Add divisor-mode detailed check coverage for factors
This commit is contained in:
Marek Lemańczyk committed
1 parent
b1c29ad16e
commit
1b28584168
7 files changed
+465
-224
No files matched your search
@@ -5,25 +5,32 @@ public class MersenneNumberResidueCpuTester
|
||||
private ModResidueTracker? _mersenneResidueTracker;
|
||||
|
||||
// CPU residue variant using tracker + unrolled residue updates (no method calls in the loop).
|
||||
public void Scan(ulong exponent, UInt128 twoP, bool lastIsSeven, UInt128 maxK, ref bool isPrime)
|
||||
{
|
||||
// Initialize/update Mersenne residue tracker and start a merge walk for ascending q divisors
|
||||
public void Scan(
|
||||
ulong exponent,
|
||||
UInt128 twoP,
|
||||
bool lastIsSeven,
|
||||
UInt128 perSetLimit,
|
||||
ulong setCount,
|
||||
UInt128 overallLimit,
|
||||
ref bool isPrime,
|
||||
ref bool divisorsExhausted)
|
||||
{
|
||||
if (setCount == 0UL || perSetLimit == UInt128.Zero || overallLimit == UInt128.Zero)
|
||||
{
|
||||
return;
|
||||
}
|
||||
|
||||
_mersenneResidueTracker ??= new(ResidueModel.Mersenne);
|
||||
// For small divisors q <= 4,000,000 use cached cycle length to fast-path checks
|
||||
// and avoid per-q powmods when possible when scanning lanes.
|
||||
_mersenneResidueTracker.BeginMerge(exponent);
|
||||
// Initialize/update Mersenne residue tracker and start a merge walk for ascending q divisors
|
||||
|
||||
// Start at k = 1: q0 = 2*p*1 + 1
|
||||
UInt128 k = UInt128.One;
|
||||
UInt128 q = twoP + UInt128.One;
|
||||
// Current residues
|
||||
q.Mod10_8_5_3(out ulong r10, out ulong r8, out ulong r5, out ulong r3);
|
||||
// Step residues for q += 2*p
|
||||
ulong step10 = (ulong)(twoP % 10UL) % 10UL;
|
||||
ulong step8 = (ulong)(twoP % 8UL);
|
||||
ulong step3 = (ulong)(twoP % 3UL);
|
||||
ulong step5 = (ulong)(twoP % 5UL);
|
||||
_mersenneResidueTracker ??= new(ResidueModel.Mersenne);
|
||||
// For small divisors q <= 4,000,000 use cached cycle length to fast-path checks
|
||||
// and avoid per-q powmods when possible when scanning lanes.
|
||||
_mersenneResidueTracker.BeginMerge(exponent);
|
||||
|
||||
ulong step10 = (ulong)(twoP % 10UL) % 10UL;
|
||||
ulong step8 = (ulong)(twoP % 8UL);
|
||||
ulong step3 = (ulong)(twoP % 3UL);
|
||||
ulong step5 = (ulong)(twoP % 5UL);
|
||||
|
||||
// Allowed last-digit sets for q depending on last digit of M_p
|
||||
// lastIsSeven == true => allow {7,9}
|
||||
@@ -31,139 +38,162 @@ public class MersenneNumberResidueCpuTester
|
||||
int allowMask = lastIsSeven ? CpuConstants.LastSevenMask10 : CpuConstants.LastOneMask10;
