torch/cuda/profiler.py

import tempfile
import torch
import contextlib
from . import cudart, check_error

__all__ = ["init", "start", "stop", "profile"]

DEFAULT_FLAGS = [
    "gpustarttimestamp",
    "gpuendtimestamp",
    "gridsize3d",
    "threadblocksize",
    "streamid",
    "enableonstart 0",
    "conckerneltrace",
]


def init(output_file, flags=None, output_mode='key_value'):
    rt = cudart()
    if not hasattr(rt, 'cudaOutputMode'):
        raise AssertionError("HIP does not support profiler initialization!")
    if hasattr(torch.version, "cuda") and torch.version.cuda is not None and int(torch.version.cuda.split(".")[0]) >= 12:
        # Check https://github.com/pytorch/pytorch/pull/91118
        # cudaProfilerInitialize is no longer needed after CUDA 12
        raise AssertionError("CUDA12+ does not need profiler initialization!")
    flags = DEFAULT_FLAGS if flags is None else flags
    if output_mode == 'key_value':
        output_mode_enum = rt.cudaOutputMode.KeyValuePair
    elif output_mode == 'csv':
        output_mode_enum = rt.cudaOutputMode.CSV
    else:
        raise RuntimeError("supported CUDA profiler output modes are: key_value and csv")
    with tempfile.NamedTemporaryFile(delete=True) as f:
        f.write(b'\n'.join(f.encode('ascii') for f in flags))
        f.flush()
        check_error(rt.cudaProfilerInitialize(f.name, output_file, output_mode_enum))


def start():
    check_error(cudart().cudaProfilerStart())


def stop():
    check_error(cudart().cudaProfilerStop())


@contextlib.contextmanager
def profile():
    try:
        start()
        yield
    finally:
        stop()