From 81830c9031bc8198b2a09dcdec85e39fc1784046 Mon Sep 17 00:00:00 2001
From: Miha Kralj <miha.kralj@outlook.com>
Date: Wed, 10 Dec 2025 15:15:20 -0500
Subject: [PATCH] Optimize memory allocation in ALMA and Conv classes; improve
 performance by using stackalloc for small periods

---
 lib/trends/alma/Alma.cs |  4 ++--
 lib/trends/conv/Conv.cs | 15 +++++----------
 lib/trends/wma/Wma.cs   |  4 +---
 3 files changed, 8 insertions(+), 15 deletions(-)

diff --git a/lib/trends/alma/Alma.cs b/lib/trends/alma/Alma.cs
index 5f78de16..a94eb6d5 100644
--- a/lib/trends/alma/Alma.cs
+++ b/lib/trends/alma/Alma.cs
@@ -210,7 +210,8 @@ public sealed class Alma : ITValuePublisher
             throw new ArgumentException("Source and output must have the same length");
 
         // Precompute weights
-        double[] weights = new double[period];
+        // Use stackalloc for small periods to avoid heap allocation
+        Span<double> weights = period <= 256 ? stackalloc double[period] : new double[period];
         double m = offset * (period - 1);
         double s = period / sigma;
         double s2 = 2 * s * s;
@@ -224,7 +225,6 @@ public sealed class Alma : ITValuePublisher
         }
 
         // Buffer for sliding window
-        // Use stackalloc for small periods
         Span<double> buffer = period <= 256 ? stackalloc double[period] : new double[period];
         int bufferIdx = 0;
         int count = 0;
diff --git a/lib/trends/conv/Conv.cs b/lib/trends/conv/Conv.cs
index c72d9fe6..9739b642 100644
--- a/lib/trends/conv/Conv.cs
+++ b/lib/trends/conv/Conv.cs
@@ -26,7 +26,6 @@ public sealed class Conv : ITValuePublisher
     private readonly RingBuffer _buffer;
 
     private double _lastValidValue;
-    private int _head;
 
     // State for bar correction
     private double _p_lastValidValue;
@@ -83,7 +82,6 @@ public sealed class Conv : ITValuePublisher
         if (isNew)
         {
             _buffer.Add(val);
-            _head = (_head + 1 == _period) ? 0 : _head + 1;
         }
         else
         {
@@ -104,10 +102,11 @@ public sealed class Conv : ITValuePublisher
             }
             else
             {
-                // Full: data is split at _head (which points to oldest)
-                int part1Len = _period - _head;
-                result = internalBuf.Slice(_head, part1Len).DotProduct(kernelSpan.Slice(0, part1Len))
-                       + internalBuf.Slice(0, _head).DotProduct(kernelSpan.Slice(part1Len));
+                // Full: data is split at StartIndex (which points to oldest)
+                int head = _buffer.StartIndex;
+                int part1Len = _period - head;
+                result = internalBuf.Slice(head, part1Len).DotProduct(kernelSpan.Slice(0, part1Len))
+                       + internalBuf.Slice(0, head).DotProduct(kernelSpan.Slice(part1Len));
             }
         }
 
@@ -165,9 +164,6 @@ public sealed class Conv : ITValuePublisher
             _buffer.Add(val);
         }
 
-        // Sync _head with buffer state
-        _head = windowSize % _period;
-
         // Set Last
         Last = new TValue(source.Times[len - 1], vSpan[len - 1]);
 
@@ -247,7 +243,6 @@ public sealed class Conv : ITValuePublisher
         _buffer.Clear();
         _lastValidValue = double.NaN;
         _p_lastValidValue = double.NaN;
-        _head = 0;
         Last = default;
     }
 }
diff --git a/lib/trends/wma/Wma.cs b/lib/trends/wma/Wma.cs
index 5e8b6bd9..2b83fe68 100644
--- a/lib/trends/wma/Wma.cs
+++ b/lib/trends/wma/Wma.cs
@@ -421,9 +421,7 @@ public sealed class Wma : ITValuePublisher
 
                 var vSums = Avx.Add(vSumState, vPS2);
 
-                var vSumsShifted = Avx2.Permute4x64(vSums.AsUInt64(), 0b_10_01_00_00).AsDouble(); // skipcq: CS-R1131
-                vSumsShifted = Avx.Blend(vSumState, vSumsShifted, 0b_1110);
-
+                var vSumsShifted = Avx.Subtract(vSums, vDeltaS);
                 var vTerm1 = Avx.Multiply(vPeriod, vNew);
                 var vU = Avx.Subtract(vTerm1, vSumsShifted);