From 9fdbbe53dcc6afee9e8fc8217fb26b9800530ac1 Mon Sep 17 00:00:00 2001
From: Saleel Kudchadker <Saleel.Kudchadker@amd.com>
Date: Fri, 20 Jan 2023 15:35:25 -0800
Subject: [PATCH] SWDEV-364604 - Add support for hipEventDisableSystemFence

Change-Id: I1a6451c873fb22729ac61e4e80f8531251e990f0


[ROCm/clr commit: 7fc5ae2226315c0ae3aab5f9a3da2e51f75529f8]
---
 projects/clr/hipamd/src/hip_event.cpp  | 26 ++++++++++++++++----------
 projects/clr/hipamd/src/hip_module.cpp | 11 +++++++++--
 projects/clr/hipamd/src/hip_stream.cpp |  8 ++++++++
 3 files changed, 33 insertions(+), 12 deletions(-)

diff --git a/projects/clr/hipamd/src/hip_event.cpp b/projects/clr/hipamd/src/hip_event.cpp
index 83cbb9ef32..7ddac56089 100644
--- a/projects/clr/hipamd/src/hip_event.cpp
+++ b/projects/clr/hipamd/src/hip_event.cpp
@@ -222,13 +222,12 @@ hipError_t Event::recordCommand(amd::Command*& command, amd::HostQueue* stream,
                                 uint32_t ext_flags ) {
   if (command == nullptr) {
     int32_t releaseFlags = ((ext_flags == 0) ? flags : ext_flags) &
-                            (hipEventReleaseToSystem | hipEventReleaseToDevice);
-    if (releaseFlags & hipEventReleaseToDevice) {
-      releaseFlags = amd::Device::kCacheStateAgent;
-    } else if (releaseFlags & hipEventReleaseToSystem) {
-      releaseFlags = amd::Device::kCacheStateSystem;
-    } else {
+                            (hipEventReleaseToDevice | hipEventReleaseToSystem |
+                             hipEventDisableSystemFence);
+    if (releaseFlags & hipEventDisableSystemFence) {
       releaseFlags = amd::Device::kCacheStateIgnore;
+    } else {
+      releaseFlags = amd::Device::kCacheStateInvalid;
     }
     // Always submit a EventMarker.
     command = new hip::EventMarker(*stream, !kMarkerDisableFlush, true, releaseFlags);
@@ -279,14 +278,21 @@ bool isValid(hipEvent_t event) {
 // ================================================================================================
 hipError_t ihipEventCreateWithFlags(hipEvent_t* event, unsigned flags) {
   unsigned supportedFlags = hipEventDefault | hipEventBlockingSync | hipEventDisableTiming |
-      hipEventReleaseToDevice | hipEventReleaseToSystem | hipEventInterprocess;
+                            hipEventReleaseToDevice | hipEventReleaseToSystem |
+                            hipEventInterprocess | hipEventDisableSystemFence;
 
-  const unsigned releaseFlags = (hipEventReleaseToDevice | hipEventReleaseToSystem);
+  const unsigned releaseFlags = (hipEventReleaseToDevice | hipEventReleaseToSystem |
+                                 hipEventDisableSystemFence);
   // can't set any unsupported flags.
-  // can't set both release flags
+  // can set only one of the release flags.
   // if hipEventInterprocess flag is set, then hipEventDisableTiming flag also must be set
   const bool illegalFlags = (flags & ~supportedFlags) ||
-                            ((flags & releaseFlags) == releaseFlags) ||
+                            ([](unsigned int num){
+                              unsigned int bitcount;
+                              for (bitcount = 0; num; bitcount++) {
+                                num &= num - 1;
+                              }
+                              return bitcount; } (flags & releaseFlags) > 1) ||
                             ((flags & hipEventInterprocess) && !(flags & hipEventDisableTiming));
   if (!illegalFlags) {
     hip::Event* e = nullptr;
diff --git a/projects/clr/hipamd/src/hip_module.cpp b/projects/clr/hipamd/src/hip_module.cpp
index 98cc7fd599..a3fa4919e7 100644
--- a/projects/clr/hipamd/src/hip_module.cpp
+++ b/projects/clr/hipamd/src/hip_module.cpp
@@ -388,11 +388,18 @@ hipError_t ihipModuleLaunchKernel(hipFunction_t f, uint32_t globalWorkSizeX,
     }
   }
 
-  command->enqueue();
-
   if (stopEvent != nullptr) {
     hip::Event* eStop = reinterpret_cast<hip::Event*>(stopEvent);
+    if (eStop->flags & hipEventDisableSystemFence) {
+      command->setEventScope(amd::Device::kCacheStateIgnore);
+    } else {
+      command->setEventScope(amd::Device::kCacheStateSystem);
+    }
+    // Enqueue Dispatch and bind the stop event
+    command->enqueue();
     eStop->BindCommand(*command, false);
+  } else {
+    command->enqueue();
   }
 
   if (command->status() == CL_INVALID_OPERATION) {
diff --git a/projects/clr/hipamd/src/hip_stream.cpp b/projects/clr/hipamd/src/hip_stream.cpp
index e6c2839a5b..80ea30133b 100644
--- a/projects/clr/hipamd/src/hip_stream.cpp
+++ b/projects/clr/hipamd/src/hip_stream.cpp
@@ -225,6 +225,14 @@ void iHipWaitActiveStreams(hip::Stream* blocking_stream, bool wait_null_stream)
       command->enqueue();
       command->release();
     }
+
+    //Reset the dirty flag for all streams now that the marker is submitted
+    for (const auto& stream : streamSet) {
+      amd::HostQueue* active_queue = stream->asHostQueue();
+      if (active_queue->vdev()->isFenceDirty()) {
+        active_queue->vdev()->resetFenceDirty();
+      }
+    }
   }
 
   // Release all active commands. It's safe after the marker was enqueued