[rocprofiler-sdk] users/mcao/fix_rocprofv_counter_yaml

projects/rocprofiler-sdk/source/lib/output/metadata.cpp

@@ -119,8 +119,20 @@ process_agent_counters(rocprofiler_agent_id_t    agent_id,
                 auto _dim_ids  = std::vector<rocprofiler_counter_dimension_id_t>{};
                 auto _dim_info = std::vector<rocprofiler_counter_record_dimension_info_t>{};
 
-                ROCPROFILER_CHECK(rocprofiler_query_counter_info(
-                    counters[i], ROCPROFILER_COUNTER_INFO_VERSION_1, &_info));
+                auto status = rocprofiler_query_counter_info(
+                    counters[i], ROCPROFILER_COUNTER_INFO_VERSION_1, &_info);
+
+                if(status != ROCPROFILER_STATUS_SUCCESS)
+                {
+                    ROCP_WARNING << fmt::format(
+                        "[{}] skipping counter metadata for handle {} because "
+                        "rocprofiler_query_counter_info(...) failed: {} :: {}",
+                        __FUNCTION__,
+                        counters[i].handle,
+                        rocprofiler_get_status_name(status),
+                        rocprofiler_get_status_string(status));
+                    continue;
+                }
 
                 if(counters_set_data != nullptr && counters_set_data->count(_info.name) == 0)
                     continue;

projects/rocprofiler-sdk/source/lib/rocprofiler-sdk/aql/helpers.cpp

@@ -44,7 +44,6 @@ get_query_info(rocprofiler_agent_id_t agent, const counters::Metric& metric)
     hsa_ven_amd_aqlprofile_id_query_t query = {metric.block().c_str(), 0, 0};
     if(aqlprofile_get_pmc_info(&profile, AQLPROFILE_INFO_BLOCK_ID, &query) != HSA_STATUS_SUCCESS)
     {
-        ROCP_DFATAL << fmt::format("AQL failed to query info for counter {}", metric);
         throw std::runtime_error(fmt::format("AQL failed to query info for counter {}", metric));
     }
     return query;

projects/rocprofiler-sdk/source/lib/rocprofiler-sdk/counters/dimensions.cpp

@@ -65,7 +65,15 @@ getBlockDimensions(rocprofiler_agent_id_t agent_id, const Metric& metric)
     {
         auto dims   = std::map<int, uint64_t>{};
         auto status = aql::get_dim_info(agent_id, event, 0, dims);
-        CHECK_EQ(status, ROCPROFILER_STATUS_SUCCESS) << rocprofiler_get_status_string(status);
+        // Event-level error handling: Skip individual events with invalid dimension info.
+        // This allows the metric to succeed partially if some events are valid.
+        if(status != ROCPROFILER_STATUS_SUCCESS)
+        {
+            ROCP_WARNING << "Failed to get dimension info for event in metric '" << metric.name()
+                         << "': " << rocprofiler_get_status_string(status)
+                         << ". Event will be skipped.";
+            continue;
+        }
 
         for(const auto& [id, extent] : dims)
         {
@@ -113,10 +121,13 @@ generate_dimensions(rocprofiler_agent_id_t agent_id)
         {
             // Generate dimensions for this specific agent
             dims.emplace(ast.out_id().handle, ast_copy.set_dimensions(agent_id));
-        } catch(std::runtime_error& e)
+        }
+        // Metric-level error handling: Skip entire metric on validation failures
+        catch(const std::runtime_error& e)
         {
-            ROCP_FATAL << metric << " has improper dimensions"
-                       << " " << e.what();
+            ROCP_WARNING << "Invalid counter '" << metric << "' in YAML configuration - "
+                         << e.what() << ". Counter will be skipped.";
+            continue;
         }
     }
     return {.id_to_dim = dims};