Fix verilator_gantt for hierarchically Verilated models (#5700)

bin/verilator_gantt

@@ -36,7 +36,7 @@ def read_data(filename):
         re_proc_cpu = re.compile(r'VLPROFPROC processor\s*:\s*(\d+)\s*$')
         re_proc_dat = re.compile(r'VLPROFPROC ([a-z_ ]+)\s*:\s*(.*)$')
         cpu = None
-        thread = None
+        thread = 0
         execGraphStart = None
 
         global LongestVcdStrValueLength
@@ -54,11 +54,11 @@ def read_data(filename):
                 if kind == "SECTION_PUSH":
                     LongestVcdStrValueLength = max(LongestVcdStrValueLength, len(payload))
                     SectionStack.append(payload)
-                    Sections.append((tick, tuple(SectionStack)))
+                    Sections[thread].append((tick, tuple(SectionStack)))
                 elif kind == "SECTION_POP":
                     assert SectionStack, "SECTION_POP without SECTION_PUSH"
                     SectionStack.pop()
-                    Sections.append((tick, tuple(SectionStack)))
+                    Sections[thread].append((tick, tuple(SectionStack)))
                 elif kind == "MTASK_BEGIN":
                     mtask, predict_start, ecpu = re_payload_mtaskBegin.match(payload).groups()
                     mtask = int(mtask)
@@ -97,6 +97,7 @@ def read_data(filename):
                     print("-Unknown execution trace record: %s" % line)
             elif re_thread.match(line):
                 thread = int(re_thread.match(line).group(1))
+                Sections.append([])
             elif re.match(r'^VLPROF(THREAD|VERSION)', line):
                 pass
             elif re_arg1.match(line):
@@ -307,23 +308,27 @@ def report_cpus():
 
 
 def report_sections():
-    if not Sections:
+    for thread, section in enumerate(Sections):
-        return
+        if section:
-    print("\nSection profile:")
+            print(f"\nSection profile for thread {thread}:")
+            report_section(section)
 
+
+def report_section(section):
     totalTime = collections.defaultdict(lambda: 0)
     selfTime = collections.defaultdict(lambda: 0)
 
     sectionTree = [0, {}, 1]  # [selfTime, childTrees, numberOfTimesEntered]
     prevTime = 0
     prevStack = ()
-    for time, stack in Sections:
+    for time, stack in section:
         if len(stack) > len(prevStack):
             scope = sectionTree
             for item in stack:
                 scope = scope[1].setdefault(item, [0, {}, 0])
             scope[2] += 1
         dt = time - prevTime
+        assert dt >= 0
         scope = sectionTree
         for item in prevStack:
             scope = scope[1].setdefault(item, [0, {}, 0])
@@ -457,10 +462,11 @@ def write_vcd(filename):
                 addValue(pcode, time, value)
 
         # Section graph
-        if Sections:
+        for thread, section in enumerate(Sections):
-            scode = getCode(LongestVcdStrValueLength * 8, "section", "trace")
+            if section:
-            dcode = getCode(32, "section", "depth")
+                scode = getCode(LongestVcdStrValueLength * 8, "section", f"t{thread}_trace")
-            for time, stack in Sections:
+                dcode = getCode(32, "section", f"t{thread}_depth")
+                for time, stack in section:
                     addValue(scode, time, stack[-1] if stack else None)
                     addValue(dcode, time, len(stack))
 

src/V3EmitCModel.cpp

@@ -432,7 +432,8 @@ class EmitCModel final : public EmitCFunc {
         puts(topModNameProtected + "__" + protect("_eval_settle") + "(&(vlSymsp->TOP));\n");
         puts("}\n");
 
