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

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
+48 -334

No files matched your search

+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]
public void Divisor_mode_accepts_large_search_limit()
{
@@ -335,7 +246,7 @@ public class ProgramTests
var forceCpuProp = typeof(GpuContextPool).GetProperty("ForceCpu");
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));
residueField.SetValue(null, new ThreadLocal<ModResidueTracker>(() => new ModResidueTracker(ResidueModel.Identity, 2UL, true), trackAllValues: true));
forceCpuProp!.SetValue(null, false);
@@ -344,7 +255,7 @@ public class ProgramTests
{
Program.IsEvenPerfectCandidate(29UL, 0UL, out bool searched, out bool detailed).Should().BeFalse();
searched.Should().BeTrue();
detailed.Should().BeTrue();
detailed.Should().BeFalse();
Program.IsEvenPerfectCandidate(31UL, 0UL, out searched, out detailed).Should().BeTrue();
searched.Should().BeTrue();
+4 -27
View File
@@ -460,8 +460,7 @@ internal static class Program
useGpuScan: mersenneOnGpu,
useGpuOrder: orderOnGpu,
useResidue: useResidue,
maxK: residueKMax,
residueDivisorSets: divisorCyclesSearchLimit128);
maxK: residueKMax);
if (!useLucas)
{
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-batch-size=<value> batch size for cycles generation (default 512)");
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(" --workaround-mod avoid '%' operator on the GPU");
@@ -1127,30 +1126,8 @@ internal static class Program
return _divisorTester!.IsPrime(p, _divisor, divisorCyclesSearchLimit, out detailedCheck);
}
bool mersennePrime = MersenneTesters.Value!.IsMersennePrime(p, out bool residueExhausted);
detailedCheck = residueExhausted;
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;
detailedCheck = MersenneTesters.Value!.IsMersennePrime(p);
return detailedCheck;
}
// Use ModResidueTracker with a small set of primes to pre-filter composite p.
@@ -69,18 +69,8 @@ public class MersenneNumberResidueCpuTesterTests
bool expectedPrime)
{
bool isPrime = true;
bool exhausted = false;
tester.Scan(
exponent,
(UInt128)exponent << 1,
LastDigitIsSeven(exponent),
(UInt128)perSetLimit,
(UInt128)setCount,
(UInt128)overallLimit,
ref isPrime,
ref exhausted);
tester.Scan(exponent, (UInt128)exponent << 1, LastDigitIsSeven(exponent), (UInt128)maxK, ref isPrime);
isPrime.Should().Be(expectedPrime);
exhausted.Should().BeTrue();
}
}
@@ -288,18 +288,8 @@ public class MersenneNumberResidueGpuTesterTests
bool expectedPrime)
{
bool isPrime = true;
bool exhausted = false;
tester.Scan(
exponent,
(UInt128)exponent << 1,
LastDigitIsSeven(exponent),
(UInt128)perSetLimit,
(UInt128)setCount,
(UInt128)overallLimit,
ref isPrime,
ref exhausted);
tester.Scan(exponent, (UInt128)exponent << 1, LastDigitIsSeven(exponent), (UInt128)maxK, ref isPrime);
isPrime.Should().Be(expectedPrime);
exhausted.Should().BeTrue();
}
}
@@ -5,21 +5,8 @@ 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 perSetLimit,
UInt128 setCount,
UInt128 overallLimit,
ref bool isPrime,
ref bool divisorsExhausted)
public void Scan(ulong exponent, UInt128 twoP, bool lastIsSeven, UInt128 maxK, ref bool isPrime)
{
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);
@@ -27,6 +14,12 @@ public class MersenneNumberResidueCpuTester
// and avoid per-q powmods when possible when scanning lanes.
