Allow building support for only a subset of variable types

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 954c053..f43e0e0 100644 (file)
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -29,8 +29,10 @@ option(NO_AFFINITY "Disable support for CPU affinity masks to avoid binding proc
 else()
 set(NO_AFFINITY 1)
 endif()
-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)
+option(BUILD_SINGLE "Single precision" OFF)
+option(BUILD_DOUBLE "Double precision" OFF)
+option(BUILD_COMPLEX "Single precision" OFF)
+option(BUILD_COMPLEX16 "Single precision" OFF)
 
 # Add a prefix or suffix to all exported symbol names in the shared library.
 # Avoids conflicts with other BLAS libraries, especially when using
@@ -108,28 +110,33 @@ endif()
 
 set(FLOAT_TYPES "")
 if (BUILD_SINGLE)
-  message(STATUS "Building Single Precision")
-  list(APPEND FLOAT_TYPES "SINGLE") # defines nothing
+  message(STATUS "Building Songle Precision")
+  list(APPEND FLOAT_TYPES "SINGLE") 
+  #    set(CCOMMON_OPT "${CCOMMON_OPT} -DBUILD_SINGLE=1")
 endif ()
 
 if (BUILD_DOUBLE)
   message(STATUS "Building Double Precision")
-  list(APPEND FLOAT_TYPES "DOUBLE") # defines DOUBLE
+  list(APPEND FLOAT_TYPES "DOUBLE")
+  #set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -DBUILD_DOUBLE=1")
 endif ()
 
 if (BUILD_COMPLEX)
   message(STATUS "Building Complex Precision")
-  list(APPEND FLOAT_TYPES "COMPLEX") # defines COMPLEX
-endif ()
+  list(APPEND FLOAT_TYPES "COMPLEX")
+  #set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -DBUILD_COMPLEX=1")
+endif () 
 
 if (BUILD_COMPLEX16)
   message(STATUS "Building Double Complex Precision")
-  list(APPEND FLOAT_TYPES "ZCOMPLEX") # defines COMPLEX and DOUBLE
+  list(APPEND FLOAT_TYPES "ZCOMPLEX") 
+  #set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -DBUILD_COMPLEX16=1")
 endif ()
 
 if (BUILD_HALF)
   message(STATUS "Building Half Precision")
-  list(APPEND FLOAT_TYPES "HALF") # defines nothing
+  list(APPEND FLOAT_TYPES "HALF")
+  set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -DBUILD_HALF")
 endif ()
 
 if (NOT DEFINED CORE OR "${CORE}" STREQUAL "UNKNOWN")
@@ -236,9 +243,6 @@ 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 b/Makefile
index 93e8af2..a9af62a 100644 (file)
--- a/Makefile
+++ b/Makefile
@@ -146,9 +146,6 @@ ifneq ($(NO_CBLAS), 1)
 ifeq ($(CPP_THREAD_SAFETY_TEST), 1)
        $(MAKE) -C cpp_thread_test all
 endif
-ifeq ($(CPP_THREAD_SAFETY_GEMV), 1)
-       $(MAKE) -C cpp_thread_test dgemv_tester
-endif
 endif
 endif
 
@@ -305,6 +302,18 @@ endif
 ifeq ($(BUILD_LAPACK_DEPRECATED), 1)
        -@echo "BUILD_DEPRECATED      = 1" >> $(NETLIB_LAPACK_DIR)/make.inc
 endif
+ifeq ($(BUILD_SINGLE), 1)
+       -@echo "BUILD_SINGLE      = 1" >> $(NETLIB_LAPACK_DIR)/make.inc
+endif
+ifeq ($(BUILD_DOUBLE), 1)
+       -@echo "BUILD_DOUBLE      = 1" >> $(NETLIB_LAPACK_DIR)/make.inc
+endif
+ifeq ($(BUILD_COMPLEX), 1)
+       -@echo "BUILD_COMPLEX      = 1" >> $(NETLIB_LAPACK_DIR)/make.inc
+endif
+ifeq ($(BUILD_COMPLEX16), 1)
+       -@echo "BUILD_COMPLEX16      = 1" >> $(NETLIB_LAPACK_DIR)/make.inc
+endif
        -@echo "LAPACKE_WITH_TMG      = 1" >> $(NETLIB_LAPACK_DIR)/make.inc
        -@cat  make.inc >> $(NETLIB_LAPACK_DIR)/make.inc
 endif

diff --git a/Makefile.rule b/Makefile.rule
index 635e02c..09dfb08 100644 (file)
--- a/Makefile.rule
+++ b/Makefile.rule
@@ -272,33 +272,17 @@ 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
-
-
-# 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
+#
+# Select 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.tail b/Makefile.tail
index cfc4a36..6410824 100644 (file)
--- a/Makefile.tail
+++ b/Makefile.tail
@@ -11,8 +11,8 @@ COMMONOBJS_P = $(COMMONOBJS:.$(SUFFIX)=.$(PSUFFIX))
 
