Use __inhibit_loop_to_libcall in all memset/memcpy implementations

This macro selects a compiler option that disables recognition of
common memset/memcpy patterns and converting those to direct
memset/memcpy calls.

Signed-off-by: Keith Packard <keithp@keithp.com>

newlib/libc/machine/cris/memcpy.c

@@ -31,6 +31,7 @@
    than what we describe.  An assembly file should be used instead.  */
 
 #include <stddef.h>
+#include "../../string/local.h"
 
 /* Break even between movem and move16 is really at 38.7 * 2, but
    modulo 44, so up to the next multiple of 44, we use ordinary code.  */
@@ -40,6 +41,7 @@
 __asm__ (".syntax no_register_prefix");
 
 void *
+__inhibit_loop_to_libcall
 memcpy(void *__restrict pdst, const void *__restrict psrc, size_t pn)
 {
   /* Now we want the parameters put in special registers.

newlib/libc/machine/riscv/memcpy.c

@@ -11,10 +11,12 @@
 
 #include <string.h>
 #include <stdint.h>
+#include "../../string/local.h"
 
 #define unlikely(X) __builtin_expect (!!(X), 0)
 
 void *
+__inhibit_loop_to_libcall
 memcpy(void *__restrict aa, const void *__restrict bb, size_t n)
 {
   #define BODY(a, b, t) { \

newlib/libc/machine/spu/memcpy.c

@@ -33,6 +33,7 @@
 #include <spu_intrinsics.h>
 #include <stddef.h>
 #include <vec_literal.h>
+#include "../../string/local.h"
 
 /* Copy n bytes from memory area src to memory area dest.
  * The memory areas may not overlap. The memcpy subroutine
@@ -42,7 +43,9 @@
  * either with prior knowledge of the alignment or special
  * casing specific optimal alignments.
  */
-void * memcpy(void * __restrict__ dest, const void * __restrict__ src, size_t n)
+void *
+__inhibit_loop_to_libcall
+memcpy(void * __restrict__ dest, const void * __restrict__ src, size_t n)
 {
   int adjust, delta;
   unsigned int soffset1, doffset1, doffset2;

newlib/libc/machine/spu/memcpy_ea.c

@@ -34,10 +34,12 @@ POSSIBILITY OF SUCH DAMAGE.
 #include "ea_internal.h"
 #include <ea.h>
 #include <spu_cache.h>
+#include "../../string/local.h"
 
 COMPAT_EA_ALIAS (memcpy_ea);
 
 __ea void *
+__inhibit_loop_to_libcall
 memcpy_ea (__ea void *dest, __ea const void *src, size_ea_t n)
 {
   __ea void *curr_dest = dest;

newlib/libc/machine/spu/memset_ea.c

@@ -34,10 +34,12 @@ POSSIBILITY OF SUCH DAMAGE.
 #include <spu_cache.h>
 #include "ea_internal.h"
 #include <ea.h>
+#include "../../string/local.h"
 
 COMPAT_EA_ALIAS (memset_ea);
 
 __ea void *
+__inhibit_loop_to_libcall
 memset_ea (__ea void *dest, int c, size_ea_t n)
 {
   __ea void *curr_dest = dest;

newlib/libc/machine/visium/memcpy.c

@@ -31,6 +31,7 @@
 
 #include <stddef.h>
 #include "memcpy.h"
+#include "../../string/local.h"
 
 #define INST_BARRIER   __asm__ __volatile__ ("":::"memory");
 
@@ -422,6 +423,7 @@ do {				\
 
 
 static inline void
+__inhibit_loop_to_libcall
 __int_memcpy (void *__restrict s1, const void *__restrict s2, size_t n) 
 {
   int value = n;
@@ -562,6 +564,7 @@ __int_memcpy (void *__restrict s1, const void *__restrict s2, size_t n)
 }
 
 static inline void
+__inhibit_loop_to_libcall
 __shrt_int_memcpy (void *__restrict s1, const void *__restrict s2, size_t n) 
 {
   int value = n;
@@ -703,6 +706,7 @@ __shrt_int_memcpy (void *__restrict s1, const void *__restrict s2, size_t n)
 
 
 static inline void
+__inhibit_loop_to_libcall
 __byte_memcpy (void *__restrict s1, const void *__restrict s2, size_t n) 
 {
   int value = n;
@@ -846,6 +850,7 @@ __byte_memcpy (void *__restrict s1, const void *__restrict s2, size_t n)
 /* Exposed interface.  */
 
 void *
+__inhibit_loop_to_libcall
 memcpy (void *__restrict s1, const void *__restrict s2, size_t n)
 {
   void *result = s1;

newlib/libc/machine/visium/memset.c

@@ -31,6 +31,7 @@
 
 #include <stddef.h>
 #include "memset.h"
+#include "../../string/local.h"
 
 #define SET_32_OBJECTS(out)	\
 do {				\
@@ -228,8 +229,8 @@ do {				\
   out += 1;			\
 } while(0)
 
-
 static inline void
+__inhibit_loop_to_libcall
 __int_memset (void *__restrict s1, int val, size_t n)
 {
   int value = n;
@@ -369,6 +370,7 @@ __int_memset (void *__restrict s1, int val, size_t n)
 }
 
 static inline void
+__inhibit_loop_to_libcall
 __short_int_memset (void *__restrict s1, int val, size_t n)
 {
   int value = n;
@@ -508,6 +510,7 @@ __short_int_memset (void *__restrict s1, int val, size_t n)
 }
 
 static inline void
+__inhibit_loop_to_libcall
 __byte_memset (void *__restrict s1, int val, size_t n)
 {
   int value = n;
@@ -650,6 +653,7 @@ __byte_memset (void *__restrict s1, int val, size_t n)
 /* Exposed interface.  */
 
 void *
+__inhibit_loop_to_libcall
 memset (void *s, int c, size_t n)
 {
   void *result = s;

newlib/libc/machine/xscale/memcpy.c

@@ -6,8 +6,10 @@
 
 #include <string.h>
 #include "xscale.h"
+#include "../../string/local.h"
 
 void *
+__inhibit_loop_to_libcall
 memcpy (void *__restrict dst0, const void *__restrict src0, size_t len)
 {
   int dummy;

newlib/libc/machine/xscale/memset.c

@@ -6,8 +6,10 @@
 
 #include <string.h>
 #include "xscale.h"
+#include "../../string/local.h"
 
 void *
+__inhibit_loop_to_libcall
 memset (void *dst, int c, size_t len)
 {
   int dummy;

newlib/libc/string/memcpy.c

@@ -29,6 +29,7 @@ QUICKREF
 
 #include <_ansi.h>
 #include <string.h>
+#include "local.h"
 
 /* Nonzero if either X or Y is not aligned on a "long" boundary.  */
 #define UNALIGNED(X, Y) \
@@ -44,6 +45,7 @@ QUICKREF
 #define TOO_SMALL(LEN)  ((LEN) < BIGBLOCKSIZE)
 
 void *
+__inhibit_loop_to_libcall
 memcpy (void *__restrict dst0,
 	const void *__restrict src0,
 	size_t len0)