nvidiagameworks / nvcloth Goto Github PK

Yesterday I met a crash when I called cloth->removeCollisionCapsules() in PhysX-3.3.
And then I find some similar code in NvCloth.
When I call cloth->setCapsules(nv::cloth::Range<uint32_t>(), 0, 1) , is srcIndices null ?
If srcIndices is null, is dstIndices[i] = srcIndices[i - first] valid?

Hi，I get trouble about understanding the wind computation. Everything is fine, but why end the function, apply the impulse to the particle by the subtraction.? Thanks a lot.

Hello,

I have recently been working on integrating PhysX 4.1 and NvCloth 1.1.6 to an engine.
I noticed that these 2 libraries have compatibility issues.
There are indeed a few minor incompatibilities that can be easily fixed, however I've encountered a crash that was quite intriguing.

Symptoms:
When I enable the CctCharacterController (more precisely a CctCapsuleController), I fall and as soon as there is a collision with some shapes (box/height field/...) it crashes, with a heap corruption error.
Note that this happens only in release and profile, debug and checked targets work fine.

Investigation and fix:
As I debugged, I saw that the crash was happening when PhysX tries to add elements to an array.
First element is added, but not the second one. Actually, what it does is quite simple, if it doesn't have enough memory in the array, then it allocates a new, bigger array, copies the previous one into the new one and deletes the previous one. However, in that case, it doesn't manage to delete the previous one.
Here is why :
(in c++ it's in
PsArray.h
PX_NOINLINE T& Array<T, Alloc>::growAndPushBack(const T& a)
see
T* newData = allocate(capacity);
the following asm code is a part of what the allocate function does )

00007FFDF9E3E168 test rdx,rdx
00007FFDF9E3E16B je physx::shdfnd::ReflectionAllocator::allocate+4Eh (07FFDF9E3E19Eh)
00007FFDF9E3E16D call qword ptr [__imp_GetNvClothAllocator (07FFDF9F0D458h)]
00007FFDF9E3E173 mov r9,rsi
00007FFDF9E3E176 mov dword ptr [rsp+20h],edi

as you can see, somehow, PhysX asks NvCloth to allocate memory. But it cannot free that memory since it belongs to another heap.
To fix it, I didn't touch PhysX, but in NvCloth I've renamed ReflectionAllocator to NvClothReflectionAllocator (and NonTrackingAllocator to NvClothNonTrackingAllocator, to be sure).
I understand that the compiler got confused because these functions have the same name, are in classes that have the same name, and are in namespaces that also have the same name.

compiler used : MSVC 14
hope that helps

Is it possible to make NvCloth work with CUDA on OSX and Linux?

Hi,

I miss the D3D9 ClothingTool from APEX 1.3
Its missing in APEX 1.4 and in nvCloth I did not find any .apx importer. Only always a plane mesh.
So how to create custom cloth based on another mesh?

Btw, is the.APX file format compatible with all APEX versions?
If the .apx file version must match the version of the used PhysX, how to create Clothing .apx files for APEX 1.4 without Maya/3DsMax?
Yet I use Blender and there I wrote my own Python exporter for exporting data.
Is there a Public File Specification of the APX file format? So that I can write my own exporter?
(I don't yet use UE4)

Thanks _

Previously, with PhysX Cloth, there was a plugin for creating .apx files in Maya, available on Windows.
Can the source code for this plugin be shared?
Would like to compile the same for Linux, for use in VFX.
thanks

Hi.

How do the collisions work when a cloth collides with another cloth ? No collision primitives are created in this case, so I am a little confused. Can you please clarify ?

Thank you and best regards !

I meet a trouble when using nvcloth to simulation. My mesh's bounding box is about 0.30.30.3, it's a very tiny mesh. Everything is fine, but only the effect of the simulation is very strange, it looks like a Jelly, not a cloth. Anyone have an idea about this, it bothers me a few days, many thanks. By the way, I can provide a short video capture to your email.

