Rect functions improvements

[itzpr3d4t0r]

trafficstars

Info

One of #3337. This PR speeds up some rect functions that recieve a list of rects as an argument. Improvements consist in the use of METH_O, as those functions just take 1 argument and PySequence_Fast, to replace the normal slower Sequence we are using now . Improvements vary and can go from 25% up to 105% depending on the situation. THIS PR IS A DRAFT AND WILL ALWAYS BE. ONCE #3230 GETS MERGED I'LL OPEN A NEW PR THAT WILL IMPLEMENT THESE CHANGES IN THE NEW FILES/DIRS.

Some performance tests (not precise. assume the rect used is always the same for each rect): OLD:

CUMULATIVE TIME: COLLIDELIST | 0.0657496 COLLIDELISTALL | 0.0851887 UNIONALL | 0.0667247 UNIONALLIP | 0.0564686001

MEAN: COLLIDELIST | 0.000131499 COLLIDELISTALL | 0.000170377 UNIONALL | 0.000133449 UNIONALLIP | 0.000112937

NEW

CUMULATIVE TIME: COLLIDELIST | 0.0327105 COLLIDELISTALL | 0.0491473 UNIONALL | 0.0377327 UNIONALLIP | 0.0290111001

MEAN: COLLIDELIST | 6.5421e-05 COLLIDELISTALL | 9.8295e-05 UNIONALL | 7.5465e-05 UNIONALLIP | 5.8022e-05

---

[itzpr3d4t0r]

Update on performance. After careful and more general testing i can say that (Best Case: all rects are normalized) the mean improvement across all tests is 41.6%. The smallest performance improvment is the one of collidelist() which stands at approx. 20%. All the others are around 45-50%. (Unlikely but worse case: mixed normalized and not normalized) The mean improvement goes down to 32,57 %, with the wost improvement still held by collidelist() at 8% improv.

[itzpr3d4t0r]

Some newer performance tests: OLD

CUMULATIVE TIME:
COLLIDELIST |  0.0465923
COLLIDELISTALL |  0.1035984
UNIONALL |  0.0792499
UNIONALLIP |  0.0672198
COLLIDEOBJECTS |  0.1883542
COLLIDEOBJECTS_ALL |  0.4826421

MEAN:
COLLIDELIST |  9.3185e-05
COLLIDELISTALL |  0.000207197
UNIONALL |  0.0001585
UNIONALLIP |  0.00013444
COLLIDEOBJECTS |  0.000376708
COLLIDEOBJECTS_ALL |  0.000965284

NEW

CUMULATIVE TIME:
COLLIDELIST |  0.0201688
COLLIDELISTALL |  0.058704
UNIONALL |  0.0394498
UNIONALLIP |  0.030145
COLLIDEOBJECTS |  0.1597061
COLLIDEOBJECTS_ALL |  0.4349253

MEAN:
COLLIDELIST |  4.0338e-05
COLLIDELISTALL |  0.000117408
UNIONALL |  7.89e-05
UNIONALLIP |  6.029e-05
COLLIDEOBJECTS |  0.000319412
COLLIDEOBJECTS_ALL |  0.000869851

Improvements

COLLIDELIST |  131% better
COLLIDELISTALL |  76.47% better
UNIONALL |  100.88% better
UNIONALLIP |  122.9% better
COLLIDEOBJECTS | 17.9%better
COLLIDEOBJECTS_ALL |  10.97% better

NOTE. This does not mean that functions will be 100% faster across the board. These results refer to a list of rects of 10 elements, all being Rect instances. Longer sequences will possibly yield smaller benefits. Like for example COLLIDELISTALL with a 20 rects list will be 64% better than the old implementation,.