diff --git a/Changes b/Changes
index ae6bd77aa..eca429ade 100644
--- a/Changes
+++ b/Changes
@@ -5,6 +5,8 @@ indicates the contributor was also the author of the fix; Thanks!
 
 * Verilator 3.6**
 
+***  Add --output-split-cfuncs for accelerating GCC. [Eugene Weber]
+
 **** Fix $signed mis-extending when input has a WIDTH violation. [Eugene Weber]
 
 * Verilator 3.622 10/17/2006 Stable
diff --git a/bin/verilator b/bin/verilator
index 60a91e7be..83dcea274 100755
--- a/bin/verilator
+++ b/bin/verilator
@@ -377,9 +377,18 @@ C++ file exceeds the specified number of operations, a new file will be
 created.  In addition, any slow routines will be placed into __Slow files.
 This accelerates compilation by as optimization can be disabled on the slow
 routines, and the remaining files can be compiled on parallel machines.
+Using --output-split should have only a trivial impact on performance.
 With GCC 3.3 on a 2GHz Opteron, --output-split 20000 will result in
 splitting into approximately one-minute-compile chunks.
 
+=item --output-split-cfuncs I<statements>
+
+Enables splitting functions in the output .cpp/.sp files into multiple
+functions.  When a generated function exceeds the specified number of
+operations, a new function will be created.  With --output-split, this will
+enable GCC to compile faster, at a small loss in performance that increases
+with smaller statement values.
+
 =item --pins64
 
 Specifies SystemC outputs of 33-64 bits wide should use uint64_t instead of
diff --git a/src/V3Options.cpp b/src/V3Options.cpp
index 6e89bd439..cc9aad580 100644
--- a/src/V3Options.cpp
+++ b/src/V3Options.cpp
@@ -348,6 +348,10 @@ void V3Options::parseOptsList(FileLine* fl, int argc, char** argv) {
 		shift;
 		m_outputSplit = atoi(argv[i]);
 	    }
+	    else if ( !strcmp (sw, "-output-split-cfuncs") ) {
+		shift;
+		m_outputSplitCFuncs = atoi(argv[i]);
+	    }
 	    else if ( !strcmp (sw, "-unroll-count") ) { // Undocumented optimization tweak
 		shift;
 		m_unrollCount = atoi(argv[i]);
@@ -581,6 +585,7 @@ V3Options::V3Options() {
 
     m_inlineMult = 2000;
     m_outputSplit = 0;
+    m_outputSplitCFuncs = 0;
     m_unrollCount = 64;
     m_unrollStmts = 20;
 
diff --git a/src/V3Options.h b/src/V3Options.h
index 29fb5d86d..676e30881 100644
--- a/src/V3Options.h
+++ b/src/V3Options.h
@@ -70,6 +70,7 @@ class V3Options {
 
     int		m_inlineMult;	// main switch: --inline-mult
     int		m_outputSplit;	// main switch: --output-split
+    int		m_outputSplitCFuncs;// main switch: --output-split-cfuncs
     int		m_unrollCount;	// main switch: --unroll-count
     int		m_unrollStmts;	// main switch: --unroll-stmts
 
@@ -152,6 +153,7 @@ class V3Options {
 
     int	   inlineMult() const { return m_inlineMult; }
     int	   outputSplit() const { return m_outputSplit; }
+    int	   outputSplitCFuncs() const { return m_outputSplitCFuncs; }
     int	   unrollCount() const { return m_unrollCount; }
     int	   unrollStmts() const { return m_unrollStmts; }
 
diff --git a/src/V3Order.cpp b/src/V3Order.cpp
index bd338bc1c..8c96e6d17 100644
--- a/src/V3Order.cpp
+++ b/src/V3Order.cpp
@@ -271,6 +271,7 @@ private:
     vector<OrderLoopEndVertex*> m_pmlLoopEndps;	// processInsLoop: End vertex for each color
     vector<OrderLoopBeginVertex*> m_pomLoopMoveps;// processMoveLoop: Loops next nodes are under
     AstCFunc*			m_pomNewFuncp;	// Current function being created
+    int				m_pomNewStmts;	// Statements in function being created
     V3Graph			m_pomGraph;	// Graph of logic elements to move
     V3List<OrderMoveVertex*>	m_pomWaiting;	// List of nodes needing inputs to become ready
 protected:
@@ -688,6 +689,7 @@ public:
 	m_settleVxp = NULL;
 	m_inputsVxp = NULL;
 	m_loopIdMax = LOOPID_FIRST;
+	m_pomNewStmts = 0;
 	if (debug()) m_graph.debug(5); // 3 is default if global debug; we want acyc debugging
     }
     virtual ~OrderVisitor() {
@@ -1374,8 +1376,10 @@ void OrderVisitor::processMoveOne(OrderMoveVertex* vertexp, OrderMoveDomScope* d
     }
     else {  // Normal logic
 	// Make or borrow a CFunc to contain the new statements
-	if (v3Global.opt.profileCFuncs()) {
-	    // Put every statement into a unique function to ease profiling
+	if (v3Global.opt.profileCFuncs()
+	    || (v3Global.opt.outputSplitCFuncs()
+		&& v3Global.opt.outputSplitCFuncs() < m_pomNewStmts)) {
+	    // Put every statement into a unique function to ease profiling or reduce function size
 	    m_pomNewFuncp = NULL;
 	}
 	if (!m_pomNewFuncp && domainp != m_deleteDomainp) {
@@ -1383,6 +1387,7 @@ void OrderVisitor::processMoveOne(OrderMoveVertex* vertexp, OrderMoveDomScope* d
 	    m_pomNewFuncp = new AstCFunc(nodep->fileline(), name, scopep);
 	    m_pomNewFuncp->argTypes(EmitCBaseVisitor::symClassVar());
 	    m_pomNewFuncp->symProlog(true);
+	    m_pomNewStmts = 0;
 	    if (domainp->hasInitial() || domainp->hasSettle()) m_pomNewFuncp->slow(true);
 	    scopep->addActivep(m_pomNewFuncp);
 	    // Where will we be adding the call?
@@ -1402,6 +1407,11 @@ void OrderVisitor::processMoveOne(OrderMoveVertex* vertexp, OrderMoveDomScope* d
 	    pushDeletep(nodep); nodep=NULL;
 	} else {
 	    m_pomNewFuncp->addStmtsp(nodep);
+	    if (v3Global.opt.outputSplitCFuncs()) {
+		// Add in the number of nodes we're adding
+		EmitCBaseCounterVisitor visitor(nodep);
+		m_pomNewStmts += visitor.count();
+	    }
 	}
     }
     processMoveDoneOne (vertexp);