_mersenneResidueTracker.BeginMerge(exponent);
// 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);
@@ -50,39 +43,18 @@ public class MersenneNumberResidueCpuTester
bool localIsPrime, isDivider;
maxK++;
ModResidueTracker tracker = _mersenneResidueTracker!;
UInt128 limitInclusive = overallLimit + UInt128.One;
for (UInt128 setIndex = UInt128.Zero; setIndex < setCount && Volatile.Read(ref isPrime); setIndex++)
while (k < maxK && (localIsPrime = Volatile.Read(ref isPrime)))
{
UInt128 setOffset = perSetLimit * setIndex;
UInt128 k = setOffset + UInt128.One;
if (k >= limitInclusive)
{
break;
}
UInt128 setLimitExclusive = setOffset + perSetLimit + UInt128.One;
if (setLimitExclusive > limitInclusive)
{
setLimitExclusive = limitInclusive;
}
UInt128 q = twoP * k + UInt128.One;
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)))
{
remaining = setLimitExclusive - k;
remaining = maxK - k;
if (remaining == UInt128.Zero)
{
break;
}
lanes = remaining > 4UL ? 4 : (int)remaining;
if (lanes <= 0)
{
break;
}
// Compute lane residues incrementally (avoid multiplications, minimize reductions)
// r10
@@ -113,6 +85,7 @@ public class MersenneNumberResidueCpuTester
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()))
{
@@ -162,12 +135,10 @@ public class MersenneNumberResidueCpuTester
if (!localIsPrime)
{
Volatile.Write(ref isPrime, false);
divisorsExhausted = true;
return;
}
else if (!Volatile.Read(ref isPrime))
{
divisorsExhausted = true;
return;
}
@@ -191,9 +162,8 @@ public class MersenneNumberResidueCpuTester
r3 += step3; if (r3 >= 3UL) r3 -= 3UL;
r5 += step5; if (r5 >= 10UL) r5 -= 10UL; if (r5 >= 5UL) r5 -= 5UL;
}
}
divisorsExhausted = true;
}
}
return;
}
}
@@ -12,29 +12,15 @@ public class MersenneNumberResidueGpuTester(bool useGpuOrder)
// via GpuKernelPool to avoid host round-trips.
// GPU residue variant: check 2^p % q == 1 for q = 2*p*k + 1.
public void Scan(
ulong exponent,
UInt128 twoP,
bool lastIsSeven,
UInt128 perSetLimit,
UInt128 setCount,
UInt128 overallLimit,
ref bool isPrime,
ref bool divisorsExhausted)
public void Scan(ulong exponent, UInt128 twoP, bool lastIsSeven, UInt128 maxK, ref bool isPrime)
{
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 stream = gpuLease.Stream;
// Ensure device has small cycles and primes tables for in-kernel lookup
var smallCyclesView = GpuKernelPool.EnsureSmallCyclesOnDevice(accelerator);
ResiduePrimeViews primeViews = GpuKernelPool.EnsureSmallPrimesOnDevice(accelerator);
int batchSize = GpuConstants.ScanBatchSize;
UInt128 kStart = 1UL;
byte last = lastIsSeven ? (byte)1 : (byte)0;
var kernel = gpuLease.Pow2ModKernel;
ulong step10 = ((exponent % 10UL) << 1) % 10UL;
@@ -52,59 +38,41 @@ public class MersenneNumberResidueGpuTester(bool useGpuOrder)
// testers such as the order kernels.
orderBuffer.MemSetToZero();
ulong[] orderArray = ArrayPool<ulong>.Shared.Rent(batchSize);
UInt128 batchSize128 = (UInt128)batchSize;
UInt128 limitInclusive = overallLimit == UInt128.MaxValue
? overallLimit
: checked(overallLimit + UInt128.One);
UInt128 remaining;
int currentSize;
int i;
UInt128 batchSize128 = (UInt128)batchSize, q;
Span<ulong> orders = orderArray.AsSpan(0, batchSize);
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);
UInt128 setStart = checked(setOffset + UInt128.One);
if (setStart >= limitInclusive)
remaining = maxK - kStart;
if (remaining > batchSize128)
{
break;
currentSize = batchSize;
}
UInt128 setLimitExclusive = checked(setOffset + perSetLimit);
setLimitExclusive = setLimitExclusive >= limitInclusive
? limitInclusive
: checked(setLimitExclusive + UInt128.One);
if (setLimitExclusive > limitInclusive)
else
{
setLimitExclusive = limitInclusive;
}
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)
{
break;
}
