Revert "add submanifold sparse feature for conv layer"

This reverts commit e02871b.

Author: vera121

FEATURES.md

@@ -103,7 +103,6 @@ pad_type (deprecated, "SAME" style padding) in ConvolutionParameter and PoolingP
 pad_l, pad_r, pad_t and pad_b (arbitrary 2D padding) in ConvolutionParameter and PoolingParameter  
 AVE_EXC_PAD (average pooling excluding the paddings), AVE_TF (deprecated, alias for AVE_EXC_PAD) in PoolingParameter  
 ceil_mode in PoolingParameter  
-submanifold_sparse in ConvolutionParameter  
 faceboxes, box_width, box_height, keras, tf and yx_order in PriorBoxParameter  
 relu6, maximum and minimum in ReLUParameter  
 

include/caffe/layers/base_conv_layer.hpp

@@ -35,7 +35,7 @@ class BaseConvolutionLayer : public Layer<Dtype> {
   // The last argument in forward_cpu_gemm is so that we can skip the im2col if
   // we just called weight_cpu_gemm with the same input.
   void forward_cpu_gemm(const Dtype* input, const Dtype* weights,
-      Dtype* output, bool skip_im2col = false, bool submanifold_sparse = false);
+      Dtype* output, bool skip_im2col = false);
   void forward_cpu_bias(Dtype* output, const Dtype* bias);
   void backward_cpu_gemm(const Dtype* input, const Dtype* weights,
       Dtype* output);
@@ -111,7 +111,6 @@ class BaseConvolutionLayer : public Layer<Dtype> {
   bool per_channel_scale_weight_; //CUSTOMIZATION
   bool per_channel_scale_output_; //CUSTOMIZATION
   int quantize_method_; //CUSTOMIZATION
-  bool submanifold_sparse_;
 
  private:
   // wrap im2col/col2im so we don't have to remember the (long) argument lists

include/caffe/util/math_functions.hpp

@@ -18,7 +18,7 @@ template <typename Dtype>
 void caffe_cpu_gemm(const CBLAS_TRANSPOSE TransA,
     const CBLAS_TRANSPOSE TransB, const int M, const int N, const int K,
     const Dtype alpha, const Dtype* A, const Dtype* B, const Dtype beta,
-    Dtype* C, const bool submanifold_sparse=false);
+    Dtype* C);
 
 template <typename Dtype>
 void caffe_cpu_gemv(const CBLAS_TRANSPOSE TransA, const int M, const int N,

src/caffe/layers/base_conv_layer.cpp

@@ -87,7 +87,6 @@ void BaseConvolutionLayer<Dtype>::LayerSetUpInternal(LayerParam conv_param,
   quantize_method_ = conv_param.quantize_method();
   per_channel_scale_weight_ = conv_param.per_channel_scale_weight();
   per_channel_scale_output_ = conv_param.per_channel_scale_output();
-  submanifold_sparse_ = conv_param.submanifold_sparse();
   //CUSTOMIZATION-->
 
   // Setup pad dimensions (pad_).
@@ -391,7 +390,7 @@ void BaseConvolutionLayer<Dtype>::Reshape(const vector<Blob<Dtype>*>& bottom,
 
 template <typename Dtype>
 void BaseConvolutionLayer<Dtype>::forward_cpu_gemm(const Dtype* input,
-    const Dtype* weights, Dtype* output, bool skip_im2col, bool submanifold_sparse) {
+    const Dtype* weights, Dtype* output, bool skip_im2col) {
   const Dtype* col_buff = input;
   if (!is_1x1_) {
     if (!skip_im2col) {
@@ -403,7 +402,7 @@ void BaseConvolutionLayer<Dtype>::forward_cpu_gemm(const Dtype* input,
     caffe_cpu_gemm<Dtype>(CblasNoTrans, CblasNoTrans, conv_out_channels_ /
         group_, conv_out_spatial_dim_, kernel_dim_,
         (Dtype)1., weights + weight_offset_ * g, col_buff + col_offset_ * g,
-        (Dtype)0., output + output_offset_ * g, submanifold_sparse);
+        (Dtype)0., output + output_offset_ * g);
   }
 }
 

