Added optimized version of memcpy.c to this directory since RTEMS makes

important distinctions between CPU models which are not made by gcc.
These distinctions help give us a more optimized memcpy().  This is important
for message queues and KA9Q.

c/src/exec/score/cpu/m68k/Makefile.in

@@ -11,7 +11,7 @@ PROJECT_ROOT = @PROJECT_ROOT@
 RELS=$(ARCH)/rtems-cpu.rel
 
 # C source names, if any, go here -- minus the .c
-C_PIECES=cpu
+C_PIECES=cpu memcpy
 C_FILES=$(C_PIECES:%=%.c)
 C_O_FILES=$(C_PIECES:%=${ARCH}/%.o)
 

c/src/exec/score/cpu/m68k/memcpy.c

@@ -0,0 +1,87 @@
+/*
+ *  C library memcpy routine
+ *
+ *  This routine has code to optimize performance on the CPU32+
+ *  and another version for other 68k machines.
+ *
+ *  It could be optimized more for machines with MOVE16 instructions.
+ *
+ *  The routine is placed in this source directory to ensure that it
+ *  is picked up by all applications.
+ *
+ *  W. Eric Norum
+ *  Saskatchewan Accelerator Laboratory
+ *  University of Saskatchewan
+ *  Saskatoon, Saskatchewan, CANADA
+ *  eric@skatter.usask.ca
+ */
+
+#include <string.h>
+#include <rtems/score/m68k.h>
+
+#if defined(__mcpu32__)
+#define COPYSETUP(n)	n--
+#define COPY(to,from,n,size) \
+	asm volatile ("1:\n" \
+	     "\tmove." size " (%0)+,(%1)+\n" \
+	     "\tdbf %2,1b\n"  \
+	     "\tsub.l #0x10000,%2\n" \
+	     "\tbpl.b 1b\n" : \
+		"=a" (from), "=a" (to), "=d" (n) :\
+		 "0" (from),  "1" (to), "2" (n) : \
+		 "cc", "memory")
+#else
+#define COPYSETUP(n)
+#define COPY(to,from,n,size) \
+	asm volatile ("1:\n" \
+	     "\tmove." size " (%0)+,(%1)+\n" \
+	     "\tsubq.l #1,%2\n\tbne.b 1b\n" : \
+		"=a" (from), "=a" (to), "=d" (n) :\
+		 "0" (from),  "1" (to), "2" (n) : \
+		 "cc", "memory")
+#endif
+
+void *
+memcpy(void *s1, const void *s2, size_t n)
+{
+	char *p1 = s1;
+	const char *p2 = s2;
+
+	if (n) {
+		if (n < 16) {
+			COPYSETUP (n);
+			COPY (p1, p2, n, "b");
+		}
+		else {
+			int nbyte;
+			int nl;
+			nbyte = (int)p1 & 0x3;
+			if (nbyte) {
+				nbyte = 4 - nbyte;
+				n -= nbyte;
+				COPYSETUP (nbyte);
+				COPY (p1, p2, nbyte, "b");
+			}
+#if (M68K_HAS_MISALIGNED == 0)
+			/*
+			 * Take care of machines that can't
+			 * handle misaligned references.
+			 */
+			if ((int)p2 & 0x1) {
+				COPYSETUP (n);
+				COPY (p1, p2, n, "b");
+				return s1;
+			}
+#endif
+			nl = (unsigned int)n >> 2;
+			COPYSETUP (nl);
+			COPY (p1, p2, nl, "l");
+			nbyte = (int)n & 0x3;
+			if (nbyte) {
+				COPYSETUP (nbyte);
+				COPY (p1, p2, nbyte, "b");
+			}
+		}
+	}
+	return s1;
+}

