Merge branch 'develop' of github.com:xianyi/OpenBLAS into develop

Author: xianyi

driver/others/blas_server.c

@@ -83,6 +83,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define ATTRIBUTE_SIZE 128
 
+extern void openblas_warning(int verbose, const char * msg);
+
 /* This is a thread server model implementation.  The threads are   */
 /* spawned at first access to blas library, and still remains until */
 /* destruction routine is called.  The number of threads are        */
@@ -921,5 +923,17 @@ int BLASFUNC(blas_thread_shutdown)(void){
   return 0;
 }
 
+/*
+https://github.com/xianyi/OpenBLAS/issues/294
+Use pthread_atfork to close blas_thread_server before fork.
+Then, re-init blas_thread_server after fork at child and parent.
+*/
+void openblas_fork_handler()
+{
+  int err;
+  err = pthread_atfork (BLASFUNC(blas_thread_shutdown), blas_thread_init, blas_thread_init);
+  if(err != 0)
+    openblas_warning(0, "OpenBLAS cannot install fork handler. You may meet hang after fork.\n");
+}
 #endif
 

driver/others/blas_server_omp.c

@@ -315,4 +315,9 @@ int exec_blas(BLASLONG num, blas_queue_t *queue){
   return 0;
 }
 
+void openblas_fork_handler()
+{
+
+}
+
 #endif

driver/others/blas_server_win32.c

@@ -498,3 +498,8 @@ void openblas_set_num_threads(int num)
 {
 	goto_set_num_threads(num);
 }
+
+void openblas_fork_handler()
+{
+
+}

driver/others/memory.c

@@ -1288,7 +1288,11 @@ void CONSTRUCTOR gotoblas_init(void) {
 #ifdef SMP
   if (blas_cpu_number == 0) blas_get_cpu_number();
 #ifdef SMP_SERVER
-  if (blas_server_avail == 0) blas_thread_init();
+  if (blas_server_avail == 0) {
+    blas_thread_init();
+    //deal with pthread and fork.
+    openblas_fork_handler();
+  }
 #endif
 #endif
 

kernel/arm/KERNEL.ARMV6

@@ -40,10 +40,10 @@ DAXPYKERNEL  = axpy_vfp.S
 CAXPYKERNEL  = axpy_vfp.S
 ZAXPYKERNEL  = axpy_vfp.S
 
-SCOPYKERNEL  = scopy_vfp.S
-DCOPYKERNEL  = dcopy_vfp.S
-CCOPYKERNEL  = ccopy_vfp.S
-ZCOPYKERNEL  = zcopy_vfp.S
+SCOPYKERNEL  = copy.c
+DCOPYKERNEL  = copy.c
+CCOPYKERNEL  = zcopy.c
+ZCOPYKERNEL  = zcopy.c
 
 SDOTKERNEL   = sdot_vfp.S
 DDOTKERNEL   = ddot_vfp.S

kernel/arm/KERNEL.ARMV7

@@ -45,10 +45,10 @@ DAXPYKERNEL  = axpy_vfp.S
 CAXPYKERNEL  = axpy_vfp.S
 ZAXPYKERNEL  = axpy_vfp.S
 
-SCOPYKERNEL  = scopy_vfp.S
-DCOPYKERNEL  = dcopy_vfp.S
-CCOPYKERNEL  = ccopy_vfp.S
-ZCOPYKERNEL  = zcopy_vfp.S
+SCOPYKERNEL  = copy.c
+DCOPYKERNEL  = copy.c
+CCOPYKERNEL  = zcopy.c
+ZCOPYKERNEL  = zcopy.c
 
 SDOTKERNEL   = sdot_vfp.S
 DDOTKERNEL   = ddot_vfp.S
@@ -66,12 +66,12 @@ CROTKERNEL   = rot_vfp.S
 ZROTKERNEL   = rot_vfp.S
 
 SSCALKERNEL  = scal_vfp.S
-DSCALKERNEL  = scal_vfp.S
+DSCALKERNEL  = scal.c
 CSCALKERNEL  = scal_vfp.S
 ZSCALKERNEL  = scal_vfp.S
 
 SGEMVNKERNEL = gemv_n_vfp.S
-DGEMVNKERNEL = gemv_n_vfp.S
+DGEMVNKERNEL = gemv_n.c
 CGEMVNKERNEL = cgemv_n_vfp.S
 ZGEMVNKERNEL = zgemv_n_vfp.S