Function main

python/2_CoreConcepts/memoryResources/memoryResources.py:210–257

()

def main():
    import argparse

    parser = argparse.ArgumentParser(
        description="Demonstrate cuda.core memory resources (Buffer + MR)"
    )
    parser.add_argument(
        "--elements",
        type=int,
        default=1 << 16,
        help="Number of float32 elements per buffer (default: 65536)",
    )
    parser.add_argument("--device", type=int, default=0, help="CUDA device id")
    args = parser.parse_args()

    device = Device(args.device)
    device.set_current()
    print_gpu_info(device)

    # PinnedMemoryResource is backed by a host memory pool, which is not
    # available on every device.
    if not device.properties.host_memory_pools_supported:
        print("Host pinned memory pools are not supported on this platform.")
        return 2

    # This sample builds host NumPy views over managed memory. Devices without
    # concurrent managed access keep managed allocations GPU-exclusive while
    # the GPU is active, so those host views fault.
    if not device.properties.concurrent_managed_access:
        print("Concurrent managed memory access is not supported on this platform.")
        return 2

    stream = device.create_stream()

    try:
        program_options = ProgramOptions(std="c++17", arch=f"sm_{device.arch}")
        program = Program(SCALE_BIAS_KERNEL, code_type="c++", options=program_options)
        module = program.compile("cubin")
        kernel = module.get_kernel("scale_and_bias")

        demo_device_and_pinned(device, stream, kernel, args.elements)
        demo_managed(device, stream, kernel, args.elements)
        demo_explicit_device_pool(device, stream, kernel, args.elements)

        print("\nDone")
        return 0
    finally:
        stream.close()


if __name__ == "__main__":