From 58b94ee42c9381a68589af6cb87256dc6510998c Mon Sep 17 00:00:00 2001
From: Chiming <chimingz@amd.com>
Date: Mon, 17 Aug 2026 16:08:29 -0700
Subject: [PATCH 1/3] add new function in xrt-smi to query the load on AIE
 array.

Signed-off-by: Chiming <chimingz@amd.com>
---
 src/runtime_src/core/common/info_platform.cpp | 21 +++++
 src/runtime_src/core/common/query_requests.h  | 27 +++++-
 src/runtime_src/core/common/smi/smi_ryzen.cpp | 44 ++--------
 .../tools/common/reports/ReportAieLoad.cpp    | 82 +++++++++++++++++++
 .../core/tools/common/reports/ReportAieLoad.h | 28 +++++++
 .../reports/platform/ReportRyzenPlatform.cpp  |  6 ++
 .../core/tools/xbutil2/SubCmdExamine.cpp      | 65 +++++++++------
 .../core/tools/xbutil2/SubCmdExamine.h        |  5 +-
 8 files changed, 212 insertions(+), 66 deletions(-)
 create mode 100644 src/runtime_src/core/tools/common/reports/ReportAieLoad.cpp
 create mode 100644 src/runtime_src/core/tools/common/reports/ReportAieLoad.h

diff --git a/src/runtime_src/core/common/info_platform.cpp b/src/runtime_src/core/common/info_platform.cpp
index 0c426249923..76541143405 100644
--- a/src/runtime_src/core/common/info_platform.cpp
+++ b/src/runtime_src/core/common/info_platform.cpp
@@ -9,6 +9,7 @@
 #include "xrt/detail/xclbin.h"
 
 #include <boost/algorithm/string.hpp>
+#include <limits>
 
 // Too much typing
 using ptree_type = boost::property_tree::ptree;
@@ -194,6 +195,25 @@ add_host_mem_info(const xrt_core::device* device, ptree_type& pt)
   pt.add("host_memory_status", get_host_mem_status(device));
 }
 
+void
+add_aie_load_info(const xrt_core::device* device, ptree_type& pt)
+{
+  try {
+    const xq::aie_load::args args{}; // sample_duration_ms = 0 -> driver default
+    const auto data = xrt_core::device_query<xq::aie_load>(device, args);
+
+    ptree_type pt_aie_load;
+    if (data.load_percent == std::numeric_limits<uint32_t>::max())
+      pt_aie_load.put("load_percent", "N/A");
+    else
+      pt_aie_load.put("load_percent", data.load_percent);
+    pt.put_child("aie_load", pt_aie_load);
+  }
+  catch (const xq::exception&) {
+    // Ignoring if not available: older driver/DLL or unsupported device
+  }
+}
+
 void
 add_status_info(const xrt_core::device* device, ptree_type& pt)
 {
@@ -431,6 +451,7 @@ add_platform_info(const xrt_core::device* device, ptree_type& pt_platform_array)
   {
     add_electrical_info(device, pt_platform);
     add_thermal_info(device, pt_platform);
+    add_aie_load_info(device, pt_platform);
     break;
   }
   default:
diff --git a/src/runtime_src/core/common/query_requests.h b/src/runtime_src/core/common/query_requests.h
index 2f71d6e9e6e..5cd20bfa9aa 100644
--- a/src/runtime_src/core/common/query_requests.h
+++ b/src/runtime_src/core/common/query_requests.h
@@ -350,7 +350,8 @@ enum class key_type
 
   aie_read,
   aie_write,
-  aie_coredump
+  aie_coredump,
+  aie_load
 };
 
 struct pcie_vendor : request
@@ -4316,6 +4317,30 @@ struct aie_coredump : request
   std::any
   get(const device*, const std::any&) const override = 0;
 };
+
+// AIE array hardware utilization, computed from two consecutive firmware
+// activity-counter snapshots separated by sample_duration_ms milliseconds.
+struct aie_load : request
+{
+  struct result_type {
+    uint32_t load_percent;          // 0-100, or UINT32_MAX if unavailable
+    uint64_t timestamp_ms;          // FW uptime when second snapshot was taken
+    uint64_t activity_counters[8];  // raw counter values from second snapshot
+    uint64_t operations_per_second; // total counter change across 8 counters per second
+  };
+
+  struct args {
+    uint32_t sample_duration_ms = 0; // 0 = driver default (XRT_AIE_LOAD_SAMPLE_INTERVAL_MS)
+  };
+
+  static const key_type key = key_type::aie_load;
+
+  virtual std::any
+  get(const device*) const override = 0;
+
+  virtual std::any
+  get(const device*, const std::any&) const override = 0;
+};
 } // query
 
 } // xrt_core
diff --git a/src/runtime_src/core/common/smi/smi_ryzen.cpp b/src/runtime_src/core/common/smi/smi_ryzen.cpp
index b8b61e1c3e0..20174656339 100644
--- a/src/runtime_src/core/common/smi/smi_ryzen.cpp
+++ b/src/runtime_src/core/common/smi/smi_ryzen.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: Apache-2.0
-// Copyright (C) 2025-2026 Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (C) 2025 Advanced Micro Devices, Inc. All rights reserved.
 
