diff --git a/lldb/packages/Python/lldbsuite/test/lldbreverse.py b/lldb/packages/Python/lldbsuite/test/lldbreverse.py
diff --git a/lldb/packages/Python/lldbsuite/test/lldbtest.py b/lldb/packages/Python/lldbsuite/test/lldbtest.py
@@ -143,6 +143,8 @@
 
 STOPPED_DUE_TO_WATCHPOINT = "Process should be stopped due to watchpoint"
 
+STOPPED_DUE_TO_HISTORY_BOUNDARY = "Process should be stopped due to history boundary"
+
 DATA_TYPES_DISPLAYED_CORRECTLY = "Data type(s) displayed correctly"
 
 VALID_BREAKPOINT = "Got a valid breakpoint"

diff --git a/lldb/source/API/SBProcess.cpp b/lldb/source/API/SBProcess.cpp
@@ -564,6 +564,10 @@ uint32_t SBProcess::GetAddressByteSize() const {
 }
 
 SBError SBProcess::Continue() {
+  return Continue(RunDirection::eRunForward);
+}
+
+SBError SBProcess::Continue(RunDirection direction) {
   LLDB_INSTRUMENT_VA(this);
 
   SBError sb_error;
@@ -574,9 +578,9 @@ SBError SBProcess::Continue() {
         process_sp->GetTarget().GetAPIMutex());
 
     if (process_sp->GetTarget().GetDebugger().GetAsyncExecution())
-      sb_error.ref() = process_sp->Resume();
+      sb_error.ref() = process_sp->Resume(direction);
     else
-      sb_error.ref() = process_sp->ResumeSynchronous(nullptr);
+      sb_error.ref() = process_sp->ResumeSynchronous(nullptr, direction);
   } else
     sb_error = Status::FromErrorString("SBProcess is invalid");
 

diff --git a/lldb/source/API/SBThread.cpp b/lldb/source/API/SBThread.cpp
@@ -172,6 +172,7 @@ size_t SBThread::GetStopReasonDataCount() {
         case eStopReasonInstrumentation:
         case eStopReasonProcessorTrace:
         case eStopReasonVForkDone:
+        case eStopReasonHistoryBoundary:
           // There is no data for these stop reasons.
           return 0;
 
@@ -233,6 +234,7 @@ uint64_t SBThread::GetStopReasonDataAtIndex(uint32_t idx) {
         case eStopReasonInstrumentation:
         case eStopReasonProcessorTrace:
         case eStopReasonVForkDone:
+        case eStopReasonHistoryBoundary:
           // There is no data for these stop reasons.
           return 0;
 

diff --git a/lldb/source/Interpreter/CommandInterpreter.cpp b/lldb/source/Interpreter/CommandInterpreter.cpp
@@ -2553,7 +2553,8 @@ bool CommandInterpreter::DidProcessStopAbnormally() const {
     const StopReason reason = stop_info->GetStopReason();
     if (reason == eStopReasonException ||
         reason == eStopReasonInstrumentation ||
-        reason == eStopReasonProcessorTrace || reason == eStopReasonInterrupt)
+        reason == eStopReasonProcessorTrace || reason == eStopReasonInterrupt ||
+        reason == eStopReasonHistoryBoundary)
       return true;
 
     if (reason == eStopReasonSignal) {

diff --git a/lldb/source/Plugins/Process/Linux/NativeThreadLinux.cpp b/lldb/source/Plugins/Process/Linux/NativeThreadLinux.cpp
@@ -82,6 +82,9 @@ void LogThreadStopInfo(Log &log, const ThreadStopInfo &stop_info,
   case eStopReasonProcessorTrace:
     log.Printf("%s: %s processor trace", __FUNCTION__, header);
     return;
+  case eStopReasonHistoryBoundary:
+    log.Printf("%s: %s history boundary", __FUNCTION__, header);
+    return;
   default:
     log.Printf("%s: %s invalid stop reason %" PRIu32, __FUNCTION__, header,
                static_cast<uint32_t>(stop_info.reason));

diff --git a/lldb/source/Plugins/Process/MacOSX-Kernel/ProcessKDP.cpp b/lldb/source/Plugins/Process/MacOSX-Kernel/ProcessKDP.cpp
@@ -402,9 +402,16 @@ lldb_private::DynamicLoader *ProcessKDP::GetDynamicLoader() {
 
 Status ProcessKDP::WillResume() { return Status(); }
 
-Status ProcessKDP::DoResume() {
+Status ProcessKDP::DoResume(RunDirection direction) {
   Status error;
   Log *log = GetLog(KDPLog::Process);
+
+  if (direction == RunDirection::eRunReverse) {
+    error.SetErrorStringWithFormatv(
+        "error: {0} does not support reverse execution of processes", GetPluginName());
+    return error;
+  }
+
   // Only start the async thread if we try to do any process control
   if (!m_async_thread.IsJoinable())
     StartAsyncThread();

diff --git a/lldb/source/Plugins/Process/MacOSX-Kernel/ProcessKDP.h b/lldb/source/Plugins/Process/MacOSX-Kernel/ProcessKDP.h
@@ -90,7 +90,7 @@ class ProcessKDP : public lldb_private::Process {
   // Process Control
   lldb_private::Status WillResume() override;
 
-  lldb_private::Status DoResume() override;
+  lldb_private::Status DoResume(lldb::RunDirection direction) override;
 
   lldb_private::Status DoHalt(bool &caused_stop) override;
 

diff --git a/lldb/source/Plugins/Process/Windows/Common/ProcessWindows.cpp b/lldb/source/Plugins/Process/Windows/Common/ProcessWindows.cpp
@@ -204,11 +204,17 @@ ProcessWindows::DoAttachToProcessWithID(lldb::pid_t pid,
   return error;
 }
 
-Status ProcessWindows::DoResume() {
+Status ProcessWindows::DoResume(RunDirection direction) {
   Log *log = GetLog(WindowsLog::Process);
   llvm::sys::ScopedLock lock(m_mutex);
   Status error;
 
+  if (direction == RunDirection::eRunReverse) {
+    error.SetErrorStringWithFormatv(
+        "error: {0} does not support reverse execution of processes", GetPluginName());
+    return error;
+  }
+
   StateType private_state = GetPrivateState();
   if (private_state == eStateStopped || private_state == eStateCrashed) {
     LLDB_LOG(log, "process {0} is in state {1}.  Resuming...",

diff --git a/lldb/source/Plugins/Process/Windows/Common/ProcessWindows.h b/lldb/source/Plugins/Process/Windows/Common/ProcessWindows.h
@@ -52,7 +52,7 @@ class ProcessWindows : public Process, public ProcessDebugger {
   Status DoAttachToProcessWithID(
       lldb::pid_t pid,
       const lldb_private::ProcessAttachInfo &attach_info) override;
-  Status DoResume() override;
+  Status DoResume(lldb::RunDirection direction) override;
   Status DoDestroy() override;
   Status DoHalt(bool &caused_stop) override;
 

diff --git a/lldb/source/Plugins/Process/gdb-remote/GDBRemoteCommunicationClient.cpp b/lldb/source/Plugins/Process/gdb-remote/GDBRemoteCommunicationClient.cpp
@@ -199,6 +199,20 @@ uint64_t GDBRemoteCommunicationClient::GetRemoteMaxPacketSize() {
   return m_max_packet_size;
 }
 
+bool GDBRemoteCommunicationClient::GetReverseContinueSupported() {
+  if (m_supports_reverse_continue == eLazyBoolCalculate) {
+    GetRemoteQSupported();
+  }
+  return m_supports_reverse_continue == eLazyBoolYes;
+}
+
+bool GDBRemoteCommunicationClient::GetReverseStepSupported() {
+  if (m_supports_reverse_step == eLazyBoolCalculate) {
+    GetRemoteQSupported();
+  }
+  return m_supports_reverse_step == eLazyBoolYes;
+}
+
 bool GDBRemoteCommunicationClient::QueryNoAckModeSupported() {
   if (m_supports_not_sending_acks == eLazyBoolCalculate) {
     m_send_acks = true;
@@ -295,6 +309,8 @@ void GDBRemoteCommunicationClient::ResetDiscoverableSettings(bool did_exec) {
     m_supports_qXfer_siginfo_read = eLazyBoolCalculate;
     m_supports_augmented_libraries_svr4_read = eLazyBoolCalculate;
     m_uses_native_signals = eLazyBoolCalculate;
+    m_supports_reverse_continue = eLazyBoolCalculate;
+    m_supports_reverse_step = eLazyBoolCalculate;
     m_supports_qProcessInfoPID = true;
     m_supports_qfProcessInfo = true;
     m_supports_qUserName = true;
@@ -348,6 +364,8 @@ void GDBRemoteCommunicationClient::GetRemoteQSupported() {
   m_supports_memory_tagging = eLazyBoolNo;
   m_supports_qSaveCore = eLazyBoolNo;
   m_uses_native_signals = eLazyBoolNo;
+  m_supports_reverse_continue = eLazyBoolNo;
+  m_supports_reverse_step = eLazyBoolNo;
 
   m_max_packet_size = UINT64_MAX; // It's supposed to always be there, but if
                                   // not, we assume no limit
@@ -401,6 +419,10 @@ void GDBRemoteCommunicationClient::GetRemoteQSupported() {
         m_supports_qSaveCore = eLazyBoolYes;
       else if (x == "native-signals+")
         m_uses_native_signals = eLazyBoolYes;
+      else if (x == "ReverseContinue+")
+        m_supports_reverse_continue = eLazyBoolYes;
+      else if (x == "ReverseStep+")
+        m_supports_reverse_step = eLazyBoolYes;
       // Look for a list of compressions in the features list e.g.
       // qXfer:features:read+;PacketSize=20000;qEcho+;SupportedCompressions=zlib-
       // deflate,lzma

diff --git a/lldb/source/Plugins/Process/gdb-remote/GDBRemoteCommunicationClient.h b/lldb/source/Plugins/Process/gdb-remote/GDBRemoteCommunicationClient.h
@@ -331,6 +331,10 @@ class GDBRemoteCommunicationClient : public GDBRemoteClientBase {
 
   bool GetMultiprocessSupported();
 
