Changeset 121382 in webkit for trunk/Source/JavaScriptCore/dfg/DFGNode.h

Timestamp:

Jun 27, 2012, 4:16:10 PM (13 years ago)

Author:

[email protected]

Message:

DFG disassembly should be easier to read
​https://bugs.webkit.org/show_bug.cgi?id=90106

Reviewed by Mark Hahnenberg.

Did a few things:

• Options::showDFGDisassembly now shows OSR exit disassembly as well.

• Phi node dumping doesn't attempt to do line wrapping since it just made the dump harder to read.

• DFG graph disassembly view shows a few additional node types that turn out to be essential for understanding OSR exits.

Put together, these changes reinforce the philosophy that anything needed for computing
OSR exit is just as important as the machine code itself. Of course, we still don't take
that philosophy to its full extreme - for example Phantom nodes are not dumped. We may
revisit that in the future.

• assembler/LinkBuffer.cpp:

(JSC::LinkBuffer::finalizeCodeWithDisassembly):

• assembler/LinkBuffer.h:

(JSC):

• dfg/DFGDisassembler.cpp:

(JSC::DFG::Disassembler::dump):

• dfg/DFGGraph.cpp:

(JSC::DFG::Graph::dumpBlockHeader):

• dfg/DFGNode.h:

(JSC::DFG::Node::willHaveCodeGenOrOSR):

• dfg/DFGOSRExitCompiler.cpp:
• jit/JIT.cpp:

(JSC::JIT::privateCompile):

File:

• trunk/Source/JavaScriptCore/dfg/DFGNode.h (modified) (1 diff)

trunk/Source/JavaScriptCore/dfg/DFGNode.h

@@ -732 +732 @@
     }
     
-    bool willHaveCodeGen()
-    {
-        return shouldGenerate() && op() != Phantom && op() != Nop;
+    bool willHaveCodeGenOrOSR()
+    {
+        switch (op()) {
+        case SetLocal:
+        case Int32ToDouble:
+        case ValueToInt32:
+        case UInt32ToNumber:
+        case DoubleAsInt32:
+            return true;
+        case Phantom:
+        case Nop:
+            return false;
+        default:
+            return shouldGenerate();
+        }
     }