 #define XRT_CORE_COMMON_SOURCE
 
@@ -15,6 +15,7 @@ config_gen_ryzen::
 config_gen_ryzen()
 {
   examine_report_desc = {
+    {"aie-load", "AIE array load utilization", "common"},
     {"aie-partitions", "AIE partition information", "common"},
     {"all", "All known reports are produced", "common"},
     {"host", "Host information (default)", "common"},
@@ -63,7 +64,7 @@ config_gen_ryzen::create_validate_subcommand()
   validate_suboptions.emplace("device", std::make_shared<option>("device", "d", "The Bus:Device.Function (e.g., 0000:d8:00.0) device of interest", "common", "", "string"));
   validate_suboptions.emplace("format", std::make_shared<option>("format", "f", "Report output format. Valid values are:\n"
                                 "\tJSON        - Latest JSON schema\n"
-                                "\tJSON-2020.2 - JSON 2020.2 schema (legacy)", "common", "JSON", "string"));
+                                "\tJSON-2020.2 - JSON 2020.2 schema", "common", "JSON", "string"));
   validate_suboptions.emplace("output", std::make_shared<option>("output", "o", "Direct the output to the given file", "common", "", "string"));
   validate_suboptions.emplace("help", std::make_shared<option>("help", "h", "Help to use this sub-command", "common", "", "none"));
   validate_suboptions.emplace("run", std::make_shared<listable_description_option>("run", "r", "Run a subset of the test suite. Valid options are:\n",
@@ -82,7 +83,7 @@ config_gen_ryzen::create_examine_subcommand()
   examine_suboptions.emplace("device", std::make_shared<option>("device", "d", "The Bus:Device.Function (e.g., 0000:d8:00.0) device of interest", "common", "", "string"));
   examine_suboptions.emplace("format", std::make_shared<option>("format", "f", "Report output format. Valid values are:\n"
                                 "\tJSON        - Latest JSON schema\n"
-                                "\tJSON-2020.2 - JSON 2020.2 schema (legacy)", "common", "JSON", "string"));
+                                "\tJSON-2020.2 - JSON 2020.2 schema", "common", "JSON", "string"));
   examine_suboptions.emplace("output", std::make_shared<option>("output", "o", "Direct the output to the given file", "common", "", "string"));
   examine_suboptions.emplace("help", std::make_shared<option>("help", "h", "Help to use this sub-command", "common", "", "none"));
   examine_suboptions.emplace("watch", std::make_shared<option>("watch", "", "Refresh interval in seconds between examine updates. Exit with Ctrl+C.", "hidden", "0", "string"));
@@ -113,6 +114,7 @@ config_gen_npu3::
 config_gen_npu3()
 {
   examine_report_desc = {
+    {"aie-load", "AIE array load utilization", "common"},
     {"aie-partitions", "AIE partition information", "common"},
     {"all", "All known reports are produced", "common"},
     {"host", "Host information", "common"},
@@ -139,42 +141,6 @@ config_gen_npu3()
   };
 }
 
