cachegrind/docs/cg-manual.xml

<?xml version="1.0"?> <!-- -*- sgml -*- -->
<!DOCTYPE chapter PUBLIC "-//OASIS//DTD DocBook XML V4.2//EN"
  "http://www.oasis-open.org/docbook/xml/4.2/docbookx.dtd"
[ <!ENTITY % vg-entities SYSTEM "../../docs/xml/vg-entities.xml"> %vg-entities; ]>


<chapter id="cg-manual" xreflabel="Cachegrind: a cache-miss profiler">
<title>Cachegrind: a cache and branch profiler</title>

<sect1 id="cg-manual.cache" xreflabel="Cache profiling">
<title>Cache and branch profiling</title>

<para>To use this tool, you must specify
<computeroutput>--tool=cachegrind</computeroutput> on the
Valgrind command line.</para>

<para>Cachegrind is a tool for finding places where programs
interact badly with typical modern superscalar processors
and run slowly as a result.
In particular, it will do a cache simulation of your program,
and optionally a branch-predictor simulation, and can
then annotate your source line-by-line with the number of cache
misses and branch mispredictions.  The following statistics are 
collected:</para>
<itemizedlist>
  <listitem>
    <para>L1 instruction cache reads and misses;</para>
  </listitem>
  <listitem>
    <para>L1 data cache reads and read misses, writes and write
    misses;</para>
  </listitem>
  <listitem>
    <para>L2 unified cache reads and read misses, writes and
    writes misses.</para>
  </listitem>
  <listitem>
    <para>Conditional branches and mispredicted conditional branches.</para>
  </listitem>
  <listitem>
    <para>Indirect branches and mispredicted indirect branches.  An
    indirect branch is a jump or call to a destination only known at
    run time.</para>
  </listitem>
</itemizedlist>

<para>On a modern machine, an L1 miss will typically cost
around 10 cycles, an L2 miss can cost as much as 200
cycles, and a mispredicted branch costs in the region of 10
to 30 cycles.  Detailed cache and branch profiling can be very useful
for improving the performance of your program.</para>

<para>Also, since one instruction cache read is performed per
instruction executed, you can find out how many instructions are
executed per line, which can be useful for traditional profiling
and test coverage.</para>

<para>Branch profiling is not enabled by default.  To use it, you must
additionally specify <computeroutput>--branch-sim=yes</computeroutput>
on the command line.</para>


<sect2 id="cg-manual.overview" xreflabel="Overview">
<title>Overview</title>

<para>First off, as for normal Valgrind use, you probably want to
compile with debugging info (the
<computeroutput>-g</computeroutput> flag).  But by contrast with
normal Valgrind use, you probably <command>do</command> want to turn
optimisation on, since you should profile your program as it will
be normally run.</para>

<para>The two steps are:</para>
<orderedlist>
  <listitem>
    <para>Run your program with <computeroutput>valgrind
    --tool=cachegrind</computeroutput> in front of the normal
    command line invocation.  When the program finishes,
    Cachegrind will print summary cache statistics. It also
    collects line-by-line information in a file
    <computeroutput>cachegrind.out.pid</computeroutput>, where
    <computeroutput>pid</computeroutput> is the program's process
    id.</para>

    <para>Branch prediction statistics are not collected by default.
    To do so, add the flag
    <computeroutput>--branch-sim=yes</computeroutput>.
    </para>

    <para>This step should be done every time you want to collect
    information about a new program, a changed program, or about
    the same program with different input.</para>
  </listitem>

  <listitem>
    <para>Generate a function-by-function summary, and possibly
    annotate source files, using the supplied
    <computeroutput>cg_annotate</computeroutput> program. Source
    files to annotate can be specified manually, or manually on
    the command line, or "interesting" source files can be
    annotated automatically with the
    <computeroutput>--auto=yes</computeroutput> option.  You can
    annotate C/C++ files or assembly language files equally
    easily.</para>

    <para>This step can be performed as many times as you like
    for each Step 2.  You may want to do multiple annotations
    showing different information each time.</para>
  </listitem>

</orderedlist>

<para>As an optional intermediate step, you can use the supplied
<computeroutput>cg_merge</computeroutput> program to sum together the
outputs of multiple Cachegrind runs, into a single file which you then
use as the input for
<computeroutput>cg_annotate</computeroutput>.</para>

<para>These steps are described in detail in the following
sections.</para>

</sect2>


<sect2 id="cache-sim" xreflabel="Cache simulation specifics">
<title>Cache simulation specifics</title>

<para>Cachegrind simulates a machine with independent
first level instruction and data caches (I1 and D1), backed by a
unified second level cache (L2).  This configuration is used by almost
all modern machines.  Some old Cyrix CPUs had a unified I and D L1
cache, but they are ancient history now.</para>

<para>Specific characteristics of the simulation are as
follows:</para>

<itemizedlist>

  <listitem>
    <para>Write-allocate: when a write miss occurs, the block
    written to is brought into the D1 cache.  Most modern caches
    have this property.</para>
  </listitem>

  <listitem>
    <para>Bit-selection hash function: the line(s) in the cache
    to which a memory block maps is chosen by the middle bits
    M--(M+N-1) of the byte address, where:</para>
    <itemizedlist>
      <listitem>
        <para>line size = 2^M bytes</para>
      </listitem>
      <listitem>
        <para>(cache size / line size) = 2^N bytes</para>
      </listitem>
    </itemizedlist> 
  </listitem>

  <listitem>
    <para>Inclusive L2 cache: the L2 cache replicates all the
    entries of the L1 cache.  This is standard on Pentium chips,
    but AMD Opterons, Athlons and Durons 
    use an exclusive L2 cache that only holds
    blocks evicted from L1.  Ditto most modern VIA CPUs.</para>
  </listitem>

</itemizedlist>

<para>The cache configuration simulated (cache size,
associativity and line size) is determined automagically using
the CPUID instruction.  If you have an old machine that (a)
doesn't support the CPUID instruction, or (b) supports it in an
early incarnation that doesn't give any cache information, then
Cachegrind will fall back to using a default configuration (that
of a model 3/4 Athlon).  Cachegrind will tell you if this
happens.  You can manually specify one, two or all three levels
(I1/D1/L2) of the cache from the command line using the
<computeroutput>--I1</computeroutput>,
<computeroutput>--D1</computeroutput> and
<computeroutput>--L2</computeroutput> options.</para>

