From 8bbbaa5bae6c19d3d388de277f37feedcf01e4f4 Mon Sep 17 00:00:00 2001
From: Jiri Pittner <jiri@pittnerovi.com>
Date: Wed, 3 Apr 2024 22:12:08 +0200
Subject: [PATCH] copyonwrite() on nested LA types will recursively call
 element copyonwrites even if the top type had count=1

---
 mat.h        | 12 ++++++++++--
 smat.h       | 12 ++++++++++--
 sparsemat.cc | 14 +++++++++++++-
 sparsemat.h  |  2 +-
 vec.h        | 11 +++++++++--
 5 files changed, 43 insertions(+), 8 deletions(-)

diff --git a/mat.h b/mat.h
index 5dbbf25..c76902b 100644
--- a/mat.h
+++ b/mat.h
@@ -126,7 +126,7 @@ public:
 	inline int getcount() const {return count?*count:0;}
 
 	//! ensure that the data of this matrix are referenced exactly once
-	void copyonwrite(bool detachonly=false);
+	void copyonwrite(bool detachonly=false, bool deep=true);
 
 	//! permute matrix elements
         const NRMat permuted_rows(const NRPerm<int> &p, const bool inverse=false) const;
@@ -1107,7 +1107,7 @@ NRMat<T> & NRMat<T>::operator|=(const NRMat<T> &rhs) {
  * @see NRMat<T>::count, NRMat<T>::operator|=() 
  ******************************************************************************/
 template <typename T>
-void NRMat<T>::copyonwrite(bool detachonly) {
+void NRMat<T>::copyonwrite(bool detachonly, bool deep) {
 	if(!count) 
                 { 
                 if(nn||mm) laerror("nonempty matrix without reference count encountered");
@@ -1115,6 +1115,14 @@ void NRMat<T>::copyonwrite(bool detachonly) {
                 return;
                 }
 
+        if(*count == 1 && !LA_traits<T>::is_plaindata() && !detachonly &&deep) //type-nested copyonwrite
+                {
+#ifdef CUDALA
+                if(location != cpu) laerror("nested types not supported on gpu memory");
+#endif
+		for(int i=0; i<nn*mm; ++i) LA_traits<T>::copyonwrite(v[i]);
+                }
+
 	if(*count > 1){
 		(*count)--;
 		count = new int;
diff --git a/smat.h b/smat.h
index 355d55a..424c115 100644
--- a/smat.h
+++ b/smat.h
@@ -168,7 +168,7 @@ public:
 	void get(int fd, bool dimensions = 1, bool transp = 0);
         void put(int fd, bool dimensions = 1, bool transp = 0) const;
 
-	void copyonwrite(bool detachonly=false);
+	void copyonwrite(bool detachonly=false, bool deep=true);
 
 	void clear() {copyonwrite(LA_traits<T>::is_plaindata()); LA_traits<T>::clear(v,NN2);}; //zero out
 	void resize(const int n);
@@ -984,13 +984,21 @@ NRSMat<T> & NRSMat<T>::operator=(const NRSMat<T> & rhs) {
  * @see NRSMat<T>::operator|=, NRSMat<T>::copyonwrite()
  ******************************************************************************/
 template <typename T>
-void NRSMat<T>::copyonwrite(bool detachonly) {
+void NRSMat<T>::copyonwrite(bool detachonly, bool deep) {
 	if(!count) 
                 { 
                 if(nn) laerror("nonempty smat without reference count encountered");
                 if(_LA_warn_empty_copyonwrite) std::cout <<"Warning: copyonwrite of empty smat\n";
                 return;
                 }
+        if(*count == 1 && !LA_traits<T>::is_plaindata() && !detachonly && deep) //type-nested copyonwrite
+                {
+#ifdef CUDALA
+                if(location != cpu) laerror("nested types not supported on gpu memory");
+#endif
+                for(int i=0; i<NN2; ++i) LA_traits<T>::copyonwrite(v[i]);
+                }
+
 	if(*count > 1){
 		(*count)--;
 		count = new int;
diff --git a/sparsemat.cc b/sparsemat.cc
index 6dc3ac6..586014d 100644
--- a/sparsemat.cc
+++ b/sparsemat.cc
@@ -143,7 +143,7 @@ while(l)
 }
 
 template <class T>
-void SparseMat<T>::copyonwrite(bool detachonly)
+void SparseMat<T>::copyonwrite(bool detachonly, bool deep)
 {
 	if(!count) 
                 { 
@@ -152,6 +152,18 @@ void SparseMat<T>::copyonwrite(bool detachonly)
                 return;
                 }
 
+        if(*count == 1 && !LA_traits<T>::is_plaindata() && !detachonly && deep) //type-nested copyonwrite
+                {
+		matel<T> *l =list;
+                while(l)
+                         {
+                         LA_traits<T>::copyonwrite(l->elem);
+                         l=l->next;
+                         }
+
+                }
+
+
         if(*count > 1)
                 {
                 (*count)--;
diff --git a/sparsemat.h b/sparsemat.h
index 69d2c35..dcfd509 100644
--- a/sparsemat.h
+++ b/sparsemat.h
@@ -139,7 +139,7 @@ public:
 	void setunsymmetric();//unwind the matrix assuming it was indeed symmetric
 	inline bool issymmetric() const {return symmetric;}
 	unsigned int length() const;
-	void copyonwrite(bool detachonly=false);
+	void copyonwrite(bool detachonly=false, bool deep=true);
 	void clear() {copyonwrite(LA_traits<T>::is_plaindata()); if(count) {delete count; count=NULL;}}
 	void simplify(const double sparseepsilon=SPARSEEPSILON);
 	const T trace() const;
diff --git a/vec.h b/vec.h
index bb2443c..34b5b0f 100644
--- a/vec.h
+++ b/vec.h
@@ -177,7 +177,7 @@ public:
 	#endif
 
 	//! create separate copy of the data corresponding to this vector 
-	void copyonwrite(bool detachonly=false);
+	void copyonwrite(bool detachonly=false, bool deep=true);
 
 	//! purge this vector 
 	void clear() { copyonwrite(LA_traits<T>::is_plaindata()); LA_traits<T>::clear(v, nn); };
@@ -1127,13 +1127,20 @@ NRVec<T>::~NRVec() {
  * make own copy of the underlying data connected with this vector
  ******************************************************************************/
 template <typename T>
-void NRVec<T>::copyonwrite(bool detachonly) {
+void NRVec<T>::copyonwrite(bool detachonly,bool deep) {
 	if(!count)
 		{
 		if(nn) laerror("nonempty vector without reference count encountered");
 		if(_LA_warn_empty_copyonwrite) std::cout <<"Warning: copyonwrite of empty vector\n";
 		return;
 		}
+	if(*count == 1 && !LA_traits<T>::is_plaindata() && !detachonly && deep) //type-nested copyonwrite
+		{
+#ifdef CUDALA
+                if(location != cpu) laerror("nested types not supported on gpu memory");
+#endif
+		for(int i=0; i<nn; ++i) LA_traits<T>::copyonwrite(v[i]);
+		}
 	if(*count > 1) {
 		(*count)--;
 		 count = new int;