mlemanczyk/even-perfect-numbers-scanner
1
0 Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Revert "Merge pull request #24 from mlemanczyk/codex/modify-evenperfectbitscanner-for-residue-mode"

Browse Source 
This reverts commit 2d024cc708, reversing
changes made to b1c29ad16e.
This commit is contained in:
Marek Lemańczyk committed 2025-09-21 02:25:18 +02:00
1 parent deb25796a8
commit 8a6fc0c6a5
7 files changed
+265 -551
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

EvenPerfectBitScanner.Tests/ProgramTests.cs
+2 -91
View File
@@ -165,95 +165,6 @@ public class ProgramTests
} }
} }
[Fact]
public void Divisor_mode_sets_detailedCheck_when_divisor_is_found()
{
var useDivisorField = typeof(Program).GetField("_useDivisor", BindingFlags.NonPublic | BindingFlags.Static)!;
var divisorField = typeof(Program).GetField("_divisor", BindingFlags.NonPublic | BindingFlags.Static)!;
var testerField = typeof(Program).GetField("_divisorTester", BindingFlags.NonPublic | BindingFlags.Static)!;
var candidatesField = typeof(MersenneNumberDivisorGpuTester).GetField("_divisorCandidates", BindingFlags.NonPublic | BindingFlags.Static)!;
var primeField = typeof(Program).GetField("PrimeTesters", BindingFlags.NonPublic | BindingFlags.Static)!;
var residueField = typeof(Program).GetField("PResidue", BindingFlags.NonPublic | BindingFlags.Static)!;
var forceCpuProp = typeof(GpuContextPool).GetProperty("ForceCpu");
useDivisorField.SetValue(null, true);
divisorField.SetValue(null, UInt128.Zero);
testerField.SetValue(null, new MersenneNumberDivisorGpuTester());
primeField.SetValue(null, new ThreadLocal<PrimeTester>(() => new PrimeTester(), trackAllValues: true));
residueField.SetValue(null, new ThreadLocal<ModResidueTracker>(() => new ModResidueTracker(ResidueModel.Identity, 2UL, true), trackAllValues: true));
candidatesField.SetValue(null, new (ulong, uint)[] { (23UL, 11U) });
forceCpuProp!.SetValue(null, true);
try
{
Program.IsEvenPerfectCandidate(11UL, 32UL, out bool searched, out bool detailedCheck).Should().BeFalse();
searched.Should().BeTrue();
detailedCheck.Should().BeTrue();
}
finally
{
useDivisorField.SetValue(null, false);
testerField.SetValue(null, null);
candidatesField.SetValue(null, null);
primeField.SetValue(null, null);
residueField.SetValue(null, null);
forceCpuProp!.SetValue(null, false);
}
}
[Fact]
public void Residue_mode_sets_detailedCheck_based_on_exhaustion()
{
var useDivisorField = typeof(Program).GetField("_useDivisor", BindingFlags.NonPublic | BindingFlags.Static)!;
var residueModeField = typeof(Program).GetField("_useResidueMode", BindingFlags.NonPublic | BindingFlags.Static)!;
var mersenneField = typeof(Program).GetField("MersenneTesters", BindingFlags.NonPublic | BindingFlags.Static)!;
var primeField = typeof(Program).GetField("PrimeTesters", BindingFlags.NonPublic | BindingFlags.Static)!;
var residueField = typeof(Program).GetField("PResidue", BindingFlags.NonPublic | BindingFlags.Static)!;
var forceCpuProp = typeof(GpuContextPool).GetProperty("ForceCpu");
const ulong exponent = 7UL;
useDivisorField.SetValue(null, false);
residueModeField.SetValue(null, true);
primeField.SetValue(null, new ThreadLocal<PrimeTester>(() => new PrimeTester(), trackAllValues: true));
residueField.SetValue(null, new ThreadLocal<ModResidueTracker>(() => new ModResidueTracker(ResidueModel.Identity, exponent, true), trackAllValues: true));
forceCpuProp!.SetValue(null, true);
try
{
mersenneField.SetValue(null, new ThreadLocal<MersenneNumberTester>(() => new MersenneNumberTester(
useGpuScan: false,
useGpuOrder: false,
useResidue: true,
maxK: 0UL,
residueDivisorSets: 0UL), trackAllValues: true));
Program.IsEvenPerfectCandidate(exponent, 0UL, out bool searched, out bool detailedCheck).Should().BeTrue();
searched.Should().BeTrue();
detailedCheck.Should().BeFalse();
mersenneField.SetValue(null, new ThreadLocal<MersenneNumberTester>(() => new MersenneNumberTester(
useGpuScan: false,
useGpuOrder: false,
useResidue: true,
maxK: 4UL,
residueDivisorSets: 2UL), trackAllValues: true));
Program.IsEvenPerfectCandidate(exponent, 0UL, out searched, out detailedCheck).Should().BeTrue();
searched.Should().BeTrue();
detailedCheck.Should().BeTrue();
}
finally
{
mersenneField.SetValue(null, null);
primeField.SetValue(null, null);
residueField.SetValue(null, null);
residueModeField.SetValue(null, false);
forceCpuProp!.SetValue(null, false);
useDivisorField.SetValue(null, false);
}
}
[Fact] [Fact]
public void Divisor_mode_accepts_large_search_limit() public void Divisor_mode_accepts_large_search_limit()
{ {
@@ -335,7 +246,7 @@ public class ProgramTests
var forceCpuProp = typeof(GpuContextPool).GetProperty("ForceCpu"); var forceCpuProp = typeof(GpuContextPool).GetProperty("ForceCpu");
useDivisorField.SetValue(null, false); useDivisorField.SetValue(null, false);
