Skip to main content

Browse the archive

https://github.com/Microsoft/CNTK
10 March 2023, 01:01:58 UTC
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
MatrixQuantizerCPU.h 
#pragma once

#include "MatrixQuantizerImpl.h"
#include "ColumnQuantizer.h"
#include "QuantizedMatrix.h"
#include "CPUMatrix.h"

#ifdef _WIN32
#ifdef MATH_EXPORTS
#define MATH_API __declspec(dllexport)
#else
#define MATH_API __declspec(dllimport)
#endif
#else // no DLLs on Linux
#define MATH_API
#endif

namespace Microsoft { namespace MSR { namespace CNTK {

//see dbn::matrix quantizer
template <class ElemType>
class MatrixQuantizerCPU final : public MatrixQuantizerImpl<ElemType>
{
public:
    MatrixQuantizerCPU();

    // Disallow copy construction and assignment
    MatrixQuantizerCPU(const MatrixQuantizerCPU&) = delete;
    MatrixQuantizerCPU& operator=(const MatrixQuantizerCPU&) = delete;

    void QuantizeAsync(const Matrix<ElemType>& inMatrix, const Matrix<ElemType>& inResidual, QuantizedMatrix<ElemType>& outQMatrix, Matrix<ElemType>& outResidual, bool zeroThresholdFor1Bit) override;
    void WaitQuantizeAsyncDone() override;

    void UnquantizeAsync(QuantizedMatrix<ElemType>& inQMatrix, Matrix<ElemType>& outMatrix, bool add = false) override;
    void WaitUnquantizeAsyncDone() override;
};
} } }
back to top