From 20db1e4eeabaa65bb75fedf41bd730c300b48160 Mon Sep 17 00:00:00 2001
From: conrad <conrad@localhost.localdomain>
Date: Thu, 20 Jul 2023 15:06:58 +1000
Subject: [PATCH] speedup by transpose based multiplication

---
 .../newarp_SparseGenMatProd_bones.hpp          |  1 +
 .../newarp_SparseGenMatProd_meat.hpp           | 18 +++++++++++++++---
 2 files changed, 16 insertions(+), 3 deletions(-)

diff --git a/include/armadillo_bits/newarp_SparseGenMatProd_bones.hpp b/include/armadillo_bits/newarp_SparseGenMatProd_bones.hpp
index fb3cefa8..2028aee2 100644
--- a/include/armadillo_bits/newarp_SparseGenMatProd_bones.hpp
+++ b/include/armadillo_bits/newarp_SparseGenMatProd_bones.hpp
@@ -27,6 +27,7 @@ class SparseGenMatProd
   private:
   
   const SpMat<eT>& op_mat;
+        SpMat<eT>  op_mat_st;
   
   
   public:
diff --git a/include/armadillo_bits/newarp_SparseGenMatProd_meat.hpp b/include/armadillo_bits/newarp_SparseGenMatProd_meat.hpp
index 0f10f125..bbe539a1 100644
--- a/include/armadillo_bits/newarp_SparseGenMatProd_meat.hpp
+++ b/include/armadillo_bits/newarp_SparseGenMatProd_meat.hpp
@@ -28,6 +28,8 @@ SparseGenMatProd<eT>::SparseGenMatProd(const SpMat<eT>& mat_obj)
   , n_cols(mat_obj.n_cols)
   {
   arma_extra_debug_sigprint();
+  
+  op_mat_st = op_mat.st(); // pre-calculate transpose
   }
 
 
@@ -41,10 +43,20 @@ SparseGenMatProd<eT>::perform_op(eT* x_in, eT* y_out) const
   {
   arma_extra_debug_sigprint();
   
-  const Col<eT> x(x_in , n_cols, false, true);
-        Col<eT> y(y_out, n_rows, false, true);
+  // // OLD METHOD
+  // 
+  // const Col<eT> x(x_in , n_cols, false, true);
+  //       Col<eT> y(y_out, n_rows, false, true);
+  // 
+  // y = op_mat * x;
   
-  y = op_mat * x;
+  
+  // NEW METHOD
+  
+  const Row<eT> x(x_in , n_cols, false, true);
+        Row<eT> y(y_out, n_rows, false, true);
+  
+  y =  x * op_mat_st;
   }