src/caffe/layers/conv_layer.cpp

@@ -117,7 +117,7 @@ void ConvolutionLayer<Dtype>::Forward_cpu(const vector<Blob<Dtype>*>& bottom,
     Dtype* top_data = top[i]->mutable_cpu_data();
     for (int n = 0; n < this->num_; ++n) {
       this->forward_cpu_gemm(bottom_data + n * this->bottom_dim_, weight,
-          top_data + n * this->top_dim_, false, this->submanifold_sparse_);
+          top_data + n * this->top_dim_);
       if (this->bias_term_) {
         const Dtype* bias = this->blobs_[1]->cpu_data();
         this->forward_cpu_bias(top_data + n * this->top_dim_, bias);

src/caffe/proto/caffe.proto

@@ -1428,7 +1428,7 @@ message ConvolutionParameter {
   optional bool per_channel_scale_weight = 39 [default = false];
   // CUSTOMIZATION, whether to have per-channel scale & zero_points for output (scale data will be stored in caffemodel)
   optional bool per_channel_scale_output = 41 [default = false];
-  optional bool submanifold_sparse = 42 [default = false];
+
 
   //<--CUSTOMIZATION
   enum SaturateMethod {
@@ -2746,7 +2746,7 @@ message SqueezeConvolutionParameter {
   // CUSTOMIZATION, whether to have per-channel scale & zero_points for weights/bias (data will be stored in caffemodel)
   optional bool per_channel_scale_weight = 59 [default = false];
   optional bool per_channel_scale_output = 61 [default = false];
-  optional bool submanifold_sparse = 62 [default = false];
+
 
   //<--CUSTOMIZATION
   enum SaturateMethod {

src/caffe/util/math_functions.cpp

@@ -20,54 +20,18 @@ template<>
 void caffe_cpu_gemm<float>(const CBLAS_TRANSPOSE TransA,
     const CBLAS_TRANSPOSE TransB, const int M, const int N, const int K,
     const float alpha, const float* A, const float* B, const float beta,
-    float* C, const bool submanifold_sparse) {
+    float* C) {
   int lda = (TransA == CblasNoTrans) ? K : M;
   int ldb = (TransB == CblasNoTrans) ? N : K;
-
   cblas_sgemm(CblasRowMajor, TransA, TransB, M, N, K, alpha, A, lda, B,
       ldb, beta, C, N);
-
-  if(submanifold_sparse)
-  {
-    // Submanifold sparse active site check
-    // weight: M*K input blob: K*N -> output blob: M*N
-    //LOG(INFO)<<"K="<<K<<", M="<<M<<", N="<<N<<std::endl;
-    LOG(INFO)<<"Use submanifold sparse."<<std::endl;
-    for (int j=0;j<N; j++)
-    {
-      bool batch_center_active = false;
-      for (int b=0; b<M; b++)
-      {
-        bool center_active = false;
-        for(int i=0; i<K/M; i++) // K=M*xyz
-        {
-          int index = (b*(K/M)+i)*N+j;
-          if(B[index] != 0)
-          {
-            if(i == (K/M)/2)
-            {
-              center_active = true;
-              //LOG(WARNING)<<"center active!!"<<std::endl;
-              //LOG(INFO)<<"index b="<<b<<", i="<<i<<", j="<<j<<", data="<<B[index]<<" ";
-            }
-          }
-        }
-        batch_center_active = batch_center_active || center_active;
-      }
-      if(!batch_center_active)
-      {
-        for(int t=0; t<M; t++)
-          C[t*N+j] = 0;
-      }
-    }
-  }
 }
 
 template<>
 void caffe_cpu_gemm<double>(const CBLAS_TRANSPOSE TransA,
     const CBLAS_TRANSPOSE TransB, const int M, const int N, const int K,
     const double alpha, const double* A, const double* B, const double beta,
-    double* C, const bool submanifold_sparse) {
+    double* C) {
   int lda = (TransA == CblasNoTrans) ? K : M;
   int ldb = (TransB == CblasNoTrans) ? N : K;
   cblas_dgemm(CblasRowMajor, TransA, TransB, M, N, K, alpha, A, lda, B,