1/*
2 * arch/xtensa/lib/hal/memcopy.S -- Core HAL library functions
3 * xthal_memcpy and xthal_bcopy
4 *
5 * This file is subject to the terms and conditions of the GNU General Public
6 * License.  See the file "COPYING" in the main directory of this archive
7 * for more details.
8 *
9 * Copyright (C) 2002 - 2005 Tensilica Inc.
10 */
11
12#include <variant/core.h>
13
14	.macro	src_b	r, w0, w1
15#ifdef __XTENSA_EB__
16	src	\r, \w0, \w1
17#else
18	src	\r, \w1, \w0
19#endif
20	.endm
21
22	.macro	ssa8	r
23#ifdef __XTENSA_EB__
24	ssa8b	\r
25#else
26	ssa8l	\r
27#endif
28	.endm
29
30
31/*
32 * void *memcpy(void *dst, const void *src, size_t len);
33 * void *memmove(void *dst, const void *src, size_t len);
34 * void *bcopy(const void *src, void *dst, size_t len);
35 *
36 * This function is intended to do the same thing as the standard
37 * library function memcpy() (or bcopy()) for most cases.
38 * However, where the source and/or destination references
39 * an instruction RAM or ROM or a data RAM or ROM, that
40 * source and/or destination will always be accessed with
41 * 32-bit load and store instructions (as required for these
42 * types of devices).
43 *
44 * !!!!!!!  XTFIXME:
45 * !!!!!!!  Handling of IRAM/IROM has not yet
46 * !!!!!!!  been implemented.
47 *
48 * The bcopy version is provided here to avoid the overhead
49 * of an extra call, for callers that require this convention.
50 *
51 * The (general case) algorithm is as follows:
52 *   If destination is unaligned, align it by conditionally
53 *     copying 1 and 2 bytes.
54 *   If source is aligned,
55 *     do 16 bytes with a loop, and then finish up with
56 *     8, 4, 2, and 1 byte copies conditional on the length;
57 *   else (if source is unaligned),
58 *     do the same, but use SRC to align the source data.
59 *   This code tries to use fall-through branches for the common
60 *     case of aligned source and destination and multiple
61 *     of 4 (or 8) length.
62 *
63 * Register use:
64 *	a0/ return address
65 *	a1/ stack pointer
66 *	a2/ return value
67 *	a3/ src
68 *	a4/ length
69 *	a5/ dst
70 *	a6/ tmp
71 *	a7/ tmp
72 *	a8/ tmp
73 *	a9/ tmp
74 *	a10/ tmp
75 *	a11/ tmp
76 */
77
78	.text
79	.align	4
80	.global	bcopy
81	.type   bcopy,@function
82bcopy:
83	entry	sp, 16		# minimal stack frame
84	# a2=src, a3=dst, a4=len
85	mov	a5, a3		# copy dst so that a2 is return value
86	mov	a3, a2
87	mov	a2, a5
88	j	.Lcommon	# go to common code for memcpy+bcopy
89
90
91/*
92 * Byte by byte copy
93 */
94	.align	4
95	.byte	0		# 1 mod 4 alignment for LOOPNEZ
96				# (0 mod 4 alignment for LBEG)
97.Lbytecopy:
98#if XCHAL_HAVE_LOOPS