 HPLOBJS_P   = $(HPLOBJS:.$(SUFFIX)=.$(PSUFFIX))
 
-BLASOBJS    = $(SHEXTOBJS) $(SHBLASOBJS)  $(SBLASOBJS)   $(DBLASOBJS)   $(CBLASOBJS)   $(ZBLASOBJS)
-BLASOBJS_P  = $(SHEXTOBJS_P) $(SHBLASOBJS_P) $(SBLASOBJS_P) $(DBLASOBJS_P) $(CBLASOBJS_P) $(ZBLASOBJS_P)
+BLASOBJS    = $(SHEXTOBJS) $(SHBLASOBJS)  $(SBLASOBJS)   $(DBLASOBJS)   $(CBLASOBJS)   $(ZBLASOBJS) $(CBAUXOBJS)
+BLASOBJS_P  = $(SHEXTOBJS_P) $(SHBLASOBJS_P) $(SBLASOBJS_P) $(DBLASOBJS_P) $(CBLASOBJS_P) $(ZBLASOBJS_P) $(CBAUXOBJS_P)
 
 ifdef EXPRECISION
 BLASOBJS   += $(QBLASOBJS)   $(XBLASOBJS)

diff --git a/common_param.h b/common_param.h
index a52de98..81b479e 100644 (file)
--- a/common_param.h
+++ b/common_param.h
@@ -146,40 +146,56 @@ BLASLONG (*ishmin_k) (BLASLONG, float *, BLASLONG);
   int    (*shlaswp_ncopy) (BLASLONG, BLASLONG, BLASLONG, float *, BLASLONG, blasint *, float *);
 
 #endif
+
+#if (BUILD_SINGLE) || (BUILD_DOUBLE) || (BUILD_COMPLEX) || (BUILD_COMPLEX16)
   int sgemm_p, sgemm_q, sgemm_r;
   int sgemm_unroll_m, sgemm_unroll_n, sgemm_unroll_mn;
+#endif
 
   int exclusive_cache;
 
+#if (BUILD_SINGLE) || (BUILD_DOUBLE) || (BUILD_COMPLEX)
   float  (*samax_k) (BLASLONG, float *, BLASLONG);
   float  (*samin_k) (BLASLONG, float *, BLASLONG);
   float  (*smax_k)  (BLASLONG, float *, BLASLONG);
   float  (*smin_k)  (BLASLONG, float *, BLASLONG);
+
 BLASLONG (*isamax_k)(BLASLONG, float *, BLASLONG);
 BLASLONG (*isamin_k)(BLASLONG, float *, BLASLONG);
 BLASLONG (*ismax_k) (BLASLONG, float *, BLASLONG);
 BLASLONG (*ismin_k) (BLASLONG, float *, BLASLONG);
-
   float  (*snrm2_k) (BLASLONG, float *, BLASLONG);
   float  (*sasum_k) (BLASLONG, float *, BLASLONG);
+#endif
+#if BUILD_SINGLE
   float  (*ssum_k)  (BLASLONG, float *, BLASLONG);
+#endif  
+#if (BUILD_SINGLE) || (BUILD_DOUBLE) || (BUILD_COMPLEX)
   int    (*scopy_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
   float  (*sdot_k)  (BLASLONG, float *, BLASLONG, float *, BLASLONG);
-  double (*dsdot_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
+  //double (*dsdot_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
 
   int    (*srot_k)  (BLASLONG, float *, BLASLONG, float *, BLASLONG, float, float);
 
   int    (*saxpy_k) (BLASLONG, BLASLONG, BLASLONG, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG);
+#endif
+#if (BUILD_SINGLE) || (BUILD_DOUBLE) || (BUILD_COMPLEX) || (BUILD_COMPLEX16)
   int    (*sscal_k) (BLASLONG, BLASLONG, BLASLONG, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG);
+#endif
+#if (BUILD_SINGLE) || (BUILD_DOUBLE) || (BUILD_COMPLEX)
   int    (*sswap_k) (BLASLONG, BLASLONG, BLASLONG, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG);
 
   int    (*sgemv_n) (BLASLONG, BLASLONG, BLASLONG, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
   int    (*sgemv_t) (BLASLONG, BLASLONG, BLASLONG, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
+#endif
+#if BUILD_SINGLE  
   int    (*sger_k)  (BLASLONG, BLASLONG, BLASLONG, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
 
   int    (*ssymv_L) (BLASLONG, BLASLONG, float,  float  *, BLASLONG, float  *, BLASLONG, float  *, BLASLONG, float *);
   int    (*ssymv_U) (BLASLONG, BLASLONG, float,  float  *, BLASLONG, float  *, BLASLONG, float  *, BLASLONG, float *);
+#endif
 
