Fix uninitialized thread PGO counters (#6167)

include/verilated_profiler.h

@@ -218,7 +218,7 @@ class VlPgoProfiler final {
     };
 
     // Counters are stored packed, all together to reduce cache effects
-    std::array<uint64_t, N_Entries> m_counters;  // Time spent on this record
+    std::array<uint64_t, N_Entries> m_counters{};  // Time spent on this record
     std::vector<Record> m_records;  // Record information
 
 public:

test_regress/t/t_pgo_threads_hier_nested.py

@@ -28,6 +28,9 @@ test.execute(all_run_flags=[
 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(

test_regress/t/t_pgo_threads_hier_nested.vlt

@@ -6,3 +6,5 @@
 
 `verilator_config
 hier_workers -module "Test" -workers 2
+hier_block -module "Check"
+hier_workers -module "Check" -workers 2