mersenneField.SetValue(null, new ThreadLocal<MersenneNumberTester>(() => new MersenneNumberTester(useIncremental: true, useResidue: true, maxK: 1_000UL, residueDivisorSets: 1UL), trackAllValues: true)); mersenneField.SetValue(null, new ThreadLocal<MersenneNumberTester>(() => new MersenneNumberTester(useIncremental: true, useResidue: true, maxK: 1_000UL), trackAllValues: true));
primeField.SetValue(null, new ThreadLocal<PrimeTester>(() => new PrimeTester(), trackAllValues: true)); primeField.SetValue(null, new ThreadLocal<PrimeTester>(() => new PrimeTester(), trackAllValues: true));
residueField.SetValue(null, new ThreadLocal<ModResidueTracker>(() => new ModResidueTracker(ResidueModel.Identity, 2UL, true), trackAllValues: true)); residueField.SetValue(null, new ThreadLocal<ModResidueTracker>(() => new ModResidueTracker(ResidueModel.Identity, 2UL, true), trackAllValues: true));
forceCpuProp!.SetValue(null, false); forceCpuProp!.SetValue(null, false);
@@ -344,7 +255,7 @@ public class ProgramTests
{ {
Program.IsEvenPerfectCandidate(29UL, 0UL, out bool searched, out bool detailed).Should().BeFalse(); Program.IsEvenPerfectCandidate(29UL, 0UL, out bool searched, out bool detailed).Should().BeFalse();
searched.Should().BeTrue(); searched.Should().BeTrue();
detailed.Should().BeTrue(); detailed.Should().BeFalse();
Program.IsEvenPerfectCandidate(31UL, 0UL, out searched, out detailed).Should().BeTrue(); Program.IsEvenPerfectCandidate(31UL, 0UL, out searched, out detailed).Should().BeTrue();
searched.Should().BeTrue(); searched.Should().BeTrue();
EvenPerfectBitScanner/Program.cs
+13 -36
View File
@@ -451,17 +451,16 @@ internal static class Program
MersenneTesters = new ThreadLocal<MersenneNumberTester>(() => MersenneTesters = new ThreadLocal<MersenneNumberTester>(() =>
{ {
var tester = new MersenneNumberTester( var tester = new MersenneNumberTester(
useIncremental: !useLucas, useIncremental: !useLucas,
useOrderCache: false, useOrderCache: false,
kernelType: kernelType, kernelType: kernelType,
useModuloWorkaround: useModuloWorkaround, useModuloWorkaround: useModuloWorkaround,
useOrder: useOrder, useOrder: useOrder,
useGpuLucas: mersenneOnGpu, useGpuLucas: mersenneOnGpu,
useGpuScan: mersenneOnGpu, useGpuScan: mersenneOnGpu,
useGpuOrder: orderOnGpu, useGpuOrder: orderOnGpu,
useResidue: useResidue, useResidue: useResidue,
maxK: residueKMax, maxK: residueKMax);
residueDivisorSets: divisorCyclesSearchLimit128);
if (!useLucas) if (!useLucas)
{ {
tester.WarmUpOrders(currentP, _orderWarmupLimitOverride ?? 5_000_000UL); tester.WarmUpOrders(currentP, _orderWarmupLimitOverride ?? 5_000_000UL);
@@ -767,7 +766,7 @@ internal static class Program
Console.WriteLine(" --divisor-cycles-device=cpu|gpu device for cycles generation (default gpu)"); Console.WriteLine(" --divisor-cycles-device=cpu|gpu device for cycles generation (default gpu)");
Console.WriteLine(" --divisor-cycles-batch-size=<value> batch size for cycles generation (default 512)"); Console.WriteLine(" --divisor-cycles-batch-size=<value> batch size for cycles generation (default 512)");
Console.WriteLine(" --divisor-cycles-continue continue divisor cycles generation"); Console.WriteLine(" --divisor-cycles-continue continue divisor cycles generation");
Console.WriteLine(" --divisor-cycles-limit=<value> cycle search iterations when --mersenne=divisor or residue (UInt128 supported; residue path honors full range)"); Console.WriteLine(" --divisor-cycles-limit=<value> cycle search iterations when --mersenne=divisor");
Console.WriteLine(" --use-order test primality via q order"); Console.WriteLine(" --use-order test primality via q order");
Console.WriteLine(" --workaround-mod avoid '%' operator on the GPU"); Console.WriteLine(" --workaround-mod avoid '%' operator on the GPU");
@@ -1127,30 +1126,8 @@ internal static class Program
return _divisorTester!.IsPrime(p, _divisor, divisorCyclesSearchLimit, out detailedCheck); return _divisorTester!.IsPrime(p, _divisor, divisorCyclesSearchLimit, out detailedCheck);
} }
bool mersennePrime = MersenneTesters.Value!.IsMersennePrime(p, out bool residueExhausted); detailedCheck = MersenneTesters.Value!.IsMersennePrime(p);
detailedCheck = residueExhausted; return detailedCheck;
return mersennePrime;
}
[MethodImpl(MethodImplOptions.AggressiveInlining)]
internal static UInt128 ResolveResidueMaxK(
bool useResidue,
bool residueMaxExplicit,
bool divisorLimitExplicit,
UInt128 residueMaxK,
UInt128 divisorLimit)
{
if (!useResidue)
{
return residueMaxK;
}
if (residueMaxExplicit || !divisorLimitExplicit)
{
return residueMaxK;
}
return divisorLimit;
} }
// Use ModResidueTracker with a small set of primes to pre-filter composite p. // Use ModResidueTracker with a small set of primes to pre-filter composite p.
