From 44ada1f189c07abcecf7a7a1df32010922251828 Mon Sep 17 00:00:00 2001
From: Vivek Kale <11766050+vlkale@users.noreply.github.com>
Date: Thu, 23 May 2024 13:05:53 -0400
Subject: [PATCH] Sampler's begin for callee's null check before fence (less
 overhead in corner case) (#253)

* kp_sampler_skip.cpp: put begin for callee check before fence

This improves performance in the case there is no callee for the kokkosp_begin_parallel_for.

This is actually done correctly in the kokkosp_begin_parallel_scan and begin_parallel_reduce.

* kp_sampler_skip.cpp: apply clang-format
---
 common/kokkos-sampler/kp_sampler_skip.cpp | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/common/kokkos-sampler/kp_sampler_skip.cpp b/common/kokkos-sampler/kp_sampler_skip.cpp
index 8df516e9e..4d8b578fb 100644
--- a/common/kokkos-sampler/kp_sampler_skip.cpp
+++ b/common/kokkos-sampler/kp_sampler_skip.cpp
@@ -200,10 +200,11 @@ void kokkosp_begin_parallel_for(const char* name, const uint32_t devID,
       std::cout << "KokkosP: sample " << *kID
                 << " calling child-begin function...\n";
     }
-    if (tool_globFence) {
-      invoke_ktools_fence(0);
-    }
+
     if (NULL != beginForCallee) {
+      if (tool_globFence) {
+        invoke_ktools_fence(0);
+      }
       uint64_t nestedkID = 0;
       (*beginForCallee)(name, devID, &nestedkID);
       if (tool_verbosity > 0) {