From: Michal Simek <[EMAIL PROTECTED]>

Signed-off-by: Michal Simek <[EMAIL PROTECTED]>
---
 arch/microblaze/lib/memcpy.c  |  159 +++++++++++++++++++++++++++++++++++++
 arch/microblaze/lib/memmove.c |  174 +++++++++++++++++++++++++++++++++++++++++
 arch/microblaze/lib/memset.c  |   83 +++++++++++++++++++
 3 files changed, 416 insertions(+), 0 deletions(-)
 create mode 100644 arch/microblaze/lib/memcpy.c
 create mode 100644 arch/microblaze/lib/memmove.c
 create mode 100644 arch/microblaze/lib/memset.c

diff --git a/arch/microblaze/lib/memcpy.c b/arch/microblaze/lib/memcpy.c
new file mode 100644
index 0000000..cc13ebd
--- /dev/null
+++ b/arch/microblaze/lib/memcpy.c
@@ -0,0 +1,159 @@
+/* Filename: memcpy.c
+ *
+ * Reasonably optimised generic C-code for memcpy on Microblaze
+ * This is generic C code to do efficient, alignment-aware memcpy.
+ *
+ * It is based on demo code originally Copyright 2001 by Intel Corp, taken from
+ * http://www.embedded.com/showArticle.jhtml?articleID=19205567
+ *
+ * Attempts were made, unsuccesfully, to contact the original
+ * author of this code (Michael Morrow, Intel).  Below is the original
+ * copyright notice.
+ *
+ * This software has been developed by Intel Corporation.
+ * Intel specifically disclaims all warranties, express or
+ * implied, and all liability, including consequential and
+ * other indirect damages, for the use of this program, including
+ * liability for infringement of any proprietary rights,
+ * and including the warranties of merchantability and fitness
+ * for a particular purpose. Intel does not assume any
+ * responsibility for and errors which may appear in this program
+ * not any responsibility to update it.
+ */
+
+#include <linux/types.h>
+#include <linux/stddef.h>
+#include <linux/compiler.h>
+
+#include <asm/string.h>
+#include <asm/system.h>
+
+/* Macro's for bytebliting unaligned data blocks */
+
+/* Big-endian MACROS */
+
+#define BYTE_BLIT_INIT(s, h, o) \
+       (h) = *((unsigned *)(s))++ << (o)
+
+#define BYTE_BLIT_STEP(d, s, h, o) \
+       { register unsigned _v_; _v_ = *((unsigned *)(s))++; \
+        *((unsigned *)(d))++ = (h) | _v_ >> (32-(o)); \
+        (h) = _v_ << (o); \
+       }
+
+void *memcpy(void *d, const void *s, __kernel_size_t c)
+{
+#if 1
+       /* This code was put into place as we transitioned from gcc3 to gcc4.
+        * gcc4 does not support the syntax *((char *)d)++ which causes the
+        * original code below to break. Short of fixing up the original code,
+        * a simpler byte oriented code was put into place.
+        */
+       /* Simple, byte oriented memcpy. */
+       const char      *src = s;
+       char            *dst = d;
+
+       while (c--) *dst++ = *src++;
+
+       return d;
+
+#else
+       /* The following code tries to optimize the copy by using unsigned
+        * alignment. This will work fine if both source and destination are
+        * aligned on the same boundary. However, if they are aligned on
+        * different boundaries shifts will be necessary. This might result in
+        * bad performance on MicroBlaze systems without a barrel shifter.
