundaries), so it hasn't been exercised through the IB patch path in
src/simulation/m_ib_patches.fpp, which also includes this shared file.srunbatch submission andsuccessful GPU binary launch after the
LD_LIBRARY_PATHfix).Checklist
Check these like this
[x]to indicate which of the below applies.See the developer guide for full coding standards.
GPU changes (expand if you modified
src/simulation/)AI code reviews
Reviews are not retriggered automatically. To request a review, comment on the PR:
@claude full review— Claude full review (also triggers on PR open/reopen/ready)claude-full-review— Claude full review via label