src/common/cnn/timlCNNAddLinearLayer.c

   1 /******************************************************************************/
   2 /*!
   3  * \file timlCNNAddLinearLayer.c
   4  */
   5 /* Copyright (C) 2015 Texas Instruments Incorporated - http://www.ti.com/
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  *
  11  *    Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  *
  14  *    Redistributions in binary form must reproduce the above copyright
  15  *    notice, this list of conditions and the following disclaimer in the
  16  *    documentation and/or other materials provided with the
  17  *    distribution.
  18  *
  19  *    Neither the name of Texas Instruments Incorporated nor the names of
  20  *    its contributors may be used to endorse or promote products derived
  21  *    from this software without specific prior written permission.
  22  *
  23  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  24  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  25  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  26  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  27  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  28  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  29  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  30  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  31  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  32  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  33  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  34  *
  35  ******************************************************************************/
  36
  37
  38 /*******************************************************************************
  39  *
  40  * INCLUDES
  41  *
  42  ******************************************************************************/
  43
  44 #include "../api/timl.h"
  45
  46
  47 /*******************************************************************************/
  48 /**
  49  * \ingroup       cnn
  50  * \brief         Add linear layer
  51  * \param[in,out] cnn    CNN
  52  * \param[in]     dim    Output 1D feature map dimension
  53  * \param[in]     params Optional parameters
  54  * \return        Error code
  55  */
  56 /*******************************************************************************/
  57
  58 int timlCNNAddLinearLayer(timlConvNeuralNetwork *cnn, int dim, timlCNNLinearParams params)
  59 {
  60    int          prevDim;
  61    timlCNNLayer *prev;
  62    timlCNNLayer *linearLayer;
  63
  64    // error checking
  65    if (cnn == NULL) {
  66       return ERROR_CNN_NULL_PTR;
  67    }
  68    if (cnn->tail == NULL) {
  69       return ERROR_CNN_EMPTY;
  70    }
  71    if (dim <= 0) {
  72       return ERROR_CNN_LINEAR_LAYER_DIM;
  73    }
  74
  75    prev    = cnn->tail;
  76    prevDim = prev->row * prev->col * prev->channel;
  77
  78    // allocate linear layer
  79    if (timlUtilMallocHost((void**)&linearLayer, sizeof(timlCNNLayer))) {
  80       return ERROR_CNN_LAYER_ALLOCATION;
  81    }
  82
  83 //   linearLayer->inputParams     = timlCNNInputParamsDefault();
  84 //   linearLayer->convParams      = timlCNNConvParamsDefault();
  85 //   linearLayer->poolingParams   = timlCNNPoolingParamsDefault();
  86 //   linearLayer->normParams      = timlCNNNormParamsDefault();
  87 //   linearLayer->nonlinearParams = timlCNNNonlinearParamsDefault();
  88
  89    // load parameter
  90    linearLayer->linearParams = params;
  91    // override parameters
  92    linearLayer->linearParams.weight          = NULL;
  93    linearLayer->linearParams.weightInc       = NULL;
  94    linearLayer->linearParams.weightGradAccum = NULL;
  95    linearLayer->linearParams.bias            = NULL;
  96    linearLayer->linearParams.biasInc         = NULL;
  97    linearLayer->linearParams.biasGradAccum   = NULL;
  98    linearLayer->linearParams.biasMultiplier  = NULL;
  99    linearLayer->linearParams.prevDim         = prevDim;
 100    linearLayer->linearParams.dim             = dim;
 101    linearLayer->linearParams.shared          = false;
 102
 103    linearLayer->type           = CNN_Linear;
 104    linearLayer->featureMap     = NULL;
 105    linearLayer->delta          = NULL;
 106    linearLayer->row            = 1;
 107    linearLayer->col            = 1;
 108    linearLayer->channel        = dim;
 109    linearLayer->batchSize      = prev->batchSize;
 110    linearLayer->maxBatchSize   = prev->maxBatchSize;
 111    linearLayer->allocatorLevel = cnn->params.allocatorLevel;
 112    linearLayer->phase          = cnn->params.phase;
 113
 114    // link the linearLayer
 115    linearLayer->cnn        = cnn;
 116    linearLayer->id         = prev->id + 1;
 117    linearLayer->prev       = prev;
 118    linearLayer->prev->next = linearLayer;
 119    linearLayer->next       = NULL;
 120    cnn->tail               = linearLayer;
 121
 122    return 0;
 123
 124 }