Inline C functions that are used only once, msg1525.

Changes

@@ -11,6 +11,8 @@ indicates the contributor was also the author of the fix; Thanks!
 
 **    SystemPerl mode is deprecated and now untested.
 
+***   Inline C functions that are used only once, msg1525. [Jie Xu]
+
 
 * Verilator 3.866 2014-11-15
 

bin/verilator

@@ -1399,6 +1399,10 @@ especially if you link in DPI code.  To enable LTO on GCC, pass "-flto" in
 both compilation and link.  Note LTO may cause excessive compile times on
 large designs.
 
+If you are using your own makefiles, you may want to compile the Verilated
+code with -DVL_INLINE_OPT=inline. This will inline functions, however this
+requires that all cpp files be compiled in a single compiler run.
+
 You may uncover further tuning possibilities by profiling the Verilog code.
 Use Verilator's --profile-cfuncs, then GCC's -g -pg.  You can then run
 either oprofile or gprof to see where in the C++ code the time is spent.

bin/verilator_includer

@@ -12,5 +12,9 @@ require 5.005;
 use warnings;
 print "// DESCR"."IPTION: Generated by verilator_includer via makefile\n";
 foreach my $param (@ARGV) {
-    print "#include \"$param\"\n"
+    if ($param =~ /^-D([^=]+)=(.*)/) {
+	print "#define $1 $2\n"
+    } else {
+	print "#include \"$param\"\n"
+    }
 }

include/verilated.mk.in

@@ -23,7 +23,9 @@ CFG_CXXFLAGS_NO_UNUSED = @CFG_CXXFLAGS_NO_UNUSED@
 # Programs
 
 SP_PREPROC	= sp_preproc
-SP_INCLUDER     = $(PERL) $(VERILATOR_ROOT)/bin/verilator_includer
+SP_INCLUDER	= $(VERILATOR_INCLUDER)
+VERILATOR_COVERAGE = $(PERL) $(VERILATOR_ROOT)/bin/verilator_coverage
+VERILATOR_INCLUDER = $(PERL) $(VERILATOR_ROOT)/bin/verilator_includer
 
 ######################################################################
 # Make checks
@@ -151,15 +153,15 @@ VK_USER_OBJS   = $(addsuffix .o, $(VM_USER_CLASSES))
 
 VK_GLOBAL_OBJS = $(addsuffix .o, $(VM_GLOBAL_FAST) $(VM_GLOBAL_SLOW))
 
-ifneq ($(VM_PARALLEL_BUILDS),1)
+ifneq ($(VM_PARALLEL_BUILDS),0)
   # Fast building, all .cpp's in one fell swoop
   # This saves about 5 sec per module, but can be slower if only a little changes
   VK_OBJS += $(VM_PREFIX)__ALLcls.o   $(VM_PREFIX)__ALLsup.o
   all_cpp:   $(VM_PREFIX)__ALLcls.cpp $(VM_PREFIX)__ALLsup.cpp
   $(VM_PREFIX)__ALLcls.cpp: $(VK_CLASSES_CPP)
-	$(SP_INCLUDER) $^ > $@
+	$(VERILATOR_INCLUDER) -DVL_INCLUDE_OPT=include $^ > $@
   $(VM_PREFIX)__ALLsup.cpp: $(VK_SUPPORT_CPP)
-	$(SP_INCLUDER) $^ > $@
+	$(VERILATOR_INCLUDER) -DVL_INCLUDE_OPT=include $^ > $@
 else
   #Slow way of building... Each .cpp file by itself
   VK_OBJS += $(addsuffix .o, $(VM_CLASSES) $(VM_SUPPORT))

include/verilatedos.h

@@ -105,6 +105,13 @@
 # define VL_UNIQUE_PTR auto_ptr
 #endif
 
+//=========================================================================
+// Optimization
+
+#ifndef VL_INLINE_OPT
+# define VL_INLINE_OPT		///< "inline" if compiling all objects in single compiler run
+#endif
+
 //=========================================================================
 // Warning disabled
 

src/V3AstNodes.h

@@ -4699,6 +4699,7 @@ private:
     bool	m_formCallTree:1;	// Make a global function to call entire tree of functions
     bool	m_slow:1;		// Slow routine, called once or just at init time
     bool	m_funcPublic:1;		// From user public task/function
+    bool	m_isInline:1;		// Inline function
     bool	m_isStatic:1;		// Function is declared static (no this)
     bool	m_symProlog:1;		// Setup symbol table for later instructions
     bool	m_entryPoint:1;		// User may call into this top level function
@@ -4719,6 +4720,7 @@ public:
 	m_formCallTree = false;
 	m_slow = false;
 	m_funcPublic = false;
+	m_isInline = false;
 	m_isStatic = true;	// Note defaults to static, later we see where thisp is needed
 	m_symProlog = false;
 	m_entryPoint = false;
@@ -4748,6 +4750,7 @@ public:
     string	rtnTypeVoid() const { return ((m_rtnType=="") ? "void" : m_rtnType); }
     bool	dontCombine() const { return m_dontCombine || funcType()!=AstCFuncType::FT_NORMAL; }
     void	dontCombine(bool flag) { m_dontCombine = flag; }
+    bool	dontInline() const { return !dontCombine() && !slow() && !skipDecl() && !funcPublic(); }
     bool	skipDecl() const { return m_skipDecl; }
     void	skipDecl(bool flag) { m_skipDecl = flag; }
     bool	declPrivate() const { return m_declPrivate; }
@@ -4762,6 +4765,8 @@ public:
     string	argTypes() const { return m_argTypes; }
     void	funcType(AstCFuncType flag) { m_funcType = flag; }
     AstCFuncType funcType() const { return m_funcType; }
+    bool	isInline() const { return m_isInline; }
+    void	isInline(bool flag) { m_isInline = flag; }
     bool	isStatic() const { return m_isStatic; }
     void	isStatic(bool flag) { m_isStatic = flag; }
     bool	symProlog() const { return m_symProlog; }