<para>On PowerPC platforms
Cachegrind cannot automatically 
determine the cache configuration, so you will 
need to specify it with the
<computeroutput>--I1</computeroutput>,
<computeroutput>--D1</computeroutput> and
<computeroutput>--L2</computeroutput> options.</para>


<para>Other noteworthy behaviour:</para>

<itemizedlist>
  <listitem>
    <para>References that straddle two cache lines are treated as
    follows:</para>
    <itemizedlist>
      <listitem>
        <para>If both blocks hit --&gt; counted as one hit</para>
      </listitem>
      <listitem>
        <para>If one block hits, the other misses --&gt; counted
        as one miss.</para>
      </listitem>
      <listitem>
        <para>If both blocks miss --&gt; counted as one miss (not
        two)</para>
      </listitem>
    </itemizedlist>
  </listitem>

  <listitem>
    <para>Instructions that modify a memory location
    (eg. <computeroutput>inc</computeroutput> and
    <computeroutput>dec</computeroutput>) are counted as doing
    just a read, ie. a single data reference.  This may seem
    strange, but since the write can never cause a miss (the read
    guarantees the block is in the cache) it's not very
    interesting.</para>

    <para>Thus it measures not the number of times the data cache
    is accessed, but the number of times a data cache miss could
    occur.</para>
  </listitem>

</itemizedlist>

<para>If you are interested in simulating a cache with different
properties, it is not particularly hard to write your own cache
simulator, or to modify the existing ones in
<computeroutput>vg_cachesim_I1.c</computeroutput>,
<computeroutput>vg_cachesim_D1.c</computeroutput>,
<computeroutput>vg_cachesim_L2.c</computeroutput> and
<computeroutput>vg_cachesim_gen.c</computeroutput>.  We'd be
interested to hear from anyone who does.</para>

</sect2>


<sect2 id="branch-sim" xreflabel="Branch simulation specifics">
<title>Branch simulation specifics</title>

<para>Cachegrind simulates branch predictors intended to be
typical of mainstream desktop/server processors of around 2004.</para>

<para>Conditional branches are predicted using an array of 16384 2-bit
saturating counters.  The array index used for a branch instruction is
computed partly from the low-order bits of the branch instruction's
address and partly using the taken/not-taken behaviour of the last few
conditional branches.  As a result the predictions for any specific
branch depend both on its own history and the behaviour of previous
branches.  This is a standard technique for improving prediction
accuracy.</para>

<para>For indirect branches (that is, jumps to unknown destinations)
Cachegrind uses a simple branch target address predictor.  Targets are
predicted using an array of 512 entries indexed by the low order 9
bits of the branch instruction's address.  Each branch is predicted to
jump to the same address it did last time.  Any other behaviour causes
a mispredict.</para>

<para>More recent processors have better branch predictors, in
particular better indirect branch predictors.  Cachegrind's predictor
design is deliberately conservative so as to be representative of the
large installed base of processors which pre-date widespread
deployment of more sophisticated indirect branch predictors.  In
particular, late model Pentium 4s (Prescott), Pentium M, Core and Core
2 have more sophisticated indirect branch predictors than modelled by
Cachegrind.  </para>

<para>Cachegrind does not simulate a return stack predictor.  It
assumes that processors perfectly predict function return addresses,
an assumption which is probably close to being true.</para>

<para>See Hennessy and Patterson's classic text "Computer
Architecture: A Quantitative Approach", 4th edition (2007), Section
2.3 (pages 80-89) for background on modern branch predictors.</para>

</sect2>


</sect1>



<sect1 id="cg-manual.profile" xreflabel="Profiling programs">
<title>Profiling programs</title>

<para>To gather cache profiling information about the program
<computeroutput>ls -l</computeroutput>, invoke Cachegrind like
this:</para>

<programlisting><![CDATA[
valgrind --tool=cachegrind ls -l]]></programlisting>

<para>The program will execute (slowly).  Upon completion,
summary statistics that look like this will be printed:</para>

<programlisting><![CDATA[
==31751== I   refs:      27,742,716
==31751== I1  misses:           276
==31751== L2  misses:           275
==31751== I1  miss rate:        0.0%
==31751== L2i miss rate:        0.0%
==31751== 
==31751== D   refs:      15,430,290  (10,955,517 rd + 4,474,773 wr)
==31751== D1  misses:        41,185  (    21,905 rd +    19,280 wr)
==31751== L2  misses:        23,085  (     3,987 rd +    19,098 wr)
==31751== D1  miss rate:        0.2% (       0.1%   +       0.4%)
==31751== L2d miss rate:        0.1% (       0.0%   +       0.4%)
==31751== 
==31751== L2 misses:         23,360  (     4,262 rd +    19,098 wr)
==31751== L2 miss rate:         0.0% (       0.0%   +       0.4%)]]></programlisting>

<para>Cache accesses for instruction fetches are summarised
first, giving the number of fetches made (this is the number of
instructions executed, which can be useful to know in its own
right), the number of I1 misses, and the number of L2 instruction
(<computeroutput>L2i</computeroutput>) misses.</para>

<para>Cache accesses for data follow. The information is similar
to that of the instruction fetches, except that the values are
also shown split between reads and writes (note each row's
<computeroutput>rd</computeroutput> and
<computeroutput>wr</computeroutput> values add up to the row's
total).</para>

<para>Combined instruction and data figures for the L2 cache
follow that.</para>



<sect2 id="cg-manual.outputfile" xreflabel="Output file">
<title>Output file</title>

<para>As well as printing summary information, Cachegrind also
writes line-by-line cache profiling information to a user-specified
file.  By default this file is named
<computeroutput>cachegrind.out.pid</computeroutput>.  This file
is human-readable, but is intended to be interpreted by the accompanying
program <computeroutput>cg_annotate</computeroutput>, described
in the next section.</para>

<para>Things to note about the
<computeroutput>cachegrind.out.pid</computeroutput>
file:</para>

