xref: /netbsd-src/sys/lib/libkern/arch/hppa/bcopy.S (revision 404fbe5fb94ca1e054339640cabb2801ce52dd30)
1/*	$NetBSD: bcopy.S,v 1.7 2008/04/28 20:24:06 martin Exp $	*/
2
3/*
4 * Copyright (c) 2002 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Matthew Fredette.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 *    notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 *    notice, this list of conditions and the following disclaimer in the
17 *    documentation and/or other materials provided with the distribution.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 * POSSIBILITY OF SUCH DAMAGE.
30 */
31
32/*
33 * Copy routines for NetBSD/hppa.
34 */
35
36#undef _LOCORE
37#define _LOCORE	/* XXX fredette - unfortunate */
38#include <machine/asm.h>
39#include <machine/frame.h>
40
41#if defined(LIBC_SCCS) && !defined(lint)
42RCSID("$NetBSD: bcopy.S,v 1.7 2008/04/28 20:24:06 martin Exp $")
43#endif /* LIBC_SCCS and not lint */
44
45/*
46 * The stbys instruction is a little asymmetric.  When (%r2 & 3)
47 * is zero, stbys,b,m %r1, 4(%r2) works like stws,ma.  You
48 * might then wish that when (%r2 & 3) == 0, stbys,e,m %r1, -4(%r2)
49 * worked like stws,mb.  But it doesn't.
50 *
51 * This macro works around this problem.  It requires that %t2
52 * hold the number of bytes that will be written by this store
53 * (meaning that it ranges from one to four).
54 *
55 * Watch the delay-slot trickery here.  The comib is used to set
56 * up which instruction, either the stws or the stbys, is run
57 * in the delay slot of the b instruction.
58 */
59#define _STBYS_E_M(r, dst_spc, dst_off)				  \
60	comib,<>	4, %t2, 4				! \
61	b		4					! \
62	stws,mb		r, -4(dst_spc, dst_off)			! \
63	stbys,e,m	r, 0(dst_spc, dst_off)
64
65/*
66 * This macro does a bulk copy with no shifting.  cmplt and m are
67 * the completer and displacement multiplier, respectively, for
68 * the load and store instructions.
69 */
70#define _COPY(src_spc, src_off, dst_spc, dst_off, count, cmplt, m) \
71								! \
72	/*							! \
73	 * Loop storing 16 bytes at a time.  Since count 	! \
74	 * may be > INT_MAX, we have to be careful and		! \
75	 * avoid comparisons that treat it as a signed 		! \
76	 * quantity, until after this loop, when count		! \
77	 * is guaranteed to be less than 16.			! \
78	 */							! \
79	comib,>>=,n	15, count, _LABEL(_skip16)		! \
80.label _LABEL(_loop16)						! \
81	addi		-16, count, count			! \
82	ldws,cmplt	m*4(src_spc, src_off), %t1		! \
83	ldws,cmplt	m*4(src_spc, src_off), %t2		! \
84	ldws,cmplt	m*4(src_spc, src_off), %t3		! \
85	ldws,cmplt	m*4(src_spc, src_off), %t4		! \
86	stws,cmplt	%t1, m*4(dst_spc, dst_off)		! \
87	stws,cmplt	%t2, m*4(dst_spc, dst_off)		! \
88	stws,cmplt	%t3, m*4(dst_spc, dst_off)		! \
89	comib,<<	15, count, _LABEL(_loop16)		! \
90	stws,cmplt	%t4, m*4(dst_spc, dst_off)		! \
91.label _LABEL(_skip16)						! \
92								! \
93	/* Loop storing 4 bytes at a time. */			! \
94	addib,<,n	-4, count, _LABEL(_skip4)		! \
95.label _LABEL(_loop4)						! \
96	ldws,cmplt	m*4(src_spc, src_off), %t1		! \
97	addib,>=	-4, count, _LABEL(_loop4)		! \
98	stws,cmplt	%t1, m*4(dst_spc, dst_off)		! \
99.label _LABEL(_skip4)						! \
100	/* Restore the correct count. */			! \
101	addi		4, count, count				! \
102								! \
103.label _LABEL(_do1)						! \
104								! \
105	/* Loop storing 1 byte at a time. */			! \
106	addib,<,n	-1, count, _LABEL(_skip1)		! \
107.label _LABEL(_loop1)						! \
108	ldbs,cmplt	m*1(src_spc, src_off), %t1		! \
109	addib,>=	-1, count, _LABEL(_loop1)		! \
110	stbs,cmplt	%t1, m*1(dst_spc, dst_off)		! \
111.label _LABEL(_skip1)						! \
112	/* Restore the correct count. */			! \
113	b		_LABEL(_done)				! \
114	addi		1, count, count
115
116/*
117 * This macro is definitely strange.  It exists purely to
118 * allow the _COPYS macro to be reused, but because it
119 * requires this long attempt to explain it, I'm starting
120 * to doubt the value of that.