+#if (BUILD_SINGLE) || (BUILD_DOUBLE) || (BUILD_COMPLEX)
 #ifdef ARCH_X86_64
   void (*sgemm_direct) (BLASLONG, BLASLONG, BLASLONG, float *, BLASLONG , float *, BLASLONG , float * , BLASLONG);
   int  (*sgemm_direct_performant) (BLASLONG M, BLASLONG N, BLASLONG K);
@@ -193,7 +209,8 @@ BLASLONG (*ismin_k) (BLASLONG, float *, BLASLONG);
   int    (*sgemm_itcopy   )(BLASLONG, BLASLONG, float *, BLASLONG, float *);
   int    (*sgemm_oncopy   )(BLASLONG, BLASLONG, float *, BLASLONG, float *);
   int    (*sgemm_otcopy   )(BLASLONG, BLASLONG, float *, BLASLONG, float *);
-
+#endif
+#if (BUILD_SINGLE) || (BUILD_DOUBLE)  
   int    (*strsm_kernel_LN)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
   int    (*strsm_kernel_LT)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
   int    (*strsm_kernel_RN)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
@@ -215,7 +232,8 @@ BLASLONG (*ismin_k) (BLASLONG, float *, BLASLONG);
   int    (*strsm_olnncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
   int    (*strsm_oltucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
   int    (*strsm_oltncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
-
+#endif
+#if BUILD_SINGLE
   int    (*strmm_kernel_RN)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
   int    (*strmm_kernel_RT)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
   int    (*strmm_kernel_LN)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
@@ -242,13 +260,18 @@ BLASLONG (*ismin_k) (BLASLONG, float *, BLASLONG);
   int    (*ssymm_iltcopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
   int    (*ssymm_outcopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
   int    (*ssymm_oltcopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
-
+#endif
+#if (BUILD_SINGLE) || (BUILD_DOUBLE)
   int   (*sneg_tcopy)   (BLASLONG, BLASLONG, float *, BLASLONG, float *);
   int    (*slaswp_ncopy) (BLASLONG, BLASLONG, BLASLONG, float *, BLASLONG, blasint *, float *);
+#endif
 
+#if (BUILD_DOUBLE) || (BUILD_COMPLEX16)
   int dgemm_p, dgemm_q, dgemm_r;
   int dgemm_unroll_m, dgemm_unroll_n, dgemm_unroll_mn;
+#endif
 
+#if (BUILD_DOUBLE) || (BUILD_COMPLEX16)
   double (*damax_k) (BLASLONG, double *, BLASLONG);
   double (*damin_k) (BLASLONG, double *, BLASLONG);
   double (*dmax_k)  (BLASLONG, double *, BLASLONG);
@@ -257,25 +280,37 @@ BLASLONG (*idamax_k)(BLASLONG, double *, BLASLONG);
 BLASLONG (*idamin_k)(BLASLONG, double *, BLASLONG);
 BLASLONG (*idmax_k) (BLASLONG, double *, BLASLONG);
 BLASLONG (*idmin_k) (BLASLONG, double *, BLASLONG);
+#endif
 
+#if (BUILD_DOUBLE) || (BUILD_COMPLEX16)
   double (*dnrm2_k) (BLASLONG, double *, BLASLONG);
   double (*dasum_k) (BLASLONG, double *, BLASLONG);
+#endif
+#if BUILD_DOUBLE
   double (*dsum_k)  (BLASLONG, double *, BLASLONG);
+#endif
+#if (BUILD_DOUBLE) || (BUILD_COMPLEX16)
   int    (*dcopy_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG);
   double (*ddot_k)  (BLASLONG, double *, BLASLONG, double *, BLASLONG);
+#endif
+#if  (BUILD_SINGLE) || (BUILD_DOUBLE)
+  double (*dsdot_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
+#endif
+#if (BUILD_DOUBLE) || (BUILD_COMPLEX16)
   int    (*drot_k)  (BLASLONG, double *, BLASLONG, double *, BLASLONG, double, double);
-
   int    (*daxpy_k) (BLASLONG, BLASLONG, BLASLONG, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG);
   int    (*dscal_k) (BLASLONG, BLASLONG, BLASLONG, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG);
   int    (*dswap_k) (BLASLONG, BLASLONG, BLASLONG, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG);
-
   int    (*dgemv_n) (BLASLONG, BLASLONG, BLASLONG, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG, double *);
   int    (*dgemv_t) (BLASLONG, BLASLONG, BLASLONG, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG, double *);
+#endif
+#if BUILD_DOUBLE
   int    (*dger_k)  (BLASLONG, BLASLONG, BLASLONG, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG, double *);
 