-subcommand
-config_gen_npu3::create_validate_subcommand()
-{
-  std::map<std::string, std::shared_ptr<option>> validate_suboptions;
-  validate_suboptions.emplace("device", std::make_shared<option>("device", "d", "The Bus:Device.Function (e.g., 0000:d8:00.0) device of interest", "common", "", "string"));
-  validate_suboptions.emplace("json", std::make_shared<option>("json", "", "JSON ABI version for file output. Valid values are:\n"
-                                "\tdefault - Latest JSON schema (default)", "common", "default", "string"));
-  validate_suboptions.emplace("output", std::make_shared<option>("output", "o", "Direct the output to the given file", "common", "", "string"));
-  validate_suboptions.emplace("help", std::make_shared<option>("help", "h", "Help to use this sub-command", "common", "", "none"));
-  validate_suboptions.emplace("run", std::make_shared<listable_description_option>("run", "r", "Run a subset of the test suite. Valid options are:\n",
-                              "common", "",  "array", get_validate_test_desc()));
-  validate_suboptions.emplace("param", std::make_shared<option>("param", "", "Extended parameter for a given test. Format: <test-name>:<key>:<value>", "param", "", "string"));
-  validate_suboptions.emplace("pmode", std::make_shared<option>("pmode", "", "Specify which power mode to run the benchmarks in. Note: Some tests might be unavailable for some modes", "hidden", "", "string"));
-  validate_suboptions.emplace("loop", std::make_shared<option>("loop", "", "Number of iterations to run the test", "hidden", "", "string"));
-
-  return {"validate", "Validates the given device by executing the platform's validate executable", "common", std::move(validate_suboptions)};
-}
-
-subcommand
-config_gen_npu3::create_examine_subcommand()
-{
-  std::map<std::string, std::shared_ptr<option>> examine_suboptions;
-  examine_suboptions.emplace("device", std::make_shared<option>("device", "d", "The Bus:Device.Function (e.g., 0000:d8:00.0) device of interest", "common", "", "string"));
-  examine_suboptions.emplace("json", std::make_shared<option>("json", "", "JSON ABI version for file output. Valid values are:\n"
-                                "\tdefault - Latest JSON schema (default)", "common", "default", "string"));
-  examine_suboptions.emplace("output", std::make_shared<option>("output", "o", "Direct the output to the given file", "common", "", "string"));
-  examine_suboptions.emplace("help", std::make_shared<option>("help", "h", "Help to use this sub-command", "common", "", "none"));
-  examine_suboptions.emplace("watch", std::make_shared<option>("watch", "", "Refresh interval in seconds between examine updates. Exit with Ctrl+C.", "hidden", "0", "string"));
-  examine_suboptions.emplace("report", std::make_shared<listable_description_option>("report", "r", "The type of report to be produced. Reports currently available are:\n", "common", "", "array", get_examine_report_desc()));
-  examine_suboptions.emplace("firmware-log", std::make_shared<option>("firmware-log", "", "Show status|watch firmware log data", "hidden", "", "string", true));
-  examine_suboptions.emplace("event-trace", std::make_shared<option>("event-trace", "", "Show status|watch event trace data", "hidden", "", "string", true));
-  examine_suboptions.emplace("context-health", std::make_shared<option>("context-health", "", "Show status|watch context health data", "hidden", "", "string", true));
-
-  return {"examine", "This command will 'examine' the state of the system/device and will generate a report of interest in a text or JSON format.", "common", std::move(examine_suboptions)};
-}
-
 static std::shared_ptr<config_gen_ryzen>
 create_config_generator(smi_hardware_config::hardware_family family)
 {
diff --git a/src/runtime_src/core/tools/common/reports/ReportAieLoad.cpp b/src/runtime_src/core/tools/common/reports/ReportAieLoad.cpp
new file mode 100644
index 00000000000..c1640c1fc88
--- /dev/null
+++ b/src/runtime_src/core/tools/common/reports/ReportAieLoad.cpp
@@ -0,0 +1,82 @@
+// SPDX-License-Identifier: Apache-2.0
+// Copyright (C) 2025 Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ReportAieLoad.h"
+
+#include "core/common/query_requests.h"
+
+#include <boost/format.hpp>
+#include <limits>
+
+void
+ReportAieLoad::getPropertyTreeInternal(const xrt_core::device* dev,
+                                       boost::property_tree::ptree& pt) const
+{
+  getPropertyTree20202(dev, pt);
+}
+
+void
+ReportAieLoad::getPropertyTree20202(const xrt_core::device* dev,
+                                    boost::property_tree::ptree& pt) const
+{
+  boost::property_tree::ptree node;
+
+  try {
+    const xrt_core::query::aie_load::args args{m_duration_ms};
+    const auto data = xrt_core::device_query<xrt_core::query::aie_load>(dev, args);
+
+    const bool unavailable = (data.load_percent == std::numeric_limits<uint32_t>::max());
+    node.put("load_percent",          unavailable ? "N/A" : std::to_string(data.load_percent));
+    node.put("timestamp_ms",          data.timestamp_ms);
+    node.put("operations_per_second", data.operations_per_second);
+
+    boost::property_tree::ptree pt_counters;
+    for (size_t i = 0; i < 8; i++) {
+      boost::property_tree::ptree pt_counter;
+      pt_counter.put("", data.activity_counters[i]);
+      pt_counters.push_back({"", pt_counter});
+    }
+    node.add_child("activity_counters", pt_counters);
+  }
+  catch (const xrt_core::query::exception& e) {
+    node.put("error", e.what());
+  }
+
+  pt.add_child("aie_load", node);
+}
+
+void
+ReportAieLoad::writeReport(const xrt_core::device* /*dev*/,
+                           const boost::property_tree::ptree& pt,
+                           const std::vector<std::string>& /*elementsFilter*/,
+                           std::ostream& output) const
+{
+  const auto& node = pt.get_child("aie_load");
+
+  // If the query failed, report the error and return.
+  auto error = node.get_optional<std::string>("error");
+  if (error) {
+    output << "AIE Load: " << *error << "\n\n";
+    return;
+  }
+
+  output << "AIE Load\n";
+
+  const auto load = node.get<std::string>("load_percent");
+  if (load == "N/A")
+    output << boost::format("  %-25s: N/A (AIE off, gated, or counters unavailable)\n") % "Utilization";
+  else
+    output << boost::format("  %-25s: %s%%\n") % "Utilization" % load;
+
+  output << boost::format("  %-25s: %s ops/s\n") % "Operations/Second"
+            % node.get<std::string>("operations_per_second");
+  output << boost::format("  %-25s: %s ms\n") % "FW Timestamp"
+            % node.get<std::string>("timestamp_ms");
+
+  output << "  Activity Counters:\n";
+  int idx = 0;
+  for (const auto& [name, counter] : node.get_child("activity_counters"))
+    output << boost::format("    [%d]: %s\n") % idx++ % counter.get_value<std::string>();
+
+  output << "\n";
+}
diff --git a/src/runtime_src/core/tools/common/reports/ReportAieLoad.h b/src/runtime_src/core/tools/common/reports/ReportAieLoad.h
new file mode 100644
index 00000000000..ab2f1a343d8
--- /dev/null
+++ b/src/runtime_src/core/tools/common/reports/ReportAieLoad.h
@@ -0,0 +1,28 @@
+// SPDX-License-Identifier: Apache-2.0
+// Copyright (C) 2025 Advanced Micro Devices, Inc. All rights reserved.
+
+#ifndef __ReportAieLoad_h_
+#define __ReportAieLoad_h_
+
+#include "tools/common/Report.h"
+
+#include <cstdint>
+
+class ReportAieLoad : public Report {
+ public:
+  ReportAieLoad() : Report("aie-load", "AIE array load utilization", true /*deviceRequired*/) {}
+
+  // Called by SubCmdExamine before produce_reports to forward --duration.
+  void setDuration(uint32_t duration_ms) { m_duration_ms = duration_ms; }
+
+ public:
+  virtual void getPropertyTreeInternal(const xrt_core::device* dev, boost::property_tree::ptree& pt) const;
+  virtual void getPropertyTree20202(const xrt_core::device* dev, boost::property_tree::ptree& pt) const;
+  virtual void writeReport(const xrt_core::device* dev, const boost::property_tree::ptree& pt,
+                           const std::vector<std::string>& elementsFilter, std::ostream& output) const;
+
+ private:
+  uint32_t m_duration_ms = 0; // 0 = driver default (XRT_AIE_LOAD_SAMPLE_INTERVAL_MS)
+};
+
+#endif
diff --git a/src/runtime_src/core/tools/common/reports/platform/ReportRyzenPlatform.cpp b/src/runtime_src/core/tools/common/reports/platform/ReportRyzenPlatform.cpp
index 7bc0a6912a5..8ce6e84cdb8 100644
--- a/src/runtime_src/core/tools/common/reports/platform/ReportRyzenPlatform.cpp
+++ b/src/runtime_src/core/tools/common/reports/platform/ReportRyzenPlatform.cpp
@@ -62,6 +62,12 @@ ReportRyzenPlatform::writeReport(const xrt_core::device* /*_pDevice*/,
 
     auto temp_c = pt_platform.get<std::string>("thermal.temp_C", "N/A");
     _output << boost::format("%-23s  : %s\n") % "Temperature (C)" % temp_c;
+
+    auto aie_load = pt_platform.get<std::string>("aie_load.load_percent", "N/A");
+    if (aie_load != "N/A")
+      _output << boost::format("%-23s  : %s%%\n") % "AIE Load" % aie_load;
+    else
+      _output << boost::format("%-23s  : %s\n") % "AIE Load" % aie_load;
   }
 
