Elamur Groupantsworld-induce-gpu
The groupant beside world/src/induce_gpu.rs in the Mova repository: what the code computes, what it never does, how to go back from its outputs, its invariants, cost and checks.
Facts
Implementation material. A Mova code model describes this code at one source revision: partial models, finite checks and their stated scope. It is not a reviewed general exposition, and its labels are shown as written, never raised.
Components
| component | label | where |
|---|---|---|
| approximant | partial | approximant |
| ast | partial | syntax.json |
| composition | partial | composition |
| cost | partial | cost |
| descriptant | present | descriptant |
| invariant | partial | invariant |
| tests | unexecuted gpu boundary | tests |
Obernants
The typed relations around the forward model: the behaviour complement (what the code never does), the reverse reconstruction (every input compatible with an output) and specialised views.
| id | type | label | where |
|---|---|---|---|
behavior_complement | behavior_complement | partial source derived model | behavior_complement |
reverse_reconstruction | reverse_reconstruction | partial source derived model | reverse_reconstruction |
world_induce_gpu_invariants | world_induce_gpu_invariants | partial source derived model | world_induce_gpu_invariants |
world_induce_gpu_resource_cost | world_induce_gpu_resource_cost | partial source derived model | world_induce_gpu_resource_cost |
world_induce_gpu_representation_join | world_induce_gpu_representation_join | partial source derived model | world_induce_gpu_representation_join |
world_induce_gpu_domain_partition | world_induce_gpu_domain_partition | partial source derived model | world_induce_gpu_domain_partition |
Parts
Model text
descriptant
world-induce-gpu
If n=groups.len()==0 or rel empty return zeros before CUDA. Else size=max(next_power_of_two(2n),1024), slot=high32((key XOR rel*K1)*K2 mod2^64)&mask. Host probes to store key,rel+1,group. One kernel thread/word enumerates1..3 subsets and probes same hash, atomicAdd cnt[group]. CUDA init/compile/load/launch/copy failure exits2.
Modeling epoch 2026-10-31
Prospective source/branch/cost hypotheses were saved externally before source edits. This partial source-derived model retains observed boundaries rather than claiming complete verification. GPU source/embedded CUDA modeled, no native compile, execution or compiled negative control. No surrogate/equivalence measurement.
approximant
Forward relation
If n=groups.len()==0 or rel empty return zeros before CUDA. Else size=max(next_power_of_two(2n),1024), slot=high32((key XOR rel*K1)*K2 mod2^64)&mask. Host probes to store key,rel+1,group. One kernel thread/word enumerates1..3 subsets and probes same hash, atomicAdd cnt[group]. CUDA init/compile/load/launch/copy failure exits2.
Domain: Static source only. Nonempty path requires GPU feature/CUDA, monotone offsets length labels+1 within code array, group ids<n, label<MAX, finite sizes/offsetsu32. No native compile/run.
No executable percent surrogate: these categorical/effect observations lack a useful approximation metric. Future supported caps1/2/5/10% need a defined domain/metric and real reference evidence; no error or speed bound is claimed.
invariant
Rust/CUDA slot wraps same64bit operations. Kernel trusts off[w]..off[w+1], group ids, nonzero21bit codes, labels+1 nonzero. Exact-count intent needs counts fittingu32 and unique group keys; not verified here.
Cost component: cost.md.
cost
Host table O(size+G*probe), transfers O(codes+offsets+labels+G). Kernel sum C(A,1..3)*probe length plus atomic contention; worst probe linearG. Host/device allocations and JIT costs unmeasured.
composition
induce::break_counts constructs eligible words/dense group ids with offsets cast tou32. Default CPU build excludes this file; CPU counting tests neither compile nor execute CUDA module.
models
behavior_complement
Absent key cannot increment with valid sentinel table. Rel0 stored1, not empty sentinel0. Duplicate keys do not update every duplicate group: first match returns. No bounds validation for off/cnt indices or narrowing casts. Errors exit process rather than return Result.
reverse_reconstruction
Count vector admits any multiset of words with same matching subset totals, reordered words and unused atom variations; allzero can mean empty labels or no matches. Counts do not recover table layout/word boundaries, and wraparound adds ambiguity beyondu32.
world_induce_gpu_invariants
Rust/CUDA slot wraps same64bit operations. Kernel trusts off[w]..off[w+1], group ids, nonzero21bit codes, labels+1 nonzero. Exact-count intent needs counts fittingu32 and unique group keys; not verified here.
world_induce_gpu_resource_cost
Host table O(size+G*probe), transfers O(codes+offsets+labels+G). Kernel sum C(A,1..3)*probe length plus atomic contention; worst probe linearG. Host/device allocations and JIT costs unmeasured.
world_induce_gpu_representation_join
induce::break_counts constructs eligible words/dense group ids with offsets cast tou32. Default CPU build excludes this file; CPU counting tests neither compile nor execute CUDA module.
world_induce_gpu_domain_partition
Static source only. Nonempty path requires GPU feature/CUDA, monotone offsets length labels+1 within code array, group ids<n, label<MAX, finite sizes/offsetsu32. No native compile/run.
tests
Finite actual-code evidence — 2026-10-31
Names: []
Selected CPU command: CARGO_BUILD_JOBS=2 CARGO_TARGET_DIR=ISOLATED_TARGET cargo test --offline --locked -p world --lib --example retention_probe [--release] wave29_worldlearn -- --test-threads=1
15 selected cases passed twice debug and twice release with identical sorted names/outcomes, exits0. No native GPU compile/execution/control. No live teacher, corpus, state activation, GPU, or full-school run. Temporary synthetic files use unique process-id paths and are removed after passing tests.
GPU source/embedded CUDA modeled, no native compile, execution or compiled negative control. No surrogate/equivalence measurement.
Source SHA256: 3c014cfba45cc70d9ce4b55049e1eff3b8f127f93f95676aab0c03185b2d0a9d. Syntax generated independently twice and compared; lexical only, no semantic proof or CUDA compilation. Strict whole-crate Clippy was not repeated merely to reproduce existing dependency/source lints; Cargo warnings retained in receipts.
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/world/src/induce_gpu.rs.groupant/