How to create 3D memory using the driver api?

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello,

in the runtime API there is a method cudaMalloc3D to allocate 1D, 2D or 3D objects in device memory.
When reading the driver API i cannot find a method which does somethign equivalent.

Is there an analog method for the cudaMalloc3D in the driver API.

Thanks
Martin

2

Is there any reason cuMemAllocPitch wouldn’t work? It’s not 3D but it will pad rows so that coalescing works properly…

3

Hello,

of course you are right. I only thought that there is a method in the runtime API so there must be a method to allocate a 3D linear memory also.

regards

Martin

4

The example for setting 3D memory with 3D texture as down:

[font="Courier New"]	

	CUarray devarr;

	CUDA_ARRAY3D_DESCRIPTOR desc;

	desc.Width	  =sizex;

	desc.Height	 =sizey;

	desc.Depth	  =sizez;

	desc.NumChannels=4;

	desc.Format	 =CU_AD_FORMAT_FLOAT;

	desc.Flags	  =0;

	CU_SAFE_CALL(cuArray3DCreate(&devarr,&desc));

	float4* voxels;

	//alloc space in device or host memory and apply source data for "voxels"

	memset(&memPro,0,sizeof(memPro));

	memPro.dstMemoryType=CU_MEMORYTYPE_ARRAY;

	memPro.dstArray		  =devarr;

	memPro.srcMemoryType=CU_MEMORYTYPE_HOST;

	memPro.srcHost		   =voxels;

	memPro.srcPitch		  =sizex*sizeof(float4);

	memPro.WidthInBytes =memPro.srcPitch;

	memPro.Height		   =sizey;

	memPro.Depth			=sizez;

	CU_SAFE_CALL(cuMemcpy3D(&memPro));

	delete [] voxels;

	CU_SAFE_CALL(cuTexRefSetArray(texref,devarr,CU_TRSA_OVERRIDE_FORMAT));	

[/font]