weirflow 0.1.0

GPU-first dataflow analysis primitives for Vyre and Santh compiler pipelines.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34

# Changelog

All notable changes to weir are documented here. The format follows
[Keep a Changelog](https://keepachangelog.com/en/1.1.0/) and the crate
versioning follows [Semantic Versioning](https://semver.org/spec/v2.0.0.html).

## [Unreleased]

### Added

- Crate-level doctest demonstrating `try_compute_dominators` on a diamond CFG.
- `fold_summary_partials` runnable example documenting the `(count, any, checksum)` triple-fold semantics (corrected: `any` is a boolean reachability flag, not a bit-OR).
- `benches/primitive_scale.rs` perf ladder at 1k/10k/100k/1M nodes for `stable_csr_layout_hash` (linear chain + fanout-2 tree), `try_compute_dominators` (linear chain), and `fold_summary_partials` (3-aligned bin lengths 513/3072/24576/196608).
- Cargo metadata for crates.io publish-readiness: `homepage`, `documentation`, `keywords`, `categories`.
- Proptest depth bumped from 256/512 to **10,000 cases** across 16 `tests/df_property_*.rs` files and the in-tree `andersen_monotonicity_proptest` (per CLAUDE.md gate-6).

### Fixed

- README mis-claim: prior text said the default feature set enabled `serde`. Cargo.toml has `default = []`. README now describes the actual minimal default and how to opt into `serde` + `cpu-parity`.

### Performance baselines captured 2026-05-22

| Primitive | 1k | 10k | 100k | 1M |
|---|---|---|---|---|
| `stable_csr_layout_hash` (linear chain) | 9.1 µs | 94 µs | 928 µs | 9.15 ms |
| `stable_csr_layout_hash` (fanout-2 tree) | 9.2 µs | 88 µs | 886 µs | 8.87 ms |
| `try_compute_dominators` (linear chain) | 59 µs | 809 µs | 44 ms | n/a |
| `fold_summary_partials` | 59 ns @513 | 322 ns @3072 | 2.5 µs @24576 | 34 µs @196608 |

The `stable_csr_layout_hash` and `fold_summary_partials` paths scale linearly. The host-CPU `try_compute_dominators` shows super-linear growth between 10k and 100k blocks (55×) on the linear-chain worst case, downstream consumers needing dominator info at linux-subsystem scale should route through `vyre_primitives::graph::dominator_tree` (the GPU bitset path that surgec already uses).

## [0.1.0]. Unreleased

Initial dataflow primitives release covering SSA, reaching-defs, IFDS (with resident batch + reachability witness), points-to (Andersen), backward slicing, liveness, dominators, range checks, callgraph, control-dependence, cross-language edge fusion, escape, must-init / SCC / value-set queries, function and loop summaries, soundness contracts, and graph-layout hashing. See `INVENTORY.md` for the full module + bench surface inventory.