Mark infrequently called functions with GCC cold attribute.

Changes

@@ -10,6 +10,8 @@ The contributors that suggested a given feature are shown in []. Thanks!
 
 ****  Fix sign-compare warning in verilated.cpp, bug1437. [Sergey Kvachonok]
 
+****  Mark infrequently called functions with GCC cold attribute.
+
 
 * Verilator 4.014 2019-05-08
 

bin/verilator

@@ -1034,15 +1034,15 @@ developers.
 
 =item --output-split I<bytes>
 
-Enables splitting the output .cpp files into multiple outputs.  When a
+Enables splitting the output .cpp files into multiple outputs.  When a C++
-C++ file exceeds the specified number of operations, a new file will be
+file exceeds the specified number of operations, a new file will be created
-created at the next function boundary.  In addition, any slow routines will
+at the next function boundary.  In addition, any infrequently executed
-be placed into __Slow files.  This accelerates compilation by as
+"cold" routines will be placed into __Slow files.  This accelerates
-optimization can be disabled on the slow routines, and the remaining files
+compilation by as optimization can be disabled on the routines in __Slow,
-can be compiled on parallel machines.  Using --output-split should have
+and the remaining files can be compiled on parallel machines.  Using
-only a trivial impact on performance.  With GCC 3.3 on a 2GHz Opteron,
+--output-split should have only a trivial impact on performance.  With GCC
---output-split 20000 will result in splitting into approximately
+3.3 on a 2GHz Opteron, --output-split 20000 will result in splitting into
-one-minute-compile chunks.
+approximately one-minute-compile chunks.
 
 =item --output-split-cfuncs I<statements>
 
@@ -1871,7 +1871,7 @@ For -cc and -sc mode, it also creates:
 
     {prefix}.cpp                        // Top level C++ file
     {prefix}.h                          // Top level header
-    {prefix}__Slow{__n}.cpp             // Constructors and infrequent routines
+    {prefix}__Slow{__n}.cpp             // Constructors and infrequent cold routines
     {prefix}{__n}.cpp                   // Additional top C++ files (--output-split)
     {prefix}{each_verilog_module}.cpp   // Lower level internal C++ files
     {prefix}{each_verilog_module}.h     // Lower level internal header files

include/verilatedos.h

@@ -36,6 +36,8 @@
 #ifdef __GNUC__
 # define VL_ATTR_ALIGNED(alignment) __attribute__ ((aligned (alignment)))
 # define VL_ATTR_ALWINLINE __attribute__ ((always_inline))
+# define VL_ATTR_COLD __attribute__ ((cold))
+# define VL_ATTR_HOT __attribute__ ((hot))
 # define VL_ATTR_NORETURN __attribute__ ((noreturn))
 # define VL_ATTR_PRINTF(fmtArgNum) __attribute__ ((format (printf, (fmtArgNum), (fmtArgNum)+1)))
 # define VL_ATTR_PURE __attribute__ ((pure))
@@ -70,6 +72,12 @@
 #ifndef VL_ATTR_ALWINLINE
 # define VL_ATTR_ALWINLINE              ///< Inline, even when not optimizing
 #endif
+#ifndef VL_ATTR_COLD
+# define VL_ATTR_COLD                   ///< Function is rarely executed
+#endif
+#ifndef VL_ATTR_HOT
+# define VL_ATTR_HOT                    ///< Function is highly executed
+#endif
 #ifndef VL_ATTR_NORETURN
 # define VL_ATTR_NORETURN               ///< Function does not ever return
 #endif

src/V3EmitC.cpp

@@ -2367,7 +2367,9 @@ void EmitCImp::emitIntFuncDecls(AstNodeModule* modp) {
             if (funcp->ifdef()!="") puts("#ifdef "+funcp->ifdef()+"\n");
             if (funcp->isStatic().trueU()) puts("static ");
             puts(funcp->rtnTypeVoid()); puts(" ");
-	    puts(funcp->name()); puts("("+cFuncArgs(funcp)+");\n");
+	    puts(funcp->name()); puts("("+cFuncArgs(funcp)+")");
+	    if (funcp->slow()) puts(" VL_ATTR_COLD");
+	    puts(";\n");
 	    if (funcp->ifdef()!="") puts("#endif // "+funcp->ifdef()+"\n");
 	}
     }