https://github.com/Microsoft/CNTK
Raw File
Tip revision: a05c3c642648373f4ede0956e4286257c3d59a61 authored by liqfu on 24 August 2018, 17:46:51 UTC
CNTK splice allows broadcast. This case is handled in the change. For noop (identity) ops, its inputs and outputs types shall be set according to upstream ops. ToBatch/ToSequence and Unpack Batch/Sequence ops added during model importing need tp be skipped. Model import need to handle ops with multiple outputs.
Tip revision: a05c3c6
GPUWatcher.h
//
// Copyright (c) Microsoft. All rights reserved.
// Licensed under the MIT license. See LICENSE.md file in the project root for full license information.
//

#pragma once

#include "GPUMatrix.h"

class MATH_API GPUWatcher
{
public:
    static size_t GetFreeMemoryOnCUDADevice(int devId);
    static int GetGPUIdWithTheMostFreeMemory();
    GPUWatcher(void);
    ~GPUWatcher(void);
};
back to top