v1.35 - Adding USE_FINE_GRAIN to a2a preset (#66)

CHANGELOG.md

@@ -3,6 +3,11 @@
 Documentation for TransferBench is available at
 [https://rocm.docs.amd.com/projects/TransferBench](https://rocm.docs.amd.com/projects/TransferBench).
 
+## v1.35
+
+### Additions
+
+* USE_FINE_GRAIN also applies to a2a preset
 
 ## v1.34
 

src/TransferBench.cpp

@@ -1876,8 +1876,8 @@ void RunAllToAllBenchmark(EnvVars const& ev, size_t const numBytesPerTransfer, i
   transfer.numSrcs     = 1;
   transfer.numDsts     = 1;
   transfer.exeType     = EXE_GPU_GFX;
-  transfer.srcType.resize(1, MEM_GPU);
-  transfer.dstType.resize(1, MEM_GPU);
+  transfer.srcType.resize(1, ev.useFineGrain ? MEM_GPU_FINE : MEM_GPU);
+  transfer.dstType.resize(1, ev.useFineGrain ? MEM_GPU_FINE : MEM_GPU);
   transfer.srcIndex.resize(1);
   transfer.dstIndex.resize(1);
 

src/include/EnvVars.hpp

@@ -29,7 +29,7 @@ THE SOFTWARE.
 #include "Compatibility.hpp"
 #include "Kernels.hpp"
 
-#define TB_VERSION "1.34"
+#define TB_VERSION "1.35"
 
 extern char const MemTypeStr[];
 extern char const ExeTypeStr[];
@@ -643,6 +643,8 @@ public:
       printf("[AllToAll Related]\n");
     PRINT_EV("A2A_DIRECT", a2aDirect,
              std::string(a2aDirect ? "Only using direct links" : "Full all-to-all"));
+    PRINT_EV("USE_FINE_GRAIN", useFineGrain,
+             std::string("Using ") + (useFineGrain ? "fine" : "coarse") + "-grained memory");
     PRINT_EV("USE_REMOTE_READ", useRemoteRead,
              std::string("Using ") + (useRemoteRead ? "DST" : "SRC") + " as executor");