Previous topic

Manual

Release Notes

1.2.0

  • New API to retrieve particle neighbor information, see NvFlexGetNeighbors()
  • New API to support shape collision filtering, collision channels can be assigned to particles and shapes to specify which pairs collide, see NvFlexPhase
  • New API to support per-shape plastic deformation, it is now possible to specify plastic creep and threshold coefficient on a per-shape basis (previous global settings), see NvFlexSetRigids()
  • New API to selectively disable solver features, this can lead to improved performance, see NvFlexFeatureMode, replaces global FlexParams::fluid option
  • New API to allow sub-range particle buffer updates, see NvFlexCopyDesc
  • New API for asynchronous compute, see NvFlexInitDesc
  • Fix for unnecessary synchronization when calling NvFlexMap() on a device buffer
  • Fix numerical precision for shape-matching constraints with large coordinates
  • Fix for uninitialized contact indices returned from NvFlexGetContacts() if NvFlexUpdateSolver() has not been called
  • Fix for SDF collision in D3D12
  • Fix for D3D warning when using NvFlexSetRigids()
  • Add D3D12 rendering support to the demo
  • Add support for device NvFlexBuffers on D3D

1.1.0

  • New API style, for consistency with other products the API has now an NvFlex prefix and follows a naming convention similar to PhysX
  • Add support for DirectX, in addition to CUDA there is now a cross platform DirectX 11 and 12 version of the Flex libraries that Windows applications can link against
  • Add support for max acceleration clamping, see NvFlexParams::maxAcceleration, can be useful to reduce popping with fast moving kinematic shapes and large interpenetration
  • Add support to querying compute device, see NvFlexGetDeviceName()
  • Add support for flushing compute queue, see NvFlexFlush()
  • Add support for multiple library instances, NvFlexInit() now returns a library which is bound to a single compute device
  • Add support for local space particle simulation, see NvFlexExtMovingFrameInit() and two new local space fluid and cloth demos
  • Add support for CUDA 8.0.44
  • Remove NvFlexError enum, errors will be reported through the NvFlexErrorCallback passed to NvFlexInit()
  • Remove NvFlexMemory enum, users should use the new NvFlexBufferType
  • Fix bug in inflatable constraint solver that could cause some shapes to generate NaNs
  • Fix for SDF contact generation when the particle lies completely outside volume, fixes a problem with ghost contacts
  • Fix for friction being incorrectly applied on trigger shapes
  • Fix for multi-phase fluid cohesion not working correctly, re-enable Rayleigh-Taylor instability test
  • Fix bug with public projects that referenced internal CUDA paths
  • Fix for calling NvFlexSetInflatables() with a count of zero
  • Fix for buoyancy parameter
  • Fix for bug in BVH builder that could cause missed collisions
  • New optional NvFlexDevice library for handling GPU selection and initialization in CUDA
  • New buffer based API, all data must now be passed to Flex through FlexBuffers created through NvFlexAllocBuffer()
  • New stable shape matching code for more robust and efficient soft bodies, removes need for NvFlexParams::inertiaBias
  • New collision shape API supports instancing of convex meshes and a much simplified API, see NvFlexSetShapes()
  • Improvements to collision detection for moving shapes, Flex will now perform a CCD check also taking into account the shape trajectory using the prev/current transforms of each shape

1.0.0

  • Added support for reporting collision shape ids, and trigger volume shapes, see flexGetContacts()
  • Optimizations to host code performance
  • Fix for potential memory leak in SDF object destruction
  • Fix for potentially missed collisions during convex shape CCD
  • Fix for incorrect bounds computation during flexSetShapes() (if not specified by user)
  • Fix for initial shape translations being incorrect when using a transform with flexExtCreateInstance()
  • Move flexExt.h header to the /include folder

0.9.5

  • Fix for NaNs being generated during shape matching with plastic deformation enabled
  • Fix for bug that could lead to translational drifting of shape matched rigid bodies
  • Fix for incorrect interaction of restitution and friction when using the collision margin parameters
  • New geometry instancing API for specifying collision shapes, adds support for transformed triangle meshes and signed distance fields, see flexSetShapes()
  • New BVH based collision detection improves performance for scenes with large triangle meshes
  • New mesh tearing API, see flexExtTearClothMesh()
  • New CCD collision detection for particle versus convex shape, helps reduce tunneling with thin convexes
  • Performance improvements for soft body cooking times in flexExtCreateSoftFromMesh()
  • Exposed intertial bias for shape matching constraints, see FlexParams::mInertiaBias
  • Exposed parameter to control the lifetime of diffuse particles, see FlexParams::mDiffuseLifetime

0.9.0

  • Added support for soft bodies, particles can now belong to multiple shape-matching clusters used to drive linear blend skinning
  • Added support for particle rest positions that can be used to filter collisions between particles that touch in their rest pose, see flexSetRestPositions()
  • Added support for spheres and capsules as collision shapes, renamed flexSetConvexes() to flexSetShapes()
  • Added support for user callbacks, these can be used to implement custom constraints, force fields, etc. See flexRegisterSolverCallback()
  • Added support for the NVIDIA control panel, Flex will now use the selected PhysX device if desired, see flexInit() for details
  • Added support for Android platforms
  • Added flexGetParams() method to retrieve default values and set defaults to sensible values
  • Improved shape-matching algorithm with better inversion handling
  • Optimizations for flexSetTriangles(), users can now pass the triangle bounds to avoid having them re-calculated explicitly
  • Rigid constraints now use quaternions as their orientation representation, see flexSetRigids() and flexGetRigidTransforms()
  • Exposed anisotropy min/max clamping for better control of particle fluid rendering, see FlexParams::mAnisotropyMin and FlexParams::mAnisotropyMax
  • Fix for smooth particles not being generated if anisotropy was not enabled
  • Fix for demo mouse picking
  • Fix for flexGetContacts() returning incorrect values, added contact visualization to the demo app
  • Fix for bug in flexGetBounds() which meant it could return bounds with one frame delay, users should now synchronize the flexGetBounds() call like other query methods
  • Fix whitespace on flex.h to play nice with Github’s code view
  • Fix for triangle mesh collision when particle moves in the plane of the triangle incorrectly reporting contact
  • Fix leak when destroying solver
  • Fix for incomplete initialization of FlexExtAssets in the extension creation methods
  • Rename FlexParams::mMaxVelocity -> FlexParams::mMaxSpeed