+  bool GetReverseContinueSupported();
+
+  bool GetReverseStepSupported();
+
   LazyBool SupportsAllocDeallocMemory() // const
   {
     // Uncomment this to have lldb pretend the debug server doesn't respond to
@@ -561,6 +565,8 @@ class GDBRemoteCommunicationClient : public GDBRemoteClientBase {
   LazyBool m_supports_memory_tagging = eLazyBoolCalculate;
   LazyBool m_supports_qSaveCore = eLazyBoolCalculate;
   LazyBool m_uses_native_signals = eLazyBoolCalculate;
+  LazyBool m_supports_reverse_continue = eLazyBoolCalculate;
+  LazyBool m_supports_reverse_step = eLazyBoolCalculate;
 
   bool m_supports_qProcessInfoPID : 1, m_supports_qfProcessInfo : 1,
       m_supports_qUserName : 1, m_supports_qGroupName : 1,

diff --git a/lldb/source/Plugins/Process/gdb-remote/GDBRemoteCommunicationServerLLGS.cpp b/lldb/source/Plugins/Process/gdb-remote/GDBRemoteCommunicationServerLLGS.cpp
@@ -716,6 +716,7 @@ static const char *GetStopReasonString(StopReason stop_reason) {
     return "vforkdone";
   case eStopReasonInterrupt:
     return "async interrupt";
+  case eStopReasonHistoryBoundary:
   case eStopReasonInstrumentation:
   case eStopReasonInvalid:
   case eStopReasonPlanComplete:

diff --git a/lldb/source/Plugins/Process/gdb-remote/ProcessGDBRemote.cpp b/lldb/source/Plugins/Process/gdb-remote/ProcessGDBRemote.cpp
@@ -169,6 +169,10 @@ class PluginProperties : public Properties {
   }
 };
 
+std::chrono::seconds ResumeTimeout() {
+  return std::chrono::seconds(5);
+}
+
 } // namespace
 
 static PluginProperties &GetGlobalPluginProperties() {
@@ -1180,10 +1184,11 @@ Status ProcessGDBRemote::WillResume() {
   return Status();
 }
 
-Status ProcessGDBRemote::DoResume() {
+Status ProcessGDBRemote::DoResume(RunDirection direction) {
   Status error;
   Log *log = GetLog(GDBRLog::Process);
-  LLDB_LOGF(log, "ProcessGDBRemote::Resume()");
+  LLDB_LOGF(log, "ProcessGDBRemote::Resume(%s)",
+            direction == RunDirection::eRunForward ? "" : "reverse");
 
   ListenerSP listener_sp(
       Listener::MakeListener("gdb-remote.resume-packet-sent"));
@@ -1197,12 +1202,21 @@ Status ProcessGDBRemote::DoResume() {
 
     StreamString continue_packet;
     bool continue_packet_error = false;
-    if (m_gdb_comm.HasAnyVContSupport()) {
+    // Number of threads continuing with "c", i.e. continuing without a signal to deliver.
+    const size_t num_continue_c_tids = m_continue_c_tids.size();
+    // Number of threads continuing with "C", i.e. continuing with a signal to deliver.
+    const size_t num_continue_C_tids = m_continue_C_tids.size();
+    // Number of threads continuing with "s", i.e. single-stepping.
+    const size_t num_continue_s_tids = m_continue_s_tids.size();
+    // Number of threads continuing with "S", i.e. single-stepping with a signal to deliver.
+    const size_t num_continue_S_tids = m_continue_S_tids.size();
+    if (direction == RunDirection::eRunForward &&
+        m_gdb_comm.HasAnyVContSupport()) {
       std::string pid_prefix;
       if (m_gdb_comm.GetMultiprocessSupported())
         pid_prefix = llvm::formatv("p{0:x-}.", GetID());
 
-      if (m_continue_c_tids.size() == num_threads ||
+      if (num_continue_c_tids == num_threads ||
           (m_continue_c_tids.empty() && m_continue_C_tids.empty() &&
            m_continue_s_tids.empty() && m_continue_S_tids.empty())) {
         // All threads are continuing
@@ -1265,14 +1279,11 @@ Status ProcessGDBRemote::DoResume() {
     } else
       continue_packet_error = true;
 
-    if (continue_packet_error) {
+    if (direction == RunDirection::eRunForward && continue_packet_error) {
       // Either no vCont support, or we tried to use part of the vCont packet
-      // that wasn't supported by the remote GDB server. We need to try and
-      // make a simple packet that can do our continue
-      const size_t num_continue_c_tids = m_continue_c_tids.size();
-      const size_t num_continue_C_tids = m_continue_C_tids.size();
-      const size_t num_continue_s_tids = m_continue_s_tids.size();
-      const size_t num_continue_S_tids = m_continue_S_tids.size();
+      // that wasn't supported by the remote GDB server, or it's the reverse
+      // direction. We need to try and make a simple packet that can do our
+      // continue.
       if (num_continue_c_tids > 0) {
         if (num_continue_c_tids == num_threads) {
           // All threads are resuming...
@@ -1363,9 +1374,41 @@ Status ProcessGDBRemote::DoResume() {
       }
     }
 
+    if (direction == RunDirection::eRunReverse && continue_packet_error) {
+      if (num_continue_C_tids > 0 || num_continue_S_tids > 0) {
+        LLDB_LOGF(log, "ProcessGDBRemote::DoResumeReverse: Signals not supported");
+        return Status::FromErrorString("can't deliver signals while running in reverse");
+      }
+
+      if (num_continue_s_tids > 0) {
+        if (num_continue_s_tids > 1) {
+          LLDB_LOGF(log, "ProcessGDBRemote::DoResumeReverse: can't step multiple threads");
+          return Status::FromErrorString("can't step multiple threads while reverse-stepping");
+        }
+
+        if (!m_gdb_comm.GetReverseStepSupported()) {
+          LLDB_LOGF(log, "ProcessGDBRemote::DoResumeReverse: target does not support reverse-stepping");
+          return Status::FromErrorString("target does not support reverse-stepping");
+        }
+
+        m_gdb_comm.SetCurrentThreadForRun(m_continue_s_tids.front());
+        continue_packet.PutCString("bs");
+      } else {
+        if (!m_gdb_comm.GetReverseContinueSupported()) {
+          LLDB_LOGF(log, "ProcessGDBRemote::DoResumeReverse: target does not support reverse-continue");
+          return Status::FromErrorString("target does not support reverse-continue");
+        }
+
+        // All threads continue whether requested or not ---
+        // we can't change how threads ran in the past.
+        continue_packet.PutCString("bc");
+      }
+
+      continue_packet_error = false;
+    }
+
     if (continue_packet_error) {
-      error =
-          Status::FromErrorString("can't make continue packet for this resume");
+      return Status::FromErrorString("can't make continue packet for this resume");
     } else {
       EventSP event_sp;
       if (!m_async_thread.IsJoinable()) {
@@ -1380,7 +1423,7 @@ Status ProcessGDBRemote::DoResume() {
           std::make_shared<EventDataBytes>(continue_packet.GetString());
       m_async_broadcaster.BroadcastEvent(eBroadcastBitAsyncContinue, data_sp);
 
-      if (!listener_sp->GetEvent(event_sp, std::chrono::seconds(5))) {
+      if (!listener_sp->GetEvent(event_sp, ResumeTimeout())) {
         error = Status::FromErrorString("Resume timed out.");
         LLDB_LOGF(log, "ProcessGDBRemote::DoResume: Resume timed out.");
       } else if (event_sp->BroadcasterIs(&m_async_broadcaster)) {
@@ -1863,6 +1906,10 @@ ThreadSP ProcessGDBRemote::SetThreadStopInfo(
           thread_sp->SetStopInfo(StopInfo::CreateStopReasonWithException(
               *thread_sp, description.c_str()));
           handled = true;
+        } else if (reason == "replaylog") {
+          thread_sp->SetStopInfo(StopInfo::CreateStopReasonHistoryBoundary(
+              *thread_sp, description.c_str()));
+          handled = true;
         } else if (reason == "exec") {
           did_exec = true;
           thread_sp->SetStopInfo(
@@ -2318,6 +2365,8 @@ StateType ProcessGDBRemote::SetThreadStopInfo(StringExtractor &stop_packet) {
         description = std::string(ostr.GetString());
       } else if (key.compare("swbreak") == 0 || key.compare("hwbreak") == 0) {
         reason = "breakpoint";
+      } else if (key.compare("replaylog") == 0) {
+        reason = "replaylog";
       } else if (key.compare("library") == 0) {
         auto error = LoadModules();
         if (error) {

diff --git a/lldb/source/Plugins/Process/gdb-remote/ProcessGDBRemote.h b/lldb/source/Plugins/Process/gdb-remote/ProcessGDBRemote.h
@@ -111,7 +111,7 @@ class ProcessGDBRemote : public Process,
   // Process Control
   Status WillResume() override;
 
