Fix PGO profiling for multi-threaded hierarchical scenarios (#5888)

include/verilated_profiler.h

@@ -214,7 +214,7 @@ public:
     // METHODS
     VlPgoProfiler() = default;
     ~VlPgoProfiler() = default;
-    void write(const char* modelp, const std::string& filename) VL_MT_SAFE;
+    void write(const char* modelp, const std::string& filename, bool firstHierCall) VL_MT_SAFE;
     void addCounter(size_t counter, const std::string& name) {
         VL_DEBUG_IF(assert(counter < N_Entries););
         m_records.emplace_back(Record{name, counter});
@@ -228,7 +228,8 @@ public:
 };
 
 template <std::size_t N_Entries>
-void VlPgoProfiler<N_Entries>::write(const char* modelp, const std::string& filename) VL_MT_SAFE {
+void VlPgoProfiler<N_Entries>::write(const char* modelp, const std::string& filename,
+                                     bool firstHierCall) VL_MT_SAFE {
     static VerilatedMutex s_mutex;
     const VerilatedLockGuard lock{s_mutex};
 
@@ -238,7 +239,7 @@ void VlPgoProfiler<N_Entries>::write(const char* modelp, const std::string& file
     // each will collect is own data correctly.  However when each is
     // destroyed we need to get all the data, not keep overwriting and only
     // get the last model's data.
-    static bool s_firstCall = true;
+    static bool s_firstCall = firstHierCall;
 
     VL_DEBUG_IF(VL_DBG_MSGF("+prof+vlt+file writing to '%s'\n", filename.c_str()););
 
@@ -246,12 +247,14 @@ void VlPgoProfiler<N_Entries>::write(const char* modelp, const std::string& file
     if (VL_UNLIKELY(!fp)) {
         VL_FATAL_MT(filename.c_str(), 0, "", "+prof+vlt+file file not writable");
     }
-    s_firstCall = false;
+    if (s_firstCall) {
+        // TODO Perhaps merge with verilated_coverage output format, so can
+        // have a common merging and reporting tool, etc.
+        fprintf(fp, "// Verilated model profile-guided optimization data dump file\n");
+        fprintf(fp, "`verilator_config\n");
+    }
 
-    // TODO Perhaps merge with verilated_coverage output format, so can
-    // have a common merging and reporting tool, etc.
-    fprintf(fp, "// Verilated model profile-guided optimization data dump file\n");
-    fprintf(fp, "`verilator_config\n");
+    s_firstCall = false;
 
     for (const Record& rec : m_records) {
         fprintf(fp, "profile_data -model \"%s\" -mtask \"%s\" -cost 64'd%" PRIu64 "\n", modelp,

src/V3EmitCSyms.cpp

@@ -731,8 +731,11 @@ void EmitCSyms::emitSymImp() {
         puts("#endif  // VM_TRACE\n");
     }
     if (v3Global.opt.profPgo()) {
+        // Do not overwrite data during the last hierarchical stage.
+        const string firstHierCall
+            = (v3Global.opt.hierBlocks().empty() || v3Global.opt.hierChild()) ? "true" : "false";
         puts("_vm_pgoProfiler.write(\"" + topClassName()
-             + "\", _vm_contextp__->profVltFilename());\n");
+             + "\", _vm_contextp__->profVltFilename(), " + firstHierCall + ");\n");
     }
     puts("}\n");
 

test_regress/t/t_hier_block_perf.py

@@ -11,7 +11,7 @@ import vltest_bootstrap
 
 test.scenarios('vlt_all')
 test.init_benchmarksim()
-test.cycles = (int(test.benchmark) if test.benchmark else 1000000)
+test.cycles = (int(test.benchmark) if test.benchmark else 100000)
 test.sim_time = test.cycles * 10 + 1000
 
 THREADS = int(os.environ["THREADS"]) if "THREADS" in os.environ else 4

test_regress/t/t_pgo_threads_hier.py

@@ -0,0 +1,40 @@
+#!/usr/bin/env python3
+# DESCRIPTION: Verilator: Verilog Test driver/expect definition
+#
+# Copyright 2025 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
+
+import vltest_bootstrap
+
+test.scenarios('vltmt')
+test.top_filename = "t/t_hier_block_perf.v"
+cycles = 100000
+test.sim_time = cycles * 10 + 1000
+
+threads = 2
+flags = ["--hierarchical", "-Wno-UNOPTFLAT", "-DSIM_CYCLES=" + str(cycles)]
+
+test.compile(benchmarksim=1, v_flags2=["--prof-pgo"] + flags, threads=threads)
+
+test.execute(all_run_flags=[
+    "+verilator+prof+exec+start+0",
+    " +verilator+prof+exec+file+/dev/null",
+    " +verilator+prof+vlt+file+" + test.obj_dir + "/profile.vlt"])  # yapf:disable
+
+test.file_grep(test.obj_dir + "/profile.vlt", r'profile_data -model "V' + test.name + '"')
+
+# Differentiate benchmarksim results
+test.name = test.name + "_optimized"
+test.compile(
+    benchmarksim=1,
+    # Intentionally no --prof-pgo here to make sure profile data can be read in
+    # without it (that is: --prof-pgo has no effect on profile_data hash names)
+    v_flags2=flags,
+    threads=threads)
+
+test.execute()
+
+test.passes()

test_regress/t/t_pgo_threads_hier_nested.py

@@ -0,0 +1,42 @@
+#!/usr/bin/env python3
+# DESCRIPTION: Verilator: Verilog Test driver/expect definition
+#
+# Copyright 2025 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
+
+import vltest_bootstrap
+
+test.scenarios('vltmt')
+test.top_filename = "t/t_hier_block_perf.v"
+cycles = 100000
+test.sim_time = cycles * 10 + 1000
+
+threads = 2
+config_file = test.t_dir + "/" + test.name + ".vlt"
+flags = [config_file, "--hierarchical", "-Wno-UNOPTFLAT", "-DSIM_CYCLES=" + str(cycles)]
+
+test.compile(benchmarksim=1, v_flags2=["--prof-pgo"] + flags, threads=threads)
+
+test.execute(all_run_flags=[
+    "+verilator+prof+exec+start+0",
+    " +verilator+prof+exec+file+/dev/null",
+    " +verilator+prof+vlt+file+" + test.obj_dir + "/profile.vlt"])  # yapf:disable
+
+test.file_grep(test.obj_dir + "/profile.vlt", r'profile_data -model "VTest"')
+test.file_grep(test.obj_dir + "/profile.vlt", r'profile_data -model "V' + test.name + '"')
+
+# Differentiate benchmarksim results
+test.name = test.name + "_optimized"
+test.compile(
+    benchmarksim=1,
+    # Intentionally no --prof-pgo here to make sure profile data can be read in
+    # without it (that is: --prof-pgo has no effect on profile_data hash names)
+    v_flags2=flags,
+    threads=threads)
+
+test.execute()
+
+test.passes()

test_regress/t/t_pgo_threads_hier_nested.vlt

@@ -0,0 +1,8 @@
+// DESCRIPTION: Verilator: Verilog Test module
+//
+// This file ONLY is placed into the Public Domain, for any use,
+// without warranty, 2025 by Antmicro.
+// SPDX-License-Identifier: CC0-1.0
+
+`verilator_config
+hier_workers -module "Test" -workers 2