I set some parameters according to the apx document but it is not useful.

Can you add some examples of getSeparationConstraints？

The CUDA version of that function updates a float per thread so threads 0,1,2,3 would update floats X,Y,Z,W of particle 0.
The DX version looks like a straight port of that but is using float4 so threads 0,1,2,3 will read/update/write all of XYZW for the same particle at the same time. Since thread 3 has a zero value for sqrIterDt you've got 4 threads writing different values back to the same location.

Should probably change sqrIterDt to
float4 sqrIterDt = float4(gFrameData.mIterDt * gFrameData.mIterDt, gFrameData.mIterDt * gFrameData.mIterDt, gFrameData.mIterDt * gFrameData.mIterDt, 0.0f);
and remove all the "*4" and "/4" calculations to leave it as one thread per particle. To make it one thread per float would require additional get/set operators for curParticles.

I am working on cloth simulation using NvCloth and I found out that NvCloth Does not support the mesh collider
contains more than 500 triangles:
in the file CuCloth.cpp:

"clamp to 500 triangles (1500 vertices) to prevent running out of shared memory"

Does anybody know how to deal with this? I want to increase the shared memory and increase triangle numbers in my mesh collider.

Best

Is there any plan on including other kinds of collider objects, such as signed distance fields? Or perhaps increasing the limit on the number of sphere or capsule colliders, so that you could approximate a non-trivial object with the union of spheres.

The main NvCloth Project got the Options to disable the CUDA and DX Solver. It would make sense to expose the same Options for the samples Project.

From the documentation:

CmakeGenerateProjects.bat <0|1:use_cuda (default is 1)> <0|1:use_dx11 (default is 1)>

When running CmakeGenerateProjects.bat:
NVGTL is unreachable or returned HTTP error

There are multible Compiler Errors when Setting USE_SELF_COLLISION_SORT to 0.

undeclared identifier 'i'

NvCloth/NvCloth/src/dx/DxSolverKernel.hlsl

Line 1566 in 145cd88

for (i = threadIdx; i < numParticles; i += blockDim)

undeclared identifier 'cdistSq'

NvCloth/NvCloth/src/dx/DxSolverKernel.hlsl

Line 1619 in 145cd88

if (distSqr > cdistSq)

undeclared identifier 'cellStartOffset'

NvCloth/NvCloth/src/dx/DxSolverKernel.hlsl

Line 1672 in 145cd88

bSelfCollisionData[cellStartOffset + key] = -1;

Is there any possible for android armv7a?

I ask for help that how NvCloth to implement collision detect either for intercollision and selfcollision?
I find collision() just need particle for pram,which cofuse me.
Thanks

In NvCloth\src\SwSolver.cpp

This line

If there is any possible that I compile for armv7 with neon, however, the cpu not supported. It will end up to this function

It will enter the if body, and use "Scalar4" type, but the "Sim4fTyped" defined as "Sim4f" in the front.

Is this some kind of bug or not?

could you provide some complete and simple examples and for using NvCloth? The code on user guide is not complete and may be different for new users(me).

Did you just delete this doc?

NvCloth 1.1.3 Release has ios.toolchain.cmake, but how to compile NvCloth for iOS?

Hi,

I'm a newbie when it comes to physics. I recently started working with PhysX 4.1 and once I got the hang of it I wanted to try adding some cloth to my simulation. After realizing that there was no "PhysX Cloth" in the code that I downloaded (turns out I was reading the documentation of PhysX 3.4, which does seem to include it) I noticed that it was deprecated in favor of NvCloth.

On the main github page it says "PhysX / PxShared compatibility" which made me think that I could merge NvCloth seamlessly into my PhysX simulation but from what I gather it seems that that's not the case. If I understand this correctly I'm supposed to create mirror capsules/spheres from my PhysX simulation and keep that updated on the solver simulation so that PhysX and NvCloth will look like they're working together. Is that correct?

