Files
FemLite/docs/v6_benchmark_matrix.md
T
shenw ee18a9b757
FemLite CI / build-test (push) Has been cancelled
Release FemLite 0.6.2 stability hardening
2026-07-02 08:26:46 +08:00

8.6 KiB

FemLite 0.6.x Benchmark And Acceptance Matrix

Date: 2026-07-01

Scope: executable evidence for the 0.6.0 nonlinear lifecycle foundation and rollback audit release.

Current local 0.6.0 evidence on build-codex-gcc:

  • cmake --build build-codex-gcc --parallel passed.
  • ctest --test-dir build-codex-gcc -R "^(solver_implicit|implicit_inp|restart|contact_mortar_report|contact_multiplier_report|cohesive_contact|particle_acceptance)$" --output-on-failure passed 7/7.
  • ctest --test-dir build-codex-gcc -L v5_acceptance --output-on-failure passed 39/39.
  • git diff --check passed with line-ending normalization warnings only.

Release Boundary

0.6.0 delivers lifecycle diagnostics and rollback audit infrastructure for the covered nonlinear subset. It does not deliver a complete industrial AL/mortar consistent tangent rewrite, generalized mixed-control continuation, cohesive contact mainline integration, DEM/FEM subcycling or a production roll-forming workflow.

0.6.0 Matrix

Area Code or output target Test/checker Acceptance evidence
Lifecycle core src/solver/fem_nonlinear_lifecycle.f90 nonlinear_lifecycle Stable event names, empty snapshot, event history, arc defaults and audit status transitions.
Static nonlinear lifecycle src/solver/fem_solver_static_nonlinear.f90 solver_implicit Increment/trial/iteration/result/commit/accept/reject/cutback events are observable without changing convergence rules.
Nonlinear Newmark lifecycle src/solver/fem_solver_newmark.f90 solver_implicit Dynamic nonlinear success and rejection paths expose lifecycle state.
Riks/arc lifecycle static nonlinear Riks path solver_implicit Accepted path records lifecycle accept, load factor, arc residual and path append state.
Material rollback audit src/element/fem_element_state.f90 solver_implicit, nonlinear_lifecycle Covered material state rollback is audited by checksum and status counters.
Contact rollback audit src/boundary/fem_contact_multiplier.f90 contact, restart Contact multiplier trial/commit/rollback contributes audit status without changing contact physics.
Cohesive audit src/boundary/fem_cohesive_state.f90 cohesive_contact Compact cohesive state rollback/commit emits lifecycle audit status.
Particle audit src/particle/fem_particle_audit.f90 particle_step Particle audit pass/fail maps into lifecycle commit/rollback-failed status.
Lifecycle CSV output continuation_path.csv solver_implicit Old path columns remain stable; lifecycle event/status, increment/iteration and accept/reject counters are appended.
Lifecycle restart restart version 5 optional block restart Accepted-increment lifecycle diagnostics round-trip; files without the optional block still read.

Regression Commands

cmake --build build-codex-gcc --parallel
ctest --test-dir build-codex-gcc -R "^(solver_implicit|implicit_inp|restart|contact_mortar_report|contact_multiplier_report|cohesive_contact|particle_acceptance)$" --output-on-failure
ctest --test-dir build-codex-gcc -L v5_acceptance --output-on-failure
git diff --check

Additional focused lifecycle evidence used during development:

ctest --test-dir build-codex-gcc -R "^(nonlinear_lifecycle|solver_implicit|restart|contact|cohesive_contact|particle_step)$" --output-on-failure

0.6.1 Diagnostic Hardening Matrix

FemLite 0.6.1 extends the 0.6.0 lifecycle foundation with additive diagnostics. It does not deliver a complete industrial AL/mortar consistent tangent rewrite, mixed-control continuation, cohesive contact mainline, DEM/FEM subcycling or production roll-forming workflow.

