Texture cache misses

Accelerated Computing CUDA CUDA Programming and Performance
#1

Hello everyone,

I have two different CUDA kernels that I have written to utilize the texture memory for all device reads. I would like to know how many texture cache misses occurred in both kernels. I have used the visual profiler but did not find any options regarding cache misses.

Does anybody know of a good method for determining the number of texture cache misses?

Thanks
Steffen

#2

Unfortunately, no.

I count the number of bytes read in each kernel and calculate an effective bandwidth to see if I’m getting close to the device capability.