Also, does this mean that it's not possible to create a, for example, horizontal piece of cloth that's hanging from 4 anchor points and then drop spheres on top so those spheres eventually gather on top of the cloth (cloth affecting physx instead of only the other way around)? Is there any other PhysX/nVidia API that will allow for something like that?

Thanks, and sorry if this is not the proper place to ask this (I'll gladly move this conversation somewhere else if pointed to the right location).

Hi,
I am trying to use GenerateProjectsOsx.sh/GenerateProjectsIOS.sh but hit the same error as follows

[NvCloth]cmake/common CMAKE_MODULE_PATH = /Users/xxx/Downloads/NvCloth-master/NvCloth/../..//sw/physx/tools/CMakeModules
CMake Error at /Users/xxx/Downloads/NvCloth-master/NvCloth/compiler/cmake/common/CMakeLists.txt:25 (MESSAGE):
Could not find CMakeModules at
/Users/xxx/Downloads/NvCloth-master/NvCloth/../..//sw/physx/tools/CMakeModules

I didn't find folder like sw/physx/tools/CMakeModules, did I miss anything in my workspace? (e.g. build Physx first?)

I am on High Sierra/Xcode 10

Thanks,
Reuben

In order to enhance cloth appearance it would be nice to have the possibility of "stitching" two Cloth objects together (i.e.: apply some sort motion constraint between some particles from cloth "a" with some particles from cloth "b", so that they are always together during the simulation).

Are there any chances to have such a feature?

Hi：
Our game has some thick clothes formed by two very thin meshes. I can't deal with this case by collision detection, because there always have some collision penetration. Any good suggestion for this case? Many thanks!

Hi,
I was trying to run NvCloth tool but when I'm compiling this project in VS, cmd.exe is keeping stopped by error 1 in 171 line. How can I solve this issue? Every tip is welcomed.

Hello.

I am trying to simulate the following mesh:

I want to preserve the original shape of the cloth, but I have got something like this (despite trying to change different parameters):

Can you please suggest a workaround to this problem ?

Hi, I am trying to compile the NvCloth Sample but I am getting many errors about missing functions and data types in the GFSDK_ShadowLib.h. I am using the NVIDIA ShadowLib 3.0 found here https://developer.nvidia.com/gameworksdownload#?dn=nvidia-shadowlib-3-x but, it is probably not the one used to create the sample so I was wondering where I could get a more up to date version of ShadowLib ?

Thanks

System info: 10.13.5

After I run BuildProjectOsx.sh, got this error:

Processing project file 'scripts/packman/packages/cmake.packman.xml'
Package 'cmake-mac-x86_64' at version '3.7.2' is missing from local storage.

I tried add log in Python script, found this:

It seems that can't download cmake 3.7.2 from cloudfront.net, include .7z, .zip

I have cmake 3.12.0 already, why nvCloth can't read it and download a new one?

I have successfully initialized factory, fabric, cloth, solver, and set properties according to the User Guide, the cloth can be rendered by my own renderer, but when I call the function Solver->simulateChunk(idx), the position of the cloth doesn't change and sometimes became Nan, I wonder what mistake could I possibly make.

I use the Physx-3.3 and the cloth1 is a PxCloth*
Why dose the application crash at cloth1->removeCollisionCapsule( 1 )
(T_T)

cloth1->addCollisionSphere( PxClothCollisionSphere( PxVec3(2.0f, 0.0f, 0.0f), 1.0f ) );
cloth1->addCollisionSphere( PxClothCollisionSphere( PxVec3(0.0f, 2.0f, 0.0f), 1.0f ) );
cloth1->addCollisionSphere( PxClothCollisionSphere( PxVec3(-2.0f, 0.0f, 0.0f), 1.0f ) );
cloth1->addCollisionSphere( PxClothCollisionSphere( PxVec3(0.0f, -2.0f, 0.0f), 1.0f ) );

cloth1->addCollisionCapsule( 0, 1 );
cloth1->addCollisionCapsule( 1, 2 );
cloth1->addCollisionCapsule( 2, 3 );
cloth1->addCollisionCapsule( 3, 1 );

cloth1->removeCollisionCapsule( 1 );

I'm not able to compile 32 bit NvCloth DLL with Visual Studio 2013.

