CUDA_CALLABLE fix and CUDA_UNCALLABLE implementation

David Mayerich
1 parent 685a889c
Showing 3 changed files with 10 additions and 10 deletions Show diff stats
stim/cuda/cudatools/callable.h
stim/math/vec3.h
stim/visualization/colormap.h
@@ -7,4 +7,10 @@
 #define CUDA_CALLABLE
 #endif
  
+#ifdef __CUDACC__
+#define CUDA_UNCALLABLE __host__ inline
+#else
+#define CUDA_UNCALLABLE
+#endif
+
 #endif
@@ -243,9 +243,9 @@ public:
 			return false;	
 	}
  
-#ifndef __NVCC__
+//#ifndef __CUDA_ARCH__
 	/// Outputs the vector as a string
-	std::string str() const{
+CUDA_UNCALLABLE	std::string str() const{
 		std::stringstream ss;
  
 		const size_t N = 3;
@@ -261,7 +261,7 @@ public:
  
 		return ss.str();
 	}
-#endif
+//#endif
  
 	size_t size(){ return 3; }
  
@@ -166,14 +166,12 @@ static void gpu2gpu(T* gpuSource, unsigned char* gpuDest, unsigned int nVals, T 
         gridX = 65535;
     }
     dim3 dimGrid(gridX, gridY);
-	//int gridDim = (nVals + blockDim - 1)/blockDim;
 	if(cm == cmGrayscale)
 		applyGrayscale<<<dimGrid, blockDim>>>(gpuSource, gpuDest, nVals, minVal, maxVal);
 	else if(cm == cmBrewer)
 	{
 		initBrewer();
 		applyBrewer<<<dimGrid, blockDim>>>(gpuSource, gpuDest, nVals, minVal, maxVal);
-		//HANDLE_ERROR(cudaMemset(gpuDest, 0, sizeof(unsigned char) * nVals * 3));
 		destroyBrewer();
 	}
  
@@ -190,13 +188,9 @@ static void gpu2cpu(T* gpuSource, unsigned char* cpuDest, unsigned int nVals, T 
     unsigned char* gpuDest;
     HANDLE_ERROR(cudaMalloc( (void**)&gpuDest, sizeof(unsigned char) * nVals * 3 ));
  
-	//HANDLE_ERROR(cudaMemset(gpuSource, 0, sizeof(T) * nVals));
-
     //create the image on the gpu
     gpu2gpu(gpuSource, gpuDest, nVals, minVal, maxVal, cm);
-
-	//HANDLE_ERROR(cudaMemset(gpuDest, 0, sizeof(unsigned char) * nVals * 3));
-
+	
     //copy the image from the GPU to the CPU
     HANDLE_ERROR(cudaMemcpy(cpuDest, gpuDest, sizeof(unsigned char) * nVals * 3, cudaMemcpyDeviceToHost));