Chemical Synthetic Turbulence#77
Open
cxflyer wants to merge 174 commits into
Open
Conversation
Merge from exapde/Exasim
Merge Teo's model template into master and update exasim APIs
Make a test that passes
update parmetis for exasim
… with the Poisson operator. The mapping as a first step is analytical but in the future it should come from the mesh deformation solver. Implementation follows Persson, Bonet and Peraire, 2009 and should be easily extensible to other flux terms.
… and Peraire, 2009. excellent agreement with exact solution. next step is to get the mapping, its deformation gradient and velocity from the mesh motion solver. this will enable solution of coupled problems where the fluid-solid interface is moving (due to deformation or ablation).
Per user request, the generated model is now a SHARED provider library
(libfrontend_model.{so,dylib}) instead of a static archive:
- exasim_add_external_builtin_model() gains a SHARED flag: the provider
takes Kokkos as compile flags/includes only and resolves Kokkos symbols
from the host executable at load time (-undefined dynamic_lookup on
macOS; host sets ENABLE_EXPORTS) — exactly one Kokkos runtime, avoiding
the duplicate-global-state double-free the static design guarded against
(see backend/Model/BuiltIn/CMakeLists.txt). The builtin fallthrough
archive (already PIC) is embedded in the provider.
- frontend-app template uses SHARED + ENABLE_EXPORTS +
CMAKE_LINK_DEPENDS_NO_SHARED, so a model change recompiles one TU and
relinks only the small provider library; exasimapp is never rebuilt.
- cmakecompile (all three languages) computes a SHA-256 over the kernel
set and rendered app sources and skips the build system entirely when it
matches the last successful build (.exasim_model_hash stamp): an
unchanged app goes straight to the solver run (~0.5s overhead).
Python: hashlib; Julia: SHA stdlib (added to Project.toml); MATLAB:
java.security.MessageDigest.
Validated: unchanged rerun = 0 TUs/0 links, cmake skipped; physics change =
1 TU + provider relink only (exasimapp mtime unchanged), wrong physics
correctly fails the QoI gate; frontend_{python,julia,matlab} all pass.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…gs and testing - EXASIM_PIP_INSTALL=ON: cmake --install runs pip install on frontends/Python with the configured interpreter and bakes the install prefix into the package (_installed_prefix.py, read by config.py), so 'import exasim' needs no PYTHONPATH or EXASIM_PREFIX afterwards. - EXASIM_JULIA_DEVELOP=ON: cmake --install runs Pkg.develop on the installed Exasim.jl, so 'using Exasim' needs no LOAD_PATH setup. Both OFF by default (they mutate user environments); forwarded through the superbuild along with EXASIM_FRONTENDS. - README: full feature-flag table with defaults, an everything-on configure line, and a 'Testing the install' section (run-tests.sh, per-test ctest invocations, running frontend tests by hand against any prefix, skip conditions, QoI gate). Validated: with both hooks ON, 'import exasim' resolves the baked prefix with no environment setup and 'using Exasim' loads via the developed package. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Upstream d440207 moved the legacy frontend flow to examples/build with a shared exasimfe executable and untracked the FrontendGenerated placeholder kernels. This branch's installed-package + external-model flow supersedes the frontend helper changes, so for the frontend flow files (cmakecompile/ runcode/exasim/preprocessing/initializepde across all three languages) the package versions are kept; upstream's multi-model compilecode->cmakecompile fix had already been applied / merges cleanly into the moved files. Taken from upstream: examples/CMakeLists.txt + exasimfeapp.cpp (legacy-flow consumer app, kept as-is), exasimapptest.m, the apps/builtinlibrary find_package modernization, the exasim_set_default_kokkos_dir() discovery helper, the FrontendGenerated kernel untracking + .gitignore additions, and the metis.h tweak. install/cmakecompile.{jl,m} and the pre-move frontend helper paths stay deleted. Full suite green after merge (consumers + all three frontends). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The repo is never written to by a build or install: - the superbuild rejects any binary dir inside the source tree (fatal, with the sibling-dir suggestion); the old lenient warning is gone - EXASIM_LIB_DIR moves from <repo>/lib to <binary-dir>/lib (Exasim/lib is deleted); the built-in model archives follow; exported link directories are wrapped in BUILD_INTERFACE/INSTALL_INTERFACE genexes so install(EXPORT) accepts the build-dir path - text2code lands in <superbuild-bin>/bin/text2code (EXASIM_T2C_OUTPUT_DIR); its location is forwarded to the inner build as EXASIM_TEXT2CODE for the builtin-model codegen and the bin/ install rule (no more <repo>/build) - tests default to the sibling build dir <repo>-build (run-tests.sh, run-consumer-tests.sh, frontend test wrapper, which now also honors EXASIM_PREFIX); the frontend prefix resolvers learn the sibling-build fallback (legacy <repo>/build/install kept as a last resort) - .gitignore: per-app .exasim/ dirs from pdeapps run inside the repo Validated: full cold build in ../Exasim-build, suite 5/5, repo clean, hygiene green; Exasim/lib and Exasim/build removed. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
-std=c++17 -> -std=c++20 on all solver targets, cxx_std_20 on the exported Exasim::headers interface (consumers now compile as C++20), and CMAKE_CXX_STANDARD 20 in the frontend-app template, the consumer tests, examples/, apps/builtinlibrary, and text2code. Vendored dependency builds (Kokkos, SymEngine, METIS) are unchanged. No removed-in-C++20 constructs in the tree; full suite green at C++20. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
