memcpy.c

  1/*
  2 * Copyright (C) 2008-2009 Michal Simek <monstr@monstr.eu>
  3 * Copyright (C) 2008-2009 PetaLogix
  4 * Copyright (C) 2007 John Williams
  5 *
  6 * Reasonably optimised generic C-code for memcpy on Microblaze
  7 * This is generic C code to do efficient, alignment-aware memcpy.
  8 *
  9 * It is based on demo code originally Copyright 2001 by Intel Corp, taken from
 10 * http://www.embedded.com/showArticle.jhtml?articleID=19205567
 11 *
 12 * Attempts were made, unsuccessfully, to contact the original
 13 * author of this code (Michael Morrow, Intel).  Below is the original
 14 * copyright notice.
 15 *
 16 * This software has been developed by Intel Corporation.
 17 * Intel specifically disclaims all warranties, express or
 18 * implied, and all liability, including consequential and
 19 * other indirect damages, for the use of this program, including
 20 * liability for infringement of any proprietary rights,
 21 * and including the warranties of merchantability and fitness
 22 * for a particular purpose. Intel does not assume any
 23 * responsibility for and errors which may appear in this program
 24 * not any responsibility to update it.
 25 */
 26
 27#include <linux/export.h>
 28#include <linux/types.h>
 29#include <linux/stddef.h>
 30#include <linux/compiler.h>
 31
 32#include <linux/string.h>
 33
 34#ifdef CONFIG_OPT_LIB_FUNCTION
 35void *memcpy(void *v_dst, const void *v_src, __kernel_size_t c)
 36{
 37	const char *src = v_src;
 38	char *dst = v_dst;
 39
 40	/* The following code tries to optimize the copy by using unsigned
 41	 * alignment. This will work fine if both source and destination are
 42	 * aligned on the same boundary. However, if they are aligned on
 43	 * different boundaries shifts will be necessary. This might result in
 44	 * bad performance on MicroBlaze systems without a barrel shifter.
 45	 */
 46	const uint32_t *i_src;
 47	uint32_t *i_dst;
 48
 49	if (likely(c >= 4)) {
 50		unsigned  value, buf_hold;
 51
 52		/* Align the destination to a word boundary. */
 53		/* This is done in an endian independent manner. */
 54		switch ((unsigned long)dst & 3) {
 55		case 1:
 56			*dst++ = *src++;
 57			--c;
 58			fallthrough;
 59		case 2:
 60			*dst++ = *src++;
 61			--c;
 62			fallthrough;
 63		case 3:
 64			*dst++ = *src++;
 65			--c;
 66		}
 67
 68		i_dst = (void *)dst;
 69
 70		/* Choose a copy scheme based on the source */
 71		/* alignment relative to destination. */
 72		switch ((unsigned long)src & 3) {
 73		case 0x0:	/* Both byte offsets are aligned */
 74			i_src  = (const void *)src;
 75
 76			for (; c >= 4; c -= 4)
 77				*i_dst++ = *i_src++;
 78
 79			src  = (const void *)i_src;
 80			break;
 81		case 0x1:	/* Unaligned - Off by 1 */
 82			/* Word align the source */
 83			i_src = (const void *) ((unsigned)src & ~3);
 84#ifndef __MICROBLAZEEL__
 85			/* Load the holding buffer */
 86			buf_hold = *i_src++ << 8;
 87
 88			for (; c >= 4; c -= 4) {
 89				value = *i_src++;
 90				*i_dst++ = buf_hold | value >> 24;
 91				buf_hold = value << 8;
 92			}
 93#else
 94			/* Load the holding buffer */
 95			buf_hold = (*i_src++ & 0xFFFFFF00) >> 8;
 96
 97			for (; c >= 4; c -= 4) {
 98				value = *i_src++;
 99				*i_dst++ = buf_hold | ((value & 0xFF) << 24);
100				buf_hold = (value & 0xFFFFFF00) >> 8;
101			}
102#endif
103			/* Realign the source */
104			src = (const void *)i_src;
105			src -= 3;
106			break;
107		case 0x2:	/* Unaligned - Off by 2 */
108			/* Word align the source */
109			i_src = (const void *) ((unsigned)src & ~3);
110#ifndef __MICROBLAZEEL__
111			/* Load the holding buffer */
112			buf_hold = *i_src++ << 16;
113
114			for (; c >= 4; c -= 4) {
115				value = *i_src++;
116				*i_dst++ = buf_hold | value >> 16;
117				buf_hold = value << 16;
118			}
119#else
120			/* Load the holding buffer */
121			buf_hold = (*i_src++ & 0xFFFF0000) >> 16;
122
123			for (; c >= 4; c -= 4) {
124				value = *i_src++;
125				*i_dst++ = buf_hold | ((value & 0xFFFF) << 16);
126				buf_hold = (value & 0xFFFF0000) >> 16;
127			}
128#endif
129			/* Realign the source */
130			src = (const void *)i_src;
131			src -= 2;
132			break;
133		case 0x3:	/* Unaligned - Off by 3 */
134			/* Word align the source */
135			i_src = (const void *) ((unsigned)src & ~3);
136#ifndef __MICROBLAZEEL__
137			/* Load the holding buffer */
138			buf_hold = *i_src++ << 24;
139
140			for (; c >= 4; c -= 4) {
141				value = *i_src++;
142				*i_dst++ = buf_hold | value >> 8;
143				buf_hold = value << 24;
144			}
145#else
146			/* Load the holding buffer */
147			buf_hold = (*i_src++ & 0xFF000000) >> 24;
148
149			for (; c >= 4; c -= 4) {
150				value = *i_src++;
151				*i_dst++ = buf_hold | ((value & 0xFFFFFF) << 8);
152				buf_hold = (value & 0xFF000000) >> 24;
153			}
154#endif
155			/* Realign the source */
156			src = (const void *)i_src;
157			src -= 1;
158			break;
159		}
160		dst = (void *)i_dst;
161	}
162
163	/* Finish off any remaining bytes */
164	/* simple fast copy, ... unless a cache boundary is crossed */
165	switch (c) {
166	case 3:
167		*dst++ = *src++;
168		fallthrough;
169	case 2:
170		*dst++ = *src++;
171		fallthrough;
172	case 1:
173		*dst++ = *src++;
174	}
175
176	return v_dst;
177}
178EXPORT_SYMBOL(memcpy);
179#endif /* CONFIG_OPT_LIB_FUNCTION */