Optimize Perl

By Martin C. Brown
2005-04-11

Using Bytecode and the Compiler Back Ends

There are three ways to use the compiler: bytecode production, full compilation, or simply as a debugging/optimizing tool. The first two methods rely on converting your original Perl source into its compiled bytecode form and storing this precompiled version for execution. This is best used through the perlcc command. These two modes follow the same basic model but produce the final result differently. In bytecode mode, the resulting compiled bytecode is written out to another Perl script. The script consists of the ByteLoader preamble, with the compiled code stored as a byte string. To create this bytecode version, use the -B option to the perlcc command. For example:

$ perlcc -B script.pl

This will create a file, a.out. The output, however, is not very Web friendly. The resulting file can be executed with any Perl executable on any platform (Perl bytecode is platform independent):

$ perl a.out

What this does is save Perl from having to compile the script from its source code into the bytecode each time. Instead, it just runs the bytecode that was generated. This is similar to the process behind Java compilation and is in fact that same one-step away from being a truly compiled form of the language. On short scripts, especially those that use a number of external modules, you probably won't notice a huge speed increase. On larger scripts that "stand alone" without a lot of external module use, you should see a noticeable improvement.

The full compilation mode is almost identical, except that instead of producing a Perl script with the compiled bytecode embedded in it, perlcc produces a version embedded into C source that is then compiled into a full-blown, standalone executable. This is not cross-platform compatible, but it does allow you to distribute an executable version of a Perl script without giving out the source. Note, however, that this doesn't convert the Perl into C, it just embeds Perl bytecode into a C-based application. This is actually the default mode of perlcc, so a simple: $ perlcc script.pl will create, and compile, a standalone application called a.out.

One of the lesser-known solutions for both debugging and optimizing your code is to use the Perl compiler with one of the many "back ends."

The back ends are actually what drive the perlcc command, and it's possible to use a back-end module directly to create a C source file that you can examine. The Perl compiler works by taking the generated bytecode and then outputting the results in a variety of different ways. Because you're looking at the opcodes generated during the compilation stage, you get to see the code after Perl's own internal optimizations have been applied. Providing you know the Perl opcodes, you can begin to identify where the potential bottlenecks might be. From a debugging perspective, go with back ends such as Terse (which is itself a wrapper on Concise) and Showlex. You can see in Listing 10 what the original Listing 1 looks like through the Terse back end.

Listing 10. Using Terse to study bytecode



LISTOP (0x306230) leave [1]

    OP (0x305f60) enter

    COP (0x3062d0) nextstate

    BINOP (0x306210) sassign

        SVOP (0x301ab0) const [7] PV (0x1809f9c) "abcdefghijklmnopqrstuvwxyz"

        OP (0x305c30) padsv [1]

    COP (0x305c70) nextstate

    BINOP (0x305c50) sassign

        SVOP (0x306330) const [8] PV (0x180be60) ""

        OP (0x306310) padsv [2]

    COP (0x305f20) nextstate

    BINOP (0x305f00) leaveloop

        LOOP (0x305d10) enteriter [3]

            OP (0x305cf0) null [3]

            UNOP (0x305cd0) null [141]

                OP (0x305e80) pushmark

                SVOP (0x3065d0) const [9] IV (0x180be30) 1

                SVOP (0x3065f0) const [10] IV (0x1801240) 999999

        UNOP (0x305ee0) null

            LOGOP (0x305ec0) and

                OP (0x305d50) iter

                LISTOP (0x305e60) lineseq

                    COP (0x305e10) nextstate

                    BINOP (0x305df0) concat [6]

                        OP (0x305d70) padsv [2]

                        OP (0x305dd0) padsv [1]

                    OP (0x305ea0) unstack

concat1.pl syntax OK

Tutorial Pages:
» Squeeze the Most From Your Code
» Sloppy Programming, Sloppy Performance
» Approaching Optimization
» Use References
» String Handling
» Loops
» Sorts
» Using Short Circuit Logic
» Use AutoLoader
» Using Bytecode and the Compiler Back Ends
» Other Tools
» Putting it All Together
» Resources

First published by IBM DeveloperWorks

Related Tutorials:
» Random subroutines in Perl
» Log Script Use
» Creating Perl Modules for Web Sites
» Bit Vector, Using Perl Vec
» Build a Perl/CGI Voting System
» Perl Range Operator