first Target infrastructure

FittingProtocol.cu

@@ -23,6 +23,17 @@ FittingProtocol::FittingProtocol() {
 	_determine_gpu_resources();
 }
 
+void FittingProtocol::loadTarget(int gpu_index){
+	if(gpu_index < _num_devices){
+		if(_device_memory_infos[gpu_index].free > target.getGpuMem()){
+				target.loadToGpu(gpu_index);
+			}
+		}
+	else{
+		std::cout << "Gpu Index " << gpu_index << " does not match the number of devices ..." << std::endl;
+	}
+}
+
 void FittingProtocol::printGPUInfo(void){
 	printf("Found %i devices ...\n",_num_devices);
 	printf("-------------------------------------");
@@ -40,9 +51,9 @@ FittingProtocol::~FittingProtocol() {
 
 void FittingProtocol::setFileTarget(const char * file_name){
 	if (file_exists(file_name)){
-//		target = Target::fromMrc(file_name);
+		target = Target::fromMrc(file_name);
 	}
 	else {
-//		std::cout << "File " << file_name << " is not found ..." << std::endl;
+		std::cout << "File " << file_name << " is not found ..." << std::endl;
 	}
 }

FittingProtocol.h

@@ -8,7 +8,7 @@
 #ifndef FITTINGPROTOCOL_H_
 #define FITTINGPROTOCOL_H_
 
-//#include <Target.h>
+#include <Target.h>
 //#include <Query.h>
 //#include <Score.h>
 #include <vector>
@@ -16,6 +16,7 @@
 #include <util.h>
 #include <nvml.h>
 #include <vector>
+#include <iostream>
 
 class FittingProtocol {
 public:
@@ -23,7 +24,8 @@ public:
 	void setFileTarget(const char * file_name);
 	virtual ~FittingProtocol();
 	void printGPUInfo(void);
-//	Target target;
+	Target target;
+	void loadTarget(int gpu_index);
 //	std::vector<Query> queries;
 //	std::vector<Score> scores;
 private:

Makefile.test

@@ -33,13 +33,13 @@ protocol:
 #	nvcc --gpu-architecture=sm_61 --device-link Particles.o PdbReader.o FittingProtocol.o Density.o Target.o CMrcReader.o --output-file link.o -lnvidia-ml $(CUDA_LINKS)
 #	g++ FittingProtocol.o Particles.o PdbReader.o Target.o Density.o CMrcReader.o ccl_test.o link.o -o protocol -lnvidia-ml -L. $(CUDA_LINKS) 
 	$(info Compiling for device architecture $(DEVICE_ARCH))
-	nvcc --gpu-architecture=$(DEVICE_ARCH) --include-path=./ --device-c FittingProtocol.cu ccl_test.cu -g -G
-	nvcc --gpu-architecture=$(DEVICE_ARCH) --device-link FittingProtocol.o ccl_test.o --output-file link.o
-	g++ FittingProtocol.o ccl_test.o link.o -o protocol -lnvidia-ml -L/usr/local/cuda-10.0/lib64/ -lcudart 
+	nvcc --gpu-architecture=$(DEVICE_ARCH) --include-path=./ --device-c PdbReader.cu Particles.cu CMrcReader.cu Density.cu Target.cu FittingProtocol.cu ccl_test.cu -g -G
+	nvcc --gpu-architecture=$(DEVICE_ARCH) --device-link PdbReader.o Particles.o CMrcReader.o Density.o Target.o FittingProtocol.o ccl_test.o --output-file link.o
+	g++ PdbReader.o Particles.o CMrcReader.o Density.o Target.o FittingProtocol.o ccl_test.o link.o -o protocol -lnvidia-ml -L/usr/local/cuda-10.0/lib64/ -lcudart -lcudadevrt
 
 
 ccl2: CMrcReader.o 
-	nvcc --gpu-architecture=sm_61 --include-path=./ --device-c Density.cu Kernels.cu PdbReader.cu Particles.cu Labeler.cu ccl_test.cu -g -G
+	nvcc --gpu-architecture=sm_61 --include-path=./ --device-c Particles.cu Density.cu Kernels.cu PdbReader.cu Particles.cu Labeler.cu ccl_test.cu -g -G
 	nvcc --gpu-architecture=sm_61 --device-link Sampler.o PdbReader.o Particles.o Labeler.o Kernels.o Density.o ccl_test.o --output-file link.o
 	g++ Sampler.o PdbReader.o Particles.o Labeler.o Kernels.o Density.o CMrcReader.o ccl_test.o link.o -o ccl_test -L. $(CUDA_LINKS) $(BOOST_LINKS)
 