<itemizedlist>
  <listitem>
    <para>It is written every time Cachegrind is run, and will
    overwrite any existing
    <computeroutput>cachegrind.out.pid</computeroutput>
    in the current directory (but that won't happen very often
    because it takes some time for process ids to be
    recycled).</para>
    <para>
    To use a basename other than the default
    <computeroutput>cachegrind.out</computeroutput>,
    use the <computeroutput>--cachegrind-out-file</computeroutput>
    switch.</para>
    <para>
    To add further qualifiers to the output filename you can use 
    the core's <computeroutput>--log-file-qualifier</computeroutput>
    flag.  This extends the file name further with the text
    <computeroutput>.lfq.</computeroutput>followed by the
    contents of the environment variable specified by 
    <computeroutput>--log-file-qualifier</computeroutput>.
      </para>
  </listitem>
  <listitem>
    <para>It can be huge: <computeroutput>ls -l</computeroutput>
    generates a file of about 350KB.  Browsing a few files and
    web pages with a Konqueror built with full debugging
    information generates a file of around 15 MB.</para>
  </listitem>
</itemizedlist>

<para>The <computeroutput>.pid</computeroutput> suffix
on the output file name serves two purposes.  Firstly, it means you 
don't have to rename old log files that you don't want to overwrite.  
Secondly, and more importantly, it allows correct profiling with the
<computeroutput>--trace-children=yes</computeroutput> option of
programs that spawn child processes.</para>

</sect2>



<sect2 id="cg-manual.cgopts" xreflabel="Cachegrind options">
<title>Cachegrind options</title>

<!-- start of xi:include in the manpage -->
<para id="cg.opts.para">Using command line options, you can 
manually specify the I1/D1/L2 cache
configuration to simulate.  For each cache, you can specify the
size, associativity and line size.  The size and line size
are measured in bytes.  The three items
must be comma-separated, but with no spaces, eg:
<literallayout>    valgrind --tool=cachegrind --I1=65535,2,64</literallayout>

You can specify one, two or three of the I1/D1/L2 caches.  Any level not
manually specified will be simulated using the configuration found in
the normal way (via the CPUID instruction for automagic cache
configuration, or failing that, via defaults).</para>

<para>Cache-simulation specific options are:</para>

<variablelist id="cg.opts.list">

  <varlistentry id="opt.I1" xreflabel="--I1">
    <term>
      <option><![CDATA[--I1=<size>,<associativity>,<line size> ]]></option>
    </term>
    <listitem>
      <para>Specify the size, associativity and line size of the level 1
      instruction cache.  </para>
    </listitem>
  </varlistentry>

  <varlistentry id="opt.D1" xreflabel="--D1">
    <term>
      <option><![CDATA[--D1=<size>,<associativity>,<line size> ]]></option>
    </term>
    <listitem>
      <para>Specify the size, associativity and line size of the level 1
      data cache.</para>
    </listitem>
  </varlistentry>

  <varlistentry id="opt.L2" xreflabel="--L2">
    <term>
      <option><![CDATA[--L2=<size>,<associativity>,<line size> ]]></option>
    </term>
    <listitem>
      <para>Specify the size, associativity and line size of the level 2
      cache.</para>
    </listitem>
  </varlistentry>

  <varlistentry id="opt.cachegrind-out-file" xreflabel="--cachegrind-out-file">
    <term>
      <option><![CDATA[--cachegrind-out-file=<basename> ]]></option>
    </term>
    <listitem>
      <para>Write the profile data to 
            <computeroutput>basename.pid</computeroutput>
            rather than to the default output file, 
            <computeroutput>cachegrind.out.pid</computeroutput>.
      </para>
    </listitem>
  </varlistentry>

  <varlistentry id="opt.cache-sim" xreflabel="--cache-sim">
    <term>
      <option><![CDATA[--cache-sim=no|yes [yes] ]]></option>
    </term>
    <listitem>
      <para>Enables or disables collection of cache access and miss
            counts.</para>
    </listitem>
  </varlistentry>

  <varlistentry id="opt.branch-sim" xreflabel="--branch-sim">
    <term>
      <option><![CDATA[--branch-sim=no|yes [no] ]]></option>
    </term>
    <listitem>
      <para>Enables or disables collection of branch instruction and
            misprediction counts.  By default this is disabled as it
            slows Cachegrind down by approximately 25%.  Note that you
            cannot specify <computeroutput>--cache-sim=no</computeroutput>
            and <computeroutput>--branch-sim=no</computeroutput>
            together, as that would leave Cachegrind with no
            information to collect.</para>
    </listitem>
  </varlistentry>

</variablelist>
<!-- end of xi:include in the manpage -->

</sect2>


  
<sect2 id="cg-manual.annotate" xreflabel="Annotating C/C++ programs">
<title>Annotating C/C++ programs</title>

<para>Before using <computeroutput>cg_annotate</computeroutput>,
it is worth widening your window to be at least 120-characters
wide if possible, as the output lines can be quite long.</para>

<para>To get a function-by-function summary, run
<computeroutput>cg_annotate --pid</computeroutput> in a directory
containing a <filename>cachegrind.out.pid</filename> file.  The
<emphasis>--pid</emphasis> is required so that
<computeroutput>cg_annotate</computeroutput> knows which log file to use
when several are present.</para>

<para>The output looks like this:</para>

<programlisting><![CDATA[
--------------------------------------------------------------------------------
I1 cache:              65536 B, 64 B, 2-way associative
D1 cache:              65536 B, 64 B, 2-way associative
L2 cache:              262144 B, 64 B, 8-way associative
Command:               concord vg_to_ucode.c
Events recorded:       Ir I1mr I2mr Dr D1mr D2mr Dw D1mw D2mw
Events shown:          Ir I1mr I2mr Dr D1mr D2mr Dw D1mw D2mw
Event sort order:      Ir I1mr I2mr Dr D1mr D2mr Dw D1mw D2mw
Threshold:             99%
Chosen for annotation:
Auto-annotation:       on

--------------------------------------------------------------------------------
Ir         I1mr I2mr Dr         D1mr   D2mr  Dw        D1mw   D2mw
--------------------------------------------------------------------------------
27,742,716  276  275 10,955,517 21,905 3,987 4,474,773 19,280 19,098  PROGRAM TOTALS