-  Status DoResume() override;
+  Status DoResume(lldb::RunDirection direction) override;
 
   Status DoHalt(bool &caused_stop) override;
 

diff --git a/lldb/source/Plugins/Process/scripted/ScriptedProcess.cpp b/lldb/source/Plugins/Process/scripted/ScriptedProcess.cpp
@@ -182,10 +182,15 @@ void ScriptedProcess::DidResume() {
   m_pid = GetInterface().GetProcessID();
 }
 
-Status ScriptedProcess::DoResume() {
+Status ScriptedProcess::DoResume(RunDirection direction) {
   LLDB_LOGF(GetLog(LLDBLog::Process), "ScriptedProcess::%s resuming process", __FUNCTION__);
 
-  return GetInterface().Resume();
+  if (direction == RunDirection::eRunForward) {
+    return GetInterface().Resume();
+  } else {
+    return Status::FromErrorStringWithFormatv(
+        "error: {0} does not support reverse execution of processes", GetPluginName());
+  }
 }
 
 Status ScriptedProcess::DoAttach(const ProcessAttachInfo &attach_info) {

diff --git a/lldb/source/Plugins/Process/scripted/ScriptedProcess.h b/lldb/source/Plugins/Process/scripted/ScriptedProcess.h
@@ -52,7 +52,7 @@ class ScriptedProcess : public Process {
 
   void DidResume() override;
 
-  Status DoResume() override;
+  Status DoResume(lldb::RunDirection direction) override;
 
   Status DoAttachToProcessWithID(lldb::pid_t pid,
                                  const ProcessAttachInfo &attach_info) override;

diff --git a/lldb/source/Target/Process.cpp b/lldb/source/Target/Process.cpp
@@ -446,7 +446,8 @@ Process::Process(lldb::TargetSP target_sp, ListenerSP listener_sp,
       m_memory_cache(*this), m_allocated_memory_cache(*this),
       m_should_detach(false), m_next_event_action_up(), m_public_run_lock(),
       m_private_run_lock(), m_currently_handling_do_on_removals(false),
-      m_resume_requested(false), m_interrupt_tid(LLDB_INVALID_THREAD_ID),
+      m_resume_requested(false), m_last_run_direction(eRunForward),
+      m_interrupt_tid(LLDB_INVALID_THREAD_ID),
       m_finalizing(false), m_destructing(false),
       m_clear_thread_plans_on_stop(false), m_force_next_event_delivery(false),
       m_last_broadcast_state(eStateInvalid), m_destroy_in_process(false),
@@ -845,6 +846,7 @@ bool Process::HandleProcessStateChangedEvent(
             switch (thread_stop_reason) {
             case eStopReasonInvalid:
             case eStopReasonNone:
+            case eStopReasonHistoryBoundary:
               break;
 
             case eStopReasonSignal: {
@@ -1352,7 +1354,7 @@ void Process::SetPublicState(StateType new_state, bool restarted) {
   }
 }
 
-Status Process::Resume() {
+Status Process::Resume(RunDirection direction) {
   Log *log(GetLog(LLDBLog::State | LLDBLog::Process));
   LLDB_LOGF(log, "(plugin = %s) -- locking run lock", GetPluginName().data());
   if (!m_public_run_lock.TrySetRunning()) {
@@ -1361,15 +1363,15 @@ Status Process::Resume() {
     return Status::FromErrorString(
         "Resume request failed - process still running.");
   }
-  Status error = PrivateResume();
+  Status error = PrivateResume(direction);
   if (!error.Success()) {
     // Undo running state change
     m_public_run_lock.SetStopped();
   }
   return error;
 }
 
-Status Process::ResumeSynchronous(Stream *stream) {
+Status Process::ResumeSynchronous(Stream *stream, RunDirection direction) {
   Log *log(GetLog(LLDBLog::State | LLDBLog::Process));
   LLDB_LOGF(log, "Process::ResumeSynchronous -- locking run lock");
   if (!m_public_run_lock.TrySetRunning()) {
@@ -1382,7 +1384,7 @@ Status Process::ResumeSynchronous(Stream *stream) {
       Listener::MakeListener(ResumeSynchronousHijackListenerName.data()));
   HijackProcessEvents(listener_sp);
 
-  Status error = PrivateResume();
+  Status error = PrivateResume(direction);
   if (error.Success()) {
     StateType state =
         WaitForProcessToStop(std::nullopt, nullptr, true, listener_sp, stream,
@@ -3239,7 +3241,7 @@ Status Process::ConnectRemote(llvm::StringRef remote_url) {
   return error;
 }
 
-Status Process::PrivateResume() {
+Status Process::PrivateResume(RunDirection direction) {
   Log *log(GetLog(LLDBLog::Process | LLDBLog::Step));
   LLDB_LOGF(log,
             "Process::PrivateResume() m_stop_id = %u, public state: %s "
@@ -3255,6 +3257,15 @@ Status Process::PrivateResume() {
   if (!GetModID().IsLastResumeForUserExpression())
     ResetExtendedCrashInfoDict();
 
+  if (m_last_run_direction != direction) {
+    // In the future we might want to support mixed-direction plans,
+    // e.g. a forward step-over stops at a breakpoint, the user does
+    // a reverse-step, then disables the breakpoint and continues forward.
+    // This code will need to be changed to support that.
+    m_thread_list.DiscardThreadPlans();
+    m_last_run_direction = direction;
+  }
+
   Status error(WillResume());
   // Tell the process it is about to resume before the thread list
   if (error.Success()) {
@@ -3272,7 +3283,7 @@ Status Process::PrivateResume() {
             "Process::PrivateResume PreResumeActions failed, not resuming.");
       } else {
         m_mod_id.BumpResumeID();
-        error = DoResume();
+        error = DoResume(direction);
         if (error.Success()) {
           DidResume();
           m_thread_list.DidResume();
@@ -3735,7 +3746,7 @@ bool Process::ShouldBroadcastEvent(Event *event_ptr) {
                     "from state: %s",
                     static_cast<void *>(event_ptr), StateAsCString(state));
           ProcessEventData::SetRestartedInEvent(event_ptr, true);
-          PrivateResume();
+          PrivateResume(m_last_run_direction);
         }
       } else {
         return_value = true;
@@ -4346,7 +4357,7 @@ void Process::ProcessEventData::DoOnRemoval(Event *event_ptr) {
     SetRestarted(true);
     // Use the private resume method here, since we aren't changing the run
     // lock state.
-    process_sp->PrivateResume();
+    process_sp->PrivateResume(process_sp->m_last_run_direction);
   } else {
     bool hijacked = process_sp->IsHijackedForEvent(eBroadcastBitStateChanged) &&
                     !process_sp->StateChangedIsHijackedForSynchronousResume();

diff --git a/lldb/source/Target/StopInfo.cpp b/lldb/source/Target/StopInfo.cpp
@@ -1212,6 +1212,30 @@ class StopInfoProcessorTrace : public StopInfo {
   }
 };
 
+// StopInfoHistoryBoundary
+
+class StopInfoHistoryBoundary : public StopInfo {
+public:
+  StopInfoHistoryBoundary(Thread &thread, const char *description)
+      : StopInfo(thread, LLDB_INVALID_UID) {
+    if (description)
+      SetDescription(description);
+  }
+
+  ~StopInfoHistoryBoundary() override = default;
+
+  StopReason GetStopReason() const override {
+    return eStopReasonHistoryBoundary;
+  }
+
+  const char *GetDescription() override {
+    if (m_description.empty())
+      return "history boundary";
+    else
+      return m_description.c_str();
+  }
+};
+
 // StopInfoThreadPlan
 
 class StopInfoThreadPlan : public StopInfo {
@@ -1439,6 +1463,11 @@ StopInfoSP StopInfo::CreateStopReasonProcessorTrace(Thread &thread,
   return StopInfoSP(new StopInfoProcessorTrace(thread, description));
 }
 
+StopInfoSP StopInfo::CreateStopReasonHistoryBoundary(Thread &thread,
+                                                     const char *description) {
+  return StopInfoSP(new StopInfoHistoryBoundary(thread, description));
+}
+
 StopInfoSP StopInfo::CreateStopReasonWithExec(Thread &thread) {
   return StopInfoSP(new StopInfoExec(thread));
 }

diff --git a/lldb/source/Target/Thread.cpp b/lldb/source/Target/Thread.cpp
@@ -624,10 +624,12 @@ void Thread::SetupForResume() {
     // what the current plan is.
 
