GPU-accelerated AprilTag detection (OpenCL), bit-exact drop-in - #9

Draft

jameshmcvay wants to merge 18 commits into

Commits on Jun 10, 2026

Add OpenCL GPU frontend: threshold, connected components, gradient clusters

jameshmcvay
and
claude
committed
46bec9d
Emit boundary records in raster order: bit-exact output by default

jameshmcvay
and
claude
committed
cbabba6
Optimize extraction: workgroup compaction, mask reuse, buffer-cache reuse

jameshmcvay
and
claude
committed
9ac8dbc

Commits on Jun 11, 2026

Add fit_quads GPU port plan with measured fp64 feasibility

jameshmcvay
and
claude
committed
8c3f7c2
Add stable GPU radix sort by cluster key (P1 validation scaffolding)

jameshmcvay
and
claude
committed
0aed0b8
Add gather permutation path: cluster-contiguous records on-device (P1b)

jameshmcvay
and
claude
committed
ffd9853
Add GPU fit preparation and exact-ptsort slope sort (P2)

jameshmcvay
and
claude
committed
3d77cc3
Note verified P3 facts: maxima qsort is order-irrelevant

jameshmcvay
and
claude
committed
c63c515
Add GPU fit_quads tail: lfps, maxima, combos, corners on-device (P3)
jameshmcvay
committed
2187bcd
Run the build walk slim when the GPU fit consumes the frame (P4)
jameshmcvay
committed
3639a09
Upload the fit sort list with a non-blocking write
jameshmcvay
committed
1b9aaa1
Accumulate the fitPrep gradient dot from local memory chunks
jameshmcvay
committed
5c596a4
Split shallow ptsort merges across lanes in the SLM sort
jameshmcvay
committed
2e4c963
Fuse the lfps moment prep and scan into one per-cluster kernel
jameshmcvay
committed
15b0f30
Add the vide live-test harness and robot setup notes
jameshmcvay
committed
02a469a
Add the fleet experiment toolkit from the live-test campaign
jameshmcvay
committed
ed55efb
Record the fleet campaign results and findings
jameshmcvay
committed
4c669ae
Add the definitive controlled cross-generation comparison
jameshmcvay
committed
ae7b335