From d0c9389bcc910aeb883941da03f3b6ee8e5328af Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Marek=20Lema=C5=84czyk?= <marek@lemanczyk.eu>
Date: Mon, 20 Oct 2025 13:58:04 +0200
Subject: [PATCH] Document GpuUInt128 MulMod shift benchmarking

---
 PerfectNumbers.Core/Gpu/GpuUInt128.cs | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/PerfectNumbers.Core/Gpu/GpuUInt128.cs b/PerfectNumbers.Core/Gpu/GpuUInt128.cs
index 3ec4ed5..5700bc7 100644
--- a/PerfectNumbers.Core/Gpu/GpuUInt128.cs
+++ b/PerfectNumbers.Core/Gpu/GpuUInt128.cs
@@ -488,6 +488,9 @@ public struct GpuUInt128 : IComparable<GpuUInt128>, IEquatable<GpuUInt128>
             }
             else
             {
+                // Copy the low word into a temporary before shifting High; assigning directly
+                // regressed the GpuUInt128MulMod benchmarks (HighWordModulus 6.45 -> 7.38 us,
+                // MixedMagnitude 4.99 -> 6.24 us) with InvocationCount=16, IterationCount=3.
                 ulong newLow = (multiplier.Low >> 1) | (multiplier.High << 63);
                 multiplier.High >>= 1;
                 multiplier.Low = newLow;