libc/string: Add BSD-optimized memset implementation.

Add lib_bsdmemset.c from newlib (Red Hat BSD license) which provides
an optimized memset using word-aligned writes and loop unrolling for
improved performance on larger buffers.

Signed-off-by: anjiahao <anjiahao@xiaomi.com>

libs/libc/string/CMakeLists.txt

@@ -59,7 +59,6 @@ set(SRCS
     lib_mempcpy.c
     lib_rawmemchr.c
     lib_memmove.c
-    lib_memset.c
     lib_strlcat.c
     lib_strlcpy.c
     lib_strnlen.c)
@@ -83,6 +82,7 @@ if(CONFIG_LIBC_NEWLIB_OPTSPEED)
     lib_bsdmemccpy.c
     lib_bsdmemcmp.c
     lib_bsdmemrchr.c
+    lib_bsdmemset.c
     lib_bsdstpncpy.c
     lib_bsdstrchr.c
     lib_bsdstrcmp.c
@@ -102,6 +102,7 @@ else()
     lib_memccpy.c
     lib_memcmp.c
     lib_memrchr.c
+    lib_memset.c
     lib_stpncpy.c
     lib_strchr.c
     lib_strcmp.c

libs/libc/string/Make.defs

@@ -34,7 +34,7 @@ CSRCS += lib_strsep.c lib_strerrorr.c lib_explicit_bzero.c lib_strsignal.c
 CSRCS += lib_index.c lib_rindex.c lib_timingsafe_bcmp.c lib_strverscmp.c
 CSRCS += lib_mempcpy.c lib_rawmemchr.c
 
-CSRCS += lib_memmove.c lib_memset.c
+CSRCS += lib_memmove.c
 CSRCS += lib_strlcat.c
 CSRCS += lib_strlcpy.c lib_strnlen.c
 
@@ -54,12 +54,12 @@ ifeq ($(CONFIG_LIBC_NEWLIB_OPTSPEED),y)
 CSRCS += lib_bsdmemccpy.c lib_bsdmemcmp.c lib_bsdmemrchr.c lib_bsdstpncpy.c
 CSRCS += lib_bsdstrchr.c lib_bsdstrcmp.c lib_bsdstrlen.c lib_bsdstrncpy.c
 CSRCS += lib_bsdmemchr.c  lib_bsdstpcpy.c lib_bsdstrcat.c lib_bsdstrchrnul.c
-CSRCS += lib_bsdstrcpy.c lib_bsdstrncmp.c lib_bsdstrrchr.c
+CSRCS += lib_bsdstrcpy.c lib_bsdstrncmp.c lib_bsdstrrchr.c lib_memset.c
 else
 CSRCS += lib_memccpy.c lib_memcmp.c lib_memrchr.c lib_stpncpy.c
 CSRCS += lib_strchr.c lib_strcmp.c lib_strlen.c lib_strncpy.c
 CSRCS += lib_memchr.c lib_stpcpy.c lib_strcat.c lib_strchrnul.c
-CSRCS += lib_strcpy.c lib_strncmp.c lib_strrchr.c
+CSRCS += lib_strcpy.c lib_strncmp.c lib_strrchr.c lib_memset.c
 endif
 
 # Add the string directory to the build

libs/libc/string/lib_bsdmemset.c

@@ -0,0 +1,113 @@
+/****************************************************************************
+ * libs/libc/string/lib_bsdmemset.c
+ *
+ * SPDX-License-Identifier: BSD
+ * SPDX-FileCopyrightText: 1994-2009  Red Hat, Inc. All rights reserved
+ *
+ * Copyright (c) 1994-2009  Red Hat, Inc. All rights reserved.
+ *
+ * This copyrighted material is made available to anyone wishing to use,
+ * modify, copy, or redistribute it subject to the terms and conditions
+ * of the BSD License.   This program is distributed in the hope that
+ * it will be useful, but WITHOUT ANY WARRANTY expressed or implied,
+ * including the implied warranties of MERCHANTABILITY or FITNESS FOR
+ * A PARTICULAR PURPOSE.  A copy of this license is available at
+ * http://www.opensource.org/licenses. Any Red Hat trademarks that are
+ * incorporated in the source code or documentation are not subject to
+ * the BSD License and may only be used or replicated with the express
+ * permission of Red Hat, Inc.
+ *
+ ****************************************************************************/
+
+/****************************************************************************
+ * Included Files
+ ****************************************************************************/
+
+#include <nuttx/config.h>
+#include <sys/types.h>
+#include <string.h>
+
+#include "libc.h"
+
+/****************************************************************************
+ * Pre-processor Definitions
+ ****************************************************************************/
+
+#define LBLOCKSIZE     (sizeof(long))
+#define UNALIGNED(x)   ((long)(x) & (LBLOCKSIZE - 1))
+#define TOO_SMALL(len) ((len) < LBLOCKSIZE)
+
+/****************************************************************************
+ * Public Functions
+ ****************************************************************************/
+
+/****************************************************************************
+ * Name: memset
+ ****************************************************************************/
+
+#if !defined(CONFIG_LIBC_ARCH_MEMSET) && defined(LIBC_BUILD_MEMSET)
+#undef memset
+no_builtin("memset")
+FAR void *memset(FAR void *m, int c, size_t n)
+{
+  FAR unsigned long *aligned_addr;
+  FAR char *s = (FAR char *)m;
+  unsigned long buffer;
+  int i;
+
+  /* To avoid sign extension, copy C to an unsigned variable.  */
+
+  while (UNALIGNED(s))
+    {
+      if (n--)
+        {
+          *s++ = c;
+        }
+      else
+        {
+          return m;
+        }
+    }
+
+  if (!TOO_SMALL(n))
+    {
+      /* If we get this far, we know that n is large and s is word-aligned. */
+
+      aligned_addr = (FAR unsigned long *)s;
+      buffer = ((unsigned int)c << 8) | c;
+      buffer |= (buffer << 16);
+      for (i = 32; i < LBLOCKSIZE * 8; i <<= 1)
+        {
+          buffer = (buffer << i) | buffer;
+        }
+
+      /* Unroll the loop.  */
+
+      while (n >= LBLOCKSIZE * 4)
+        {
+          *aligned_addr++ = buffer;
+          *aligned_addr++ = buffer;
+          *aligned_addr++ = buffer;
+          *aligned_addr++ = buffer;
+          n -= 4 * LBLOCKSIZE;
+        }
+
+      while (n >= LBLOCKSIZE)
+        {
+          *aligned_addr++ = buffer;
+          n -= LBLOCKSIZE;
+        }
+
+      /* Pick up the remainder with a bytewise loop.  */
+
+      s = (FAR char *)aligned_addr;
+    }
+
+  while (n--)
+    {
+      *s++ = c;
+    }
+
+  return m;
+}
+#endif