Re^2: Fastest way to turn coordinate list into array

Those results are different than what I would expect.

On perl, v5.10.0 built for i486-linux-gnu-thread-multi, I get what I'd expect:

             Rate         OP while_shift     regex      ike2 chromatic
+      ike1
OP          245/s         --        -17%      -28%      -38%      -48%
+      -50%
while_shift 297/s        21%          --      -13%      -26%      -36%
+      -39%
regex       340/s        39%         15%        --      -15%      -27%
+      -31%
ike2        398/s        63%         34%       17%        --      -15%
+      -19%
chromatic   467/s        91%         57%       37%       17%        --
+       -5%
ike1        490/s       100%         65%       44%       23%        5%
+        --
[download]

On perl 5, version 12, subversion 0 (v5.12.0-RC1) built for i686-linux, I the same results.

             Rate         OP while_shift     regex      ike2 chromatic
+      ike1
OP          117/s         --        -22%      -28%      -42%      -52%
+      -54%
while_shift 151/s        29%          --       -7%      -25%      -38%
+      -41%
regex       163/s        39%          8%        --      -19%      -33%
+      -36%
ike2        202/s        72%         33%       24%        --      -18%
+      -21%
chromatic   244/s       109%         62%       50%       21%        --
+       -4%
ike1        254/s       117%         68%       56%       26%        4%
+        --
[download]

The two extra tests are:

ike1 => sub {
   my @poly = map [ (split /,/, $_)[0, 1] ], split ' ', $coords;
},
ike2 => sub {
   my @poly = map [ /([^,]*),([^,]*)/ ], split ' ', $coords;
},
[download]

Comment on Re^2: Fastest way to turn coordinate list into array Select or Download Code

Replies are listed 'Best First'.
Re^3: Fastest way to turn coordinate list into array by brancusi (Novice) on Apr 09, 2010 at 19:01 UTC
Brilliant, thank you all for your speedy replies. FWIW the regex was simply hanging apache with real data at my end (some lists have 300K points) but the chromatic and ikegami solutions are definitely helping. It has highlighted that actually the real problem is algorithmic rather than performance of this section of code as the speed up, whilst beneficial, isn't the main problem. I can think of a way of whittling down the data to test in another part of the code so I'll try that out next. Again, many thanks - and ikegami's code will remain in as the winner so far. Roger	[reply]
Re^3: Fastest way to turn coordinate list into array by kennethk (Abbot) on Apr 09, 2010 at 18:59 UTC
As I've pointed out in the update above, I reran and maintained ordering. I thought perhaps it was an 5.8/10 issue, but ran on my Linux box and got the following: Rate while_shift OP regex ike2 double_split chromatic + jwkrahn ike1 while_shift 15.0/s -- -11% -30% -40% -46% -51% + -54% -54% OP 16.8/s 12% -- -22% -33% -40% -45% + -48% -48% regex 21.5/s 43% 28% -- -14% -23% -29% + -34% -34% ike2 25.0/s 67% 49% 16% -- -10% -18% + -23% -23% double_split 27.9/s 86% 66% 30% 11% -- -8% + -14% -14% chromatic 30.4/s 102% 81% 41% 22% 9% -- + -6% -6% jwkrahn 32.4/s 116% 93% 50% 29% 16% 6% + -- -0% ike1 32.5/s 116% 93% 51% 30% 17% 7% + 0% -- This is perl, v5.8.8 built for x86_64-linux-gnu-thread-multi [download] Anyone have an idea why my regex engine is so fast for ActiveState Perl 5.8.9?	[reply] [d/l]