+        */
+       void *r = d;
+
+       if (c >= 4) {
+               unsigned x, a, h, align;
+
+               /* Align the destination to a word boundry. */
+               /* This is done in an endian independant manner. */
+               switch ((unsigned) d & 3) {
+               case 1:
+                       *((char *) d)++ = *((char *) s)++;
+                       c--;
+               case 2:
+                       *((char *) d)++ = *((char *) s)++;
+                       c--;
+               case 3:
+                       *((char *) d)++ = *((char *) s)++;
+                       c--;
+               }
+               /* Choose a copy scheme based on the source */
+               /* alignment relative to destination. */
+               switch ((unsigned) s & 3) {
+               case 0x0:       /* Both byte offsets are aligned */
+
+                       for (; c >= 4; c -= 4)
+                               *((unsigned *) d)++ = *((unsigned *) s)++;
+
+                       break;
+
+               case 0x1:       /* Unaligned - Off by 1 */
+                       /* Word align the source */
+                       a = (unsigned) s & ~3;
+
+                       /* Load the holding buffer */
+                       BYTE_BLIT_INIT(a, h, 8);
+
+                       for (; c >= 4; c -= 4)
+                               BYTE_BLIT_STEP(d, a, h, 8);
+
+                       /* Realign the source */
+                       (unsigned) s = a - 3;
+                       break;
+
+               case 0x2:       /* Unaligned - Off by 2 */
+                       /* Word align the source */
+                       a = (unsigned) s & ~3;
+
+                       /* Load the holding buffer */
+                       BYTE_BLIT_INIT(a, h, 16);
+
+                       for (; c >= 4; c -= 4)
+                               BYTE_BLIT_STEP(d, a, h, 16);
+
+                       /* Realign the source */
+                       (unsigned) s = a - 2;
+                       break;
+
+               case 0x3:       /* Unaligned - Off by 3 */
+                       /* Word align the source */
+                       a = (unsigned) s & ~3;
+
+                       /* Load the holding buffer */
+                       BYTE_BLIT_INIT(a, h, 24);
+
+                       for (; c >= 4; c -= 4)
+                               BYTE_BLIT_STEP(d, a, h, 24);
+
+                       /* Realign the source */
+                       (unsigned) s = a - 1;
+                       break;
+               }
+
+       }
+
+       /* Finish off any remaining bytes */
+       /* simple fast copy, ... unless a cache boundry is crossed */
+       switch (c) {
+       case 3:
+               *((char *) d)++ = *((char *) s)++;
+       case 2:
+               *((char *) d)++ = *((char *) s)++;
+       case 1:
+               *((char *) d)++ = *((char *) s)++;
+       }
+
+       return r;
+#endif
+}
+
+
+void *cacheable_memcpy(void *d, const void *s, __kernel_size_t c)
+{
+       return memcpy(d, s, c);
+}
diff --git a/arch/microblaze/lib/memmove.c b/arch/microblaze/lib/memmove.c
new file mode 100644
index 0000000..e9f8f04
--- /dev/null
+++ b/arch/microblaze/lib/memmove.c
@@ -0,0 +1,174 @@
+/* Filename: memmove.c
+ *
+ * Reasonably optimised generic C-code for memcpy on Microblaze
+ * This is generic C code to do efficient, alignment-aware memmove.
+ *
+ * It is based on demo code originally Copyright 2001 by Intel Corp, taken from
+ * http://www.embedded.com/showArticle.jhtml?articleID=19205567
+ *
+ * Attempts were made, unsuccesfully, to contact the original
+ * author of this code (Michael Morrow, Intel).  Below is the original
+ * copyright notice.
+ *
+ * This software has been developed by Intel Corporation.
+ * Intel specifically disclaims all warranties, express or
+ * implied, and all liability, including consequential and
+ * other indirect damages, for the use of this program, including
+ * liability for infringement of any proprietary rights,
+ * and including the warranties of merchantability and fitness
+ * for a particular purpose. Intel does not assume any
+ * responsibility for and errors which may appear in this program
+ * not any responsibility to update it.