--------------------------------------------------------------------------------
Ir        I1mr I2mr Dr        D1mr  D2mr  Dw        D1mw   D2mw    file:function
--------------------------------------------------------------------------------
8,821,482    5    5 2,242,702 1,621    73 1,794,230      0      0  getc.c:_IO_getc
5,222,023    4    4 2,276,334    16    12   875,959      1      1  concord.c:get_word
2,649,248    2    2 1,344,810 7,326 1,385         .      .      .  vg_main.c:strcmp
2,521,927    2    2   591,215     0     0   179,398      0      0  concord.c:hash
2,242,740    2    2 1,046,612   568    22   448,548      0      0  ctype.c:tolower
1,496,937    4    4   630,874 9,000 1,400   279,388      0      0  concord.c:insert
  897,991   51   51   897,831    95    30        62      1      1  ???:???
  598,068    1    1   299,034     0     0   149,517      0      0  ../sysdeps/generic/lockfile.c:__flockfile
  598,068    0    0   299,034     0     0   149,517      0      0  ../sysdeps/generic/lockfile.c:__funlockfile
  598,024    4    4   213,580    35    16   149,506      0      0  vg_clientmalloc.c:malloc
  446,587    1    1   215,973 2,167   430   129,948 14,057 13,957  concord.c:add_existing
  341,760    2    2   128,160     0     0   128,160      0      0  vg_clientmalloc.c:vg_trap_here_WRAPPER
  320,782    4    4   150,711   276     0    56,027     53     53  concord.c:init_hash_table
  298,998    1    1   106,785     0     0    64,071      1      1  concord.c:create
  149,518    0    0   149,516     0     0         1      0      0  ???:tolower@@GLIBC_2.0
  149,518    0    0   149,516     0     0         1      0      0  ???:fgetc@@GLIBC_2.0
   95,983    4    4    38,031     0     0    34,409  3,152  3,150  concord.c:new_word_node
   85,440    0    0    42,720     0     0    21,360      0      0  vg_clientmalloc.c:vg_bogus_epilogue]]></programlisting>


<para>First up is a summary of the annotation options:</para>
                    
<itemizedlist>

  <listitem>
    <para>I1 cache, D1 cache, L2 cache: cache configuration.  So
    you know the configuration with which these results were
    obtained.</para>
  </listitem>

  <listitem>
    <para>Command: the command line invocation of the program
      under examination.</para>
  </listitem>

  <listitem>
   <para>Events recorded: event abbreviations are:</para>
   <itemizedlist>
     <listitem>
       <para><computeroutput>Ir</computeroutput>: I cache reads
       (ie. instructions executed)</para>
     </listitem>
     <listitem>
       <para><computeroutput>I1mr</computeroutput>: I1 cache read
       misses</para>
     </listitem>
     <listitem>
       <para><computeroutput>I2mr</computeroutput>: L2 cache
       instruction read misses</para>
     </listitem>
     <listitem>
       <para><computeroutput>Dr</computeroutput>: D cache reads
       (ie. memory reads)</para>
     </listitem>
     <listitem>
       <para><computeroutput>D1mr</computeroutput>: D1 cache read
       misses</para>
     </listitem>
     <listitem>
       <para><computeroutput>D2mr</computeroutput>: L2 cache data
       read misses</para>
     </listitem>
     <listitem>
       <para><computeroutput>Dw</computeroutput>: D cache writes
       (ie. memory writes)</para>
     </listitem>
     <listitem>
       <para><computeroutput>D1mw</computeroutput>: D1 cache write
       misses</para>
     </listitem>
     <listitem>
       <para><computeroutput>D2mw</computeroutput>: L2 cache data
       write misses</para>
     </listitem>
     <listitem>
       <para><computeroutput>Bc</computeroutput>: Conditional branches
       executed</para>
     </listitem>
     <listitem>
       <para><computeroutput>Bcm</computeroutput>: Conditional branches
       mispredicted</para>
     </listitem>
     <listitem>
       <para><computeroutput>Bi</computeroutput>: Indirect branches
       executed</para>
     </listitem>
     <listitem>
       <para><computeroutput>Bim</computeroutput>: Conditional branches
       mispredicted</para>
     </listitem>
   </itemizedlist>

   <para>Note that D1 total accesses is given by
   <computeroutput>D1mr</computeroutput> +
   <computeroutput>D1mw</computeroutput>, and that L2 total
   accesses is given by <computeroutput>I2mr</computeroutput> +
   <computeroutput>D2mr</computeroutput> +
   <computeroutput>D2mw</computeroutput>.</para>
 </listitem>

 <listitem>
   <para>Events shown: the events shown, which is a subset of the events
   gathered.  This can be adjusted with the
   <computeroutput>--show</computeroutput> option.</para>
  </listitem>

  <listitem>
    <para>Event sort order: the sort order in which functions are
    shown.  For example, in this case the functions are sorted
    from highest <computeroutput>Ir</computeroutput> counts to
    lowest.  If two functions have identical
    <computeroutput>Ir</computeroutput> counts, they will then be
    sorted by <computeroutput>I1mr</computeroutput> counts, and
    so on.  This order can be adjusted with the
    <computeroutput>--sort</computeroutput> option.</para>

    <para>Note that this dictates the order the functions appear.
    It is <command>not</command> the order in which the columns
    appear; that is dictated by the "events shown" line (and can
    be changed with the <computeroutput>--show</computeroutput>
    option).</para>
  </listitem>

  <listitem>
    <para>Threshold: <computeroutput>cg_annotate</computeroutput>
    by default omits functions that cause very low counts
    to avoid drowning you in information.  In this case,
    cg_annotate shows summaries the functions that account for
    99% of the <computeroutput>Ir</computeroutput> counts;
    <computeroutput>Ir</computeroutput> is chosen as the
    threshold event since it is the primary sort event.  The
    threshold can be adjusted with the
    <computeroutput>--threshold</computeroutput>
    option.</para>
  </listitem>

  <listitem>
    <para>Chosen for annotation: names of files specified
    manually for annotation; in this case none.</para>
  </listitem>

  <listitem>
    <para>Auto-annotation: whether auto-annotation was requested
    via the <computeroutput>--auto=yes</computeroutput>
    option. In this case no.</para>
  </listitem>

</itemizedlist>

<para>Then follows summary statistics for the whole
program. These are similar to the summary provided when running
<computeroutput>valgrind --tool=cachegrind</computeroutput>.</para>
  
<para>Then follows function-by-function statistics. Each function
is identified by a
<computeroutput>file_name:function_name</computeroutput> pair. If
a column contains only a dot it means the function never performs
that event (eg. the third row shows that
<computeroutput>strcmp()</computeroutput> contains no
instructions that write to memory). The name
<computeroutput>???</computeroutput> is used if the the file name
and/or function name could not be determined from debugging
information. If most of the entries have the form
<computeroutput>???:???</computeroutput> the program probably
wasn't compiled with <computeroutput>-g</computeroutput>.  If any
code was invalidated (either due to self-modifying code or
unloading of shared objects) its counts are aggregated into a
single cost centre written as
<computeroutput>(discarded):(discarded)</computeroutput>.</para>

