diff --git a/CMakeLists.txt b/CMakeLists.txt
index f43e0e0fc..a6cf2ef83 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -29,10 +29,8 @@ option(NO_AFFINITY "Disable support for CPU affinity masks to avoid binding proc
 else()
 set(NO_AFFINITY 1)
 endif()
-option(BUILD_SINGLE "Single precision" OFF)
-option(BUILD_DOUBLE "Double precision" OFF)
-option(BUILD_COMPLEX "Single precision" OFF)
-option(BUILD_COMPLEX16 "Single precision" OFF)
+option(CPP_THREAD_SAFETY_TEST "Run a massively parallel DGEMM test to confirm thread safety of the library (requires OpenMP and about 1.3GB of RAM)" OFF)
+option(CPP_THREAD_SAFETY_GEMV "Run a massively parallel DGEMV test to confirm thread safety of the library (requires OpenMP)" OFF)
 
 # Add a prefix or suffix to all exported symbol names in the shared library.
 # Avoids conflicts with other BLAS libraries, especially when using
@@ -91,13 +89,13 @@ if (NOT NO_LAPACK)
   list(APPEND SUBDIRS lapack)
 endif ()
 
-if (NOT DEFINED BUILD_HALF)
- set (BUILD_HALF false)
+if (NOT DEFINED BUILD_BFLOAT16)
+ set (BUILD_BFLOAT16 false)
 endif ()
 # set which float types we want to build for
 if (NOT DEFINED BUILD_SINGLE AND NOT DEFINED BUILD_DOUBLE AND NOT DEFINED BUILD_COMPLEX AND NOT DEFINED BUILD_COMPLEX16)
   # if none are defined, build for all
-#  set(BUILD_HALF true)
+#  set(BUILD_BFLOAT16 true)
   set(BUILD_SINGLE true)
   set(BUILD_DOUBLE true)
   set(BUILD_COMPLEX true)
@@ -110,33 +108,28 @@ endif()
 
 set(FLOAT_TYPES "")
 if (BUILD_SINGLE)
-  message(STATUS "Building Songle Precision")
-  list(APPEND FLOAT_TYPES "SINGLE") 
-  #    set(CCOMMON_OPT "${CCOMMON_OPT} -DBUILD_SINGLE=1")
+  message(STATUS "Building Single Precision")
+  list(APPEND FLOAT_TYPES "SINGLE") # defines nothing
 endif ()
 
 if (BUILD_DOUBLE)
   message(STATUS "Building Double Precision")
-  list(APPEND FLOAT_TYPES "DOUBLE")
-  #set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -DBUILD_DOUBLE=1")
+  list(APPEND FLOAT_TYPES "DOUBLE") # defines DOUBLE
 endif ()
 
 if (BUILD_COMPLEX)
   message(STATUS "Building Complex Precision")
-  list(APPEND FLOAT_TYPES "COMPLEX")
-  #set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -DBUILD_COMPLEX=1")
-endif () 
+  list(APPEND FLOAT_TYPES "COMPLEX") # defines COMPLEX
+endif ()
 
 if (BUILD_COMPLEX16)
   message(STATUS "Building Double Complex Precision")
-  list(APPEND FLOAT_TYPES "ZCOMPLEX") 
-  #set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -DBUILD_COMPLEX16=1")
+  list(APPEND FLOAT_TYPES "ZCOMPLEX") # defines COMPLEX and DOUBLE
 endif ()
 
-if (BUILD_HALF)
+if (BUILD_BFLOAT16)
   message(STATUS "Building Half Precision")
-  list(APPEND FLOAT_TYPES "HALF")
-  set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -DBUILD_HALF")
+  list(APPEND FLOAT_TYPES "BFLOAT16") # defines nothing
 endif ()
 
 if (NOT DEFINED CORE OR "${CORE}" STREQUAL "UNKNOWN")
@@ -243,6 +236,9 @@ if (NOT MSVC AND NOT NOFORTRAN)
     add_subdirectory(ctest)
   endif()
   add_subdirectory(lapack-netlib/TESTING)
+	  if (CPP_THREAD_SAFETY_TEST OR CPP_THREAD_SAFETY_GEMV)
+	  	add_subdirectory(cpp_thread_test)
+    	  endif()
 endif()
 
 set_target_properties(${OpenBLAS_LIBNAME} PROPERTIES
diff --git a/Makefile.rule b/Makefile.rule
index 09dfb0881..67d183936 100644
--- a/Makefile.rule
+++ b/Makefile.rule
@@ -272,17 +272,33 @@ COMMON_PROF = -pg
 # work at all.
 #
 # CPP_THREAD_SAFETY_TEST = 1
+#
+# use this to run only the less memory-hungry GEMV test
+# CPP_THREAD_SAFETY_GEMV = 1
 
