merging

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

@@ -390,10 +390,10 @@ __global__ void vecMul(float* A, float* B, int n){
 void initRandValues(Tensor* bias, int error_scale){
 
   float scaling_values[20];
-
+  
   // FIXIT: Error knob 0 should be 0 zero
-  scaling_values[0] = 0.007;
-  scaling_values[1] = 0.015;
+  scaling_values[0] = 0.000;
+  scaling_values[1] = 0.01;
   scaling_values[2] = 0.03;
   scaling_values[3] = 0.06;
   scaling_values[4] = 0.08;
@@ -583,7 +583,7 @@ void* addRoundError(void* x_ptr, int error_scale){
 // Routine for Adding Gaussian Error
 void* addGaussianError(void* x_ptr, int error_scale){
 
-  if(error_scale > 11 || error_scale < 0){
+  if(error_scale > 20 || error_scale < 0){
     ERROR("Error Scale out of bounds \n");
   }
       
@@ -627,6 +627,10 @@ void* addGaussianError(void* x_ptr, int error_scale){
 
   //Norm_t* norms = calculateNorms2(x, x_original);
   Norm_t* norms = calculateNormsGPU(x, x_original);
+
+  
+  freeTensor(x_original);
+  freeTensor(bias);
   
   
   profileEvent("tensorAddError_end", true);

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

@@ -14,6 +14,7 @@
 #include <random>
 #include <string>
 #include <unordered_map>
+#include <cuda_runtime.h>
 
 
 
@@ -47,6 +48,8 @@ extern "C"{
 
   void profileEvent(char* event_name, bool compare_previous = false){
 
+    checkCudaErrors(cudaDeviceSynchronize());
+
     auto it = func_counters.find(event_name);
     if(it == func_counters.end()){
       func_counters[event_name] = 1;