src/V3Branch.cpp

@@ -21,6 +21,9 @@
 //	At each IF/(IF else).
 //	   Count underneath $display/$stop statements.
 //	   If more on if than else, this branch is unlikely, or vice-versa.
+//	At each FTASKREF,
+//	   Count calls into the function
+//	Then, if FTASK is called only once, add inline attribute
 //
 //*************************************************************************
 
@@ -40,9 +43,18 @@
 
 class BranchVisitor : public AstNVisitor {
 private:
+    // NODE STATE
+    // Entire netlist:
+    //  AstFTask::user1()	-> int.  Number of references
+    AstUser1InUse	m_inuser1;
+
+    // TYPES
+    typedef vector<AstCFunc*> CFuncVec;
+
     // STATE
     int		m_likely;	// Excuses for branch likely taken
     int		m_unlikely;	// Excuses for branch likely not taken
+    CFuncVec	m_cfuncsp;	// List of all tasks
 
     // METHODS
     static int debug() {
@@ -55,6 +67,12 @@ private:
 	m_likely = false;
 	m_unlikely = false;
     }
+    void checkUnlikely(AstNode* nodep) {
+	if (nodep->isUnlikely()) {
+	    UINFO(4,"  UNLIKELY: "<<nodep<<endl);
+	    m_unlikely++;
+	}
+    }
 
     // VISITORS
     virtual void visit(AstNodeIf* nodep, AstNUser*) {
@@ -83,20 +101,37 @@ private:
 	m_likely = lastLikely;
 	m_unlikely = lastUnlikely;
     }
-    virtual void visit(AstNode* nodep, AstNUser*) {
-	// Default: Just iterate
-	if (nodep->isUnlikely()) {
-	    UINFO(4,"  UNLIKELY: "<<nodep<<endl);
-	    m_unlikely++;
-	}
+    virtual void visit(AstCCall* nodep, AstNUser*) {
+	checkUnlikely(nodep);
+	nodep->funcp()->user1Inc();
 	nodep->iterateChildren(*this);
     }
+    virtual void visit(AstCFunc* nodep, AstNUser*) {
+	checkUnlikely(nodep);
+	m_cfuncsp.push_back(nodep);
+	nodep->iterateChildren(*this);
+    }
+    virtual void visit(AstNode* nodep, AstNUser*) {
+	checkUnlikely(nodep);
+	nodep->iterateChildren(*this);
+    }
+
+    // METHODS
+    void calc_tasks() {
+	for (CFuncVec::iterator it=m_cfuncsp.begin(); it!=m_cfuncsp.end(); ++it) {
+	    AstCFunc* nodep = *it;
+	    if (!nodep->dontInline()) {
+		nodep->isInline(true);
+	    }
+	}
+    }
 
 public:
     // CONSTUCTORS
     BranchVisitor(AstNetlist* nodep) {
 	reset();
 	nodep->iterateChildren(*this);
+	calc_tasks();
     }
     virtual ~BranchVisitor() {}
 };

src/V3EmitC.cpp

@@ -829,6 +829,7 @@ class EmitCImp : EmitCStmts {
 	splitSizeInc(nodep);
 
 	puts("\n");
+	if (nodep->isInline()) puts("VL_INLINE_OPT ");
 	puts(nodep->rtnTypeVoid()); puts(" ");
 	puts(modClassName(m_modp)+"::"+nodep->name()
 	     +"("+cFuncArgs(nodep)+") {\n");

test_regress/Makefile_obj

@@ -52,7 +52,7 @@ endif
 
 #Our own compile rules; Faster compile, all in one file
 $(VM_PREFIX)__ALLboth.cpp: $(VK_CLASSES_CPP) $(VK_SUPPORT_CPP)
-	$(SP_INCLUDER) $^ > $@
+	$(VERILATOR_INCLUDER) -DVL_INLINE_OPT=inline $^ > $@
 
 $(VM_PREFIX)__ALLboth.o: $(VM_PREFIX)__ALLboth.cpp
 	$(OBJCACHE) $(CXX) $(CXXFLAGS) $(CPPFLAGS) $(OPT_FAST) -c -o $@ $<