   int    (*dsymv_L) (BLASLONG, BLASLONG, double,  double  *, BLASLONG, double  *, BLASLONG, double  *, BLASLONG, double *);
   int    (*dsymv_U) (BLASLONG, BLASLONG, double,  double  *, BLASLONG, double  *, BLASLONG, double  *, BLASLONG, double *);
-
+#endif
+#if (BUILD_DOUBLE) || (BUILD_COMPLEX16)
   int    (*dgemm_kernel   )(BLASLONG, BLASLONG, BLASLONG, double, double *, double *, double *, BLASLONG);
   int    (*dgemm_beta     )(BLASLONG, BLASLONG, BLASLONG, double, double *, BLASLONG, double *, BLASLONG, double  *, BLASLONG);
 
@@ -283,7 +318,8 @@ BLASLONG (*idmin_k) (BLASLONG, double *, BLASLONG);
   int    (*dgemm_itcopy   )(BLASLONG, BLASLONG, double *, BLASLONG, double *);
   int    (*dgemm_oncopy   )(BLASLONG, BLASLONG, double *, BLASLONG, double *);
   int    (*dgemm_otcopy   )(BLASLONG, BLASLONG, double *, BLASLONG, double *);
-
+#endif
+#if BUILD_DOUBLE
   int    (*dtrsm_kernel_LN)(BLASLONG, BLASLONG, BLASLONG, double, double *, double *, double *, BLASLONG, BLASLONG);
   int    (*dtrsm_kernel_LT)(BLASLONG, BLASLONG, BLASLONG, double, double *, double *, double *, BLASLONG, BLASLONG);
   int    (*dtrsm_kernel_RN)(BLASLONG, BLASLONG, BLASLONG, double, double *, double *, double *, BLASLONG, BLASLONG);
@@ -335,7 +371,7 @@ BLASLONG (*idmin_k) (BLASLONG, double *, BLASLONG);
 
   int   (*dneg_tcopy)   (BLASLONG, BLASLONG, double *, BLASLONG, double *);
   int    (*dlaswp_ncopy) (BLASLONG, BLASLONG, BLASLONG, double *, BLASLONG, blasint *, double *);
-
+#endif
 #ifdef EXPRECISION
 
   int qgemm_p, qgemm_q, qgemm_r;
@@ -430,22 +466,29 @@ BLASLONG (*iqmin_k) (BLASLONG, xdouble *, BLASLONG);
 
 #endif
 
+#if (BUILD_COMPLEX) || (BUILD_COMPLEX16)
   int cgemm_p, cgemm_q, cgemm_r;
   int cgemm_unroll_m, cgemm_unroll_n, cgemm_unroll_mn;
-
   float (*camax_k) (BLASLONG, float *, BLASLONG);
   float (*camin_k) (BLASLONG, float *, BLASLONG);
 BLASLONG (*icamax_k)(BLASLONG, float *, BLASLONG);
 BLASLONG (*icamin_k)(BLASLONG, float *, BLASLONG);
+#endif
+#if BUILD_COMPLEX
 
   float (*cnrm2_k) (BLASLONG, float *, BLASLONG);
   float (*casum_k) (BLASLONG, float *, BLASLONG);
   float (*csum_k)  (BLASLONG, float *, BLASLONG);
+#endif
+#if (BUILD_COMPLEX)|| (BUILD_COMPLEX16)
   int    (*ccopy_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
   openblas_complex_float (*cdotu_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
   openblas_complex_float (*cdotc_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
+#endif
+#if BUILD_COMPLEX
   int    (*csrot_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG, float, float);
-
+#endif
+#if (BUILD_COMPLEX)|| (BUILD_COMPLEX16)
   int    (*caxpy_k) (BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG);
   int    (*caxpyc_k)(BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG);
   int    (*cscal_k) (BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG);
@@ -459,6 +502,8 @@ BLASLONG (*icamin_k)(BLASLONG, float *, BLASLONG);
   int    (*cgemv_u) (BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
   int    (*cgemv_s) (BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
   int    (*cgemv_d) (BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
+#endif
+#if (BUILD_COMPLEX)
   int    (*cgeru_k) (BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
   int    (*cgerc_k) (BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
   int    (*cgerv_k) (BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
@@ -470,13 +515,14 @@ BLASLONG (*icamin_k)(BLASLONG, float *, BLASLONG);
   int    (*chemv_U) (BLASLONG, BLASLONG, float,  float, float  *, BLASLONG, float  *, BLASLONG, float  *, BLASLONG, float *);
   int    (*chemv_M) (BLASLONG, BLASLONG, float,  float, float  *, BLASLONG, float  *, BLASLONG, float  *, BLASLONG, float *);
   int    (*chemv_V) (BLASLONG, BLASLONG, float,  float, float  *, BLASLONG, float  *, BLASLONG, float  *, BLASLONG, float *);
+#endif
+#if (BUILD_COMPLEX) || (BUILD_COMPLEX16)
 
