add visibility=default to exported functions, visibility=hidden to most assembler

common.h

@@ -424,6 +424,29 @@ please https://github.com/xianyi/OpenBLAS/issues/246
 #define BLAS3_MEM_ALLOC_THRESHOLD 32 
 #endif
 
+/***
+ For GCC/Clang, always use -fvisibility=hidden. Then mark exported function
+ implementations with OPENBLAS_EXPORT. For MSVC, using `__declspec` onces makes
+ the default atrribute for any function in the entire shared object hidden
+ (observed behaviour, documentation source needed).
+**/
+#if defined (_WIN32) || defined (__CYGWIN__)
+#  if defined (__GNUC__)
+     /* GCC */
+#    define OPENBLAS_EXPORT __attribute__ ((dllexport))
+#    define OPENBLAS_IMPORT __attribute__ ((dllimport))
+#  else
+     /* MSVC */
+#    define OPENBLAS_EXPORT __declspec(dllexport)
+#    define OPENBLAS_IMPORT __declspec(dllimport)
+#  endif
+#else
+   /* All other platforms. */
+#  define OPENBLAS_EXPORT __attribute__ ((visibility ("default")))
+#  define OPENBLAS_IMPORT
+#endif
+
+
 #ifdef QUAD_PRECISION
 #include "common_quad.h"
 #endif

common_loongarch64.h

@@ -267,6 +267,14 @@ static inline int get_cpu_model(char *model_name) {
 #if defined(ASSEMBLER) && !defined(NEEDPARAM)
 
 #define PROLOGUE \
+    .text ;\
+    .align 5 ;\
+    .globl  REALNAME ;\
+	.hidden REALNAME ;\
+    .type   REALNAME, @function ;\
+REALNAME: ;\
+
+#define PROLOGUE_EXPORT \
     .text ;\
     .align 5 ;\
     .globl  REALNAME ;\

common_mips.h

@@ -76,6 +76,12 @@ static inline int blas_quickdivide(blasint x, blasint y){
 #define PROLOGUE \
 	.arm		 ;\
 	.global	REALNAME ;\
+	.hidden REALNAME ;\
+REALNAME:
+
+#define PROLOGUE_EXPORT \
+	.arm		 ;\
+	.global	REALNAME ;\
 REALNAME:
 
 #define EPILOGUE

common_mips64.h

@@ -206,6 +206,18 @@ static inline int blas_quickdivide(blasint x, blasint y){
 #endif
 
 #define PROLOGUE \
+	.text ;\
+	.set	ASSEMBLER_ARCH ;\
+	.align 5 ;\
+	.globl	REALNAME ;\
+	.hidden REALNAME ;\
+	.ent	REALNAME ;\
+	.type	REALNAME, @function ;\
+REALNAME: ;\
+	.set	noreorder ;\
+	.set	nomacro
+
+#define PROLOGUE_EXPORT \
 	.text ;\
 	.set	ASSEMBLER_ARCH ;\
 	.align 5 ;\

common_power.h

@@ -532,6 +532,13 @@ static inline int blas_quickdivide(blasint x, blasint y){
 #if defined(OS_LINUX) || defined(OS_FREEBSD)
 #ifndef __64BIT__
 #define PROLOGUE \
+	.section .text;\
+	.align 6;\
+	.globl	REALNAME;\
+	.hidden REALNAME ;\
+	.type	REALNAME, @function;\
+REALNAME:
+#define PROLOGUE_EXPORT \
 	.section .text;\
 	.align 6;\
 	.globl	REALNAME;\
@@ -541,6 +548,13 @@ static inline int blas_quickdivide(blasint x, blasint y){
 #else
 #if _CALL_ELF == 2
 #define PROLOGUE \
+	.section .text;\
+	.align 6;\
+	.globl	REALNAME;\
+	.hidden REALNAME ;\
+	.type	REALNAME, @function;\
+REALNAME:
+#define PROLOGUE_EXPORT \
 	.section .text;\
 	.align 6;\
 	.globl	REALNAME;\
@@ -549,6 +563,21 @@ static inline int blas_quickdivide(blasint x, blasint y){
 #define EPILOGUE	.size	REALNAME, .-REALNAME
 #else
 #define PROLOGUE \
+	.section .text;\
+	.align 5;\
+	.globl REALNAME;\
+	.hidden REALNAME ;\
+	.section	".opd","aw";\
+	.align 3;\
+REALNAME:;\
+	.quad	.REALNAME, .TOC.@tocbase, 0;\
+	.previous;\
+	.size	REALNAME, 24;\
+	.type	.REALNAME, @function;\
+	.globl	.REALNAME;\
+	.hidden .REALNAME ;\
+.REALNAME:
+#define PROLOGUE_EXPORT \
 	.section .text;\
 	.align 5;\
 	.globl REALNAME;\

common_x86.h

@@ -297,7 +297,8 @@ static __inline int blas_quickdivide(unsigned int x, unsigned int y){
 #endif
 
