04-Mar-2013 16:17, Namespace пишет:
With reserving of 60% of the file size in memory I get a big speed boost.
------------------------
Total time (ms): 17544.6
Repetitions : 200
Sample mode : 87 (121 ocurrences)
Median time : 87.3835
Avg time : 87.7232
Std dev. : 1.77905
Minimum : 84.238
Maximum : 101.919
95% conf.int. : [84.2363, 91.21] e = 3.48687
99% conf.int. : [83.1407, 92.3057] e = 4.58252
EstimatedAvg95%: [87.4766, 87.9697] e = 0.246559
EstimatedAvg99%: [87.3991, 88.0472] e = 0.324033
But I'm still away from:
Avg time : 69.3088
http://forum.dlang.org/thread/dpdgcycrgfspcxenz...@forum.dlang.org?page=8#post-abncwoqtpnaoohihkliw:40forum.dlang.org
Keep in mind the CPU you are testing on. What's you chip?
In any case Dscanner has been improved to take far far less time then
60ms. Brain listed graphs showing under 20 ms on his CPU. for all and
<10 for every module in std except std.datetime.
I've tried also 10%, 40% and 80% but 60% seems optimal.
And I think I reached my performance limit. I could implement the
LazyForwardRange and, if anyone can confirm the statement of char <->
dchar decoding and that the usage of ubyte would be (crucial) better, I
will try this too.
--
Dmitry Olshansky