   int    (*cgemm_kernel_n )(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG);
   int    (*cgemm_kernel_l )(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG);
   int    (*cgemm_kernel_r )(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG);
   int    (*cgemm_kernel_b )(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG);
   int    (*cgemm_beta     )(BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float  *, BLASLONG);
-
   int    (*cgemm_incopy   )(BLASLONG, BLASLONG, float *, BLASLONG, float *);
   int    (*cgemm_itcopy   )(BLASLONG, BLASLONG, float *, BLASLONG, float *);
   int    (*cgemm_oncopy   )(BLASLONG, BLASLONG, float *, BLASLONG, float *);
@@ -507,6 +553,8 @@ BLASLONG (*icamin_k)(BLASLONG, float *, BLASLONG);
   int    (*ctrsm_olnncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
   int    (*ctrsm_oltucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
   int    (*ctrsm_oltncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+#endif
+#if (BUILD_COMPLEX)
 
   int    (*ctrmm_kernel_RN)(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG, BLASLONG);
   int    (*ctrmm_kernel_RT)(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG, BLASLONG);
@@ -590,10 +638,13 @@ BLASLONG (*icamin_k)(BLASLONG, float *, BLASLONG);
   int    (*chemm3m_olcopyr)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float, float, float *);
   int    (*chemm3m_oucopyi)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float, float, float *);
   int    (*chemm3m_olcopyi)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float, float, float *);
-
+#endif  
+#if (BUILD_COMPLEX) || (BUILD_COMPLEX16)
   int   (*cneg_tcopy)   (BLASLONG, BLASLONG, float *, BLASLONG, float *);
   int    (*claswp_ncopy) (BLASLONG, BLASLONG, BLASLONG, float *, BLASLONG, blasint *, float *);
+#endif
 
+#if BUILD_COMPLEX16
   int zgemm_p, zgemm_q, zgemm_r;
   int zgemm_unroll_m, zgemm_unroll_n, zgemm_unroll_mn;
 
@@ -757,6 +808,7 @@ BLASLONG (*izamin_k)(BLASLONG, double *, BLASLONG);
 
   int   (*zneg_tcopy)   (BLASLONG, BLASLONG, double *, BLASLONG, double *);
   int    (*zlaswp_ncopy) (BLASLONG, BLASLONG, BLASLONG, double *, BLASLONG, blasint *, double *);
+#endif
 
 #ifdef EXPRECISION
 
@@ -930,22 +982,34 @@ BLASLONG (*ixamin_k)(BLASLONG, xdouble *, BLASLONG);
   void (*init)(void);
 
   int snum_opt, dnum_opt, qnum_opt;
-
+#if BUILD_SINGLE
   int    (*saxpby_k)     (BLASLONG, float, float*, BLASLONG,float, float*, BLASLONG);
+#endif
+#if BUILD_DOUBLE
   int    (*daxpby_k)     (BLASLONG, double, double*, BLASLONG,double, double*, BLASLONG);
+#endif
+#if BUILD_COMPLEX
   int    (*caxpby_k)     (BLASLONG, float, float,  float*, BLASLONG,float,float, float*, BLASLONG);
+#endif
+#if BUILD_COMPLEX16
   int    (*zaxpby_k)     (BLASLONG, double, double,  double*, BLASLONG,double,double, double*, BLASLONG);
+#endif
 
+#if BUILD_SINGLE  
   int    (*somatcopy_k_cn)     (BLASLONG, BLASLONG, float, float*, BLASLONG, float*, BLASLONG);
   int    (*somatcopy_k_ct)     (BLASLONG, BLASLONG, float, float*, BLASLONG, float*, BLASLONG);
   int    (*somatcopy_k_rn)     (BLASLONG, BLASLONG, float, float*, BLASLONG, float*, BLASLONG);
   int    (*somatcopy_k_rt)     (BLASLONG, BLASLONG, float, float*, BLASLONG, float*, BLASLONG);
+#endif
 
