use atomic operations

* regcomp.c (onig_chain_link_add): use atomic operation instead of
  mutex.
* regint.h (ONIG_STATE_{INC,DEC}_THREAD): ditto.
* regparse.c (PopFreeNode, node_recycle): ditto.

git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@52995 b2dd03c8-39d4-4d8f-98ff-823fe69b080e

6 files changed, 80 insertions, 50 deletions

diff --git a/ChangeLog b/ChangeLog
index eb9c051b8c..1c44cd8c00 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,12 @@
+Wed Dec  9 15:48:12 2015  Nobuyoshi Nakada  <nobu@ruby-lang.org>
+
+	* regcomp.c (onig_chain_link_add): use atomic operation instead of
+	  mutex.
+
+	* regint.h (ONIG_STATE_{INC,DEC}_THREAD): ditto.
+
+	* regparse.c (PopFreeNode, node_recycle): ditto.
+
 Wed Dec  9 14:45:27 2015  Koichi Sasada  <ko1@atdot.net>
 
 	* gc.c (gc_mark_stacked_objects): fix typo.
diff --git a/common.mk b/common.mk
index dde72e30ee..78127429c9 100644
--- a/common.mk
+++ b/common.mk
@@ -1345,6 +1345,7 @@ enc/unicode.$(OBJEXT): {$(VPATH)}missing.h
 enc/unicode.$(OBJEXT): {$(VPATH)}oniguruma.h
 enc/unicode.$(OBJEXT): {$(VPATH)}regenc.h
 enc/unicode.$(OBJEXT): {$(VPATH)}regint.h
+enc/unicode.$(OBJEXT): {$(VPATH)}ruby_atomic.h
 enc/unicode.$(OBJEXT): {$(VPATH)}st.h
 enc/unicode.$(OBJEXT): {$(VPATH)}subst.h
 enc/us_ascii.$(OBJEXT): {$(VPATH)}config.h
@@ -1987,6 +1988,7 @@ re.$(OBJEXT): {$(VPATH)}re.h
 re.$(OBJEXT): {$(VPATH)}regenc.h
 re.$(OBJEXT): {$(VPATH)}regex.h
 re.$(OBJEXT): {$(VPATH)}regint.h
+re.$(OBJEXT): {$(VPATH)}ruby_atomic.h
 re.$(OBJEXT): {$(VPATH)}st.h
 re.$(OBJEXT): {$(VPATH)}subst.h
 re.$(OBJEXT): {$(VPATH)}util.h
@@ -1999,6 +2001,7 @@ regcomp.$(OBJEXT): {$(VPATH)}oniguruma.h
 regcomp.$(OBJEXT): {$(VPATH)}regcomp.c
 regcomp.$(OBJEXT): {$(VPATH)}regenc.h
 regcomp.$(OBJEXT): {$(VPATH)}regint.h
+regcomp.$(OBJEXT): {$(VPATH)}ruby_atomic.h
 regcomp.$(OBJEXT): {$(VPATH)}regparse.h
 regcomp.$(OBJEXT): {$(VPATH)}st.h
 regcomp.$(OBJEXT): {$(VPATH)}subst.h
@@ -2011,6 +2014,7 @@ regenc.$(OBJEXT): {$(VPATH)}oniguruma.h
 regenc.$(OBJEXT): {$(VPATH)}regenc.c
 regenc.$(OBJEXT): {$(VPATH)}regenc.h
 regenc.$(OBJEXT): {$(VPATH)}regint.h
+regenc.$(OBJEXT): {$(VPATH)}ruby_atomic.h
 regenc.$(OBJEXT): {$(VPATH)}st.h
 regenc.$(OBJEXT): {$(VPATH)}subst.h
 regerror.$(OBJEXT): $(hdrdir)/ruby/ruby.h
@@ -2022,6 +2026,7 @@ regerror.$(OBJEXT): {$(VPATH)}oniguruma.h
 regerror.$(OBJEXT): {$(VPATH)}regenc.h
 regerror.$(OBJEXT): {$(VPATH)}regerror.c
 regerror.$(OBJEXT): {$(VPATH)}regint.h
+regerror.$(OBJEXT): {$(VPATH)}ruby_atomic.h
 regerror.$(OBJEXT): {$(VPATH)}st.h
 regerror.$(OBJEXT): {$(VPATH)}subst.h
 regexec.$(OBJEXT): $(hdrdir)/ruby/ruby.h
@@ -2033,6 +2038,7 @@ regexec.$(OBJEXT): {$(VPATH)}oniguruma.h
 regexec.$(OBJEXT): {$(VPATH)}regenc.h
 regexec.$(OBJEXT): {$(VPATH)}regexec.c
 regexec.$(OBJEXT): {$(VPATH)}regint.h
+regexec.$(OBJEXT): {$(VPATH)}ruby_atomic.h
 regexec.$(OBJEXT): {$(VPATH)}st.h
 regexec.$(OBJEXT): {$(VPATH)}subst.h
 regparse.$(OBJEXT): $(hdrdir)/ruby/ruby.h
@@ -2045,6 +2051,7 @@ regparse.$(OBJEXT): {$(VPATH)}regenc.h
 regparse.$(OBJEXT): {$(VPATH)}regint.h
 regparse.$(OBJEXT): {$(VPATH)}regparse.c
 regparse.$(OBJEXT): {$(VPATH)}regparse.h
+regparse.$(OBJEXT): {$(VPATH)}ruby_atomic.h
 regparse.$(OBJEXT): {$(VPATH)}st.h
 regparse.$(OBJEXT): {$(VPATH)}subst.h
 regsyntax.$(OBJEXT): $(hdrdir)/ruby/ruby.h
@@ -2056,6 +2063,7 @@ regsyntax.$(OBJEXT): {$(VPATH)}oniguruma.h
 regsyntax.$(OBJEXT): {$(VPATH)}regenc.h
 regsyntax.$(OBJEXT): {$(VPATH)}regint.h
 regsyntax.$(OBJEXT): {$(VPATH)}regsyntax.c
+regsyntax.$(OBJEXT): {$(VPATH)}ruby_atomic.h
 regsyntax.$(OBJEXT): {$(VPATH)}st.h
 regsyntax.$(OBJEXT): {$(VPATH)}subst.h
 ruby.$(OBJEXT): $(CCAN_DIR)/check_type/check_type.h
diff --git a/regcomp.c b/regcomp.c
index 23cb0ead9a..176aa6b8ef 100644
--- a/regcomp.c
+++ b/regcomp.c
@@ -5663,10 +5663,11 @@ onig_transfer(regex_t* to, regex_t* from)
 extern void
 onig_chain_link_add(regex_t* to, regex_t* add)
 {
-  THREAD_ATOMIC_START;
-  REGEX_CHAIN_HEAD(to);
-  to->chain = add;
-  THREAD_ATOMIC_END;
+  /* THREAD_ATOMIC_START; */
+  do {
+    REGEX_CHAIN_HEAD(to);
+  } while (IS_NOT_NULL(ATOMIC_PTR_CAS(to->chain, (regex_t* )NULL, add)));
+  /* THREAD_ATOMIC_END; */
 }
 