Area Code or output target Test/checker Acceptance evidence
Lifecycle summary fem_nonlinear_lifecycle, diagnostics API nonlinear_lifecycle, solver_implicit One compact query explains latest increment event, reason, convergence, residual, linear solver and audit state.
Cutback taxonomy fem_control, static nonlinear solver nonlinear_lifecycle, solver_implicit Last actual cutback reason is separate from final convergence failure reason.
Lifecycle report nonlinear_lifecycle.csv nonlinear_lifecycle, solver_implicit Empty, accepted and rejected/cutback rows include lifecycle, residual, cutback and linear solver fields.
Contact state diagnostics fem_contact_state, contact CSV contact Stable open/stick/slip/restick/separating/failed-projection counters and gap/slip/friction summaries.
Contact long-path smoke compact contact benchmark checkers contact two-block, sliding, shell-shell, shell-solid examples Existing compact paths fail loudly if state/residual/multiplier fields disappear.
AL/friction summaries multiplier report and restart checks contact_multiplier_report, restart Multiplier residual, augmented multiplier and supported friction summaries are reported and restart-stable.
Tangent audit report shell fem_tangent_audit_report contact_mortar_tangent Shared audit schema covers contact and non-contact categories with pass/missing/not-applicable counters.
Riks/arc diagnostics continuation_path.csv solver_implicit, implicit_inp Path rows distinguish load Riks and displacement control and expose arc residual, direction and reversal indicators.
Restart after accepted increments restart optional lifecycle block restart Two accepted increments are written/read through fresh models, then continued one more increment.
Linear solver coupling lifecycle/path rows solver_implicit, linear_solver_cg Method, preconditioner, iterations, status, residuals and fail class are visible in nonlinear rows.

Final local 0.6.1 evidence on build-codex-gcc:

cmake --build build-codex-gcc --parallel
ctest --test-dir build-codex-gcc -R "^(nonlinear_lifecycle|solver_implicit|implicit_inp|restart|contact|contact_mortar_report|contact_multiplier_report|cohesive_contact|particle_acceptance|linear_solver_cg)$" --output-on-failure
ctest --test-dir build-codex-gcc -R "^example_(contact_two_blocks_compression|contact_block_sliding|shell_shell_contact|shell_solid_contact)_benchmark$" --output-on-failure
ctest --test-dir build-codex-gcc -L v5_acceptance --output-on-failure
git diff --check

Results:

  • Targeted 0.6.1 regression command passed 10/10.
  • V5 acceptance subset passed 39/39.
  • Contact compact benchmark smoke passed 4/4.
  • Version smoke checks passed 2/2 after raising the public version to 0.6.1.

0.6.2 Stability Gate Matrix

FemLite 0.6.2 is a stability release for the 0.6.1 diagnostic surface. It does not add new algorithms. The release gate is intentionally small enough to run locally and broad enough to catch source-size regressions, schema breaks, restart incompatibilities and stale solver/contact diagnostics.

Gate item Required command Stability coverage
Build cmake --build build-codex-gcc --parallel Confirms all modules and tests still compile after hardening refactors.
Source-size hard gate ctest --test-dir build-codex-gcc -R "^source_size$" --output-on-failure Blocks oversized routines and would catch the 0.6.2 entry failures before release.
Targeted stability regression `ctest --test-dir build-codex-gcc -R "^(nonlinear_lifecycle solver_implicit
V5 acceptance subset ctest --test-dir build-codex-gcc -L v5_acceptance --output-on-failure Guards the established V5 executable acceptance set while 0.6.2 hardens diagnostics.
Diff hygiene git diff --check Rejects whitespace and merge-marker mistakes before release.

The 0.6.2 gate is not a replacement for future long-run research validation. It is the required local quality bar for the release hardening scope.

Final local 0.6.2 evidence on build-codex-gcc:

  • Build passed.
  • source_size passed 1/1.
  • Targeted stability regression passed 13/13.
  • V5 acceptance subset passed 39/39.
  • Version smoke checks passed 2/2.
  • git diff --check passed with line-ending normalization warnings only.

0.6.2 does not claim new algorithms. Complete industrial AL/mortar consistent tangents, mixed-control continuation, cohesive contact mainline, DEM/FEM subcycling, full shell/high-order geometric stiffness expansion, nonzero harmonic cyclic solving and production roll-forming workflows remain deferred.