From 442e722c1e09a5370728ba7b307d90d28e9b4fa1 Mon Sep 17 00:00:00 2001
From: Siddharth Bhat <siddu.druid@gmail.com>
Date: Tue, 1 Aug 2017 14:36:24 +0000
Subject: [PATCH] [GPUJIT] Call `cuProfilerStop` before destroying the context
 to flush profiler cache.

This is necessary to get accurate traces from `nvprof` / `nvcc`.
Otherwise, we lose some profiling information.

Differential Revision: https://reviews.llvm.org/D35940

llvm-svn: 309682
---
 polly/tools/GPURuntime/GPUJIT.c | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/polly/tools/GPURuntime/GPUJIT.c b/polly/tools/GPURuntime/GPUJIT.c
index 02a91a7913d3..a26bc7c01b2d 100644
--- a/polly/tools/GPURuntime/GPUJIT.c
+++ b/polly/tools/GPURuntime/GPUJIT.c
@@ -958,6 +958,9 @@ static CuMemFreeFcnTy *CuMemFreeFcnPtr;
 typedef CUresult CUDAAPI CuModuleUnloadFcnTy(CUmodule);
 static CuModuleUnloadFcnTy *CuModuleUnloadFcnPtr;
 
+typedef CUresult CUDAAPI CuProfilerStopFcnTy();
+static CuProfilerStopFcnTy *CuProfilerStopFcnPtr;
+
 typedef CUresult CUDAAPI CuCtxDestroyFcnTy(CUcontext);
 static CuCtxDestroyFcnTy *CuCtxDestroyFcnPtr;
 
@@ -1085,6 +1088,9 @@ static int initialDeviceAPIsCUDA() {
   CuModuleUnloadFcnPtr =
       (CuModuleUnloadFcnTy *)getAPIHandleCUDA(HandleCuda, "cuModuleUnload");
 
+  CuProfilerStopFcnPtr =
+      (CuProfilerStopFcnTy *)getAPIHandleCUDA(HandleCuda, "cuProfilerStop");
+
   CuCtxDestroyFcnPtr =
       (CuCtxDestroyFcnTy *)getAPIHandleCUDA(HandleCuda, "cuCtxDestroy");
 
@@ -1416,6 +1422,7 @@ static void freeContextCUDA(PollyGPUContext *Context) {
 
   CUDAContext *Ctx = (CUDAContext *)Context->Context;
   if (Ctx->Cuda) {
+    CuProfilerStopFcnPtr();
     CuCtxDestroyFcnPtr(Ctx->Cuda);
     free(Ctx);
     free(Context);