|
||||
|
||||
// Predeclare temps to avoid redeclaration overhead in the loop
|
||||
UInt128 remaining;
|
||||
int i, lanes;
|
||||
// per-lane residues
|
||||
ulong r10_0, r10_1, r10_2, r10_3, r8_0, r8_1, r8_2, r8_3, r3_0, r3_1, r3_2, r3_3, r5_0, r5_1, r5_2, r5_3;
|
||||
// q lanes
|
||||
UInt128 twoP2 = twoP + twoP,
|
||||
twoP3 = twoP2 + twoP,
|
||||
twoP4 = twoP3 + twoP,
|
||||
qIncremental;
|
||||
UInt128 remaining;
|
||||
int i, lanes;
|
||||
// per-lane residues
|
||||
ulong r10_0, r10_1, r10_2, r10_3, r8_0, r8_1, r8_2, r8_3, r3_0, r3_1, r3_2, r3_3, r5_0, r5_1, r5_2, r5_3;
|
||||
// q lanes
|
||||
UInt128 twoP2 = twoP + twoP,
|
||||
twoP3 = twoP2 + twoP,
|
||||
twoP4 = twoP3 + twoP,
|
||||
qIncremental;
|
||||
|
||||
bool localIsPrime, isDivider;
|
||||
bool localIsPrime, isDivider;
|
||||
|
||||
maxK++;
|
||||
ModResidueTracker tracker = _mersenneResidueTracker!;
|
||||
UInt128 limitInclusive = overallLimit + UInt128.One;
|
||||
for (ulong setIndex = 0; setIndex < setCount && Volatile.Read(ref isPrime); setIndex++)
|
||||
{
|
||||
UInt128 setOffset = perSetLimit * (UInt128)setIndex;
|
||||
UInt128 k = setOffset + UInt128.One;
|
||||
if (k >= limitInclusive)
|
||||
{
|
||||
break;
|
||||
}
|
||||
|
||||
ModResidueTracker tracker = _mersenneResidueTracker!;
|
||||
while (k < maxK && (localIsPrime = Volatile.Read(ref isPrime)))
|
||||
{
|
||||
remaining = maxK - k;
|
||||
if (remaining == UInt128.Zero)
|
||||
{
|
||||
break;
|
||||
}
|
||||
UInt128 setLimitExclusive = setOffset + perSetLimit + UInt128.One;
|
||||
if (setLimitExclusive > limitInclusive)
|
||||
{
|
||||
setLimitExclusive = limitInclusive;
|
||||
}
|
||||
|
||||
lanes = remaining > 4UL ? 4 : (int)remaining;
|
||||
UInt128 q = twoP * k + UInt128.One;
|
||||
q.Mod10_8_5_3(out ulong r10, out ulong r8, out ulong r5, out ulong r3);
|
||||
|
||||
// Compute lane residues incrementally (avoid multiplications, minimize reductions)
|
||||
// r10
|
||||
r10_0 = r10;
|
||||
r10_1 = r10_0 + step10; if (r10_1 >= 20UL) r10_1 -= 20UL; if (r10_1 >= 10UL) r10_1 -= 10UL;
|
||||
r10_2 = r10_1 + step10; if (r10_2 >= 20UL) r10_2 -= 20UL; if (r10_2 >= 10UL) r10_2 -= 10UL;
|
||||
r10_3 = r10_2 + step10; if (r10_3 >= 20UL) r10_3 -= 20UL; if (r10_3 >= 10UL) r10_3 -= 10UL;
|
||||
while (k < setLimitExclusive && (localIsPrime = Volatile.Read(ref isPrime)))
|
||||
{
|
||||
remaining = setLimitExclusive - k;
|
||||
if (remaining == UInt128.Zero)
|
||||
{
|
||||
break;
|
||||
}
|
||||
|
||||
// r8
|
||||
r8_0 = r8 & 7UL;
|
||||
r8_1 = (r8_0 + step8) & 7UL;
