I’m trying to use the CUDA Driver API to construct a texture object but I keep getting an invalid value error. The texture will be indexed in the kernel code without needing any interpolation. The code I’m using is

CUDA_ARRAY_DESCRIPTOR array_desc;

array_desc.Width = size;
array_desc.Height = 1;
array_desc.Format = CU_AD_FORMAT_FLOAT;
array_desc.NumChannels = 1;

CUDA_RESOURCE_DESC resource_desc;
CUDA_TEXTURE_DESC texture_desc;

memset(&resource_desc, 0, sizeof(CUDA_RESOURCE_DESC));
memset(&texture_desc, 0, sizeof(CUDA_TEXTURE_DESC));

resource_desc.resType = CU_RESOURCE_TYPE_ARRAY;
texture_desc.addressMode[0] = CU_TR_ADDRESS_MODE_BORDER;
texture_desc.addressMode[1] = CU_TR_ADDRESS_MODE_BORDER;
texture_desc.addressMode[2] = CU_TR_ADDRESS_MODE_BORDER;

cuArrayCreate(&resource_desc.res.array.hArray, &array_desc);

CUtexObject tex;
cuTexObjectCreate(&tex, &resource_desc, &texture_desc, NULL); // This is giving me an invalid argument error.

When I run this cuTexObjectCreate gives me an CUDA_ERROR_INVALID_VALUE error.

An additional question are:

When I destruct this object, do I need to explicit destruct the CUArray object also?

Why are the CUDA_RESOURCE_DESC and CUDA_TEXTURE_DESC passed by pointers instead of value or references? Is it ok if these objects go out of scope or do I need to keep them in memory for the life time off the CUtexObject?