PicoGK Runtime is the C++ backbone to PicoGK a compact, robust geometry kernel for Computational Engineering.
This repo is hosted at: https://github.com/leap71/PicoGKRuntime
You will usually not clone this directory, unless you want to compile the PicoGK Runtime yourself instead of using the available installers.
PicoGK Runtime is based on the OpenVDB and GLFW open source projects.
For more information, please see the main PicoGK repository.
PicoGK was developed by Lin Kayser, and is maintained by LEAP 71. PicoGK and PicoGK Runtime are released under the Apache License 2.0 (SPDX Apache-2.0).
Voxels_RenderImplicit evaluates the signed-distance callback on oneTBB worker threads by default. The callback must be pure in the sample point: no shared mutable state and no thread-affine resources. A stateful implicit can race. Set PICOGK_IMPLICIT_SERIAL to force the original single-threaded loop. Voxels_IntersectImplicit calls RenderImplicit, so it uses the same callback contract.
PICOGK_IMPLICIT_SERIAL is read with getenv on every RenderImplicit call. It is not cached.
The variable is trimmed of ASCII whitespace (space, tab, CR, LF, VT, FF) and compared case-insensitively:
- Serial:
1,true,yes,on - Parallel: unset, empty,
0,false,no,off
Any other value selects the serial path and prints one warning to stderr.
Two behaviors of the parallel path are part of the contract. A callback that samples the field it is rendering sees that field's state from before the current chunk is written back. If a worker throws, chunks already written stay written and the chunk in flight is not, which is a different partial grid than the serial loop. The serial loop writes each voxel before the next sample.
Voxels_RenderMesh keeps the scalar OpenVDB mesh voxelizer unless PICOGK_MESH_AVX2 is set to 1, true, yes, or on (trimmed, case-insensitive) on an x86-64 machine with AVX2. Unset, empty, 0, false, no, and off stay on the scalar path. Any other value stays on the scalar path and prints one warning to stderr. The switch is read once per process. Degenerate triangles stay on the scalar path either way. Other architectures compile the scalar path only.