Use ENTRY and EXIT macros for strcmp, memcpy, atexit.

Without this change strcmp size is zero (not set), and it gets
ignored by Valgrind. Changes to memcpy and atexit don't affect the
generated binary in any way.

Change-Id: I05818cb5951f75901dc8c0eef02807a2e83a9231

libc/arch-arm/bionic/atexit.S

@@ -37,16 +37,10 @@
 	.eabi_attribute 26, 2
 	.eabi_attribute 30, 4
 	.eabi_attribute 18, 4
-	.code	16
-	.section	.text.atexit,"ax",%progbits
-	.align	2
-	.global	atexit
 	.hidden	atexit
 	.code	16
 	.thumb_func
-	.type	atexit, %function
+ENTRY(atexit)
-atexit:
-	.fnstart
 .LFB0:
 	.save	{r4, lr}
 	push	{r4, lr}
@@ -64,6 +58,5 @@ atexit:
 .L3:
 	.word	__dso_handle-(.LPIC0+4)
 .LFE0:
-	.fnend
+END(atexit)
-	.size	atexit, .-atexit
 #endif

libc/arch-arm/bionic/memcpy.S

@@ -34,16 +34,11 @@
         .text
         .fpu    neon
 
-        .global memcpy
-        .type memcpy, %function
-        .align 4
-
 /* a prefetch distance of 4 cache-lines works best experimentally */
 #define CACHE_LINE_SIZE     64
 #define PREFETCH_DISTANCE   (CACHE_LINE_SIZE*4)
 
-memcpy:
+ENTRY(memcpy)
-        .fnstart
         .save       {r0, lr}
         stmfd       sp!, {r0, lr}
 

libc/arch-arm/bionic/strcmp.S

@@ -28,13 +28,10 @@
  */
 
 #include <machine/cpu-features.h>
+#include <machine/asm.h>
 
 	.text
 
-	.global strcmp
-	.type strcmp, %function
-	.align 4
-
 #ifdef __ARMEB__
 #define SHFT2LSB lsl
 #define SHFT2LSBEQ lsleq
@@ -54,8 +51,7 @@
 #define magic1(REG) REG
 #define magic2(REG) REG, lsl #7
 
-strcmp:
+ENTRY(strcmp)
-	.fnstart
 	PLD(r0, #0)
 	PLD(r1, #0)
 	eor	r2, r0, r1
@@ -136,7 +132,6 @@ strcmp:
 #endif
 	ldr	r4, [sp], #4
 	bx	lr
-	.fnend
 
 .Lstrcmp_unaligned:
 	wp1 .req r0
@@ -319,3 +314,4 @@ strcmp:
 	ldr	r4, [sp], #4
 	ldr	r5, [sp], #4
 	bx	lr
+END(strcmp)