• lzzll (58/58) Apr 05 2012 Hey, I wrote some code to benchmark dict list and string in D,
• Andrej Mitrovic (34/36) Apr 05 2012 Tables to the rescue:
• Andrei Alexandrescu (31/37) Apr 05 2012 [snip]

"lzzll" <ownrepos gmail.com> writes:

Hey, I wrote some code to benchmark dict list and string in D, 
python and lua.
D is a great language, but some library look comparatively slow.
Some where may be wrong, I start D just today.

D test:
code: http://pastebin.com/qF7N4wLx
dict set: 2.251s
dict get: 1.957s
dict cleat: 0.00s
list set: 0.085s
list get: 0.007s
list each: 0.014s
str find: 1.254s
str replace: 1.870s

Python test:
code: http://pastebin.com/m3qMjPcM
dict set: 1.05s
dict get: 0.75s
dict cleat: 0.11s
list set: 0.02s
list get: 0.09s
list each: 0.09s
str find: 0.94s
str replace: 0.67s

Lua test:
code: http://pastebin.com/TFTnsWzX
dict set: 2.11s
dict get: 0.87s
dict cleat: 0.19s
list set: 0.33s
list get: 0.03s
list each: 0.09s
str find: 0.94s
str replace: 1.10s

Compare (> mean faster than):
dict set: python > lua > D
dict get: python > lua > D
dict clear: D > lua > python
list set: python > D > lua
list get: D > lua > python
list each: D > lua = python
str find: python = lua > D
str replace: python > lua > D

C algorithm (with -O3):
map tree set: 0.70s
map tree get: 0.39s
map tree clear: 0.23
chain set: 0.09s
chain get: too large ...
chain each: 0.05s
variable length list set: 0.025s
variable length list get: 0.01s
variable length list each: 0.01s
quick search find: 0.29s
remalloc replace: 0.40s
chain replace: 0.40s

I think the one of problems is to!string(int) too slow, snprintf 
will better.

Andrej Mitrovic <andrej.mitrovich gmail.com> writes:

On 4/5/12, lzzll <ownrepos gmail.com> wrote:
 I think the one of problems is to!string(int) too slow, snprintf
 will better.

Tables to the rescue:

__gshared string[int] table;
shared static this()
{
    foreach (i; 0 .. MAX_RANGE)
        table[i] = to!string(i);
}

Then use table[i] instead of to!string(i). Here's a comparison:

to!string:
---- test_dict -----
test dict set (1000000)...      1.281s
test dict get (1000000)...      1.047s
test dict clear (1000000)...    0.000s
---- test_list -----
test list set (1000000)...      0.125s
test list get (1000000)...      0.000s
test list each (1000000)...     0.000s
---- test_str -----
test str find (1000000)...      0.703s
test str replace (1000000)...   1.016s

table (http://pastebin.com/43Z4EwMj)
test dict set (1000000)...      0.891s
test dict get (1000000)...      0.109s
test dict clear (1000000)...    0.000s
---- test_list -----
test list set (1000000)...      0.078s
test list get (1000000)...      0.000s
test list each (1000000)...     0.000s
---- test_str -----
test str find (1000000)...      0.125s
test str replace (1000000)...   1.203s

But it seems the str replace test got slower even though all the other
tests got faster.

Andrei Alexandrescu <SeeWebsiteForEmail erdani.org> writes:

On 4/5/12 2:00 PM, lzzll wrote:
 Hey, I wrote some code to benchmark dict list and string in D, python
 and lua.
 D is a great language, but some library look comparatively slow.
 Some where may be wrong, I start D just today.

[snip]

Thanks for the benchmark, good comparisons are always helpful.

 I think the one of problems is to!string(int) too slow, snprintf
 will better.

You're right. In fact I took a minute to adapt some C++ code I wrote for 
work into a faster to!string routine. On my machine I get:

$ ./test.py
---- test_dict -----
test dict set (1000000)...	0.64s
test dict get (1000000)...	0.40s
test dict clear (1000000)...	0.09s

---- test_list -----
test list set (1000000)...	0.02s
test list get (1000000)...	0.08s
test list each (1000000)...	0.06s

---- test_str -----
test str find (1000000)...	0.47s
test str replace (1000000)...	0.48s

$ ./test
---- test_dict -----
test dict set (1000000)...	0.729s
test dict get (1000000)...	0.261s
test dict clear (1000000)...	0.000s
---- test_list -----
test list set (1000000)...	0.073s
test list get (1000000)...	0.001s
test list each (1000000)...	0.001s
---- test_str -----
test str find (1000000)...	0.593s
test str replace (1000000)...	0.804s

Not too shoddy! I'll work on adding the better conversion routine to Phobos.


Andrei