Chapter 5. Profiling

Table of Contents
5.1. Cost centres and cost-centre stacks
5.2. Compiler options for profiling
5.3. Time and allocation profiling
5.4. Profiling memory usage
5.5. Graphical time/allocation profile
5.6. hp2ps––heap profile to PostScript
5.7. Using “ticky-ticky” profiling (for implementors)

Glasgow Haskell comes with a time and space profiling system. Its purpose is to help you improve your understanding of your program's execution behaviour, so you can improve it.

Any comments, suggestions and/or improvements you have are welcome. Recommended “profiling tricks” would be especially cool!

Profiling a program is a three-step process:

  1. Re-compile your program for profiling with the -prof option, and probably one of the -auto or -auto-all options. These options are described in more detail in Section 5.2

  2. Run your program with one of the profiling options, eg. +RTS -p -RTS. This generates a file of profiling information.

  3. Examine the generated profiling information, using one of GHC's profiling tools. The tool to use will depend on the kind of profiling information generated.

5.1. Cost centres and cost-centre stacks

GHC's profiling system assigns costs to cost centres. A cost is simply the time or space required to evaluate an expression. Cost centres are program annotations around expressions; all costs incurred by the annotated expression are assigned to the enclosing cost centre. Furthermore, GHC will remember the stack of enclosing cost centres for any given expression at run-time and generate a call-graph of cost attributions.

Let's take a look at an example:

main = print (nfib 25)
nfib n = if n < 2 then 1 else nfib (n-1) + nfib (n-2)

Compile and run this program as follows:

$ ghc -prof -auto-all -o Main Main.hs
$ ./Main +RTS -p
121393
$

When a GHC-compiled program is run with the -p RTS option, it generates a file called <prog>.prof. In this case, the file will contain something like this:

          Fri May 12 14:06 2000 Time and Allocation Profiling Report  (Final)

           Main +RTS -p -RTS

        total time  =        0.14 secs   (7 ticks @ 20 ms)
        total alloc =   8,741,204 bytes  (excludes profiling overheads)

COST CENTRE          MODULE     %time %alloc

nfib                 Main       100.0  100.0


                                              individual     inherited
COST CENTRE              MODULE      entries %time %alloc   %time %alloc

MAIN                     MAIN             0    0.0   0.0    100.0 100.0
 main                    Main             0    0.0   0.0      0.0   0.0
 CAF                     PrelHandle       3    0.0   0.0      0.0   0.0
 CAF                     PrelAddr         1    0.0   0.0      0.0   0.0
 CAF                     Main             6    0.0   0.0    100.0 100.0
  main                   Main             1    0.0   0.0    100.0 100.0
   nfib                  Main        242785  100.0 100.0    100.0 100.0

The first part of the file gives the program name and options, and the total time and total memory allocation measured during the run of the program (note that the total memory allocation figure isn't the same as the amount of live memory needed by the program at any one time; the latter can be determined using heap profiling, which we will describe shortly).

The second part of the file is a break-down by cost centre of the most costly functions in the program. In this case, there was only one significant function in the program, namely nfib, and it was responsible for 100% of both the time and allocation costs of the program.

The third and final section of the file gives a profile break-down by cost-centre stack. This is roughly a call-graph profile of the program. In the example above, it is clear that the costly call to nfib came from main.

The time and allocation incurred by a given part of the program is displayed in two ways: “individual”, which are the costs incurred by the code covered by this cost centre stack alone, and “inherited”, which includes the costs incurred by all the children of this node.

The usefulness of cost-centre stacks is better demonstrated by modifying the example slightly:

main = print (f 25 + g 25)
f n  = nfib n
g n  = nfib (n `div` 2)
nfib n = if n < 2 then 1 else nfib (n-1) + nfib (n-2)

Compile and run this program as before, and take a look at the new profiling results:

COST CENTRE              MODULE         scc  %time %alloc   %time %alloc

MAIN                     MAIN             0    0.0   0.0    100.0 100.0
 main                    Main             0    0.0   0.0      0.0   0.0
 CAF                     PrelHandle       3    0.0   0.0      0.0   0.0
 CAF                     PrelAddr         1    0.0   0.0      0.0   0.0
 CAF                     Main             9    0.0   0.0    100.0 100.0
  main                   Main             1    0.0   0.0    100.0 100.0
   g                     Main             1    0.0   0.0      0.0   0.2
    nfib                 Main           465    0.0   0.2      0.0   0.2
   f                     Main             1    0.0   0.0    100.0  99.8
    nfib                 Main        242785  100.0  99.8    100.0  99.8

Now although we had two calls to nfib in the program, it is immediately clear that it was the call from f which took all the time.

The actual meaning of the various columns in the output is:

entries

The number of times this particular point in the call graph was entered.

individual %time

The percentage of the total run time of the program spent at this point in the call graph.

individual %alloc

The percentage of the total memory allocations (excluding profiling overheads) of the program made by this call.

inherited %time

The percentage of the total run time of the program spent below this point in the call graph.

inherited %alloc

The percentage of the total memory allocations (excluding profiling overheads) of the program made by this call and all of its sub-calls.

In addition you can use the -P RTS option to get the following additional information:

ticks

The raw number of time “ticks” which were attributed to this cost-centre; from this, we get the %time figure mentioned above.

bytes

Number of bytes allocated in the heap while in this cost-centre; again, this is the raw number from which we get the %alloc figure mentioned above.

What about recursive functions, and mutually recursive groups of functions? Where are the costs attributed? Well, although GHC does keep information about which groups of functions called each other recursively, this information isn't displayed in the basic time and allocation profile, instead the call-graph is flattened into a tree. The XML profiling tool (described in Section 5.5) will be able to display real loops in the call-graph.

5.1.1. Inserting cost centres by hand

Cost centres are just program annotations. When you say -auto-all to the compiler, it automatically inserts a cost centre annotation around every top-level function in your program, but you are entirely free to add the cost centre annotations yourself.

The syntax of a cost centre annotation is

     {-# SCC "name" #-} <expression>

where "name" is an aribrary string, that will become the name of your cost centre as it appears in the profiling output, and <expression> is any Haskell expression. An SCC annotation extends as far to the right as possible when parsing.

5.1.2. Rules for attributing costs

The cost of evaluating any expression in your program is attributed to a cost-centre stack using the following rules:

  • If the expression is part of the one-off costs of evaluating the enclosing top-level definition, then costs are attributed to the stack of lexically enclosing SCC annotations on top of the special CAF cost-centre.

  • Otherwise, costs are attributed to the stack of lexically-enclosing SCC annotations, appended to the cost-centre stack in effect at the call site of the current top-level definition[1]. Notice that this is a recursive definition.

What do we mean by one-off costs? Well, Haskell is a lazy language, and certain expressions are only ever evaluated once. For example, if we write:

x = nfib 25

then x will only be evaluated once (if at all), and subsequent demands for x will immediately get to see the cached result. The definition x is called a CAF (Constant Applicative Form), because it has no arguments.

For the purposes of profiling, we say that the expression nfib 25 belongs to the one-off costs of evaluating x.

Since one-off costs aren't strictly speaking part of the call-graph of the program, they are attributed to a special top-level cost centre, CAF. There may be one CAF cost centre for each module (the default), or one for each top-level definition with any one-off costs (this behaviour can be selected by giving GHC the -caf-all flag).

If you think you have a weird profile, or the call-graph doesn't look like you expect it to, feel free to send it (and your program) to us at .

Notes

[1]

The call-site is just the place in the source code which mentions the particular function or variable.