Refine Montgomery GPU helpers to avoid kernel branching

This commit is contained in:
Marek Lemańczyk committed 2025-10-22 01:43:10 +02:00
1 parent f12919e220
commit 24eaec7a21
5 files changed
+46 -22

No files matched your search

@@ -17,7 +17,7 @@ internal static class DivisorByDivisorKernels
}
ulong exponent = exponents[index];
hits[index] = ULongExtensions.Pow2MontgomeryModWindowedGpu(divisor, exponent, keepMontgomery: false) == 1UL ? (byte)1 : (byte)0;
hits[index] = ULongExtensions.Pow2MontgomeryModWindowedGpuConvertToStandard(divisor, exponent) == 1UL ? (byte)1 : (byte)0;
}
public static void ComputeRemainderDeltasKernel(Index1D index, ArrayView<ulong> gaps, ArrayView<byte> deltas, byte modulus)
@@ -149,6 +149,6 @@ internal static class DivisorByDivisorKernels
ulong exponent = exponents[index];
results[index] = ULongExtensions.Pow2MontgomeryModWindowedGpu(divisor, exponent, keepMontgomery: true);
results[index] = ULongExtensions.Pow2MontgomeryModWindowedGpuKeepMontgomery(divisor, exponent);
}
}
@@ -12,7 +12,7 @@ internal static class Pow2MontgomeryKernels
MontgomeryDivisorData divisor,
ArrayView1D<ulong, Stride1D.Dense> results)
{
results[index] = ULongExtensions.Pow2MontgomeryModWindowedGpu(divisor, exponents[index], keepMontgomery: true);
results[index] = ULongExtensions.Pow2MontgomeryModWindowedGpuKeepMontgomery(divisor, exponents[index]);
}
public static void Pow2MontgomeryKernelConvertToStandard(
@@ -21,6 +21,6 @@ internal static class Pow2MontgomeryKernels
MontgomeryDivisorData divisor,
ArrayView1D<ulong, Stride1D.Dense> results)
{
results[index] = ULongExtensions.Pow2MontgomeryModWindowedGpu(divisor, exponents[index], keepMontgomery: false);
results[index] = ULongExtensions.Pow2MontgomeryModWindowedGpuConvertToStandard(divisor, exponents[index]);
}
}
@@ -347,7 +347,7 @@ internal static partial class PrimeOrderGpuHeuristics
}
ulong reduced = phi / factor;
if (ULongExtensions.Pow2MontgomeryModWindowedGpu(divisor, reduced, keepMontgomery: false) == 1UL)
if (ULongExtensions.Pow2MontgomeryModWindowedGpuConvertToStandard(divisor, reduced) == 1UL)
{
return false;
}
@@ -1131,7 +1131,7 @@ internal static partial class PrimeOrderGpuHeuristics
private static bool Pow2EqualsOneKernel(ulong exponent, in MontgomeryDivisorData divisor)
{
return ULongExtensions.Pow2MontgomeryModWindowedGpu(divisor, exponent, keepMontgomery: false) == 1UL;
return ULongExtensions.Pow2MontgomeryModWindowedGpuConvertToStandard(divisor, exponent) == 1UL;
}
private static ulong CalculateByDoublingKernel(ulong prime)
@@ -1156,7 +1156,7 @@ internal static partial class PrimeOrderGpuHeuristics
internal static void Pow2ModKernel(Index1D index, ArrayView1D<ulong, Stride1D.Dense> exponents, MontgomeryDivisorData divisor, ArrayView1D<ulong, Stride1D.Dense> remainders)
{
ulong exponent = exponents[index];
remainders[index] = ULongExtensions.Pow2MontgomeryModWindowedGpu(divisor, exponent, keepMontgomery: false);
remainders[index] = ULongExtensions.Pow2MontgomeryModWindowedGpuConvertToStandard(divisor, exponent);
}
private static Action<AcceleratorStream, Index1D, ArrayView1D<GpuUInt128, Stride1D.Dense>, GpuUInt128, ArrayView1D<GpuUInt128, Stride1D.Dense>> GetPow2ModWideKernel(Accelerator accelerator)