Beefy Boxes and Bandwidth Generously Provided by pair Networks
Just another Perl shrine

Re^2: Given When Syntax

by Laurent_R (Canon)
on Mar 16, 2014 at 22:58 UTC ( #1078555=note: print w/replies, xml ) Need Help??

in reply to Re: Given When Syntax
in thread Given When Syntax

Hashes with dispatch tables (slow sub calls) and if-elsif-chains (slow linear testing) shouldn't be as fast as this approach.

Dear Rolf, I know that you are interested in functional programming possibilities in Perl, so please stop taking for granted what HOP opponents are hammering constantly. Subroutine calls of course add some time penalty, but not as much as many people think or say. In fact, as soon as the routine is really doing something, the sub call penalty becomes almost anecdotal or even almost negligible compared to the rest or the processing. In the benchmark below, the dispatch table ranks second best, immediately after direct array search, although it does really nothing more than returning a value.
use strict; use warnings; use Benchmark qw/cmpthese/; my @translation = qw / Zero One Two Three/; my %trans = (1 => "One", 2 => "Two", 3 => "Three"); my @dispatch = ( sub {return "Zero"}, sub {return "One"}, sub {return +"Two"}, sub {return "Three"} ); sub test1 { my $var2 = shift; return ("One") if ($var2 =~ /^1/ ); return ("Two") if ($var2 =~ /^2/ ); return ("Three") if ($var2 =~ /^3/ ); return undef; } sub test2 { my $var2 = substr shift, 0, 1; return ("One") if ($var2 == 1 ); return ("Two") if ($var2 == 2 ); return ("Three") if ($var2 == 3 ); return undef; } sub test3 { return $translation[(substr shift, 0, 1)]; } sub test4 { my $var = shift; return $trans{$1} if $var =~ /^(\d)\./ ; } sub test5 { my $var = substr shift, 0, 1; eval { goto "_$var" } or return "Other"; _1: return "One" ; _2: return "Two" ; _3: return "Three"; } sub test6 { return $dispatch[(substr shift, 0, 1)]->(); } cmpthese( -1, { _linear_1 => q {test1("2.01.000")}, _linear_2 => q {test2("2.01.000")}, _array => q {test3("2.01.000")}, _hash_regex => q {test4("2.01.000")}, _goto => q {test5("2.01.000")}, _dispatch => q {test6("2.01.000")}, } )
And the results:
$ perl Rate _hash_regex _goto linear_1 linear_2 _dispatc +h _array _hash_regex 831494/s -- -29% -46% -52% -54 +% -70% _goto 1173439/s 41% -- -24% -32% -36 +% -57% _linear_1 1538868/s 85% 31% -- -10% -1 +6% -44% _linear_2 1714704/s 106% 46% 11% -- - +6% -37% _dispatch 1827212/s 120% 56% 19% 7% - +- -33% _array 2735932/s 229% 133% 78% 60% 50 +% --
The dispatch table approach, with its sub call penalty, is 33% slower than the direct array access, but still quicker than any of the other tested approaches.

Replies are listed 'Best First'.
Re^3: Given When Syntax
by LanX (Sage) on Mar 17, 2014 at 00:16 UTC
    Dear Laurent,

    Thanks for providing timings, I appreciate this!

    I said "should" because I was too busy to measure it myself.

    But maybe someone has the time to do these benchmarks properly? :)

    Cheers Rolf

    ( addicted to the Perl Programming Language)

      Hi Rolf, there may be some errors in the benchmark, but then you should say what is wrong, rather than just implying it was not done properly.
        I'll post benchmarks after GPW.

        Sorry but ATM you are mangling different things and effectively comparing apples with plastic food.

        BTW as it seems you are also "proving" that gotos to labels with code execution are faster then simple hash look ups. (which would support using goto)


        I'm not sure but I somehow remember benchmarks indicating that goto was not very efficiently implemented.

        By having a linear complexity, it could be in small cases faster than the constant overhead of hash lookups.

        Cheers Rolf

        ( addicted to the Perl Programming Language)

Log In?

What's my password?
Create A New User
Domain Nodelet?
Node Status?
node history
Node Type: note [id://1078555]
and the web crawler heard nothing...

How do I use this? | Other CB clients
Other Users?
Others chilling in the Monastery: (5)
As of 2022-08-10 12:07 GMT
Find Nodes?
    Voting Booth?

    No recent polls found