【CUDA开发】CUDA面内存拷贝用法总结

标签（空格分隔）：【CUDA开发】

主要是在调试CUDA硬解码并用D3D9或者D3D11显示的时候遇到了一些代码，如下所示：

CUdeviceptr    g_pRgba = 0;
CUDA_MEMCPY2D memcpy2D = { 0 };
memcpy2D.srcMemoryType = CU_MEMORYTYPE_DEVICE;
memcpy2D.srcDevice = g_pRgba;
memcpy2D.srcPitch = nWidth * 4;
memcpy2D.dstMemoryType = CU_MEMORYTYPE_ARRAY;
memcpy2D.dstArray = array;
memcpy2D.dstPitch = nWidth * 4;
memcpy2D.WidthInBytes = nWidth * 4;
memcpy2D.Height = nHeight;

// clear the surface to solid white
checkCudaErrors(cuMemcpy2D(&memcpy2D));

将设备内存g_pRgba中的数据拷贝到面内存中；还有另外一种情况：

src GMEM pointer : dpSrc  
src GMEM layout  : 100x100  
dst GMEM pointer : dpDst  
dst GMEM layout  : 50*50  
将src GMEM按行序初始化为：0~9999的值  
CUDA_MEMCPY2D planeMem;  
memset(&planeMem,0,sizeof(planeMem));  
planeMem.srcMemoryType=CU_MEMORYTYPE_DEVICE;  
planeMem.srcDevice    =dpSrc;  
planeMem.srcXInBytes  =25*sizeof(float);  
planeMem.srcY         =25;  
planeMem.srcPitch     =100*sizeof(float);  
planeMem.dstMemoryType=CU_MEMORYTYPE_DEVICE;  
planeMem.dstDevice    =dpDst;  
planeMem.dstXInBytes  =0;  
planeMem.dstY         =0;  
planeMem.dstPitch     =50*sizeof(float);  
planeMem.WidthInBytes =planeMem.dstPitch;  
planeMem.Height       =50;  
cuMemcpy2DUnaligned(&planeMem); //如果数据已经对齐则最好使用cuMemcpy2D，否则必须使用该函数，另外当内存是使用cuMemAllocPitch分配的时候，如果内存布局本事不是2的次幂，则需要将planeMem的srcPitch和dstPitch设置为通过cuMemAllocPitch得到的pitch参数，而不是内存本身的布局大小*sizeof(TYPE)

2017-01-26 23:28
张朋艺 pyZhangBIT2010@126.com