Fixing #SubExec bug in P2PBenchmark (#47)

CHANGELOG.md

 # Changelog for TransferBench
 
+## v1.25
+### Fixed
+- Fixed bug in P2P bidirectional benchmark using incorrect number of subExecutors for CPU<->GPU tests
+
 ## v1.24
 ### Added
 - New All-To-All GPU benchmark accessed by preset "a2a"

src/TransferBench.cpp

@@ -1508,7 +1508,7 @@ double GetPeakBandwidth(EnvVars const& ev, size_t const N,
   transfers[0].exeIndex = (ev.useRemoteRead ? dstIndex : srcIndex);
   transfers[1].exeIndex = (ev.useRemoteRead ? srcIndex : dstIndex);
   transfers[0].numSubExecs = IsGpuType(transfers[0].exeType) ? ev.numGpuSubExecs : ev.numCpuSubExecs;
-  transfers[1].numSubExecs = IsGpuType(transfers[0].exeType) ? ev.numGpuSubExecs : ev.numCpuSubExecs;
+  transfers[1].numSubExecs = IsGpuType(transfers[1].exeType) ? ev.numGpuSubExecs : ev.numCpuSubExecs;
 
   // Remove (DST->SRC) if not bidirectional
   transfers.resize(isBidirectional + 1);

src/include/EnvVars.hpp

@@ -29,7 +29,7 @@ THE SOFTWARE.
 #include "Compatibility.hpp"
 #include "Kernels.hpp"
 
-#define TB_VERSION "1.24"
+#define TB_VERSION "1.25"
 
 extern char const MemTypeStr[];
 extern char const ExeTypeStr[];