+ */
+
+#include <linux/types.h>
+#include <linux/stddef.h>
+#include <linux/compiler.h>
+
+#include <asm/string.h>
+
+/* Macro's for bytebliting unaligned data blocks */
+
+/* Big-endian MACROS */
+
+#define BYTE_BLIT_INIT(s, h, o) \
+       (h) = *(--(unsigned *)(s)) >> (32-(o))
+
+#define BYTE_BLIT_STEP(d, s, h, o) \
+       { register unsigned _v_; _v_ = *(--(unsigned *)(s)); \
+        *(--(unsigned *)(d)) = _v_ << (o) | (h); \
+        (h) = _v_ >> (32-(o)); \
+       }
+
+void *memmove(void *d, const void *s, __kernel_size_t c)
+{
+#if 1
+       /* This code was put into place as we transitioned from gcc3 to gcc4.
+        * gcc4 does not support the syntax *((char *)d)++ which causes the
+        * original code below to break. Short of fixing up the original code,
+        * a simpler byte oriented code was put into place.
+        */
+       /* Simple, byte oriented memmove. */
+       if (d <= s) {
+               return memcpy(d, s, c);
+       } else {
+               const char      *src = s;
+               char            *dst = d;
+
+               /* copy backwards, from end to beginning */
+               src += c;
+               dst += c;
+
+               while (c--) *--dst = *--src;
+
+               return d;
+       }
+
+#else
+       /* The following code tries to optimize the copy by using unsigned
+        * alignment. This will work fine if both source and destination are
+        * aligned on the same boundary. However, if they are aligned on
+        * different boundaries shifts will be necessary. This might result in
+        * bad performance on MicroBlaze systems without a barrel shifter.
+        */
+       void *r = d;
+
+       /* Use memcpy when source is higher than dest */
+       if (d <= s)
+               return memcpy(d, s, c);
+
+       /* Do a descending copy - this is a bit trickier! */
+       d += c;
+       s += c;
+
+       if (c >= 4) {
+               unsigned x, a, h, align;
+
+               /* Align the destination to a word boundry. */
+               /* This is done in an endian independant manner. */
+               switch ((unsigned) d & 3) {
+               case 3:
+                       *(--(char *) d) = *(--(char *) s);
+                       c--;
+               case 2:
+                       *(--(char *) d) = *(--(char *) s);
+                       c--;
+               case 1:
+                       *(--(char *) d) = *(--(char *) s);
+                       c--;
+               }
+               /* Choose a copy scheme based on the source */
+               /* alignment relative to destination. */
+               switch ((unsigned) s & 3) {
+               case 0x0:       /* Both byte offsets are aligned */
+
+                       for (; c >= 4; c -= 4)
+                               *(--(unsigned *) d) = *(--(unsigned *) s);
+
+                       break;
+
+               case 0x1:       /* Unaligned - Off by 1 */
+                       /* Word align the source */
+                       a = (unsigned) (s + 4) & ~3;
+
+                       /* Load the holding buffer */
+                       BYTE_BLIT_INIT(a, h, 8);
+
+                       for (; c >= 4; c -= 4)
+                               BYTE_BLIT_STEP(d, a, h, 8);
+
+                       /* Realign the source */
+                       (unsigned) s = a + 1;
+                       break;
+
+               case 0x2:       /* Unaligned - Off by 2 */
+                       /* Word align the source */
+                       a = (unsigned) (s + 4) & ~3;
+
+                       /* Load the holding buffer */
+                       BYTE_BLIT_INIT(a, h, 16);
+
+                       for (; c >= 4; c -= 4)
+                               BYTE_BLIT_STEP(d, a, h, 16);
+
+                       /* Realign the source */
+                       (unsigned) s = a + 2;
+                       break;
+
+               case 0x3:       /* Unaligned - Off by 3 */
+                       /* Word align the source */
+                       a = (unsigned) (s + 4) & ~3;
+
+                       /* Load the holding buffer */
+                       BYTE_BLIT_INIT(a, h, 24);
+
+                       for (; c >= 4; c -= 4)
+                               BYTE_BLIT_STEP(d, a, h, 24);
+
+                       /* Realign the source */
+                       (unsigned) s = a + 3;
+                       break;
+               }
+
+#if 0
+               /* Finish off any remaining bytes */
+               c &= 3;
+               goto finish;
+#endif
+
+       }
+
+       /* simple fast copy, ... unless a cache boundry is crossed */
+       switch (c) {
+       case 4:
+               *(--(char *) d) = *(--(char *) s);
+       case 3:
+               *(--(char *) d) = *(--(char *) s);
+       case 2:
+               *(--(char *) d) = *(--(char *) s);
+       case 1:
+               *(--(char *) d) = *(--(char *) s);
+       }
+       return r;
+#endif
+}
diff --git a/arch/microblaze/lib/memset.c b/arch/microblaze/lib/memset.c
new file mode 100644
index 0000000..5d3d83e
--- /dev/null
+++ b/arch/microblaze/lib/memset.c
@@ -0,0 +1,83 @@
+/* Filename: memset.c
+ *
+ * Reasonably optimised generic C-code for memset on Microblaze
+ * This is generic C code to do efficient, alignment-aware memcpy.
