|
|
@@ -339,9 +339,13 @@
|
|
|
// However, compiling this option in does add some additional runtime
|
|
|
// overhead even if it is not used. By default, we enable pipelining
|
|
|
// whenever threads are enabled, assuming that if you have threads,
|
|
|
-// you also want to use piplining. We also enable it at OPTIMIZE
|
|
|
+// you also want to use pipelining. We also enable it at OPTIMIZE
|
|
|
// level 1, since that enables additional runtime checks.
|
|
|
-#defer DO_PIPELINING $[or $[<= $[OPTIMIZE], 1],$[HAVE_THREADS]]
|
|
|
+//#defer DO_PIPELINING $[or $[<= $[OPTIMIZE], 1],$[HAVE_THREADS]]
|
|
|
+
|
|
|
+// Actually, let's *not* assume that threading implies pipelining, at
|
|
|
+// least not until pipelining is less of a performance hit.
|
|
|
+#defer DO_PIPELINING $[<= $[OPTIMIZE], 1]
|
|
|
|
|
|
// Do you want to use one of the alternative malloc implementations?
|
|
|
// This is almost always a good idea on Windows, where the standard
|