mirror of
https://github.com/allaunthefox/Research-Stack.git
synced 2026-07-31 03:05:21 +00:00
Squash the four overlapping feature branches into a single change set against main, eliminating cross-PR merge conflicts and the duplicated CI-fix scripts. What this brings in (merge order #79 -> #80 -> #81 -> #89): - #79 refactor(infra): shared utilities (4-Infrastructure/lib/*: q16, hashing, jsonl, fraction_utils) + the scripts/math-first/* validators that the math-check CI requires. - #80 feat(lean): Semantics.E8Sidon (1025 lines) -- Eisenstein coefficient identity E4^2 = E8 and the Sidon framework. E4_sq_eq_E8_coeff is fully proved (all Fourier-coefficient extraction machine-checked); the single residual gap is pinned to E4_sq_eq_E8_qExpansion (Mathlib lacks the valence formula / dim M8 = 1). 4 sorries + 1 axiom (e8_additive_completeness), all TODO(lean-port). - #81 refactor(lean): Float-free FixedPoint core (integer-only sqrt/log2/expNeg). E8Sidon.lean kept at #80's final 1025-line version (the #81 intermediate 438-line copy was overridden by merge order). - #89 feat(lean): Semantics.RRC.PolyFactorIdentity -- short-sleeve polynomial detection at the zerocopy limb boundary; now imports Semantics.E8Sidon for sigma3/sigma7/convolutionLHS (single source of truth) instead of inlining them. Conflict resolution: - flake.nix -> canonical rs-surface removal (Garnix shutdown). - scripts/math-first/* -> byte-identical across branches, clean. - .cursorrules / AGENTS.md -> unified; baselines + sorry inventory refreshed. Verification: - lake build (default aggregator): 3573 jobs, 0 errors. - lake build Semantics.RRC.PolyFactorIdentity (E8Sidon + FixedPoint + PolyFactor): 3655 jobs, 0 errors. Witnesses verified (sigma7 4 = 16513, convolutionLHS 6 = 2350). - Python tests: 68/68 pass. Note: the "Workers Builds: researchstack" check is a preexisting external Cloudflare build unrelated to this change (no branch touches 4-Infrastructure/cloudflare/). Build: 3573 jobs (default), 3655 jobs (narrow), 0 errors Co-Authored-By: Allaun Silverfox <bigdataiscoming+9i37y6j2@protonmail.com>
183 lines
5.7 KiB
Python
183 lines
5.7 KiB
Python
#!/usr/bin/env python3
|
|
"""
|
|
Q16_16 LUT Generator - Python fallback for LUT generation when WGSL runtime unavailable.
|
|
|
|
This script computes all combinations of Q16_16 operations and generates a lookup table.
|
|
For each pair (a, b) in Q16_16 space (65,536² = 4,294,967,296 combinations), it computes:
|
|
- add(a, b)
|
|
- sub(a, b)
|
|
- mul(a, b)
|
|
- div(a, b) (when b != 0)
|
|
- max(a, b)
|
|
- min(a, b)
|
|
- neg(a)
|
|
- abs(a)
|
|
|
|
The output is a packed LUT that can be used for fast hardware implementation.
|
|
"""
|
|
|
|
import numpy as np
|
|
import json
|
|
from pathlib import Path
|
|
|
|
|
|
def q16_add(a, b):
|
|
"""Q16_16 addition (wrapping)."""
|
|
return (a + b) & 0xFFFFFFFF
|
|
|
|
def q16_sub(a, b):
|
|
"""Q16_16 subtraction (wrapping)."""
|
|
return (a - b) & 0xFFFFFFFF
|
|
|
|
def q16_mul(a, b):
|
|
"""Q16_16 multiplication (high 32 bits of 64-bit product)."""
|
|
prod = (a * b) >> 16
|
|
return prod & 0xFFFFFFFF
|
|
|
|
def q16_div(a, b):
|
|
"""Q16_16 division (with division by zero handling)."""
|
|
if b == 0:
|
|
return 0xFFFFFFFF
|
|
numerator = (a << 16) & 0xFFFFFFFFFFFFFFFF
|
|
return (numerator // b) & 0xFFFFFFFF
|
|
|
|
def q16_max(a, b):
|
|
"""Q16_16 maximum (unsigned comparison)."""
|
|
return a if a > b else b
|
|
|
|
def q16_min(a, b):
|
|
"""Q16_16 minimum (unsigned comparison)."""
|
|
return a if a < b else b
|
|
|
|
def q16_neg(a):
|
|
"""Q16_16 negation (two's complement)."""
|
|
return (-a) & 0xFFFFFFFF
|
|
|
|
def q16_abs(a):
|
|
"""Q16_16 absolute value (sign bit check)."""
|
|
if a & 0x80000000:
|
|
return q16_neg(a)
|
|
return a
|
|
|
|
def generate_lut_sample():
|
|
"""Generate a sample LUT (not full 4.3B combinations due to memory constraints)."""
|
|
Q16_SPACE = 65536
|
|
SAMPLE_SIZE = 1000 # Sample size for demonstration
|
|
|
|
print(f"Generating sample Q16_16 LUT ({SAMPLE_SIZE} combinations)...")