+#if BUILD_DOUBLE 
   int    (*domatcopy_k_cn)     (BLASLONG, BLASLONG, double, double*, BLASLONG, double*, BLASLONG);
   int    (*domatcopy_k_ct)     (BLASLONG, BLASLONG, double, double*, BLASLONG, double*, BLASLONG);
   int    (*domatcopy_k_rn)     (BLASLONG, BLASLONG, double, double*, BLASLONG, double*, BLASLONG);
   int    (*domatcopy_k_rt)     (BLASLONG, BLASLONG, double, double*, BLASLONG, double*, BLASLONG);
+#endif
 
+#if BUILD_COMPLEX
   int    (*comatcopy_k_cn)     (BLASLONG, BLASLONG, float, float, float*, BLASLONG, float*, BLASLONG);
   int    (*comatcopy_k_ct)     (BLASLONG, BLASLONG, float, float, float*, BLASLONG, float*, BLASLONG);
   int    (*comatcopy_k_rn)     (BLASLONG, BLASLONG, float, float, float*, BLASLONG, float*, BLASLONG);
@@ -955,7 +1019,9 @@ BLASLONG (*ixamin_k)(BLASLONG, xdouble *, BLASLONG);
   int    (*comatcopy_k_ctc)    (BLASLONG, BLASLONG, float, float, float*, BLASLONG, float*, BLASLONG);
   int    (*comatcopy_k_rnc)    (BLASLONG, BLASLONG, float, float, float*, BLASLONG, float*, BLASLONG);
   int    (*comatcopy_k_rtc)    (BLASLONG, BLASLONG, float, float, float*, BLASLONG, float*, BLASLONG);
+#endif
 
+#if BUILD_COMPLEX16
   int    (*zomatcopy_k_cn)     (BLASLONG, BLASLONG, double, double, double*, BLASLONG, double*, BLASLONG);
   int    (*zomatcopy_k_ct)     (BLASLONG, BLASLONG, double, double, double*, BLASLONG, double*, BLASLONG);
   int    (*zomatcopy_k_rn)     (BLASLONG, BLASLONG, double, double, double*, BLASLONG, double*, BLASLONG);
@@ -965,17 +1031,23 @@ BLASLONG (*ixamin_k)(BLASLONG, xdouble *, BLASLONG);
   int    (*zomatcopy_k_ctc)    (BLASLONG, BLASLONG, double, double, double*, BLASLONG, double*, BLASLONG);
   int    (*zomatcopy_k_rnc)    (BLASLONG, BLASLONG, double, double, double*, BLASLONG, double*, BLASLONG);
   int    (*zomatcopy_k_rtc)    (BLASLONG, BLASLONG, double, double, double*, BLASLONG, double*, BLASLONG);
+#endif
 
+#if BUILD_SINGLE
   int    (*simatcopy_k_cn)     (BLASLONG, BLASLONG, float, float*, BLASLONG);
   int    (*simatcopy_k_ct)     (BLASLONG, BLASLONG, float, float*, BLASLONG);
   int    (*simatcopy_k_rn)     (BLASLONG, BLASLONG, float, float*, BLASLONG);
   int    (*simatcopy_k_rt)     (BLASLONG, BLASLONG, float, float*, BLASLONG);
+#endif
 
+#if BUILD_DOUBLE
   int    (*dimatcopy_k_cn)     (BLASLONG, BLASLONG, double, double*, BLASLONG);
   int    (*dimatcopy_k_ct)     (BLASLONG, BLASLONG, double, double*, BLASLONG);
   int    (*dimatcopy_k_rn)     (BLASLONG, BLASLONG, double, double*, BLASLONG);
   int    (*dimatcopy_k_rt)     (BLASLONG, BLASLONG, double, double*, BLASLONG);
+#endif
 
+#if BUILD_COMPLEX
   int    (*cimatcopy_k_cn)     (BLASLONG, BLASLONG, float, float, float*, BLASLONG);
   int    (*cimatcopy_k_ct)     (BLASLONG, BLASLONG, float, float, float*, BLASLONG);
   int    (*cimatcopy_k_rn)     (BLASLONG, BLASLONG, float, float, float*, BLASLONG);
@@ -985,7 +1057,9 @@ BLASLONG (*ixamin_k)(BLASLONG, xdouble *, BLASLONG);
   int    (*cimatcopy_k_ctc)    (BLASLONG, BLASLONG, float, float, float*, BLASLONG);
   int    (*cimatcopy_k_rnc)    (BLASLONG, BLASLONG, float, float, float*, BLASLONG);
   int    (*cimatcopy_k_rtc)    (BLASLONG, BLASLONG, float, float, float*, BLASLONG);
+#endif
 
