在cuda中,使用自定義結構體,

cudaimprocess.h

//cuda頭文件

struct CoreDataGPU
{
	// host cpu
	float *H_X;
	float *H_Y ;
	float *H_Z ;

	//device gpu 
	float *D_X;
	float *D_Y ;
	float *D_Z ;
};

//開闢內存空間.
void initPointGPU(PointGpu **_HostPointGPU,int arraySize);

//釋放內存空間
void freePointGPU(PointGpu &_HostPointGPU);

//核心算子.
__global__ void addPoint(PointGpu* _PointGPU,int arraySize);

//對上面結構體進行處理測試.
void PointGpuMethod_Test();

//測試二:上面結構體,假設裏面結構體不是指針,只是一個變量.,那麼做一個結構體數組
//
void PointGpuMethod_dim1(); 

cudaimproces.cu

///////////////////////
void initPointGPU(PointGpu **_HostPointGPU, int arraySize)
{
	//開闢空間. 在本地開闢空間.
	cudaMallocHost((void**)(_HostPointGPU), sizeof(PointGpu));
	
	//在cpu上開闢空間,這裏考慮的是:後續進行類封裝的時候,直接從gpu上對應的地址拷貝到cpu對應的數據內存,
	cudaMallocHost((void**)&(*_HostPointGPU)->host_X,sizeof(float)*arraySize);
	cudaMallocHost((void**)&(*_HostPointGPU)->host_Y, sizeof(float)*arraySize);
	cudaMallocHost((void**)&(*_HostPointGPU)->host_Z, sizeof(float)*arraySize);

//在gpu上開闢空間
	cudaMalloc((void**)&(*_HostPointGPU)->device_X,sizeof(float)*arraySize);
	cudaMalloc((void**)&(*_HostPointGPU)->device_Y, sizeof(float)*arraySize);
	cudaMalloc((void**)&(*_HostPointGPU)->device_Z, sizeof(float)*arraySize);

}
void freePointGPU(PointGpu &_HostPointGPU)
{
	cudaFreeHost(_HostPointGPU.host_X);
	cudaFreeHost(_HostPointGPU.host_Y);
	cudaFreeHost(_HostPointGPU.host_Z);
	cudaFree(_HostPointGPU.device_X);
	cudaFree(_HostPointGPU.device_Y);
	cudaFree(_HostPointGPU.device_Z);
	cudaFreeHost(&_HostPointGPU);
}

//核函數,我就直接賦值了,不做計算.
__global__ void addPoint(PointGpu *_PointGPU, int arraySize)
{
	int tid = threadIdx.x;
	if (_PointGPU != nullptr && tid < arraySize)
	{
		_PointGPU->device_X[tid] = 9;
		_PointGPU->device_Y[tid] = 8;
		_PointGPU->device_Z[tid] = 10;
	}
}

//接下來就是如何使用了.
void PointGpuMethod_Test()
{

////////這裏是單一的數據

	const int arraySize = 10;
	float X[arraySize] = {0,};
	float Y[arraySize] = { 0, };
	float Z[arraySize] = { 0 ,};
	memset(X, 0, sizeof(float)*arraySize);
	memset(Y, 0, sizeof(float)*arraySize);
	memset(Z, 0, sizeof(float)*arraySize);

	PointGpu* source = {nullptr};
	initPointGPU(&source,arraySize);  //初始化數據.

	addPoint << < 1,arraySize >> > (source, arraySize); //計算
	cudaDeviceSynchronize(); //同步 或者在覈函數裏面使用 __syncthreads();

	cudaMemcpy(source->host_Z, source->device_Z, arraySize*sizeof(float), cudaMemcpyKind::cudaMemcpyDeviceToHost); //下載數據.
	cudaMemcpy(source->host_Y, source->device_Y, arraySize * sizeof(float), cudaMemcpyKind::cudaMemcpyDeviceToHost); //下載數據.
	cudaMemcpy(source->host_X, source->device_X, arraySize * sizeof(float), cudaMemcpyKind::cudaMemcpyDeviceToHost); //下載數據.

	memcpy(Z, source->host_Z,sizeof(float)*arraySize);
	memcpy(Y, source->host_Y, sizeof(float)*arraySize);
	memcpy(X, source->host_X, sizeof(float)*arraySize);

	//輸出：
	std::cout << "GPU下載的數據:" << std::endl;
	for (int i = 0; i < arraySize; i++)
	{
		std::cout << Z[i]<<" "<<X[i]<<" "<<Y[i] << " | ";
	}
	std::cout << std::endl;

	freePointGPU(*source);

///////////////////GPU 下載數據數組. 

}

void PointGpuMethod_dim1()
{
	const int arraySize = 10;
	float X[arraySize] = { 0, };
	float Y[arraySize] = { 0, };
	float Z[arraySize] = { 0 , };
	memset(X, 0, sizeof(float)*arraySize);
	memset(Y, 0, sizeof(float)*arraySize);
	memset(Z, 0, sizeof(float)*arraySize);
	
	const int BUfferNumber = 10;
	PointGpu** source = new PointGpu*[BUfferNumber]; //創建二級指針
	for (int i = 0; i < BUfferNumber; i++)
	{
		initPointGPU((source+i), 110000);  //初始化數據. //每級指針開闢空間
	}

	//初始化之後，賦值.

	for (int i = 0; i < BUfferNumber; i++)
	{
		addPoint << < 1, 10 >> > (source[i], 1);
		cudaDeviceSynchronize();

		//獲取數值.賦值，並輸出數據.
		float buffer[3][2] = {0};
		cudaMemcpy(buffer[0], source[i]->device_X, sizeof(float),cudaMemcpyKind::cudaMemcpyDeviceToHost);
		cudaMemcpy(buffer[1], source[i]->device_Y, sizeof(float), cudaMemcpyKind::cudaMemcpyDeviceToHost);
		cudaMemcpy(buffer[2], source[i]->device_Z, sizeof(float), cudaMemcpyKind::cudaMemcpyDeviceToHost);
		std::cout << buffer[0][0] << " " << buffer[1][0] << " " << buffer[2][0] << " "<<std::endl;
	}
	

	//釋放空間
	for (int i = 0; i < BUfferNumber; i++)
	{
		freePointGPU(*source[i]);
	}

	delete []source;
}

///以上經過測試,連續GPU內存以及使用情況,穩定.

這樣,cuda自定義結構體測試成功,這個只是最基本的,其它都可以在這個基礎上做改動.