Textures and Memcpy handling of textures with changing memor

Accelerated Computing CUDA CUDA Programming and Performance
#1

Hallo , i have following Scenario:

1)I do a memcpy of data to some area on the device.
2)Then i bind a texture to that memory area.

now i use the texture for some time. Texture cache gets filled and so on.

  1. Now i execute a memcopy again and fill memory of the texture with new data.

Now the question:
Do i get now old cache issues ? This means by accessing old address i get old results from cache rather using new data or does the cache gets automatically invalidated?

Do i need to unbind the texture before step 3 and bind it after step 3 again ?

What do you think about it ?

#2

AFAIK texture anc constant cache is invalidated between kernel calls.

#3

If it is true, then it solves all my problems, especially of bind and unbind overhead.