Rev |
Age |
Author |
Path |
Log message |
Diff |
1222 |
3037d 08h |
werner |
/ |
public SVN |
|
1221 |
3037d 08h |
werner |
/branches/ |
public SVN |
|
1220 |
3037d 08h |
werner |
/branches/version1.0/ |
Version 1.0 |
|
1218 |
3037d 10h |
werner |
/ |
clean up repository - make 1.0 trunk again! |
|
1217 |
3037d 10h |
werner |
/ |
clean up repository - make 1.0 trunk again! |
|
1164 |
3253d 08h |
werner |
/branches/iland_regen/src/ |
found unexpected behavior of irandom(): obviously, it return *never* the upper bound.
eg. irandom(0,3) -> values 0,1, or 2, but never 3.
changed calls to irandom() accordingly.
added also a better random selection of species in the establishment routine |
|
1104 |
3325d 14h |
werner |
/branches/iland_regen/ |
Branch dedicated to playing around with establishment/regeneration.
Goals are:
(i) speed up
(ii) introduction of new features for Yellowstone |
|
1002 |
3675d 04h |
werner |
/branches/port_qt5_64bit/src/ |
new optimized establishment routine.
not finished yet, but looks promising in my first tests. |
|
779 |
4309d 09h |
werner |
/branches/port_qt5_64bit/ |
|
|
758 |
4535d 08h |
werner |
/trunk/src/ |
tests with the random generator; (see also the blog post about random numbers). Now the generations of the generator are deactivated (i.e. set to 0). The ignition calculation of the fire module is now based on the resource unit (instead of the single fire pixel). |
|
757 |
4537d 03h |
werner |
/trunk/src/ |
after performing random number tests; still trying to understand the fire ignition patterns. |
|
707 |
4840d 11h |
werner |
/trunk/src/ |
rework of the random numbers in iland.
Now, many (200000) random numbers are precalculated using a given algorithm (there are three or four choices), then the drandom() etc functions read from this array. After some rotations, the numbers are renewed. The point is: this is now reentrant, so the per-thread storage of MTRand objects is obsolete. |
|
706 |
4841d 04h |
werner |
/trunk/src/ |
some performance tests with random number generators. Results so far are not too indicative. |
|
705 |
4841d 06h |
werner |
/trunk/src/ |
series of performance tests / modifications.
Using "valgrind" on the linux virtual machine, some performance
measurements were made.
Some candidates for optimization were tested. (1) the applyLIP()
function (about 30% of total run time in runYear()): here several
inlining options were tested; testing the value of the stamp before
using it (in order to omit the other steps) prooved to be slower than
unsing all stamp values.
(2) the random number generator; here the random number generation is
about 5% of the total run time; the commit contains two/three other
variants. |
|