CI (smoke-cpu): explicit Configure+Build, Test, and Test-summary steps — failures are attributable to a stage at a glance, and every run's summary page shows a per-test pass/fail/skip table (ctest --output-junit rendered into GITHUB_STEP_SUMMARY). Builds out of tree into ../Exasim-build; dep caches follow (new -oot cache keys). README: feature-flag table gains a consequences column (esp. what each option means for using the frontends); a test-inventory table marks exactly what runs on CI vs locally; 'The install prefix, in one place' documents the single prefix notion, who writes/reads it, and the EXASIM_PREFIX override; all build commands shown out-of-tree (cmake -S Exasim -B Exasim-build). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
pip/setuptools drops build/ and exasim.egg-info/ next to the package source when installing from a directory; remove them in the install step and gitignore them. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
cache-warm.yml restores the vendored Kokkos (serial) and SymEngine caches twice a week (rebuilding only on a key miss), refreshing GitHub's 7-day eviction clock so smoke-cpu never pays the ~15-20 min cold dependency build. Scheduled workflows run from the default branch only and only default-branch caches are repo-global, so the schedule takes effect once this reaches master; until then it is manually dispatchable. Keys are shared with smoke-cpu.yml (cross-referenced in both files). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…utes
The out-of-tree move silently broke dep caching: actions/cache rejects
relative paths containing '..' ('Invalid pattern ... not allowed' in the
post-job save), so every run rebuilt Kokkos and SymEngine cold. Point
BUILD_DIR and both cache paths at runner.workspace/Exasim-build (the
workspace parent, absolute, dot-free) in smoke-cpu and cache-warm.
docs: tighten the pip wording in the README/example Python usage notes.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Per-user cache (task: reuse model dynamic libraries across app dirs): - every successful build stores the (libfrontend_model, exasimapp) pair under ~/.exasim/cache/<modelID>/<digest>/ (EXASIM_CACHE_DIR overrides); any app directory with the same digest reuses it with zero compilation - relocatable pairs: the SHARED provider gets an @rpath install name and exasimapp gets @loader_path/$ORIGIN BUILD_RPATH - the digest hashes the kernel set, the app TEMPLATES (not the rendered files, which embed absolute paths and would be directory-specific), the variant + model ID, and the install identity (prefix + ExasimTargets mtime) — implemented in all three frontends - new ctest frontend_python_modelcache: a second app dir must hit the cache with no compilation; the test wrapper now exports a hermetic EXASIM_CACHE_DIR so tests never touch ~/.exasim - pyproject: include-package-data + exasim/**/*.bin (a pip install was missing masternodes.bin/gaussnodes.bin) CI: the runner.workspace expression context does not exist and actions/cache rejects '..' paths, so the previous layout broke workflow parsing / silently skipped cache saves. Both workflows now check the repo out into src/ with the build in build/ beside it — everything workspace-relative and dot-free. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…er-frontends-and-misc-cleanup
…rce change readpdeapp trimmed even an explicitly-set exasimpath at the last 'Exasim' substring, mangling any path that doesn't end in Exasim — on CI the checkout lives at .../Exasim/Exasim/src, so the builtin-model codegen got -I<workspace>/backend/... and failed to find SymEngineFunctionWrappers.hpp. An explicit path that already looks like an Exasim root (has backend/) is now used verbatim; the substring trim remains only as a fallback for inside-the-tree paths, and the cwd derivation still applies when no path was set. (Local runs never tripped this because the repo dir name ends in 'Exasim', making the trim a no-op.) Also give text2code_build BUILD_ALWAYS like exasim_build, so text2code source changes actually propagate into existing build dirs instead of being frozen by the ExternalProject stamp. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…c-cleanup Frontends as installed packages: import exasim / using Exasim / exasim_setup.m on the external-model build path
Python (24 files): drop the cwd/setpath.py exec hack; import exasim and use exasim.* / exasim.<Subpackage>.* call sites per the package exports. Julia (23 files): drop setpath.jl + bare submodule usings for 'using Exasim'; qualify Mesh/Gencode helpers not re-exported at top level. MATLAB (112 files): replace the strfind(cdir)/setpath.m header with exasim_setup.m via a depth-correct relative run(fullfile(...)). Mechanical migration only; no behavioral changes intended. Flagged oddballs kept working as-is (gpucompiler paths derived from cwd in poisson3d/test3d, preprocess_mesh6.m cdir reuse, ConvectionDiffusion/1D references a mesh1D that never existed in the new frontend). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Convert the MATLAB pdemodels (3D Poisson, axisymmetric Poisson, ALE
Euler vortex, linear elasticity) to pdemodel{N}.txt + pdeapp{N}.txt so
the build regenerates them via text2code instead of using the legacy
checked-in kernels (the existing pdeapp{N}.txt-presence switch in
backend/Model/BuiltIn/CMakeLists.txt picks them up automatically).
