Conv1d minor changes and type casting

Author: AnirudhBHarish

c_reference/include/conv1d.h

@@ -10,8 +10,8 @@
  * @var   B    pointer to the bias vector for the convolution, size = out_channels
  */
 typedef struct ConvLayers_Params {
-  float* W;
-  float* B;
+  const float* const W;
+  const float* const B;
 } ConvLayers_Params;
 
 /**
@@ -36,7 +36,7 @@ typedef struct ConvLayers_Params {
  */
 int conv1d(float* output_signal, unsigned out_time, unsigned out_channels, const float* input_signal, 
   unsigned in_time, unsigned in_channels, unsigned padding, unsigned kernel_size, 
-  const void* params, unsigned stride, int activation);
+  const void* params, unsigned stride, unsigned activation);
 
 /**
  * @brief Model definition for the 1D Depthwise Convolution Layer. Currently only for dilation = 1
@@ -60,7 +60,7 @@ int conv1d(float* output_signal, unsigned out_time, unsigned out_channels, const
  */
 int conv1d_depth(float* output_signal, unsigned out_time, const float* input_signal, 
   unsigned in_time, unsigned in_channels, unsigned padding, unsigned kernel_size, 
-  const void* params, unsigned stride, int activation);
+  const void* params, unsigned stride, unsigned activation);
 
 /**
  * @brief Model parameters for the 1D Low Rank Convolution Layer.
@@ -70,9 +70,9 @@ int conv1d_depth(float* output_signal, unsigned out_time, const float* input_sig
  * @var    rank    rank of the weight tensor. A low-rank decomposition typically used to reduce computation and storage
  */
 typedef struct ConvLayers_LR_Params {
-  float* W1;
-  float* W2;
-  float* B;
+  const float* const W1;
+  const float* const W2;
+  const float* const B;
   unsigned rank;
 } ConvLayers_LR_Params;
 
@@ -100,7 +100,7 @@ typedef struct ConvLayers_LR_Params {
  */
 int conv1d_lr(float* output_signal, unsigned out_time, unsigned out_channels, const float* input_signal, 
   unsigned in_time, unsigned in_channels, unsigned padding, unsigned kernel_size, 
-  const void* params, unsigned stride, int activation);
+  const void* params, unsigned stride, unsigned activation);
 
 /**
  * @brief Model definition for the 1D Low-Rank Depthwise Convolution Layer. Currently only for dilation = 1
@@ -126,7 +126,7 @@ int conv1d_lr(float* output_signal, unsigned out_time, unsigned out_channels, co
  */
 int conv1d_depth_lr(float* output_signal, unsigned out_time, const float* input_signal, 
   unsigned in_time, unsigned in_channels, unsigned padding, unsigned kernel_size, 
-  const void* params, unsigned stride, int activation);
+  const void* params, unsigned stride, unsigned activation);
 
 // Auxiliary Layers
 /**
@@ -150,7 +150,7 @@ int conv1d_depth_lr(float* output_signal, unsigned out_time, const float* input_
  */
 int avgpool1d(float* output_signal, unsigned out_time, const float* input_signal,
   unsigned in_time, unsigned in_channels,
-  unsigned padding, unsigned kernel_size, unsigned stride, int activation);
+  unsigned padding, unsigned kernel_size, unsigned stride, unsigned activation);
 
 /**
  * @brief Model definition for the 1D batch Normalization Layer
@@ -168,7 +168,8 @@ int avgpool1d(float* output_signal, unsigned out_time, const float* input_signal
  */
 int batchnorm1d(float* output_signal, float* input_signal,
   unsigned in_time, unsigned in_channels,
-  float* mean, float* var, unsigned affine, float* gamma , float* beta,
+  const float* const mean, const float* const var, 
+  unsigned affine, const float* const gamma , const float* const beta,
   unsigned in_place, float eps);
 
 #endif

c_reference/include/dscnn.h

@@ -30,9 +30,10 @@
  */
 int phon_pred_lr_cnn(float* output_signal, float* input_signal,
   unsigned in_time, unsigned in_channels,
-  float* mean, float* var, unsigned affine, float* gamma, float* beta, unsigned in_place,
+  const float* const mean, const float* const var,
+  unsigned affine, float* gamma, float* beta, unsigned in_place,
   unsigned cnn_hidden, unsigned cnn_padding, unsigned cnn_kernel_size,
-  const void* cnn_params, unsigned cnn_stride, int cnn_activation);
+  const void* cnn_params, unsigned cnn_stride, unsigned cnn_activation);
 