c/src/lib/libcpu/m68k/shared/misc/memcpy.c

@@ -0,0 +1,87 @@
+/*
+ *  C library memcpy routine
+ *
+ *  This routine has code to optimize performance on the CPU32+
+ *  and another version for other 68k machines.
+ *
+ *  It could be optimized more for machines with MOVE16 instructions.
+ *
+ *  The routine is placed in this source directory to ensure that it
+ *  is picked up by all applications.
+ *
+ *  W. Eric Norum
+ *  Saskatchewan Accelerator Laboratory
+ *  University of Saskatchewan
+ *  Saskatoon, Saskatchewan, CANADA
+ *  eric@skatter.usask.ca
+ */
+
+#include <string.h>
+#include <rtems/score/m68k.h>
+
+#if defined(__mcpu32__)
+#define COPYSETUP(n)	n--
+#define COPY(to,from,n,size) \
+	asm volatile ("1:\n" \
+	     "\tmove." size " (%0)+,(%1)+\n" \
+	     "\tdbf %2,1b\n"  \
+	     "\tsub.l #0x10000,%2\n" \
+	     "\tbpl.b 1b\n" : \
+		"=a" (from), "=a" (to), "=d" (n) :\
+		 "0" (from),  "1" (to), "2" (n) : \
+		 "cc", "memory")
+#else
+#define COPYSETUP(n)
+#define COPY(to,from,n,size) \
+	asm volatile ("1:\n" \
+	     "\tmove." size " (%0)+,(%1)+\n" \
+	     "\tsubq.l #1,%2\n\tbne.b 1b\n" : \
+		"=a" (from), "=a" (to), "=d" (n) :\
+		 "0" (from),  "1" (to), "2" (n) : \
+		 "cc", "memory")
+#endif
+
+void *
+memcpy(void *s1, const void *s2, size_t n)
+{
+	char *p1 = s1;
+	const char *p2 = s2;
+
+	if (n) {
+		if (n < 16) {
+			COPYSETUP (n);
+			COPY (p1, p2, n, "b");
+		}
+		else {
+			int nbyte;
+			int nl;
+			nbyte = (int)p1 & 0x3;
+			if (nbyte) {
+				nbyte = 4 - nbyte;
+				n -= nbyte;
+				COPYSETUP (nbyte);
+				COPY (p1, p2, nbyte, "b");
+			}
+#if (M68K_HAS_MISALIGNED == 0)
+			/*
+			 * Take care of machines that can't
+			 * handle misaligned references.
+			 */
+			if ((int)p2 & 0x1) {
+				COPYSETUP (n);
+				COPY (p1, p2, n, "b");
+				return s1;
+			}
+#endif
+			nl = (unsigned int)n >> 2;
+			COPYSETUP (nl);
+			COPY (p1, p2, nl, "l");
+			nbyte = (int)n & 0x3;
+			if (nbyte) {
+				COPYSETUP (nbyte);
+				COPY (p1, p2, nbyte, "b");
+			}
+		}
+	}
+	return s1;
+}

cpukit/score/cpu/m68k/memcpy.c

@@ -0,0 +1,87 @@
+/*
+ *  C library memcpy routine
+ *
+ *  This routine has code to optimize performance on the CPU32+
+ *  and another version for other 68k machines.
+ *
+ *  It could be optimized more for machines with MOVE16 instructions.
+ *
+ *  The routine is placed in this source directory to ensure that it
+ *  is picked up by all applications.
+ *
+ *  W. Eric Norum
+ *  Saskatchewan Accelerator Laboratory
+ *  University of Saskatchewan
+ *  Saskatoon, Saskatchewan, CANADA
+ *  eric@skatter.usask.ca
+ */
+
+#include <string.h>
+#include <rtems/score/m68k.h>
+
+#if defined(__mcpu32__)
+#define COPYSETUP(n)	n--
+#define COPY(to,from,n,size) \
+	asm volatile ("1:\n" \
+	     "\tmove." size " (%0)+,(%1)+\n" \
+	     "\tdbf %2,1b\n"  \
+	     "\tsub.l #0x10000,%2\n" \
+	     "\tbpl.b 1b\n" : \
+		"=a" (from), "=a" (to), "=d" (n) :\
+		 "0" (from),  "1" (to), "2" (n) : \
+		 "cc", "memory")
+#else
+#define COPYSETUP(n)
+#define COPY(to,from,n,size) \
+	asm volatile ("1:\n" \
+	     "\tmove." size " (%0)+,(%1)+\n" \
+	     "\tsubq.l #1,%2\n\tbne.b 1b\n" : \
+		"=a" (from), "=a" (to), "=d" (n) :\
+		 "0" (from),  "1" (to), "2" (n) : \
+		 "cc", "memory")
+#endif
+
+void *
+memcpy(void *s1, const void *s2, size_t n)
+{
+	char *p1 = s1;
+	const char *p2 = s2;
+
+	if (n) {
+		if (n < 16) {
+			COPYSETUP (n);
+			COPY (p1, p2, n, "b");
+		}
+		else {
+			int nbyte;
+			int nl;
+			nbyte = (int)p1 & 0x3;
+			if (nbyte) {
+				nbyte = 4 - nbyte;
+				n -= nbyte;
+				COPYSETUP (nbyte);
+				COPY (p1, p2, nbyte, "b");
+			}
+#if (M68K_HAS_MISALIGNED == 0)
+			/*
+			 * Take care of machines that can't
+			 * handle misaligned references.
+			 */
+			if ((int)p2 & 0x1) {
+				COPYSETUP (n);
+				COPY (p1, p2, n, "b");
+				return s1;
+			}
+#endif
+			nl = (unsigned int)n >> 2;
+			COPYSETUP (nl);
+			COPY (p1, p2, nl, "l");
+			nbyte = (int)n & 0x3;
+			if (nbyte) {
+				COPYSETUP (nbyte);
+				COPY (p1, p2, nbyte, "b");
+			}
+		}
+	}
+	return s1;
+}