 extern void
diff --git a/regint.h b/regint.h
index 9fda5509a9..79867f7b6d 100644
--- a/regint.h
+++ b/regint.h
@@ -95,6 +95,7 @@
 
 #ifndef RUBY_DEFINES_H
 #include "ruby/ruby.h"
+#include "ruby_atomic.h"
 #undef xmalloc
 #undef xrealloc
 #undef xcalloc
@@ -238,6 +239,10 @@ extern pthread_mutex_t gOnigMutex;
 #define ONIG_STATE_INC(reg) (reg)->state++
 #define ONIG_STATE_DEC(reg) (reg)->state--
 
+#if SIZEOF_ATOMIC_T == SIZEOF_INT
+#define ONIG_STATE_INC_THREAD(reg) (ATOMIC_INC((reg)->state))
+#define ONIG_STATE_DEC_THREAD(reg) (ATOMIC_DEC((reg)->state))
+#else
 #define ONIG_STATE_INC_THREAD(reg) do {\
   THREAD_ATOMIC_START;\
   (reg)->state++;\
@@ -248,6 +253,7 @@ extern pthread_mutex_t gOnigMutex;
   (reg)->state--;\
   THREAD_ATOMIC_END;\
 } while(0)
+#endif
 #else
 #define ONIG_STATE_INC(reg)         /* Nothing */
 #define ONIG_STATE_DEC(reg)         /* Nothing */
diff --git a/regparse.c b/regparse.c
index 094332f90e..0baf02c145 100644
--- a/regparse.c
+++ b/regparse.c
@@ -1031,7 +1031,43 @@ typedef struct _FreeNode {
 } FreeNode;
 
 static FreeNode* FreeNodeList = (FreeNode* )NULL;
+
+#define PopFreeNode(n, popped) \
+  do { \
+    FreeNode* n = FreeNodeList; \
+    while (IS_NOT_NULL(n)) { \
+      FreeNode* next = n->next; \
+      FreeNode *head = ATOMIC_PTR_CAS(FreeNodeList, n, next); \
+      if (head == n) { \
+	popped; \
+	n = next; \
+      } \
+      else { \
+	n = head; /* modified, retry */ \
+      } \
+    } \
+  } while (0)
+
+#endif
+
+static void
+node_recycle(Node* node)
+{
+#ifdef USE_PARSE_TREE_NODE_RECYCLE
+  FreeNode* n = (FreeNode* )node;
+  FreeNode* list = FreeNodeList;
+  FreeNode* l;
+
+  /* THREAD_ATOMIC_START; */
+  do {
+    n->next = l = list;
+    list = ATOMIC_PTR_CAS(FreeNodeList, list, n);
+  } while (list != l);
+  /* THREAD_ATOMIC_END; */
+#else
+  xfree(node);
 #endif
+}
 
 extern void
 onig_node_free(Node* node)
@@ -1053,18 +1089,7 @@ onig_node_free(Node* node)
     {
       Node* next_node = NCDR(node);
 
