Fix gate optimization out of memory, add --gate-stmts, bug1260.

Changes

@@ -4,6 +4,8 @@ The contributors that suggested a given feature are shown in []. Thanks!
 
 * Verilator 3.919 devel
 
+****  Fix gate optimization out of memory, add --gate-stmts, bug1260. [Alex Solomatnikov]
+
 ****  Fix compile error on public real parameters by suppressing, bug1261. [Alex Solomatnikov]
 
 ****  Fix input-only tristate comparisons, bug1267. [Alexis G]

bin/verilator

@@ -289,6 +289,7 @@ descriptions in the next sections for more information.
     --getenv <var>              Get environment variable with defaults
     --help                      Display this help
      -I<dir>                    Directory to search for includes
+    --gate-stmts <value>        Tune gate optimizer depth
     --if-depth <value>          Tune IFDEPTH warning
      +incdir+<dir>              Directory to search for includes
     --inhibit-sim               Create function to turn off sim
@@ -750,6 +751,11 @@ them properly, e.g. as -GSTR="\"My String\"" or -GSTR='"My String"'.
 
 =back
 
+=item --gate-stmts I<value>
+
+Rarely needed.  Set the maximum number of statements that may be present in
+an equation for the gate substitution optimization to inline that equation.
+
 =item --gdb
 
 Run Verilator underneath an interactive GDB (or VERILATOR_GDB environment

src/V3Gate.cpp

@@ -195,6 +195,7 @@ private:
     bool		m_buffersOnly;	// Set when we only allow simple buffering, no equations (for clocks)
     AstNodeVarRef*	m_lhsVarRef;	// VarRef on lhs of assignment (what we're replacing)
     bool		m_dedupe;	// Set when we use isGateDedupable instead of isGateOptimizable
+    int                 m_ops;  // Operation count
 
     // METHODS
     void clearSimple(const char* because) {
@@ -205,6 +206,7 @@ private:
     }
     // VISITORS
     virtual void visit(AstNodeVarRef* nodep) {
+        ++m_ops;
 	nodep->iterateChildren(*this);
 	// We only allow a LHS ref for the var being set, and a RHS ref for something else being read.
 	if (nodep->varScopep()->varp()->isSc()) {
@@ -254,6 +256,9 @@ private:
     virtual void visit(AstNode* nodep) {
 	// *** Special iterator
 	if (!m_isSimple) return;	// Fastpath
+        if (++m_ops > v3Global.opt.gateStmts()) {
+            clearSimple("--gate-stmts exceeded");
+        }
 	if (!(m_dedupe ? nodep->isGateDedupable() : nodep->isGateOptimizable())
 	    || !nodep->isPure()
 	    || nodep->isBrancher()) {
@@ -270,6 +275,7 @@ public:
 	m_buffersOnly = buffersOnly;
 	m_lhsVarRef = NULL;
 	m_dedupe = dedupe;
+        m_ops = 0;
 	// Iterate
 	nodep->accept(*this);
 	// Check results
@@ -327,7 +333,7 @@ private:
     // METHODS
     void iterateNewStmt(AstNode* nodep, const char* nonReducibleReason, const char* consumeReason) {
 	if (m_scopep) {
-	    UINFO(4,"   STMT "<<nodep<<endl);
+            UINFO(5,"   STMT "<<nodep<<endl);
 	    // m_activep is null under AstCFunc's, that's ok.
 	    m_logicVertexp = new GateLogicVertex(&m_graph, m_scopep, nodep, m_activep, m_inSlow);
 	    if (nonReducibleReason) {

src/V3Options.cpp

@@ -787,6 +787,10 @@ void V3Options::parseOptsList(FileLine* fl, const string& optdir, int argc, char
 	    else if ( !strncmp (sw, "-G", strlen("-G"))) {
 		addParameter(string (sw+strlen("-G")), false);
 	    }
+            else if ( !strcmp (sw, "-gate-stmts") && (i+1)<argc ) {
+                shift;
+                m_gateStmts = atoi(argv[i]);
+            }
 	    else if ( !strcmp (sw, "-getenv") && (i+1)<argc ) {
 		shift;
 		cout<<V3Options::getenvBuiltins(argv[i])<<endl;
@@ -1252,6 +1256,7 @@ V3Options::V3Options() {
 
     m_convergeLimit = 100;
     m_dumpTree = 0;
+    m_gateStmts = 100;
     m_ifDepth = 0;
     m_inlineMult = 2000;
     m_moduleRecursion = 100;

src/V3Options.h

@@ -107,6 +107,7 @@ class V3Options {
 
     int		m_convergeLimit;// main switch: --converge-limit
     int		m_dumpTree;	// main switch: --dump-tree
+    int         m_gateStmts;    // main switch: --gate-stmts
     int		m_ifDepth;	// main switch: --if-depth
     int		m_inlineMult;	// main switch: --inline-mult
     int		m_moduleRecursion;// main switch: --module-recursion-depth
@@ -254,6 +255,7 @@ class V3Options {
 
     int	   convergeLimit() const { return m_convergeLimit; }
     int    dumpTree() const { return m_dumpTree; }
+    int    gateStmts() const { return m_gateStmts; }
     int	   ifDepth() const { return m_ifDepth; }
     int	   inlineMult() const { return m_inlineMult; }
     int	   moduleRecursionDepth() const { return m_moduleRecursion; }

test_regress/t/t_gate_chained.pl

@@ -0,0 +1,56 @@
+#!/usr/bin/perl
+if (!$::Driver) { use FindBin; exec("$FindBin::Bin/bootstrap.pl", @ARGV, $0); die; }
+# DESCRIPTION: Verilator: Verilog Test driver/expect definition
+#
+# Copyright 2003 by Wilson Snyder. This program is free software; you can
+# redistribute it and/or modify it under the terms of either the GNU
+# Lesser General Public License Version 3 or the Perl Artistic License
+# Version 2.0.
+use IO::File;
+
+sub gen {
+    my $filename = shift;
+
+    my $fh = IO::File->new(">$filename");
+    $fh->print("// Generated by t_gate_chained.pl\n");
+    $fh->print("module t (clk,i,sel,o);\n");
+    $fh->print("  input clk;\n");
+    $fh->print("  input [63:0] i;\n");
+    $fh->print("  input [15:0] sel;\n");
+    $fh->print("  output [63:0] o;\n");
+    $fh->print("\n");
+    my $prev = "i";
+    my $n = 9000;
+    for (my $i=1; $i<$n; ++$i) {
+        $fh->printf("  wire [63:0] ass%04x = (sel == 16'h%04x) ? 64'h0 : $prev;\n", $i, $i);
+        $prev = sprintf("ass%04x", $i);
+    }
+
+    $fh->print("\n");
+    $fh->print("  wire [63:0] o = $prev;\n");
+
+    $fh->print("\n");
+    $fh->print("  always @ (posedge clk) begin\n");
+    $fh->print('    $write("*-* All Finished *-*\n");',"\n");
+    $fh->print('    $finish;',"\n");
+    $fh->print("  end\n");
+    $fh->print("endmodule\n");
+}
+
+top_filename("$Self->{obj_dir}/t_gate_chained.v");
+
+gen($Self->{top_filename});
+
+compile (
+    verilator_flags2=>["--stats --x-assign fast --x-initial fast"],
+    );
+
+execute (
+    check_finished=>1,
+    );
+
+# Must be <<9000 above to prove this worked
+file_grep ($Self->{stats}, qr/Optimizations, Gate sigs deleted\s+(\d+)/i, 8575);
+
+ok(1);
+1;