+#if BUILD_COMPLEX16
   int    (*zimatcopy_k_cn)     (BLASLONG, BLASLONG, double, double, double*, BLASLONG);
   int    (*zimatcopy_k_ct)     (BLASLONG, BLASLONG, double, double, double*, BLASLONG);
   int    (*zimatcopy_k_rn)     (BLASLONG, BLASLONG, double, double, double*, BLASLONG);
@@ -995,12 +1069,20 @@ BLASLONG (*ixamin_k)(BLASLONG, xdouble *, BLASLONG);
   int    (*zimatcopy_k_ctc)    (BLASLONG, BLASLONG, double, double, double*, BLASLONG);
   int    (*zimatcopy_k_rnc)    (BLASLONG, BLASLONG, double, double, double*, BLASLONG);
   int    (*zimatcopy_k_rtc)    (BLASLONG, BLASLONG, double, double, double*, BLASLONG);
+#endif
 
+#if BUILD_SINGLE
   int    (*sgeadd_k) (BLASLONG, BLASLONG, float, float *, BLASLONG, float, float *, BLASLONG); 
+#endif
+#if BUILD_DOUBLE
   int    (*dgeadd_k) (BLASLONG, BLASLONG, double, double *, BLASLONG, double, double *, BLASLONG); 
+#endif
+#if BUILD_COMPLEX
   int    (*cgeadd_k) (BLASLONG, BLASLONG, float, float,  float *,  BLASLONG, float, float, float *, BLASLONG); 
+#endif
+#if BUILD_COMPLEX16
   int    (*zgeadd_k) (BLASLONG, BLASLONG, double, double, double *, BLASLONG, double, double, double *, BLASLONG); 
-
+#endif
 } gotoblas_t;
 
 extern gotoblas_t *gotoblas;
@@ -1021,19 +1103,31 @@ extern gotoblas_t *gotoblas;
 #define        SHGEMM_UNROLL_MN        gotoblas -> shgemm_unroll_mn
 #endif
 
+#if  (BUILD_SINGLE)
 #define        SGEMM_P         gotoblas -> sgemm_p
 #define        SGEMM_Q         gotoblas -> sgemm_q
 #define        SGEMM_R         gotoblas -> sgemm_r
 #define        SGEMM_UNROLL_M  gotoblas -> sgemm_unroll_m
 #define        SGEMM_UNROLL_N  gotoblas -> sgemm_unroll_n
 #define SGEMM_UNROLL_MN        gotoblas -> sgemm_unroll_mn
+#endif
 
+#if  (BUILD_DOUBLE)
 #define        DGEMM_P         gotoblas -> dgemm_p
 #define        DGEMM_Q         gotoblas -> dgemm_q
 #define        DGEMM_R         gotoblas -> dgemm_r
 #define        DGEMM_UNROLL_M  gotoblas -> dgemm_unroll_m
 #define        DGEMM_UNROLL_N  gotoblas -> dgemm_unroll_n
 #define DGEMM_UNROLL_MN        gotoblas -> dgemm_unroll_mn
+#if ! (BUILD_SINGLE)
+#define        SGEMM_P         gotoblas -> sgemm_p
+#define        SGEMM_Q         gotoblas -> sgemm_q
+#define        SGEMM_R         gotoblas -> sgemm_r
+#define        SGEMM_UNROLL_M  gotoblas -> sgemm_unroll_m
+#define        SGEMM_UNROLL_N  gotoblas -> sgemm_unroll_n
+#define SGEMM_UNROLL_MN        gotoblas -> sgemm_unroll_mn
+#endif
+#endif
 
 #define        QGEMM_P         gotoblas -> qgemm_p
 #define        QGEMM_Q         gotoblas -> qgemm_q
@@ -1042,19 +1136,47 @@ extern gotoblas_t *gotoblas;
 #define        QGEMM_UNROLL_N  gotoblas -> qgemm_unroll_n
 #define QGEMM_UNROLL_MN        gotoblas -> qgemm_unroll_mn
 
+#if BUILD_COMPLEX
 #define        CGEMM_P         gotoblas -> cgemm_p
 #define        CGEMM_Q         gotoblas -> cgemm_q
 #define        CGEMM_R         gotoblas -> cgemm_r
 #define        CGEMM_UNROLL_M  gotoblas -> cgemm_unroll_m
 #define        CGEMM_UNROLL_N  gotoblas -> cgemm_unroll_n
 #define CGEMM_UNROLL_MN        gotoblas -> cgemm_unroll_mn
+#ifndef BUILD_SINGLE
+#define        SGEMM_P         gotoblas -> sgemm_p
+#define        SGEMM_Q         gotoblas -> sgemm_q
+#define        SGEMM_R         1024
+#define        SGEMM_UNROLL_M  gotoblas -> sgemm_unroll_m
+#define        SGEMM_UNROLL_N  gotoblas -> sgemm_unroll_n
+#define SGEMM_UNROLL_MN        gotoblas -> sgemm_unroll_mn
+#endif
+#endif
 
