valgrind-cputil

cputil - a valgrind tool for estimating CPU utilization

Copyright (C) 2016-2017,2021,2023 -- Matthew Wette.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.3 or any later version published by the Free Software Foundation; with no Invariant Sections, no Front-Cover Texts, and no Back-Cover Texts. A copy of the license is included with the distribution as COPYING.DOC.

The cputil program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version.

Introduction

This valgrind tool provides approximate cycle counts for user-defined regions of code. It works on a per-thread basis. The following program, when executed under cputil/valgrind will report the estimated number of clock cycles required to execute foo(). The default processor model is a PowerPC750, but can be changed with a user-loadable table.

#include <stdio.h>
#include "cputil.h"
int main() {
  long cycles;
  CU_REGTHR();                /* register the current thread */
  for (i = 0; i < 5; i++) {
    CU_CLRCTR();              /* clear the cycle counter */
    foo(i);
    cycles += CU_GETCTR();    /* read the cycle counter */
    bar(i);
  }
  printf("clk-cycles=%d\n", cycles)
}

Installation

This works with valgrind-3.22.0. I assume it will work with 3.22.X. To work with older versions of valgrind, try other git branches.

To install:

1. Download and unpack valgrind-3.22.X

2. Clone or otherwise install this cputil distribution in a directory called cputil under valgrind-3.22.X

3. In the valgrind top-level directory execute:

   $ cputil/upd-vg-files
   $ ./autogen.sh
   

4a. In the valgrind top directory run configure, make and install:

   $ ./configure --prefix=/usr/local
   $ make 
   $ sudo make install

On macOS, you may need to perform the following steps:

   $ CFLAGS="" ./configure --prefix=/usr/local \
     ./configure --enable-only64bit --prefix=/usr/local

4b. Optionally, if you want a minimal distribution, use this sequence,

using /var/tmp/mindist as a place to install the distribution:

   $ ./configure --prefix=/usr/local
   $ make 
   $ mkdir -p /var/tmp/mindist
   $ make DESTDIR=/var/tmp/mindist/ install
   $ sh cputil/mindist DESTDIR=/var/tmp/mindist

The above will generate a distribution. To install, become root if necessary, then execute the following:

   # cd /var/tmp/mindist
   # cd usr/local
   # tar cf - . | (cd /usr/local; tar xvf -)

In this case, the command to execute is cputil' instead ofvalgrind --tool=cputil`.

dumping the op-count table

To dump the internal table execute

   $ valgrind --tool=cputil --dump-op-table=dump.cut true
     [OR, for min dist, cputil --dump-op-table=dump.cut true]

There is no way to dump the op-table w/o specifing a program to so we use use /bin/true.

loading user-defined op-count table

A dumped op table can be edited for clock counts and used for another run:

   $ valgrind --tool=cputil --load-op-table=dump.cut my_program

Threads

Support for multi-threaded code has been added.

Demo

Look in cputil/tests/demo1.c

Other issues

If you see the following error during compilation then your version of valgrind still does works to MPI1.

libmpiwrap.c: In function 'showTy': libmpiwrap.c:281:19: error: 'MPI_UB' undeclared (first use in this function);

  did you mean 'MPI_IO'?
281 |    else if (ty == MPI_UB)             fprintf(f,"UB");
    |                   ^~~~~~
    |                   MPI_IO

The included patch

0001-Drop-MPI-1-support.patch

from

https://bugs.kde.org/show_bug.cgi?id=401416#c3

should help. To apply, from valgring top directory,

   $ patch -b -p1 < cputil/patch/0001-Drop-MPI-1-support.patch