PTO Runtime - Task Runtime Execution Framework

Modular runtime for building and executing task dependency graphs on Ascend devices with coordinated AICPU and AICore execution. Three independently compiled programs (Host .so, AICPU .so, AICore .o) work together through clearly defined APIs.

Quick Start

# Clone the repository
git clone <repo-url>
cd simpler

# Run the vector example (simulation, no hardware required)
python examples/scripts/run_example.py \
  -k examples/a2a3/host_build_graph/vector_example/kernels \
  -g examples/a2a3/host_build_graph/vector_example/golden.py \
  -p a2a3sim

PTO ISA headers are automatically cloned on first run. See Getting Started for manual setup and troubleshooting.

Platforms

Platform	Description	Requirements
a2a3	Real Ascend A2/A3 hardware	CANN toolkit (ccec, aarch64 cross-compiler)
a2a3sim	Thread-based A2/A3 simulation	gcc/g++ only (no Ascend SDK needed)
a5	Real Ascend A5 hardware	CANN toolkit (ccec, aarch64 cross-compiler)
a5sim	Thread-based A5 simulation	gcc/g++ only (no Ascend SDK needed)

Runtime Variants

Three runtimes under src/{arch}/runtime/, each with a different graph-building strategy:

Runtime	Graph built on	Use case
host_build_graph	Host CPU	Development, debugging
aicpu_build_graph	AICPU (device)	Reduced host-device transfer
tensormap_and_ringbuffer	AICPU (device)	Production workloads

See runtime docs per arch: a2a3, a5.

Testing

# Simulation tests (no hardware)
./ci.sh -p a2a3sim

# Hardware tests (requires Ascend device)
./ci.sh -p a2a3 -d 4-7 --parallel

# Python unit tests
pytest tests -m "not requires_hardware" -v

# C++ unit tests
cmake -B tests/ut/cpp/build -S tests/ut/cpp && cmake --build tests/ut/cpp/build && ctest --test-dir tests/ut/cpp/build --output-on-failure

See Testing Guide for details.

Environment Setup

source /usr/local/Ascend/ascend-toolkit/latest/bin/setenv.bash
export ASCEND_HOME_PATH=/usr/local/Ascend/ascend-toolkit/latest

Documentation

Document	Description
Chip-Level Architecture	L2 single-chip: three-program model (host/AICPU/AICore), API layers, handshake protocol
Distributed Level Runtime	L0–L6 level model, component composition (Orchestrator / Scheduler / Worker)
Task Flow	End-to-end data flow: Callable / TaskArgs / CallConfig handles, IWorker interface
Orchestrator	DAG submission internals: submit flow, TensorMap, Scope, Ring, task state machine
Scheduler	DAG dispatch internals: wiring/ready/completion queues, dispatch loop
Worker Manager	Worker pool, WorkerThread, THREAD/PROCESS modes, fork + mailbox mechanics
Roadmap	Hierarchical-runtime refactor — what has landed and what is still in flight
Getting Started	Setup, prerequisites, build process, configuration
Developer Guide	Directory structure, role ownership, conventions
Testing Guide	CI pipeline, test types, writing new tests

Per-arch docs

Document	a2a3 arch	a5 arch
Runtimes	a2a3/docs/runtimes.md	a5/docs/runtimes.md
Platform	a2a3/docs/platform.md	a5/docs/platform.md

License

This project is licensed under the CANN Open Software License Agreement Version 2.0. See the LICENSE file for the full license text.

References

• src/a2a3/platform/ - Platform implementations
• src/a2a3/runtime/ - Runtime implementations
• examples/a2a3/ - Examples organized by runtime
• examples/scripts/ - Test framework
• python/ - Python bindings and compiler