PerfectNumbers.Core.Tests/MersenneNumberResidueCpuTesterTests.cs
+1 -11
View File
@@ -69,18 +69,8 @@ public class MersenneNumberResidueCpuTesterTests
bool expectedPrime) bool expectedPrime)
{ {
bool isPrime = true; bool isPrime = true;
bool exhausted = false; tester.Scan(exponent, (UInt128)exponent << 1, LastDigitIsSeven(exponent), (UInt128)maxK, ref isPrime);
tester.Scan(
exponent,
(UInt128)exponent << 1,
LastDigitIsSeven(exponent),
(UInt128)perSetLimit,
(UInt128)setCount,
(UInt128)overallLimit,
ref isPrime,
ref exhausted);
isPrime.Should().Be(expectedPrime); isPrime.Should().Be(expectedPrime);
exhausted.Should().BeTrue();
} }
} }
PerfectNumbers.Core.Tests/MersenneNumberResidueGpuTesterTests.cs
+1 -11
View File
@@ -288,18 +288,8 @@ public class MersenneNumberResidueGpuTesterTests
bool expectedPrime) bool expectedPrime)
{ {
bool isPrime = true; bool isPrime = true;
bool exhausted = false; tester.Scan(exponent, (UInt128)exponent << 1, LastDigitIsSeven(exponent), (UInt128)maxK, ref isPrime);
tester.Scan(
exponent,
(UInt128)exponent << 1,
LastDigitIsSeven(exponent),
(UInt128)perSetLimit,
(UInt128)setCount,
(UInt128)overallLimit,
ref isPrime,
ref exhausted);
isPrime.Should().Be(expectedPrime); isPrime.Should().Be(expectedPrime);
exhausted.Should().BeTrue();
} }
} }
PerfectNumbers.Core/Cpu/MersenneNumberResidueCpuTester.cs
+136 -166
View File
@@ -5,32 +5,25 @@ public class MersenneNumberResidueCpuTester
private ModResidueTracker? _mersenneResidueTracker; private ModResidueTracker? _mersenneResidueTracker;
// CPU residue variant using tracker + unrolled residue updates (no method calls in the loop). // CPU residue variant using tracker + unrolled residue updates (no method calls in the loop).
public void Scan( public void Scan(ulong exponent, UInt128 twoP, bool lastIsSeven, UInt128 maxK, ref bool isPrime)
ulong exponent, {
UInt128 twoP, // Initialize/update Mersenne residue tracker and start a merge walk for ascending q divisors
bool lastIsSeven,
UInt128 perSetLimit,
UInt128 setCount,
UInt128 overallLimit,
ref bool isPrime,
ref bool divisorsExhausted)
{
if (setCount == UInt128.Zero || perSetLimit == UInt128.Zero || overallLimit == UInt128.Zero)
{
return;
}
// Initialize/update Mersenne residue tracker and start a merge walk for ascending q divisors _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);
_mersenneResidueTracker ??= new(ResidueModel.Mersenne); // Start at k = 1: q0 = 2*p*1 + 1
// For small divisors q <= 4,000,000 use cached cycle length to fast-path checks UInt128 k = UInt128.One;
// and avoid per-q powmods when possible when scanning lanes. UInt128 q = twoP + UInt128.One;
_mersenneResidueTracker.BeginMerge(exponent); // Current residues
q.Mod10_8_5_3(out ulong r10, out ulong r8, out ulong r5, out ulong r3);
ulong step10 = (ulong)(twoP % 10UL) % 10UL; // Step residues for q += 2*p
ulong step8 = (ulong)(twoP % 8UL); ulong step10 = (ulong)(twoP % 10UL) % 10UL;
ulong step3 = (ulong)(twoP % 3UL); ulong step8 = (ulong)(twoP % 8UL);
ulong step5 = (ulong)(twoP % 5UL); ulong step3 = (ulong)(twoP % 3UL);
ulong step5 = (ulong)(twoP % 5UL);
// Allowed last-digit sets for q depending on last digit of M_p // Allowed last-digit sets for q depending on last digit of M_p
// lastIsSeven == true => allow {7,9} // lastIsSeven == true => allow {7,9}
@@ -38,162 +31,139 @@ public class MersenneNumberResidueCpuTester
int allowMask = lastIsSeven ? CpuConstants.LastSevenMask10 : CpuConstants.LastOneMask10; int allowMask = lastIsSeven ? CpuConstants.LastSevenMask10 : CpuConstants.LastOneMask10;
// Predeclare temps to avoid redeclaration overhead in the loop // Predeclare temps to avoid redeclaration overhead in the loop
UInt128 remaining; UInt128 remaining;
