Le 06/09/2010 18:29, Matthieu Fertré a écrit :
> uint l,r,t,b;
> body->GetTestRect(l,r,t,b);
> SetTestRect(l,r,t,b);
In fact I think it's trickier. The test rect for body is always the same
(see their definition in data/body/movements.xml); this can be verified
when moving: the rectangle doesn't
Hi,
Le 05/09/2010 20:54, Mikko Vartiainen a écrit :
> On Sun, Sep 5, 2010 at 7:31 PM, Mikko Vartiainen
> wrote:
>
>> Animation seems to be the big obstacle on 4x10 case. I made crude test
>> (see below) with animations, and this alone changes completely
>> unplayable game to about 10 fps game
Attached is patch which has all of my optimisations from yesterday and today.
src/character/character.cpp:
This change is not in finished form, but just to test the effect if
FootsInVacuum wasn't calculated two times for every cycle. Gain is
very noticeable. To finish it, FootsInVacuum result shou
On Mon, Sep 6, 2010 at 9:43 AM, Kurosu wrote:
>> BTW, Cortex-A8 also lacks integer division which needs to be done in
>> software, that's why you see __aeabi_uidivmod and the likes...
>
> Sounds like *unsigned* integer division, but still I didn't know about it.
> Worse would be that for some rea
On Mon, Sep 6, 2010 at 11:59 AM, Kurosu wrote:
> Hi,
>
>
> - Mikko Vartiainen a écrit :
>> I made more refined test. Here we simply stop animation if character
>> is not visible. It makes a huge difference when there is lot a
>> characters.
>
> The patch by itself seems fine, except that appl
Hi,
- Mikko Vartiainen a écrit :
> I made more refined test. Here we simply stop animation if character
> is not visible. It makes a huge difference when there is lot a
> characters.
The patch by itself seems fine, except that applying movements are /probably/
needed for the game to update
- Mikko Vartiainen a écrit :
> 54560 8.8264 wormux
> PhysicalObj::CollidedObjectXY(Vector2 const&) const
I'm wondering if this isn't AI computing its trajectories.
This made me think about this algorithm for AI:
1) Estimate each strategy damage from its explosion, but do not estimate
Hi,
- RCL a écrit :
> It looks like the game is doing type conversions too often.
>
> This particular function (__aeabi_f2ulz) converts floats to unsigned long.
> (see here for complete list
> http://infocenter.arm.com/help/topic/com.arm.doc.ihi0043a/IHI0043A_rtabi.pdf)
That's what a google
Hi,
On Sun, Sep 5, 2010 at 10:17 PM, Mikko Vartiainen wrote:
> 65309 10.5653 libgcc_s.so.1 __aeabi_f2ulz
It looks like the game is doing type conversions too often.
This particular function (__aeabi_f2ulz) converts floats to unsigned long.
(see here for complete list
http://infoc
On Sun, Sep 5, 2010 at 11:17 PM, Mikko Vartiainen wrote:
> Here is new profile with libgcc debug symbols. This was all computer
> 4x10 game with the patch from
> https://mail.gna.org/public/wormux-dev/2010-09/msg00013.html
Here is graphical call graph
http://img266.imageshack.us/img266/8382/callg
On Sun, Sep 5, 2010 at 7:31 PM, Mikko Vartiainen wrote:
> On Sun, Sep 5, 2010 at 9:15 PM, Kurosu wrote:
>> Top callers are:
>> 95053 20.8283 libgcc_s.so.1 /lib/libgcc_s.so.1
>> 75516 16.5473 libSDL-1.2.so.0.11.1 BlitNtoNPixelAlpha
>> 41306 9.0511 libc-2.5.so
On Sun, Sep 5, 2010 at 7:31 PM, Mikko Vartiainen wrote:
> Animation seems to be the big obstacle on 4x10 case. I made crude test
> (see below) with animations, and this alone changes completely
> unplayable game to about 10 fps game.
I made more refined test. Here we simply stop animation if char
On Sun, Sep 5, 2010 at 9:15 PM, Kurosu wrote:
> Top callers are:
> 95053 20.8283 libgcc_s.so.1 /lib/libgcc_s.so.1
> 75516 16.5473 libSDL-1.2.so.0.11.1 BlitNtoNPixelAlpha
> 41306 9.0511 libc-2.5.so /lib/libc-2.5.so
>
> I'm a bit surprised by first one, as I'
Hi,
sorry for the offlist reply, I didn't notice you provided onlist a
opreport log, which I'm more used to deal with.
Le 05/09/2010 16:33, Mikko Vartiainen a écrit :
> Attached is oprofile report from Nokia N900 (600 MHz ARM Cortex-A8).
> Wormux was compiled with -mcpu=cortex-a8 -mfpu=neon
> -
Here is another case with 4x10 teams (1 Human) for comparison. Game
wasn't really playable because screen updates were quite random and
computer wasn't able to make a single shot (or at least I didn't see
one) because it run out of thinking time.
Counted GPTIMER_CYCLES events (32KiHz timer clock c
15 matches
Mail list logo