 /**
  * @brief Model definition for the 1D Convolution block applied after the RNN
@@ -47,7 +48,6 @@ int phon_pred_lr_cnn(float* output_signal, float* input_signal,
  * @param[in]    gamma                  pointer to the scaling factors for the post-norm affine operation, size = in_channels
  * @param[in]    beta                   pointer to the offsets for the post-norm affine operation, size = in_channels
  * @param[in]    in_place               in-place computation of the batchnorm. Storage efficient
- * @param[in]    depth_cnn_hidden       hidden state/out_channels dimensions for the depth CNN
  * @param[in]    depth_cnn_padding      padding for the depth CNN layer. Note: applied to both sides of the input to the depth CNN
  * @param[in]    depth_cnn_kernel_size  kernel size of the depth CNN
  * @param[in]    depth_cnn_params       weights, bias and other essential parameters used to describe the depth CNN
@@ -78,11 +78,12 @@ int phon_pred_lr_cnn(float* output_signal, float* input_signal,
  */
 int phon_pred_depth_point_lr_cnn(float* output_signal, float* input_signal,
   unsigned in_time, unsigned in_channels,
-  float* mean, float* var, unsigned affine, float* gamma, float* beta, unsigned in_place,
-  unsigned depth_cnn_hidden, unsigned depth_cnn_padding, unsigned depth_cnn_kernel_size,
-  const void* depth_cnn_params, unsigned depth_cnn_stride, int depth_cnn_activation,
+  const float* const mean, const float* const var,
+  unsigned affine, const float* const gamma, const float* const beta, unsigned in_place,
+  unsigned depth_cnn_padding, unsigned depth_cnn_kernel_size,
+  const void* depth_cnn_params, unsigned depth_cnn_stride, unsigned depth_cnn_activation,
   unsigned point_cnn_hidden, unsigned point_cnn_padding, unsigned point_cnn_kernel_size,
-  const void* point_cnn_params, unsigned point_cnn_stride, int point_cnn_activation,
-  unsigned pool_padding, unsigned pool_kernel_size, unsigned pool_stride, int pool_activation);
+  const void* point_cnn_params, unsigned point_cnn_stride, unsigned point_cnn_activation,
+  unsigned pool_padding, unsigned pool_kernel_size, unsigned pool_stride, unsigned pool_activation);
 
 #endif

c_reference/include/rnn_bricked.h

@@ -4,8 +4,8 @@
 #ifndef __RNN_BRICKED_H__
 #define __RNN_BRICKED_H__
 
-// Function Pointer for the RNN to be passed as a parameter
-typedef int (*rnn_t)(float* const, unsigned, const float* const, unsigned, 
+// Function pointer for the RNN to be passed as a parameter
+typedef int (*rnn_layer)(float* const, unsigned, const float* const, unsigned, 
                       unsigned, const void*, void*, int, int);
 
 // NOTES for bi-direction
@@ -52,8 +52,8 @@ typedef int (*rnn_t)(float* const, unsigned, const float* const, unsigned,
  */
 int forward_bricked_rnn(float* output_signal, unsigned rnn_hidden, float* input_signal,
   unsigned in_time, unsigned in_dims, unsigned window, unsigned hop,
-  rnn_t rnn, const void* params, void* buffers,
-  int bi_direction, int sample_first_brick, int normalize);
+  rnn_layer rnn, const void* params, void* buffers,
+  unsigned bi_direction, unsigned sample_first_brick, int normalize);
 
 /** Backward Bricking and application of the backward RNN for an input signal
  * @param[out]       output_signal        pointer to output signal. size = out_time * rnn_hidden
@@ -73,8 +73,8 @@ int forward_bricked_rnn(float* output_signal, unsigned rnn_hidden, float* input_
  */
 int backward_bricked_rnn(float* output_signal, unsigned rnn_hidden, float* input_signal,
   unsigned in_time, unsigned in_dims, unsigned window, unsigned hop,
-  rnn_t rnn, const void* params, void* buffers,
-  int bi_direction, int sample_last_brick, int normalize);
+  rnn_layer rnn, const void* params, void* buffers,
+  unsigned bi_direction, unsigned sample_last_brick, int normalize);
 
