Oilpan: Fix a performance regression in CSS.ClassInvalidation

Source/platform/heap/ThreadState.cpp

Index: Source/platform/heap/ThreadState.cpp
diff --git a/Source/platform/heap/ThreadState.cpp b/Source/platform/heap/ThreadState.cpp
index 8d2eb9f6f6f2495de33ea25adf3dc679516bd960..98b05b029e71104926deaec26a83b7df72ff826e 100644
--- a/Source/platform/heap/ThreadState.cpp
+++ b/Source/platform/heap/ThreadState.cpp
@@ -631,7 +631,7 @@ bool ThreadState::shouldForceMemoryPressureGC()
 
     // If we're consuming too much memory, trigger a conservative GC
     // aggressively. This is a safe guard to avoid OOM.
-    return judgeGCThreshold(0, 1.5);
+    return judgeGCThreshold(1024 * 1024, 1.5);

[haraken, 2015/09/20 13:36:41]

Note: This change will be harmless, so I'd like to just land this.

Note: 0 causes the regression. Any value larger than 0 fixes the regression. It
seems that we're hitting some compiler bug?

[sof, 2015/09/21 05:59:28]

This is a good find; how about changing judgeGCThreshold()'s first argument (the
minimum allocation it requires) to non-zero values.

 }
 
 bool ThreadState::shouldScheduleIdleGC()