summaryrefslogtreecommitdiffstats
path: root/generic/tclThreadAlloc.c
diff options
context:
space:
mode:
authordgp <dgp@users.sourceforge.net>2015-04-15 19:57:41 (GMT)
committerdgp <dgp@users.sourceforge.net>2015-04-15 19:57:41 (GMT)
commitf32d7a27baaf82615453a772b3371d58909f341f (patch)
treec00ffa340ccef6a6eb22f5f93ef47bb0c4c03ab1 /generic/tclThreadAlloc.c
parent84d6b9aefa8ba771283aa4103db1079b659ee359 (diff)
downloadtcl-f32d7a27baaf82615453a772b3371d58909f341f.zip
tcl-f32d7a27baaf82615453a772b3371d58909f341f.tar.gz
tcl-f32d7a27baaf82615453a772b3371d58909f341f.tar.bz2
Revise the zippy dance that pushes Blocks and Tcl_Objs back the shared pool
so that the overal operation remains a FIFO. This help preserve cache locality where it exists. The price (for now) is a bit more time walking to the end of free lists. If this is important, the addition of a lastPtr field could fix it without difficulty.
Diffstat (limited to 'generic/tclThreadAlloc.c')
-rw-r--r--generic/tclThreadAlloc.c100
1 files changed, 85 insertions, 15 deletions
diff --git a/generic/tclThreadAlloc.c b/generic/tclThreadAlloc.c
index 560556d..92bec44 100644
--- a/generic/tclThreadAlloc.c
+++ b/generic/tclThreadAlloc.c
@@ -135,6 +135,7 @@ static int GetBlocks(Cache *cachePtr, int bucket);
static Block * Ptr2Block(char *ptr);
static char * Block2Ptr(Block *blockPtr, int bucket, unsigned int reqSize);
static void MoveObjs(Cache *fromPtr, Cache *toPtr, int numMove);
+static void PutObjs(Cache *fromPtr, int numMove);
/*
* Local variables defined in this file and initialized at startup.
@@ -271,9 +272,7 @@ TclFreeAllocCache(
*/
if (cachePtr->numObjects > 0) {
- Tcl_MutexLock(objLockPtr);
- MoveObjs(cachePtr, sharedPtr, cachePtr->numObjects);
- Tcl_MutexUnlock(objLockPtr);
+ PutObjs(cachePtr, cachePtr->numObjects);
}
/*
@@ -632,9 +631,7 @@ TclThreadFreeObj(
*/
if (cachePtr->numObjects > NOBJHIGH) {
- Tcl_MutexLock(objLockPtr);
- MoveObjs(cachePtr, sharedPtr, NOBJALLOC);
- Tcl_MutexUnlock(objLockPtr);
+ PutObjs(cachePtr, NOBJALLOC);
}
}
@@ -739,6 +736,60 @@ MoveObjs(
/*
*----------------------------------------------------------------------
*
+ * PutObjs --
+ *
+ * Move Tcl_Obj's from thread cache to shared cache.
+ *
+ * Results:
+ * None.
+ *
+ * Side effects:
+ * None.
+ *
+ *----------------------------------------------------------------------
+ */
+
+static void
+PutObjs(
+ Cache *fromPtr,
+ int numMove)
+{
+ int keep = fromPtr->numObjects - numMove;
+ Tcl_Obj *firstPtr, *lastPtr;
+
+ fromPtr->numObjects = keep;
+ firstPtr = fromPtr->firstObjPtr;
+ if (keep == 0) {
+ fromPtr->firstObjPtr = NULL;
+ } else {
+ do {
+ lastPtr = firstPtr;
+ firstPtr = firstPtr->internalRep.twoPtrValue.ptr1;
+ } while (--keep > 0);
+ lastPtr->internalRep.twoPtrValue.ptr1 = NULL;
+ }
+
+ /* TODO: We could avoid this walk to lastPtr if we kept a lastPtr field */
+ lastPtr = firstPtr;
+ while (lastPtr->internalRep.twoPtrValue.ptr1) {
+ lastPtr = lastPtr->internalRep.twoPtrValue.ptr1;
+ }
+
+ /*
+ * Move all objects as a block - they are already linked to each other, we
+ * just have to update the first and last.
+ */
+
+ Tcl_MutexLock(objLockPtr);
+ lastPtr->internalRep.twoPtrValue.ptr1 = sharedPtr->firstObjPtr;
+ sharedPtr->firstObjPtr = firstPtr;
+ sharedPtr->numObjects += numMove;
+ Tcl_MutexUnlock(objLockPtr);
+}
+
+/*
+ *----------------------------------------------------------------------
+ *
* Block2Ptr, Ptr2Block --
*
* Convert between internal blocks and user pointers.
@@ -848,20 +899,39 @@ PutBlocks(
int bucket,
int numMove)
{
- register Block *lastPtr, *firstPtr;
- register int n = numMove;
-
/*
- * Before acquiring the lock, walk the block list to find the last block
- * to be moved.
+ * We have numFree. Want to shed numMove. So compute how many
+ * Blocks to keep.
*/
- firstPtr = lastPtr = cachePtr->buckets[bucket].firstPtr;
- while (--n > 0) {
+ int keep = cachePtr->buckets[bucket].numFree - numMove;
+ Block *lastPtr, *firstPtr;
+
+ cachePtr->buckets[bucket].numFree = keep;
+ firstPtr = cachePtr->buckets[bucket].firstPtr;
+ if (keep == 0) {
+ cachePtr->buckets[bucket].firstPtr = NULL;
+ } else {
+ do {
+ lastPtr = firstPtr;
+ firstPtr = firstPtr->nextBlock;
+ } while (--keep > 0);
+ lastPtr->nextBlock = NULL;
+ }
+
+ /*
+ * firstPtr now points to the first Block to return to shared.
+ */
+
+ /* TODO: We could avoid this walk to lastPtr if we kept a lastPtr field */
+ lastPtr = firstPtr;
+ while (lastPtr->nextBlock) {
lastPtr = lastPtr->nextBlock;
}
- cachePtr->buckets[bucket].firstPtr = lastPtr->nextBlock;
- cachePtr->buckets[bucket].numFree -= numMove;
+
+ /*
+ * lastPtr now points to the last Block to return to shared.
+ */
/*
* Aquire the lock and place the list of blocks at the front of the shared