 #ifdef OS_DARWIN
-#define PROLOGUE .text;.align 5; .globl REALNAME; REALNAME:
+#define PROLOGUE .text;.align 5; .globl REALNAME; .hidden REALNAME; REALNAME:
+#define PROLOGUE_EXPORT .text;.align 5; .globl REALNAME; REALNAME:
 #define EPILOGUE	.subsections_via_symbols
 #define PROFCODE
 #endif
@@ -319,6 +320,14 @@ static __inline int blas_quickdivide(unsigned int x, unsigned int y){
 
 #if defined(OS_WINNT) || defined(OS_CYGWIN_NT) || defined(OS_INTERIX)
 #define PROLOGUE \
+	.text; \
+	.align 16; \
+	.globl REALNAME ;\
+	.hidden REALNAME ;\
+	.def REALNAME;.scl	2;.type	32;.endef; \
+REALNAME:
+
+#define PROLOGUE_EXPORT \
 	.text; \
 	.align 16; \
 	.globl REALNAME ;\
@@ -339,7 +348,16 @@ static __inline int blas_quickdivide(unsigned int x, unsigned int y){
 	.text; \
 	.align 16; \
 	.globl REALNAME ;\
-       .type REALNAME, @function; \
+	.hidden REALNAME ;\
+    .type REALNAME, @function; \
+REALNAME: \
+	_CET_ENDBR
+
+#define PROLOGUE_EXPORT \
+	.text; \
+	.align 16; \
+	.globl REALNAME ;\
+    .type REALNAME, @function; \
 REALNAME: \
 	_CET_ENDBR
 

common_x86_64.h

@@ -400,7 +400,8 @@ static __inline unsigned int blas_quickdivide(unsigned int x, unsigned int y){
 #endif
 
 #ifdef OS_DARWIN
-#define PROLOGUE .text;.align 5; .globl REALNAME; REALNAME:
+#define PROLOGUE .text;.align 5; .globl REALNAME; .hidden REALNAME; REALNAME:
+#define PROLOGUE_EXPORT .text;.align 5; .globl REALNAME; REALNAME:
 #define EPILOGUE	.subsections_via_symbols
 #define PROFCODE
 #endif
@@ -441,6 +442,14 @@ static __inline unsigned int blas_quickdivide(unsigned int x, unsigned int y){
 	.text; \
 	.align 16; \
 	.globl REALNAME ;\
+    .hidden REALNAME ; \
+	.def REALNAME;.scl	2;.type	32;.endef; \
+REALNAME:
+
+#define PROLOGUE_EXPORT \
+	.text; \
+	.align 16; \
+	.globl REALNAME ;\
 	.def REALNAME;.scl	2;.type	32;.endef; \
 REALNAME:
 
@@ -454,7 +463,16 @@ static __inline unsigned int blas_quickdivide(unsigned int x, unsigned int y){
 	.text; \
 	.align 512; \
 	.globl REALNAME ;\
-       .type REALNAME, @function; \
+	.hidden REALNAME ;\
+    .type REALNAME, @function; \
+REALNAME: \
+	_CET_ENDBR
+
+#define PROLOGUE_EXPORT \
+	.text; \
+	.align 512; \
+	.globl REALNAME ;\
+    .type REALNAME, @function; \
 REALNAME: \
 	_CET_ENDBR
 

driver/others/blas_server.c

@@ -861,6 +861,7 @@ if (openblas_threads_callback_) {
   return 0;
 }
 
+OPENBLAS_EXPORT
 void goto_set_num_threads(int num_threads) {
 
   long i;
@@ -933,6 +934,7 @@ void goto_set_num_threads(int num_threads) {
 
 }
 
+OPENBLAS_EXPORT
 void openblas_set_num_threads(int num_threads) {
 	goto_set_num_threads(num_threads);
 

driver/others/memory.c

@@ -500,6 +500,7 @@ int blas_get_cpu_number(void){
 #endif
 
 
+OPENBLAS_EXPORT
 int openblas_get_num_procs(void) {
 #ifndef SMP
   return 1;
@@ -508,6 +509,7 @@ int openblas_get_num_procs(void) {
 #endif
 }
 
+OPENBLAS_EXPORT
 int openblas_get_num_threads(void) {
 #ifndef SMP
   return 1;
@@ -2071,6 +2073,7 @@ int blas_get_cpu_number(void){
 #endif
 
 
+OPENBLAS_EXPORT
 int openblas_get_num_procs(void) {
 #ifndef SMP
   return 1;
@@ -2079,6 +2082,7 @@ int openblas_get_num_procs(void) {
 #endif
 }
 