|
||||
r8_2 = (r8_1 + step8) & 7UL;
|
||||
r8_3 = (r8_2 + step8) & 7UL;
|
||||
lanes = remaining > 4UL ? 4 : (int)remaining;
|
||||
if (lanes <= 0)
|
||||
{
|
||||
break;
|
||||
}
|
||||
|
||||
// r3 (pattern 0,1,2,0)
|
||||
r3_0 = r3; if (r3_0 >= 3UL) r3_0 -= 3UL;
|
||||
r3_1 = r3_0 + step3; if (r3_1 >= 3UL) r3_1 -= 3UL;
|
||||
r3_2 = r3_1 + step3; if (r3_2 >= 3UL) r3_2 -= 3UL;
|
||||
r3_3 = r3; if (r3_3 >= 3UL) r3_3 -= 3UL;
|
||||
// Compute lane residues incrementally (avoid multiplications, minimize reductions)
|
||||
// r10
|
||||
r10_0 = r10;
|
||||
r10_1 = r10_0 + step10; if (r10_1 >= 20UL) r10_1 -= 20UL; if (r10_1 >= 10UL) r10_1 -= 10UL;
|
||||
r10_2 = r10_1 + step10; if (r10_2 >= 20UL) r10_2 -= 20UL; if (r10_2 >= 10UL) r10_2 -= 10UL;
|
||||
r10_3 = r10_2 + step10; if (r10_3 >= 20UL) r10_3 -= 20UL; if (r10_3 >= 10UL) r10_3 -= 10UL;
|
||||
|
||||
// r5 (pattern 0,1,2,3)
|
||||
r5_0 = r5; if (r5_0 >= 5UL) r5_0 -= 5UL;
|
||||
r5_1 = r5_0 + step5; if (r5_1 >= 10UL) r5_1 -= 10UL; if (r5_1 >= 5UL) r5_1 -= 5UL;
|
||||
r5_2 = r5_1 + step5; if (r5_2 >= 10UL) r5_2 -= 10UL; if (r5_2 >= 5UL) r5_2 -= 5UL;
|
||||
r5_3 = r5_2 + step5; if (r5_3 >= 10UL) r5_3 -= 10UL; if (r5_3 >= 5UL) r5_3 -= 5UL;
|
||||
// r8
|
||||
r8_0 = r8 & 7UL;
|
||||
r8_1 = (r8_0 + step8) & 7UL;
|
||||
r8_2 = (r8_1 + step8) & 7UL;
|
||||
r8_3 = (r8_2 + step8) & 7UL;
|
||||
|
||||
// lane 0
|
||||
if (lanes >= 1 &&
|
||||
(((allowMask >> (int)r10_0) & 1) != 0 && (r8_0 == 1UL || r8_0 == 7UL) && r3_0 != 0UL && r5_0 != 0UL))
|
||||
{
|
||||
// cycle-based quick check for small q
|
||||
if (localIsPrime && tracker.MergeOrAppend(exponent, q, out isDivider) && isDivider &&
|
||||
(q <= ulong.MaxValue ? ((ulong)q).IsPrimeCandidate() : q.IsPrimeCandidate()))
|
||||
{
|
||||
localIsPrime = false;
|
||||
}
|
||||
}
|
||||
// r3 (pattern 0,1,2,0)
|
||||
r3_0 = r3; if (r3_0 >= 3UL) r3_0 -= 3UL;
|
||||
r3_1 = r3_0 + step3; if (r3_1 >= 3UL) r3_1 -= 3UL;
|
||||
r3_2 = r3_1 + step3; if (r3_2 >= 3UL) r3_2 -= 3UL;
|
||||
r3_3 = r3; if (r3_3 >= 3UL) r3_3 -= 3UL;
|
||||
|
||||
// lane 1
|
||||
qIncremental = q + twoP;
|
||||
if (lanes >= 2 &&
|
||||
((allowMask >> (int)r10_1) & 1) != 0 && (r8_1 == 1UL || r8_1 == 7UL) && r3_1 != 0UL && r5_1 != 0UL &&
|
||||
localIsPrime)
|
||||
{
|
||||
if (localIsPrime && tracker.MergeOrAppend(exponent, qIncremental, out isDivider) && isDivider &&
|
||||
(qIncremental <= ulong.MaxValue ? ((ulong)qIncremental).IsPrimeCandidate() : qIncremental.IsPrimeCandidate()))
|
||||
{
|
||||
localIsPrime = false;
|
||||
}
|
||||
}