int i, lanes; int i, lanes;
// per-lane residues // 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; 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 // q lanes
UInt128 twoP2 = twoP + twoP, UInt128 twoP2 = twoP + twoP,
twoP3 = twoP2 + twoP, twoP3 = twoP2 + twoP,
twoP4 = twoP3 + twoP, twoP4 = twoP3 + twoP,
qIncremental; qIncremental;
bool localIsPrime, isDivider; bool localIsPrime, isDivider;
ModResidueTracker tracker = _mersenneResidueTracker!; maxK++;
UInt128 limitInclusive = overallLimit + UInt128.One;
for (UInt128 setIndex = UInt128.Zero; setIndex < setCount && Volatile.Read(ref isPrime); setIndex++)
{
UInt128 setOffset = perSetLimit * setIndex;
UInt128 k = setOffset + UInt128.One;
if (k >= limitInclusive)
{
break;
}
UInt128 setLimitExclusive = setOffset + perSetLimit + UInt128.One; ModResidueTracker tracker = _mersenneResidueTracker!;
if (setLimitExclusive > limitInclusive) while (k < maxK && (localIsPrime = Volatile.Read(ref isPrime)))
{ {
setLimitExclusive = limitInclusive; remaining = maxK - k;
} if (remaining == UInt128.Zero)
{
break;
}
UInt128 q = twoP * k + UInt128.One; lanes = remaining > 4UL ? 4 : (int)remaining;
q.Mod10_8_5_3(out ulong r10, out ulong r8, out ulong r5, out ulong r3);
while (k < setLimitExclusive && (localIsPrime = Volatile.Read(ref isPrime))) // Compute lane residues incrementally (avoid multiplications, minimize reductions)
{ // r10
remaining = setLimitExclusive - k; r10_0 = r10;
if (remaining == UInt128.Zero) 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;
break; r10_3 = r10_2 + step10; if (r10_3 >= 20UL) r10_3 -= 20UL; if (r10_3 >= 10UL) r10_3 -= 10UL;
}
lanes = remaining > 4UL ? 4 : (int)remaining; // r8
if (lanes <= 0) r8_0 = r8 & 7UL;
{ r8_1 = (r8_0 + step8) & 7UL;
break; r8_2 = (r8_1 + step8) & 7UL;
} r8_3 = (r8_2 + step8) & 7UL;
// Compute lane residues incrementally (avoid multiplications, minimize reductions) // r3 (pattern 0,1,2,0)
// r10 r3_0 = r3; if (r3_0 >= 3UL) r3_0 -= 3UL;
r10_0 = r10; r3_1 = r3_0 + step3; if (r3_1 >= 3UL) r3_1 -= 3UL;
r10_1 = r10_0 + step10; if (r10_1 >= 20UL) r10_1 -= 20UL; if (r10_1 >= 10UL) r10_1 -= 10UL; r3_2 = r3_1 + step3; if (r3_2 >= 3UL) r3_2 -= 3UL;
r10_2 = r10_1 + step10; if (r10_2 >= 20UL) r10_2 -= 20UL; if (r10_2 >= 10UL) r10_2 -= 10UL; r3_3 = r3; if (r3_3 >= 3UL) r3_3 -= 3UL;
r10_3 = r10_2 + step10; if (r10_3 >= 20UL) r10_3 -= 20UL; if (r10_3 >= 10UL) r10_3 -= 10UL;
// r8 // r5 (pattern 0,1,2,3)
r8_0 = r8 & 7UL; r5_0 = r5; if (r5_0 >= 5UL) r5_0 -= 5UL;
r8_1 = (r8_0 + step8) & 7UL; r5_1 = r5_0 + step5; if (r5_1 >= 10UL) r5_1 -= 10UL; if (r5_1 >= 5UL) r5_1 -= 5UL;
r8_2 = (r8_1 + step8) & 7UL; r5_2 = r5_1 + step5; if (r5_2 >= 10UL) r5_2 -= 10UL; if (r5_2 >= 5UL) r5_2 -= 5UL;
r8_3 = (r8_2 + step8) & 7UL; r5_3 = r5_2 + step5; if (r5_3 >= 10UL) r5_3 -= 10UL; if (r5_3 >= 5UL) r5_3 -= 5UL;
// r3 (pattern 0,1,2,0) // lane 0
r3_0 = r3; if (r3_0 >= 3UL) r3_0 -= 3UL; if (lanes >= 1 &&
r3_1 = r3_0 + step3; if (r3_1 >= 3UL) r3_1 -= 3UL; (((allowMask >> (int)r10_0) & 1) != 0 && (r8_0 == 1UL || r8_0 == 7UL) && r3_0 != 0UL && r5_0 != 0UL))
r3_2 = r3_1 + step3; if (r3_2 >= 3UL) r3_2 -= 3UL; {
r3_3 = r3; if (r3_3 >= 3UL) r3_3 -= 3UL; // 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;
}
}
// r5 (pattern 0,1,2,3) // lane 1
r5_0 = r5; if (r5_0 >= 5UL) r5_0 -= 5UL; qIncremental = q + twoP;
r5_1 = r5_0 + step5; if (r5_1 >= 10UL) r5_1 -= 10UL; if (r5_1 >= 5UL) r5_1 -= 5UL; if (lanes >= 2 &&
r5_2 = r5_1 + step5; if (r5_2 >= 10UL) r5_2 -= 10UL; if (r5_2 >= 5UL) r5_2 -= 5UL; ((allowMask >> (int)r10_1) & 1) != 0 && (r8_1 == 1UL || r8_1 == 7UL) && r3_1 != 0UL && r5_1 != 0UL &&
r5_3 = r5_2 + step5; if (r5_3 >= 10UL) r5_3 -= 10UL; if (r5_3 >= 5UL) r5_3 -= 5UL; localIsPrime)
{