<para>It is worth noting that functions will come both from
the profiled program (eg. <filename>concord.c</filename>)
and from libraries (eg. <filename>getc.c</filename>)</para>

<para>There are two ways to annotate source files -- by choosing
them manually, or with the
<computeroutput>--auto=yes</computeroutput> option. To do it
manually, just specify the filenames as arguments to
<computeroutput>cg_annotate</computeroutput>. For example, the
output from running <filename>cg_annotate concord.c</filename>
for our example produces the same output as above followed by an
annotated version of <filename>concord.c</filename>, a section of
which looks like:</para>

<programlisting><![CDATA[
--------------------------------------------------------------------------------
-- User-annotated source: concord.c
--------------------------------------------------------------------------------
Ir        I1mr I2mr Dr      D1mr  D2mr  Dw      D1mw   D2mw

[snip]

        .    .    .       .     .     .       .      .      .  void init_hash_table(char *file_name, Word_Node *table[])
        3    1    1       .     .     .       1      0      0  {
        .    .    .       .     .     .       .      .      .      FILE *file_ptr;
        .    .    .       .     .     .       .      .      .      Word_Info *data;
        1    0    0       .     .     .       1      1      1      int line = 1, i;
        .    .    .       .     .     .       .      .      .
        5    0    0       .     .     .       3      0      0      data = (Word_Info *) create(sizeof(Word_Info));
        .    .    .       .     .     .       .      .      .
    4,991    0    0   1,995     0     0     998      0      0      for (i = 0; i < TABLE_SIZE; i++)
    3,988    1    1   1,994     0     0     997     53     52          table[i] = NULL;
        .    .    .       .     .     .       .      .      .
        .    .    .       .     .     .       .      .      .      /* Open file, check it. */
        6    0    0       1     0     0       4      0      0      file_ptr = fopen(file_name, "r");
        2    0    0       1     0     0       .      .      .      if (!(file_ptr)) {
        .    .    .       .     .     .       .      .      .          fprintf(stderr, "Couldn't open '%s'.\n", file_name);
        1    1    1       .     .     .       .      .      .          exit(EXIT_FAILURE);
        .    .    .       .     .     .       .      .      .      }
        .    .    .       .     .     .       .      .      .
  165,062    1    1  73,360     0     0  91,700      0      0      while ((line = get_word(data, line, file_ptr)) != EOF)
  146,712    0    0  73,356     0     0  73,356      0      0          insert(data->;word, data->line, table);
        .    .    .       .     .     .       .      .      .
        4    0    0       1     0     0       2      0      0      free(data);
        4    0    0       1     0     0       2      0      0      fclose(file_ptr);
        3    0    0       2     0     0       .      .      .  }]]></programlisting>

<para>(Although column widths are automatically minimised, a wide
terminal is clearly useful.)</para>
  
<para>Each source file is clearly marked
(<computeroutput>User-annotated source</computeroutput>) as
having been chosen manually for annotation.  If the file was
found in one of the directories specified with the
<computeroutput>-I / --include</computeroutput> option, the directory
and file are both given.</para>

<para>Each line is annotated with its event counts.  Events not
applicable for a line are represented by a dot.  This is useful
for distinguishing between an event which cannot happen, and one
which can but did not.</para>

<para>Sometimes only a small section of a source file is
executed.  To minimise uninteresting output, Cachegrind only shows
annotated lines and lines within a small distance of annotated
lines.  Gaps are marked with the line numbers so you know which
part of a file the shown code comes from, eg:</para>

<programlisting><![CDATA[
(figures and code for line 704)
-- line 704 ----------------------------------------
-- line 878 ----------------------------------------
(figures and code for line 878)]]></programlisting>

<para>The amount of context to show around annotated lines is
controlled by the <computeroutput>--context</computeroutput>
option.</para>

<para>To get automatic annotation, run
<computeroutput>cg_annotate --auto=yes</computeroutput>.
cg_annotate will automatically annotate every source file it can
find that is mentioned in the function-by-function summary.
Therefore, the files chosen for auto-annotation are affected by
the <computeroutput>--sort</computeroutput> and
<computeroutput>--threshold</computeroutput> options.  Each
source file is clearly marked (<computeroutput>Auto-annotated
source</computeroutput>) as being chosen automatically.  Any
files that could not be found are mentioned at the end of the
output, eg:</para>

<programlisting><![CDATA[
------------------------------------------------------------------
The following files chosen for auto-annotation could not be found:
------------------------------------------------------------------
  getc.c
  ctype.c
  ../sysdeps/generic/lockfile.c]]></programlisting>

<para>This is quite common for library files, since libraries are
usually compiled with debugging information, but the source files
are often not present on a system.  If a file is chosen for
annotation <command>both</command> manually and automatically, it
is marked as <computeroutput>User-annotated
source</computeroutput>. Use the <computeroutput>-I /
--include</computeroutput> option to tell Valgrind where to look
for source files if the filenames found from the debugging
information aren't specific enough.</para>

<para>Beware that cg_annotate can take some time to digest large
<computeroutput>cachegrind.out.pid</computeroutput> files,
e.g. 30 seconds or more.  Also beware that auto-annotation can
produce a lot of output if your program is large!</para>

</sect2>


<sect2 id="cg-manual.assembler" xreflabel="Annotating assembler programs">
<title>Annotating assembler programs</title>

<para>Valgrind can annotate assembler programs too, or annotate
the assembler generated for your C program.  Sometimes this is
useful for understanding what is really happening when an
interesting line of C code is translated into multiple
instructions.</para>

<para>To do this, you just need to assemble your
<computeroutput>.s</computeroutput> files with assembler-level
debug information.  gcc doesn't do this, but you can use the GNU
assembler with the <computeroutput>--gstabs</computeroutput>
option to generate object files with this information, eg:</para>

<programlisting><![CDATA[
as --gstabs foo.s]]></programlisting>

<para>You can then profile and annotate source files in the same
way as for C/C++ programs.</para>

