understand MAGMA output

Open discussion for MAGMA library (Matrix Algebra on GPU and Multicore Architectures)

understand MAGMA output

Postby ar_wwu » Wed Jan 20, 2016 3:38 am

Hello,

I am currently testing the MAGMA MIC library and have problems to understand the test cases' output. For every case, a runtime in ms is depicted. Is this the pure kernel runtime (not including host-to-device/device-to-host data transfers) or rather the total runtime, i.e., host-to-device/device-to-host data transfers + kernel runtime?

Thank you very much in advance.

Best regards
ar_wwu
 
Posts: 2
Joined: Wed Jan 20, 2016 3:33 am

Re: understand MAGMA output

Postby mgates3 » Mon Jan 25, 2016 4:35 pm

It depends on the tester:

For BLAS routines, such as magma_zgemm, it should just be the function itself, excluding data transfers.

For _mic interface routines, such as magma_zgetrf_mic, it is also just the function itself, as the matrix A starts and ends on the MIC. There are, however, data transfers during the function that are counted.

For CPU interface routines, such as magma_zgetrf, the time includes copying the matrix to the MIC and back, as the matrix A starts and ends in CPU memory.

-mark
mgates3
 
Posts: 738
Joined: Fri Jan 06, 2012 2:13 pm

Re: understand MAGMA output

Postby ar_wwu » Mon Jan 25, 2016 5:23 pm

Thank you very much.
Do you know what is the case for the "testing_sgemv" test-case of the mic library? Is the depicted runtime referring to the function itself only, excluding data transfers?

Best regards
ar_wwu
 
Posts: 2
Joined: Wed Jan 20, 2016 3:33 am


Return to User discussion

Who is online

Users browsing this forum: No registered users and 1 guest

cron