if (localIsPrime && tracker.MergeOrAppend(exponent, qIncremental, out isDivider) && isDivider &&
(qIncremental <= ulong.MaxValue ? ((ulong)qIncremental).IsPrimeCandidate() : qIncremental.IsPrimeCandidate()))
{
localIsPrime = false;
}
}
// lane 0 // lane 2
if (lanes >= 1 && qIncremental += twoP;
(((allowMask >> (int)r10_0) & 1) != 0 && (r8_0 == 1UL || r8_0 == 7UL) && r3_0 != 0UL && r5_0 != 0UL)) if (lanes >= 3 &&
{ ((allowMask >> (int)r10_2) & 1) != 0 && (r8_2 == 1UL || r8_2 == 7UL) && r3_2 != 0UL && r5_2 != 0UL &&
if (localIsPrime && tracker.MergeOrAppend(exponent, q, out isDivider) && isDivider && localIsPrime)
(q <= ulong.MaxValue ? ((ulong)q).IsPrimeCandidate() : q.IsPrimeCandidate())) {
{ if (localIsPrime && tracker.MergeOrAppend(exponent, qIncremental, out isDivider) && isDivider &&
localIsPrime = false; (qIncremental <= ulong.MaxValue ? ((ulong)qIncremental).IsPrimeCandidate() : qIncremental.IsPrimeCandidate()))
} {
} localIsPrime = false;
}
}
// lane 1 // lane 3
qIncremental = q + twoP; qIncremental += twoP;
if (lanes >= 2 && if (lanes >= 4 &&
((allowMask >> (int)r10_1) & 1) != 0 && (r8_1 == 1UL || r8_1 == 7UL) && r3_1 != 0UL && r5_1 != 0UL && ((allowMask >> (int)r10_3) & 1) != 0 && (r8_3 == 1UL || r8_3 == 7UL) && r3_3 != 0UL && r5_3 != 0UL &&
localIsPrime) localIsPrime)
{ {
if (localIsPrime && tracker.MergeOrAppend(exponent, qIncremental, out isDivider) && isDivider && if (localIsPrime && tracker.MergeOrAppend(exponent, qIncremental, out isDivider) && isDivider &&
(qIncremental <= ulong.MaxValue ? ((ulong)qIncremental).IsPrimeCandidate() : qIncremental.IsPrimeCandidate())) (qIncremental <= ulong.MaxValue ? ((ulong)qIncremental).IsPrimeCandidate() : qIncremental.IsPrimeCandidate()))
{ {
localIsPrime = false; localIsPrime = false;
} }
} }
// lane 2 if (!localIsPrime)
qIncremental += twoP; {
if (lanes >= 3 && Volatile.Write(ref isPrime, false);
((allowMask >> (int)r10_2) & 1) != 0 && (r8_2 == 1UL || r8_2 == 7UL) && r3_2 != 0UL && r5_2 != 0UL && return;
localIsPrime) }
{ else if (!Volatile.Read(ref isPrime))
if (localIsPrime && tracker.MergeOrAppend(exponent, qIncremental, out isDivider) && isDivider && {
(qIncremental <= ulong.MaxValue ? ((ulong)qIncremental).IsPrimeCandidate() : qIncremental.IsPrimeCandidate())) return;
{ }
localIsPrime = false;
}
}
// lane 3 // advance base by processed lanes
qIncremental += twoP; k += (UInt128)lanes;
if (lanes >= 4 && q += lanes switch
((allowMask >> (int)r10_3) & 1) != 0 && (r8_3 == 1UL || r8_3 == 7UL) && r3_3 != 0UL && r5_3 != 0UL && {
localIsPrime) 0 => 0UL,
{ 1 => twoP,
if (localIsPrime && tracker.MergeOrAppend(exponent, qIncremental, out isDivider) && isDivider && 2 => twoP2,
(qIncremental <= ulong.MaxValue ? ((ulong)qIncremental).IsPrimeCandidate() : qIncremental.IsPrimeCandidate())) 3 => twoP3,
{ 4 => twoP4,
localIsPrime = false; _ => throw new ArgumentException("Unsupproted value", nameof(twoP))
} };
}
if (!localIsPrime) // advance residues by 'lanes' steps (consistent with base advance)
{ for (i = 0; i < lanes; i++)
Volatile.Write(ref isPrime, false); {
divisorsExhausted = true; r10 += step10; if (r10 >= 20UL) r10 -= 20UL; if (r10 >= 10UL) r10 -= 10UL;
return; r8 = (r8 + step8) & 7UL;
} r3 += step3; if (r3 >= 3UL) r3 -= 3UL;
else if (!Volatile.Read(ref isPrime)) r5 += step5; if (r5 >= 10UL) r5 -= 10UL; if (r5 >= 5UL) r5 -= 5UL;
{ }
divisorsExhausted = true; }
return;
}
// advance base by processed lanes return;
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;
}
}
}
} }
PerfectNumbers.Core/Gpu/MersenneNumberResidueGpuTester.cs
+50 -86
View File
@@ -12,36 +12,22 @@ public class MersenneNumberResidueGpuTester(bool useGpuOrder)
// via GpuKernelPool to avoid host round-trips. // via GpuKernelPool to avoid host round-trips.