     lldb::RegisterContextSP reg_ctx_sp(GetRegisterContext());
-    if (reg_ctx_sp) {
+    ProcessSP process_sp(GetProcess());
+    if (reg_ctx_sp && process_sp &&
+        process_sp->GetLastRunDirection() == eRunForward) {
       const addr_t thread_pc = reg_ctx_sp->GetPC();
       BreakpointSiteSP bp_site_sp =
-          GetProcess()->GetBreakpointSiteList().FindByAddress(thread_pc);
+          process_sp->GetBreakpointSiteList().FindByAddress(thread_pc);
       if (bp_site_sp) {
         // Note, don't assume there's a ThreadPlanStepOverBreakpoint, the
         // target may not require anything special to step over a breakpoint.
@@ -1732,6 +1734,8 @@ std::string Thread::StopReasonAsString(lldb::StopReason reason) {
     return "processor trace";
   case eStopReasonInterrupt:
     return "async interrupt";
+  case eStopReasonHistoryBoundary:
+    return "history boundary";
   }
 
   return "StopReason = " + std::to_string(reason);

diff --git a/lldb/test/API/functionalities/reverse-execution/Makefile b/lldb/test/API/functionalities/reverse-execution/Makefile
@@ -0,0 +1,3 @@
+C_SOURCES := main.c
+
+include Makefile.rules
diff --git a/lldb/test/API/functionalities/reverse-execution/TestReverseContinueBreakpoints.py b/lldb/test/API/functionalities/reverse-execution/TestReverseContinueBreakpoints.py
@@ -0,0 +1,115 @@
+import lldb
+import time
+import unittest
+from lldbsuite.test.lldbtest import *
+from lldbsuite.test.decorators import *
+from lldbsuite.test.gdbclientutils import *
+from lldbsuite.test.lldbreverse import ReverseTestBase
+from lldbsuite.test import lldbutil
+
+
+class TestReverseContinueBreakpoints(ReverseTestBase):
+    NO_DEBUG_INFO_TESTCASE = True
+
+    def test_reverse_continue(self):
+        self.reverse_continue_internal(async_mode=False)
+
+    def test_reverse_continue_async(self):
+        self.reverse_continue_internal(async_mode=True)
+
+    def reverse_continue_internal(self, async_mode):
+        target, process, initial_threads = self.setup_recording(async_mode)
+
+        # Reverse-continue. We'll stop at the point where we started recording.
+        status = process.Continue(lldb.eRunReverse)
+        self.assertSuccess(status)
+        self.expect_async_state_changes(async_mode, process, [lldb.eStateRunning, lldb.eStateStopped])
+        self.expect(
+            "thread list",
+            STOPPED_DUE_TO_HISTORY_BOUNDARY,
+            substrs=["stopped", "stop reason = history boundary"],
+        )
+
+        # Continue forward normally until the target exits.
+        status = process.Continue()
+        self.expect_async_state_changes(async_mode, process, [lldb.eStateRunning, lldb.eStateExited])
+        self.assertSuccess(status)
+        self.assertState(process.GetState(), lldb.eStateExited)
+        self.assertEqual(process.GetExitStatus(), 0)
+
+    def test_reverse_continue_breakpoint(self):
+        self.reverse_continue_breakpoint_internal(async_mode=False)
+
+    def test_reverse_continue_breakpoint_async(self):
+        self.reverse_continue_breakpoint_internal(async_mode=True)
+
+    def reverse_continue_breakpoint_internal(self, async_mode):
+        target, process, initial_threads = self.setup_recording(async_mode)
+
+        # Reverse-continue to the function "trigger_breakpoint".
+        trigger_bkpt = target.BreakpointCreateByName("trigger_breakpoint", None)
+        status = process.Continue(lldb.eRunReverse)
+        self.assertSuccess(status)
+        self.expect_async_state_changes(async_mode, process, [lldb.eStateRunning, lldb.eStateStopped])
+        threads_now = lldbutil.get_threads_stopped_at_breakpoint(process, trigger_bkpt)
+        self.assertEqual(threads_now, initial_threads)
+
+    def test_reverse_continue_skip_breakpoint(self):
+        self.reverse_continue_skip_breakpoint_internal(async_mode=False)
+
+    def test_reverse_continue_skip_breakpoint_async(self):
+        self.reverse_continue_skip_breakpoint_internal(async_mode=True)
+
+    def reverse_continue_skip_breakpoint_internal(self, async_mode):
+        target, process, initial_threads = self.setup_recording(async_mode)
+
+        # Reverse-continue over a breakpoint at "trigger_breakpoint" whose
+        # condition is false.
+        # This tests that we continue in the correct direction after hitting
+        # the breakpoint.
+        trigger_bkpt = target.BreakpointCreateByName("trigger_breakpoint", None)
+        trigger_bkpt.SetCondition("false_condition")
+        status = process.Continue(lldb.eRunReverse)
+        self.expect_async_state_changes(async_mode, process, [lldb.eStateRunning, lldb.eStateStopped])
+        self.assertSuccess(status)
+        self.expect(
+            "thread list",
+            STOPPED_DUE_TO_HISTORY_BOUNDARY,
+            substrs=["stopped", "stop reason = history boundary"],
+        )
+
+    def setup_recording(self, async_mode):
+        """
+        Record execution of code between "start_recording" and "stop_recording" breakpoints.
+
+        Returns with the target stopped at "stop_recording", with recording disabled,
+        ready to reverse-execute.
+        """
+        self.build()
+        target = self.dbg.CreateTarget("")
+        process = self.connect(target)
+
+        # Record execution from the start of the function "start_recording"
+        # to the start of the function "stop_recording". We want to keep the
+        # interval that we record as small as possible to minimize the run-time
+        # of our single-stepping recorder.
+        start_recording_bkpt = target.BreakpointCreateByName("start_recording", None)
+        initial_threads = lldbutil.continue_to_breakpoint(process, start_recording_bkpt)
+        self.assertEqual(len(initial_threads), 1)
+        target.BreakpointDelete(start_recording_bkpt.GetID())
+        self.start_recording()
+        stop_recording_bkpt = target.BreakpointCreateByName("stop_recording", None)
+        lldbutil.continue_to_breakpoint(process, stop_recording_bkpt)
+        target.BreakpointDelete(stop_recording_bkpt.GetID())
+        self.stop_recording()
+
+        self.dbg.SetAsync(async_mode)
+        self.expect_async_state_changes(async_mode, process, [lldb.eStateStopped])
+
+        return target, process, initial_threads
+
+    def expect_async_state_changes(self, async_mode, process, states):
+        if not async_mode:
+            return
+        listener = self.dbg.GetListener()
+        lldbutil.expect_state_changes(self, listener, process, states)
diff --git a/lldb/test/API/functionalities/reverse-execution/TestReverseContinueNotSupported.py b/lldb/test/API/functionalities/reverse-execution/TestReverseContinueNotSupported.py
@@ -0,0 +1,30 @@
+import lldb
+import unittest
+from lldbsuite.test.lldbtest import *
+from lldbsuite.test.decorators import *
+from lldbsuite.test import lldbutil
+
+
+class TestReverseContinueNotSupported(TestBase):
+    NO_DEBUG_INFO_TESTCASE = True
+
+    def test_reverse_continue_not_supported(self):
+        self.build()
+        exe = self.getBuildArtifact("a.out")
+        target = self.dbg.CreateTarget(exe)
+        self.assertTrue(target, VALID_TARGET)
+
+        main_bkpt = target.BreakpointCreateByName("main", None)
+        self.assertTrue(main_bkpt, VALID_BREAKPOINT)
+
+        process = target.LaunchSimple(None, None, self.get_process_working_directory())
+        self.assertTrue(process, PROCESS_IS_VALID)
+
+        # This will fail gracefully.
+        status = process.Continue(lldb.eRunReverse)
+        self.assertFailure(status, "target does not support reverse-continue")
+
+        status = process.Continue()
+        self.assertSuccess(status)
+        self.assertState(process.GetState(), lldb.eStateExited)
+        self.assertEqual(process.GetExitStatus(), 0)
diff --git a/lldb/test/API/functionalities/reverse-execution/main.c b/lldb/test/API/functionalities/reverse-execution/main.c
@@ -0,0 +1,14 @@
+volatile int false_condition = 0;
+
+static void start_recording() {}
+
+static void trigger_breakpoint() {}
+
+static void stop_recording() {}
+
+int main() {
+  start_recording();
+  trigger_breakpoint();
+  stop_recording();
+  return 0;
+}
diff --git a/lldb/tools/lldb-dap/JSONUtils.cpp b/lldb/tools/lldb-dap/JSONUtils.cpp
@@ -1045,6 +1045,9 @@ llvm::json::Value CreateThreadStopped(lldb::SBThread &thread,
   case lldb::eStopReasonProcessorTrace:
     body.try_emplace("reason", "processor trace");
     break;
+  case lldb::eStopReasonHistoryBoundary:
+    body.try_emplace("reason", "history boundary");
+    break;
   case lldb::eStopReasonSignal:
   case lldb::eStopReasonException:
     body.try_emplace("reason", "exception");

diff --git a/lldb/tools/lldb-dap/LLDBUtils.cpp b/lldb/tools/lldb-dap/LLDBUtils.cpp
@@ -111,6 +111,7 @@ bool ThreadHasStopReason(lldb::SBThread &thread) {
   case lldb::eStopReasonVFork:
   case lldb::eStopReasonVForkDone:
   case lldb::eStopReasonInterrupt:
+  case lldb::eStopReasonHistoryBoundary:
     return true;
   case lldb::eStopReasonThreadExiting:
   case lldb::eStopReasonInvalid:

diff --git a/llvm/docs/Coroutines.rst b/llvm/docs/Coroutines.rst
@@ -312,6 +312,7 @@ lowered to a constant representing the size required for the coroutine frame.
 The `coro.begin`_ intrinsic initializes the coroutine frame and returns the
 coroutine handle. The second parameter of `coro.begin` is given a block of memory
 to be used if the coroutine frame needs to be allocated dynamically.
+
 The `coro.id`_ intrinsic serves as coroutine identity useful in cases when the
 `coro.begin`_ intrinsic get duplicated by optimization passes such as
 jump-threading.
@@ -749,6 +750,65 @@ and python iterator `__next__` would look like:
     return *(int*)coro.promise(hdl, 4, false);
   }
 