121 *
122 * Part of the expansion of the _COPYS macro below are loops
123 * that copy four words or one word at a time, performing shifts
124 * to get data to line up correctly in the destination buffer.
125 *
126 * The _COPYS macro is used when copying backwards, as well
127 * as forwards.  The 4-word loop always loads into %t1, %t2, %t3,
128 * and %t4 in that order.  This means that when copying forward,
129 * %t1 will have the word from the lowest address, and %t4 will
130 * have the word from the highest address.  When copying
131 * backwards, the opposite is true.
132 *
133 * The shift instructions need pairs of registers with adjacent
134 * words, with the register containing the word from the lowest
135 * address *always* coming first.  It is this assymetry that
136 * gives rise to this macro - depending on which direction
137 * we're copying in, these ordered pairs are different.
138 *
139 * Fortunately, we can compute those register numbers at compile
140 * time, and assemble them manually into a shift instruction.
141 * That's what this macro does.
142 *
143 * This macro takes two arguments.  n ranges from 0 to 3 and
144 * is the "shift number", i.e., n = 0 means we're doing the
145 * shift for what will be the first store.
146 *
147 * m is the displacement multiplier from the _COPYS macro call.
148 * This is 1 for a forward copy and -1 for a backwards copy.
149 * So, the ((m + 1) / 2) term yields 0 for a backwards copy and
150 * 1 for a forward copy, and the ((m - 1) / 2) term yields
151 * 0 for a forward copy, and -1 for a backwards copy.
152 * These terms are used to discriminate the register computations
153 * below.
154 *
155 * When copying forward, then, the first register used with
156 * the first vshd will be 19 + (3 - ((0 - 1) & 3)), or %t4,
157 * which matches _COPYS' requirement that the word last loaded
158 * be in %t4.  The first register used for the second vshd
159 * will then "wrap" around to 19 + (3 - ((1 - 1) & 3)), or %t1.
160 * And so on to %t2 and %t3.
161 *
162 * When copying forward, the second register used with the first
163 * vshd will be (19 + (3 - ((n + 0) & 3)), or %t1.  It will
164 * continue to be %t2, then %t3, and finally %t4.
165 *
166 * When copying backwards, the values for the first and second
167 * register for each vshd are reversed from the forwards case.
168 * (Symmetry reclaimed!)  Proving this is "left as an exercise
169 * for the reader" (remember the different discriminating values!)
170 */
171#define _VSHD(n, m, t)						  \
172	.word (0xd0000000					| \
173	((19 + (3 - ((n - 1 * ((m + 1) / 2)) & 3))) << 16)	| \
174	((19 + (3 - ((n + 1 * ((m - 1) / 2)) & 3))) << 21)	| \
175	(t))
176
177/*
178 * This macro does a bulk copy with shifting.  cmplt and m are
179 * the completer and displacement multiplier, respectively, for
180 * the load and store instructions.  It is assumed that the
181 * word last loaded is already in %t4.