// GPU residue variant: check 2^p % q == 1 for q = 2*p*k + 1. // GPU residue variant: check 2^p % q == 1 for q = 2*p*k + 1.
public void Scan( public void Scan(ulong exponent, UInt128 twoP, bool lastIsSeven, UInt128 maxK, ref bool isPrime)
ulong exponent, {
UInt128 twoP, var gpuLease = GpuKernelPool.GetKernel(_useGpuOrder);
bool lastIsSeven,
UInt128 perSetLimit,
UInt128 setCount,
UInt128 overallLimit,
ref bool isPrime,
ref bool divisorsExhausted)
{
if (setCount == UInt128.Zero || perSetLimit == UInt128.Zero || overallLimit == UInt128.Zero)
{
return;
}
var gpuLease = GpuKernelPool.GetKernel(_useGpuOrder);
var execution = gpuLease.EnterExecutionScope();
var accelerator = gpuLease.Accelerator; var accelerator = gpuLease.Accelerator;
var stream = gpuLease.Stream; // Ensure device has small cycles and primes tables for in-kernel lookup
// Ensure device has small cycles and primes tables for in-kernel lookup var smallCyclesView = GpuKernelPool.EnsureSmallCyclesOnDevice(accelerator);
var smallCyclesView = GpuKernelPool.EnsureSmallCyclesOnDevice(accelerator); ResiduePrimeViews primeViews = GpuKernelPool.EnsureSmallPrimesOnDevice(accelerator);
ResiduePrimeViews primeViews = GpuKernelPool.EnsureSmallPrimesOnDevice(accelerator); int batchSize = GpuConstants.ScanBatchSize;
int batchSize = GpuConstants.ScanBatchSize; UInt128 kStart = 1UL;
byte last = lastIsSeven ? (byte)1 : (byte)0; byte last = lastIsSeven ? (byte)1 : (byte)0;
var kernel = gpuLease.Pow2ModKernel; var kernel = gpuLease.Pow2ModKernel;
ulong step10 = ((exponent % 10UL) << 1) % 10UL; ulong step10 = ((exponent % 10UL) << 1) % 10UL;
ulong step8 = ((exponent & 7UL) << 1) & 7UL; ulong step8 = ((exponent & 7UL) << 1) & 7UL;
ulong step3 = ((exponent % 3UL) << 1) % 3UL; ulong step3 = ((exponent % 3UL) << 1) % 3UL;
ulong step5 = ((exponent % 5UL) << 1) % 5UL; ulong step5 = ((exponent % 5UL) << 1) % 5UL;
GpuUInt128 twoPGpu = (GpuUInt128)twoP; GpuUInt128 twoPGpu = (GpuUInt128)twoP;
var orderBuffer = accelerator.Allocate1D<ulong>(batchSize); var orderBuffer = accelerator.Allocate1D<ulong>(batchSize);
// Device memory returned by Allocate1D is not guaranteed to be zeroed. // Device memory returned by Allocate1D is not guaranteed to be zeroed.
@@ -52,79 +38,57 @@ public class MersenneNumberResidueGpuTester(bool useGpuOrder)
// testers such as the order kernels. // testers such as the order kernels.
orderBuffer.MemSetToZero(); orderBuffer.MemSetToZero();
ulong[] orderArray = ArrayPool<ulong>.Shared.Rent(batchSize); ulong[] orderArray = ArrayPool<ulong>.Shared.Rent(batchSize);
UInt128 batchSize128 = (UInt128)batchSize; UInt128 remaining;
UInt128 limitInclusive = overallLimit == UInt128.MaxValue int currentSize;
? overallLimit int i;
: checked(overallLimit + UInt128.One); UInt128 batchSize128 = (UInt128)batchSize, q;
Span<ulong> orders = orderArray.AsSpan(0, batchSize); Span<ulong> orders = orderArray.AsSpan(0, batchSize);
try try
{ {
for (UInt128 setIndex = UInt128.Zero; setIndex < setCount && Volatile.Read(ref isPrime); setIndex++) while (kStart < maxK && Volatile.Read(ref isPrime))
{ {
UInt128 setOffset = checked(perSetLimit * setIndex); remaining = maxK - kStart;
UInt128 setStart = checked(setOffset + UInt128.One); if (remaining > batchSize128)
if (setStart >= limitInclusive) {
currentSize = batchSize;
}
else
{
currentSize = (int)remaining;
orders = orderArray.AsSpan(0, currentSize);
}
q = twoP * kStart + UInt128.One;
q.Mod10_8_5_3(out ulong q0m10, out ulong q0m8, out ulong q0m5, out ulong q0m3);
var ra = new ResidueAutomatonArgs(q0m10, step10, q0m8, step8, q0m3, step3, q0m5, step5);
kernel(currentSize, exponent, twoPGpu, (GpuUInt128)kStart, last, 0UL,
ra, orderBuffer.View, smallCyclesView, primeViews.LastOne, primeViews.LastSeven, primeViews.LastOnePow2, primeViews.LastSevenPow2);
accelerator.Synchronize();
orderBuffer.View.CopyToCPU(ref MemoryMarshal.GetReference(orders), currentSize);
if (!Volatile.Read(ref isPrime))
{ {
break; break;
} }
UInt128 setLimitExclusive = checked(setOffset + perSetLimit); for (i = 0; i < currentSize; i++)
setLimitExclusive = setLimitExclusive >= limitInclusive
? limitInclusive
: checked(setLimitExclusive + UInt128.One);
if (setLimitExclusive > limitInclusive)
{ {
setLimitExclusive = limitInclusive; if (orders[i] != 0UL)
}
UInt128 kStart = setStart;
while (kStart < setLimitExclusive && Volatile.Read(ref isPrime))
{
UInt128 remaining = checked(setLimitExclusive - kStart);
int currentSize = remaining > batchSize128 ? batchSize : (int)remaining;