 
 #endif

c_reference/include/utils.h

@@ -76,7 +76,7 @@ void softmax(const float* const input, unsigned len, float* const ret);
    in_channels      number of input channels. The output will have the half the number of input channels. 
                     Necessary for in_channels % 2 == 0
  */
-int semi_sigmoid_tanh(float* output_signal, float* input_signal,
+void semi_sigmoid_tanh(float* output_signal, const float* const input_signal,
   unsigned in_time, unsigned in_channels);
 
 #endif

c_reference/src/conv1d.c

@@ -8,7 +8,7 @@
 
 int conv1d_lr(float* output_signal, unsigned out_time, unsigned out_channels, const float* input_signal,
   unsigned in_time, unsigned in_channels, unsigned padding, unsigned kernel_size,
-  const void* params, unsigned stride, int activation) {
+  const void* params, unsigned stride, unsigned activation) {
 
   const ConvLayers_LR_Params* tparams= (ConvLayers_LR_Params*) params;
 
@@ -51,7 +51,7 @@ int conv1d_lr(float* output_signal, unsigned out_time, unsigned out_channels, co
 
 int conv1d_depth_lr(float* output_signal, unsigned out_time, const float* input_signal,
   unsigned in_time, unsigned in_channels, unsigned padding, unsigned kernel_size,
-  const void* params, unsigned stride, int activation) {
+  const void* params, unsigned stride, unsigned activation) {
 
   const ConvLayers_LR_Params* tparams= (ConvLayers_LR_Params*) params;
 
@@ -90,11 +90,9 @@ int conv1d_depth_lr(float* output_signal, unsigned out_time, const float* input_
   return 0;
 }
 
-
-
 int conv1d(float* output_signal, unsigned out_time, unsigned out_channels, const float* input_signal,
   unsigned in_time, unsigned in_channels, unsigned padding, unsigned kernel_size,
-  const void* params, unsigned stride, int activation) {
+  const void* params, unsigned stride, unsigned activation) {
 
   const ConvLayers_Params* tparams= (ConvLayers_Params*) params;
 
@@ -133,7 +131,7 @@ int conv1d(float* output_signal, unsigned out_time, unsigned out_channels, const
 
 int conv1d_depth(float* output_signal, unsigned out_time, const float* input_signal,
   unsigned in_time, unsigned in_channels, unsigned padding, unsigned kernel_size,
-  const void* params, unsigned stride, int activation) {
+  const void* params, unsigned stride, unsigned activation) {
 
   const ConvLayers_Params* tparams= (ConvLayers_Params*) params;
 
@@ -170,7 +168,7 @@ int conv1d_depth(float* output_signal, unsigned out_time, const float* input_sig
 
 int avgpool1d(float* output_signal, unsigned out_time, const float* input_signal,
   unsigned in_time, unsigned in_channels,
-  unsigned padding, unsigned kernel_size, unsigned stride, int activation) {
+  unsigned padding, unsigned kernel_size, unsigned stride, unsigned activation) {
 
   // Iterate over the time steps and average them. Similar to Conv1D_Dept with a filter kernel of ones
   for (unsigned t_in = 0, t_out = 0; t_out < out_time; t_out++, t_in += stride) {
@@ -203,7 +201,8 @@ int avgpool1d(float* output_signal, unsigned out_time, const float* input_signal
 
 int batchnorm1d(float* output_signal, float* input_signal,
   unsigned in_time, unsigned in_channels,
-  float* mean, float* var, unsigned affine, float* gamma , float* beta,
+  const float* const mean, const float* const var,
+  unsigned affine, const float* const gamma , const float* const beta,
   unsigned in_place, float eps) {
   // Check if affine values was learnt
   if (affine) {

c_reference/src/dscnn.c

@@ -10,9 +10,10 @@
 
 int phon_pred_lr_cnn(float* output_signal, float* input_signal,
   unsigned in_time, unsigned in_channels,
-  float* mean, float* var, unsigned affine, float* gamma, float* beta, unsigned in_place,
+  const float* const mean, const float* const var,
+  unsigned affine, float* gamma, float* beta, unsigned in_place,
   unsigned cnn_hidden, unsigned cnn_padding, unsigned cnn_kernel_size,
-  const void* cnn_params, unsigned cnn_stride, int cnn_activation) {
+  const void* cnn_params, unsigned cnn_stride, unsigned cnn_activation) {
 
