From 0ae70a4c8dec0ac97a1a184ebc5526cffebe5e67 Mon Sep 17 00:00:00 2001
From: Sourabh Betigeri <sourabh.betigeri@amd.com>
Date: Mon, 7 Mar 2022 12:08:26 -0800
Subject: [PATCH] SWDEV-313503 - Returns appropriate error code when destroyed
 exec graphs are launched

Change-Id: I3ca025494fd27f9ed0cd4534c740f7e19e1c66b3
---
 hipamd/src/hip_graph.cpp          |  2 +-
 hipamd/src/hip_graph_internal.cpp | 10 ++++++++++
 hipamd/src/hip_graph_internal.hpp | 12 +++++++++++-
 3 files changed, 22 insertions(+), 2 deletions(-)

diff --git a/hipamd/src/hip_graph.cpp b/hipamd/src/hip_graph.cpp
index fc07149498..17890e3dda 100644
--- a/hipamd/src/hip_graph.cpp
+++ b/hipamd/src/hip_graph.cpp
@@ -997,7 +997,7 @@ hipError_t ihipGraphLaunch(hipGraphExec_t graphExec, hipStream_t stream) {
 
 hipError_t hipGraphLaunch(hipGraphExec_t graphExec, hipStream_t stream) {
   HIP_INIT_API(hipGraphLaunch, graphExec, stream);
-  if (graphExec == nullptr || !hip::isValid(stream)) {
+  if (graphExec == nullptr || !hip::isValid(stream) || !hipGraphExec::isGraphExecValid(graphExec)) {
     HIP_RETURN(hipErrorInvalidValue);
   }
   HIP_RETURN_DURATION(ihipGraphLaunch(graphExec, stream));
diff --git a/hipamd/src/hip_graph_internal.cpp b/hipamd/src/hip_graph_internal.cpp
index 28a824b101..7569f1a639 100644
--- a/hipamd/src/hip_graph_internal.cpp
+++ b/hipamd/src/hip_graph_internal.cpp
@@ -51,6 +51,8 @@ std::unordered_set<hipGraphNode*> hipGraphNode::nodeSet_;
 amd::Monitor hipGraphNode::nodeSetLock_{"Guards global node set"};
 std::unordered_set<ihipGraph*> ihipGraph::graphSet_;
 amd::Monitor ihipGraph::graphSetLock_{"Guards global graph set"};
+std::unordered_set<hipGraphExec*> hipGraphExec::graphExecSet_;
+amd::Monitor hipGraphExec::graphExecSetLock_{"Guards global exec graph set"};
 
 hipError_t hipGraphMemcpyNode1D::ValidateParams(void* dst, const void* src, size_t count,
                                                 hipMemcpyKind kind) {
@@ -673,6 +675,14 @@ ihipGraph* ihipGraph::clone() const{
   return clone(clonedNodes);
 }
 
+bool hipGraphExec::isGraphExecValid(hipGraphExec* pGraphExec) {
+  amd::ScopedLock lock(graphExecSetLock_);
+  if (graphExecSet_.find(pGraphExec) == graphExecSet_.end()) {
+    return false;
+  }
+  return true;
+}
+
 hipError_t hipGraphExec::CreateQueues(size_t numQueues) {
   parallelQueues_.reserve(numQueues);
   for (size_t i = 0; i < numQueues; i++) {
diff --git a/hipamd/src/hip_graph_internal.hpp b/hipamd/src/hip_graph_internal.hpp
index cfb3594699..11dda0bfaa 100644
--- a/hipamd/src/hip_graph_internal.hpp
+++ b/hipamd/src/hip_graph_internal.hpp
@@ -290,6 +290,8 @@ struct hipGraphExec {
   uint currentQueueIndex_;
   std::unordered_map<Node, Node> clonedNodes_;
   amd::Command* lastEnqueuedCommand_;
+  static std::unordered_set<hipGraphExec*> graphExecSet_;
+  static amd::Monitor graphExecSetLock_ ;
 
  public:
   hipGraphExec(std::vector<Node>& levelOrder, std::vector<std::vector<Node>>& lists,
@@ -300,7 +302,10 @@ struct hipGraphExec {
         nodeWaitLists_(nodeWaitLists),
         clonedNodes_(clonedNodes),
         lastEnqueuedCommand_(nullptr),
-        currentQueueIndex_(0) {}
+        currentQueueIndex_(0) {
+    amd::ScopedLock lock(graphExecSetLock_);
+    graphExecSet_.insert(this);
+        }
 
   ~hipGraphExec() {
     // new commands are launched for every launch they are destroyed as and when command is
@@ -309,6 +314,8 @@ struct hipGraphExec {
       queue->release();
     }
     for (auto it = clonedNodes_.begin(); it != clonedNodes_.end(); it++) delete it->second;
+    amd::ScopedLock lock(graphExecSetLock_);
+    graphExecSet_.erase(this);
   }
 
   Node GetClonedNode(Node node) {
@@ -321,6 +328,9 @@ struct hipGraphExec {
     return clonedNode;
   }
 
+  // check executable graphs validity
+  static bool isGraphExecValid(hipGraphExec* pGraphExec);
+
   std::vector<Node>& GetNodes() { return levelOrder_; }
 
   amd::HostQueue* GetAvailableQueue() { return parallelQueues_[currentQueueIndex_++]; }