if (currentSize <= 0)
{ {
Volatile.Write(ref isPrime, false);
break; break;
} }
orders = orderArray.AsSpan(0, currentSize);
UInt128 q = checked(twoP * kStart);
q = checked(q + UInt128.One);
q.Mod10_8_5_3(out ulong q0m10, out ulong q0m8, out ulong q0m5, out ulong q0m3);
var ra = new ResidueAutomatonArgs(q0m10, step10, q0m8, step8, q0m3, step3, q0m5, step5);
kernel(stream, currentSize, exponent, twoPGpu, (GpuUInt128)kStart, last, 0UL,
ra, orderBuffer.View, smallCyclesView, primeViews.LastOne, primeViews.LastSeven, primeViews.LastOnePow2, primeViews.LastSevenPow2);
stream.Synchronize();
orderBuffer.View.CopyToCPU(ref MemoryMarshal.GetReference(orders), currentSize);
stream.Synchronize();
if (!Volatile.Read(ref isPrime))
{
break;
}
for (int i = 0; i < currentSize; i++)
{
if (orders[i] != 0UL)
{
Volatile.Write(ref isPrime, false);
break;
}
}
kStart = checked(kStart + (UInt128)currentSize);
} }
kStart += batchSize128;
} }
} }
finally finally
{ {
ArrayPool<ulong>.Shared.Return(orderArray); ArrayPool<ulong>.Shared.Return(orderArray);
orderBuffer.Dispose(); orderBuffer.Dispose();
execution.Dispose(); gpuLease.Dispose();
gpuLease.Dispose(); }
} }
divisorsExhausted = true;
}
} }
PerfectNumbers.Core/MersenneNumberTester.cs
+62 -150
View File
@@ -16,28 +16,26 @@ public enum GpuKernelType
} }
public sealed class MersenneNumberTester( public sealed class MersenneNumberTester(
bool useIncremental = true, bool useIncremental = true,
bool useOrderCache = false, bool useOrderCache = false,
GpuKernelType kernelType = GpuKernelType.Incremental, GpuKernelType kernelType = GpuKernelType.Incremental,
bool useModuloWorkaround = false, bool useModuloWorkaround = false,
bool useOrder = false, bool useOrder = false,
bool useGpuLucas = true, bool useGpuLucas = true,
bool useGpuScan = true, bool useGpuScan = true,
bool useGpuOrder = true, bool useGpuOrder = true,
bool useResidue = true, bool useResidue = true,
UInt128? maxK = null, ulong maxK = 5_000_000UL)
UInt128? residueDivisorSets = null)
{ {
private readonly bool _useResidue = useResidue; private readonly bool _useResidue = useResidue;
private readonly bool _useIncremental = useIncremental && !useResidue; private readonly bool _useIncremental = useIncremental && !useResidue;
private readonly UInt128 _maxK = maxK ?? (UInt128)5_000_000UL; private readonly ulong _maxK = maxK;
private readonly UInt128 _residueDivisorSetCount = residueDivisorSets ?? (UInt128)PerfectNumberConstants.ExtraDivisorCycleSearchLimit;
private readonly GpuKernelType _kernelType = kernelType; private readonly GpuKernelType _kernelType = kernelType;
private readonly bool _useModuloWorkaround = useModuloWorkaround; private readonly bool _useModuloWorkaround = useModuloWorkaround;
private readonly bool _useGpuLucas = useGpuLucas; private readonly bool _useGpuLucas = useGpuLucas;
private readonly bool _useGpuScan = useGpuScan; // device for pow2mod/incremental scanning private readonly bool _useGpuScan = useGpuScan; // device for pow2mod/incremental scanning
private readonly bool _useGpuOrder = useGpuOrder; // device for order computations private readonly bool _useGpuOrder = useGpuOrder; // device for order computations
private static readonly ConcurrentDictionary<UInt128, ulong> OrderCache = new(); private static readonly ConcurrentDictionary<UInt128, ulong> OrderCache = new();
private readonly MersenneNumberIncrementalGpuTester _incrementalGpuTester = new(kernelType, useGpuOrder); private readonly MersenneNumberIncrementalGpuTester _incrementalGpuTester = new(kernelType, useGpuOrder);
private readonly MersenneNumberIncrementalCpuTester _incrementalCpuTester = new(kernelType); private readonly MersenneNumberIncrementalCpuTester _incrementalCpuTester = new(kernelType);
private readonly MersenneNumberOrderGpuTester _orderGpuTester = new(kernelType, useGpuOrder); private readonly MersenneNumberOrderGpuTester _orderGpuTester = new(kernelType, useGpuOrder);
@@ -50,24 +48,7 @@ public sealed class MersenneNumberTester(
// TODO: Ensure Program passes useResidue correctly and that residue-vs-incremental-vs-LL selection respects CLI flags. // TODO: Ensure Program passes useResidue correctly and that residue-vs-incremental-vs-LL selection respects CLI flags.
// Also consider injecting a shared cycles cache (MersenneDivisorCycles.Shared) to both CPU and GPU testers. // Also consider injecting a shared cycles cache (MersenneDivisorCycles.Shared) to both CPU and GPU testers.