</sect2>

</sect1>


<sect1 id="cg-manual.annopts" xreflabel="cg_annotate options">
<title><computeroutput>cg_annotate</computeroutput> options</title>

<itemizedlist>

  <listitem id="pid">
    <para><computeroutput>--pid</computeroutput></para>
    <para>Indicates that profile data should be read from 
    the file
    <computeroutput>cachegrind.out.pid</computeroutput>.
    read.  
    Note that you must specify either
    <computeroutput>--pid</computeroutput>
    or <computeroutput>--cachegrind-out-file=filename</computeroutput>
    exactly once.
    </para>
  </listitem>
    
  <listitem id="cachegrind-out-file">
    <para><computeroutput>--cachegrind-out-file=filename</computeroutput></para>
    <para>Indicates that profile data
    should be read from <computeroutput>filename</computeroutput>.
    Note that you must specify either
    <computeroutput>--pid</computeroutput>
    or <computeroutput>--cachegrind-out-file=filename</computeroutput>
    exactly once.
    </para>
  </listitem>
    
  <listitem>
    <para><computeroutput>-h, --help</computeroutput></para>
    <para><computeroutput>-v, --version</computeroutput></para>
    <para>Help and version, as usual.</para>
  </listitem>

  <listitem id="sort">
    <para><computeroutput>--sort=A,B,C</computeroutput> [default:
    order in
    <computeroutput>cachegrind.out.pid</computeroutput>]</para>
    <para>Specifies the events upon which the sorting of the
    function-by-function entries will be based.  Useful if you
    want to concentrate on eg. I cache misses
    (<computeroutput>--sort=I1mr,I2mr</computeroutput>), or D
    cache misses
    (<computeroutput>--sort=D1mr,D2mr</computeroutput>), or L2
    misses
    (<computeroutput>--sort=D2mr,I2mr</computeroutput>).</para>
  </listitem>

  <listitem id="show">
    <para><computeroutput>--show=A,B,C</computeroutput> [default:
    all, using order in
    <computeroutput>cachegrind.out.pid</computeroutput>]</para>
    <para>Specifies which events to show (and the column
    order). Default is to use all present in the
    <computeroutput>cachegrind.out.pid</computeroutput> file (and
    use the order in the file).</para>
  </listitem>

  <listitem id="threshold">
    <para><computeroutput>--threshold=X</computeroutput>
    [default: 99%]</para>
    <para>Sets the threshold for the function-by-function
    summary.  Functions are shown that account for more than X%
    of the primary sort event.  If auto-annotating, also affects
    which files are annotated.</para>
      
    <para>Note: thresholds can be set for more than one of the
    events by appending any events for the
    <computeroutput>--sort</computeroutput> option with a colon
    and a number (no spaces, though).  E.g. if you want to see
    the functions that cover 99% of L2 read misses and 99% of L2
    write misses, use this option:</para>
    <para><computeroutput>--sort=D2mr:99,D2mw:99</computeroutput></para>
  </listitem>

  <listitem id="auto">
    <para><computeroutput>--auto=no</computeroutput> [default]</para>
    <para><computeroutput>--auto=yes</computeroutput></para>
    <para>When enabled, automatically annotates every file that
    is mentioned in the function-by-function summary that can be
    found.  Also gives a list of those that couldn't be found.</para>
  </listitem>

  <listitem id="context">
    <para><computeroutput>--context=N</computeroutput> [default:
    8]</para>
    <para>Print N lines of context before and after each
    annotated line.  Avoids printing large sections of source
    files that were not executed.  Use a large number
    (eg. 10,000) to show all source lines.</para>
  </listitem>

  <listitem id="include">
    <para><computeroutput>-I&lt;dir&gt;,
      --include=&lt;dir&gt;</computeroutput> [default: empty
      string]</para>
    <para>Adds a directory to the list in which to search for
    files.  Multiple -I/--include options can be given to add
    multiple directories.</para>
  </listitem>

</itemizedlist>
  


<sect2>
<title>Warnings</title>

<para>There are a couple of situations in which
<computeroutput>cg_annotate</computeroutput> issues
warnings.</para>

<itemizedlist>
  <listitem>
    <para>If a source file is more recent than the
    <computeroutput>cachegrind.out.pid</computeroutput> file.
    This is because the information in
    <computeroutput>cachegrind.out.pid</computeroutput> is only
    recorded with line numbers, so if the line numbers change at
    all in the source (eg.  lines added, deleted, swapped), any
    annotations will be incorrect.</para>
  </listitem>
  <listitem>
    <para>If information is recorded about line numbers past the
    end of a file.  This can be caused by the above problem,
    ie. shortening the source file while using an old
    <computeroutput>cachegrind.out.pid</computeroutput> file.  If
    this happens, the figures for the bogus lines are printed
    anyway (clearly marked as bogus) in case they are
    important.</para>
  </listitem>
</itemizedlist>

</sect2>



<sect2>
<title>Things to watch out for</title>

<para>Some odd things that can occur during annotation:</para>

<itemizedlist>
  <listitem>
    <para>If annotating at the assembler level, you might see
    something like this:</para>
<programlisting><![CDATA[
      1    0    0  .    .    .  .    .    .          leal -12(%ebp),%eax
      1    0    0  .    .    .  1    0    0          movl %eax,84(%ebx)
      2    0    0  0    0    0  1    0    0          movl $1,-20(%ebp)
      .    .    .  .    .    .  .    .    .          .align 4,0x90
      1    0    0  .    .    .  .    .    .          movl $.LnrB,%eax
      1    0    0  .    .    .  1    0    0          movl %eax,-16(%ebp)]]></programlisting>

    <para>How can the third instruction be executed twice when
    the others are executed only once?  As it turns out, it
    isn't.  Here's a dump of the executable, using
    <computeroutput>objdump -d</computeroutput>:</para>
<programlisting><![CDATA[
      8048f25:       8d 45 f4                lea    0xfffffff4(%ebp),%eax
      8048f28:       89 43 54                mov    %eax,0x54(%ebx)
      8048f2b:       c7 45 ec 01 00 00 00    movl   $0x1,0xffffffec(%ebp)
      8048f32:       89 f6                   mov    %esi,%esi
      8048f34:       b8 08 8b 07 08          mov    $0x8078b08,%eax
      8048f39:       89 45 f0                mov    %eax,0xfffffff0(%ebp)]]></programlisting>