+OPENBLAS_EXPORT
 int openblas_get_num_threads(void) {
 #ifndef SMP
   return 1;

driver/others/openblas_get_config.c

@@ -78,6 +78,7 @@ char *gotoblas_corename(void);
 static char tmp_config_str[256];
 int openblas_get_parallel(void);
 
+OPENBLAS_EXPORT
 char* CNAME(void) {
 char tmpstr[20];
   strcpy(tmp_config_str, openblas_config_str);
@@ -93,6 +94,7 @@ char tmpstr[20];
 }
 
 
+OPENBLAS_EXPORT
 char* openblas_get_corename(void) {
 #ifndef DYNAMIC_ARCH
   return CHAR_CORENAME;

driver/others/openblas_get_num_procs.c

@@ -35,6 +35,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 extern int openblas_get_num_procs(void);
 
+OPENBLAS_EXPORT
 int openblas_get_num_procs_(void) {
   return openblas_get_num_procs();
 }

driver/others/openblas_get_num_threads.c

@@ -35,6 +35,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 extern int openblas_get_num_threads(void);
 
+OPENBLAS_EXPORT
 int openblas_get_num_threads_(void) {
   return openblas_get_num_threads();
 }

driver/others/openblas_get_parallel.c

@@ -41,11 +41,14 @@ static int parallel = 0;
 #endif
 
 
+OPENBLAS_EXPORT
+
 #ifdef NEEDBUNDERSCORE
 int CNAME(void) {
   return parallel;
 }
 
+OPENBLAS_EXPORT
 int NAME(void) {
   return parallel;
 }

driver/others/openblas_set_num_threads.c

@@ -38,6 +38,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 extern  void openblas_set_num_threads(int num_threads) ;
 extern  int openblas_get_num_threads(void) ;
 
+OPENBLAS_EXPORT
 void openblas_set_num_threads_(int* num_threads){
 	openblas_set_num_threads(*num_threads);
 }
@@ -53,9 +54,11 @@ int openblas_set_num_threads_local(int num_threads){
 #else
 //Single thread
 
+OPENBLAS_EXPORT
 void openblas_set_num_threads(int num_threads) {
 }
 
+OPENBLAS_EXPORT
 void openblas_set_num_threads_(int* num_threads){
 
 }

interface/Makefile

@@ -29,6 +29,7 @@
 TOPDIR	= ..
 include $(TOPDIR)/Makefile.system
 
+
 SUPPORT_GEMM3M = 0
 
 ifeq ($(ARCH), x86)

interface/asum.c

@@ -42,6 +42,7 @@
 #include "functable.h"
 #endif
 
+OPENBLAS_EXPORT
 #ifndef CBLAS
 
 FLOATRET NAME(blasint *N, FLOAT *x, blasint *INCX){

interface/axpby.c

@@ -36,6 +36,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "functable.h"
 #endif
 
+OPENBLAS_EXPORT
 #ifndef CBLAS
 
 void NAME(blasint *N, FLOAT *ALPHA, FLOAT *x, blasint *INCX, FLOAT *BETA, FLOAT *y, blasint *INCY)

interface/axpy.c

@@ -46,6 +46,9 @@
 #else
 #define MULTI_THREAD_MINIMAL  10000
 #endif
+
+
+OPENBLAS_EXPORT
 #ifndef CBLAS
 
 void NAME(blasint *N, FLOAT *ALPHA, FLOAT *x, blasint *INCX, FLOAT *y, blasint *INCY){

interface/bf16dot.c

@@ -4,6 +4,7 @@
 #include "functable.h"
 #endif
 
+OPENBLAS_EXPORT
 #ifndef CBLAS
 float NAME(blasint *N, bfloat16 *x, blasint *INCX, bfloat16 *y, blasint *INCY){
    BLASLONG n    = *N;

interface/bf16to.c

@@ -11,6 +11,7 @@
 #else
 #endif
 
+OPENBLAS_EXPORT
 #ifndef CBLAS
 void NAME(blasint *N, bfloat16 *in, blasint *INC_IN, FLOAT_TYPE *out, blasint *INC_OUT){
   BLASLONG n    = *N;

interface/copy.c

@@ -42,6 +42,7 @@
 #include "functable.h"
 #endif
 
+OPENBLAS_EXPORT
 #ifndef CBLAS
 
 void NAME(blasint *N, FLOAT *x, blasint *INCX, FLOAT *y, blasint *INCY){

interface/dot.c

@@ -42,6 +42,7 @@
 #include "functable.h"
 #endif
 
+OPENBLAS_EXPORT
 #ifndef CBLAS
 
 FLOATRET NAME(blasint *N, FLOAT *x, blasint *INCX, FLOAT *y, blasint *INCY){