diff --git a/driver/others/blas_server.c b/driver/others/blas_server.c
index 2afcb742e..cc1015b7e 100644
--- a/driver/others/blas_server.c
+++ b/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
 
diff --git a/driver/others/blas_server_omp.c b/driver/others/blas_server_omp.c
index 0a484f3e4..090590e6a 100644
--- a/driver/others/blas_server_omp.c
+++ b/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
diff --git a/driver/others/blas_server_win32.c b/driver/others/blas_server_win32.c
index 100ca34f7..68630a679 100644
--- a/driver/others/blas_server_win32.c
+++ b/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()
+{
+
+}
diff --git a/driver/others/memory.c b/driver/others/memory.c
index 35758d13c..4faf82f29 100644
--- a/driver/others/memory.c
+++ b/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
 
diff --git a/kernel/arm/KERNEL.ARMV6 b/kernel/arm/KERNEL.ARMV6
index f47a843f3..02627cbd0 100644
--- a/kernel/arm/KERNEL.ARMV6
+++ b/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
diff --git a/kernel/arm/KERNEL.ARMV7 b/kernel/arm/KERNEL.ARMV7
index 507f9813c..df66388db 100644
--- a/kernel/arm/KERNEL.ARMV7
+++ b/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
 
diff --git a/kernel/x86_64/dgemv_n.S b/kernel/x86_64/dgemv_n.S
index 65e228584..5f4c40467 100644
--- a/kernel/x86_64/dgemv_n.S
+++ b/kernel/x86_64/dgemv_n.S
@@ -111,6 +111,9 @@
 #define MM	M
 #endif
 
+#define TMP_M   %r15
+#define Y2      %rbx
+	
 	PROLOGUE
 	PROFCODE
 
@@ -170,8 +173,9 @@
 	jge	.L00t
 
 	movq	MMM,M
-	addq	I,M
+	addq	M, I
 	jle	.L999x
+	movq	I, M
 	
 .L00t:
 	movq	XX,X
@@ -2463,21 +2467,23 @@
 	cmpq	Y, BUFFER
 	je	.L999
 #endif
-
+	movq	M,   TMP_M
+	movq    Y,    Y1
+	
 	cmpq	$SIZE, INCY
 	jne	.L950
 
-	testq	$SIZE, Y
+	testq	$SIZE, Y1
 	je	.L910
 
-	movsd	(Y), %xmm0
+	movsd	(Y1), %xmm0
 	addsd	(BUFFER), %xmm0
-	movsd	%xmm0, (Y)
+	movsd	%xmm0, (Y1)
 
-	addq	$SIZE, Y
+	addq	$SIZE, Y1
 	addq	$SIZE, BUFFER
 
-	decq	M
+	decq	TMP_M
 	jle	.L999
 	ALIGN_4
 
@@ -2485,20 +2491,20 @@
 	testq	$SIZE, BUFFER
 	jne	.L920
 
-	movq	M,   %rax
+	movq	TMP_M,   %rax
 	sarq	$3,  %rax
 	jle	.L914
 	ALIGN_3
 
 .L912:
 #ifdef PREFETCHW
-	PREFETCHW	(PREFETCHSIZE) * 4 + PREOFFSET(Y)
+	PREFETCHW	(PREFETCHSIZE) * 4 + PREOFFSET(Y1)
 #endif
 
-	movapd	0 * SIZE(Y), %xmm0
-	movapd	2 * SIZE(Y), %xmm1
-	movapd	4 * SIZE(Y), %xmm2
-	movapd	6 * SIZE(Y), %xmm3
+	movapd	0 * SIZE(Y1), %xmm0
+	movapd	2 * SIZE(Y1), %xmm1
+	movapd	4 * SIZE(Y1), %xmm2
+	movapd	6 * SIZE(Y1), %xmm3
 
 	movapd	0 * SIZE(BUFFER), %xmm4
 	movapd	2 * SIZE(BUFFER), %xmm5
@@ -2514,12 +2520,12 @@
 	addpd	%xmm6, %xmm2
 	addpd	%xmm7, %xmm3
 
