ARM64: Convert all labels to local labels

While debugging/profiling applications using perf or other tools, the
kernels appear scattered in the profile reports. This is because the labels
within the kernels are not local and each label is shown as a separate
function.

To avoid this, all the labels within the kernels are changed to local
labels.

Author: Ashwin Sekhar T K

kernel/arm64/amax.S

@@ -160,88 +160,88 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	PROLOGUE
 
 	cmp	N, xzr
-	ble	amax_kernel_zero
+	ble	.Lamax_kernel_zero
 	cmp	INC_X, xzr
-	ble	amax_kernel_zero
+	ble	.Lamax_kernel_zero
 
 	cmp	INC_X, #1
-	bne	amax_kernel_S_BEGIN
+	bne	.Lamax_kernel_S_BEGIN
 
-amax_kernel_F_BEGIN:
+.Lamax_kernel_F_BEGIN:
 
 	asr	I, N, #2
 	cmp	I, xzr
-	beq	amax_kernel_F1_INIT
+	beq	.Lamax_kernel_F1_INIT
 
 	INIT_F4
 	subs	I, I, #1
-	beq	amax_kernel_F1
+	beq	.Lamax_kernel_F1
 
-amax_kernel_F4:
+.Lamax_kernel_F4:
 
 	KERNEL_F4
 
 	subs	I, I, #1
-	bne	amax_kernel_F4
+	bne	.Lamax_kernel_F4
 
-amax_kernel_F1:
+.Lamax_kernel_F1:
 
 	ands	I, N, #3
-	ble	amax_kernel_L999
+	ble	.Lamax_kernel_L999
 
-amax_kernel_F10:
+.Lamax_kernel_F10:
 
 	KERNEL_F1
 
 	subs    I, I, #1
-        bne     amax_kernel_F10
+        bne     .Lamax_kernel_F10
 
 	ret
 
-amax_kernel_F1_INIT:
+.Lamax_kernel_F1_INIT:
 
 	INIT_F1
 	subs	N, N, #1
-	b	amax_kernel_F1
+	b	.Lamax_kernel_F1
 
-amax_kernel_S_BEGIN:
+.Lamax_kernel_S_BEGIN:
 
 	INIT_S
 
 	subs	N, N, #1
-	ble	amax_kernel_L999
+	ble	.Lamax_kernel_L999
 
 	asr	I, N, #2
 	cmp	I, xzr
-	ble	amax_kernel_S1
+	ble	.Lamax_kernel_S1
 
-amax_kernel_S4:
+.Lamax_kernel_S4:
 
 	KERNEL_S1
 	KERNEL_S1
 	KERNEL_S1
 	KERNEL_S1
 
 	subs	I, I, #1
-	bne	amax_kernel_S4
+	bne	.Lamax_kernel_S4
 
-amax_kernel_S1:
+.Lamax_kernel_S1:
 
 	ands	I, N, #3
-	ble	amax_kernel_L999
+	ble	.Lamax_kernel_L999
 
-amax_kernel_S10:
+.Lamax_kernel_S10:
 
 	KERNEL_S1
 
 	subs    I, I, #1
-        bne     amax_kernel_S10
+        bne     .Lamax_kernel_S10
 
-amax_kernel_L999:
+.Lamax_kernel_L999:
 
 	ret
 
-amax_kernel_zero:
+.Lamax_kernel_zero:
 
 	fmov	MAXF, REG0
 	ret

kernel/arm64/asum.S

@@ -122,72 +122,72 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #endif
 
 	cmp	N, xzr
-	ble	asum_kernel_L999
+	ble	.Lasum_kernel_L999
 	cmp	INC_X, xzr
-	ble	asum_kernel_L999
+	ble	.Lasum_kernel_L999
 
 	cmp	INC_X, #1
-	bne	asum_kernel_S_BEGIN
+	bne	.Lasum_kernel_S_BEGIN
 
-asum_kernel_F_BEGIN:
+.Lasum_kernel_F_BEGIN:
 
 	asr	I, N, #3
 	cmp	I, xzr
-	beq	asum_kernel_F1
+	beq	.Lasum_kernel_F1
 
-asum_kernel_F8:
+.Lasum_kernel_F8:
 
 	KERNEL_F8
 
 	subs	I, I, #1
-	bne	asum_kernel_F8
+	bne	.Lasum_kernel_F8
 
 	KERNEL_F8_FINALIZE
 
-asum_kernel_F1:
+.Lasum_kernel_F1:
 
 	ands	I, N, #7
-	ble	asum_kernel_L999
+	ble	.Lasum_kernel_L999
 
-asum_kernel_F10:
+.Lasum_kernel_F10:
 
 	KERNEL_F1
 
 	subs    I, I, #1
-        bne     asum_kernel_F10
+        bne     .Lasum_kernel_F10
 
-asum_kernel_L999:
+.Lasum_kernel_L999:
 	ret
 
-asum_kernel_S_BEGIN:
+.Lasum_kernel_S_BEGIN:
 
 	INIT_S
 
 	asr	I, N, #2
 	cmp	I, xzr
-	ble	asum_kernel_S1
+	ble	.Lasum_kernel_S1
 
-asum_kernel_S4:
+.Lasum_kernel_S4:
 
 	KERNEL_S1
 	KERNEL_S1
 	KERNEL_S1
 	KERNEL_S1
 
 	subs	I, I, #1
-	bne	asum_kernel_S4
+	bne	.Lasum_kernel_S4
 
-asum_kernel_S1:
+.Lasum_kernel_S1:
 
 	ands	I, N, #3
-	ble	asum_kernel_L999
+	ble	.Lasum_kernel_L999
 
-asum_kernel_S10:
+.Lasum_kernel_S10:
 