   _output << std::endl;
diff --git a/src/runtime_src/core/tools/xbutil2/SubCmdExamine.cpp b/src/runtime_src/core/tools/xbutil2/SubCmdExamine.cpp
index 4dacf699710..d3891cffd43 100644
--- a/src/runtime_src/core/tools/xbutil2/SubCmdExamine.cpp
+++ b/src/runtime_src/core/tools/xbutil2/SubCmdExamine.cpp
@@ -44,7 +44,9 @@
 #include "tools/common/reports/platform/ReportPlatforms.h"
 #include "tools/common/reports/ReportPsKernels.h"
 #include "tools/common/reports/ReportQspiStatus.h"
+#include "tools/common/reports/ReportAieLoad.h"
 #include "tools/common/reports/ReportTelemetry.h"
+#include <cstdint>
 #include "tools/common/reports/ReportThermal.h"
 
 #include <filesystem>
@@ -66,6 +68,7 @@ SubCmdExamine::SubCmdExamine(bool _isHidden, bool _isDepricated, bool _isPrelimi
   uniqueReportCollection = {
   // Common reports
     std::make_shared<ReportAie>(),
+    std::make_shared<ReportAieLoad>(),
     std::make_shared<ReportAieShim>(),
     std::make_shared<ReportAieMem>(),
     std::make_shared<ReportAiePartitions>(),
@@ -104,17 +107,22 @@ SubCmdExamine::SubCmdExamine(bool _isHidden, bool _isDepricated, bool _isPrelimi
     option->setExecutable(getExecutableName());
     option->setCommand(getName());
   }
+
+  // Register --duration directly so it is always recognized without --advanced.
+  m_commonOptions.add_options()
+    ("duration", po::value<uint32_t>()->default_value(0),
+     "Sampling window in milliseconds for the aie-load report (0 = driver default)");
 }
 
 void SubCmdExamine::fill_option_values(const po::variables_map& vm, SubCmdExamineOptions& options) const
 {
   options.m_device = vm.count("device") ? vm["device"].as<std::string>() : "";
-  options.m_format = vm.find("format") != vm.end() ? vm["format"].as<std::string>() : "JSON";
-  options.m_json = vm.find("json") != vm.end() ? vm["json"].as<std::string>() : "default";
+  options.m_format = vm.count("format") ? vm["format"].as<std::string>() : "JSON";
   options.m_output = vm.count("output") ? vm["output"].as<std::string>() : "";
   options.m_reportNames = vm.count("report") ? vm["report"].as<std::vector<std::string>>() : std::vector<std::string>();
   options.m_help = vm.count("help") ? vm["help"].as<bool>() : false;
   options.m_elementsFilter = vm.count("element") ? vm["element"].as<std::vector<std::string>>() : std::vector<std::string>();
+  options.m_duration = vm["duration"].as<uint32_t>();
   options.m_watchIntervalSec.reset();
   if (vm.count("watch")) {
     const auto& s = vm["watch"].as<std::string>();
@@ -129,9 +137,7 @@ SubCmdExamine::setOptionConfig(const boost::property_tree::ptree &config)
   try{
     m_jsonConfig.addProgramOptions(m_commonOptions, "common", getName());
     m_jsonConfig.addProgramOptions(m_hiddenOptions, "hidden", getName());
-    m_json_abi_platform = m_jsonConfig.has_option(getName(), "json")
-                     && !m_jsonConfig.has_option(getName(), "format");
-  }
+  } 
   catch (const std::exception& e) {
     std::cerr << "Error: " << e.what() << std::endl;
   }
@@ -218,11 +224,7 @@ SubCmdExamine::execute(const SubCmdOptions& _options) const
     return;
   }
 
-  Report::SchemaVersion schema_version = Report::SchemaVersion::unknown;
-  const bool json_platform = vm.count("json") || m_json_abi_platform;
-  const std::string schema_label = json_platform
-    ? (options.m_json.empty() ? "default" : options.m_json)
-    : options.m_format;
+  Report::SchemaVersion schemaVersion = Report::getSchemaDescription(options.m_format).schemaVersion;
   try{
     if (vm.count("output") && options.m_output.empty())
       throw xrt_core::error("Output file not specified");
@@ -233,20 +235,20 @@ SubCmdExamine::execute(const SubCmdOptions& _options) const
     if (vm.count("element") && options.m_elementsFilter.empty())
       throw xrt_core::error("No element filter given to be produced");
 
-    schema_version = Report::resolve_json_abi(json_platform, vm.count("json") || vm.count("format"), schema_label);
-    if (schema_version == Report::SchemaVersion::unknown)
-      throw xrt_core::error((boost::format("Unknown JSON ABI version: '%s'") % schema_label).str());
+    if (schemaVersion == Report::SchemaVersion::unknown) 
+      throw xrt_core::error((boost::format("Unknown output format: '%s'") % options.m_format).str());
 
-    // Either JSON output selector requires an accompanying output file
-    if ((vm.count("format") || vm.count("json")) && options.m_output.empty())
+    // DRC check
+    // When json is specified, make sure an accompanying output file is also specified
+    if (vm.count("format") && options.m_output.empty())
       throw xrt_core::error("Please specify an output file to redirect the json to");
 
     if (!options.m_output.empty() && std::filesystem::exists(options.m_output) && !XBU::getForce())
       throw xrt_core::error((boost::format("The output file '%s' already exists. Please either remove it or execute this command again with the '--force' option to overwrite it") % options.m_output).str());
 
     if (options.m_watchIntervalSec) {
-      if (vm.count("format") || vm.count("json"))
-        throw xrt_core::error("Watch mode cannot be additionally formatted; output is text only.");
+      if (vm.count("format"))
+        throw xrt_core::error("Watch mode cannot be used with --format; output is text only.");
       if (!options.m_output.empty())
         throw xrt_core::error("Watch mode cannot be used with --output.");
     }
@@ -258,9 +260,12 @@ SubCmdExamine::execute(const SubCmdOptions& _options) const
     throw xrt_core::error(std::errc::operation_canceled);
   }
 
