From bdf4b84f82e7b0e864aaa06010a438fbe63a3f3c Mon Sep 17 00:00:00 2001
From: Sean Keely <Sean.Keely@amd.com>
Date: Thu, 18 Apr 2019 19:59:16 -0500
Subject: [PATCH] Don't create blits when copy profiling is enabled.

Change-Id: I879827133957ee610c3381ea30c536ec7d10ffab


[ROCm/ROCR-Runtime commit: 1251842900e5927a615a66bd0d286328eb25ec94]
---
 .../hsa-runtime/core/runtime/amd_gpu_agent.cpp      |  2 +-
 .../hsa-runtime/core/runtime/hsa_ext_amd.cpp        | 13 +++++++------
 .../runtime/hsa-runtime/core/util/lazy_ptr.h        |  5 ++++-
 3 files changed, 12 insertions(+), 8 deletions(-)

diff --git a/projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_gpu_agent.cpp b/projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_gpu_agent.cpp
index d3dc21cc87..245bc4d64a 100644
--- a/projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_gpu_agent.cpp
+++ b/projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_gpu_agent.cpp
@@ -676,7 +676,7 @@ hsa_status_t GpuAgent::EnableDmaProfiling(bool enable) {
   }
 
   for (int i = 0; i < BlitCount; ++i) {
-    if (blits_[i] != NULL) {
+    if (blits_[i].created()) {
       const hsa_status_t stat = blits_[i]->EnableProfiling(enable);
       if (stat != HSA_STATUS_SUCCESS) {
         return stat;
diff --git a/projects/rocr-runtime/runtime/hsa-runtime/core/runtime/hsa_ext_amd.cpp b/projects/rocr-runtime/runtime/hsa-runtime/core/runtime/hsa_ext_amd.cpp
index b4fd546431..d804d81558 100644
--- a/projects/rocr-runtime/runtime/hsa-runtime/core/runtime/hsa_ext_amd.cpp
+++ b/projects/rocr-runtime/runtime/hsa-runtime/core/runtime/hsa_ext_amd.cpp
@@ -328,12 +328,13 @@ hsa_status_t hsa_amd_profiling_async_copy_enable(bool enable) {
   TRY;
   IS_OPEN();
 
-  return core::Runtime::runtime_singleton_->IterateAgent(
-      [](hsa_agent_t agent_handle, void* data) -> hsa_status_t {
-        const bool enable = *(reinterpret_cast<bool*>(data));
-        return core::Agent::Convert(agent_handle)->profiling_enabled(enable);
-      },
-      reinterpret_cast<void*>(&enable));
+  hsa_status_t ret = HSA_STATUS_SUCCESS;
+  for (core::Agent* agent : core::Runtime::runtime_singleton_->gpu_agents()) {
+    hsa_status_t err = agent->profiling_enabled(enable);
+    if (err != HSA_STATUS_SUCCESS) ret = err;
+  }
+  return ret;
+
   CATCH;
 }
 
diff --git a/projects/rocr-runtime/runtime/hsa-runtime/core/util/lazy_ptr.h b/projects/rocr-runtime/runtime/hsa-runtime/core/util/lazy_ptr.h
index 7837200d89..3e00b74db3 100644
--- a/projects/rocr-runtime/runtime/hsa-runtime/core/util/lazy_ptr.h
+++ b/projects/rocr-runtime/runtime/hsa-runtime/core/util/lazy_ptr.h
@@ -88,10 +88,13 @@ template <typename T> class lazy_ptr {
 
   /*
    * Ensures that the object is created or is being created.
-   * This is useful when early consruction of the object is required.
+   * This is useful when early construction of the object is required.
    */
   void touch() const { make(false); }
 
+  // Tells if the lazy object has been constructed or not.
+  bool created() const { return obj != nullptr; }
+
  private:
   mutable std::unique_ptr<T> obj;
   mutable std::function<T*(void)> func;