-#ifdef USE_PARSE_TREE_NODE_RECYCLE
-      {
-	FreeNode* n = (FreeNode* )node;
-
-        THREAD_ATOMIC_START;
-	n->next = FreeNodeList;
-	FreeNodeList = n;
-        THREAD_ATOMIC_END;
-      }
-#else
-      xfree(node);
-#endif
+      node_recycle(node);
       node = next_node;
       goto start;
     }
@@ -1101,32 +1126,15 @@ onig_node_free(Node* node)
     break;
   }
 
-#ifdef USE_PARSE_TREE_NODE_RECYCLE
-  {
-    FreeNode* n = (FreeNode* )node;
-
-    THREAD_ATOMIC_START;
-    n->next = FreeNodeList;
-    FreeNodeList = n;
-    THREAD_ATOMIC_END;
-  }
-#else
-  xfree(node);
-#endif
+  node_recycle(node);
 }
 
 #ifdef USE_PARSE_TREE_NODE_RECYCLE
 extern int
 onig_free_node_list(void)
 {
-  FreeNode* n;
-
   /* THREAD_ATOMIC_START; */
-  while (IS_NOT_NULL(FreeNodeList)) {
-    n = FreeNodeList;
-    FreeNodeList = FreeNodeList->next;
-    xfree(n);
-  }
+  PopFreeNode(n, xfree(n));
   /* THREAD_ATOMIC_END; */
   return 0;
 }
@@ -1138,14 +1146,9 @@ node_new(void)
   Node* node;
 
 #ifdef USE_PARSE_TREE_NODE_RECYCLE
-  THREAD_ATOMIC_START;
-  if (IS_NOT_NULL(FreeNodeList)) {
-    node = (Node* )FreeNodeList;
-    FreeNodeList = FreeNodeList->next;
-    THREAD_ATOMIC_END;
-    return node;
-  }
-  THREAD_ATOMIC_END;
+  /* THREAD_ATOMIC_START; */
+  PopFreeNode(n, return (Node* )n);
+  /* THREAD_ATOMIC_END; */
 #endif
 
   node = (Node* )xmalloc(sizeof(Node));
@@ -1155,17 +1158,14 @@ node_new(void)
 
 #if defined(USE_MULTI_THREAD_SYSTEM) && \
     defined(USE_SHARED_CCLASS_TABLE) && \
-    defined(USE_PARSE_TREE_NODE_RECYCLE)
+    defined(USE_PARSE_TREE_NODE_RECYCLE) && \
+    0
 static Node*
 node_new_locked(void)
 {
   Node* node;
 
-  if (IS_NOT_NULL(FreeNodeList)) {
-    node = (Node* )FreeNodeList;
-    FreeNodeList = FreeNodeList->next;
-    return node;
-  }
+  PopFreeNode(n, return (Node* )n);
 
   node = (Node* )xmalloc(sizeof(Node));
   /* xmemset(node, 0, sizeof(Node)); */
@@ -1195,7 +1195,8 @@ node_new_cclass(void)
 
 #if defined(USE_MULTI_THREAD_SYSTEM) && \
     defined(USE_SHARED_CCLASS_TABLE) && \
-    defined(USE_PARSE_TREE_NODE_RECYCLE)
+    defined(USE_PARSE_TREE_NODE_RECYCLE) && \
+    0
 static Node*
 node_new_cclass_locked(void)
 {
diff --git a/ruby_atomic.h b/ruby_atomic.h
index 4bc9f37e0d..52b93efbd1 100644
--- a/ruby_atomic.h
+++ b/ruby_atomic.h
@@ -42,6 +42,7 @@ typedef unsigned int rb_atomic_t; /* Anything OK */
 #pragma intrinsic(_InterlockedOr)
 #endif
 typedef LONG rb_atomic_t;
+#define SIZEOF_ATOMIC_T SIZEOF_LONG
 
 # define ATOMIC_SET(var, val) InterlockedExchange(&(var), (val))
 # define ATOMIC_INC(var) InterlockedIncrement(&(var))
@@ -143,6 +144,10 @@ ruby_atomic_size_exchange(size_t *ptr, size_t val)
 }
 #endif
 
+#ifndef SIZEOF_ATOMIC_T
+#define SIZEOF_ATOMIC_T SIZEOF_INT
+#endif
+
 #ifndef ATOMIC_SIZE_INC
 # define ATOMIC_SIZE_INC(var) ATOMIC_INC(var)
 #endif