Capture and Replay: Add support for xrt::runlist APIs

Signed-off-by: Harsh Wardhan <[email protected]>

Author: theharshwardhan

src/runtime_src/core/tools/xbreplay/CMakeLists.txt

@@ -22,6 +22,7 @@ add_library(xbreplay_objects OBJECT
   src/replay_xrt/replay_xrt_kernel.cpp
   src/replay_xrt/replay_xrt_run.cpp
   src/replay_xrt/replay_xrt_xclbin.cpp
+  src/replay_xrt/replay_xrt_runlist.cpp
 )
 
 target_include_directories(xbreplay_objects PRIVATE

src/runtime_src/core/tools/xbreplay/src/replay_xrt/replay_xrt.hpp

@@ -6,6 +6,7 @@
 #include "experimental/xrt_device.h"
 #include "experimental/xrt_xclbin.h"
 #include "experimental/xrt_ext.h"
+#include "experimental/xrt_kernel.h"
 #include "xrt/xrt_kernel.h"
 #include "xrt.h"
 #include "xclbin.h"
@@ -61,6 +62,9 @@ class replay_xrt
   /*Map between handle from log and bo */
   std::unordered_map<uint64_t, std::shared_ptr<xrt::xclbin>> m_xclbin_hndle_map;
 
+  /*Map between handle from log and runlist */
+  std::unordered_map<uint64_t, std::shared_ptr<xrt::runlist>> m_runlist_hndle_map;
+
   /*Map between group id */
   std::unordered_map<uint64_t, xrt::memory_group> m_kernel_grp_id;
 
@@ -87,6 +91,9 @@ class replay_xrt
   /*Register hw_ctx class API's*/
   void register_hwctxt_class_func();
 
+  /* Registers runlist class API's */
+  void register_runlist_class_func();
+
   /* Validate's file path */
   bool is_file_exist(const std::string& fileName)
   {
@@ -124,6 +131,9 @@ class replay_xrt
 
     /* Register HW CTX Class APIs*/
     register_hwctxt_class_func();
+
+    /*Register Runlist class API's */
+    register_runlist_class_func();
   }
 
   /*
@@ -159,6 +169,7 @@ class replay_xrt
     m_xclbin_hndle_map.clear();
     m_hwctx_hndle_map.clear();
     m_device_hndle_map.clear();
+    m_runlist_hndle_map.clear();
   }
 
   /*

src/runtime_src/core/tools/xbreplay/src/replay_xrt/replay_xrt_runlist.cpp

@@ -0,0 +1,87 @@
+// SPDX-License-Identifier: Apache-2.0
+// Copyright (C) 2024 Advanced Micro Devices, Inc. All rights reserved.
+
+#include "replay_xrt.hpp"
+
+namespace xrt_core::tools::xbreplay {
+
+/**
+ * Replay maintains a map where each member function of the XRT classes is associated
+ * with a corresponding lambda function. This API adds a lambda function entry for each
+ * corresponding member of the xrt::runlist class.
+ */
+void replay_xrt::register_runlist_class_func()
+{
+  m_api_map["xrt::runlist::runlist(const xrt::hw_context&)"] =
+  [this] (std::shared_ptr<utils::message> msg)
+  {
+    const std::vector<std::pair<std::string, std::string>>& args = msg->m_args;
+    auto hwctx_handle = std::stoull(args[0].second, nullptr, utils::base_hex);
+    auto phwctx = m_hwctx_hndle_map[hwctx_handle];
+
+    if (phwctx)
+    {
+      const xrt::hw_context& hwctx = *phwctx;
+      auto runlist_ptr = std::make_shared<xrt::runlist>(hwctx);
+      m_runlist_hndle_map[msg->m_ret_val] = runlist_ptr;
+    }
+    else
+      throw std::runtime_error("Failed to get hardware context handle");
+  };
+
+  m_api_map["xrt::runlist::add(const xrt::run&)"] =
+  [this] (std::shared_ptr<utils::message> msg)
+  {
+    const std::vector<std::pair<std::string, std::string>>& args = msg->m_args;
+    auto runlist_handle = std::stoull(args[0].second, nullptr, utils::base_hex);
+    auto run_handle = std::stoull(args[1].second, nullptr, utils::base_hex);
+
+    auto runlist_ptr = m_runlist_hndle_map[runlist_handle];
+    auto run_ptr = m_run_hndle_map[run_handle];
+
+    if (runlist_ptr && run_ptr)
+    {
+      const xrt::run& run = *run_ptr;
+      runlist_ptr->add(run);
+    }
+    else
+      throw std::runtime_error("Failed to get runlist or run handle");
+  };
+
+  m_api_map["xrt::runlist::execute()"] =
+  [this] (std::shared_ptr<utils::message> msg)
+  {
+    auto runlist_ptr = m_runlist_hndle_map[msg->m_handle];
+
+    if (runlist_ptr)
+      runlist_ptr->execute();
+    else
+      throw std::runtime_error("Failed to get runlist handle");
+  };
+
+  m_api_map["xrt::runlist::wait(const std::chrono::milliseconds&)"] =
+  [this] (std::shared_ptr<utils::message> msg)
+  {
+    const std::vector<std::pair<std::string, std::string>>& args = msg->m_args;
+    auto runlist_ptr = m_runlist_hndle_map[msg->m_handle];
+    auto time = std::stoll(args[0].second);
+    const std::chrono::milliseconds timeout(time);
+    
+    if (runlist_ptr)
+      runlist_ptr->wait(timeout);
+    else
+      throw std::runtime_error("Failed to get runlist handle");
+  };
+
+  m_api_map["xrt::runlist::reset()"] =
+  [this] (std::shared_ptr<utils::message> msg)
+  {
+    auto runlist_ptr = m_runlist_hndle_map[msg->m_handle];
+
+    if (runlist_ptr)
+      runlist_ptr->reset();
+    else
+      throw std::runtime_error("Failed to get runlist handle");
+  };
+}
+}

