Commit Graph

40 Commits

Author SHA1 Message Date
Erwin Coumans
b53cc5edb8 disable Mpr on Mac OSX (need to split the kernel again...) 2014-01-15 11:24:35 -08:00
Erwin Coumans
472a4068fc Enable MPR by default
Add the contact point from MPR, in addition to SAT/clipping contacts. Added a new kernel to clear/reset the number of contacts in pairs (stored in the z component)
Always sample unit sphere directions, if there are more edge-edge combinations than unit sphere directions (162 by default)
Remember last running demo for Bullet 3 (and save it in a text file, Bullet
Enable the testFileFracture.bullet in the Bullet2FileDemo
2014-01-15 10:35:51 -08:00
erwin coumans
2e55d03fd9 more fixes towards working convex-convex, in case of (too) many edge-edge cases:
1) sample directions over a unit sphere and
2) add a contact, using mpr
add missing bullet.pc.cmake file (for old Bullet 2.x)
2014-01-09 16:22:34 -08:00
erwincoumans
ed73bce9da Move Bullet 2.x demos in a single demo 'AllBullet2Demos', for now only BasicDemo and FeatherstoneDemo is implemented. 2014-01-05 21:58:30 -08:00
erwincoumans
6f9a1c0da2 disable MPR until some issues are fixed (on an NVIDIA GPU, sat axis are not properly computed, resulting in instability) 2014-01-04 21:44:53 -08:00
Erwin Coumans
5eb8fd33ad enable MPR by default, needs more testing on various OpenCL configurations (OS, drive, GPU etc) 2014-01-04 21:33:16 -08:00
erwincoumans
271f458837 Ported Minkowski Portal Refinement mpr.c from libccd to OpenCL, for bettwe edge-edge performance (and additional contact point for degenerate/high detailed convex shapes)
Removed b3RigidBodyCL, replace by b3RigidBodyData and b3RigidBodyData_t shared between C++ host and OpenCL,
Same for b3InertiaCL -> b3InertiaData
2014-01-04 20:54:27 -08:00
erwin coumans
09ba86ea36 fix cpu version of findCompoundPairs, in order to fix Mac OSX OpenCL compound issues 2013-12-17 19:25:53 -08:00
erwin coumans
7964f2d8c1 split OpenCL kernels: fatal error C1091: compiler limit: string exceeds 65535 bytes in length 2013-12-17 12:33:38 -08:00
Erwin Coumans
7e86932edf avoid out-of-bounds issue for some OpenCL kernel, hanging Mac OSX (should not happen, need to check why)
split kernel for debugging
2013-12-17 10:44:41 -08:00
erwin coumans
7b55ffd237 more rewriting to get this #@!#@!&*( Mac OSX GPU to work 2013-12-16 15:23:43 -08:00
erwin coumans
1dce953d76 one more debug option for a kernel, to fix a Mac OSX issue (one more to go...) 2013-12-16 12:36:22 -08:00
erwincoumans
fc474a0e62 more debugging to fix Mac OSX OpenCL issue on certain GPUs. 2013-12-16 10:53:38 -08:00
erwin coumans
9ebef9552b implement 'new contact reduction' kernel on CPU , for debugging 2013-12-14 10:54:25 -08:00
erwin coumans
d24b1eaae9 better host check code, to determine failing OpenCL kernels on certain (Apple) platforms 2013-12-13 13:27:00 -08:00
erwin coumans
571b14787a implement clipFacesAndFindContactsCPU, to track down OpenCL issue on Mac OSX 2013-12-13 10:05:07 -08:00
erwin coumans
3fe969c4ee b3Solver -> pass pointer to source instead of 0 (was left over from a debugging session), thanks to David for the report
Break up clipHullHullConcaveConvexKernel into multiple stages, so it might 'fit' in Apple's OpenCL implementation
Implemented bvhTraversalKernel and findConcaveSeparatingAxis on CPU (debugging, possible future CPU version)
2013-12-13 07:52:41 -08:00
erwin coumans
c155e126d0 move parts of collision pipeline to shared header files (work-in-progress) 2013-12-12 11:03:55 -08:00
erwincoumans
1d5c651753 Use more shared data structures in src\Bullet3OpenCL\NarrowphaseCollision\kernels\satClipHullContacts.cl
Revert default error func (was broken)
2013-11-19 17:52:59 -08:00
erwin coumans
26dfaa441e add some command-line options
--use_jacobi
--allow_opencl_cpu

