From 0dadf681db8c42ccbcd70efd3caf607f00bb08cd Mon Sep 17 00:00:00 2001
From: Ben Sander <ben.sander@amd.com>
Date: Mon, 26 Sep 2016 16:32:35 -0500
Subject: [PATCH] Fix signal resource issue.

Remove memory leak with new hc::completion_future.
Implement HIP_LAUNCH_BLOCKING with queue-level wait.

Change-Id: I45975f81c4d239fdeed7776970988d28449865dc


[ROCm/hip commit: 225e37fdc9eaec5ae4d33c5b10863786c1ff51f6]
---
 projects/hip/include/hcc_detail/hip_hcc.h |  2 +-
 projects/hip/src/hip_hcc.cpp              | 15 ++++++++-------
 projects/hip/src/hip_memory.cpp           |  5 +++--
 3 files changed, 12 insertions(+), 10 deletions(-)

diff --git a/projects/hip/include/hcc_detail/hip_hcc.h b/projects/hip/include/hcc_detail/hip_hcc.h
index 6c061b01a9..9d4b70c73f 100644
--- a/projects/hip/include/hcc_detail/hip_hcc.h
+++ b/projects/hip/include/hcc_detail/hip_hcc.h
@@ -373,7 +373,7 @@ typedef uint64_t SeqNum_t ;
     //---
     // Member functions that begin with locked_ are thread-safe accessors - these acquire / release the critical mutex.
     LockedAccessor_StreamCrit_t  lockopen_preKernelCommand();
-    void                 lockclose_postKernelCommand(hc::completion_future &kernel_future);
+    void                 lockclose_postKernelCommand(hc::accelerator_view *av);
 
 
     void                 locked_wait(bool assertQueueEmpty=false);
diff --git a/projects/hip/src/hip_hcc.cpp b/projects/hip/src/hip_hcc.cpp
index 1058d96412..fcbc553f83 100644
--- a/projects/hip/src/hip_hcc.cpp
+++ b/projects/hip/src/hip_hcc.cpp
@@ -253,11 +253,12 @@ LockedAccessor_StreamCrit_t ihipStream_t::lockopen_preKernelCommand()
 
 //---
 // Must be called after kernel finishes, this releases the lock on the stream so other commands can submit.
-void ihipStream_t::lockclose_postKernelCommand(hc::completion_future &kernelFuture)
+void ihipStream_t::lockclose_postKernelCommand(hc::accelerator_view *av)
 {
 
     if (HIP_LAUNCH_BLOCKING) {
-        kernelFuture.wait();
+        // TODO - fix this so it goes through proper stream::wait() call.
+        av->wait();  // direct wait OK since we know the stream is locked. 
         tprintf(DB_SYNC, " %s LAUNCH_BLOCKING for kernel completion\n", ToString(this).c_str());
     }
 
@@ -1163,7 +1164,7 @@ hipStream_t ihipPreLaunchKernel(hipStream_t stream, dim3 grid, dim3 block, grid_
 
     auto crit = stream->lockopen_preKernelCommand();
     lp->av = &(crit->_av);
-    lp->cf = new hc::completion_future;
+    lp->cf = nullptr;
     ihipPrintKernelLaunch(kernelNameStr, lp, stream);
 
     return (stream);
@@ -1185,7 +1186,7 @@ hipStream_t ihipPreLaunchKernel(hipStream_t stream, size_t grid, dim3 block, gri
 
     auto crit = stream->lockopen_preKernelCommand();
     lp->av = &(crit->_av);
-    lp->cf = new hc::completion_future;
+    lp->cf = nullptr;
     ihipPrintKernelLaunch(kernelNameStr, lp, stream);
     return (stream);
 }
@@ -1206,7 +1207,7 @@ hipStream_t ihipPreLaunchKernel(hipStream_t stream, dim3 grid, size_t block, gri
 
     auto crit = stream->lockopen_preKernelCommand();
     lp->av = &(crit->_av);
-    lp->cf = new hc::completion_future;
+    lp->cf = nullptr;
     ihipPrintKernelLaunch(kernelNameStr, lp, stream);
     return (stream);
 }
@@ -1227,7 +1228,7 @@ hipStream_t ihipPreLaunchKernel(hipStream_t stream, size_t grid, size_t block, g
 
     auto crit = stream->lockopen_preKernelCommand();
     lp->av = &(crit->_av);
-    lp->cf = new hc::completion_future; // TODO, is this necessary?
+    lp->cf = nullptr;
 
 
     ihipPrintKernelLaunch(kernelNameStr, lp, stream);
@@ -1240,7 +1241,7 @@ hipStream_t ihipPreLaunchKernel(hipStream_t stream, size_t grid, size_t block, g
 //This releases the lock on the stream.
 void ihipPostLaunchKernel(hipStream_t stream, grid_launch_parm &lp)
 {
-    stream->lockclose_postKernelCommand(*(lp.cf));
+    stream->lockclose_postKernelCommand(lp.av);
 }
 
 
diff --git a/projects/hip/src/hip_memory.cpp b/projects/hip/src/hip_memory.cpp
index d2bb084fc5..08c4b4392f 100644
--- a/projects/hip/src/hip_memory.cpp
+++ b/projects/hip/src/hip_memory.cpp
@@ -772,7 +772,7 @@ hipError_t hipMemsetAsync(void* dst, int  value, size_t sizeBytes, hipStream_t s
             }
         }
 
-        stream->lockclose_postKernelCommand(cf);
+        stream->lockclose_postKernelCommand(&crit->_av);
 
 
         if (HIP_LAUNCH_BLOCKING) {
@@ -822,9 +822,10 @@ hipError_t hipMemset(void* dst, int  value, size_t sizeBytes )
                 e = hipErrorInvalidValue;
             }
         }
+        // TODO - is hipMemset supposed to be async?
         cf.wait();
 
-        stream->lockclose_postKernelCommand(cf);
+        stream->lockclose_postKernelCommand(&crit->_av);
 
 
         if (HIP_LAUNCH_BLOCKING) {