src/runtime_src/core/tools/xbtracer/src/lib/capture.cpp

@@ -105,6 +105,13 @@ const static std::unordered_map < std::string, void **> fname2fptr_map = {
   {"xrt::run::set_arg_at_index(int, xrt::bo const&)", (void **) &dtbl.run.set_arg2},
   {"xrt::run::update_arg_at_index(int, void const*, unsigned long)", (void **) &dtbl.run.update_arg3},
   {"xrt::run::update_arg_at_index(int, xrt::bo const&)", (void **) &dtbl.run.update_arg2},
+  
+  /* runlist class maps */
+  {"xrt::runlist::runlist(xrt::hw_context const&)", (void **) &dtbl.runlist.ctor},
+  {"xrt::runlist::add(xrt::run const&)", (void **) &dtbl.runlist.add},
+  {"xrt::runlist::execute()", (void **) &dtbl.runlist.execute},
+  {"xrt::runlist::wait(std::chrono::duration<long, std::ratio<1l, 1000l> > const&) const", (void **) &dtbl.runlist.wait},
+  {"xrt::runlist::state()", (void **) &dtbl.runlist.reset},
 
   /* kernel class maps */
   {"xrt::kernel::kernel(xrt::device const&, xrt::uuid const&, std::string const&, xrt::kernel::cu_access_mode)", (void **) &dtbl.kernel.ctor},

src/runtime_src/core/tools/xbtracer/src/lib/capture.h

@@ -19,6 +19,7 @@ class xrt_ftbl
   xrt_bo_ftbl         bo;
   xrt_kernel_ftbl     kernel;
   xrt_run_ftbl        run;
+  xrt_runlist_ftbl    runlist;
   xrt_xclbin_ftbl     xclbin;
   xrt_hw_context_ftbl hw_context;
   xrt_ext_ftbl        ext;

src/runtime_src/core/tools/xbtracer/src/lib/logger.cpp

@@ -174,6 +174,7 @@ void logger::synth_dtor_trace_fn()
     run |= check_ref_count(m_krnl_ref_tracker);
     run |= check_ref_count(m_bo_ref_tracker);
     run |= check_ref_count(m_hw_cnxt_ref_tracker);
+    run |= check_ref_count(m_runlist_ref_tracker);
 
     if (m_is_destructing == false)
 	run = true;

src/runtime_src/core/tools/xbtracer/src/lib/logger.h

@@ -19,6 +19,7 @@
 #include "xrt/xrt_bo.h"
 #include "xrt/xrt_device.h"
 #include "xrt/xrt_kernel.h"
+#include "experimental/xrt_kernel.h"
 
 #ifdef _WIN32
 # include <windows.h>
@@ -103,6 +104,8 @@ class logger
                          std::string>> m_krnl_ref_tracker;
   std::vector<std::tuple<std::shared_ptr<run_impl>, std::thread::id,
                          std::string>> m_run_ref_tracker;
+  std::vector<std::tuple<std::shared_ptr<runlist_impl>, std::thread::id,
+                         std::string>> m_runlist_ref_tracker;
   std::vector<std::tuple<std::shared_ptr<bo_impl>, std::thread::id,
                          std::string>> m_bo_ref_tracker;
   std::vector<std::tuple<std::shared_ptr<hw_context_impl>, std::thread::id,
@@ -178,6 +181,12 @@ class logger
           std::this_thread::get_id(), "xrt::run::~run()"));
   }
 
+  void set_pimpl(std::shared_ptr<runlist_impl> hpimpl)
+  {
+    m_runlist_ref_tracker.emplace_back(std::make_tuple(hpimpl,
+          std::this_thread::get_id(), "xrt::runlist::~runlist()"));
+  }
+
   void set_pimpl(std::shared_ptr<hw_context_impl> hpimpl)
   {
     m_hw_cnxt_ref_tracker.emplace_back(std::make_tuple(hpimpl,

src/runtime_src/core/tools/xbtracer/src/lib/xrt_kernel_inst.cpp

@@ -4,6 +4,7 @@
 #include <iostream>
 
 #define XCL_DRIVER_DLL_EXPORT
+#define XRT_API_SOURCE
 #include "capture.h"
 #include "logger.h"
 