b3LauncherCL constructor takes string, to make it easier to determine failing OpenCL kernel
b3SetCustomErrorMessageFunc, printf error and exit(0)
2013-11-19 13:42:53 -08:00
erwincoumans
5ce0b3938f Re-introduce GPU parallel Jacobi solver, called btGpuJacobiContactSolver. There are still some issues, but for basic scene's it works.
To avoid confusion, rename GPU contact solver to btGpuContactSolver, and constraint (non-contact) solver to btPgsConstraintSolver.
2013-11-11 21:15:06 -08:00
erwincoumans
a6bc61ad1a tweaked the particle demo a little bit 2013-11-08 23:21:01 -08:00
erwin coumans
a9a758dd54 share more code between OpenCL kernels and C++ by moving kernel data structures/code to the 'shared' folders 2013-11-06 19:57:36 -08:00
erwincoumans
677722bba3 support compound versus compound collision shape acceleration on GPU, using aabb tree versus aabb tree.
Remove constructor from b3Vector3,  to make it a POD type, so it can go into a union (and more compatible with OpenCL float4)
Use b3MakeVector3 instead of constructor
Share some code between C++ and GPU in a shared file: see b3TransformAabb2 in src/Bullet3Collision/BroadPhaseCollision/shared/b3Aabb.h
Improve PairBench a bit, show timings and #overlapping pairs.
Increase shadowmap default size to 8192x8192 (hope the GPU supports it)
2013-08-20 03:19:59 -07:00
erwincoumans
41ba48b10d fix compound-compound tree/tree collision on the host side (still need to implement it for all cases, and on GPU/OpenCL) 2013-08-16 10:33:38 -07:00
erwin coumans
ef224370ab implement compound versus compound BVH acceleration (quantized tree-versus-tree, using subtrees and quantization) on host 2013-08-16 08:58:52 -07:00
erwin coumans
b32ae0c75c move some files to shared folders
use b3Aabb for compound vs compound on host (for testing, towards BVH)
2013-08-14 17:48:12 -07:00
erwincoumans
678f634699 fix compound collision (host side) 2013-08-11 20:02:07 -07:00
erwincoumans
ac23dbc4be contact normal should point from B to A (to be consistent with Bullet 2.x)
stringify: pre-allocate bigger buffer (10MB, workaround)
2013-08-10 12:08:15 -07:00
erwincoumans
3bf003ace1 change lcpp Lua preprocessor, to keep #defines and comments, remove empty lines
remove duplicate data in b3Contact4 (now in btContact4Data shared between CPU/C++ and OpenCL)
OpenCL kernels use #include "Bullet3Collision/NarrowPhaseCollision/shared/b3Contact4Data.h"
Increase number of batches back to 250 (from 50), need to fix this hard coded number (see https://github.com/erwincoumans/bullet3/issues/12)
Work towards GJK/EPA, in addition to SAT/clipping (early on)
2013-08-08 12:24:09 -07:00
erwincoumans
906415429c added (and stripped) a simple C/C++ preprocessor (written in Lua), so the stringifier can handle the #include directive,
and embed the included files directly in the stringified files.
We need this, because we start sharing struct definitions and code between C/C++ and OpenCL (and potentially other languages)
preprocessor is from http://github.com/willsteel/lcpp
2013-08-01 21:05:19 -07:00
erwincoumans
34de49d8a4 add gjk/epa (host only), possibly improve convex-convex with many edge-edge tests
more preparation towards persistent/incremental contact cache
2013-07-31 23:22:43 -07:00
erwincoumans
7992ff816b use int4 for broadphase pair, it allows to store persistent information in the future
(contact cache, applied impulse/warm starting info etc)
2013-07-31 09:58:15 -07:00
erwincoumans
1c0fc6bc2d report when compound pair capacity is exceeded (and avoid crash)
add tetrahedron, reorganized demos
2013-07-30 17:00:50 -07:00
erwincoumans
ab125fbb6d implement GPU breakable constraints
add GPU fixed constraint
fix performance issue with concave meshes (didn't clear the number of concave-convex pairs, so it increased every frame)
2013-07-17 16:11:54 -07:00
erwin coumans
0fcdd290f5 remove dynamic OpenCL allocations at run-time 2013-07-12 16:00:25 -07:00
erwin coumans
a69ba48de4 Move b3Quickprof.* from Bullet 3.x src folder to btgui/Timing
The Bullet 3.x B3_PROFILE can be customized using b3SetCustomEnterProfileZoneFunc/b3SetCustomLeaveProfileZoneFunc defined in Bullet3Common/b3Logging,
so you can hook Bullet 3.x up to your profiler of choice.
The Demos3/BasicGpuDemo will show the Bullet 3.x timings inside the Bullet 2.x btQuickprof profiler.
2013-06-19 22:08:03 -07:00
erwin coumans
70b4ead641 remove some warnings 2013-06-03 10:16:50 -07:00
erwin coumans
6ee9eb9bb5 fix bug in assignment of contact constraints to solver grid (always use dynamic body to determine constraint assignment, otherwise write conflicts can occur)
implement CPU version of narrowphase convex collision, for comparison/debug purposes
start towards cpu/gpu sync, for adding/removing bodies (work in progress)
2013-05-02 09:49:16 -07:00
erwincoumans
3ac332f3a7 reorder files, in preparation for Bullet 3 -> Bullet 2 merge 2013-04-29 19:04:08 -07:00