 
 # If you want to enable the experimental BFLOAT16 support
-# BUILD_HALF = 1
-#
-# Select if you need to build only select types
-# BUILD_SINGLE = 1
-# BUILD_DOUBLE = 1
-# BUILD_COMPLEX = 1
-# BUILD_COMPLEX16 = 1
-#
-#
+# BUILD_BFLOAT16 = 1
+
+
+# Set the thread number threshold beyond which the job array for the threaded level3 BLAS
+# will be allocated on the heap rather than the stack. (This array alone requires 
+# NUM_THREADS*NUM_THREADS*128 bytes of memory so should not pose a problem at low cpu
+# counts, but obviously it is not the only item that ends up on the stack.
+# The default value of 32 ensures that the overall requirement is compatible
+# with the default 1MB stacksize imposed by having the Java VM loaded without use
+# of its -Xss parameter.
+# The value of 160 formerly used from about version 0.2.7 until 0.3.10 is easily compatible
+# with the common Linux stacksize of 8MB but will cause crashes with unwary use of the java
+# VM e.g. in Octave or with the java-based libhdfs in numpy or scipy code
+# BLAS3_MEM_ALLOC_THRESHOLD = 160
+
+
+
+# the below is not yet configurable, use cmake if you need to build only select types
+BUILD_SINGLE = 1
+BUILD_DOUBLE = 1
+BUILD_COMPLEX = 1
+BUILD_COMPLEX16 = 1
 #  End of user configuration
 #
diff --git a/Makefile.system b/Makefile.system
index eb6e14a98..461f7370b 100644
--- a/Makefile.system
+++ b/Makefile.system
@@ -1232,8 +1232,8 @@ ifeq ($(USE_TLS), 1)
 CCOMMON_OPT += -DUSE_TLS
 endif
 
-ifeq ($(BUILD_HALF), 1)
-CCOMMON_OPT += -DBUILD_HALF
+ifeq ($(BUILD_BFLOAT16), 1)
+CCOMMON_OPT += -DBUILD_BFLOAT16
 endif
 ifeq ($(BUILD_SINGLE), 1)
 CCOMMON_OPT += -DBUILD_SINGLE=1
@@ -1521,10 +1521,10 @@ export KERNELDIR
 export FUNCTION_PROFILE
 export TARGET_CORE
 export NO_AVX512
-export BUILD_HALF
+export BUILD_BFLOAT16
 
-export SHGEMM_UNROLL_M
-export SHGEMM_UNROLL_N
+export SBGEMM_UNROLL_M
+export SBGEMM_UNROLL_N
 export SGEMM_UNROLL_M
 export SGEMM_UNROLL_N
 export DGEMM_UNROLL_M
diff --git a/Makefile.tail b/Makefile.tail
index 641082450..b14689fc7 100644
--- a/Makefile.tail
+++ b/Makefile.tail
@@ -24,14 +24,14 @@ BLASOBJS   += $(QBLASOBJS)   $(XBLASOBJS)
 BLASOBJS_P += $(QBLASOBJS_P) $(XBLASOBJS_P)
 endif
 
-$(SHBLASOBJS) $(SHBLASOBJS_P) : override CFLAGS += -DHALF -UDOUBLE  -UCOMPLEX
+$(SHBLASOBJS) $(SHBLASOBJS_P) : override CFLAGS += -DBFLOAT16 -UDOUBLE  -UCOMPLEX
 $(SBLASOBJS) $(SBLASOBJS_P) : override CFLAGS += -UDOUBLE  -UCOMPLEX
 $(DBLASOBJS) $(DBLASOBJS_P) : override CFLAGS += -DDOUBLE  -UCOMPLEX
 $(QBLASOBJS) $(QBLASOBJS_P) : override CFLAGS += -DXDOUBLE -UCOMPLEX
 $(CBLASOBJS) $(CBLASOBJS_P) : override CFLAGS += -UDOUBLE  -DCOMPLEX
 $(ZBLASOBJS) $(ZBLASOBJS_P) : override CFLAGS += -DDOUBLE  -DCOMPLEX
 $(XBLASOBJS) $(XBLASOBJS_P) : override CFLAGS += -DXDOUBLE -DCOMPLEX
-$(SHEXTOBJS) $(SHEXTOBJS_P) : override CFLAGS += -DHALF -UDOUBLE  -UCOMPLEX
+$(SHEXTOBJS) $(SHEXTOBJS_P) : override CFLAGS += -DBFLOAT16 -UDOUBLE  -UCOMPLEX
 
 $(SHBLASOBJS_P) : override CFLAGS += -DPROFILE $(COMMON_PROF)
 $(SBLASOBJS_P) : override CFLAGS += -DPROFILE $(COMMON_PROF)