0.8.0

  • Upgrade to CUDA 7.0
  • Linux binaries are now compiled with -fPIC
  • Fix for extensions library not writing velocities correctly
  • Fix for memory leak when using moving convex shapes
  • Optimizations for inflatable constraints to reduce number of kernel launches
  • Added option to control the max number of neighbors stored per-particle, this can reduce memory usage, see flexCreateSolver()

0.2.5

  • Fix for vertex normal calculation when using sparse active particle set
  • Fix for division by zero in rest density calculation when fluid rest distance was 0.0
  • Fix for infinite mass particles being moved by kinematic objects
  • Added CUDA force fields to extensions library, see flexExtSetForceFields()
  • Added particle contact reporting API, see flexGetContacts()
  • Added early out to avoid Laplacian smoothing and anisotropy calculation if not using them
  • Added global damping parameter to model viscous drag on particles (not just cloth), see FlexParams::mDamping
  • Added user manual in pdf and chm formats
  • Made enum names consistent, eFlexHost -> eFlexMemoryHost
  • Upgraded to CUB 1.3.2
  • Extended particle phase parameters to allow fine grained control over self-collision and fluid properties, see flexMakePhase()

0.2.3

  • Switched to bindless textures, 50% reduction in CPU usage, because of this change Flex now requires an SM3.0 or above GPU
  • Optimized convex grid rasterization by assigning 4 warps/shape, reduces time from 1ms->0.25ms for large shapes
  • Added error mesasge callback, see flexInit()
  • Added flexSetFence()/flexWaitFence() for more fine grained synchronization with Flex GPU work, flexSynchronize() has been removed
  • Added static/dynamic flags for convex shapes to allow prioritising static contacts over dynamic, can be useful to prevent tunnelling through thin shapes
  • Added local/global relaxation to improve convergence for some scenes (see FlexParams::eRelaxationMode)
  • Removed flexGetVorticities(), allows optimizations for vorticity confinement calculation
  • Fix for flexSetSprings(), flexSetTriangles(), and flexSetConvexes() reallocing more often than necessary
  • Fix for flexGetDensities(), and flexGetAnisotropy() not returning the full particle array if using sparse active set
  • Fix for memory leak when instance creation fails in flexExtCreateInstance()
  • Fix for memory leak when setting rigid bodies multiple times
  • Fix for potential GPU crash when using plastic deformation on rigid bodies
  • Smooth vertex normals are now computed for particles belonging to dynamic triangles (cloth), they can be retrieved via flexGetNormals()
  • Optimized CompactObjects in flexExtPushToDevice()
  • Made headers compile with /W4 on MSVC
  • Added debug capture functionality

0.2.2

  • Fix for infinite mass particles gaining velocity during collisions
  • Fix bug with planar triangle mesh rasterization
  • Fix for zero-length distance constraints
  • Fix cloth drag and lift causing instability at high velocities
  • Fix for convex grid perf. when there is a single large convex body
  • Improved behaviour when particles are overconstrained (constraint averaging), can reduce jitter for pinned cloth
  • Improved adhesion behaviour for fluids against solids
  • Improved restitution behavior
  • Improved fluid rendering performance by using a smaller bilateral filter kernel
  • Change particle friction to be an absolute parameter instead of a multiplier on dynamic friction
  • Change sort to use async device-device memcpys to avoid stalling host
  • Exposed new parameter to control collision margin against shapes see FlexParams::mShapeCollisionMargin
  • Warning fixes for Linux and Windows
  • Fix z-precision rendering issues in demo
  • Added “Ext” prefix to extensions types
  • Clamp particle velocity to maximum user velocity before prediction step, improves behaviour when interacting with force fields

0.2.1

  • Add support for lower dimensional particle rigid bodies (rods, planes)
  • Allow specifying rigid rotations through the solver API
  • Allow separate control of fluid and solid rest distances
  • Add GUI to demo

0.2.0

  • Fix for rest density calculation when number of neighbors > than max
  • Add support for async memory copies back to pinned host memory
  • Add functions to allocate pinned host memory, see flexAlloc()
  • Add rigid body support to FlexExtensions
  • Add rigid body SDF collision support
  • Add position level friction
  • Static triangle grid now created on the GPU, see flexSetTriangles
  • Unify triangle collision with rest of the pipline
  • Exposed a new parameter to run a second collision pass against triangle meshes, see FlexParam::mEnableCCD
  • Optimized rigid body solve using CUB parallel reductions
  • Fix for solve springs when some particles have infinite mass
  • Reduce allocations in flexSetConvexes
  • Fix plastic deformation for rigids
  • Switch to /MT for FlexExtensions library

0.1.0

Initial release