-	movapd	%xmm0,  0 * SIZE(Y)
-	movapd	%xmm1,  2 * SIZE(Y)
-	movapd	%xmm2,  4 * SIZE(Y)
-	movapd	%xmm3,  6 * SIZE(Y)
+	movapd	%xmm0,  0 * SIZE(Y1)
+	movapd	%xmm1,  2 * SIZE(Y1)
+	movapd	%xmm2,  4 * SIZE(Y1)
+	movapd	%xmm3,  6 * SIZE(Y1)
 
-	addq	$8 * SIZE, Y
+	addq	$8 * SIZE, Y1
 	addq	$8 * SIZE, BUFFER
 
 	decq	%rax
@@ -2527,14 +2533,14 @@
 	ALIGN_3
 
 .L914:
-	testq	$7, M
+	testq	$7, TMP_M
 	jle	.L999
 
-	testq	$4, M
+	testq	$4, TMP_M
 	jle	.L915
 
-	movapd	0 * SIZE(Y), %xmm0
-	movapd	2 * SIZE(Y), %xmm1
+	movapd	0 * SIZE(Y1), %xmm0
+	movapd	2 * SIZE(Y1), %xmm1
 
 	movapd	0 * SIZE(BUFFER), %xmm4
 	movapd	2 * SIZE(BUFFER), %xmm5
@@ -2542,40 +2548,40 @@
 	addpd	%xmm4, %xmm0
 	addpd	%xmm5, %xmm1
 
-	movapd	%xmm0,  0 * SIZE(Y)
-	movapd	%xmm1,  2 * SIZE(Y)
+	movapd	%xmm0,  0 * SIZE(Y1)
+	movapd	%xmm1,  2 * SIZE(Y1)
 
-	addq	$4 * SIZE, Y
+	addq	$4 * SIZE, Y1
 	addq	$4 * SIZE, BUFFER
 	ALIGN_3
 
 .L915:
-	testq	$2, M
+	testq	$2, TMP_M
 	jle	.L916
 
-	movapd	(Y), %xmm0
+	movapd	(Y1), %xmm0
 
 	movapd	(BUFFER), %xmm4
 
 	addpd	%xmm4, %xmm0
 
-	movapd	%xmm0,  (Y)
+	movapd	%xmm0,  (Y1)
 
-	addq	$2 * SIZE, Y
+	addq	$2 * SIZE, Y1
 	addq	$2 * SIZE, BUFFER
 	ALIGN_3
 
 .L916:
-	testq	$1, M
+	testq	$1, TMP_M
 	jle	.L999
 
-	movsd	(Y), %xmm0
+	movsd	(Y1), %xmm0
 
 	movsd	0 * SIZE(BUFFER), %xmm4
 
 	addsd	%xmm4, %xmm0
 
-	movlpd	%xmm0, (Y)
+	movlpd	%xmm0, (Y1)
 	ALIGN_3
 
 	jmp	.L999
@@ -2584,20 +2590,20 @@
 .L920:
 	movapd	-1 * SIZE(BUFFER), %xmm4
 
-	movq	M,   %rax
+	movq	TMP_M,   %rax
 	sarq	$3,  %rax
 	jle	.L924
 	ALIGN_3
 
 .L922:
 #ifdef PREFETCHW
-	PREFETCHW	(PREFETCHSIZE) * 4 + PREOFFSET(Y)
+	PREFETCHW	(PREFETCHSIZE) * 4 + PREOFFSET(Y1)
 #endif
 
-	movapd	 0 * SIZE(Y), %xmm0
-	movapd	 2 * SIZE(Y), %xmm1
-	movapd	 4 * SIZE(Y), %xmm2
-	movapd	 6 * SIZE(Y), %xmm3
+	movapd	 0 * SIZE(Y1), %xmm0
+	movapd	 2 * SIZE(Y1), %xmm1
+	movapd	 4 * SIZE(Y1), %xmm2
+	movapd	 6 * SIZE(Y1), %xmm3
 
 	movapd	1 * SIZE(BUFFER), %xmm5
 	movapd	3 * SIZE(BUFFER), %xmm6
@@ -2618,14 +2624,14 @@
 	addpd	%xmm6, %xmm2
 	addpd	%xmm7, %xmm3
 
