vyre-conform 0.1.0

Conformance suite for vyre backends — proves byte-identical output to CPU reference
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
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54

# Regression Format

A regression is a previously-failing input persisted so it can be replayed in every future run.

## File Layout

Regressions are stored under `regressions/` in a directory tree keyed by operation ID:

```text
regressions/
  primitive.bitwise.xor/
    2025-04-12T06-00-00Z.json
    2025-04-12T06-05-00Z.json
  engine.dfa/
    2025-04-12T07-00-00Z.json
```

## JSON Schema

Each regression file contains:

```json
{
  "op_id": "primitive.bitwise.xor",
  "op_version": 1,
  "generator": "edge_cases",
  "input_label": "max_xor_max",
  "input_bytes": [255, 255, 255, 255, 255, 255, 255, 255],
  "gpu_output": [0, 0, 0, 0],
  "cpu_output": [255, 255, 255, 255],
  "message": "byte 0 differs: gpu=0x00, cpu=0xFF",
  "workgroup_size": 64,
  "timestamp": "2025-04-12T06:00:00Z"
}
```

## Replay Policy

Every regression is replayed before or alongside newly generated inputs. The regression suite only grows. A fixed bug that is not replayed could silently return.

## Coverage

Skipped regressions are coverage failures. The regression coverage metric is:

```text
regression_coverage = replayed_regressions / total_regressions
```

## Migration

When an operation version increments, old-version regressions may be:
- Re-checked under the new version (if behavior is compatible)
- Archived (if behavior intentionally changed)
- Re-tagged with the new version (if the fix resolved the issue)