[MethodImpl(MethodImplOptions.AggressiveInlining)] public void WarmUpOrders(ulong exponent, ulong limit = 5_000_000UL)
private static UInt128 DivideRoundUp(UInt128 value, UInt128 divisor)
{
if (divisor == UInt128.Zero)
{
return UInt128.Zero;
}
UInt128 result = value / divisor;
if (result * divisor == value)
{
return result;
}
return result + UInt128.One;
}
public void WarmUpOrders(ulong exponent, ulong limit = 5_000_000UL)
{ {
bool cacheEnabled = useOrderCache; bool cacheEnabled = useOrderCache;
if (!cacheEnabled) if (!cacheEnabled)
@@ -218,116 +199,51 @@ public sealed class MersenneNumberTester(
} }
} }
public bool IsMersennePrime(ulong exponent) => IsMersennePrime(exponent, out _); public bool IsMersennePrime(ulong exponent)
{
// Safe early rejections using known orders for tiny primes:
// 7 | M_p iff 3 | p. Avoid rejecting p == 3 where M_3 == 7 is prime.
if ((exponent % 3UL) == 0UL && exponent != 3UL)
{
return false;
}
public bool IsMersennePrime(ulong exponent, out bool divisorsExhausted) if ((exponent & 3UL) == 1UL && exponent.SharesFactorWithExponentMinusOne())
{ {
divisorsExhausted = false; return false;
}
// Safe early rejections using known orders for tiny primes: // Skip even exponents: for even p > 2, M_p is composite; here p is large.
// 7 | M_p iff 3 | p. Avoid rejecting p == 3 where M_3 == 7 is prime. if ((exponent & 1UL) == 0UL)
if ((exponent % 3UL) == 0UL && exponent != 3UL) {
{ return false;
divisorsExhausted = true; }
return false;
}
if ((exponent & 3UL) == 1UL && exponent.SharesFactorWithExponentMinusOne())
{
divisorsExhausted = true;
return false;
}
// Skip even exponents: for even p > 2, M_p is composite; here p is large.
if ((exponent & 1UL) == 0UL)
{
divisorsExhausted = true;
return false;
}
bool prePrime = true; bool prePrime = true;
UInt128 twoP = (UInt128)exponent << 1; // 2 * p UInt128 twoP = (UInt128)exponent << 1; // 2 * p
// UInt128 maxDivisor = new(ulong.MaxValue, ulong.MaxValue); // UInt128 maxDivisor = new(ulong.MaxValue, ulong.MaxValue);
UInt128 maxK = UInt128Numbers.Two.Pow(exponent) / 2 + 1; UInt128 maxK = UInt128Numbers.Two.Pow(exponent) / 2 + 1;
bool lastIsSeven = (exponent & 3UL) == 3UL; bool lastIsSeven = (exponent & 3UL) == 3UL;
// In residue mode the residue scan is the final primality check. // In residue mode the residue scan is the final primality check.
if (_useResidue) if (_useResidue)
{ {
if (_maxK < maxK) if ((UInt128)_maxK < maxK)
{ {
maxK = _maxK; maxK = _maxK;
} }
if (_maxK == UInt128.Zero || _residueDivisorSetCount == UInt128.Zero) if (_useGpuScan)
{ {
return prePrime; _residueGpuTester!.Scan(exponent, twoP, lastIsSeven, maxK, ref prePrime);
} }
else
{
_residueCpuTester!.Scan(exponent, twoP, lastIsSeven, maxK, ref prePrime);
}
UInt128 totalLimit = _maxK; return prePrime;
if (totalLimit > maxK) }
{
totalLimit = maxK;
}
if (totalLimit == UInt128.Zero)
{
return true;
}
UInt128 requestedSets = _residueDivisorSetCount;
if (_useGpuScan)
{
UInt128 maxUsefulSets = DivideRoundUp(totalLimit, (UInt128)GpuConstants.ScanBatchSize);
if (maxUsefulSets == UInt128.Zero)
{
maxUsefulSets = UInt128.One;
}
if (requestedSets == UInt128.Zero || requestedSets > maxUsefulSets)
{
requestedSets = maxUsefulSets;
}
}
if (requestedSets == UInt128.Zero)
{
requestedSets = UInt128.One;
}
UInt128 perSetLimit = DivideRoundUp(totalLimit, requestedSets);
if (perSetLimit == UInt128.Zero)
{
perSetLimit = UInt128.One;
}
UInt128 setsNeeded = DivideRoundUp(totalLimit, perSetLimit);
UInt128 effectiveSets = requestedSets;
if (setsNeeded < effectiveSets)
{
effectiveSets = setsNeeded;
}
bool scanCompleted = false;
if (_useGpuScan)
{
_residueGpuTester!.Scan(exponent, twoP, lastIsSeven, perSetLimit, effectiveSets, totalLimit, ref prePrime, ref scanCompleted);
}
else
{
_residueCpuTester!.Scan(exponent, twoP, lastIsSeven, perSetLimit, effectiveSets, totalLimit, ref prePrime, ref scanCompleted);
}
if (!prePrime)
{
divisorsExhausted = true;
return false;
}
divisorsExhausted = scanCompleted;
return true;
}
if (!_useIncremental) if (!_useIncremental)
{ {
@@ -347,15 +263,12 @@ public sealed class MersenneNumberTester(
if (!prePrime) if (!prePrime)
{ {
divisorsExhausted = true;
return false; return false;
} }
bool lucasPrime = _useGpuLucas return _useGpuLucas
? _lucasLehmerGpuTester.IsPrime(exponent, runOnGpu: true) ? _lucasLehmerGpuTester.IsPrime(exponent, runOnGpu: true)
: _lucasLehmerCpuTester.IsPrime(exponent); : _lucasLehmerCpuTester.IsPrime(exponent);
divisorsExhausted = true;
return lucasPrime;
} }
bool isPrime = true; bool isPrime = true;
@@ -383,9 +296,8 @@ public sealed class MersenneNumberTester(
} }
} }
divisorsExhausted = true; return isPrime;
return isPrime; }
}
// Limit of k to scan in the pre-Lucas prescan // Limit of k to scan in the pre-Lucas prescan
public const int PreLucasPreScanK = 5_000_000; public const int PreLucasPreScanK = 5_000_000;