-	movapd	%xmm0,  0 * SIZE(Y)
-	movapd	%xmm1,  2 * SIZE(Y)
-	movapd	%xmm2,  4 * SIZE(Y)
-	movapd	%xmm3,  6 * SIZE(Y)
+	movapd	%xmm0,  0 * SIZE(Y1)
+	movapd	%xmm1,  2 * SIZE(Y1)
+	movapd	%xmm2,  4 * SIZE(Y1)
+	movapd	%xmm3,  6 * SIZE(Y1)
 
 	movapd	%xmm8, %xmm4
 
-	addq	$8 * SIZE, Y
+	addq	$8 * SIZE, Y1
 	addq	$8 * SIZE, BUFFER
 
 	decq	%rax
@@ -2633,14 +2639,14 @@
 	ALIGN_3
 
 .L924:
-	testq	$7, M
+	testq	$7, TMP_M
 	jle	.L999
 
-	testq	$4, M
+	testq	$4, TMP_M
 	jle	.L925
 
-	movapd	 0 * SIZE(Y), %xmm0
-	movapd	 2 * SIZE(Y), %xmm1
+	movapd	 0 * SIZE(Y1), %xmm0
+	movapd	 2 * SIZE(Y1), %xmm1
 
 	movapd	1 * SIZE(BUFFER), %xmm5
 	movapd	3 * SIZE(BUFFER), %xmm6
@@ -2651,20 +2657,20 @@
 	addpd	%xmm4, %xmm0
 	addpd	%xmm5, %xmm1
 
-	movapd	%xmm0,  0 * SIZE(Y)
-	movapd	%xmm1,  2 * SIZE(Y)
+	movapd	%xmm0,  0 * SIZE(Y1)
+	movapd	%xmm1,  2 * SIZE(Y1)
 
 	movapd	%xmm6, %xmm4
 
-	addq	$4 * SIZE, Y
+	addq	$4 * SIZE, Y1
 	addq	$4 * SIZE, BUFFER
 	ALIGN_3
 
 .L925:
-	testq	$2, M
+	testq	$2, TMP_M
 	jle	.L926
 
-	movapd	 (Y), %xmm0
+	movapd	 (Y1), %xmm0
 
 	movapd	 1 * SIZE(BUFFER), %xmm5
 
@@ -2672,25 +2678,25 @@
 
 	addpd	%xmm4, %xmm0
 
-	movapd	%xmm0,  (Y)
+	movapd	%xmm0,  (Y1)
 
 	movaps	%xmm5, %xmm4
 
-	addq	$2 * SIZE, Y
+	addq	$2 * SIZE, Y1
 	addq	$2 * SIZE, BUFFER
 	ALIGN_3
 
 .L926:
-	testq	$1, M
+	testq	$1, TMP_M
 	jle	.L999
 
-	movsd	(Y), %xmm0
+	movsd	(Y1), %xmm0
 
 	shufpd	$1, %xmm4, %xmm4
 
 	addsd	%xmm4, %xmm0
 
-	movlpd	%xmm0, (Y)
+	movlpd	%xmm0, (Y1)
 	ALIGN_3
 
 	jmp	.L999
@@ -2700,53 +2706,53 @@
 	testq	$SIZE, BUFFER
 	je	.L960
 
-	movsd	(Y), %xmm0
+	movsd	(Y1), %xmm0
 	addsd	(BUFFER), %xmm0
-	movsd	%xmm0, (Y)
+	movsd	%xmm0, (Y1)
 
