cudaBindTextureToArray in PyCuda

Question

Is-there a way to bind an array that is already on the gpu to a texture using PyCuda ?

There is already a cuda.bind_array_to_texref(cuda.make_multichannel_2d_array(...), texref) that binds an array on the CPU to a texture, but I couldn't find the equivalent of cudaBindTextureToArray in PyCuda if the array is already on the device. For example, doing :

myArray = [1, 2, 3]
myArray_d = gpu.to_gpu(myArray)  # then performs some computations on  it, and then
cuda.bind_texture_to_array(myArray_d, texref)

Isn't `pycuda.driver.TextureReference.set_array()` what you want? — talonmies, Nov 19 '12 at 21:27
great, thanks! :) (you can add that as an answer and I'll validate) — nbonneel, Nov 19 '12 at 22:05
the array passed to set_array is an array on the GPU, right? — nbonneel, Nov 19 '12 at 22:06
Yes, a `pycuda.driver.Array`. If you have linear memory use `set_address` or `set_address_2d`for pitched linear memory. — talonmies, Nov 19 '12 at 22:11

score 3 · Accepted Answer · answered Nov 19 '12 at 22:12

If you want to bind an existing CUDA array in GPU memory to a texture reference, then pycuda.driver.TextureReference.set_array() is probably what you want. Note the PyCUDA is built on the driver API, so the call you are looking for is actually cuTexRefSetArray rather than cudaBindTextureToArray.

cudaBindTextureToArray in PyCuda

1 Answers1