+Custom ABIs and Plugin Libraries
+--------------------------------
+
+Plugin libraries can extend coroutine lowering enabling a wide variety of users
+to utilize the coroutine transformation passes. An existing coroutine lowering
+is extended by:
+
+#. defining custom ABIs that inherit from the existing ABIs,
+#. give a list of generators for the custom ABIs when constructing the `CoroSplit`_ pass, and
+#. use `coro.begin.custom.abi`_ in place of `coro.begin`_ that has an additional parameter for the index of the generator/ABI to be used for the coroutine.
+
+A custom ABI overriding the SwitchABI's materialization looks like:
+
+.. code-block:: c++
+
+  class CustomSwitchABI : public coro::SwitchABI {
+  public:
+    CustomSwitchABI(Function &F, coro::Shape &S)
+      : coro::SwitchABI(F, S, ExtraMaterializable) {}
+  };
+
+Giving a list of custom ABI generators while constructing the `CoroSplit`
+pass looks like:
+
+.. code-block:: c++
+
+  CoroSplitPass::BaseABITy GenCustomABI = [](Function &F, coro::Shape &S) {
+    return std::make_unique<CustomSwitchABI>(F, S);
+  };
+
+  CGSCCPassManager CGPM;
+  CGPM.addPass(CoroSplitPass({GenCustomABI}));
+
+The LLVM IR for a coroutine using a Coroutine with a custom ABI looks like:
+
+.. code-block:: llvm
+
+  define ptr @f(i32 %n) presplitcoroutine_custom_abi {
+  entry:
+    %id = call token @llvm.coro.id(i32 0, ptr null, ptr null, ptr null)
+    %size = call i32 @llvm.coro.size.i32()
+    %alloc = call ptr @malloc(i32 %size)
+    %hdl = call noalias ptr @llvm.coro.begin.custom.abi(token %id, ptr %alloc, i32 0)
+    br label %loop
+  loop:
+    %n.val = phi i32 [ %n, %entry ], [ %inc, %loop ]
+    %inc = add nsw i32 %n.val, 1
+    call void @print(i32 %n.val)
+    %0 = call i8 @llvm.coro.suspend(token none, i1 false)
+    switch i8 %0, label %suspend [i8 0, label %loop
+                                  i8 1, label %cleanup]
+  cleanup:
+    %mem = call ptr @llvm.coro.free(token %id, ptr %hdl)
+    call void @free(ptr %mem)
+    br label %suspend
+  suspend:
+    %unused = call i1 @llvm.coro.end(ptr %hdl, i1 false, token none)
+    ret ptr %hdl
+  }
 
 Intrinsics
 ==========
@@ -1007,6 +1067,36 @@ with small positive and negative offsets).
 
 A frontend should emit exactly one `coro.begin` intrinsic per coroutine.
 
+.. _coro.begin.custom.abi:
+
+'llvm.coro.begin.custom.abi' Intrinsic
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+::
+
+  declare ptr @llvm.coro.begin.custom.abi(token <id>, ptr <mem>, i32)
+
+Overview:
+"""""""""
+
+The '``llvm.coro.begin.custom.abi``' intrinsic is used in place of the
+`coro.begin` intrinsic that has an additional parameter to specify the custom
+ABI for the coroutine. The return is identical to that of the `coro.begin`
+intrinsic.
+
+Arguments:
+""""""""""
+
+The first and second arguments are identical to those of the `coro.begin`
+intrinsic.
+
+The third argument is an i32 index of the generator list given to the
+`CoroSplit` pass specifying the custom ABI generator lor this coroutine.
+
+Semantics:
+""""""""""
+
+The semantics are identical to those of the `coro.begin` intrinsic.
+
 .. _coro.free:
 
 'llvm.coro.free' Intrinsic

diff --git a/llvm/include/llvm/Transforms/Vectorize/SandboxVectorizer/DependencyGraph.h b/llvm/include/llvm/Transforms/Vectorize/SandboxVectorizer/DependencyGraph.h
@@ -1,4 +1,4 @@
-//===- DependencyGraph.h ----------------------------------*- C++ -*-===//
+//===- DependencyGraph.h ----------------------------------------*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
@@ -96,9 +96,6 @@ class DGNode {
   // TODO: Use a PointerIntPair for SubclassID and I.
   /// For isa/dyn_cast etc.
   DGNodeID SubclassID;
-  // TODO: Move MemPreds to MemDGNode.
-  /// Memory predecessors.
-  DenseSet<MemDGNode *> MemPreds;
 
   DGNode(Instruction *I, DGNodeID ID) : I(I), SubclassID(ID) {}
   friend class MemDGNode; // For constructor.
@@ -170,17 +167,6 @@ class DGNode {
   }
 
   Instruction *getInstruction() const { return I; }
-  void addMemPred(MemDGNode *PredN) { MemPreds.insert(PredN); }
-  /// \Returns all memory dependency predecessors.
-  iterator_range<DenseSet<MemDGNode *>::const_iterator> memPreds() const {
-    return make_range(MemPreds.begin(), MemPreds.end());
-  }
-  /// \Returns true if there is a memory dependency N->this.
-  bool hasMemPred(DGNode *N) const {
-    if (auto *MN = dyn_cast<MemDGNode>(N))
-      return MemPreds.count(MN);
-    return false;
-  }
 
 #ifndef NDEBUG
   virtual void print(raw_ostream &OS, bool PrintDeps = true) const;
@@ -198,6 +184,9 @@ class DGNode {
 class MemDGNode final : public DGNode {
   MemDGNode *PrevMemN = nullptr;
   MemDGNode *NextMemN = nullptr;
+  /// Memory predecessors.
+  DenseSet<MemDGNode *> MemPreds;
+  friend class PredIterator; // For MemPreds.
 
   void setNextNode(MemDGNode *N) { NextMemN = N; }
   void setPrevNode(MemDGNode *N) { PrevMemN = N; }
@@ -222,6 +211,21 @@ class MemDGNode final : public DGNode {
   MemDGNode *getPrevNode() const { return PrevMemN; }
   /// \Returns the next Mem DGNode in instruction order.
   MemDGNode *getNextNode() const { return NextMemN; }
+  /// Adds the mem dependency edge PredN->this.
+  void addMemPred(MemDGNode *PredN) { MemPreds.insert(PredN); }
+  /// \Returns true if there is a memory dependency N->this.
+  bool hasMemPred(DGNode *N) const {
+    if (auto *MN = dyn_cast<MemDGNode>(N))
+      return MemPreds.count(MN);
+    return false;
+  }
+  /// \Returns all memory dependency predecessors. Used by tests.
+  iterator_range<DenseSet<MemDGNode *>::const_iterator> memPreds() const {
+    return make_range(MemPreds.begin(), MemPreds.end());
+  }
+#ifndef NDEBUG
+  virtual void print(raw_ostream &OS, bool PrintDeps = true) const override;
+#endif // NDEBUG
 };
 
 /// Convenience builders for a MemDGNode interval.
@@ -266,7 +270,7 @@ class DependencyGraph {
 
   /// Go through all mem nodes in \p SrcScanRange and try to add dependencies to
   /// \p DstN.
-  void scanAndAddDeps(DGNode &DstN, const Interval<MemDGNode> &SrcScanRange);
+  void scanAndAddDeps(MemDGNode &DstN, const Interval<MemDGNode> &SrcScanRange);
 
 public:
   DependencyGraph(AAResults &AA)

diff --git a/llvm/lib/Target/NVPTX/NVPTXInstrInfo.td b/llvm/lib/Target/NVPTX/NVPTXInstrInfo.td
@@ -139,6 +139,8 @@ def hasVote : Predicate<"Subtarget->hasVote()">;
 def hasDouble : Predicate<"Subtarget->hasDouble()">;
 def hasLDG : Predicate<"Subtarget->hasLDG()">;
 def hasLDU : Predicate<"Subtarget->hasLDU()">;
+def hasPTXASUnreachableBug : Predicate<"Subtarget->hasPTXASUnreachableBug()">;
+def noPTXASUnreachableBug : Predicate<"!Subtarget->hasPTXASUnreachableBug()">;
 
 def doF32FTZ : Predicate<"useF32FTZ()">;
 def doNoF32FTZ : Predicate<"!useF32FTZ()">;
@@ -3736,9 +3738,10 @@ def Callseq_End :
             [(callseq_end timm:$amt1, timm:$amt2)]>;
 
