From d6bb2dcbbf08be80b4e9a08392de3c433cc77d68 Mon Sep 17 00:00:00 2001
From: sunwen <sunwen@realcan.com>
Date: Thu, 8 Jun 2023 13:34:45 +0800
Subject: [PATCH] speed up repmat.

---
 src/Function1D.cpp | 15 ++++++++++-----
 1 file changed, 10 insertions(+), 5 deletions(-)

diff --git a/src/Function1D.cpp b/src/Function1D.cpp
index 041b01a..337074c 100644
--- a/src/Function1D.cpp
+++ b/src/Function1D.cpp
@@ -377,18 +377,23 @@ Matrix Aurora::repmat(const Matrix& aMatrix,int aRowTimes, int aColumnTimes)
     int column = aMatrix.getDimSize(1);
     double* originalData = aMatrix.getData();
     double* resultDataTemp = resultData;
+    size_t step = row*complexStep;
+    #pragma omp parallel for
     for(int i=0; i<column; ++i)
     {
-        for(int j=1; j<=aRowTimes; ++j)
+        double* origninalStart = originalData + i * step;
+        for(int j=0; j<aRowTimes; ++j)
         {
-            cblas_dcopy(row*complexStep,originalData, 1, resultDataTemp, 1);
-            resultDataTemp += row*complexStep;
+            double* copyStart = resultData + step * (i*aRowTimes + j);
+            cblas_dcopy(step, origninalStart, 1, copyStart, 1);
+            //resultDataTemp += row*complexStep;
         }
-        originalData += row*complexStep;
+        //originalData += step;
     }
 
     resultDataTemp = resultData;
-    int step = originalDataSize * aRowTimes;
+    step = originalDataSize * aRowTimes;
+    #pragma omp parallel for
     for(int i=1; i<aColumnTimes; ++i)
     {
         cblas_dcopy(step, resultData, 1, resultData + i*step, 1);