|
seams-core v2.14.0
libyodaLib, the C++ engine of d-SEAMS
|
Run the analysis of a resident frame batch on the device. More...
Go to the source code of this file.
Classes | |
| struct | gpu::BatchResult |
Namespaces | |
| namespace | gpu |
Functions | |
| BatchResult | gpu::analyzeResident (const double *xyz, const double *box, int nAtoms, int nFrames, double rc=5.5, const double *boxLow=nullptr, int nBox=3) |
| xyz is frame-major, atom, xyz. | |
Run the analysis of a resident frame batch on the device.
Coordinates for every accepted frame are uploaded once. The device path is the TUM ice score: linkcell (k)-nearest, the mutual four-nearest graph, primitive six-rings, and the HC/DDC affiliation predicates. CHILL and (q_{lm}) stay on the host. Only cage labels come back.
Every device counter is incremented before its bound check. A frame whose rings, atom rows or basal pairs overflow the planned buffers reruns with those buffers grown to the counted sizes. A frame with an atom row longer than the kernels' 32 entries, or whose grown buffers cannot be allocated after a completed pass, is evaluated on the host from its mutual graph. A batch the planner rejects, or whose first allocation fails, returns BatchResult::error.
Definition in file gpu_batch.hpp.