From c9fb07ffe57372f982dccd368a649c96effc880b Mon Sep 17 00:00:00 2001
From: Cloud Han <cloudhan@outlook.com>
Date: Fri, 28 Jan 2022 11:55:42 +0800
Subject: [PATCH] remove blocking collect

---
 TracyOpenCL.hpp                              | 13 +++----------
 examples/OpenCLVectorAdd/OpenCLVectorAdd.cpp |  4 ++--
 2 files changed, 5 insertions(+), 12 deletions(-)

diff --git a/TracyOpenCL.hpp b/TracyOpenCL.hpp
index fb60431c..d477158d 100644
--- a/TracyOpenCL.hpp
+++ b/TracyOpenCL.hpp
@@ -21,7 +21,6 @@
 #define TracyCLZoneSetEvent(e)
 
 #define TracyCLCollect(c)
-#define TracyCLBlockingCollect(c)
 
 namespace tracy
 {
@@ -136,7 +135,7 @@ namespace tracy {
             Profiler::QueueSerialFinish();
         }
 
-        void Collect(bool blocking)
+        void Collect()
         {
             ZoneScopedC(Color::Red4);
 
@@ -166,12 +165,7 @@ namespace tracy {
                     assert(false && "clGetEventInfo failed, maybe a TracyCLZone is not paired with TracyCLZoneSetEvent");
                     continue;
                 }
-                if (eventStatus != CL_COMPLETE)
-                {
-                    if (!blocking)
-                        return;
-                    TRACY_CL_CHECK_ERROR(clWaitForEvents(1, &eventInfo.event));
-                }
+                if (eventStatus != CL_COMPLETE) return;
 
                 cl_int eventInfoQuery = (eventInfo.phase == EventPhase::Begin)
                     ? CL_PROFILING_COMMAND_START
@@ -359,8 +353,7 @@ using TracyCLCtx = tracy::OpenCLCtx*;
 #define TracyCLNamedZoneSetEvent(varname, event) varname.SetEvent(event)
 #define TracyCLZoneSetEvent(event) __tracy_gpu_zone.SetEvent(event)
 
-#define TracyCLCollect(ctx) ctx->Collect(/*blocking=*/false)
-#define TracyCLBlockingCollect(ctx) ctx->Collect(/*blocking=*/true)
+#define TracyCLCollect(ctx) ctx->Collect()
 
 #endif
 
diff --git a/examples/OpenCLVectorAdd/OpenCLVectorAdd.cpp b/examples/OpenCLVectorAdd/OpenCLVectorAdd.cpp
index ce5adece..a2560eae 100644
--- a/examples/OpenCLVectorAdd/OpenCLVectorAdd.cpp
+++ b/examples/OpenCLVectorAdd/OpenCLVectorAdd.cpp
@@ -200,8 +200,8 @@ int main()
     }) / (durations.size() - 1.0f);
     std::cout << "VectorAdd runtime avg: " << avg << "us, std: " << sqrt(stddev2) << "us over " << numLaunchs << " runs." << std::endl;
 
-    // Use blocking collect will ensure all queued events is finished
-    TracyCLBlockingCollect(tracyCLCtx);
+    // User should ensure all events are finished, in this case, collect after the clFinish will do the trick.
+    TracyCLCollect(tracyCLCtx);
 
     {
         ZoneScopedN("Checking results");