   unsigned out_time = in_time - cnn_kernel_size + 2 * cnn_padding + 1;
   if (in_place) {
@@ -44,29 +45,31 @@ int phon_pred_lr_cnn(float* output_signal, float* input_signal,
 
 int phon_pred_depth_point_lr_cnn(float* output_signal, float* input_signal,
   unsigned in_time, unsigned in_channels,
-  float* mean, float* var, unsigned affine, float* gamma, float* beta, unsigned in_place,
-  unsigned depth_cnn_hidden, unsigned depth_cnn_padding, unsigned depth_cnn_kernel_size,
-  const void* depth_cnn_params, unsigned depth_cnn_stride, int depth_cnn_activation,
+  const float* const mean, const float* const var,
+  unsigned affine, const float* const gamma, const float* const beta, unsigned in_place,
+  unsigned depth_cnn_padding, unsigned depth_cnn_kernel_size,
+  const void* depth_cnn_params, unsigned depth_cnn_stride, unsigned depth_cnn_activation,
   unsigned point_cnn_hidden, unsigned point_cnn_padding, unsigned point_cnn_kernel_size,
-  const void* point_cnn_params, unsigned point_cnn_stride, int point_cnn_activation,
-  unsigned pool_padding, unsigned pool_kernel_size, unsigned pool_stride, int pool_activation) {
+  const void* point_cnn_params, unsigned point_cnn_stride, unsigned point_cnn_activation,
+  unsigned pool_padding, unsigned pool_kernel_size, unsigned pool_stride, unsigned pool_activation) {
 
   // Activation
-  unsigned out_time;
+  
   float* act_out= (float*)malloc(in_time * (in_channels >> 1) * sizeof(float));
   semi_sigmoid_tanh(act_out, input_signal, in_time, in_channels);
 
   in_channels >>= 1;
   float* depth_out;
+  unsigned out_time = in_time - depth_cnn_kernel_size + 2 * depth_cnn_padding + 1;
   if (in_place) {
     // Norm
     batchnorm1d(0, act_out,
       in_time, in_channels, 
-      mean, var, affine, gamma, beta,
+      mean, var,
+      affine, gamma, beta,
       in_place, 0.00001);
     // Depth CNN
-    out_time = in_time - depth_cnn_kernel_size + 2 * depth_cnn_padding + 1;
-    depth_out = (float*)malloc(out_time * depth_cnn_hidden * sizeof(float));
+    depth_out = (float*)malloc(out_time * in_channels * sizeof(float));
     conv1d_depth(depth_out, out_time, act_out, 
       in_time, in_channels, depth_cnn_padding, depth_cnn_kernel_size, 
       depth_cnn_params, depth_cnn_stride, depth_cnn_activation);
@@ -77,12 +80,12 @@ int phon_pred_depth_point_lr_cnn(float* output_signal, float* input_signal,
     float* norm_out = (float*)malloc(in_time * in_channels * sizeof(float));
     batchnorm1d(norm_out, act_out,
       in_time, in_channels, 
-      mean, var, affine, gamma, beta,
+      mean, var,
+      affine, gamma, beta,
       in_place, 0.00001);
     free(act_out);
     // Depth CNN
-    out_time = in_time - depth_cnn_kernel_size + 2 * depth_cnn_padding + 1;
-    depth_out = (float*)malloc(out_time * depth_cnn_hidden * sizeof(float));
+    depth_out = (float*)malloc(out_time * in_channels * sizeof(float));
     conv1d_depth(depth_out, out_time, norm_out, 
       in_time, in_channels, depth_cnn_padding, depth_cnn_kernel_size, 
       depth_cnn_params, depth_cnn_stride, depth_cnn_activation);
@@ -94,7 +97,7 @@ int phon_pred_depth_point_lr_cnn(float* output_signal, float* input_signal,
   out_time = in_time - point_cnn_kernel_size + 2 * point_cnn_padding + 1;
   float* point_out = (float*)malloc(out_time * point_cnn_hidden * sizeof(float));
   conv1d_lr(point_out, out_time, point_cnn_hidden, depth_out, 
-    in_time, depth_cnn_hidden, point_cnn_padding, point_cnn_kernel_size, 
+    in_time, in_channels, point_cnn_padding, point_cnn_kernel_size, 
     point_cnn_params, point_cnn_stride, point_cnn_activation);
   free(depth_out);
 

