pyproject.toml

[build-system]
requires = ["setuptools>=64", "setuptools_scm>=8", "wheel"]
build-backend = "setuptools.build_meta"

[project]
name = "amd-aorta"
# The version is not hard-coded here: it is derived from git tags at build time
# by setuptools_scm (see [tool.setuptools_scm]). ``dynamic`` tells setuptools the
# version is supplied by the build backend, so there is a single source of truth
# (the ``vX.Y.Z`` tag) that both released wheels and local/editable installs read.
dynamic = ["version"]
description = "PyTorch compute-communication overlap debugging toolkit with GPU hardware queue evaluation"
readme = "README.md"
license = {text = "MIT"}
requires-python = ">=3.10"
authors = [
    {name = "AMD ROCm Team"}
]
classifiers = [
    "Development Status :: 4 - Beta",
    "Intended Audience :: Developers",
    "License :: OSI Approved :: MIT License",
    "Programming Language :: Python :: 3",
    "Programming Language :: Python :: 3.10",
    "Programming Language :: Python :: 3.11",
    "Programming Language :: Python :: 3.12",
    "Topic :: Scientific/Engineering :: Artificial Intelligence",
    "Topic :: System :: Hardware",
]
keywords = ["gpu", "rocm", "amd", "pytorch", "cuda", "benchmark", "fsdp", "distributed-training", "hardware-queue"]

# Base dependencies (PyTorch installed separately - see requirements.txt)
dependencies = [
    "pyyaml>=6.0",
    "click>=8.0.0",  # For aorta CLI
]

[project.scripts]
aorta = "aorta.cli:main"

# Plugin discovery: workloads register here via aorta.workloads entry-point group.
# Public workloads (when they land) register here; private workloads
# register from downstream packages' pyproject.toml under the same group.
#
# ``_subprocess`` is platform-internal: it wraps an opaque user launch
# command for ``aorta probe`` (issue #188). The leading underscore both
# (a) prevents collision with user-facing workload names and (b) marks
# the entry as visibly platform-internal in ``aorta triage list-workloads``
# when that command lands. Direct ``aorta run --workload _subprocess``
# invocations fail at ``SubprocessWorkload.setup()`` because the
# reserved argv config key is only injected by ``aorta probe``.
[project.entry-points."aorta.workloads"]
_subprocess = "aorta.workloads._subprocess:SubprocessWorkload"
hrx = "aorta.workloads.hrx:HrxWorkload"
hrx_perf = "aorta.workloads.hrx_perf:HrxPerfWorkload"
inference = "aorta.workloads.inference:InferenceWorkload"
llm_determinism = "aorta.workloads.llm_determinism:LlmDeterminismWorkload"
race = "aorta.workloads.race:RaceWorkload"
# Single-process GPU smoke test (no torchrun); ideal for emulated/CI runs.
gpu_smoke = "aorta.workloads.gpu_smoke:GpuSmokeWorkload"
training = "aorta.workloads.training:TrainingWorkload"

# Side-effect-free isolation metadata. The dispatcher reads these entry-point
# values without importing workload modules, so ``--dry-run`` and ``auto``
# policy resolution cannot trigger torch/native-library imports. Third-party
# process-capable workloads register the same group alongside aorta.workloads.
[project.entry-points."aorta.workload_policies"]
_subprocess = "aorta.run.validation:IN_PROCESS_REQUIRED_POLICY"
llm_determinism = "aorta.run.validation:PROCESS_OPTIONAL_POLICY"
race = "aorta.run.validation:PROCESS_REQUIRED_POLICY"

[project.entry-points."aorta.workload_startup_env"]
race = "aorta.run.validation:race_startup_env"

[project.optional-dependencies]
# For analysis/visualization scripts
analysis = [
    "matplotlib>=3.7.0",
]

# For aorta-report CLI (full functionality)
report = [
    "numpy>=1.20.0",
    "pandas>=1.3.0",
    "openpyxl>=3.0.0",
    "matplotlib>=3.7.0",
    "seaborn>=0.12.0",
    "beautifulsoup4>=4.12.0",
]

# For hw_queue_eval subpackage
hw-queue = [
    "numpy>=1.20.0",
    "pandas>=1.3.0",
    "tabulate>=0.9.0",
]

# For hw_queue_eval profiling features
hw-queue-profiling = [
    "amd-aorta[hw-queue]",
    "matplotlib>=3.5.0",
    "seaborn>=0.12.0",
]