 // trap instruction
+def trapinst : NVPTXInst<(outs), (ins), "trap;", [(trap)]>, Requires<[noPTXASUnreachableBug]>;
 // Emit an `exit` as well to convey to ptxas that `trap` exits the CFG.
 // This won't be necessary in a future version of ptxas.
-def trapinst : NVPTXInst<(outs), (ins), "trap; exit;", [(trap)]>;
+def trapexitinst : NVPTXInst<(outs), (ins), "trap; exit;", [(trap)]>, Requires<[hasPTXASUnreachableBug]>;
 // brkpt instruction
 def debugtrapinst : NVPTXInst<(outs), (ins), "brkpt;", [(debugtrap)]>;
 

diff --git a/llvm/lib/Target/NVPTX/NVPTXSubtarget.h b/llvm/lib/Target/NVPTX/NVPTXSubtarget.h
@@ -95,6 +95,14 @@ class NVPTXSubtarget : public NVPTXGenSubtargetInfo {
   bool hasDotInstructions() const {
     return SmVersion >= 61 && PTXVersion >= 50;
   }
+  // Prior to CUDA 12.3 ptxas did not recognize that the trap instruction
+  // terminates a basic block. Instead, it would assume that control flow
+  // continued to the next instruction. The next instruction could be in the
+  // block that's lexically below it. This would lead to a phantom CFG edges
+  // being created within ptxas. This issue was fixed in CUDA 12.3. Thus, when
+  // PTX ISA versions 8.3+ we can confidently say that the bug will not be
+  // present.
+  bool hasPTXASUnreachableBug() const { return PTXVersion < 83; }
   bool hasCvtaParam() const { return SmVersion >= 70 && PTXVersion >= 77; }
   unsigned int getFullSmVersion() const { return FullSmVersion; }
   unsigned int getSmVersion() const { return getFullSmVersion() / 10; }

diff --git a/llvm/lib/Target/NVPTX/NVPTXTargetMachine.cpp b/llvm/lib/Target/NVPTX/NVPTXTargetMachine.cpp
@@ -367,9 +367,13 @@ void NVPTXPassConfig::addIRPasses() {
     addPass(createSROAPass());
   }
 
-  const auto &Options = getNVPTXTargetMachine().Options;
-  addPass(createNVPTXLowerUnreachablePass(Options.TrapUnreachable,
-                                          Options.NoTrapAfterNoreturn));
+  if (ST.hasPTXASUnreachableBug()) {
+    // Run LowerUnreachable to WAR a ptxas bug. See the commit description of
+    // 1ee4d880e8760256c606fe55b7af85a4f70d006d for more details.
+    const auto &Options = getNVPTXTargetMachine().Options;
+    addPass(createNVPTXLowerUnreachablePass(Options.TrapUnreachable,
+                                            Options.NoTrapAfterNoreturn));
+  }
 }
 
 bool NVPTXPassConfig::addInstSelector() {

diff --git a/llvm/lib/Transforms/Instrumentation/PGOCtxProfLowering.cpp b/llvm/lib/Transforms/Instrumentation/PGOCtxProfLowering.cpp
@@ -154,29 +154,28 @@ CtxInstrumentationLowerer::CtxInstrumentationLowerer(Module &M,
   StartCtx = cast<Function>(
       M.getOrInsertFunction(
            CompilerRtAPINames::StartCtx,
-           FunctionType::get(ContextNodeTy->getPointerTo(),
-                             {ContextRootTy->getPointerTo(), /*ContextRoot*/
+           FunctionType::get(PointerTy,
+                             {PointerTy, /*ContextRoot*/
                               I64Ty, /*Guid*/ I32Ty,
                               /*NumCounters*/ I32Ty /*NumCallsites*/},
                              false))
           .getCallee());
   GetCtx = cast<Function>(
       M.getOrInsertFunction(CompilerRtAPINames::GetCtx,
-                            FunctionType::get(ContextNodeTy->getPointerTo(),
+                            FunctionType::get(PointerTy,
                                               {PointerTy, /*Callee*/
                                                I64Ty,     /*Guid*/
                                                I32Ty,     /*NumCounters*/
                                                I32Ty},    /*NumCallsites*/
                                               false))
           .getCallee());
   ReleaseCtx = cast<Function>(
-      M.getOrInsertFunction(
-           CompilerRtAPINames::ReleaseCtx,
-           FunctionType::get(Type::getVoidTy(M.getContext()),
-                             {
-                                 ContextRootTy->getPointerTo(), /*ContextRoot*/
-                             },
-                             false))
+      M.getOrInsertFunction(CompilerRtAPINames::ReleaseCtx,
+                            FunctionType::get(Type::getVoidTy(M.getContext()),
+                                              {
+                                                  PointerTy, /*ContextRoot*/
+                                              },
+                                              false))
           .getCallee());
 
   // Declare the TLSes we will need to use.
@@ -264,7 +263,7 @@ bool CtxInstrumentationLowerer::lowerFunction(Function &F) {
         auto *Index = Builder.CreateAnd(CtxAsInt, Builder.getInt64(1));
         // The GEPs corresponding to that index, in the respective TLS.
         ExpectedCalleeTLSAddr = Builder.CreateGEP(
-            Builder.getInt8Ty()->getPointerTo(),
+            PointerType::getUnqual(F.getContext()),
             Builder.CreateThreadLocalAddress(ExpectedCalleeTLS), {Index});
         CallsiteInfoTLSAddr = Builder.CreateGEP(
             Builder.getInt32Ty(),
@@ -277,7 +276,7 @@ bool CtxInstrumentationLowerer::lowerFunction(Function &F) {
       // with counters) stays the same.
       RealContext = Builder.CreateIntToPtr(
           Builder.CreateAnd(CtxAsInt, Builder.getInt64(-2)),
-          ThisContextType->getPointerTo());
+          PointerType::getUnqual(F.getContext()));
       I.eraseFromParent();
       break;
     }

diff --git a/llvm/lib/Transforms/Vectorize/SandboxVectorizer/DependencyGraph.cpp b/llvm/lib/Transforms/Vectorize/SandboxVectorizer/DependencyGraph.cpp
@@ -23,7 +23,8 @@ PredIterator::value_type PredIterator::operator*() {
   // or a mem predecessor.
   if (OpIt != OpItE)
     return DAG->getNode(cast<Instruction>((Value *)*OpIt));
-  assert(MemIt != cast<MemDGNode>(N)->memPreds().end() &&
+  // It's a MemDGNode with OpIt == end, so we need to use MemIt.
+  assert(MemIt != cast<MemDGNode>(N)->MemPreds.end() &&
          "Cant' dereference end iterator!");
   return *MemIt;
 }
@@ -45,7 +46,8 @@ PredIterator &PredIterator::operator++() {
     OpIt = skipNonInstr(OpIt, OpItE);
     return *this;
   }
-  assert(MemIt != cast<MemDGNode>(N)->memPreds().end() && "Already at end!");
+  // It's a MemDGNode with OpIt == end, so we need to increment MemIt.
+  assert(MemIt != cast<MemDGNode>(N)->MemPreds.end() && "Already at end!");
   ++MemIt;
   return *this;
 }
@@ -57,10 +59,14 @@ bool PredIterator::operator==(const PredIterator &Other) const {
 }
 
 #ifndef NDEBUG
-void DGNode::print(raw_ostream &OS, bool PrintDeps) const {
+void DGNode::print(raw_ostream &OS, bool PrintDeps) const { I->dumpOS(OS); }
+void DGNode::dump() const {
+  print(dbgs());
+  dbgs() << "\n";
+}
+void MemDGNode::print(raw_ostream &OS, bool PrintDeps) const {
   I->dumpOS(OS);
   if (PrintDeps) {
-    OS << "\n";
     // Print memory preds.
     static constexpr const unsigned Indent = 4;
     for (auto *Pred : MemPreds) {
@@ -70,10 +76,6 @@ void DGNode::print(raw_ostream &OS, bool PrintDeps) const {
     }
   }
 }
-void DGNode::dump() const {
-  print(dbgs());
-  dbgs() << "\n";
-}
 #endif // NDEBUG
 
 Interval<MemDGNode>
@@ -179,7 +181,7 @@ bool DependencyGraph::hasDep(Instruction *SrcI, Instruction *DstI) {
   llvm_unreachable("Unknown DependencyType enum");
 }
 
