diff --git a/configure.ac b/configure.ac
index b758f5b..7f64855 100755
--- a/configure.ac
+++ b/configure.ac
@@ -93,7 +93,8 @@ AS_IF([test "$ac_cv_header_cuda_runtime_h" = "yes"], [
         [AC_MSG_ERROR([Library containing cudaMalloc symbol not found])])
     ])
 ])
-AM_CONDITIONAL([USE_CUDA], [test x$with_cuda = xyes])
+AM_CONDITIONAL([HAVE_CUDA], [test x$with_cuda = xyes])
+AM_COND_IF([HAVE_CUDA],[AC_DEFINE([HAVE_CUDA], [], [CUDA GPU API found])])
 
 # Check for GPUDirect
 AC_ARG_WITH([gpuDirect],
@@ -116,7 +117,7 @@ AS_IF([test "$ac_cv_header_cufile_h" = "yes"], [
     ])
 ])
 AM_CONDITIONAL([HAVE_GPU_DIRECT], [test x$with_gpuDirect = xyes])
-
+AM_COND_IF([HAVE_GPU_DIRECT],[AC_DEFINE([HAVE_GPU_DIRECT], [], [GPUDirect API found])])
 
 
 # Check for system capabilities
diff --git a/src/Makefile.am b/src/Makefile.am
index fdf746f..037433c 100755
--- a/src/Makefile.am
+++ b/src/Makefile.am
@@ -41,10 +41,14 @@ extraLDFLAGS  += -L/opt/hadoop-2.2.0/lib/native
 extraLDADD    += -lhdfs
 endif
 
-if USE_CUDA
+if HAVE_CUDA
 extraLDADD    += -lcudart
 endif
 
+if HAVE_GPU_DIRECT
+extraLDADD    += -lcufile
+endif
+
 if USE_HDF5_AIORI
 extraSOURCES  += aiori-HDF5.c
 extraLDADD    += -lhdf5 -lz
@@ -164,5 +168,5 @@ libaiori_a_CPPFLAGS = $(extraCPPFLAGS)
 .PHONY: build.conf
 all-local: build.conf
 build.conf:
-	@echo LDFLAGS=$(LDFLAGS) $(extraLDFLAGS) $(extraLDADD) > build.conf
+	@echo LDFLAGS=$(LDFLAGS) $(extraLDFLAGS) $(extraLDADD) $(LIBS) > build.conf
 	@echo CFLAGS=$(CFLAGS) $(extraCPPFLAGS) >> build.conf
diff --git a/src/utilities.c b/src/utilities.c
index 9ab5432..5972b27 100755
--- a/src/utilities.c
+++ b/src/utilities.c
@@ -86,7 +86,7 @@ void update_write_memory_pattern(uint64_t item, char * buf, size_t bytes, int bu
 void generate_memory_pattern(char * buf, size_t bytes, int buff_offset, int rank){
   uint64_t * buffi = (uint64_t*) buf;
   // first half of 64 bits use the rank
-  const uint64_t ranki = (uint64_t)(rank + 1) << 32 + buff_offset;
+  const uint64_t ranki = ((uint64_t)(rank + 1) << 32) + buff_offset;
   const size_t size = bytes / 8;
   // the first 8 bytes are set to item number
   for(size_t i=1; i < size; i++){
@@ -106,7 +106,7 @@ int verify_memory_pattern(int item, char * buffer, size_t bytes, int buff_offset
 
   uint64_t * buffi = (uint64_t*) buffer;
   // first half of 64 bits use the rank, here need to apply rank shifting
-  uint64_t rank_mod = (uint64_t)(pretendRank + 1) << 32 + buff_offset;
+  uint64_t rank_mod = ((uint64_t)(pretendRank + 1) << 32) + buff_offset;
   // the first 8 bytes are set to item number
   for(size_t i=1; i < bytes/8; i++){
     uint64_t exp = (i + 1) + rank_mod;