Reading results while kernel is executing... "Streams" is the right way?

Accelerated Computing CUDA CUDA Programming and Performance
1

I have a kernel that keeps producing results as they are available…

I want to see if some result is available on it… (just peek)…while the kernel is still executing…
Can I use another “stream” to copy the result-set and peek into it? (assuming hardware supports concurrent copy and exec).

The bottomline of my question is : Will another “stream” share the same context as the “executing kernel” ?

THanks,

2

You can try, but I am 99% sure there are no guarantees in the concurrency model that it will work.

By definition, that is always the case - streams exist within a single context. But it doesn’t imply you can successfully do what you are asking about.

3

You can try, but I am 99% sure there are no guarantees in the concurrency model that it will work.

By definition, that is always the case - streams exist within a single context. But it doesn’t imply you can successfully do what you are asking about.

4

Ok… Thanks!

Say, I use “threadfence” kindaa memory fence functions…, I should be able to read back the intermediate results… no? (OR) Is this what u call “undefined behavior” ?

5

Ok… Thanks!

Say, I use “threadfence” kindaa memory fence functions…, I should be able to read back the intermediate results… no? (OR) Is this what u call “undefined behavior” ?