Add --reloop-limit argument (#2960)

Add --reloop-limit argument

Changes

@@ -34,6 +34,7 @@ Verilator 4.202 2021-04-24
 * Add VerilatedCovContext::forcePerInstance (#2793). [Kevin Laeufer]
 * Add FST SystemC tracing (#2806). [Alex Torregrosa]
 * Add PINNOTFOUND warning in place of error (#2868). [Udi Finkelstein]
+* Add --reloop-limit argument (#2943). [Geza Lore]
 * Support overlaps in priority case statements (#2864). [Rupert Swarbrick]
 * Support for null ports (#2875). [Udi Finkelstein]
 * Optimize large lookup tables to static data (#2925). [Geza Lore]

bin/verilator

@@ -378,6 +378,7 @@ detailed descriptions of these arguments.
     --quiet-exit                Don't print the command on failure
     --relative-includes         Resolve includes relative to current file
     --no-relative-cfuncs        Disallow 'this->' in generated functions
+    --reloop-limit              Minimum iterations for forming loops
     --report-unoptflat          Extra diagnostics for UNOPTFLAT
     --rr                        Run Verilator and record with rr
     --savable                   Enable model save-restore

docs/guide/exe_verilator.rst

@@ -915,6 +915,15 @@ Summary:
    the path of the referencing file, instead of relative to the current
    directory.
 
+.. option:: --reloop-limit
+
+   Rarely needed. Verilator attempts to turn some common sequences of
+   statements into loops in the output. This argument specifies the minimum
+   number of iterations the resulting loop needs to have in order to perform
+   this transformation. Default limit is 40. A smaller number may slightly
+   improve C++ compilation time on designs where these sequences are common,
+   however effect on model performance requires benchmarking.
+
 .. option:: --report-unoptflat
 
    Extra diagnostics for UNOPTFLAT warnings. This includes for each loop,

src/V3Options.cpp

@@ -1242,6 +1242,10 @@ void V3Options::parseOptsList(FileLine* fl, const string& optdir, int argc, char
             fl->v3warn(DEPRECATED, "Deprecated --no-relative-cfuncs, unnecessary with C++11.");
     });
     DECL_OPTION("-relative-includes", OnOff, &m_relativeIncludes);
+    DECL_OPTION("-reloop-limit", CbVal, [this, fl](const char* valp) {
+        m_reloopLimit = std::atoi(valp);
+        if (m_reloopLimit < 2) { fl->v3error("--reloop-limit must be >= 2: " << valp); }
+    });
     DECL_OPTION("-report-unoptflat", OnOff, &m_reportUnoptflat);
     DECL_OPTION("-rr", CbCall, []() {});  // Processed only in bin/verilator shell
 

src/V3Options.h

@@ -295,6 +295,7 @@ private:
     int         m_outputSplitCFuncs = -1;  // main switch: --output-split-cfuncs
     int         m_outputSplitCTrace = -1;  // main switch: --output-split-ctrace
     int         m_pinsBv = 65;       // main switch: --pins-bv
+    int         m_reloopLimit = 40; // main switch: --reloop-limit
     VOptionBool m_skipIdentical;  // main switch: --skip-identical
     int         m_threads = 0;      // main switch: --threads (0 == --no-threads)
     int         m_threadsMaxMTasks = 0;  // main switch: --threads-max-mtasks
@@ -491,6 +492,7 @@ public:
     int outputSplitCFuncs() const { return m_outputSplitCFuncs; }
     int outputSplitCTrace() const { return m_outputSplitCTrace; }
     int pinsBv() const { return m_pinsBv; }
+    int reloopLimit() const { return m_reloopLimit; }
     VOptionBool skipIdentical() const { return m_skipIdentical; }
     int threads() const { return m_threads; }
     int threadsMaxMTasks() const { return m_threadsMaxMTasks; }

src/V3Reloop.cpp

@@ -39,8 +39,6 @@
 
 #include <algorithm>
 
-constexpr unsigned RELOOP_MIN_ITERS = 40;  // Need at least this many loops to do this optimization
-
 //######################################################################
 
 class ReloopVisitor final : public AstNVisitor {
@@ -85,7 +83,7 @@ private:
             uint32_t items = m_mgIndexHi - m_mgIndexLo + 1;
             UINFO(9, "End merge iter=" << items << " " << m_mgIndexHi << ":" << m_mgIndexLo << " "
                                        << m_mgOffset << " " << m_mgAssignps[0] << endl);
-            if (items >= RELOOP_MIN_ITERS) {
+            if (items >= static_cast<uint32_t>(v3Global.opt.reloopLimit())) {
                 UINFO(6, "Reloop merging items=" << items << " " << m_mgIndexHi << ":"
                                                  << m_mgIndexLo << " " << m_mgOffset << " "
                                                  << m_mgAssignps[0] << endl);

test_regress/t/t_reloop_offset_lim_63.pl

@@ -0,0 +1,37 @@
+#!/usr/bin/env 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.
+# SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0
+
+scenarios(simulator => 1);
+
+top_filename("t/t_reloop_offset.v");
+golden_filename("t/t_reloop_offset.out");
+
+compile(
+    verilator_flags2 => ["-unroll-count 1024",
+                         $Self->wno_unopthreads_for_few_cores(),
+                         "--reloop-limit 63",
+                         "--stats"],
+    );
+
+execute(
+    check_finished => 1,
+    expect_filename => $Self->{golden_filename},
+    );
+
+if ($Self->{vlt}) {
+    # Note, with vltmt this might be split differently, so only checking vlt
+    file_grep($Self->{stats}, qr/Optimizations, Reloop iterations\s+(\d+)/i,
+              63);
+    file_grep($Self->{stats}, qr/Optimizations, Reloops\s+(\d+)/i,
+              1);
+}
+
+ok(1);
+1;