99	loopnez	a4, .Lbytecopydone
100#else /* !XCHAL_HAVE_LOOPS */
101	beqz	a4, .Lbytecopydone
102	add	a7, a3, a4	# a7 = end address for source
103#endif /* !XCHAL_HAVE_LOOPS */
104.Lnextbyte:
105	l8ui	a6, a3, 0
106	addi	a3, a3, 1
107	s8i	a6, a5, 0
108	addi	a5, a5, 1
109#if !XCHAL_HAVE_LOOPS
110	blt	a3, a7, .Lnextbyte
111#endif /* !XCHAL_HAVE_LOOPS */
112.Lbytecopydone:
113	retw
114
115/*
116 * Destination is unaligned
117 */
118
119	.align	4
120.Ldst1mod2:	# dst is only byte aligned
121	_bltui	a4, 7, .Lbytecopy	# do short copies byte by byte
122
123	# copy 1 byte
124	l8ui	a6, a3,  0
125	addi	a3, a3,  1
126	addi	a4, a4, -1
127	s8i	a6, a5,  0
128	addi	a5, a5,  1
129	_bbci.l	a5, 1, .Ldstaligned	# if dst is now aligned, then
130					# return to main algorithm
131.Ldst2mod4:	# dst 16-bit aligned
132	# copy 2 bytes
133	_bltui	a4, 6, .Lbytecopy	# do short copies byte by byte
134	l8ui	a6, a3,  0
135	l8ui	a7, a3,  1
136	addi	a3, a3,  2
137	addi	a4, a4, -2
138	s8i	a6, a5,  0
139	s8i	a7, a5,  1
140	addi	a5, a5,  2
141	j	.Ldstaligned	# dst is now aligned, return to main algorithm
142
143	.align	4
144	.global	memcpy
145	.type   memcpy,@function
146memcpy:
147	.global	memmove
148	.type   memmove,@function
149memmove:
150
151	entry	sp, 16		# minimal stack frame
152	# a2/ dst, a3/ src, a4/ len
153	mov	a5, a2		# copy dst so that a2 is return value
154.Lcommon:
155	_bbsi.l	a2, 0, .Ldst1mod2	# if dst is 1 mod 2
156	_bbsi.l	a2, 1, .Ldst2mod4	# if dst is 2 mod 4
157.Ldstaligned:	# return here from .Ldst?mod? once dst is aligned
158	srli	a7, a4, 4	# number of loop iterations with 16B
159				# per iteration
160	movi	a8, 3		# if source is not aligned,
161	_bany	a3, a8, .Lsrcunaligned	# then use shifting copy
162	/*
163	 * Destination and source are word-aligned, use word copy.
164	 */
165	# copy 16 bytes per iteration for word-aligned dst and word-aligned src
166#if XCHAL_HAVE_LOOPS
167	loopnez	a7, .Loop1done
168#else /* !XCHAL_HAVE_LOOPS */
169	beqz	a7, .Loop1done
170	slli	a8, a7, 4
171	add	a8, a8, a3	# a8 = end of last 16B source chunk
172#endif /* !XCHAL_HAVE_LOOPS */
173.Loop1:
174	l32i	a6, a3,  0
175	l32i	a7, a3,  4
176	s32i	a6, a5,  0
177	l32i	a6, a3,  8
178	s32i	a7, a5,  4
179	l32i	a7, a3, 12
180	s32i	a6, a5,  8
181	addi	a3, a3, 16
182	s32i	a7, a5, 12
183	addi	a5, a5, 16
184#if !XCHAL_HAVE_LOOPS
185	blt	a3, a8, .Loop1
186#endif /* !XCHAL_HAVE_LOOPS */
187.Loop1done:
188	bbci.l	a4, 3, .L2
189	# copy 8 bytes
190	l32i	a6, a3,  0
191	l32i	a7, a3,  4
192	addi	a3, a3,  8
193	s32i	a6, a5,  0
194	s32i	a7, a5,  4
195	addi	a5, a5,  8
196.L2:
197	bbsi.l	a4, 2, .L3
198	bbsi.l	a4, 1, .L4
199	bbsi.l	a4, 0, .L5