Generated kernels spot-verified equivalent to the checked-in ones.
Also remove the redundant pow(Expression, Expression) wrapper from
SymEngineFunctionWrappers.hpp: symengine/expression.h already defines
that exact signature (found via ADL), so every unqualified pow call in
a model was ambiguous — model 5 is the first builtin source to use pow.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Convert the MATLAB 2D axisymmetric 5-species reacting Navier-Stokes model (fluxaxial2d/sourceaxial2d/eosnd/fbouhdgaxialnd + the CNS5air thermo, transport, and kinetics stack) to pdemodel4.txt: NASA9 thermodynamics with tanh range switches, Blottner viscosity, Gupta diffusion, Wilke mixing, 5-reaction Park kinetics, and 9 HDG boundary conditions (incl. the partial-catalytic wall family, reproducing the legacy double-counted gamma*mdot quirk in BC9 verbatim). text2code: teach the generator to emit EoS and Sourcew kernels (KokkosEoS/EoSdu/EoSdw/HdgEoS, KokkosSourcew/HdgSourcew/ HdgSourcewonly) with the element-DG npe-layout ABI the discretization drivers expect; these outputs had never been exercised by the txt pipeline. Model 1 regenerates byte-identical; model 12 regenerates cleanly. Validation: generated kernels match the checked-in MATLAB-generated ones at machine precision (35 kernel comparisons incl. all analytic jacobians at a 4500 K test state); full suite 6/6. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…on CI) Promote the model 4 conversion harness to a first-class ctest: compile the legacy MATLAB-generated kernels (backend/Model/BuiltIn/model4/, kept as the reference) and the kernels THIS build regenerated from pdemodel4.txt into one binary (Kokkos parallel_for shimmed to a serial loop) and compare 35 outputs at a physical reacting-flow state (T~4500 K, reentry reference scales): every kernel family, all 9 HDG boundary conditions, and the analytic jacobians. Gate rtol 2e-9; observed agreement ~1e-13. Documented in tests/README.md and the workflow header so the verification is visible to reviewers. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Migrate examples to the frontend packages; complete builtin models 2-6 as text2code sources
Also update examples/exasimfe/CMakeLists.txt to prefer the PSAAP repo-local Exasim package under local/lib/cmake/Exasim before calling find_package(Exasim). This fixes the standalone exasimfe build by avoiding ABI mismatches caused by CMake resolving an older Exasim package from a different checkout.
…ging Unify Matlab, Python, and Julia frontend build behavior around the shared examples/exasimfe app and the local Exasim package layout. - simplify frontend cmakecompile flows to build the example app directly instead of rendering installed frontend-app templates - add shared/local build directory helpers and sharedbuild state across Matlab, Python, and Julia frontends - improve Exasim prefix discovery in source-tree workflows, including repo-local installs under Exasim/local - update Matlab, Python, and Julia startup/entry-point code to populate EXASIM_PREFIX automatically when missing - make backend runtime data lookup honor EXASIM_PREFIX before falling back to the compiled install path - update builtinlibrary and exasimfe CMake configuration to resolve the intended local Exasim package and support current frontend app linkage - revise exasimapptest to run Matlab/Julia/Python frontend smoke tests and builtinlibrary checks against the local package layout - adjust install rules to generate frontend setup/entry files with the install-time prefix for Matlab, Python, and Julia - stop the superbuild from installing exasim_build during cmake --build and route installation through the outer cmake --install step
- add a new shared-library example app under apps/sharedlibrary - add shared-library provider wiring for ExasimSolver setup - fix shared-library ABI selection to use the text2code ABI table - update Exasim solver headers for shared-library/provider support - refresh text2code-generated model sources and generated helper code - update text2code compiler logic to match the new generated interfaces - adjust Matlab setup output and ignore generated shared-library build artifacts
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
5 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.
Need copilot to review