182 */
183#define _COPYS(src_spc, src_off, dst_spc, dst_off, count, cmplt, m) \
184								! \
185	/*							! \
186	 * Loop storing 16 bytes at a time.  Since count 	! \
187	 * may be > INT_MAX, we have to be careful and		! \
188	 * avoid comparisons that treat it as a signed 		! \
189	 * quantity, until after this loop, when count		! \
190	 * is guaranteed to be less than 16.			! \
191	 */							! \
192	comib,>>=,n	15, count, _LABEL(S_skip16)		! \
193.label _LABEL(S_loop16)						! \
194	addi		-16, count, count			! \
195	ldws,cmplt	m*4(src_spc, src_off), %t1		! \
196	ldws,cmplt	m*4(src_spc, src_off), %t2		! \
197	ldws,cmplt	m*4(src_spc, src_off), %t3		! \
198	_VSHD(0, m, 1)	/* vshd %t4, %t1, %r1 */		! \
199	ldws,cmplt	m*4(src_spc, src_off), %t4		! \
200	_VSHD(1, m, 22)	/* vshd %t1, %t2, %t1 */		! \
201	_VSHD(2, m, 21)	/* vshd %t2, %t3, %t2 */		! \
202	_VSHD(3, m, 20)	/* vshd %t3, %t4, %t3 */		! \
203	stws,cmplt	%r1, m*4(dst_spc, dst_off)		! \
204	stws,cmplt	%t1, m*4(dst_spc, dst_off)		! \
205	stws,cmplt	%t2, m*4(dst_spc, dst_off)		! \
206	comib,<<	15, count, _LABEL(S_loop16)		! \
207	stws,cmplt	%t3, m*4(dst_spc, dst_off)		! \
208.label _LABEL(S_skip16)						! \
209								! \
210	/* Loop storing 4 bytes at a time. */			! \
211	addib,<,n	-4, count, _LABEL(S_skip4)		! \
212.label _LABEL(S_loop4)						! \
213	ldws,cmplt	m*4(src_spc, src_off), %t1		! \
214	_VSHD(0, m, 1)	/* into %r1 (1) */			! \
215	copy		%t1, %t4				! \
216	addib,>=	-4, count, _LABEL(S_loop4)		! \
217	stws,cmplt	%r1, m*4(dst_spc, dst_off)		! \
218.label _LABEL(S_skip4)						! \
219								! \
220	/*							! \
221 	 * We now need to "back up" src_off by the		! \
222	 * number of bytes remaining in the FIFO		! \
223	 * (i.e., the number of bytes remaining in %t4),	! \
224	 * because (the correct) count still includes		! \
225	 * these bytes, and we intent to keep it that		! \
226	 * way, and finish with the single-byte copier.		! \
227	 *							! \
228	 * The number of bytes remaining in the FIFO is		! \
229	 * related to the shift count, so recover it,		! \
230	 * restoring the correct count at the same time.	! \
231	 */							! \
232	mfctl	%cr11, %t1					! \
233	addi	4, count, count					! \
234	shd	%r0, %t1, 3, %t1				! \
235								! \
236	/*							! \
237	 * If we're copying forward, the shift count		! \
238	 * is the number of bytes remaining in the		! \
239	 * FIFO, and we want to subtract it from src_off.	! \
240	 * If we're copying backwards, (4 - shift count)	! \
241	 * is the number of bytes remaining in the FIFO,	! \
242	 * and we want to add it to src_off.			! \
243	 *							! \
244	 * We observe that x + (4 - y) = x - (y - 4),		! \
245	 * and introduce this instruction to add -4 when	! \
246	 * m is -1, although this does mean one extra		! \
247	 * instruction in the forward case.			! \
248	 */							! \
249	addi	4*((m - 1) / 2), %t1, %t1			! \
250								! \
251	/* Now branch to the byte-at-a-time loop. */		! \
252	b	_LABEL(_do1)					! \
253	sub	src_off, %t1, src_off
254
255/*
256 * This macro copies a region in the forward direction.