I run CmakeGenerateProjects.bat to create vc12win32-cmake, compiling the solution at the beginning VS show the error 'cmd.exe exit with code 3'

I also tried to made makelist by hand with, VS takes a while, but end with error C2220 and some warnings.

3dmax Whether to support this version nvcloth
How to use these features（backstop ,latch to nearest）

I'm not able to compile 32 bit NvCloth DLL with Visual Studio 2013.

I run CmakeGenerateProjects.bat to create vc12win32-cmake, compiling the solution at the beginning VS show the error 'cmd.exe exit with code 3'

I also tried to made makelist by hand with, VS takes a while, but end with error C2220 and some warnings.

Can you help me?

i encountered a problem when i compiled the sample:
-- Could NOT find ASSIMP (missing: ASSIMP_PATH ASSIMP_LIB ASSIMP_LIB_DEBUG)

i cloned the src code by "git clone https://github.com/NVIDIAGameWorks/NvCloth.git NvCloth", i can not find the assimp-4.1.0 folder,how can i get it?

I'm trying to use NvCloth for a cloth-body animation, where a cloth with 1k vertices will interact a deformable human model with about 10k faces, the triangle collision seems to be far too slow, does NvCloth support this kind of simulation? if it doesn't, is there any alternative method?

I have test a same example both on IOS and Android
IOS(IPhone 7) : run 10000 points cloth simulation still can get 30 fps
Android(Goole Pixel 2XL) : run a simple flag without skinning abount 600 simulation point seem to be very slow, the fps is 50 when cloth is off and the fps is 6 when enable cloth simulation
I had checked the android lib compile, the neon is enabled and the compile use - O2 option
Is that situation normal？
It seems that nvcloth on andorid is very slowly....

Hi.

there's a restriction In the CuCloth.cpp ~ cloth::CuCloth::clampTriangleCount:

cloth::Range<const PxVec3> cloth::CuCloth::clampTriangleCount(Range<const PxVec3> range, uint32_t replaceSize)
{
	// clamp to 500 triangles (1500 vertices) to prevent running out of shared memory
	uint32_t removedSize = mStartCollisionTriangles.size() - replaceSize;
	const PxVec3* clamp = range.begin() + 500 * 3 - removedSize;

	if (range.end() > clamp)
	{
		NV_CLOTH_LOG_WARNING("Too many collision triangles specified for cloth, dropping all but first 500 * 3.\n");
	}

	return Range<const PxVec3>(range.begin(), std::min(range.end(), clamp));
}

How to create more collision triangles ?
Or how to remove this restriction without causing consequent crushes?

Thank you.

PS.
Two additional questions.

As far as I understand, the shared memory on the GPU is limited. So what if more complicated cloth is composed from the primitive cloth-objects having 500 collision triangles each ?

What is an alternative way instead of using shared memory ?

g++ version 7 >= 4.7
Processing project file 'scripts/packman/packages/cmake.packman.xml'
-- The C compiler identification is GNU 7.3.0
-- The CXX compiler identification is GNU 7.3.0
-- Check for working C compiler: /usr/bin/cc
-- Check for working C compiler: /usr/bin/cc -- works
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Detecting C compile features
-- Detecting C compile features - done
-- Check for working CXX compiler: /usr/bin/c++
-- Check for working CXX compiler: /usr/bin/c++ -- works
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Detecting CXX compile features
-- Detecting CXX compile features - done
[NvCloth]cmake/linux/CMakeList.txt
[NvCloth]cmake/common/CMakeList.txt
[NvCloth]cmake/common CMAKE_MODULE_PATH = /home/gugu/project/nvidia/NvCloth/NvCloth/..//Externals/CMakeModules
NvCloth Build Platform: linux
Using CXX Compiler: /usr/bin/c++
Appending config to output names
[NvCloth]cmake/common/CMakeList.txt END
-- Found PxShared: /home/gugu/project/nvidia/NvCloth/PxShared
CMake Error at CMakeLists.txt:50 (ADD_SUBDIRECTORY):
add_subdirectory given source
"/home/gugu/project/nvidia/NvCloth/PxShared/src/compiler/cmake/linux" which
is not an existing directory.

