ISlideComputer abstracts batched directional-distance computation so GPU
implementations can process all slide offsets in a single kernel launch.
GpuSlideComputer uses ILGPU with prepared edge data (precomputed inverse
deltas and min/max bounds) and caches stationary/moving buffers across
calls. GpuEvaluatorFactory exposes a singleton factory method.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The GPU pair evaluator reported false-positive overlaps for all
candidates because the bitmap coordinate system didn't account for
Part.CreateAtOrigin's Location offset. When rotation produced negative
coordinates, CreateAtOrigin sets Location = -bbox.Location (non-zero),
but the offset formula assumed Location was always (0,0).
Two fixes:
- Rasterize bitmaps from Part.CreateAtOrigin directly (new FromPart
method) instead of separately rotating polygons and computing bbox,
eliminating any Polygon.Rotate vs Program.Rotate mismatch
- Correct offset formula to include the Location shift:
(Part2Offset - partB.Location) instead of raw Part2Offset
Also optimized post-kernel bounding computation: pre-compute vertices
once per rotation group and process results with Parallel.For, matching
the CPU evaluator's concurrency.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The GPU bitmap evaluator produces false overlap detections due to
discretization errors at cell boundaries. Use the CPU PairEvaluator
(exact geometric intersection) for now. Also remove the double-counted
spacing dilation from GpuPairEvaluator for when GPU is revisited.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
GPU factory logic belongs with the GPU implementation, not the UI.
Changed from internal to public and updated namespace to OpenNest.Gpu.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace grid-cell-based dimensions with geometry-aware computation using
convex hull and rotating calipers to determine minimum bounding rectangle
for valid pair candidates. Overlap results now short-circuit with zeroed
metrics instead of using stale grid dimensions.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Introduces the OpenNest.Gpu class library with ILGPU dependencies and a
PartBitmap class that rasterizes Drawing closed shapes into integer grids
for GPU-based overlap testing. Supports rotation and spacing dilation.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