+  // Determine report level
   std::vector<std::string> reportsToRun(options.m_reportNames);
-  if (reportsToRun.empty())
+  if (reportsToRun.empty()) {
+    // Default report with or without --advanced (advanced only unlocks hidden options/reports).
     reportsToRun.emplace_back("host");
+  }
 
   if ((std::find(reportsToRun.begin(), reportsToRun.end(), "all") != reportsToRun.end()) && (reportsToRun.size() > 1)) {
     std::cerr << "ERROR: The 'all' value for the reports to run cannot be used with any other named reports.\n";
@@ -287,13 +292,21 @@ SubCmdExamine::execute(const SubCmdOptions& _options) const
   if (device){
     const xrt_core::smi::tuple_vector& reportList = xrt_core::device_query<xrt_core::query::xrt_smi_lists>(device, xrt_core::query::xrt_smi_lists::type::examine_reports);
     runnableReports = getReportsList(reportList);
-  } 
+  }
   else {
     runnableReports = uniqueReportCollection;
   }
+
+  // When the user explicitly names reports with -r, validate against the full set compiled
+  // into this binary rather than the device-queried list. The device list may be from an
+  // older installed DLL that pre-dates reports added to this build of xrt-smi.
+  const ReportCollection& validationSource = reportsToRun.empty()
+      ? runnableReports
+      : uniqueReportCollection;
+
   // Collect the reports to be processed
   try {
-    XBU::collect_and_validate_reports(runnableReports, reportsToRun, reportsToProcess);
+    XBU::collect_and_validate_reports(validationSource, reportsToRun, reportsToProcess);
   } catch (const xrt_core::error& e) {
     std::cerr << boost::format("ERROR: %s\n") % e.what();
     printHelp();
@@ -330,6 +343,12 @@ SubCmdExamine::execute(const SubCmdOptions& _options) const
     }
   }
 
+  // Propagate --duration to ReportAieLoad before running reports.
+  for (const auto& report : reportsToProcess) {
+    if (auto* aieLoad = dynamic_cast<ReportAieLoad*>(report.get()))
+      aieLoad->setDuration(options.m_duration);
+  }
+
   // Create the report
   std::ostringstream oSchemaOutput;
   try {
@@ -338,13 +357,13 @@ SubCmdExamine::execute(const SubCmdOptions& _options) const
       const auto examine_watch_snapshot =
           [&](const xrt_core::device*) {
             std::ostringstream console;
-            XBU::produce_reports(device, reportsToProcess, schema_version, schema_label, {}, console, oSchemaOutput);
+            XBU::produce_reports(device, reportsToProcess, schemaVersion, {}, console, oSchemaOutput);
             return console.str();
           };
       smi_watch_mode::run_watch_mode(device.get(), std::cout, examine_watch_snapshot,
           *options.m_watchIntervalSec);
     } else {
-      XBU::produce_reports(device, reportsToProcess, schema_version, schema_label, {}, std::cout, oSchemaOutput);
+      XBU::produce_reports(device, reportsToProcess, schemaVersion, {}, std::cout, oSchemaOutput);
     }
   } catch (const std::exception&) {
     // Exception is thrown at the end of this function to allow for report writing
@@ -362,7 +381,7 @@ SubCmdExamine::execute(const SubCmdOptions& _options) const
 
     fOutput << oSchemaOutput.str();
 
-    std::cout << boost::format("Successfully wrote the JSON file: %s") % options.m_output << std::endl;
+    std::cout << boost::format("Successfully wrote the %s file: %s") % options.m_format % options.m_output << std::endl;
   }
 
   if (!is_report_output_valid)
diff --git a/src/runtime_src/core/tools/xbutil2/SubCmdExamine.h b/src/runtime_src/core/tools/xbutil2/SubCmdExamine.h
index 8cfd2af6b83..7eb56f10a94 100644
--- a/src/runtime_src/core/tools/xbutil2/SubCmdExamine.h
+++ b/src/runtime_src/core/tools/xbutil2/SubCmdExamine.h
@@ -1,6 +1,6 @@
 // SPDX-License-Identifier: Apache-2.0
 // Copyright (C) 2020 Xilinx, Inc
-// Copyright (C) 2023-2026 Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (C) 2023 Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef __SubCmdExamine_h_
 #define __SubCmdExamine_h_
@@ -25,15 +25,14 @@ struct SubCmdExamineOptions {
   std::vector<std::string>  m_reportNames;
   std::vector<std::string>  m_elementsFilter;
   std::string               m_format;
-  std::string               m_json;
   std::string               m_output;
   bool                      m_help;
   std::optional<unsigned>   m_watchIntervalSec;
+  uint32_t                  m_duration = 0; // sampling window in ms for aie-load (0 = driver default)
 };
 class SubCmdExamine : public SubCmd {
   ReportCollection uniqueReportCollection;
   std::vector<std::shared_ptr<OptionOptions>> m_optionOptionsCollection;
-  bool m_json_abi_platform = false;
 
   void fill_option_values(const boost::program_options::variables_map& vm, SubCmdExamineOptions& options) const;
   std::vector<std::shared_ptr<Report>> getReportsList(const xrt_core::smi::tuple_vector&) const;

From 46c61365b4af76c4d64aab2e32a0f544520135fc Mon Sep 17 00:00:00 2001
From: Chiming <chimingz@amd.com>
Date: Tue, 18 Aug 2026 10:50:56 -0700
Subject: [PATCH 2/3] clean up the code.