|
||||
// r5 (pattern 0,1,2,3)
|
||||
r5_0 = r5; if (r5_0 >= 5UL) r5_0 -= 5UL;
|
||||
r5_1 = r5_0 + step5; if (r5_1 >= 10UL) r5_1 -= 10UL; if (r5_1 >= 5UL) r5_1 -= 5UL;
|
||||
r5_2 = r5_1 + step5; if (r5_2 >= 10UL) r5_2 -= 10UL; if (r5_2 >= 5UL) r5_2 -= 5UL;
|
||||
r5_3 = r5_2 + step5; if (r5_3 >= 10UL) r5_3 -= 10UL; if (r5_3 >= 5UL) r5_3 -= 5UL;
|
||||
|
||||
// lane 2
|
||||
qIncremental += twoP;
|
||||
if (lanes >= 3 &&
|
||||
((allowMask >> (int)r10_2) & 1) != 0 && (r8_2 == 1UL || r8_2 == 7UL) && r3_2 != 0UL && r5_2 != 0UL &&
|
||||
localIsPrime)
|
||||
{
|
||||
if (localIsPrime && tracker.MergeOrAppend(exponent, qIncremental, out isDivider) && isDivider &&
|
||||
(qIncremental <= ulong.MaxValue ? ((ulong)qIncremental).IsPrimeCandidate() : qIncremental.IsPrimeCandidate()))
|
||||
{
|
||||
localIsPrime = false;
|
||||
}
|
||||
}
|
||||
// lane 0
|
||||
if (lanes >= 1 &&
|
||||
(((allowMask >> (int)r10_0) & 1) != 0 && (r8_0 == 1UL || r8_0 == 7UL) && r3_0 != 0UL && r5_0 != 0UL))
|
||||
{
|
||||
if (localIsPrime && tracker.MergeOrAppend(exponent, q, out isDivider) && isDivider &&
|
||||
(q <= ulong.MaxValue ? ((ulong)q).IsPrimeCandidate() : q.IsPrimeCandidate()))
|
||||
{
|
||||
localIsPrime = false;
|
||||
}
|
||||
}
|
||||
|
||||
// lane 3
|
||||
qIncremental += twoP;
|
||||
if (lanes >= 4 &&
|
||||
((allowMask >> (int)r10_3) & 1) != 0 && (r8_3 == 1UL || r8_3 == 7UL) && r3_3 != 0UL && r5_3 != 0UL &&
|
||||
localIsPrime)
|
||||
{
|
||||
if (localIsPrime && tracker.MergeOrAppend(exponent, qIncremental, out isDivider) && isDivider &&
|
||||
(qIncremental <= ulong.MaxValue ? ((ulong)qIncremental).IsPrimeCandidate() : qIncremental.IsPrimeCandidate()))
|
||||
{
|
||||
localIsPrime = false;
|
||||
}
|
||||
}
|
||||
// lane 1
|
||||
qIncremental = q + twoP;
|
||||
if (lanes >= 2 &&
|
||||
((allowMask >> (int)r10_1) & 1) != 0 && (r8_1 == 1UL || r8_1 == 7UL) && r3_1 != 0UL && r5_1 != 0UL &&
|
||||
localIsPrime)
|
||||
{
|
||||
if (localIsPrime && tracker.MergeOrAppend(exponent, qIncremental, out isDivider) && isDivider &&
|
||||
(qIncremental <= ulong.MaxValue ? ((ulong)qIncremental).IsPrimeCandidate() : qIncremental.IsPrimeCandidate()))
|
||||
{
|
||||
localIsPrime = false;
|
||||
}
|
||||
}
|
||||
|
||||
if (!localIsPrime)
|
||||
{
|
||||
Volatile.Write(ref isPrime, false);
|
||||
return;
|
||||
}
|
||||
else if (!Volatile.Read(ref isPrime))
|
||||
{
|
||||
return;
|
||||
}
|
||||
// lane 2
|
||||
qIncremental += twoP;
|
||||
if (lanes >= 3 &&
|
||||
((allowMask >> (int)r10_2) & 1) != 0 && (r8_2 == 1UL || r8_2 == 7UL) && r3_2 != 0UL && r5_2 != 0UL &&
