Fixing USE_INTERACTIVE bug (#142)

CHANGELOG.md

@@ -3,6 +3,10 @@
 Documentation for TransferBench is available at
 [https://rocm.docs.amd.com/projects/TransferBench](https://rocm.docs.amd.com/projects/TransferBench).
 
+## v1.56
+### Fixed
+- Fixed bug when using interactive mode.  Interactive mode now starts prior to all warmup iterations
+
 ## v1.55
 ### Fixed
 - Fixed missing header error when compiling on CentOS

CMakeLists.txt

@@ -7,7 +7,7 @@ else()
 endif()
 cmake_minimum_required(VERSION 3.5)
 
-project(TransferBench VERSION 1.55.0 LANGUAGES CXX)
+project(TransferBench VERSION 1.56.0 LANGUAGES CXX)
 
 # Default GPU architectures to build
 #==================================================================================================

Makefile

@@ -18,7 +18,7 @@ endif
 
 CXXFLAGS = -I$(ROCM_PATH)/include -lnuma -L$(ROCM_PATH)/lib -lhsa-runtime64
 NVFLAGS  = -x cu -lnuma -arch=native
-COMMON_FLAGS = -g -O3 --std=c++20 -I./src/header -I./src/client -I./src/client/Presets
+COMMON_FLAGS = -O3 --std=c++20 -I./src/header -I./src/client -I./src/client/Presets
 LDFLAGS += -lpthread
 
 all: $(EXE)

src/client/Client.hpp

@@ -23,7 +23,7 @@ THE SOFTWARE.
 #pragma once
 
 // TransferBench client version
-#define CLIENT_VERSION "1.55.00"
+#define CLIENT_VERSION "1.56.00"
 
 #include "TransferBench.hpp"
 #include "EnvVars.hpp"

src/header/TransferBench.hpp

@@ -49,7 +49,7 @@ namespace TransferBench
   using std::set;
   using std::vector;
 
-  constexpr char VERSION[] = "1.54";
+  constexpr char VERSION[] = "1.56";
 
   /**
    * Enumeration of supported Executor types
@@ -2226,6 +2226,26 @@ namespace {
       }
     }
 
+    // Pause before starting when running in iteractive mode
+    if (cfg.general.useInteractive) {
+      printf("Memory prepared:\n");
+
+      for (int i = 0; i < transfers.size(); i++) {
+        ExeInfo const& exeInfo = executorMap[transfers[i].exeDevice];
+        printf("Transfer %03d:\n", i);
+        for (int iSrc = 0; iSrc < transfers[i].srcs.size(); ++iSrc)
+          printf("  SRC %0d: %p\n", iSrc, transferResources[i]->srcMem[iSrc]);
+        for (int iDst = 0; iDst < transfers[i].dsts.size(); ++iDst)
+          printf("  DST %0d: %p\n", iDst, transferResources[i]->dstMem[iDst]);
+      }
+      printf("Hit <Enter> to continue: ");
+      if (scanf("%*c") != 0) {
+        printf("[ERROR] Unexpected input\n");
+        exit(1);
+      }
+      printf("\n");
+    }
+
     // Perform iterations
     size_t numTimedIterations = 0;
     double totalCpuTimeSec = 0.0;
@@ -2234,25 +2254,6 @@ namespace {
       if (cfg.general.numIterations > 0 && iteration >= cfg.general.numIterations) break;
       if (cfg.general.numIterations < 0 && totalCpuTimeSec > -cfg.general.numIterations) break;
 
-      // Pause before starting first timed iteration in iteractive mode
-      if (cfg.general.useInteractive && iteration == 0) {
-        printf("Memory prepared:\n");
-
-        for (int i = 0; i < transfers.size(); i++) {
-          ExeInfo const& exeInfo = executorMap[transfers[i].exeDevice];
-          printf("Transfer %03d:\n", i);
-          for (int iSrc = 0; iSrc < transfers[i].srcs.size(); ++iSrc)
-            printf("  SRC %0d: %p\n", iSrc, exeInfo.resources[i].srcMem[iSrc]);
-          for (int iDst = 0; iDst < transfers[i].dsts.size(); ++iDst)
-            printf("  DST %0d: %p\n", iDst, exeInfo.resources[i].dstMem[iDst]);
-        }
-        printf("Hit <Enter> to continue: ");
-        if (scanf("%*c") != 0) {
-          printf("[ERROR] Unexpected input\n");
-          exit(1);
-        }
-        printf("\n");
-      }
 
       // Start CPU timing for this iteration
       auto cpuStart = std::chrono::high_resolution_clock::now();