+#if BUILD_COMPLEX16
 #define        ZGEMM_P         gotoblas -> zgemm_p
 #define        ZGEMM_Q         gotoblas -> zgemm_q
 #define        ZGEMM_R         gotoblas -> zgemm_r
 #define        ZGEMM_UNROLL_M  gotoblas -> zgemm_unroll_m
 #define        ZGEMM_UNROLL_N  gotoblas -> zgemm_unroll_n
 #define ZGEMM_UNROLL_MN        gotoblas -> zgemm_unroll_mn
+#ifndef BUILD_DOUBLE
+#define        DGEMM_P         gotoblas -> dgemm_p
+#define        DGEMM_Q         gotoblas -> dgemm_q
+#define        DGEMM_R         1024
+#define        DGEMM_UNROLL_M  gotoblas -> dgemm_unroll_m
+#define        DGEMM_UNROLL_N  gotoblas -> dgemm_unroll_n
+#define DGEMM_UNROLL_MN        gotoblas -> dgemm_unroll_mn
+#endif
+#ifndef BUILD_COMPLEX
+#define        CGEMM_P         gotoblas -> cgemm_p
+#define        CGEMM_Q         gotoblas -> cgemm_q
+#define        CGEMM_R         gotoblas -> cgemm_r
+#define        CGEMM_UNROLL_M  gotoblas -> cgemm_unroll_m
+#define        CGEMM_UNROLL_N  gotoblas -> cgemm_unroll_n
+#define CGEMM_UNROLL_MN        gotoblas -> cgemm_unroll_mn
+#endif
+#endif
 
 #define        XGEMM_P         gotoblas -> xgemm_p
 #define        XGEMM_Q         gotoblas -> xgemm_q
@@ -1222,7 +1344,7 @@ extern gotoblas_t *gotoblas;
 #endif
 
 #ifndef COMPLEX
-#if   defined(XDOUBLE)
+#if   (XDOUBLE)
 #define GEMM_P                 QGEMM_P
 #define GEMM_Q                 QGEMM_Q
 #define GEMM_R                 QGEMM_R
@@ -1246,7 +1368,7 @@ extern gotoblas_t *gotoblas;
 #define GEMM_DEFAULT_R         DGEMM_DEFAULT_R
 #define GEMM_DEFAULT_UNROLL_M  DGEMM_DEFAULT_UNROLL_M
 #define GEMM_DEFAULT_UNROLL_N  DGEMM_DEFAULT_UNROLL_N
-#elif defined(HALF)
+#elif (HALF)
 #define GEMM_P                 SHGEMM_P
 #define GEMM_Q                 SHGEMM_Q
 #define GEMM_R                 SHGEMM_R
@@ -1272,7 +1394,7 @@ extern gotoblas_t *gotoblas;
 #define GEMM_DEFAULT_UNROLL_N  SGEMM_DEFAULT_UNROLL_N
 #endif
 #else
-#if   defined(XDOUBLE)
+#if   (XDOUBLE)
 #define GEMM_P                 XGEMM_P
 #define GEMM_Q                 XGEMM_Q
 #define GEMM_R                 XGEMM_R
@@ -1386,7 +1508,7 @@ extern gotoblas_t *gotoblas;
 #ifndef GEMM3M_P
 #ifdef XDOUBLE
 #define GEMM3M_P       XGEMM3M_P
-#elif defined(DOUBLE)
+#elif defined (DOUBLE)
 #define GEMM3M_P       ZGEMM3M_P
 #else
 #define GEMM3M_P       CGEMM3M_P
@@ -1396,7 +1518,7 @@ extern gotoblas_t *gotoblas;
 #ifndef GEMM3M_Q
 #ifdef XDOUBLE
 #define GEMM3M_Q       XGEMM3M_Q
-#elif defined(DOUBLE)
+#elif defined (DOUBLE)
 #define GEMM3M_Q       ZGEMM3M_Q
 #else
 #define GEMM3M_Q       CGEMM3M_Q
@@ -1406,7 +1528,7 @@ extern gotoblas_t *gotoblas;
 #ifndef GEMM3M_R
 #ifdef XDOUBLE
 #define GEMM3M_R       XGEMM3M_R
-#elif defined(DOUBLE)
+#elif defined (DOUBLE)
 #define GEMM3M_R       ZGEMM3M_R
 #else
 #define GEMM3M_R       CGEMM3M_R