From 2f4269d06d803541057c6d4ef9ec192660d22fbc Mon Sep 17 00:00:00 2001
From: gilbertlee-amd <44450918+gilbertlee-amd@users.noreply.github.com>
Date: Thu, 12 Dec 2019 15:20:54 -0700
Subject: [PATCH] Adding new sleep after sync capability for data fabric
 profiling (#162)

Fixing missing header include for ROCM 3.0 changes
---
 tools/TransferBench/TransferBench.cpp | 18 +++++++++++++-----
 tools/TransferBench/TransferBench.hpp |  2 ++
 2 files changed, 15 insertions(+), 5 deletions(-)

diff --git a/tools/TransferBench/TransferBench.cpp b/tools/TransferBench/TransferBench.cpp
index afb548ff58..016ff80ed4 100644
--- a/tools/TransferBench/TransferBench.cpp
+++ b/tools/TransferBench/TransferBench.cpp
@@ -27,6 +27,7 @@ THE SOFTWARE.
 #include <cstdio>
 #include <cstdlib>
 #include <set>
+#include <unistd.h>
 
 #include <hip/hip_runtime.h>
 #include "copy_kernel.h"
@@ -49,12 +50,13 @@ int main(int argc, char **argv)
     printf("\n");
     printf("Environment variables:\n");
     printf("======================\n");
-    printf(" USE_HIP_CALL    - Use hip calls (hipMemcpyAsync/hipMemset) instead of kernel\n");
-    printf(" USE_MEMSET      - Write constant value (instead of doing a copy)\n");
-    printf(" USE_COARSE_MEM  - Use coarse-grained dst GPU memory (instead of fine-grained)\n");
-    printf(" USE_SINGLE_SYNC - Only synchronize once at end of iterations (disables GPU times)\n");
-    printf(" USE_INTERACTIVE - Waits for user-input prior to start and after transfer loop (for profiling)\n");
+    printf(" USE_HIP_CALL     - Use hip calls (hipMemcpyAsync/hipMemset) instead of kernel\n");
+    printf(" USE_MEMSET       - Write constant value (instead of doing a copy)\n");
+    printf(" USE_COARSE_MEM   - Use coarse-grained dst GPU memory (instead of fine-grained)\n");
+    printf(" USE_SINGLE_SYNC  - Only synchronize once at end of iterations (disables GPU times)\n");
+    printf(" USE_INTERACTIVE  - Waits for user-input prior to start and after transfer loop (for profiling)\n");
     printf(" USE_ITERATIONS=N - Sets number of iterations to run (default is 10)\n");
+    printf(" USE_SLEEP        - Adds a 100ms sleep after sync (for profiling)\n");
     exit(0);
   }
 
@@ -74,6 +76,7 @@ int main(int argc, char **argv)
   bool useCoarseMem = getenv("USE_COARSE_MEM");
   bool useSingleSync = getenv("USE_SINGLE_SYNC");
   bool useInteractive = getenv("USE_INTERACTIVE");
+  bool useSleep = getenv("USE_SLEEP");
 
   int numWarmups = 3;
   int numIterations = getenv("USE_ITERATIONS") ? atoi(getenv("USE_ITERATIONS")) : 10;
@@ -99,6 +102,10 @@ int main(int argc, char **argv)
     printf("Running in interactive mode (USE_INTERACTIVE)\n");
   else
     printf("Running in non-interactive mode (enable interactive mode via USE_INTERACTIVE)\n");
+  if (useSleep)
+    printf("Adding 100ms sleep after sync (USE_SLEEP)\n");
+  else
+    printf("No sleep per sync (enable sleep via USE_SLEEP)\n");
 
   printf("Executing %d warmup iteration(s), and %d timed iteration(s) (Set via USE_ITERATION=#)\n",
          numWarmups, numIterations);
@@ -314,6 +321,7 @@ int main(int argc, char **argv)
 
       auto cpuDelta = std::chrono::high_resolution_clock::now() - cpuStart;
       double deltaSec = std::chrono::duration_cast<std::chrono::duration<double>>(cpuDelta).count();
+      if (useSleep) usleep(100000);
 
       if (iteration >= 0)
       {
diff --git a/tools/TransferBench/TransferBench.hpp b/tools/TransferBench/TransferBench.hpp
index 5fde8f58fe..92c596d672 100644
--- a/tools/TransferBench/TransferBench.hpp
+++ b/tools/TransferBench/TransferBench.hpp
@@ -20,6 +20,8 @@ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
 THE SOFTWARE.
 */
 
+#include <sstream>
+
 // Helper macro for catching HIP errors
 #define HIP_CALL(cmd)                                                   \
     do {                                                                \