def get_rels_elems1(self, tcs):
if __debug__:
assert len(tcs) >= 1, tcs
aeterms = tcs[0].keys()
#Find rels among array elements
logger.debug('Find linear eqts over {} array elems'
.format(len(aeterms)))
aeterms = [SR(1)] + aeterms
from dig_polynomials import Eqt
solverE = Eqt(aeterms, tcs, self.xinfo)
logger.info('Select traces (note: |tcs|=|terms|={})'.format(len(aeterms)))
ntcs_extra = 200
solverE.tcs, solverE.tcs_extra = get_traces(tcs,len(aeterms),ntcs_extra=200)
solverE.do_refine=False
solverE._go()
from dig_refine import Refine
rf = Refine(solverE.sols)
rf.rfilter(tcs=solverE.tcs_extra)
ps = [p.p for p in rf.ps]
return ps
def mk_traces(self):
# will be modified in preprocess
tcs_all = Miscs.keys_to_str(self.tcs)
tcs,_ = get_traces(tcs_all,
ntcs=1,#using only 1 array
ntcs_extra = 0)
tcs_extra = tcs_all #refine on all tcs
return tcs, tcs_extra
def mk_traces(self):
"""
No need to do too much filtering because
most eqts obtained are valid
"""
tpercent = 150 # terms % for inv gen
ntcs = int(round(len(self.terms) * tpercent / 100.00))
ntcs_extra = 200
return get_traces(self.tcs, ntcs, ntcs_extra)
def mk_traces(self):
"""
No need to do too much filtering because
most eqts obtained are valid
"""
tpercent = 150 # terms % for inv gen
ntcs = int(round(len(self.terms) * tpercent / 100.00))
ntcs_extra = 200
return get_traces(self.tcs, ntcs, ntcs_extra)
def mk_traces(self):
"""
E.g.,
|tcs| = 100 then 80 for invgen, 20 for filter
|tcs| = 1000 then 300 for invgen, 700 for filter
|tcs| = 10000 then 300 for invgen, 1000 for filter
"""
max_ntcs = 300
max_ntcs_extra = 1000
if len(self.tcs) <= 100:
tpercent = 100
else:
tpercent = 80 # traces % for inv gen
ntcs = int(round(len(self.tcs) * tpercent / 100.00))
ntcs = min(ntcs, max_ntcs)
ntcs_extra = len(self.tcs) - ntcs
ntcs_extra = min(ntcs_extra, max_ntcs_extra)
return get_traces(self.tcs, ntcs, ntcs_extra)
def mk_traces(self):
"""
E.g.,
|tcs| = 100 then 80 for invgen, 20 for filter
|tcs| = 1000 then 300 for invgen, 700 for filter
|tcs| = 10000 then 300 for invgen, 1000 for filter
"""
max_ntcs = 300
max_ntcs_extra = 1000
if len(self.tcs)<=100:
tpercent=100
else:
tpercent = 80 # traces % for inv gen
ntcs = int(round(len(self.tcs) * tpercent / 100.00))
ntcs = min(ntcs,max_ntcs)
ntcs_extra = len(self.tcs) - ntcs
ntcs_extra = min(ntcs_extra,max_ntcs_extra)
return get_traces(self.tcs, ntcs, ntcs_extra)
def get_rels_elems2(self, tcs, tsinfo):
"""
Instead of solving a large N+M eqts once, we can solve
just a tuple of size N (N is the number of arrays) N*M times.
This has a O(N^M) complexity, which is lower than solving N+M
equations when M = 2.
Doesn't work really well in practice (even when M=2) because
the large number of invocations to the equation solver.
"""
from dig_polynomials import Eqt
from itertools import product
ps = []
for i,aeterms in enumerate(product(*tsinfo.values())):
#Find rels among array elements
logger.debug('{}. Find linear eqts over {} array elems {}'
.format(i, len(aeterms), aeterms))
aeterms = [SR(1)] + list(aeterms)
solverE = Eqt(aeterms, tcs, self.xinfo)
logger.info('Select traces (note: |tcs|=|terms|={})'
.format(len(aeterms)))
ntcs_extra = 200
solverE.tcs, solverE.tcs_extra = get_traces(tcs,len(aeterms),ntcs_extra=200)
solverE.do_refine = False
solverE._go()
ps.extend(solverE.sols)
from dig_refine import Refine
rf = Refine(ps)
rf.rfilter(tcs=tcs)
ps = rf.ps
ps = [p.p for p in ps]
return ps