Signed-off-by: Chiming <chimingz@amd.com>
---
 src/runtime_src/core/common/query_requests.h          | 11 +++++++----
 src/runtime_src/core/common/smi/smi_ryzen.h           |  6 ------
 .../core/tools/common/reports/ReportAieLoad.cpp       |  2 +-
 .../core/tools/common/reports/ReportAieLoad.h         |  8 ++++----
 src/runtime_src/core/tools/xbutil2/SubCmdExamine.cpp  |  8 ++++++--
 5 files changed, 18 insertions(+), 17 deletions(-)

diff --git a/src/runtime_src/core/common/query_requests.h b/src/runtime_src/core/common/query_requests.h
index 5cd20bfa9aa..47cb9305f5b 100644
--- a/src/runtime_src/core/common/query_requests.h
+++ b/src/runtime_src/core/common/query_requests.h
@@ -13,6 +13,7 @@
 #include "core/include/ert.h"
 #include "core/include/xclerr_int.h"
 
+#include <array>
 #include <cstdint>
 #include <iomanip>
 #include <map>
@@ -4322,11 +4323,13 @@ struct aie_coredump : request
 // activity-counter snapshots separated by sample_duration_ms milliseconds.
 struct aie_load : request
 {
+  static constexpr size_t num_activity_counters = 8;
+
   struct result_type {
     uint32_t load_percent;          // 0-100, or UINT32_MAX if unavailable
     uint64_t timestamp_ms;          // FW uptime when second snapshot was taken
-    uint64_t activity_counters[8];  // raw counter values from second snapshot
-    uint64_t operations_per_second; // total counter change across 8 counters per second
+    std::array<uint64_t, num_activity_counters> activity_counters; // raw counter values from second snapshot
+    uint64_t operations_per_second; // total counter change across all activity counters per second
   };
 
   struct args {
@@ -4335,10 +4338,10 @@ struct aie_load : request
 
   static const key_type key = key_type::aie_load;
 
-  virtual std::any
+  std::any
   get(const device*) const override = 0;
 
-  virtual std::any
+  std::any
   get(const device*, const std::any&) const override = 0;
 };
 } // query
diff --git a/src/runtime_src/core/common/smi/smi_ryzen.h b/src/runtime_src/core/common/smi/smi_ryzen.h
index a5bfa81dc87..38b07dcb1bf 100644
--- a/src/runtime_src/core/common/smi/smi_ryzen.h
+++ b/src/runtime_src/core/common/smi/smi_ryzen.h
@@ -76,12 +76,6 @@ class config_gen_npu3 : public config_gen_ryzen {
   {
     return validate_test_desc;
   }
-
-  xrt_core::smi::subcommand
-  create_validate_subcommand() override;
-
-  xrt_core::smi::subcommand
-  create_examine_subcommand() override;
 };
 
 void
diff --git a/src/runtime_src/core/tools/common/reports/ReportAieLoad.cpp b/src/runtime_src/core/tools/common/reports/ReportAieLoad.cpp
index c1640c1fc88..7d253cdeba9 100644
--- a/src/runtime_src/core/tools/common/reports/ReportAieLoad.cpp
+++ b/src/runtime_src/core/tools/common/reports/ReportAieLoad.cpp
@@ -31,7 +31,7 @@ ReportAieLoad::getPropertyTree20202(const xrt_core::device* dev,
     node.put("operations_per_second", data.operations_per_second);
 
     boost::property_tree::ptree pt_counters;
-    for (size_t i = 0; i < 8; i++) {
+    for (size_t i = 0; i < xrt_core::query::aie_load::num_activity_counters; i++) {
       boost::property_tree::ptree pt_counter;
       pt_counter.put("", data.activity_counters[i]);
       pt_counters.push_back({"", pt_counter});
diff --git a/src/runtime_src/core/tools/common/reports/ReportAieLoad.h b/src/runtime_src/core/tools/common/reports/ReportAieLoad.h
index ab2f1a343d8..4d2ec7c06e2 100644
--- a/src/runtime_src/core/tools/common/reports/ReportAieLoad.h
+++ b/src/runtime_src/core/tools/common/reports/ReportAieLoad.h
@@ -16,10 +16,10 @@ class ReportAieLoad : public Report {
   void setDuration(uint32_t duration_ms) { m_duration_ms = duration_ms; }
 
  public:
-  virtual void getPropertyTreeInternal(const xrt_core::device* dev, boost::property_tree::ptree& pt) const;
-  virtual void getPropertyTree20202(const xrt_core::device* dev, boost::property_tree::ptree& pt) const;
-  virtual void writeReport(const xrt_core::device* dev, const boost::property_tree::ptree& pt,
-                           const std::vector<std::string>& elementsFilter, std::ostream& output) const;
+  void getPropertyTreeInternal(const xrt_core::device* dev, boost::property_tree::ptree& pt) const override;
+  void getPropertyTree20202(const xrt_core::device* dev, boost::property_tree::ptree& pt) const override;
+  void writeReport(const xrt_core::device* dev, const boost::property_tree::ptree& pt,
+                   const std::vector<std::string>& elementsFilter, std::ostream& output) const override;
 
  private:
   uint32_t m_duration_ms = 0; // 0 = driver default (XRT_AIE_LOAD_SAMPLE_INTERVAL_MS)
diff --git a/src/runtime_src/core/tools/xbutil2/SubCmdExamine.cpp b/src/runtime_src/core/tools/xbutil2/SubCmdExamine.cpp
index d3891cffd43..1cb822b1fa0 100644
--- a/src/runtime_src/core/tools/xbutil2/SubCmdExamine.cpp
+++ b/src/runtime_src/core/tools/xbutil2/SubCmdExamine.cpp
@@ -357,13 +357,17 @@ SubCmdExamine::execute(const SubCmdOptions& _options) const
       const auto examine_watch_snapshot =
           [&](const xrt_core::device*) {
             std::ostringstream console;
-            XBU::produce_reports(device, reportsToProcess, schemaVersion, {}, console, oSchemaOutput);
+            XBU::produce_reports(device, reportsToProcess, schemaVersion,
+                                 Report::getSchemaDescription(options.m_format).optionName,
+                                 options.m_elementsFilter, console, oSchemaOutput);
             return console.str();
           };
       smi_watch_mode::run_watch_mode(device.get(), std::cout, examine_watch_snapshot,
           *options.m_watchIntervalSec);
     } else {
-      XBU::produce_reports(device, reportsToProcess, schemaVersion, {}, std::cout, oSchemaOutput);
+      XBU::produce_reports(device, reportsToProcess, schemaVersion,
+                           Report::getSchemaDescription(options.m_format).optionName,
+                           options.m_elementsFilter, std::cout, oSchemaOutput);
     }
   } catch (const std::exception&) {
     // Exception is thrown at the end of this function to allow for report writing

