OPCODE -

Opcode is a small collision detection library. It works on all meshes (convex or non-convex), even polygon soups.

Here are some key features of "OPCODE":

· C++ interface, developed for Windows systems using VC++ 6.0
· Works on arbitrary meshes (convex or non-convex), even polygon soups
· Current implementation uses AABB-trees
· Introduces Primitive-BV overlap tests during recursive collision queries (whereas standard libraries only rely on Primitive-Primitive and BV-BV tests)
· Introduces no-leaf trees, i.e. collision trees whose leaf nodes have been removed
· Supports collision queries on quantized trees (decompressed on-the-fly)
· Supports "first contact" or "all contacts" modes (à la RAPID)
· Uses temporal coherence for "first contact" mode (~10 to 20 times faster, useful in rigid body simulation during bisection)
· Memory footprint is 7.2 times smaller than RAPID's one, which is ideal for console games with limited ram (actually, if you use the unmodified RAPID code using double precision, it's more like 13 times smaller...)
· And yet it often runs faster than RAPID (according to RDTSC, sometimes more than 5 times faster when objects are deeply overlapping)
· Performance is usually close to RAPID's one in close-proximity situations
· Stabbing, planes & volume queries (sphere, AABB, OBB, LSS)
· Sweep-and-prune and radix-based box pruner
· Now works with deformable meshes (OPCODE 1.3)
· Hybrid trees (OPCODE 1.3) keep a maximum of 16 triangles per leaf and reorganize client triangle lists, to eventually need roughly 16 times less ram than OPCODE's standard trees. In the best case, this goes down to 1.25 byte / triangle, which is 115 times smaller than RAPID's OBB trees (using floats ! else it's 168 times). Speed hit is often negligible, and volume queries can actually run faster than OPCODE 1.2 due to less cache misses (reorganizing clients arrays also helps in this regard). They're also faster for deformable meshes.



Product's homepage

Here are some key features of "OPCODE":

· C interface, developed for Windows systems using VC 6.0
· Works on arbitrary meshes (convex or non-convex), even polygon soups
· Current implementation uses AABB-trees
· Introduces Primitive-BV overlap tests during recursive collision queries (whereas standard libraries only rely on Primitive-Primitive and BV-BV tests)
· Introduces no-leaf trees, i.e. collision trees whose leaf nodes have been removed
· Supports collision queries on quantized trees (decompressed on-the-fly)
· Supports "first contact" or "all contacts" modes (à la RAPID)
· Uses temporal coherence for "first contact" mode (~10 to 20 times faster, useful in rigid body simulation during bisection)
· Memory footprint is 7.2 times smaller than RAPID's one, which is ideal for console games with limited ram (actually, if you use the unmodified RAPID code using double precision, it's more like 13 times smaller...)
· And yet it often runs faster than RAPID (according to RDTSC, sometimes more than 5 times faster when objects are deeply overlapping)
· Performance is usually close to RAPID's one in close-proximity situations
· Stabbing, planes & volume queries (sphere, AABB, OBB, LSS)
· Sweep-and-prune and radix-based box pruner
· Now works with deformable meshes (OPCODE 1.3)
· Hybrid trees (OPCODE 1.3) keep a maximum of 16 triangles per leaf and reorganize client triangle lists, to eventually need roughly 16 times less ram than OPCODE's standard trees. In the best case, this goes down to 1.25 byte / triangle, which is 115 times smaller than RAPID's OBB trees (using floats ! else it's 168 times). Speed hit is often negligible, and volume queries can actually run faster than OPCODE 1.2 due to less cache misses (reorganizing clients arrays also helps in this regard). They're also faster for deformable meshes.