naming

diff --git a/src/caffeine/layers/inner_product_layer.cu b/src/caffeine/layers/inner_product_layer.cu
index baeca0e..4cae30c 100644 (file)
--- a/src/caffeine/layers/inner_product_layer.cu
+++ b/src/caffeine/layers/inner_product_layer.cu
@@ -101,7 +101,7 @@ Dtype InnerProductLayer<Dtype>::Backward_cpu(const vector<Blob<Dtype>*>& top,
 }
 
 template <typename Dtype>
-__global__ void BroadcastCopy(const int total, const int vec_len,
+__global__ void BroadcastRow(const int total, const int vec_len,
        const Dtype* in_vec, Dtype* out_matrix) {
   int index = threadIdx.x + blockIdx.x * blockDim.x;
   if (index < total) {
@@ -110,6 +110,8 @@ __global__ void BroadcastCopy(const int total, const int vec_len,
   }
 }
 
+
+
 template <typename Dtype>
 void InnerProductLayer<Dtype>::Forward_gpu(const vector<Blob<Dtype>*>& bottom,
     vector<Blob<Dtype>*>* top) {
@@ -123,7 +125,7 @@ void InnerProductLayer<Dtype>::Forward_gpu(const vector<Blob<Dtype>*>& bottom,
        beta = 1.;
        const int count = (*top)[0]->count();
        // we pre-copy the bias to the results, and then call gemm.
-       BroadcastCopy<<<CAFFEINE_GET_BLOCKS(count), CAFFEINE_CUDA_NUM_THREADS>>>(
+       BroadcastRow<<<CAFFEINE_GET_BLOCKS(count), CAFFEINE_CUDA_NUM_THREADS>>>(
                        count, N_, bias, top_data);
   }
   switch(sizeof(Dtype)) {