Jukka Aho d676ab3bba perf(benchmark): Add CPU nodal assembly scalability benchmark ...

New benchmark testing nodal assembly performance on CPU:
- 438 lines implementing three execution modes
- Single-threaded baseline (reference performance)
- Multi-threaded using @threads (measures scaling efficiency)
- Partitioned mode (simulates multi-GPU with explicit partitions)

Features:
- Hex8 mesh generation (structured hexahedral elements)
- Node-to-element inverse connectivity building
- Mesh partitioning with ghost nodes and interface detection
- Performance metrics: throughput (Mnodes/s), speedup, efficiency
- Correctness verification (compares results to baseline)

Test mesh sizes: 20³, 40³, 60³ (8K to 216K nodes)
Measures: execution time, speedup vs baseline, parallel efficiency
Interface overhead calculation for partitioned mode

Run with: julia --project=. -t 8 benchmarks/nodal_assembly_scalability.jl

2025-11-09 16:04:39 +02:00

..

field_storage_comparison.jl

perf(benchmarks): Add field storage performance comparison script

2025-11-09 11:07:53 +02:00

multigpu_mpi_benchmark.jl

feat(gpu): Multi-GPU MPI benchmark with nodal assembly

2025-11-09 15:59:00 +02:00

multigpu_results_2025-11-09.md

feat(gpu): Multi-GPU MPI benchmark with nodal assembly

2025-11-09 15:59:00 +02:00

nodal_assembly_scalability.jl

perf(benchmark): Add CPU nodal assembly scalability benchmark

2025-11-09 16:04:39 +02:00

tet10_derivatives_benchmark.jl

docs: Benchmark manual vs AD derivatives for Tet10

2025-11-09 03:41:47 +02:00

VALIDATION_RESULTS.md

docs(benchmarks): Add validation results for field storage design

2025-11-09 11:08:22 +02:00