[NvCloth]cmake/linux/NvCloth.cmake
NV_CLOTH_ENABLE_CUDA = 0
[NvCloth]cmake/common/NvCloth.cmake
[NvCloth]cmake/common/NvCloth.cmake END
[NvCloth]cmake/linux/NvCloth.cmake END
-- Configuring incomplete, errors occurred!
See also "/home/gugu/project/nvidia/NvCloth/NvCloth/compiler/linux64-debug-cmake/CMakeFiles/CMakeOutput.log".

when build with VS 2015 ,a mistake happened.
Is there any guidance for samples?

Is it posssible to integrate the cloth solver in Xcode Scenekit?

If yes, is there any “artist friendly” tutorial or documentation?

Thanks

I'm a game engine beginner.

I found crash on win32 release.

It seems like mClothData.mCurParticles is not
128-bit alignment. How can I fix it. It's ok on android x86,arm arch and win x64.

Hi, recently I make an editor for nvCloth. I give a very strong wind to the cloth, and the cloth disappears. Then I shut down the wind, and I want to reset the cloth back, so I set cloth particle's positions to animation position preposition, clearParticleAccelerations(),clearInterpolation(),clearInertia(),clearMotionConstraints(),clearSeparationConstraints(). But it not works. So is there any method to reset the cloth particle to the normal simulation state. And any method I can detect this at runtime (I want to auto reset cloth if this happens).

Hi.

I am using this function in order to initialize a custom mesh:

template<typename PositionType, typename IndexType>
bool ClothMeshData::InitializeFromData(nv::cloth::BoundedData positions, nv::cloth::BoundedData indices, physx::PxMat44 transform)

This is the usage code of InitializeFromData:

    std::vector<float> vrt;
    for (int i=0; i < mesh.vertices.size(); i++)
    {
        vrt.push_back(float(mesh.vertices[i].Position.x));
        vrt.push_back(float(mesh.vertices[i].Position.y));
        vrt.push_back(float(mesh.vertices[i].Position.z));
    }
    
    std::vector<uint32_t> ind;
    for (int i=0; i < mesh.indices.size(); i++)
    {
        ind.push_back(uint32_t(mesh.indices[i]));
    }
    
       
    bool success = clothMesh.InitializeFromData<float, uint32_t>(ClothMeshData::ToBoundedData(vrt), ClothMeshData::ToBoundedData(ind), transform);
    if (!success)
    {
        std::cout << "Failed to initialize cloth" << std::endl;
        exit(0);
    }

I have such behavior:

When I use ClothMeshData::GeneratePlaneCloth then it works fine but when I use ClothMeshData::InitializeFromData it looks like there are no constraints between the particles.

Can you please advice how to resolve the issue ?

Thanks.

PS.

I have a guess that the provided mesh is not correctly defined for simulation.

Hi，
It's me again _
I'm trying to make cloth can collide with the scene object. And I make it work with PhysX. I wrap the cloth's bounding box as a PxBoxGeometry. And then use the box to collect all triangles nearby, store it into std::vector TriangleList, and use the setTriangles to make collision detection work.
And now I found a strange phenomenon, I have two box model each one can work well with the cloth, but if I put them both, the cloth will disappear. I check my code over and over but found nothing wrong. Do I have any misunderstanding about him? NvCloth's API?
Many many thanks.

there is no CmakeGenerateAll.bat , when i'm trying use "CmakeGenerateProjects" to generate, the error happen.

Is there any date when we can see OpenGL supported ? Would be great to finally fully integrate it with our engine ;)