|
||||
localIsPrime)
|
||||
{
|
||||
if (localIsPrime && tracker.MergeOrAppend(exponent, qIncremental, out isDivider) && isDivider &&
|
||||
(qIncremental <= ulong.MaxValue ? ((ulong)qIncremental).IsPrimeCandidate() : qIncremental.IsPrimeCandidate()))
|
||||
{
|
||||
localIsPrime = false;
|
||||
}
|
||||
}
|
||||
|
||||
// advance base by processed lanes
|
||||
k += (UInt128)lanes;
|
||||
q += lanes switch
|
||||
{
|
||||
0 => 0UL,
|
||||
1 => twoP,
|
||||
2 => twoP2,
|
||||
3 => twoP3,
|
||||
4 => twoP4,
|
||||
_ => throw new ArgumentException("Unsupproted value", nameof(twoP))
|
||||
};
|
||||
// lane 3
|
||||
qIncremental += twoP;
|
||||
if (lanes >= 4 &&
|
||||
((allowMask >> (int)r10_3) & 1) != 0 && (r8_3 == 1UL || r8_3 == 7UL) && r3_3 != 0UL && r5_3 != 0UL &&
|
||||
localIsPrime)
|
||||
{
|
||||
if (localIsPrime && tracker.MergeOrAppend(exponent, qIncremental, out isDivider) && isDivider &&
|
||||
(qIncremental <= ulong.MaxValue ? ((ulong)qIncremental).IsPrimeCandidate() : qIncremental.IsPrimeCandidate()))
|
||||
{
|
||||
localIsPrime = false;
|
||||
}
|
||||
}
|
||||
|
||||
// advance residues by 'lanes' steps (consistent with base advance)
|
||||
for (i = 0; i < lanes; i++)
|
||||
{
|
||||
r10 += step10; if (r10 >= 20UL) r10 -= 20UL; if (r10 >= 10UL) r10 -= 10UL;
|
||||
r8 = (r8 + step8) & 7UL;
|
||||
r3 += step3; if (r3 >= 3UL) r3 -= 3UL;
|
||||
r5 += step5; if (r5 >= 10UL) r5 -= 10UL; if (r5 >= 5UL) r5 -= 5UL;
|
||||
}
|
||||
}
|
||||
if (!localIsPrime)
|
||||
{
|
||||
Volatile.Write(ref isPrime, false);
|
||||
divisorsExhausted = true;
|
||||
return;
|
||||
}
|
||||
else if (!Volatile.Read(ref isPrime))
|
||||
{
|
||||
divisorsExhausted = true;
|
||||
return;
|
||||
}
|
||||
|
||||
return;
|
||||
}
|
||||
// advance base by processed lanes
|
||||
k += (UInt128)lanes;
|
||||
q += lanes switch
|
||||
{
|
||||
0 => 0UL,
|
||||
1 => twoP,
|
||||
2 => twoP2,
|
||||
3 => twoP3,
|
||||
4 => twoP4,
|
||||
_ => throw new ArgumentException("Unsupproted value", nameof(twoP))
|
||||
};
|
||||
|
||||
// advance residues by 'lanes' steps (consistent with base advance)
|
||||
for (i = 0; i < lanes; i++)
|
||||
{
|
||||
r10 += step10; if (r10 >= 20UL) r10 -= 20UL; if (r10 >= 10UL) r10 -= 10UL;
|
||||
r8 = (r8 + step8) & 7UL;
|
||||
r3 += step3; if (r3 >= 3UL) r3 -= 3UL;
|
||||
r5 += step5; if (r5 >= 10UL) r5 -= 10UL; if (r5 >= 5UL) r5 -= 5UL;
|
||||
}
|
||||
|
||||
divisorsExhausted = true;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
Reference in new issue
Block a user