-void DependencyGraph::scanAndAddDeps(DGNode &DstN,
+void DependencyGraph::scanAndAddDeps(MemDGNode &DstN,
                                      const Interval<MemDGNode> &SrcScanRange) {
   assert(isa<MemDGNode>(DstN) &&
          "DstN is the mem dep destination, so it must be mem");

diff --git a/llvm/test/CodeGen/NVPTX/unreachable.ll b/llvm/test/CodeGen/NVPTX/unreachable.ll
@@ -1,48 +1,107 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
 ; RUN: llc < %s -march=nvptx -mcpu=sm_20 -verify-machineinstrs -trap-unreachable=false \
-; RUN:     | FileCheck %s  --check-prefix=CHECK --check-prefix=CHECK-NOTRAP
+; RUN:     | FileCheck %s --check-prefixes=CHECK,NO-TRAP-UNREACHABLE
 ; RUN: llc < %s -march=nvptx64 -mcpu=sm_20 -verify-machineinstrs -trap-unreachable=false \
-; RUN:     | FileCheck %s  --check-prefix=CHECK --check-prefix=CHECK-NOTRAP
+; RUN:     | FileCheck %s --check-prefixes=CHECK,NO-TRAP-UNREACHABLE
 ; RUN: llc < %s -march=nvptx -mcpu=sm_20 -verify-machineinstrs -trap-unreachable -no-trap-after-noreturn \
-; RUN:     | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-NOTRAP
+; RUN:     | FileCheck %s --check-prefixes=CHECK,NO-TRAP-AFTER-NORETURN
 ; RUN: llc < %s -march=nvptx64 -mcpu=sm_20 -verify-machineinstrs -trap-unreachable -no-trap-after-noreturn \
-; RUN:     | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-NOTRAP
+; RUN:     | FileCheck %s --check-prefixes=CHECK,NO-TRAP-AFTER-NORETURN
 ; RUN: llc < %s -march=nvptx -mcpu=sm_20 -verify-machineinstrs -trap-unreachable -no-trap-after-noreturn=false \
-; RUN:     | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-TRAP
+; RUN:     | FileCheck %s --check-prefixes=CHECK,TRAP
 ; RUN: llc < %s -march=nvptx64 -mcpu=sm_20 -verify-machineinstrs -trap-unreachable -no-trap-after-noreturn=false \
-; RUN:     | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-TRAP
+; RUN:     | FileCheck %s --check-prefixes=CHECK,TRAP
+; RUN: llc < %s -march=nvptx64 -mcpu=sm_20 -verify-machineinstrs -trap-unreachable -mattr=+ptx83 \
+; RUN:     | FileCheck %s --check-prefixes=BUG-FIXED
 ; RUN: %if ptxas && !ptxas-12.0 %{ llc < %s -march=nvptx -mcpu=sm_20 -verify-machineinstrs | %ptxas-verify %}
 ; RUN: %if ptxas %{ llc < %s -march=nvptx64 -mcpu=sm_20 -verify-machineinstrs | %ptxas-verify %}
 
-; CHECK: .extern .func throw
+target triple = "nvptx-unknown-cuda"
+
 declare void @throw() #0
 declare void @llvm.trap() #0
 
-; CHECK-LABEL: .entry kernel_func
 define void @kernel_func() {
-; CHECK: call.uni
-; CHECK: throw,
+; NO-TRAP-UNREACHABLE-LABEL: kernel_func(
+; NO-TRAP-UNREACHABLE:       {
+; NO-TRAP-UNREACHABLE-EMPTY:
+; NO-TRAP-UNREACHABLE-EMPTY:
+; NO-TRAP-UNREACHABLE-NEXT:  // %bb.0:
+; NO-TRAP-UNREACHABLE-NEXT:    { // callseq 0, 0
+; NO-TRAP-UNREACHABLE-NEXT:    call.uni
+; NO-TRAP-UNREACHABLE-NEXT:    throw,
+; NO-TRAP-UNREACHABLE-NEXT:    (
+; NO-TRAP-UNREACHABLE-NEXT:    );
+; NO-TRAP-UNREACHABLE-NEXT:    } // callseq 0
+; NO-TRAP-UNREACHABLE-NEXT:    // begin inline asm
+; NO-TRAP-UNREACHABLE-NEXT:    exit;
+; NO-TRAP-UNREACHABLE-NEXT:    // end inline asm
+;
+; NO-TRAP-AFTER-NORETURN-LABEL: kernel_func(
+; NO-TRAP-AFTER-NORETURN:       {
+; NO-TRAP-AFTER-NORETURN-EMPTY:
+; NO-TRAP-AFTER-NORETURN-EMPTY:
+; NO-TRAP-AFTER-NORETURN-NEXT:  // %bb.0:
+; NO-TRAP-AFTER-NORETURN-NEXT:    { // callseq 0, 0
+; NO-TRAP-AFTER-NORETURN-NEXT:    call.uni
+; NO-TRAP-AFTER-NORETURN-NEXT:    throw,
+; NO-TRAP-AFTER-NORETURN-NEXT:    (
+; NO-TRAP-AFTER-NORETURN-NEXT:    );
+; NO-TRAP-AFTER-NORETURN-NEXT:    } // callseq 0
+; NO-TRAP-AFTER-NORETURN-NEXT:    // begin inline asm
+; NO-TRAP-AFTER-NORETURN-NEXT:    exit;
+; NO-TRAP-AFTER-NORETURN-NEXT:    // end inline asm
+; NO-TRAP-AFTER-NORETURN-NEXT:    trap; exit;
+;
+; TRAP-LABEL: kernel_func(
+; TRAP:       {
+; TRAP-EMPTY:
+; TRAP-EMPTY:
+; TRAP-NEXT:  // %bb.0:
+; TRAP-NEXT:    { // callseq 0, 0
+; TRAP-NEXT:    call.uni
+; TRAP-NEXT:    throw,
+; TRAP-NEXT:    (
+; TRAP-NEXT:    );
+; TRAP-NEXT:    } // callseq 0
+; TRAP-NEXT:    trap; exit;
+;
+; BUG-FIXED-LABEL: kernel_func(
+; BUG-FIXED:       {
+; BUG-FIXED-EMPTY:
+; BUG-FIXED-EMPTY:
+; BUG-FIXED-NEXT:  // %bb.0:
+; BUG-FIXED-NEXT:    { // callseq 0, 0
+; BUG-FIXED-NEXT:    call.uni
+; BUG-FIXED-NEXT:    throw,
+; BUG-FIXED-NEXT:    (
+; BUG-FIXED-NEXT:    );
+; BUG-FIXED-NEXT:    } // callseq 0
+; BUG-FIXED-NEXT:    trap;
   call void @throw()
-; CHECK-TRAP-NOT: exit;
-; CHECK-TRAP: trap;
-; CHECK-NOTRAP-NOT: trap;
-; CHECK: exit;
   unreachable
 }
 
-; CHECK-LABEL: kernel_func_2
 define void @kernel_func_2() {
-; CHECK: trap; exit;
+; CHECK-LABEL: kernel_func_2(
+; CHECK:       {
+; CHECK-EMPTY:
+; CHECK-EMPTY:
+; CHECK-NEXT:  // %bb.0:
+; CHECK-NEXT:    trap; exit;
+;
+; BUG-FIXED-LABEL: kernel_func_2(
+; BUG-FIXED:       {
+; BUG-FIXED-EMPTY:
+; BUG-FIXED-EMPTY:
+; BUG-FIXED-NEXT:  // %bb.0:
+; BUG-FIXED-NEXT:    trap;
   call void @llvm.trap()
-
-;; Make sure we avoid emitting two trap instructions.
-; CHECK-NOT: trap;
-; CHECK-NOT: exit;
+; Make sure we avoid emitting two trap instructions.
   unreachable
 }
 
 attributes #0 = { noreturn }
 
-
 !nvvm.annotations = !{!1}
-
 !1 = !{ptr @kernel_func, !"kernel", i32 1}
diff --git a/llvm/unittests/Transforms/Vectorize/SandboxVectorizer/DependencyGraphTest.cpp b/llvm/unittests/Transforms/Vectorize/SandboxVectorizer/DependencyGraphTest.cpp
@@ -50,10 +50,10 @@ struct DependencyGraphTest : public testing::Test {
     return *AA;
   }
   /// \Returns true if there is a dependency: SrcN->DstN.
-  bool dependency(sandboxir::DGNode *SrcN, sandboxir::DGNode *DstN) {
-    const auto &Preds = DstN->memPreds();
-    auto It = find(Preds, SrcN);
-    return It != Preds.end();
+  bool memDependency(sandboxir::DGNode *SrcN, sandboxir::DGNode *DstN) {
+    if (auto *MemDstN = dyn_cast<sandboxir::MemDGNode>(DstN))
+      return MemDstN->hasMemPred(SrcN);
+    return false;
   }
 };
 
@@ -230,9 +230,10 @@ define void @foo(ptr %ptr, i8 %v0, i8 %v1) {
   EXPECT_EQ(Span.top(), &*BB->begin());
   EXPECT_EQ(Span.bottom(), BB->getTerminator());
 
-  sandboxir::DGNode *N0 = DAG.getNode(S0);
-  sandboxir::DGNode *N1 = DAG.getNode(S1);
-  sandboxir::DGNode *N2 = DAG.getNode(Ret);
+  auto *N0 = cast<sandboxir::MemDGNode>(DAG.getNode(S0));
+  auto *N1 = cast<sandboxir::MemDGNode>(DAG.getNode(S1));
+  auto *N2 = DAG.getNode(Ret);
+
   // Check getInstruction().
   EXPECT_EQ(N0->getInstruction(), S0);
   EXPECT_EQ(N1->getInstruction(), S1);
@@ -247,7 +248,7 @@ define void @foo(ptr %ptr, i8 %v0, i8 %v1) {
   // Check memPreds().
   EXPECT_TRUE(N0->memPreds().empty());
   EXPECT_THAT(N1->memPreds(), testing::ElementsAre(N0));
-  EXPECT_TRUE(N2->memPreds().empty());
+  EXPECT_TRUE(N2->preds(DAG).empty());
 }
 