MemoryResources.h

@@ -16,7 +16,7 @@ struct MemoryResources{
 	size_t CPU;
 };
 
-enum SourceType {MRC,PDB};
+enum SourceType {MRC,PDB,NONE};
 typedef float TDensity;
 
 #endif /* MEMORYRESOURCES_H_ */

Target.cu

@@ -21,6 +21,18 @@ size_t Target::getCpuMem(){
 	return _data_byte_vol;
 }
 
+int Target::loadToGpu(int gpu_index){
+	int ret = 0;
+	if (!_cpu_initiated){
+		ret += initCpu();
+	}
+	if (!_gpu_initiated){
+		ret += initGpu(gpu_index);
+	}
+	toGpu(gpu_index);
+	return ret;
+}
+
 int Target::initCpu(){
 	int ret = 0;
 	if (_source_type == MRC){
@@ -30,26 +42,33 @@ int Target::initCpu(){
 	return ret;
 }
 
-int Target::initGpu(){
+int Target::initGpu(int gpu_index){
 	int ret = 0;
 	if (_source_type == MRC){
-		ret = int(cudaSuccess != cudaMalloc((void **)&_density.d_data,_data_byte_vol));
+		ret += int(cudaSuccess != cudaSetDevice(gpu_index));
+		ret += int(cudaSuccess != cudaMalloc((void **)&_density.d_data,_data_byte_vol));
 		_gpu_initiated = true;
 		return ret;
 	}
 	return 0;
 }
 
-int Target::toGpu(){
+int Target::toGpu(int gpu_index){
+	int ret = 0;
 	if (_source_type == MRC && _cpu_initiated && _gpu_initiated){
-			return int(cudaSuccess != cudaMemcpy(_density.d_data,_density.d_data,_data_byte_vol,cudaMemcpyHostToDevice));
+			ret += int(cudaSuccess != cudaSetDevice(gpu_index));
+			ret += int(cudaSuccess != cudaMemcpy(_density.d_data,_density.d_data,_data_byte_vol,cudaMemcpyHostToDevice));
+			return ret;
 	}
 	return 0;
 }
 
-int Target::fromGpu(){
+int Target::fromGpu(int gpu_index){
+	int ret = 0;
 	if (_source_type == MRC && _cpu_initiated && _gpu_initiated){
-			return int(cudaSuccess != cudaMemcpy(_density.d_data,_density.d_data,_data_byte_vol,cudaMemcpyDeviceToHost));
+			ret += int(cudaSuccess != cudaSetDevice(gpu_index));
+			ret += int(cudaSuccess != cudaMemcpy(_density.d_data,_density.d_data,_data_byte_vol,cudaMemcpyDeviceToHost));
+			return ret;
 	}
 	return 0;
 }

Target.h

@@ -18,15 +18,16 @@ public:
 	static Target fromMrc(const char * fileName);
 	size_t getCpuMem();
 	size_t getGpuMem();
+	int loadToGpu(int gpu_index);
 	int initCpu();
-	int initGpu();
-	int toGpu();
-	int fromGpu();
+	int initGpu(int gpu_index);
+	int toGpu(int gpu_index);
+	int fromGpu(int gpu_index);
 	virtual ~Target();
 private:
 	Density _density;
 	size_t _data_byte_vol = 0;
-	SourceType _source_type;
+	SourceType _source_type = NONE;
 	bool _cpu_initiated = false;
 	bool _gpu_initiated = false;
 };