Question

Does the data in cache persist across kernel launches? I’ve around 5 kernels which refer same data again and again so I bound them to texture but can anybody tell me whether data in texture cache persists across kernel launches?

Answer 1

Section 3.2.10.4 of the CUDA C Programming Guide says on this:

a thread can safely read some texture or surface memory location
only if this memory location has been updated by a previous kernel
call or memory copy

i.e. the texture cache is invalidated between kernels (btw. it would probably have been faster to consult the documentation than to wait for an answer on stackoverflow).

However the lifetime of your data in the texture cache probably is much shorter than that, given there are only a handful of bytes available for each thread.