23.10.2014 Views

PGPROF User's Guide - The Portland Group

PGPROF User's Guide - The Portland Group

PGPROF User's Guide - The Portland Group

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Using <strong>PGPROF</strong><br />

Figure 11 Profile with Hardware Event Counter<br />

2.9. Profiling GPU Programs<br />

You can use <strong>PGPROF</strong> to analyze the performance of GPU programs. GPU performance data<br />

is included in the profile, pgprof.out, when a GPU program is run using pgcollect. PGI<br />

provides two methods of programming GPUs: OpenACC, which uses programs and directives to<br />

tell the compiler how to generate GPU code, and CUDA Fortran, which is used to program the<br />

GPU more directly.<br />

<strong>The</strong> next section describes how to use pgcollect with OpenACC programs, and the subsequent<br />

section describes using it with CUDA Fortran programs.<br />

2.9.1. Profiling OpenACC Programs<br />

For OpenACC the profiling procedure is the same as for host-only programs, except that<br />

<strong>PGPROF</strong> provides an Accelerator Performance tab that allows you to review profiling<br />

information provided by the accelerator. You do not need to build or run with any special options<br />

to collect accelerator performance data.<br />

PGI Profiler User <strong>Guide</strong> 22

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!