I've run some of the shootout tests with disabled optimizations and found some surprising results: Pike: pike PikeNP: pike -dp (without peephole optimization) PikeNT: pike -dt (without tail recursion) PikeNTP: pike -dp -dt
Pike PikeNP PikeNT PikeNTP 1: 42.81 32.83 37.49 34.75 (ackermann) 2: 34.92 37.05 37.64 37.80 (ary3) 3: 148.92 304.90 118.21 128.94 (fibo)
Is the test flawed or is there something fishy in the optimization. The compilation overhead between Pike with and without optimizations is at most 0.13s.
Hm, it runs way faster without peephole and tail recursing in some applications? Gra?
Probably code where the time spent optimizing is on the order of the time saved by the optimizations.
/ Henrik Grubbström (Lysator)
Previous text:
2002-11-04 17:57: Subject: Re: Optimizations
Hm, it runs way faster without peephole and tail recursing in some applications? Gra?
/ Brevbäraren
Nilsson's statement "The compilation overhead between Pike with and without optimizations is at most 0.13s" seems to suggest otherwise. But perhaps I misinterpreted it...
/ Leif Stensson, Lysator
Previous text:
2002-11-04 18:21: Subject: Re: Optimizations
Probably code where the time spent optimizing is on the order of the time saved by the optimizations.
/ Henrik Grubbström (Lysator)
No, as I said the difference in compilation time is not an issue. I first saw the strangeness in the same tests with more modest values and decided to rerun them greater values to see if the time difference scaled up as well. It did as you can see. Anyway, given that the tests are about 10-20 lines of code it should not take ten seconds to optimize on a 1800 MHz computer.
/ Martin Nilsson (Fake Build Master)
Previous text:
2002-11-04 18:21: Subject: Re: Optimizations
Probably code where the time spent optimizing is on the order of the time saved by the optimizations.
/ Henrik Grubbström (Lysator)
True.
/ Henrik Grubbström (Lysator)
Previous text:
2002-11-04 18:43: Subject: Re: Optimizations
No, as I said the difference in compilation time is not an issue. I first saw the strangeness in the same tests with more modest values and decided to rerun them greater values to see if the time difference scaled up as well. It did as you can see. Anyway, given that the tests are about 10-20 lines of code it should not take ten seconds to optimize on a 1800 MHz computer.
/ Martin Nilsson (Fake Build Master)
pike-devel@lists.lysator.liu.se
-
Henrik Grubbstr�m (Lysator) ＠ Pike (-) developers forum -
Leif Stensson, Lysator ＠ Pike developers forum
-
Martin Nilsson (Fake Build Master) ＠ Pike (-) developers forum
-
Mirar