[Starpu-devel] A question regarding the STARPU_REDUX flag

[Mirko Myllykoski <mirkom@cs.umu.se>]

Hi,

I have a question regarding STARPU_REDUX flag.

I am confused over the following two statement in the StarPU Handbook:

 a) "StarPU provides a mode STARPU_REDUX, which permits to optimize 
this case: it will allocate a buffer on each memory node, and accumulate 
intermediate results there. When the data is eventually accessed in the 
normal mode STARPU_R, StarPU will collect the intermediate results in 
just one buffer." [1]

 b) "Set the codelets to be used for handle when it is accessed in the 
mode STARPU_REDUX. Per-worker buffers will be initialized with the 
codelet init_cl, and reduction between per-worker buffers will be done 
with the codelet redux_cl." [2]

My understanding is the following:

 * Statement (a) states that StarPU maintains a reduction buffer for 
each *memory node*. That is, in shared memory configuration, there is 
going to be just one reduction buffer and all tasks that access a 
STARPU_REDUX handle are executed sequentially.

 * Statement (b) states that StarPU maintains a reduction buffer for 
each *worker*. That is, in shared memory configuration, there is going 
to be several reduction buffers and all tasks that access a STARPU_REDUX 
handle are executed in parallel.

Which one of these these statements is correct? I am getting results 
that seem to indicate that the statement (a) is correct.

Best regards,
Mirko Myllykoski

[1]: 
http://starpu.gforge.inria.fr/doc/html/DataManagement.html#DataReduction
[2]: 
http://starpu.gforge.inria.fr/doc/html/group__API__Data__Management.html#ga1f5ba1c1cfefc1f81a4095cf3c213e54

--
By sending an email to Umeå University, the University processes your 
personal data. For more information, please read: https://www.umu.se/en/gdpr