    <para>Notice the extra <computeroutput>mov
    %esi,%esi</computeroutput> instruction.  Where did this come
    from?  The GNU assembler inserted it to serve as the two
    bytes of padding needed to align the <computeroutput>movl
    $.LnrB,%eax</computeroutput> instruction on a four-byte
    boundary, but pretended it didn't exist when adding debug
    information.  Thus when Valgrind reads the debug info it
    thinks that the <computeroutput>movl
    $0x1,0xffffffec(%ebp)</computeroutput> instruction covers the
    address range 0x8048f2b--0x804833 by itself, and attributes
    the counts for the <computeroutput>mov
    %esi,%esi</computeroutput> to it.</para>
  </listitem>

  <listitem>
    <para>Inlined functions can cause strange results in the
    function-by-function summary.  If a function
    <computeroutput>inline_me()</computeroutput> is defined in
    <filename>foo.h</filename> and inlined in the functions
    <computeroutput>f1()</computeroutput>,
    <computeroutput>f2()</computeroutput> and
    <computeroutput>f3()</computeroutput> in
    <filename>bar.c</filename>, there will not be a
    <computeroutput>foo.h:inline_me()</computeroutput> function
    entry.  Instead, there will be separate function entries for
    each inlining site, ie.
    <computeroutput>foo.h:f1()</computeroutput>,
    <computeroutput>foo.h:f2()</computeroutput> and
    <computeroutput>foo.h:f3()</computeroutput>.  To find the
    total counts for
    <computeroutput>foo.h:inline_me()</computeroutput>, add up
    the counts from each entry.</para>

    <para>The reason for this is that although the debug info
    output by gcc indicates the switch from
    <filename>bar.c</filename> to <filename>foo.h</filename>, it
    doesn't indicate the name of the function in
    <filename>foo.h</filename>, so Valgrind keeps using the old
    one.</para>
  </listitem>

  <listitem>
    <para>Sometimes, the same filename might be represented with
    a relative name and with an absolute name in different parts
    of the debug info, eg:
    <filename>/home/user/proj/proj.h</filename> and
    <filename>../proj.h</filename>.  In this case, if you use
    auto-annotation, the file will be annotated twice with the
    counts split between the two.</para>
  </listitem>

  <listitem>
    <para>Files with more than 65,535 lines cause difficulties
    for the Stabs-format debug info reader.  This is because the line
    number in the <computeroutput>struct nlist</computeroutput>
    defined in <filename>a.out.h</filename> under Linux is only a
    16-bit value.  Valgrind can handle some files with more than
    65,535 lines correctly by making some guesses to identify
    line number overflows.  But some cases are beyond it, in
    which case you'll get a warning message explaining that
    annotations for the file might be incorrect.</para>
    
    <para>If you are using gcc 3.1 or later, this is most likely
    irrelevant, since gcc switched to using the more modern DWARF2 
    format by default at version 3.1.  DWARF2 does not have any such
    limitations on line numbers.</para>
  </listitem>

  <listitem>
    <para>If you compile some files with
    <computeroutput>-g</computeroutput> and some without, some
    events that take place in a file without debug info could be
    attributed to the last line of a file with debug info
    (whichever one gets placed before the non-debug-info file in
    the executable).</para>
  </listitem>

</itemizedlist>

<para>This list looks long, but these cases should be fairly
rare.</para>

</sect2>



<sect2>
<title>Accuracy</title>

<para>Valgrind's cache profiling has a number of
shortcomings:</para>

<itemizedlist>
  <listitem>
    <para>It doesn't account for kernel activity -- the effect of
    system calls on the cache contents is ignored.</para>
  </listitem>

  <listitem>
    <para>It doesn't account for other process activity.
    This is probably desirable when considering a single
    program.</para>
  </listitem>

  <listitem>
    <para>It doesn't account for virtual-to-physical address
    mappings.  Hence the simulation is not a true
    representation of what's happening in the
    cache.  Most caches are physically indexed, but Cachegrind
    simulates caches using virtual addresses.</para>
  </listitem>

  <listitem>
    <para>It doesn't account for cache misses not visible at the
    instruction level, eg. those arising from TLB misses, or
    speculative execution.</para>
  </listitem>

  <listitem>
    <para>Valgrind will schedule
    threads differently from how they would be when running natively.
    This could warp the results for threaded programs.</para>
  </listitem>

  <listitem>
    <para>The x86/amd64 instructions <computeroutput>bts</computeroutput>,
    <computeroutput>btr</computeroutput> and
    <computeroutput>btc</computeroutput> will incorrectly be
    counted as doing a data read if both the arguments are
    registers, eg:</para>
<programlisting><![CDATA[
    btsl %eax, %edx]]></programlisting>

    <para>This should only happen rarely.</para>
  </listitem>

  <listitem>
    <para>x86/amd64 FPU instructions with data sizes of 28 and 108 bytes
    (e.g.  <computeroutput>fsave</computeroutput>) are treated as
    though they only access 16 bytes.  These instructions seem to
    be rare so hopefully this won't affect accuracy much.</para>
  </listitem>

</itemizedlist>

<para>Another thing worth noting is that results are very sensitive.
Changing the size of the the executable being profiled, or the sizes
of any of the shared libraries it uses, or even the length of their
file names, can perturb the results.  Variations will be small, but
don't expect perfectly repeatable results if your program changes at
all.</para>

<para>More recent GNU/Linux distributions do address space
randomisation, in which identical runs of the same program have their
shared libraries loaded at different locations, as a security measure.
This also perturbs the results.</para>

<para>While these factors mean you shouldn't trust the results to
be super-accurate, hopefully they should be close enough to be
useful.</para>

</sect2>

</sect1>



<sect1 id="cg-manual.cg_merge" xreflabel="cg_merge">
<title>Merging profiles with <computeroutput>cg_merge</computeroutput></title>

<para>
<computeroutput>cg_merge</computeroutput> is a simple program which
reads multiple profile files, as created by cachegrind, merges them
together, and writes the results into another file in the same format.
You can then examine the merged results using
<computeroutput>cg_annotate
--cachegrind-out-file=outputfile</computeroutput>, as
described above.  The merging functionality might be useful if you
want to aggregate costs over multiple runs of the same program, or
from a single parallel run with multiple instances of the same
program.</para>

<para>
<computeroutput>cg_merge</computeroutput> is invoked as follows:
</para>

<programlisting><![CDATA[
cg_merge -o outputfile file1 file2 file3 ...]]></programlisting>

