Skip to content

Execution resource controls (WP4)

Summary

lic build gains --cores=N and --threads-per-core=M to size the native parallel runtime team, documents jobs vs cores vs threads vs SIMD lanes, and adds li-tests/execution_resources/ smoke tests.

Agent continuation

  1. Read compiler/common/include/li/resource_options.hpp, docs/language/decorators.md resource table.
  2. Run ./scripts/build.sh and ./li-tests/run_all.sh execution_resources.
  3. Stack PR on feat/lic-native-parallel (#208); merge #208 before this branch when landing.
  4. Blocked on: none for local dev; CI needs LLVM 22 + built lic.

Changed

  • compiler/common/resource_options.{hpp,cpp} — --cores, --threads-per-core, effective_runtime_team_size(), --threads precedence warning
  • compiler/lic/main.cpp — usage table; runtime_team_size from flags (no LI_OMP_THREADS setenv)
  • compiler/codegen/emit.cpp — li_parallel_for_i64 team constant; vectorized SIMD comments
  • compiler/mir/include/li/mir.hpp, compiler/mir/lower.cpp — MirDecorator.vectorized
  • runtime/li_rt.c — document build-time team_size preference
  • std/execution/decorators.li, docs/language/decorators.md, docs/compiler/build-pipeline.md
  • li-tests/execution_resources/, li-tests/run_all.sh (script_ok), li-tests/manifest.toml

Not changed

  • OpenMP removal / pthread pool implementation (parent #208)
  • --jobs compile parallelism semantics
  • @parallel(disjoint=) proof obligations (G-par still open)

Breaking

N/A — additive flags; LI_OMP_THREADS remains deprecated fallback when build omits team flags.

Security

N/A — no new trusted surface; LI_MAX_THREADS cap unchanged.

Performance

N/A — bench row not required; team size is user-configurable up to 64 threads.

Downstream

  • li-cursor-agents agent prompts may reference --cores after merge
  • Benchmarks CSV may add cores×tpc columns in a follow-up