From 6aa0ff6b753e186ae3016c14d328d1ab0ceea395 Mon Sep 17 00:00:00 2001
From: Chiming <chimingz@amd.com>
Date: Tue, 18 Aug 2026 14:20:21 -0700
Subject: [PATCH 3/3] remove aie-load standalone option from xrt-smi.

Signed-off-by: Chiming <chimingz@amd.com>
---
 src/runtime_src/core/common/smi/smi_ryzen.cpp |  2 -
 .../tools/common/reports/ReportAieLoad.cpp    | 82 -------------------
 .../core/tools/common/reports/ReportAieLoad.h | 28 -------
 .../core/tools/xbutil2/SubCmdExamine.cpp      | 14 ----
 .../core/tools/xbutil2/SubCmdExamine.h        |  1 -
 5 files changed, 127 deletions(-)
 delete mode 100644 src/runtime_src/core/tools/common/reports/ReportAieLoad.cpp
 delete mode 100644 src/runtime_src/core/tools/common/reports/ReportAieLoad.h

diff --git a/src/runtime_src/core/common/smi/smi_ryzen.cpp b/src/runtime_src/core/common/smi/smi_ryzen.cpp
index 20174656339..1c1cf52e4e8 100644
--- a/src/runtime_src/core/common/smi/smi_ryzen.cpp
+++ b/src/runtime_src/core/common/smi/smi_ryzen.cpp
@@ -15,7 +15,6 @@ config_gen_ryzen::
 config_gen_ryzen()
 {
   examine_report_desc = {
-    {"aie-load", "AIE array load utilization", "common"},
     {"aie-partitions", "AIE partition information", "common"},
     {"all", "All known reports are produced", "common"},
     {"host", "Host information (default)", "common"},
@@ -114,7 +113,6 @@ config_gen_npu3::
 config_gen_npu3()
 {
   examine_report_desc = {
-    {"aie-load", "AIE array load utilization", "common"},
     {"aie-partitions", "AIE partition information", "common"},
     {"all", "All known reports are produced", "common"},
     {"host", "Host information", "common"},
diff --git a/src/runtime_src/core/tools/common/reports/ReportAieLoad.cpp b/src/runtime_src/core/tools/common/reports/ReportAieLoad.cpp
deleted file mode 100644
index 7d253cdeba9..00000000000
--- a/src/runtime_src/core/tools/common/reports/ReportAieLoad.cpp
+++ /dev/null
@@ -1,82 +0,0 @@
-// SPDX-License-Identifier: Apache-2.0
-// Copyright (C) 2025 Advanced Micro Devices, Inc. All rights reserved.
-
-#include "ReportAieLoad.h"
-
-#include "core/common/query_requests.h"
-
-#include <boost/format.hpp>
-#include <limits>
-
-void
-ReportAieLoad::getPropertyTreeInternal(const xrt_core::device* dev,
-                                       boost::property_tree::ptree& pt) const
-{
-  getPropertyTree20202(dev, pt);
-}
-
-void
-ReportAieLoad::getPropertyTree20202(const xrt_core::device* dev,
-                                    boost::property_tree::ptree& pt) const
-{
-  boost::property_tree::ptree node;
-
-  try {
-    const xrt_core::query::aie_load::args args{m_duration_ms};
-    const auto data = xrt_core::device_query<xrt_core::query::aie_load>(dev, args);
-
-    const bool unavailable = (data.load_percent == std::numeric_limits<uint32_t>::max());
-    node.put("load_percent",          unavailable ? "N/A" : std::to_string(data.load_percent));
-    node.put("timestamp_ms",          data.timestamp_ms);
-    node.put("operations_per_second", data.operations_per_second);
-
-    boost::property_tree::ptree pt_counters;
-    for (size_t i = 0; i < xrt_core::query::aie_load::num_activity_counters; i++) {
-      boost::property_tree::ptree pt_counter;
-      pt_counter.put("", data.activity_counters[i]);
-      pt_counters.push_back({"", pt_counter});
-    }
-    node.add_child("activity_counters", pt_counters);
-  }
-  catch (const xrt_core::query::exception& e) {
-    node.put("error", e.what());
-  }
-
-  pt.add_child("aie_load", node);
-}
-
-void
-ReportAieLoad::writeReport(const xrt_core::device* /*dev*/,
-                           const boost::property_tree::ptree& pt,
-                           const std::vector<std::string>& /*elementsFilter*/,
-                           std::ostream& output) const
-{
-  const auto& node = pt.get_child("aie_load");
-
-  // If the query failed, report the error and return.
-  auto error = node.get_optional<std::string>("error");
-  if (error) {
-    output << "AIE Load: " << *error << "\n\n";
-    return;
-  }
-
-  output << "AIE Load\n";
-
-  const auto load = node.get<std::string>("load_percent");
-  if (load == "N/A")
-    output << boost::format("  %-25s: N/A (AIE off, gated, or counters unavailable)\n") % "Utilization";
-  else
-    output << boost::format("  %-25s: %s%%\n") % "Utilization" % load;
-
-  output << boost::format("  %-25s: %s ops/s\n") % "Operations/Second"
-            % node.get<std::string>("operations_per_second");
-  output << boost::format("  %-25s: %s ms\n") % "FW Timestamp"
-            % node.get<std::string>("timestamp_ms");
-
-  output << "  Activity Counters:\n";
-  int idx = 0;
-  for (const auto& [name, counter] : node.get_child("activity_counters"))
-    output << boost::format("    [%d]: %s\n") % idx++ % counter.get_value<std::string>();
-
-  output << "\n";
-}
diff --git a/src/runtime_src/core/tools/common/reports/ReportAieLoad.h b/src/runtime_src/core/tools/common/reports/ReportAieLoad.h
deleted file mode 100644
index 4d2ec7c06e2..00000000000
--- a/src/runtime_src/core/tools/common/reports/ReportAieLoad.h
+++ /dev/null
@@ -1,28 +0,0 @@
-// SPDX-License-Identifier: Apache-2.0
-// Copyright (C) 2025 Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef __ReportAieLoad_h_
-#define __ReportAieLoad_h_
-
-#include "tools/common/Report.h"
-
-#include <cstdint>
-
-class ReportAieLoad : public Report {
- public:
-  ReportAieLoad() : Report("aie-load", "AIE array load utilization", true /*deviceRequired*/) {}
-
-  // Called by SubCmdExamine before produce_reports to forward --duration.
-  void setDuration(uint32_t duration_ms) { m_duration_ms = duration_ms; }
-
- public:
-  void getPropertyTreeInternal(const xrt_core::device* dev, boost::property_tree::ptree& pt) const override;
-  void getPropertyTree20202(const xrt_core::device* dev, boost::property_tree::ptree& pt) const override;
-  void writeReport(const xrt_core::device* dev, const boost::property_tree::ptree& pt,
-                   const std::vector<std::string>& elementsFilter, std::ostream& output) const override;
-
- private:
-  uint32_t m_duration_ms = 0; // 0 = driver default (XRT_AIE_LOAD_SAMPLE_INTERVAL_MS)
-};
-
-#endif
diff --git a/src/runtime_src/core/tools/xbutil2/SubCmdExamine.cpp b/src/runtime_src/core/tools/xbutil2/SubCmdExamine.cpp
index 1cb822b1fa0..b43eaee4939 100644
--- a/src/runtime_src/core/tools/xbutil2/SubCmdExamine.cpp
+++ b/src/runtime_src/core/tools/xbutil2/SubCmdExamine.cpp
@@ -44,7 +44,6 @@
 #include "tools/common/reports/platform/ReportPlatforms.h"
 #include "tools/common/reports/ReportPsKernels.h"
 #include "tools/common/reports/ReportQspiStatus.h"
-#include "tools/common/reports/ReportAieLoad.h"
 #include "tools/common/reports/ReportTelemetry.h"
 #include <cstdint>
 #include "tools/common/reports/ReportThermal.h"
@@ -68,7 +67,6 @@ SubCmdExamine::SubCmdExamine(bool _isHidden, bool _isDepricated, bool _isPrelimi
   uniqueReportCollection = {
   // Common reports
     std::make_shared<ReportAie>(),
-    std::make_shared<ReportAieLoad>(),
     std::make_shared<ReportAieShim>(),
     std::make_shared<ReportAieMem>(),
     std::make_shared<ReportAiePartitions>(),
@@ -107,11 +105,6 @@ SubCmdExamine::SubCmdExamine(bool _isHidden, bool _isDepricated, bool _isPrelimi
     option->setExecutable(getExecutableName());
     option->setCommand(getName());
   }
-
-  // Register --duration directly so it is always recognized without --advanced.
-  m_commonOptions.add_options()
-    ("duration", po::value<uint32_t>()->default_value(0),
-     "Sampling window in milliseconds for the aie-load report (0 = driver default)");
 }
 
