FEniCS / ffcx / 33203028276

86%

main: 85%

Build Type

Pull #865

github

Committed by web-flow

Commit Message

Add benchmark harness and three codegen memory/dead-store fixes (#868)

* Add golden-master numeric test for HyperElasticity's cell kernels

The demo's residual and Jacobian forms (P2 vector displacement, nonlinear
strain energy, Jacobian built via ufl.derivative()) are the deepest-nested
tensor composition anywhere in the test/demo corpus, and had zero numeric
correctness coverage -- demo/test_demos.py only checks that the demo
compiles, never asserts a result. Any future change to how nested tensor
expressions get scalarized or factorized had nothing to catch a silent
wrong-value regression against.

Freezes the current compiler's output for both cell kernels against fixed,
deterministic, non-degenerate coefficient/constant/coordinate data (stored in
test/data/hyperelasticity_golden.npz alongside the inputs, so the fixture is
self-contained and reproducible). Verified the check actually catches a
regression by perturbing one frozen value and confirming the test fails with
a clear mismatch report, then restoring it.

* Add generated-code performance benchmark harness

FFCx has no performance harness at all -- only -O1 correctness tests and
demo/test_demos.py, which merely compiles. Recent perf-stack work
(loop-nest ordering, quadrature-loop splitting) needed ad hoc measurement
to validate, and at least one attempted optimisation had to be reverted
after being found to regress performance. bench/ compiles each case at
-O3 -march=native (unlike the -O1 correctness tests, which hide the
vectoriser behaviour these benchmarks target) and times it in a tight C
loop via ctypes, avoiding Python/FFI call overhead that would otherwise
swamp the cheapest kernels. Static stack usage is read from a
-fstack-usage sidecar of the exact cached generated source.

Run with `python -m bench.run [name-substring ...]`.

* Remove dead zero-initialisers for fw and LICM temp arrays

Both `fw` scalars (integral_generator.py) and LICM's hoisted `temp_N`
arrays (optimizer.py) w... (continued)

Pull Request Pull Request #865: Stack codegen performance fixes: quadrature-loop split, tensor contraction, CSE, coordinate unrolling, LICM fusion

Coverage Stats

335 of 350 new or added lines in 7 files covered. (95.71%)

2 existing lines in 2 files now uncovered.

4529 of 5296 relevant lines covered (85.52%)

0.86 hits per line