Separating cuda dependencies and adding CPU CMakeLists

llvm/projects/hpvm-tensor-rt/CMakeLists_cpu.txt

+cmake_minimum_required (VERSION 2.6)
+project (approxhpvm-tensorRt-cpu)
+
+
+# Addresses a bug where code is not compiled as C++11 in non-CUDA code and older g++ versions
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 " )
+ 
+
+# Adding new rule for building a cuDNN runtime library
+add_library(tensor_cpu_runtime tensor_runtime/src/tensor_cpu_runtime.cc)
+target_link_libraries(tensor_cpu_runtime)
+
+
+#**** CPU sources
+add_executable(fc2_cpu  dnn_sources/src/fc2_cpu.cc)
+target_link_libraries(fc2_cpu  tensor_cpu_runtime)
+
+
+

llvm/projects/hpvm-tensor-rt/dnn_sources/include/utils_cpu.h

@@ -7,7 +7,7 @@
 #include <sstream>
 #include <vector>
 #include <bits/stdc++.h>
-#include "../../tensor_runtime/include/tensor.h"
+#include "../../tensor_runtime/include/tensor_cpu.h"
 #include "types.h"
 #include <cmath>
 
@@ -88,10 +88,10 @@ struct Tensor* readTrainedWeightsCPU(const char* file_name, int data_type,
   printf("size in bytes = %lu, bytes read = %lu \n", size_in_bytes, bytes_read);
 
   fclose(file);
-  
+
   
   struct Tensor* weights = (struct Tensor*) create4DTensorCPU(data_type, nchw, dim1_size, dim2_size,
-					                   dim3_size, dim4_size);
+					                      dim3_size, dim4_size);
   
   initTensorData(weights, tensor_data, size_in_bytes);
   //compareValues(weights, tensor_data, num_elems);

llvm/projects/hpvm-tensor-rt/tensor_runtime/include/tensor_cpu.h

+
+
+#ifndef TENSOR_HEADER
+#define TENSOR_HEADER
+
+
+struct Dimension{
+  int num_dims;
+  size_t* dim_sizes;
+};
+
+
+struct Tensor{
+  int data_type;
+  int data_format;
+  void* host_data;
+  void* gpu_data; // The pointers should not be device specific per se - TODO: Better design needed
+  size_t num_elems; // Total elements
+  size_t size_in_bytes; // Total size in bytes
+  struct Dimension dims;
+};
+
+
+#endif
+

llvm/projects/hpvm-tensor-rt/tensor_runtime/include/tensor_cpu_runtime.h

@@ -4,7 +4,6 @@
 #include <cmath>
 #include <memory>
 #include <string>
-//#include "runtime_types.h"
 
 
 #ifndef CUDNN_HEADER
@@ -41,10 +40,12 @@ extern "C"{
 			 int vertical_stride, int horizontal_stride);
 
   void* tensorGemmCPU(void* lhs, void* rhs);
-  // NOTE: In place operation
+
   void* tensorAddCPU(void* x, void* bias);
-  // NOTE: In-place operation
+
   void* tensorReluCPU(void* input);
+
+  void* tensorRelu2CPU(void* input);
   
   void* tensorTanhCPU(void* input);
   

llvm/projects/hpvm-tensor-rt/tensor_runtime/src/tensor_cpu_runtime.cc

@@ -23,7 +23,8 @@
 
 // Tensor runtime header files
 #include "../include/tensor_cpu_runtime.h"
-#include "../include/tensor.h"
+#include "../include/tensor_cpu.h"
+
 
 
 void llvm_hpvm_initTensorRt(int gpuid){