Source code 'roadmap'?

[Aphex]

Hi all,

Not sure if I've missed this somewhere, but is there some sort of a 'roadmap' or moving guide to the new source? I'm looking at moving from Newton 2 to Open Source. From what I can surmise:
coreLibrary_200 is the stable source?
coreLibrary_300 is alpha/new unstable stuff?
packages seems to be the old Newton 2 interface stuff. Is this for backward compatibility and lib builds only? How does it co-exist with the new source?

WRT 'packages' I use the old multi-body vehicle joint - should I continue using the packaged one or move to the one in the new source? I.e. are the packages going to be deprecated or anything?
Same question for joints - there seemed to be a move towards using a customJoint for everything in Newton2 - does the same apply here?

[Julio Jerez]

core 200 and core 300 hundred are compatible at teh ineface level.
core 300 hundred is far mor stable whiol faster on teh mutthreade portion.

also core three hundred has major improvemen on the broadphase.
in core three hundred all engene parts are mutithreaded, not longer there is a maximun or optimal thread count limit.
core thread hundred maximize teh CPU usage, independe of teh thread count and with out estreesing the the CPU load.
the last thing is the major problem wih core 200 hundred in muthreading when the thread count is hight.

yes custom joint is the final joint solution for core 300.

[KingSnail]

Is the 300 branch usable right now?

[carli2]

Is the 300 branch usable right now?

no, unless it does not compile under linux.

[Aphex]

Is the dJointLibrary package dead? I try to build it and get:

Code: Select all

1>..\..\dCustomJoints\CustomWheelVehicle.cpp(131): error C3861: 'NewtonDestroyCollision': identifier not found
1>..\..\dCustomJoints\CustomWheelVehicle.cpp(705): error C3861: 'NewtonBodyApplyImpulsePair': identifier not found
1>..\..\dCustomJoints\CustomWheelVehicle.cpp(708): error C3861: 'NewtonBodyIntegrateVelocity': identifier not found
1>..\..\dCustomJoints\CustomWheelVehicle.cpp(733): error C3861: 'NewtonBodyGetInertiaMatrix': identifier not found
1>..\..\dCustomJoints\CustomPlayerController.cpp(168): error C3861: 'NewtonDestroyCollision': identifier not found


None of which appear to be implemented in the code..?

[Julio Jerez]

No it is not, as a matter of fact it is becoming much more important in core 300.
I have not gotting to revisitiy yet because I am still working on the overhold of the collision system.

the new joint library will have soem more new features, they will make it eassy to make local objects.
I am hopping I can bring the collision to a stabel point this week, so that I can move to thr joint library.

these functions

Code: Select all

>..\..\dCustomJoints\CustomWheelVehicle.cpp(131): error C3861: 'NewtonDestroyCollision': identifier not found
1>..\..\dCustomJoints\CustomWheelVehicle.cpp(705): error C3861: 'NewtonBodyApplyImpulsePair': identifier not found
1>..\..\dCustomJoints\CustomWheelVehicle.cpp(708): error C3861: 'NewtonBodyIntegrateVelocity': identifier not found
1>..\..\dCustomJoints\CustomWheelVehicle.cpp(733): error C3861: 'NewtonBodyGetInertiaMatrix': identifier not found
1>..\..\dCustomJoints\CustomPlayerController.cpp(168): error C3861: 'NewtonDestroyCollision': identifier not found

should all be there, what version are you building?

[Aphex]

Ah I was using core200 from latest svn - I see they're in core300... will try that, cheers.

[Julio Jerez]

yes give core 300 a test drive, so that you can see.

[Aphex]

Hmm getting errors and warnings all over the place.
Think I'll wait til you've managed to look at the double precision core300 build (think you said it was imminent in another post).

[Aphex]

I have no idea how to fix the SIMD stuff:

Code: Select all

..\..\..\source\core\dgPolyhedra.cpp(1188): error C2664: 'dgVector::dgVector(const simd_128 &)' : cannot convert parameter 1 from 'dgFloat64 *' to 'const simd_128 &'
1>          Reason: cannot convert from 'dgFloat64 *' to 'const simd_128'

[Julio Jerez]

is this in core 200? when you post please state what library you are compiling so that is eassy for me to find out.

anyway I just check in all core 300 with uniform scale on all regular primitives.
the new thong aboput thso version is that all close form primitive are now special case with fats collision calculation.
this is different than core 200 an previus verisions where only sphere/sphere and sphere/capsule sphere/box, and capsule/capsule where special cased.

now I will copmplete the non unifrom scale, which for the much part should be working, but ther are some funtions that requre special treatement.
then I will make continue teh close form primitive to use teh fast collison routine as well,
the fast collison routines generate high qaulity contacts, than teh genral conve contact solber because teh make usne of the features of the shapes.

finally I will enable scale on polygon mesh collision (collision tree and heightmap) and this will conclude the solid collision engine for core 300.

[Aphex]

sorry: it's core300, double precision build, win vs2010.

[Aphex]

I do see some SIMD 256 stuff in there though its turned off for Double Precision builds?
Also it uses __m256, not __m256d?

[Julio Jerez]

yes I have not made dowble presition simd class.

I never did because SSE2 you cna only do tow floats, and when I tryed I codu never get better performace than not using it, so I thopugh i was a wast of time.
It is a long time since I try again,
however with 256 register simd now is 4 double, so it may be good ides to make simd_256d

btw the 256 avx code does no really mak ethe code faster at all, it is just a prove of compset, I beleiev thsi si liek when SSE came out,
it took 9 years for intel to double the bus side and it was not until the core duo that SSE became effective.

anyway I alway feel that with VS specially in 64 bit, both float and dwouble sse makes much less differenct than when comipling in 32 bit
It takes a lot of wok\rk to make SSE code really effective, for the must cases it is a net lost if the data is not prepared.