From 8fd038f88e6741763d5c54835bdb651c5fea0f72 Mon Sep 17 00:00:00 2001 From: Wilson Snyder Date: Tue, 30 Apr 2024 19:46:54 -0400 Subject: [PATCH] Add `--localize-max-size` option and optimization (#5072). --- Changes | 3 ++- bin/verilator | 1 + docs/guide/exe_verilator.rst | 5 ++++ src/V3Localize.cpp | 9 ++++--- src/V3Options.cpp | 7 +++--- src/V3Options.h | 2 ++ test_regress/t/t_opt_localize_max_size.pl | 25 +++++++++++++++++++ test_regress/t/t_opt_localize_max_size.v | 12 +++++++++ test_regress/t/t_opt_localize_max_size_1.pl | 27 +++++++++++++++++++++ 9 files changed, 84 insertions(+), 7 deletions(-) create mode 100755 test_regress/t/t_opt_localize_max_size.pl create mode 100644 test_regress/t/t_opt_localize_max_size.v create mode 100755 test_regress/t/t_opt_localize_max_size_1.pl diff --git a/Changes b/Changes index 5a954ecd7..c120a6715 100644 --- a/Changes +++ b/Changes @@ -15,9 +15,10 @@ Verilator 5.025 devel * Support __en/__out signals on top level inout ports (#4812) (#4856). [Paul Wright] * Support empty queue as dynarray default value (#5055). [Arkadiusz Kozdra, Antmicro Ltd.] +* Add error on zero width select (#5028). * Add CITATION.cff (#5057) (#5058). [Gijs Burghoorn] * Add VPI eval needed tracking (#5065). [Todd Strader] -* Add error on zero width select (#5028). +* Add `--localize-max-size` option and optimization (#5072). * Fix missing flex include path variable (#4970) (#4971). [Christopher Taylor] * Fix missing parameters with comma to be errors (#4979) (#5012). [Paul Swirhun] * Fix bound queue printing (#5032). [Aleksander Kiryk, Antmicro Ltd.] diff --git a/bin/verilator b/bin/verilator index 31f172144..264f0d6c2 100755 --- a/bin/verilator +++ b/bin/verilator @@ -387,6 +387,7 @@ detailed descriptions of these arguments. --lib-create Create a DPI library +libext++[ext]... Extensions for finding modules --lint-only Lint, but do not make output + --localize-max-size Tune localize optimization variable size --make Generate scripts for specified build tool -MAKEFLAGS Arguments to pass to make during --build --main Generate C++ main() file diff --git a/docs/guide/exe_verilator.rst b/docs/guide/exe_verilator.rst index a7d7f2b9d..453baa1ab 100644 --- a/docs/guide/exe_verilator.rst +++ b/docs/guide/exe_verilator.rst @@ -817,6 +817,11 @@ Summary: If the design is not to be completely Verilated, see also the :vlopt:`--bbox-sys` and :vlopt:`--bbox-unsup` options. +.. option:: --localize-max-size + + Rarely needed. Set the maximum variable size in bytes for it to be + subject to localizing-to-stack optimization. Defaults to 1024. + .. option:: --make Generates a script for the specified build tool. diff --git a/src/V3Localize.cpp b/src/V3Localize.cpp index 7a16c5352..0b59c8888 100644 --- a/src/V3Localize.cpp +++ b/src/V3Localize.cpp @@ -63,9 +63,12 @@ class LocalizeVisitor final : public VNVisitor { // METHODS bool isOptimizable(AstVarScope* nodep) { - return !nodep->user1() || // Not marked as not optimizable, or ... - (nodep->varp()->varType() == VVarType::BLOCKTEMP - && m_accessors(nodep).size() == 1); // .. a block temp used in a single CFunc + return ((!nodep->user1() // Not marked as not optimizable, or ... + // .. a block temp used in a single CFunc + || (nodep->varp()->varType() == VVarType::BLOCKTEMP + && m_accessors(nodep).size() == 1)) + // and under size limit + && nodep->varp()->dtypep()->widthTotalBytes() <= v3Global.opt.localizeMaxSize()); } static bool existsNonLeaf(const std::unordered_set& funcps) { diff --git a/src/V3Options.cpp b/src/V3Options.cpp index 9c09d5e1a..3b828c712 100644 --- a/src/V3Options.cpp +++ b/src/V3Options.cpp @@ -1307,6 +1307,9 @@ void V3Options::parseOptsList(FileLine* fl, const string& optdir, int argc, }); DECL_OPTION("-LDFLAGS", CbVal, callStrSetter(&V3Options::addLdLibs)); + DECL_OPTION("-l2-name", Set, &m_l2Name); + DECL_OPTION("-no-l2name", CbCall, [this]() { m_l2Name = ""; }).undocumented(); // Historical + DECL_OPTION("-l2name", CbCall, [this]() { m_l2Name = "v"; }).undocumented(); // Historical const auto setLang = [this, fl](const char* valp) { const V3LangCode optval{valp}; if (optval.legal()) { @@ -1323,9 +1326,7 @@ void V3Options::parseOptsList(FileLine* fl, const string& optdir, int argc, DECL_OPTION("-language", CbVal, setLang); DECL_OPTION("-lib-create", Set, &m_libCreate); DECL_OPTION("-lint-only", OnOff, &m_lintOnly); - DECL_OPTION("-l2-name", Set, &m_l2Name); - DECL_OPTION("-no-l2name", CbCall, [this]() { m_l2Name = ""; }).undocumented(); // Historical - DECL_OPTION("-l2name", CbCall, [this]() { m_l2Name = "v"; }).undocumented(); // Historical + DECL_OPTION("-localize-max-size", Set, &m_localizeMaxSize); DECL_OPTION("-main-top-name", Set, &m_mainTopName); DECL_OPTION("-MAKEFLAGS", CbVal, callStrSetter(&V3Options::addMakeFlags)); diff --git a/src/V3Options.h b/src/V3Options.h index 4439551a3..4907b8534 100644 --- a/src/V3Options.h +++ b/src/V3Options.h @@ -309,6 +309,7 @@ private: int m_instrCountDpi = 200; // main switch: --instr-count-dpi bool m_jsonEditNums = true; // main switch: --no-json-edit-nums bool m_jsonIds = true; // main switch: --no-json-ids + int m_localizeMaxSize = 1024; // main switch: --localize-max-size VOptionBool m_makeDepend; // main switch: -MMD int m_maxNumWidth = 65536; // main switch: --max-num-width int m_moduleRecursion = 100; // main switch: --module-recursion-depth @@ -546,6 +547,7 @@ public: int ifDepth() const { return m_ifDepth; } int inlineMult() const { return m_inlineMult; } int instrCountDpi() const { return m_instrCountDpi; } + int localizeMaxSize() const { return m_localizeMaxSize; } bool jsonEditNums() const { return m_jsonEditNums; } bool jsonIds() const { return m_jsonIds; } VOptionBool makeDepend() const { return m_makeDepend; } diff --git a/test_regress/t/t_opt_localize_max_size.pl b/test_regress/t/t_opt_localize_max_size.pl new file mode 100755 index 000000000..a7a0b1e7b --- /dev/null +++ b/test_regress/t/t_opt_localize_max_size.pl @@ -0,0 +1,25 @@ +#!/usr/bin/env perl +if (!$::Driver) { use FindBin; exec("$FindBin::Bin/bootstrap.pl", @ARGV, $0); die; } +# DESCRIPTION: Verilator: Verilog Test driver/expect definition +# +# Copyright 2003 by Wilson Snyder. This program is free software; you +# can redistribute it and/or modify it under the terms of either the GNU +# Lesser General Public License Version 3 or the Perl Artistic License +# Version 2.0. +# SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0 + +scenarios(vlt => 1); + +compile( + verilator_flags2 => ["--stats"], + ); + +execute( + check_finished => 1, + ); + +# Value must differ from that in t_opt_localize_max_size.pl +file_grep($Self->{stats}, qr/Optimizations, Vars localized\s+(\d+)/i, 5); + +ok(1); +1; diff --git a/test_regress/t/t_opt_localize_max_size.v b/test_regress/t/t_opt_localize_max_size.v new file mode 100644 index 000000000..b25bbf7ab --- /dev/null +++ b/test_regress/t/t_opt_localize_max_size.v @@ -0,0 +1,12 @@ +// DESCRIPTION: Verilator: Verilog Test module +// +// This file ONLY is placed under the Creative Commons Public Domain, for +// any use, without warranty, 2016 by Wilson Snyder. +// SPDX-License-Identifier: CC0-1.0 + +module t; + initial begin + $write("*-* All Finished *-*\n"); + $finish; + end +endmodule diff --git a/test_regress/t/t_opt_localize_max_size_1.pl b/test_regress/t/t_opt_localize_max_size_1.pl new file mode 100755 index 000000000..f59d5f62e --- /dev/null +++ b/test_regress/t/t_opt_localize_max_size_1.pl @@ -0,0 +1,27 @@ +#!/usr/bin/env perl +if (!$::Driver) { use FindBin; exec("$FindBin::Bin/bootstrap.pl", @ARGV, $0); die; } +# DESCRIPTION: Verilator: Verilog Test driver/expect definition +# +# Copyright 2003 by Wilson Snyder. This program is free software; you +# can redistribute it and/or modify it under the terms of either the GNU +# Lesser General Public License Version 3 or the Perl Artistic License +# Version 2.0. +# SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0 + +scenarios(vlt => 1); + +top_filename("t/t_opt_localize_max_size.v"); + +compile( + verilator_flags2 => ["--stats --localize-max-size 1"], + ); + +execute( + check_finished => 1, + ); + +# Value must differ from that in t_opt_localize_max_size.pl +file_grep($Self->{stats}, qr/Optimizations, Vars localized\s+(\d+)/i, 4); + +ok(1); +1;