# For aorta agent closed-loop (LiteLLM proposer)
agent = [
    "litellm>=1.0.0",
]

# For aorta.ebpf kernel tracing module.
# The runtime requirements are the bpftrace binary and CAP_BPF/CAP_PERFMON
# (or sudo); no Python deps beyond the base.
ebpf = []

# Test-only tooling (pytest + plugins). Kept separate from `dev` so the CI
# test jobs can install a slim environment without the lint/type tooling.
tests = [
    "pytest>=8.0.0",
    "pytest-cov>=4.1.0",
    "pytest-timeout>=2.1.0",
    "pytest-xdist>=3.3.0",
    "pytest-forked>=1.6.0",  # per-test process isolation for the CPU CI gate
]

# Development dependencies (tests + lint/type tooling)
dev = [
    "amd-aorta[tests]",
    "pre-commit>=3.0.0",
    "black>=23.0.0",
    "isort>=5.12.0",
    "mypy>=1.0.0",
    "ruff>=0.1.0",
]

# Full installation
all = [
    "amd-aorta[analysis]",
    "amd-aorta[report]",
    "amd-aorta[hw-queue]",
    "amd-aorta[hw-queue-profiling]",
    "amd-aorta[agent]",
    "amd-aorta[dev]",
]

[project.urls]
Homepage = "https://github.com/ROCm/aorta"
Documentation = "https://github.com/ROCm/aorta#readme"
Repository = "https://github.com/ROCm/aorta"

[tool.setuptools_scm]
# Single source of truth for the version: the ``vX.Y.Z`` git tags.
#   * a checkout sitting exactly on a tag builds a clean ``X.Y.Z``;
#   * a checkout ahead of the latest tag builds ``X.Y.(Z+1).devN+g<sha>`` so a
#     local ``pip install .`` / ``pip install git+...`` always reflects the most
#     recent release (plus how far past it the tree is), never a stale literal.
# Only exact ``vMAJOR.MINOR.PATCH`` tags are matched (same shape as
# ``scripts/bump_version.py``), so neither the rolling ``dev-wheels`` nightly tag
# nor a stray ``v1`` / ``v1beta`` / ``v0.2.0rc1`` / ``v1.2.3.4`` tag can be
# mistaken for a release. The ``--match`` glob is only a coarse first pass (glob
# ``*`` also spans dots, so it alone still admits ``v0.2.0rc1`` and
# ``v1.2.3.4``); ``--exclude`` drops any tag carrying a non-digit/non-dot
# character or a fourth numeric segment, and ``tag_regex`` then anchors the
# extracted version to exactly three numeric components with no leading zeros
# (so it round-trips through the integer parsing in ``bump_version.py`` and the
# release gate unchanged). A tree with no git
# metadata (e.g. an unpacked sdist without .git) falls back to the literal below;
# built sdists/wheels embed the resolved version in their metadata, so installing
# from those still reports the correct version.
tag_regex = "^v(?P<version>(?:0|[1-9][0-9]*)\\.(?:0|[1-9][0-9]*)\\.(?:0|[1-9][0-9]*))$"
git_describe_command = [
    "git",
    "describe",
    "--dirty",
    "--tags",
    "--long",
    "--match",
    "v[0-9]*.[0-9]*.[0-9]*",
    "--exclude",
    "v*[!0-9.]*",
    "--exclude",
    "v*.*.*.*",
]
fallback_version = "0.0.0+unknown"

[tool.setuptools.packages.find]
where = ["src"]
include = ["aorta*"]

[tool.setuptools.package-data]
aorta = [
    "py.typed",
    "ebpf/scripts/*.bt",
    "ebpf/scripts/PROVENANCE.md",
    "workloads/hrx_kernels/*.cpp",
    "workloads/hrx_kernels/README.md",
]

[tool.pytest.ini_options]
testpaths = ["tests"]
python_files = ["test_*.py"]
python_functions = ["test_*"]
addopts = "-v --tb=short"
markers = [
    "slow: marks tests as slow (deselect with '-m \"not slow\"')",
    "gpu: marks tests that require GPU",
    "rocm: marks tests that require ROCm specifically",
    "hw_queue: marks hw_queue_eval tests",
]

[tool.black]
line-length = 100
target-version = ['py310', 'py311', 'py312']
include = '\.pyi?$'

[tool.isort]
profile = "black"
line_length = 100

[tool.mypy]
python_version = "3.10"
warn_return_any = true
warn_unused_configs = true
ignore_missing_imports = true

[tool.ruff]
line-length = 100
select = ["E", "F", "W", "I", "N", "UP", "B", "C4"]
ignore = ["E501"]