 	KERNEL_S1
 
 	subs    I, I, #1
-        bne     asum_kernel_S10
+        bne     .Lasum_kernel_S10
 
 	ret
 

kernel/arm64/axpy.S

@@ -135,75 +135,75 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	PROLOGUE
 
 	cmp	N, xzr
-	ble	axpy_kernel_L999
+	ble	.Laxpy_kernel_L999
 
 	fcmp	DA, #0.0
-	beq	axpy_kernel_L999
+	beq	.Laxpy_kernel_L999
 
 	cmp	INC_X, #1
-	bne	axpy_kernel_S_BEGIN
+	bne	.Laxpy_kernel_S_BEGIN
 	cmp	INC_Y, #1
-	bne	axpy_kernel_S_BEGIN
+	bne	.Laxpy_kernel_S_BEGIN
 
-axpy_kernel_F_BEGIN:
+.Laxpy_kernel_F_BEGIN:
 
 	asr	I, N, #3
 	cmp	I, xzr
-	beq	axpy_kernel_F1
+	beq	.Laxpy_kernel_F1
 
-axpy_kernel_F8:
+.Laxpy_kernel_F8:
 
 	KERNEL_F8
 
 	subs	I, I, #1
-	bne	axpy_kernel_F8
+	bne	.Laxpy_kernel_F8
 
-axpy_kernel_F1:
+.Laxpy_kernel_F1:
 
 	ands	I, N, #7
-	ble	axpy_kernel_L999
+	ble	.Laxpy_kernel_L999
 
-axpy_kernel_F10:
+.Laxpy_kernel_F10:
 
 	KERNEL_F1
 
 	subs    I, I, #1
-        bne     axpy_kernel_F10
+        bne     .Laxpy_kernel_F10
 
 	mov	w0, wzr
 	ret
 
-axpy_kernel_S_BEGIN:
+.Laxpy_kernel_S_BEGIN:
 
 	INIT_S
 
 	asr	I, N, #2
 	cmp	I, xzr
-	ble	axpy_kernel_S1
+	ble	.Laxpy_kernel_S1
 
-axpy_kernel_S4:
+.Laxpy_kernel_S4:
 
 	KERNEL_S1
 	KERNEL_S1
 	KERNEL_S1
 	KERNEL_S1
 
 	subs	I, I, #1
-	bne	axpy_kernel_S4
+	bne	.Laxpy_kernel_S4
 
-axpy_kernel_S1:
+.Laxpy_kernel_S1:
 
 	ands	I, N, #3
-	ble	axpy_kernel_L999
+	ble	.Laxpy_kernel_L999
 
-axpy_kernel_S10:
+.Laxpy_kernel_S10:
 
 	KERNEL_S1
 
 	subs    I, I, #1
-        bne     axpy_kernel_S10
+        bne     .Laxpy_kernel_S10
 
-axpy_kernel_L999:
+.Laxpy_kernel_L999:
 
 	mov	w0, wzr
 	ret

kernel/arm64/casum.S

@@ -98,72 +98,72 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	fmov	s1, SUMF
 
 	cmp	N, xzr
-	ble	asum_kernel_L999
+	ble	.Lcasum_kernel_L999
 	cmp	INC_X, xzr
-	ble	asum_kernel_L999
+	ble	.Lcasum_kernel_L999
 
 	cmp	INC_X, #1
-	bne	asum_kernel_S_BEGIN
+	bne	.Lcasum_kernel_S_BEGIN
 
-asum_kernel_F_BEGIN:
+.Lcasum_kernel_F_BEGIN:
 
 	asr	I, N, #3
 	cmp	I, xzr
-	beq	asum_kernel_F1
+	beq	.Lcasum_kernel_F1
 
-asum_kernel_F8:
+.Lcasum_kernel_F8:
 
 	KERNEL_F8
 
 	subs	I, I, #1
-	bne	asum_kernel_F8
+	bne	.Lcasum_kernel_F8
 
 	KERNEL_F8_FINALIZE
 
-asum_kernel_F1:
+.Lcasum_kernel_F1:
 
 	ands	I, N, #7
-	ble	asum_kernel_L999
+	ble	.Lcasum_kernel_L999
 
-asum_kernel_F10:
+.Lcasum_kernel_F10:
 
 	KERNEL_F1
 
 	subs    I, I, #1
-        bne     asum_kernel_F10
+        bne     .Lcasum_kernel_F10
 
-asum_kernel_L999:
+.Lcasum_kernel_L999:
 	ret
 
-asum_kernel_S_BEGIN:
+.Lcasum_kernel_S_BEGIN:
 
 	INIT_S
 
 	asr	I, N, #2
 	cmp	I, xzr
-	ble	asum_kernel_S1
+	ble	.Lcasum_kernel_S1
 
-asum_kernel_S4:
+.Lcasum_kernel_S4:
 
 	KERNEL_S1
 	KERNEL_S1
 	KERNEL_S1
 	KERNEL_S1
 
 	subs	I, I, #1
-	bne	asum_kernel_S4
+	bne	.Lcasum_kernel_S4
 
-asum_kernel_S1:
+.Lcasum_kernel_S1:
 
 	ands	I, N, #3
-	ble	asum_kernel_L999
+	ble	.Lcasum_kernel_L999
 
-asum_kernel_S10:
+.Lcasum_kernel_S10:
 
 	KERNEL_S1
 
 	subs    I, I, #1
-        bne     asum_kernel_S10
+        bne     .Lcasum_kernel_S10
 
 	ret