From 452a6c1b4929d77aef6a88da07e4e24dbb7ed5b3 Mon Sep 17 00:00:00 2001
From: conrad <conrad@localhost.localdomain>
Date: Sat, 13 Jun 2026 00:29:21 +1000
Subject: [PATCH] use partially unrolled loops

---
 include/armadillo_bits/glue_max_meat.hpp | 38 ++++++++++++++++++++----
 include/armadillo_bits/glue_min_meat.hpp | 38 ++++++++++++++++++++----
 2 files changed, 66 insertions(+), 10 deletions(-)

diff --git a/include/armadillo_bits/glue_max_meat.hpp b/include/armadillo_bits/glue_max_meat.hpp
index 15995a7f..93404704 100644
--- a/include/armadillo_bits/glue_max_meat.hpp
+++ b/include/armadillo_bits/glue_max_meat.hpp
@@ -75,13 +75,41 @@ glue_max::apply(Mat<eT>& out, const Proxy<T1>& PA, const Proxy<T2>& PB)
     
     const uword N = PA.get_n_elem();
     
-    for(uword i=0; i<N; ++i)
+    #if defined(ARMA_SIMPLE_LOOPS)
       {
-      const eT Ai = A[i];
-      const eT Bi = B[i];
-      
-      out_mem[i] = comparator(Ai,Bi) ? Ai : Bi;
+      for(uword i=0; i<N; ++i)
+        {
+        const eT Ai = A[i];
+        const eT Bi = B[i];
+        
+        out_mem[i] = comparator(Ai,Bi) ? Ai : Bi;
+        }
       }
+    #else
+      {
+      uword i,j;
+      
+      for(i=0, j=1; j < N; i+=2, j+=2)
+        {
+        const eT Ai = A[i];
+        const eT Aj = A[j];
+        
+        const eT Bi = B[i];
+        const eT Bj = B[j];
+        
+        (*out_mem) = comparator(Ai,Bi) ? Ai : Bi;  ++out_mem;
+        (*out_mem) = comparator(Aj,Bj) ? Aj : Bj;  ++out_mem;
+        }
+      
+      if(i < N)
+        {
+        const eT Ai = A[i];
+        const eT Bi = B[i];
+        
+        (*out_mem) = comparator(Ai,Bi) ? Ai : Bi;
+        }
+      }
+    #endif
     }
   else
     {
diff --git a/include/armadillo_bits/glue_min_meat.hpp b/include/armadillo_bits/glue_min_meat.hpp
index 17039996..82dd55a7 100644
--- a/include/armadillo_bits/glue_min_meat.hpp
+++ b/include/armadillo_bits/glue_min_meat.hpp
@@ -75,13 +75,41 @@ glue_min::apply(Mat<eT>& out, const Proxy<T1>& PA, const Proxy<T2>& PB)
     
     const uword N = PA.get_n_elem();
     
-    for(uword i=0; i<N; ++i)
+    #if defined(ARMA_SIMPLE_LOOPS)
       {
-      const eT Ai = A[i];
-      const eT Bi = B[i];
-      
-      out_mem[i] = comparator(Ai,Bi) ? Ai : Bi;
+      for(uword i=0; i<N; ++i)
+        {
+        const eT Ai = A[i];
+        const eT Bi = B[i];
+        
+        out_mem[i] = comparator(Ai,Bi) ? Ai : Bi;
+        }
       }
+    #else
+      {
+      uword i,j;
+      
+      for(i=0, j=1; j < N; i+=2, j+=2)
+        {
+        const eT Ai = A[i];
+        const eT Aj = A[j];
+        
+        const eT Bi = B[i];
+        const eT Bj = B[j];
+        
+        (*out_mem) = comparator(Ai,Bi) ? Ai : Bi;  ++out_mem;
+        (*out_mem) = comparator(Aj,Bj) ? Aj : Bj;  ++out_mem;
+        }
+      
+      if(i < N)
+        {
+        const eT Ai = A[i];
+        const eT Bi = B[i];
+        
+        (*out_mem) = comparator(Ai,Bi) ? Ai : Bi;
+        }
+      }
+    #endif
     }
   else
     {