GPU-native AtomSpace Phase 6: class substitution

[plankatron]

Summary

GPU kernel for word class substitution — the final stage of the learning pipeline that applies clustering results back into the data structures.

• gpu-substitute.cl — Class substitution with two operations:
  • Pair merge: replaces word indices with class indices in PairPool, consolidating co-occurrence statistics
  • Section rewrite: updates section connectors to use class IDs, enabling grammar generalization
  • Uses atomic reference counting for safe concurrent class membership updates

Pipeline Position

This is Phase 6 of the GPU-native AtomSpace pipeline:

 Phase 0-1: Hash Table + Pools (#3, dependency)
 Phase 2-3: Counting + MI      (#4)
 Phase 4-5: Sections + Cosine  (#5)
[Phase 6:   Substitution] ← THIS PR

Depends on PR #3 — this kernel references WordPool, PairPool, SectionPool, and hash table structs defined in gpu-atomspace.cl and gpu-hashtable.cl. PR #3 should be merged first. No dependencies on PRs #4 or #5.

Files

File	Lines	Description
gpu-substitute.cl	232	Pair merge + section rewrite kernel
test-substitute.c	767	Substitution unit tests

Performance

• 454.6M pairs/sec throughput on substitute benchmark

Building & Testing

cd opencog/opencl/atomspace

# Substitution tests
gcc -o test-substitute test-substitute.c -lOpenCL -lm
./test-substitute

Relationship to Prior Work

Builds on PR #2 (binary caching + dispatch thread build), which was merged as 2a9c544. These kernels benefit from the binary caching infrastructure for faster program load times.

🤖 Generated with Claude Code