 void SubCmdExamine::fill_option_values(const po::variables_map& vm, SubCmdExamineOptions& options) const
@@ -122,7 +115,6 @@ void SubCmdExamine::fill_option_values(const po::variables_map& vm, SubCmdExamin
   options.m_reportNames = vm.count("report") ? vm["report"].as<std::vector<std::string>>() : std::vector<std::string>();
   options.m_help = vm.count("help") ? vm["help"].as<bool>() : false;
   options.m_elementsFilter = vm.count("element") ? vm["element"].as<std::vector<std::string>>() : std::vector<std::string>();
-  options.m_duration = vm["duration"].as<uint32_t>();
   options.m_watchIntervalSec.reset();
   if (vm.count("watch")) {
     const auto& s = vm["watch"].as<std::string>();
@@ -343,12 +335,6 @@ SubCmdExamine::execute(const SubCmdOptions& _options) const
     }
   }
 
-  // Propagate --duration to ReportAieLoad before running reports.
-  for (const auto& report : reportsToProcess) {
-    if (auto* aieLoad = dynamic_cast<ReportAieLoad*>(report.get()))
-      aieLoad->setDuration(options.m_duration);
-  }
-
   // Create the report
   std::ostringstream oSchemaOutput;
   try {
diff --git a/src/runtime_src/core/tools/xbutil2/SubCmdExamine.h b/src/runtime_src/core/tools/xbutil2/SubCmdExamine.h
index 7eb56f10a94..5d1af47c4d9 100644
--- a/src/runtime_src/core/tools/xbutil2/SubCmdExamine.h
+++ b/src/runtime_src/core/tools/xbutil2/SubCmdExamine.h
@@ -28,7 +28,6 @@ struct SubCmdExamineOptions {
   std::string               m_output;
   bool                      m_help;
   std::optional<unsigned>   m_watchIntervalSec;
-  uint32_t                  m_duration = 0; // sampling window in ms for aie-load (0 = driver default)
 };
 class SubCmdExamine : public SubCmd {
   ReportCollection uniqueReportCollection;