200	retw
201.L3:
202	# copy 4 bytes
203	l32i	a6, a3,  0
204	addi	a3, a3,  4
205	s32i	a6, a5,  0
206	addi	a5, a5,  4
207	bbsi.l	a4, 1, .L4
208	bbsi.l	a4, 0, .L5
209	retw
210.L4:
211	# copy 2 bytes
212	l16ui	a6, a3,  0
213	addi	a3, a3,  2
214	s16i	a6, a5,  0
215	addi	a5, a5,  2
216	bbsi.l	a4, 0, .L5
217	retw
218.L5:
219	# copy 1 byte
220	l8ui	a6, a3,  0
221	s8i	a6, a5,  0
222	retw
223
224/*
225 * Destination is aligned, Source is unaligned
226 */
227
228	.align	4
229.Lsrcunaligned:
230	_beqz	a4, .Ldone	# avoid loading anything for zero-length copies
231	# copy 16 bytes per iteration for word-aligned dst and unaligned src
232	ssa8	a3		# set shift amount from byte offset
233#define SIM_CHECKS_ALIGNMENT	1	/* set to 1 when running on ISS (simulator) with the
234					   lint or ferret client, or 0 to save a few cycles */
235#if XCHAL_UNALIGNED_LOAD_EXCEPTION || SIM_CHECKS_ALIGNMENT
236	and	a11, a3, a8	# save unalignment offset for below
237	sub	a3, a3, a11	# align a3
238#endif
239	l32i	a6, a3, 0	# load first word
240#if XCHAL_HAVE_LOOPS
241	loopnez	a7, .Loop2done
242#else /* !XCHAL_HAVE_LOOPS */
243	beqz	a7, .Loop2done
244	slli	a10, a7, 4
245	add	a10, a10, a3	# a10 = end of last 16B source chunk
246#endif /* !XCHAL_HAVE_LOOPS */
247.Loop2:
248	l32i	a7, a3,  4
249	l32i	a8, a3,  8
250	src_b	a6, a6, a7
251	s32i	a6, a5,  0
252	l32i	a9, a3, 12
253	src_b	a7, a7, a8
254	s32i	a7, a5,  4
255	l32i	a6, a3, 16
256	src_b	a8, a8, a9
257	s32i	a8, a5,  8
258	addi	a3, a3, 16
259	src_b	a9, a9, a6
260	s32i	a9, a5, 12
261	addi	a5, a5, 16
262#if !XCHAL_HAVE_LOOPS
263	blt	a3, a10, .Loop2
264#endif /* !XCHAL_HAVE_LOOPS */
265.Loop2done:
266	bbci.l	a4, 3, .L12
267	# copy 8 bytes
268	l32i	a7, a3,  4
269	l32i	a8, a3,  8
270	src_b	a6, a6, a7
271	s32i	a6, a5,  0
272	addi	a3, a3,  8
273	src_b	a7, a7, a8
274	s32i	a7, a5,  4
275	addi	a5, a5,  8
276	mov	a6, a8
277.L12:
278	bbci.l	a4, 2, .L13
279	# copy 4 bytes
280	l32i	a7, a3,  4
281	addi	a3, a3,  4
282	src_b	a6, a6, a7
283	s32i	a6, a5,  0
284	addi	a5, a5,  4
285	mov	a6, a7
286.L13:
287#if XCHAL_UNALIGNED_LOAD_EXCEPTION || SIM_CHECKS_ALIGNMENT
288	add	a3, a3, a11	# readjust a3 with correct misalignment
289#endif
290	bbsi.l	a4, 1, .L14
291	bbsi.l	a4, 0, .L15
292.Ldone:	retw
293.L14:
294	# copy 2 bytes
295	l8ui	a6, a3,  0
296	l8ui	a7, a3,  1
297	addi	a3, a3,  2
298	s8i	a6, a5,  0
299	s8i	a7, a5,  1
300	addi	a5, a5,  2
301	bbsi.l	a4, 0, .L15
302	retw
303.L15:
304	# copy 1 byte
305	l8ui	a6, a3,  0
306	s8i	a6, a5,  0
307	retw
308
309/*
310 * Local Variables:
311 * mode:fundamental
312 * comment-start: "# "
313 * comment-start-skip: "# *"
314 * End:
315 */
316