micro-optimization from drh-micro-optimization branch: Knock perhaps 1% off execution time: guard on TclAsyncReady more efficient when decrementing to zero.

1 files changed, 11 insertions, 9 deletions

diff --git a/generic/tclExecute.c b/generic/tclExecute.c
index d4077f5..0e98222 100644
--- a/generic/tclExecute.c
+++ b/generic/tclExecute.c
@@ -35,14 +35,14 @@
 #endif
 
 /*
- * A mask (should be 2**n-1) that is used to work out when the bytecode engine
- * should call Tcl_AsyncReady() to see whether there is a signal that needs
- * handling.
+ * A counter that is used to work out when the bytecode engine should call
+ * Tcl_AsyncReady() to see whether there is a signal that needs handling, and
+ * other expensive periodic operations.
  */
 
-#ifndef ASYNC_CHECK_COUNT_MASK
-#   define ASYNC_CHECK_COUNT_MASK	63
-#endif /* !ASYNC_CHECK_COUNT_MASK */
+#ifndef ASYNC_CHECK_COUNT
+#   define ASYNC_CHECK_COUNT	64
+#endif /* !ASYNC_CHECK_COUNT */
 
 /*
  * Boolean flag indicating whether the Tcl bytecode interpreter has been
@@ -2115,7 +2115,8 @@ TEBCresume(
      * sporadically: no special need for speed.
      */
 
-    int instructionCount = 0;	/* Counter that is used to work out when to
+    int instructionCount = ASYNC_CHECK_COUNT;
+				/* Counter that is used to work out when to
 				 * call Tcl_AsyncReady() */
     const char *curInstName;
 #ifdef TCL_COMPILE_DEBUG
@@ -2314,10 +2315,11 @@ TEBCresume(
 
     /*
      * Check for asynchronous handlers [Bug 746722]; we do the check every
-     * ASYNC_CHECK_COUNT_MASK instruction, of the form (2**n-1).
+     * ASYNC_CHECK_COUNT instructions.
      */
 
-    if ((instructionCount++ & ASYNC_CHECK_COUNT_MASK) == 0) {
+    if (!(--instructionCount)) {
+	instructionCount = ASYNC_CHECK_COUNT;
 	DECACHE_STACK_INFO();
 	if (TclAsyncReady(iPtr)) {
 	    result = Tcl_AsyncInvoke(interp, result);