使用 OpenMP 的 OProfile

Question

我通过执行以下操作将 OProfile 用于 OpenMP 并行化代码，

$ gcc -I/usr/include/hdf5/serial/ -std=c11 -O3 -fopt-info -fopenmp sp_linsvm.c -o sp_linsvm -lhdf5_serial
$ sudo ocount --events=CPU_CLK_UNHALTED,LLC_MISSES,LLC_REFS,MEM_INST_RETIRED,BR_MISP_EXEC, ./sp_linsvm
Events were actively counted for 22.0 seconds.
Event counts (scaled) for /home/aidan/progs/linsvm/sp_linsvm:
    Event                    Count                    % time counted
    BR_MISP_EXEC             6,523,181                80.00
    CPU_CLK_UNHALTED         225,384,009,348          80.00
    LLC_MISSES               276,587,407              80.02
    LLC_REFS                 1,098,236,806            80.00
    MEM_INST_RETIRED         51,754,855,734           79.99

我如何知道事件是按 CPU 还是整体计算的？我很确定它作为一个整体，因为如果我在没有 OpenMP 的情况下编译，它们接近数字，但我想确定。

Answer 1

ocount ... ./program 的默认模式是 "command"。据我了解，如果没有 -t (--separate-thread) 或 -c (--separate-cpu) 选项，来自所有线程的数据将被聚合。

所以，检查文档 http://oprofile.sourceforge.net/doc/controlling-counter.html#controlling-ocount 并尝试 -t / -c 选项...

--separate-thread / -t This option can be used in conjunction with either the --process-list or --thread-list option to display event counts on a per-thread (per-process) basis. Without this option, all counts are aggregated.

--separate-cpu / -c This option can be used in conjunction with either the --system-wide or --cpu-list option to display event counts on a per-cpu basis. Without this option, all counts are aggregated.

使用 OpenMP 的 OProfile

OProfile with OpenMP

parallel-processing

profiling

openmp

oprofile