PureOne's picture
Release UPMV 1.0.1: standalone research, data and code
860030d verified
Raw
History Blame Contribute Delete
3.08 kB
"""A small runnable accounting compiler, not a physical sequence/CAD generator."""
import json,math,argparse
from pathlib import Path
def gf4mul(a,b):
v=0
while b:
if b&1:v^=a
b>>=1;a<<=1
if a&4:a^=7
return v
def affine_code():
return [[gf4mul(a,x)^b for x in range(4)] for a in range(4) for b in range(4)]
def compile_conflicts(spec):
if not spec.get('retirement_verified',False):
raise ValueError('Palette reuse requires an explicit qualified-retirement assumption')
code=affine_code();output=[]
for stage in spec['stages']:
vertices=stage['classes'];neighbors={v:set() for v in vertices}
for a,b in stage['conflicts']:
if a not in neighbors or b not in neighbors or a==b:raise ValueError('Invalid edge')
neighbors[a].add(b);neighbors[b].add(a)
color={}
for v in sorted(vertices,key=lambda v:(-len(neighbors[v]),v)):
used={color[n] for n in neighbors[v] if n in color}
c=next(c for c in range(len(code)) if c not in used)
color[v]=c
assert all(color[a]!=color[b] for a,b in stage['conflicts'])
output.append(dict(stage=stage['stage'],palette_size=max(color.values())+1,
assignment={v:code[c] for v,c in color.items()},
risk_bound=stage['joins']*stage['join_error_bound']+stage['retirements']*stage['retirement_error_bound']))
return dict(target=spec['target'],status='logical accounting only; physical assumptions unverified',
stages=output,maximum_active_palette=max(x['palette_size'] for x in output),
total_union_risk_bound=min(1,sum(x['risk_bound'] for x in output)),
elementary_recognition_pairs=4,slots_per_port=4,
warnings=['No DNA sequences generated','No collision graph inferred from geometry',
'Retirement flag is a user-supplied contract, not proof'])
def adaptive_pitch(sensitivity,epsilon,a=1,d=3,hmin=.001,hmax=1.):
# Unit-volume equal-volume cells; sensitivity supplied per volume.
n=len(sensitivity)
def pitch(lam):
return [hmax if s<=0 else min(hmax,max(hmin,(d/(lam*a*s))**(1/(a+d)))) for s in sensitivity]
def err(h):return sum(s*x**a for s,x in zip(sensitivity,h))/n
if err([hmin]*n)>epsilon:raise ValueError('Infeasible error budget at minimum pitch')
lo=1e-20;hi=1.
while err(pitch(hi))>epsilon:hi*=2
for _ in range(100):
mid=(lo+hi)/2
if err(pitch(mid))>epsilon:lo=mid
else:hi=mid
h=pitch(hi)
return dict(pitches=h,bounded_error=err(h),count_density=sum(x**(-d) for x in h)/n)
if __name__=='__main__':
ap=argparse.ArgumentParser();ap.add_argument('input',type=Path);ap.add_argument('output',type=Path);args=ap.parse_args()
result=compile_conflicts(json.loads(args.input.read_text()))
result['adaptive_example']=adaptive_pitch([1.,1.,.001,.001],.1,hmin=.01,hmax=1.)
args.output.write_text(json.dumps(result,indent=2))
print('Compiled',len(result['stages']),'stages; active palette',result['maximum_active_palette'])