From b0c3a1790f911b09eb9d83de5ac9445670d255ce Mon Sep 17 00:00:00 2001
From: gilbertlee-amd <44450918+gilbertlee-amd@users.noreply.github.com>
Date: Wed, 4 Aug 2021 10:53:41 -0600
Subject: [PATCH] [TransferBench] Removing dependency on hip_fp16 header,
 fixing swapped output CSV header (#416)

[ROCm/rccl commit: 1ed272e5f072833c21a398798b2b9d40e6de3454]
---
 projects/rccl/tools/TransferBench/TransferBench.cpp | 2 +-
 projects/rccl/tools/TransferBench/TransferBench.hpp | 3 +--
 2 files changed, 2 insertions(+), 3 deletions(-)

diff --git a/projects/rccl/tools/TransferBench/TransferBench.cpp b/projects/rccl/tools/TransferBench/TransferBench.cpp
index 852661776b..899c71fc55 100644
--- a/projects/rccl/tools/TransferBench/TransferBench.cpp
+++ b/projects/rccl/tools/TransferBench/TransferBench.cpp
@@ -94,7 +94,7 @@ int main(int argc, char **argv)
   // Print CSV header
   if (ev.outputToCsv)
   {
-    printf("Test,NumBytes,Executor,SrcMem,DstMem,CUs,BW(GB/s),Time(ms),LinkDesc,SrcAddr,DstAddr,ByteOffset,numWarmups,numIters,useHipCall,useMemSet,useSingleSync,combinedTiming\n");
+    printf("Test,NumBytes,SrcMem,Executor,DstMem,CUs,BW(GB/s),Time(ms),LinkDesc,SrcAddr,DstAddr,ByteOffset,numWarmups,numIters,useHipCall,useMemSet,useSingleSync,combinedTiming\n");
   }
 
   // Loop over each line in the Link configuration file
diff --git a/projects/rccl/tools/TransferBench/TransferBench.hpp b/projects/rccl/tools/TransferBench/TransferBench.hpp
index 129ee198c3..4529d0f291 100644
--- a/projects/rccl/tools/TransferBench/TransferBench.hpp
+++ b/projects/rccl/tools/TransferBench/TransferBench.hpp
@@ -34,13 +34,12 @@ THE SOFTWARE.
 #include <hip/hip_runtime.h>
 #include <hip/hip_ext.h>
 #include <hsa/hsa_ext_amd.h>
-#include <hip/hcc_detail/hip_fp16.h>
 
 // Include common_kernel.h from RCCL for copy kernel
 // However define some variables to avoid extra includes / missing defines
 #define NCCL_DEVICE_H_   // Avoid loading devcomm.h
 #define WARP_SIZE 64
-
+typedef float half;  // TransferBench doesn't actually operate on half-precision floats
 typedef uint64_t PackType;
 typedef ulong2 Pack128;
 typedef struct