currentSize = (int)remaining;
orders = orderArray.AsSpan(0, currentSize);
}
UInt128 q = checked(twoP * kStart);
q = checked(q + UInt128.One);
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(stream, currentSize, exponent, twoPGpu, (GpuUInt128)kStart, last, 0UL,
kernel(currentSize, exponent, twoPGpu, (GpuUInt128)kStart, last, 0UL,
ra, orderBuffer.View, smallCyclesView, primeViews.LastOne, primeViews.LastSeven, primeViews.LastOnePow2, primeViews.LastSevenPow2);
stream.Synchronize();
accelerator.Synchronize();
orderBuffer.View.CopyToCPU(ref MemoryMarshal.GetReference(orders), currentSize);
stream.Synchronize();
if (!Volatile.Read(ref isPrime))
{
break;
}
for (int i = 0; i < currentSize; i++)
for (i = 0; i < currentSize; i++)
{
if (orders[i] != 0UL)
{
@@ -113,18 +81,14 @@ public class MersenneNumberResidueGpuTester(bool useGpuOrder)
}
}
kStart = checked(kStart + (UInt128)currentSize);
}
kStart += batchSize128;
}
}
finally
{
ArrayPool<ulong>.Shared.Return(orderArray);
orderBuffer.Dispose();
execution.Dispose();
gpuLease.Dispose();
}
divisorsExhausted = true;
}
}
+8 -96
View File
@@ -25,13 +25,11 @@ public sealed class MersenneNumberTester(
bool useGpuScan = true,
bool useGpuOrder = true,
bool useResidue = true,
UInt128? maxK = null,
UInt128? residueDivisorSets = null)
ulong maxK = 5_000_000UL)
{
private readonly bool _useResidue = useResidue;
private readonly bool _useIncremental = useIncremental && !useResidue;
private readonly UInt128 _maxK = maxK ?? (UInt128)5_000_000UL;
private readonly UInt128 _residueDivisorSetCount = residueDivisorSets ?? (UInt128)PerfectNumberConstants.ExtraDivisorCycleSearchLimit;
private readonly ulong _maxK = maxK;
private readonly GpuKernelType _kernelType = kernelType;
private readonly bool _useModuloWorkaround = useModuloWorkaround;
private readonly bool _useGpuLucas = useGpuLucas;
@@ -50,23 +48,6 @@ public sealed class MersenneNumberTester(
// 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.
[MethodImpl(MethodImplOptions.AggressiveInlining)]
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;
@@ -218,30 +199,23 @@ public sealed class MersenneNumberTester(
}
}
public bool IsMersennePrime(ulong exponent) => IsMersennePrime(exponent, out _);
public bool IsMersennePrime(ulong exponent, out bool divisorsExhausted)
public bool IsMersennePrime(ulong exponent)
{
divisorsExhausted = false;
// 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)
{
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;
}
@@ -254,79 +228,21 @@ public sealed class MersenneNumberTester(
// In residue mode the residue scan is the final primality check.
if (_useResidue)
{
if (_maxK < maxK)
if ((UInt128)_maxK < maxK)
{
maxK = _maxK;
}
if (_maxK == UInt128.Zero || _residueDivisorSetCount == UInt128.Zero)
{
return prePrime;
}
UInt128 totalLimit = _maxK;
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);
_residueGpuTester!.Scan(exponent, twoP, lastIsSeven, maxK, ref prePrime);
}
else
{
_residueCpuTester!.Scan(exponent, twoP, lastIsSeven, perSetLimit, effectiveSets, totalLimit, ref prePrime, ref scanCompleted);
_residueCpuTester!.Scan(exponent, twoP, lastIsSeven, maxK, ref prePrime);
}
if (!prePrime)
{
divisorsExhausted = true;
return false;
}
divisorsExhausted = scanCompleted;
return true;
return prePrime;
}
if (!_useIncremental)
@@ -347,15 +263,12 @@ public sealed class MersenneNumberTester(
if (!prePrime)
{
divisorsExhausted = true;
return false;
}
bool lucasPrime = _useGpuLucas
return _useGpuLucas
? _lucasLehmerGpuTester.IsPrime(exponent, runOnGpu: true)
: _lucasLehmerCpuTester.IsPrime(exponent);
divisorsExhausted = true;
return lucasPrime;
}
bool isPrime = true;
@@ -383,7 +296,6 @@ public sealed class MersenneNumberTester(
}
}
divisorsExhausted = true;
return isPrime;
}