Summary

• Reorganize tests/ut/cpp/ into domain-based subdirectories: hierarchical/, pto2_a2a3/, pto2_a5/, types/, hardware/, platform/
• Expand C++ GoogleTest coverage from 10 to 30+ test binaries covering PTO2 runtime internals: ring buffer (incl. edge cases), scheduler (state + edge), tensormap (incl. edge), shared memory, ready queue, runtime graph/lifecycle/status, handshake, coupling, orchestrator fatal/submit, boundary edge, dispatch payload, core/submit types, worker manager, and platform stubs
• Add shared test_helpers.h header and stubs/ directory for reusable test infrastructure
• Add 5 new Python UT modules: test_elf_parser.py, test_env_manager.py, test_kernel_compiler.py, test_runtime_compiler.py, test_toolchain_setup.py
• Update conftest.py with improved path setup for isolated UT execution
• Update docs/testing.md, docs/ci.md for new UT layout; add docs/ut-test-suite.md describing the full test inventory
• Minor fixes in runtime_compiler.py for testability

C++ test directory layout

tests/ut/cpp/
├── hardware/          # HCCL comm tests (requires_hardware label)
��── hierarchical/      # Orchestrator, ring, scheduler, scope, tensormap, worker_manager
├── platform/          # Platform host_log, memory_allocator stubs
├── pto2_a2a3/         # PTO2 A2A3 runtime: ring_buffer, scheduler, tensormap,
│                      #   shared_memory, ready_queue, coupling, handshake, types, edge cases
├── pto2_a5/           # PTO2 A5 runtime fatal tests
├── stubs/             # Shared test stubs (test_stubs.cpp)
├── types/             # Child memory, PTO types, tensor
├── test_helpers.h     # Shared test helpers
└── CMakeLists.txt     # All test targets with no_hardware / requires_hardware labels

Known defect-exposing tests

Several tests intentionally verify known src defects (documented in the test suite doc). These are preserved as regression anchors per project policy — they will pass once the underlying issues are fixed:

• TensorMap check_overlap() dimension mismatch (HIGH)
• TensorMap cleanup slot collision on large windows (HIGH)
• TensorMap lookup saturation with 16+ producers (MEDIUM)
• SharedMemory missing task_window_size >= 1 validation (LOW)

Test plan

• C++ UT build succeeds
• C++ UT no_hardware tests run
• Python UT passes (44 tests)
• CI ut-cpp job passes on ubuntu-latest
• Simulation tests unaffected (no runtime/platform code changes)