c_reference/src/rnn_bricked.c

@@ -10,8 +10,8 @@
 // Forward Pass
 int forward_bricked_rnn(float* output_signal, unsigned rnn_hidden, float* input_signal,
   unsigned in_time, unsigned in_dims, unsigned window, unsigned hop,
-  rnn_t rnn, const void* params, void* buffers,
-  int bi_direction, int sample_first_brick, int normalize) {
+  rnn_layer rnn, const void* params, void* buffers,
+  unsigned bi_direction, unsigned sample_first_brick, int normalize) {
   unsigned out_index = 0, t; // t is an index, but we want to remember the value after the loop. Hence we define it outside
 
   unsigned rnn_assign_offset = rnn_hidden;
@@ -61,8 +61,8 @@ int forward_bricked_rnn(float* output_signal, unsigned rnn_hidden, float* input_
 // Backward Pass
 int backward_bricked_rnn(float* output_signal, unsigned rnn_hidden, float* input_signal,
   unsigned in_time, unsigned in_dims, unsigned window, unsigned hop,
-  rnn_t rnn, const void* params, void* buffers,
-  int bi_direction, int sample_last_brick, int normalize) {
+  rnn_layer rnn, const void* params, void* buffers,
+  unsigned bi_direction, unsigned sample_last_brick, int normalize) {
   unsigned out_index = 0, t;
 
   // When bi-direction = 1, an offset of "rnn_hidden" will need to be provided during the function call(to the output_signal). 
@@ -88,7 +88,7 @@ int backward_bricked_rnn(float* output_signal, unsigned rnn_hidden, float* input
     out_index += window / hop;
   // If sample_last_block = 1, sample every hop index only for the last window 
   // Else the final hidden state(in reverse) is calculated
-  int stop_time = t;
+  unsigned stop_time = t;
   memset(temp_hiddenstate, 0, rnn_hidden * sizeof(float));
   for (t = in_time - 1; t >= stop_time; t--) {
     rnn(temp_hiddenstate, rnn_hidden,

c_reference/src/utils.c

@@ -78,8 +78,9 @@ void matmul(const float* const matA, const float* const matB,
   for (unsigned row = 0; row < nrows; row++) {
     for (unsigned col = 0; col < ncols; col++) {
       float sum = 0;
-      for (int k = 0; k < ncommon; k++)
+      for (unsigned k = 0; k < ncommon; k++) {
         sum += (matA[row * ncommon + k] * matB[k * ncols + col]);
+      }
       ret[row * ncols + col] = alpha * ret[row * ncols + col] + beta * sum;
     }
   }
@@ -135,7 +136,7 @@ void softmax(const float* const input, unsigned len, float* const ret) {
     ret[i] = expf(input[i] - offset);
 }
 
-int semi_sigmoid_tanh(float* output_signal, float* input_signal, 
+void semi_sigmoid_tanh(float* output_signal, const float* const input_signal, 
   unsigned in_time, unsigned in_channels) {
   unsigned piv = in_channels >> 1;
   for (unsigned t = 0; t < in_time; t++) {
@@ -144,5 +145,4 @@ int semi_sigmoid_tanh(float* output_signal, float* input_signal,
                                    * tanh(input_signal[t * in_channels + d + piv]);
     }
   }
-  return 0;
 }

c_reference/tests/conv1d/conv1d_depthwise/conv_param_depth.h

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a68ad85b8af772dcbd486752a8eb73ff667f01e5e8ef6b1b3014bfe1e756cefe
-size 39054
+oid sha256:d705c8b29a9eaf7255e15fb96314cc5b541d13e8a44494921fa0d00fbe46beee
+size 39066

c_reference/tests/conv1d/conv1d_lr/conv_param_lr.h

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8ec14100aaf0cfb424640b49c95bb9138344ff526d1125a0b09de62a129834da
-size 66193
+oid sha256:a822abfe5a1bd6b3439e7a5d994ef52ceedbd107431e3af8fed1fe28748ea425
+size 66211