@@ -13,7 +14,7 @@ namespace xtx = xrt::tools::xbtracer;
 const xtx::xrt_ftbl& dtbl = xtx::xrt_ftbl::get_instance();
 
 /*
- *  kernel/run class instrumented methods
+ *  kernel/run/runlist class instrumented methods
  * */
 namespace xrt {
 XCL_DRIVER_DLLESPEC
@@ -254,4 +255,52 @@ xrt::xclbin kernel::get_xclbin() const
   return xclbin;
 }
 
+XRT_API_EXPORT
+runlist::runlist(const xrt::hw_context& hwctx)
+{
+  auto func = "xrt::runlist::runlist(const xrt::hw_context&)";
+  XRT_TOOLS_XBT_CALL_CTOR(dtbl.runlist.ctor, this, hwctx);
+  /* As pimpl will be updated only after ctor call */
+  XRT_TOOLS_XBT_FUNC_ENTRY(func, hwctx.get_handle().get());
+  XRT_TOOLS_XBT_FUNC_EXIT(func);
+}
+
+XRT_API_EXPORT
+void runlist::add(const xrt::run& run)
+{
+  auto func = "xrt::runlist::add(const xrt::run&)";
+  XRT_TOOLS_XBT_FUNC_ENTRY(func, run.get_handle().get());
+  XRT_TOOLS_XBT_CALL_METD(dtbl.runlist.add, run);
+  XRT_TOOLS_XBT_FUNC_EXIT(func);
+}
+
+XRT_API_EXPORT
+void runlist::execute()
+{
+  auto func = "xrt::runlist::execute()";
+  XRT_TOOLS_XBT_FUNC_ENTRY(func, this);
+  XRT_TOOLS_XBT_CALL_METD(dtbl.runlist.execute);
+  XRT_TOOLS_XBT_FUNC_EXIT(func);
+}
+
+XRT_API_EXPORT
+std::cv_status runlist::wait(const std::chrono::milliseconds& timeout) const
+{
+  auto func = "xrt::runlist::wait(const std::chrono::milliseconds&)";
+  XRT_TOOLS_XBT_FUNC_ENTRY(func, timeout.count());
+  std::cv_status status = std::cv_status::no_timeout;
+  XRT_TOOLS_XBT_CALL_METD_RET(dtbl.runlist.wait, status, timeout);
+  XRT_TOOLS_XBT_FUNC_EXIT_RET(func, (int)status);
+  return status;
+}
+
+XRT_API_EXPORT
+void runlist::reset()
+{
+  auto func = "xrt::runlist::reset()";
+  XRT_TOOLS_XBT_FUNC_ENTRY(func, this);
+  XRT_TOOLS_XBT_CALL_METD(dtbl.runlist.reset);
+  XRT_TOOLS_XBT_FUNC_EXIT(func);
+}
+
 }  // namespace xrt

src/runtime_src/core/tools/xbtracer/src/lib/xrt_kernel_inst.h

@@ -4,6 +4,7 @@
 #pragma once
 
 #include "xrt/xrt_kernel.h"
+#include "experimental/xrt_kernel.h"
 
 /*
  * Run class method aliases.
@@ -63,7 +64,7 @@ using xrt_kernel_ctor = xrt::kernel* (*)(void*, const xrt::device&,\
   const xrt::uuid&, const std::string&, xrt::kernel::cu_access_mode);
 using xrt_kernel_ctor2 = xrt::kernel* (*)(void*, const xrt::hw_context&,\
   const std::string&);
-/*TODO: Marked obselete - should be keep this ?*/
+/*TODO: Marked obselete - should we keep this ?*/
 using xrt_kernel_ctor_obs = xrt::kernel* (*)(void* cxt, xclDeviceHandle,\
   const xrt::uuid&, const std::string&, xrt::kernel::cu_access_mode);
 using xrt_kernel_group_id = int (xrt::kernel::*)(int) const;
@@ -90,3 +91,25 @@ struct xrt_kernel_ftbl
   xrt_kernel_get_xclbin get_xclbin;
   xrt_kernel_dtor dtor;
 };
+
+/*
+ * Runlist class method aliases.
+ */
+using xrt_runlist_ctor = xrt::runlist (*)(void*, const xrt::hw_context&);
+using xrt_runlist_add = void (xrt::runlist::*)(const xrt::run&);
+using xrt_runlist_execute = void (xrt::runlist::*)(void);
+using xrt_runlist_wait = std::cv_status (xrt::runlist::*)\
+  (const std::chrono::milliseconds&) const;
+using xrt_runlist_reset = void (xrt::runlist::*)(void);
+
+/*
+ * struct xrt_runlist_ftbl definition.
+ */
+struct xrt_runlist_ftbl
+{
+  xrt_runlist_ctor ctor;
+  xrt_runlist_add add;
+  xrt_runlist_execute execute;
+  xrt_runlist_wait wait;
+  xrt_runlist_reset reset;
+};