Initialize profile file once

Signed-off-by: Bartłomiej Chmiel <bchmiel@antmicro.com>

include/verilated_profiler.h

@@ -225,7 +225,10 @@ public:
     // METHODS
     VlPgoProfiler() = default;
     ~VlPgoProfiler() = default;
-    void write(const char* modelp, const std::string& filename, bool firstHierCall) VL_MT_SAFE;
+    VL_UNMOVABLE(VlPgoProfiler);
+    VL_UNCOPYABLE(VlPgoProfiler);
+    void configure(const std::string& filename) VL_MT_SAFE;
+    void write(const char* modelp, const std::string& filename) 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});
@@ -239,8 +242,7 @@ public:
 };
 
 template <std::size_t N_Entries>
-void VlPgoProfiler<N_Entries>::write(const char* modelp, const std::string& filename,
+void VlPgoProfiler<N_Entries>::configure(const std::string& filename) VL_MT_SAFE {
-                                     bool firstHierCall) VL_MT_SAFE {
     static VerilatedMutex s_mutex;
     const VerilatedLockGuard lock{s_mutex};
 
@@ -250,22 +252,33 @@ 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 = firstHierCall;
 
-    VL_DEBUG_IF(VL_DBG_MSGF("+prof+vlt+file writing to '%s'\n", filename.c_str()););
+    FILE* const fp = std::fopen(filename.c_str(), "w");
-
-    FILE* const fp = std::fopen(filename.c_str(), s_firstCall ? "w" : "a");
     if (VL_UNLIKELY(!fp)) {
         VL_FATAL_MT(filename.c_str(), 0, "", "+prof+vlt+file file not writable");
     }
-    if (s_firstCall) {
+
-        // TODO Perhaps merge with verilated_coverage output format, so can
+    VL_DEBUG_IF(VL_DBG_MSGF("+prof+vlt+file initializing '%s'\n", filename.c_str()););
-        // have a common merging and reporting tool, etc.
+
-        fprintf(fp, "// Verilated model profile-guided optimization data dump file\n");
+    // TODO Perhaps merge with verilated_coverage output format, so can
-        fprintf(fp, "`verilator_config\n");
+    // have a common merging and reporting tool, etc.
+    fprintf(fp, "// Verilated model profile-guided optimization data dump file\n");
+    fprintf(fp, "`verilator_config\n");
+
+    std::fclose(fp);
+}
+
+template <std::size_t N_Entries>
+void VlPgoProfiler<N_Entries>::write(const char* modelp, const std::string& filename) VL_MT_SAFE {
+    static VerilatedMutex s_mutex;
+    const VerilatedLockGuard lock{s_mutex};
+
+    FILE* const fp = std::fopen(filename.c_str(), "a");
+    if (VL_UNLIKELY(!fp)) {
+        VL_FATAL_MT(filename.c_str(), 0, "", "+prof+vlt+file file not writable");
     }
 
-    s_firstCall = false;
+    VL_DEBUG_IF(VL_DBG_MSGF("+prof+vlt+file writing to '%s'\n", filename.c_str()););
 
     for (const Record& rec : m_records) {
         fprintf(fp, "profile_data -model \"%s\" -mtask \"%s\" -cost 64'd%" PRIu64 "\n", modelp,

src/V3EmitCSyms.cpp

@@ -731,11 +731,8 @@ 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(), " + firstHierCall + ");\n");
+             + "\", _vm_contextp__->profVltFilename());\n");
     }
     puts("}\n");
 
@@ -837,6 +834,9 @@ void EmitCSyms::emitSymImp() {
 
     if (v3Global.opt.profPgo()) {
         puts("// Configure profiling for PGO\n");
+        if (!v3Global.opt.hierChild()) {
+            puts("_vm_pgoProfiler.configure(_vm_contextp__->profVltFilename());\n");
+        }
         if (v3Global.opt.mtasks()) {
             v3Global.rootp()->topModulep()->foreach([&](const AstExecGraph* execGraphp) {
                 for (const V3GraphVertex& vtx : execGraphp->depGraphp()->vertices()) {

test_regress/t/t_hier_block_perf.v

@@ -18,6 +18,7 @@ module t (/*AUTOARG*/
 
    generate
       for (genvar i = 0; i < `CORES; ++i) Core core(clk);
+      for (genvar i = 0; i < `CORES; ++i) CoreHier hierCore(clk);
    endgenerate
 endmodule
 
@@ -42,6 +43,14 @@ module Core(input clk);
 
       Check check(.clk(clk), .crc(crc), .result(result), .rdata(rdata), .rdata2(rdata2));
 endmodule
+module CoreHier(input clk);
+   // Dummy logic to have two different hier blocks at the same level.
+   integer cyc = 0;
+   always @(posedge clk) begin
+      cyc += 1;
+      if (cyc == 1) $display("%d", clk);
+   end
+endmodule
 
 module Check(
       input clk,

test_regress/t/t_pgo_threads_hier.py

@@ -11,11 +11,12 @@ import vltest_bootstrap
 
 test.scenarios('vltmt')
 test.top_filename = "t/t_hier_block_perf.v"
-cycles = 100000
+cycles = 100
 test.sim_time = cycles * 10 + 1000
 
 threads = 2
-flags = ["--hierarchical", "-Wno-UNOPTFLAT", "-DSIM_CYCLES=" + str(cycles)]
+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)
 
@@ -24,15 +25,21 @@ test.execute(all_run_flags=[
     " +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 "VCheck"')
+test.file_grep(test.obj_dir + "/profile.vlt", r'profile_data -model "VCoreHier"')
 test.file_grep(test.obj_dir + "/profile.vlt", r'profile_data -model "V' + test.name + '"')
 
+# Check for cost rollovers
+test.file_grep_not(test.obj_dir + "/profile.vlt", r'.*cost 64\'d\d{18}.*')
+
 # 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,
+    v_flags2=[test.obj_dir + "/profile.vlt", "-Wno-PROFOUTOFDATE"] + flags,
     threads=threads)
 
 test.execute()

test_regress/t/t_pgo_threads_hier.vlt

@@ -8,3 +8,5 @@
 hier_workers -module "Test" -workers 2
 hier_block -module "Check"
 hier_workers -module "Check" -workers 2
+hier_block -module "CoreHier"
+hier_workers -module "CoreHier" -workers 2

test_regress/t/t_pgo_threads_hier_nested.py

@@ -1,45 +0,0 @@
-#!/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 + '"')
-
-# Check for cost rollovers
-test.file_grep_not(test.obj_dir + "/profile.vlt", r'.*cost 64\'d\d{18}.*')
-
-# 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()