<para>
It reads and checks <computeroutput>file1</computeroutput>, then read
and checks <computeroutput>file2</computeroutput> and merges it into
the running totals, then the same with
<computeroutput>file3</computeroutput>, etc.  The final results are
written to <computeroutput>outputfile</computeroutput>, or to standard
out if no output file is specified.</para>

<para>
Costs are summed on a per-function, per-line and per-instruction
basis.  Because of this, the order in which the input files does not
matter, although you should take care to only mention each file once,
since any file mentioned twice will be added in twice.</para>

<para>
<computeroutput>cg_merge</computeroutput> does not attempt to check
that the input files come from runs of the same executable.  It will
happily merge together profile files from completely unrelated
programs.  It does however check that the
<computeroutput>Events:</computeroutput> lines of all the inputs are
identical, so as to ensure that the addition of costs makes sense.
For example, it would be nonsensical for it to add a number indicating
D1 read references to a number from a different file indicating L2
write misses.</para>

<para>
A number of other syntax and sanity checks are done whilst reading the
inputs.  <computeroutput>cg_merge</computeroutput> will stop and
attempt to print a helpful error message if any of the input files
fail these checks.</para>

</sect1>


<sect1>
<title>Acting on Cachegrind's information</title>
<para>
So, you've managed to profile your program with Cachegrind.  Now what?
What's the best way to actually act on the information it provides to speed
up your program?  Here are some rules of thumb that we have found to be
useful.</para>

<para>
First of all, the global hit/miss rate numbers are not that useful.  If you
have multiple programs or multiple runs of a program, comparing the numbers
might identify if any are outliers and worthy of closer investigation.
Otherwise, they're not enough to act on.</para>

<para>
The line-by-line source code annotations are much more useful.  In our
experience, the best place to start is by looking at the
<computeroutput>Ir</computeroutput> numbers.  They simply measure how many
instructions were executed for each line, and don't include any cache
information, but they can still be very useful for identifying
bottlenecks.</para>

<para>
After that, we have found that L2 misses are typically a much bigger source
of slow-downs than L1 misses.  So it's worth looking for any snippets of
code that cause a high proportion of the L2 misses.  If you find any, it's
still not always easy to work out how to improve things.  You need to have a
reasonable understanding of how caches work, the principles of locality, and
your program's data access patterns.  Improving things may require
redesigning a data structure, for example.</para>

<para>
In short, Cachegrind can tell you where some of the bottlenecks in your code
are, but it can't tell you how to fix them.  You have to work that out for
yourself.  But at least you have the information!
</para>

</sect1>

<sect1>
<title>Implementation details</title>
<para>
This section talks about details you don't need to know about in order to
use Cachegrind, but may be of interest to some people.
</para>

<sect2>
<title>How Cachegrind works</title>
<para>The best reference for understanding how Cachegrind works is chapter 3 of
"Dynamic Binary Analysis and Instrumentation", by Nicholas Nethercote.  It
is available on the <ulink url="&vg-pubs;">Valgrind publications
page</ulink>.</para>
</sect2>

<sect2>
<title>Cachegrind output file format</title>
<para>The file format is fairly straightforward, basically giving the
cost centre for every line, grouped by files and
functions.  Total counts (eg. total cache accesses, total L1
misses) are calculated when traversing this structure rather than
during execution, to save time; the cache simulation functions
are called so often that even one or two extra adds can make a
sizeable difference.</para>

<para>The file format:</para>
<programlisting><![CDATA[
file         ::= desc_line* cmd_line events_line data_line+ summary_line
desc_line    ::= "desc:" ws? non_nl_string
cmd_line     ::= "cmd:" ws? cmd
events_line  ::= "events:" ws? (event ws)+
data_line    ::= file_line | fn_line | count_line
file_line    ::= "fl=" filename
fn_line      ::= "fn=" fn_name
count_line   ::= line_num ws? (count ws)+
summary_line ::= "summary:" ws? (count ws)+
count        ::= num | "."]]></programlisting>

<para>Where:</para>
<itemizedlist>
  <listitem>
    <para><computeroutput>non_nl_string</computeroutput> is any
    string not containing a newline.</para>
  </listitem>
  <listitem>
    <para><computeroutput>cmd</computeroutput> is a string holding the
    command line of the profiled program.</para>
  </listitem>
  <listitem>
    <para><computeroutput>event</computeroutput> is a string containing
    no whitespace.</para>
  </listitem>
  <listitem>
    <para><computeroutput>filename</computeroutput> and
    <computeroutput>fn_name</computeroutput> are strings.</para>
  </listitem>
  <listitem>
    <para><computeroutput>num</computeroutput> and
    <computeroutput>line_num</computeroutput> are decimal
    numbers.</para>
  </listitem>
  <listitem>
    <para><computeroutput>ws</computeroutput> is whitespace.</para>
  </listitem>
</itemizedlist>

<para>The contents of the "desc:" lines are printed out at the top
of the summary.  This is a generic way of providing simulation
specific information, eg. for giving the cache configuration for
cache simulation.</para>

<para>More than one line of info can be presented for each file/fn/line number.
In such cases, the counts for the named events will be accumulated.</para>

<para>Counts can be "." to represent zero.  This makes the files easier for
humans to read.</para>

<para>The number of counts in each
<computeroutput>line</computeroutput> and the
<computeroutput>summary_line</computeroutput> should not exceed
the number of events in the
<computeroutput>event_line</computeroutput>.  If the number in
each <computeroutput>line</computeroutput> is less, cg_annotate
treats those missing as though they were a "." entry.  This saves space.
</para>

<para>A <computeroutput>file_line</computeroutput> changes the
current file name.  A <computeroutput>fn_line</computeroutput>
changes the current function name.  A
<computeroutput>count_line</computeroutput> contains counts that
pertain to the current filename/fn_name.  A "fn="
<computeroutput>file_line</computeroutput> and a
<computeroutput>fn_line</computeroutput> must appear before any
<computeroutput>count_line</computeroutput>s to give the context
of the first <computeroutput>count_line</computeroutput>s.</para>

<para>Each <computeroutput>file_line</computeroutput> will normally be
immediately followed by a <computeroutput>fn_line</computeroutput>.  But it
doesn't have to be.</para>


</sect2>

</sect1>
</chapter>