 TEST_F(DependencyGraphTest, Preds) {
@@ -399,12 +400,14 @@ define void @foo(ptr %ptr, i8 %v0, i8 %v1) {
   sandboxir::DependencyGraph DAG(getAA(*LLVMF));
   DAG.extend({&*BB->begin(), BB->getTerminator()});
   auto It = BB->begin();
-  auto *Store0N = DAG.getNode(cast<sandboxir::StoreInst>(&*It++));
-  auto *Store1N = DAG.getNode(cast<sandboxir::StoreInst>(&*It++));
+  auto *Store0N = cast<sandboxir::MemDGNode>(
+      DAG.getNode(cast<sandboxir::StoreInst>(&*It++)));
+  auto *Store1N = cast<sandboxir::MemDGNode>(
+      DAG.getNode(cast<sandboxir::StoreInst>(&*It++)));
   auto *RetN = DAG.getNode(cast<sandboxir::ReturnInst>(&*It++));
   EXPECT_TRUE(Store0N->memPreds().empty());
   EXPECT_THAT(Store1N->memPreds(), testing::ElementsAre(Store0N));
-  EXPECT_TRUE(RetN->memPreds().empty());
+  EXPECT_TRUE(RetN->preds(DAG).empty());
 }
 
 TEST_F(DependencyGraphTest, NonAliasingStores) {
@@ -422,13 +425,15 @@ define void @foo(ptr noalias %ptr0, ptr noalias %ptr1, i8 %v0, i8 %v1) {
   sandboxir::DependencyGraph DAG(getAA(*LLVMF));
   DAG.extend({&*BB->begin(), BB->getTerminator()});
   auto It = BB->begin();
-  auto *Store0N = DAG.getNode(cast<sandboxir::StoreInst>(&*It++));
-  auto *Store1N = DAG.getNode(cast<sandboxir::StoreInst>(&*It++));
+  auto *Store0N = cast<sandboxir::MemDGNode>(
+      DAG.getNode(cast<sandboxir::StoreInst>(&*It++)));
+  auto *Store1N = cast<sandboxir::MemDGNode>(
+      DAG.getNode(cast<sandboxir::StoreInst>(&*It++)));
   auto *RetN = DAG.getNode(cast<sandboxir::ReturnInst>(&*It++));
   // We expect no dependencies because the stores don't alias.
   EXPECT_TRUE(Store0N->memPreds().empty());
   EXPECT_TRUE(Store1N->memPreds().empty());
-  EXPECT_TRUE(RetN->memPreds().empty());
+  EXPECT_TRUE(RetN->preds(DAG).empty());
 }
 
 TEST_F(DependencyGraphTest, VolatileLoads) {
@@ -446,12 +451,14 @@ define void @foo(ptr noalias %ptr0, ptr noalias %ptr1) {
   sandboxir::DependencyGraph DAG(getAA(*LLVMF));
   DAG.extend({&*BB->begin(), BB->getTerminator()});
   auto It = BB->begin();
-  auto *Ld0N = DAG.getNode(cast<sandboxir::LoadInst>(&*It++));
-  auto *Ld1N = DAG.getNode(cast<sandboxir::LoadInst>(&*It++));
+  auto *Ld0N = cast<sandboxir::MemDGNode>(
+      DAG.getNode(cast<sandboxir::LoadInst>(&*It++)));
+  auto *Ld1N = cast<sandboxir::MemDGNode>(
+      DAG.getNode(cast<sandboxir::LoadInst>(&*It++)));
   auto *RetN = DAG.getNode(cast<sandboxir::ReturnInst>(&*It++));
   EXPECT_TRUE(Ld0N->memPreds().empty());
   EXPECT_THAT(Ld1N->memPreds(), testing::ElementsAre(Ld0N));
-  EXPECT_TRUE(RetN->memPreds().empty());
+  EXPECT_TRUE(RetN->preds(DAG).empty());
 }
 
 TEST_F(DependencyGraphTest, VolatileSotres) {
@@ -469,12 +476,14 @@ define void @foo(ptr noalias %ptr0, ptr noalias %ptr1, i8 %v) {
   sandboxir::DependencyGraph DAG(getAA(*LLVMF));
   DAG.extend({&*BB->begin(), BB->getTerminator()});
   auto It = BB->begin();
-  auto *Store0N = DAG.getNode(cast<sandboxir::StoreInst>(&*It++));
-  auto *Store1N = DAG.getNode(cast<sandboxir::StoreInst>(&*It++));
+  auto *Store0N = cast<sandboxir::MemDGNode>(
+      DAG.getNode(cast<sandboxir::StoreInst>(&*It++)));
+  auto *Store1N = cast<sandboxir::MemDGNode>(
+      DAG.getNode(cast<sandboxir::StoreInst>(&*It++)));
   auto *RetN = DAG.getNode(cast<sandboxir::ReturnInst>(&*It++));
   EXPECT_TRUE(Store0N->memPreds().empty());
   EXPECT_THAT(Store1N->memPreds(), testing::ElementsAre(Store0N));
-  EXPECT_TRUE(RetN->memPreds().empty());
+  EXPECT_TRUE(RetN->preds(DAG).empty());
 }
 
 TEST_F(DependencyGraphTest, Call) {
@@ -498,12 +507,12 @@ define void @foo(float %v1, float %v2) {
   DAG.extend({&*BB->begin(), BB->getTerminator()->getPrevNode()});
 
   auto It = BB->begin();
-  auto *Call1N = DAG.getNode(&*It++);
+  auto *Call1N = cast<sandboxir::MemDGNode>(DAG.getNode(&*It++));
   auto *AddN = DAG.getNode(&*It++);
-  auto *Call2N = DAG.getNode(&*It++);
+  auto *Call2N = cast<sandboxir::MemDGNode>(DAG.getNode(&*It++));
 
   EXPECT_THAT(Call1N->memPreds(), testing::ElementsAre());
-  EXPECT_THAT(AddN->memPreds(), testing::ElementsAre());
+  EXPECT_THAT(AddN->preds(DAG), testing::ElementsAre());
   EXPECT_THAT(Call2N->memPreds(), testing::ElementsAre(Call1N));
 }
 
@@ -534,8 +543,8 @@ define void @foo() {
   auto *AllocaN = DAG.getNode(&*It++);
   auto *StackRestoreN = DAG.getNode(&*It++);
 
-  EXPECT_TRUE(dependency(AllocaN, StackRestoreN));
-  EXPECT_TRUE(dependency(StackSaveN, AllocaN));
+  EXPECT_TRUE(memDependency(AllocaN, StackRestoreN));
+  EXPECT_TRUE(memDependency(StackSaveN, AllocaN));
 }
 
 // Checks that stacksave and stackrestore depend on other mem instrs.
@@ -567,9 +576,9 @@ define void @foo(i8 %v0, i8 %v1, ptr %ptr) {
   auto *StackRestoreN = DAG.getNode(&*It++);
   auto *Store1N = DAG.getNode(&*It++);
 
-  EXPECT_TRUE(dependency(Store0N, StackSaveN));
-  EXPECT_TRUE(dependency(StackSaveN, StackRestoreN));
-  EXPECT_TRUE(dependency(StackRestoreN, Store1N));
+  EXPECT_TRUE(memDependency(Store0N, StackSaveN));
+  EXPECT_TRUE(memDependency(StackSaveN, StackRestoreN));
+  EXPECT_TRUE(memDependency(StackRestoreN, Store1N));
 }
 
 // Make sure there is a dependency between a stackrestore and an alloca.
@@ -596,7 +605,7 @@ define void @foo(ptr %ptr) {
   auto *StackRestoreN = DAG.getNode(&*It++);
   auto *AllocaN = DAG.getNode(&*It++);
 
-  EXPECT_TRUE(dependency(StackRestoreN, AllocaN));
+  EXPECT_TRUE(memDependency(StackRestoreN, AllocaN));
 }
 
 // Make sure there is a dependency between the alloca and stacksave
@@ -623,7 +632,7 @@ define void @foo(ptr %ptr) {
   auto *AllocaN = DAG.getNode(&*It++);
   auto *StackSaveN = DAG.getNode(&*It++);
 
-  EXPECT_TRUE(dependency(AllocaN, StackSaveN));
+  EXPECT_TRUE(memDependency(AllocaN, StackSaveN));
 }
 
 // A non-InAlloca in a stacksave-stackrestore region does not need extra
@@ -655,6 +664,6 @@ define void @foo() {
   auto *AllocaN = DAG.getNode(&*It++);
   auto *StackRestoreN = DAG.getNode(&*It++);
 
-  EXPECT_FALSE(dependency(StackSaveN, AllocaN));
-  EXPECT_FALSE(dependency(AllocaN, StackRestoreN));
+  EXPECT_FALSE(memDependency(StackSaveN, AllocaN));
+  EXPECT_FALSE(memDependency(AllocaN, StackRestoreN));
 }