-        if (v3Global.opt.profExec()) puts("vlSymsp->__Vm_executionProfilerp->configure();\n");
+        if (v3Global.opt.profExec() && !v3Global.opt.hierChild())
+            puts("vlSymsp->__Vm_executionProfilerp->configure();\n");
 
         puts("VL_DEBUG_IF(VL_DBG_MSGF(\"+ Eval\\n\"););\n");
         puts(topModNameProtected + "__" + protect("_eval") + "(&(vlSymsp->TOP));\n");

src/V3Order.cpp

@@ -123,8 +123,10 @@ AstCFunc* V3Order::order(AstNetlist* netlistp,  //
     }();
 
     if (v3Global.opt.profExec()) {
-        funcp->addStmtsp(new AstCStmt{flp, "VL_EXEC_TRACE_ADD_RECORD(vlSymsp).sectionPush(\"func "
+        const string name
-                                               + tag + "\");\n"});
+            = (v3Global.opt.hierChild() ? (v3Global.opt.topModule() + " ") : "") + "func " + tag;
+        funcp->addStmtsp(new AstCStmt{flp, "VL_EXEC_TRACE_ADD_RECORD(vlSymsp).sectionPush(\""
+                                               + name + "\");\n"});
     }
 
     // Build the OrderGraph

src/V3Sched.cpp

@@ -156,7 +156,9 @@ AstNodeStmt* checkIterationLimit(AstNetlist* netlistp, const string& name, AstVa
     return ifp;
 }
 
-AstNodeStmt* profExecSectionPush(FileLine* flp, const string& name) {
+AstNodeStmt* profExecSectionPush(FileLine* flp, const string& section) {
+    const string name
+        = (v3Global.opt.hierChild() ? (v3Global.opt.topModule() + " ") : "") + section;
     return new AstCStmt{flp, "VL_EXEC_TRACE_ADD_RECORD(vlSymsp).sectionPush(\"" + name + "\");\n"};
 }
 

test_regress/t/t_gantt_hier.py

@@ -0,0 +1,52 @@
+#!/usr/bin/env python3
+# DESCRIPTION: Verilator: Verilog Test driver/expect definition
+#
+# Copyright 2024 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
+
+# Test for bin/verilator_gantt,
+
+import vltest_bootstrap
+
+test.scenarios('vlt_all')
+test.top_filename = "t/t_gen_alw.v"  # Any, as long as runs a few cycles
+
+test.compile(
+    v_flags2=["--prof-exec", "--hierarchical"],
+    # Checks below care about thread count, so use 2 (minimum reasonable)
+    threads=(2 if test.vltmt else 1))
+
+test.execute(all_run_flags=[
+    "+verilator+prof+exec+start+2",
+    " +verilator+prof+exec+window+2",
+    " +verilator+prof+exec+file+" + test.obj_dir + "/profile_exec.dat",
+    " +verilator+prof+vlt+file+" + test.obj_dir + "/profile.vlt"])  # yapf:disable
+
+# For now, verilator_gantt still reads from STDIN
+#  (probably it should take a file, gantt.dat like verilator_profcfunc)
+# The profiling data still goes direct to the runtime's STDOUT
+#  (maybe that should go to a separate file - gantt.dat?)
+test.run(cmd=[
+    os.environ["VERILATOR_ROOT"] + "/bin/verilator_gantt", test.obj_dir +
+    "/profile_exec.dat", "--vcd " + test.obj_dir + "/profile_exec.vcd", "| tee " + test.obj_dir +
+    "/gantt.log"
+])
+
+if test.vltmt:
+    test.file_grep(test.obj_dir + "/gantt.log", r'Total threads += 2')
+    test.file_grep(test.obj_dir + "/gantt.log", r'Total mtasks += 8')
+    # Predicted thread utilization should be less than 100%
+    test.file_grep_not(test.obj_dir + "/gantt.log", r'Thread utilization =\s*\d\d\d+\.\d+%')
+else:
+    test.file_grep(test.obj_dir + "/gantt.log", r'Total threads += 1')
+    test.file_grep(test.obj_dir + "/gantt.log", r'Total mtasks += 0')
+
+test.file_grep(test.obj_dir + "/gantt.log", r'\|\s+2\s+\|\s+2\.0+\s+\|\s+eval')
+
+# Diff to itself, just to check parsing
+test.vcd_identical(test.obj_dir + "/profile_exec.vcd", test.obj_dir + "/profile_exec.vcd")
+
+test.passes()

test_regress/t/t_gen_alw.v

@@ -59,7 +59,7 @@ endmodule
 module Test (/*AUTOARG*/
    // Inputs
    clk, in
-   );
+   ); /*verilator hier_block*/
    input clk;
    input [9:0] in;