+ *
+ * It is based on demo code originally Copyright 2001 by Intel Corp, taken from
+ * http://www.embedded.com/showArticle.jhtml?articleID=19205567
+ *
+ * Attempts were made, unsuccesfully, to contact the original
+ * author of this code (Michael Morrow, Intel).  Below is the original
+ * copyright notice.
+ *
+ * This software has been developed by Intel Corporation.
+ * Intel specifically disclaims all warranties, express or
+ * implied, and all liability, including consequential and
+ * other indirect damages, for the use of this program, including
+ * liability for infringement of any proprietary rights,
+ * and including the warranties of merchantability and fitness
+ * for a particular purpose. Intel does not assume any
+ * responsibility for and errors which may appear in this program
+ * not any responsibility to update it.
+ */
+
+/* Filename: memcpy.c
+ *
+ * Reasonably optimised generic C-code for memset on Microblaze
+ * Based on demo code originally Copyright 2001 by Intel Corp.
+ *
+ * This software has been developed by Intel Corporation.
+ * Intel specifically disclaims all warranties, express or
+ * implied, and all liability, including consequential and
+ * other indirect damages, for the use of this program, including
+ * liability for infringement of any proprietary rights,
+ * and including the warranties of merchantability and fitness
+ * for a particular purpose. Intel does not assume any
+ * responsibility for and errors which may appear in this program
+ * not any responsibility to update it.
+ */
+
+#include <linux/types.h>
+#include <linux/stddef.h>
+#include <linux/compiler.h>
+
+#include <asm/string.h>
+
+void *memset(void *s, int c, __kernel_size_t n)
+{
+       void *r = s;
+       unsigned int w32;
+
+       /* Truncate c to 8 bits */
+       w32 = c = (c & 0xFF);
+
+       /* Make a repeating word out of it */
+       w32 |= w32 << 8;
+       w32 |= w32 << 8;
+       w32 |= w32 << 8;
+
+       if (n >= 4) {
+               /* Align the destination to a word boundary. */
+               /* This is done in an endian independant manner. */
+               switch ((unsigned) s & 3) {
+               case 1: *(char *)s = c; s = (char *)s+1; n--;
+               case 2: *(char *)s = c; s = (char *)s+1; n--;
+               case 3: *(char *)s = c; s = (char *)s+1; n--;
+               }
+
+               /* Do as many full-word copies as we can */
+               for (; n >= 4; n -= 4) {
+                       *(unsigned *)s = w32;
+                       s = (unsigned *)s + 1;
+               }
+
+       }
+
+       /* Finish off the rest as byte sets */
+       while (n--) {
+               *(char *)s = c;
+               s = (char *)s+1;
+       }
+       return r;
+}
+
-- 
1.5.4.rc4.14.g6fc74

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Reply via email to