-	addq	INCY, Y
+	addq	INCY, Y1
 	addq	$SIZE, BUFFER
 
-	decq	M
+	decq	TMP_M
 	jle	.L999
 	ALIGN_4
 
 .L960:
-	movq	Y,  Y1
+	movq	Y1,  Y2
 
-	movq	M,   %rax
+	movq	TMP_M,   %rax
 	sarq	$3,  %rax
 	jle	.L964
 	ALIGN_3
 
 .L962:
-	movsd	(Y), %xmm0
-	addq	INCY, Y
-	movhpd	(Y), %xmm0
-	addq	INCY, Y
+	movsd	(Y2), %xmm0
+	addq	INCY, Y2
+	movhpd	(Y2), %xmm0
+	addq	INCY, Y2
 
 	movapd	0 * SIZE(BUFFER), %xmm4
 
-	movsd	(Y), %xmm1
-	addq	INCY, Y
-	movhpd	(Y), %xmm1
-	addq	INCY, Y
+	movsd	(Y2), %xmm1
+	addq	INCY, Y2
+	movhpd	(Y2), %xmm1
+	addq	INCY, Y2
 
 	movapd	2 * SIZE(BUFFER), %xmm5
 
-	movsd	(Y), %xmm2
-	addq	INCY, Y
-	movhpd	(Y), %xmm2
-	addq	INCY, Y
+	movsd	(Y2), %xmm2
+	addq	INCY, Y2
+	movhpd	(Y2), %xmm2
+	addq	INCY, Y2
 
 	movapd	4 * SIZE(BUFFER), %xmm6
 
 	addpd	%xmm4, %xmm0
 
-	movsd	(Y), %xmm3
-	addq	INCY, Y
-	movhpd	(Y), %xmm3
-	addq	INCY, Y
+	movsd	(Y2), %xmm3
+	addq	INCY, Y2
+	movhpd	(Y2), %xmm3
+	addq	INCY, Y2
 
 	movapd	6 * SIZE(BUFFER), %xmm7
 
@@ -2781,23 +2787,23 @@
 	ALIGN_3
 
 .L964:
-	testq	$7, M
+	testq	$7, TMP_M
 	jle	.L999
 
-	testq	$4, M
+	testq	$4, TMP_M
 	jle	.L965
 
-	movsd	(Y), %xmm0
-	addq	INCY, Y
-	movhpd	(Y), %xmm0
-	addq	INCY, Y
+	movsd	(Y2), %xmm0
+	addq	INCY, Y2
+	movhpd	(Y2), %xmm0
+	addq	INCY, Y2
 
 	movapd	0 * SIZE(BUFFER), %xmm4
 
-	movsd	(Y), %xmm1
-	addq	INCY, Y
-	movhpd	(Y), %xmm1
-	addq	INCY, Y
+	movsd	(Y2), %xmm1
+	addq	INCY, Y2
+	movhpd	(Y2), %xmm1
+	addq	INCY, Y2
 
 	movapd	2 * SIZE(BUFFER), %xmm5
 
@@ -2817,13 +2823,13 @@
 	ALIGN_3
 
 .L965:
-	testq	$2, M
+	testq	$2, TMP_M
 	jle	.L966
 
-	movsd	(Y), %xmm0
-	addq	INCY, Y
-	movhpd	(Y), %xmm0
-	addq	INCY, Y
+	movsd	(Y2), %xmm0
+	addq	INCY, Y2
+	movhpd	(Y2), %xmm0
+	addq	INCY, Y2
 
 	movapd	0 * SIZE(BUFFER), %xmm4
 
@@ -2838,10 +2844,10 @@
 	ALIGN_3
 
 .L966:
-	testq	$1, M
+	testq	$1, TMP_M
 	jle	.L999
 
-	movsd	(Y), %xmm0
+	movsd	(Y2), %xmm0
 
 	movsd	0 * SIZE(BUFFER), %xmm4
 
@@ -2853,6 +2859,9 @@
 .L999:
 	leaq	(, M, SIZE), %rax
 	addq %rax,AA
+	movq	STACK_INCY, INCY
+	imulq	INCY, %rax
+	addq %rax, Y
 	jmp .L0t
 	ALIGN_4