|
|
print(f"Full LUT would require {Q16_SPACE * Q16_SPACE:,} combinations (4.3 billion)")
|
|
print(f"Full LUT size: {(Q16_SPACE * Q16_SPACE * 32):,} bytes ({(Q16_SPACE * Q16_SPACE * 32 / (1024**3)):.2f} GB)")
|
|
|
|
# Generate sample LUT
|
|
lut = []
|
|
for i in range(SAMPLE_SIZE):
|
|
a = (i * 137) % Q16_SPACE # Pseudo-random a
|
|
b = (i * 251) % Q16_SPACE # Pseudo-random b
|
|
|
|
entry = {
|
|
'a': a,
|
|
'b': b,
|
|
'add': q16_add(a, b),
|
|
'sub': q16_sub(a, b),
|
|
'mul': q16_mul(a, b),
|
|
'div': q16_div(a, b),
|
|
'max': q16_max(a, b),
|
|
'min': q16_min(a, b),
|
|
'neg': q16_neg(a),
|
|
'abs': q16_abs(a)
|
|
}
|
|
lut.append(entry)
|
|
|
|
# Save sample LUT
|
|
output_path = Path('/home/allaun/Documents/Research Stack/out/q16_lut_sample.json')
|
|
with open(output_path, 'w') as f:
|
|
json.dump({
|
|
'metadata': {
|
|
'q16_space': Q16_SPACE,
|
|
'total_combinations': Q16_SPACE * Q16_SPACE,
|
|
'sample_size': SAMPLE_SIZE,
|
|
'full_lut_size_bytes': Q16_SPACE * Q16_SPACE * 32,
|
|
'full_lut_size_gb': (Q16_SPACE * Q16_SPACE * 32) / (1024**3)
|
|
},
|
|
'lut': lut
|
|
}, f, indent=2)
|
|
|
|
print(f"Sample LUT saved to {output_path}")
|
|
print(f"Sample entries: {len(lut)}")
|
|
|
|
# Statistics
|
|
adds = [e['add'] for e in lut]
|
|
muls = [e['mul'] for e in lut]
|
|
divs = [e['div'] for e in lut]
|
|
|
|
print(f"\nStatistics:")
|
|
print(f" Add range: [{min(adds)}, {max(adds)}]")
|
|
print(f" Mul range: [{min(muls)}, {max(muls)}]")
|
|
print(f" Div by zero: {divs.count(0xFFFFFFFF)}")
|
|
|
|
return lut
|
|
|
|
def generate_full_lut_chunked():
|
|
"""
|
|
Generate full LUT in chunks to avoid memory issues.
|
|
This would generate the actual 4.3B combinations but is disabled by default.
|
|
"""
|
|
Q16_SPACE = 65536
|
|
CHUNK_SIZE = 10000 # Process in chunks
|
|
|
|
print(f"Generating full Q16_16 LUT in chunks of {CHUNK_SIZE}...")
|
|
print(f"Total combinations: {Q16_SPACE * Q16_SPACE:,}")
|
|
|
|
output_dir = Path('/home/allaun/Documents/Research Stack/out/q16_lut_chunks')
|
|
output_dir.mkdir(exist_ok=True)
|
|
|
|
chunk_count = 0
|
|
for chunk_start in range(0, Q16_SPACE * Q16_SPACE, CHUNK_SIZE):
|
|
chunk_end = min(chunk_start + CHUNK_SIZE, Q16_SPACE * Q16_SPACE)
|
|
|
|
lut_chunk = []
|
|
for idx in range(chunk_start, chunk_end):
|
|
a = idx // Q16_SPACE
|
|
b = idx % Q16_SPACE
|
|
|
|
entry = {
|
|
'a': a,
|
|
'b': b,
|
|
'add': q16_add(a, b),
|
|
'sub': q16_sub(a, b),
|
|
'mul': q16_mul(a, b),
|
|
'div': q16_div(a, b),
|
|
'max': q16_max(a, b),
|
|
'min': q16_min(a, b),
|
|
'neg': q16_neg(a),
|
|
'abs': q16_abs(a)
|
|
}
|
|
lut_chunk.append(entry)
|
|
|
|
# Save chunk
|
|
chunk_file = output_dir / f'chunk_{chunk_count:06d}.json'
|
|
with open(chunk_file, 'w') as f:
|
|
json.dump(lut_chunk, f)
|
|
|
|
chunk_count += 1
|
|
if chunk_count % 100 == 0:
|
|
print(f" Processed {chunk_end:,} / {Q16_SPACE * Q16_SPACE:,} combinations ({chunk_end / (Q16_SPACE * Q16_SPACE) * 100:.2f}%)")
|
|
|
|
print(f"Full LUT saved to {output_dir} in {chunk_count} chunks")
|
|
|
|
def main():
|
|
print("Q16_16 LUT Generator")
|
|
print("=" * 50)
|
|
|
|
# Generate sample LUT (for demonstration)
|
|
generate_lut_sample()
|
|
|
|
# Note: Full LUT generation is disabled by default due to memory constraints
|
|
# To generate the full 4.3B combination LUT, uncomment the following line:
|
|
# generate_full_lut_chunked()
|
|
|
|
print("\nNote: Full LUT generation (4.3B combinations) requires ~137 GB of storage.")
|
|
print("Use the WGSL shader (q16_lut_generator.wgsl) with GPU for full LUT generation.")
|
|
|
|
if __name__ == '__main__':
|
|
main()
|