257 */
258#define _COPY_FORWARD(src_spc, src_off, dst_spc, dst_off, count)  \
259								! \
260	/*							! \
261	 * Since in the shifting-left case we will		! \
262	 * load 8 bytes before checking count, to		! \
263	 * keep things simple, branch to the byte 		! \
264	 * copier unless we're copying at least 8.		! \
265	 */							! \
266	comib,>>,n	8, count, _LABEL(_do1)			! \
267								! \
268	/*							! \
269	 * Once we 4-byte align the source offset, 		! \
270	 * figure out how many bytes from the region		! \
271	 * will be in the first 4-byte word we read.		! \
272	 * Ditto for writing the destination offset.		! \
273	 */							! \
274	extru		src_off, 31, 2, %t1			! \
275	extru		dst_off, 31, 2, %t2			! \
276	subi		4, %t1, %t1				! \
277	subi		4, %t2, %t2				! \
278								! \
279	/*							! \
280	 * Calculate the byte shift required.  A 		! \
281	 * positive value means a source 4-byte word 		! \
282	 * has to be shifted to the right to line up 		! \
283	 * as a destination 4-byte word.			! \
284	 */							! \
285	sub		%t1, %t2, %t1				! \
286								! \
287	/* 4-byte align src_off. */				! \
288	depi		0, 31, 2, src_off			! \
289								! \
290	/*							! \
291	 * It's somewhat important to note that this		! \
292	 * code thinks of count as "the number of bytes		! \
293	 * that haven't been stored yet", as opposed to		! \
294	 * "the number of bytes that haven't been copied	! \
295	 * yet".  The distinction is subtle, but becomes	! \
296	 * apparent at the end of the shifting code, where	! \
297	 * we "back up" src_off to correspond to count,		! \
298	 * as opposed to flushing the FIFO.			! \
299	 *							! \
300	 * We calculated above how many bytes our first		! \
301	 * store will store, so update count now.		! \
302	 *							! \
303	 * If the shift is zero, strictly as an optimization	! \
304	 * we use a copy loop that does no shifting.		! \
305	 */							! \
306	comb,<>		%r0, %t1, _LABEL(_shifting)		! \
307	sub		count, %t2, count			! \
308								! \
309	/* Load and store the first word. */			! \
310	ldws,ma		4(src_spc, src_off), %t4		! \
311	stbys,b,m	%t4, 4(dst_spc, dst_off)		! \
312								! \
313	/* Do the rest of the copy. */				! \
314	_COPY(src_spc,src_off,dst_spc,dst_off,count,ma,1)	! \
315								! \
316.label _LABEL(_shifting)					! \
317								! \
318	/*							! \
319	 * If shift < 0, we need to shift words to the		! \
320	 * left.  Since we can't do this directly, we		! \
321	 * adjust the shift so it's a shift to the right	! \
322	 * and load the first word into the high word of	! \
323	 * the FIFO.  Otherwise, we load a zero into the	! \
324	 * high word of the FIFO.				! \
325	 */							! \
326	comb,<=		%r0, %t1, _LABEL(_shiftingrt)		! \
327	copy		%r0, %t3				! \
328	addi		4, %t1, %t1				! \
329	ldws,ma		4(src_spc, src_off), %t3		! \
330.label _LABEL(_shiftingrt)					! \
331								! \
332	/*							! \
333	 * Turn the shift byte count into a bit count,		! \
334	 * load the next word, set the Shift Amount 		! \
335	 * Register, and form and store the first word.		! \
336	 */							! \
337	sh3add		%t1, %r0, %t1				! \
338	ldws,ma		4(src_spc, src_off), %t4		! \
339	mtctl		%t1, %cr11				! \
340	vshd		%t3, %t4, %r1				! \
341	stbys,b,m	%r1, 4(dst_spc, dst_off)		! \
342								! \
343	/* Do the rest of the copy. */				! \
344	_COPYS(src_spc,src_off,dst_spc,dst_off,count,ma,1)
345
346/* This macro copies a region in the reverse direction. */
347#define _COPY_REVERSE(src_spc, src_off, dst_spc, dst_off, count)  \
348								! \
349	/* Immediately add count to both offsets. */		! \
350	add	src_off, count, src_off				! \
351	add	dst_off, count, dst_off				! \
352								! \
353	/*							! \
354	 * Since in the shifting-right case we 			! \
355	 * will load 8 bytes before checking 			! \
356	 * count, to keep things simple, branch 		! \
357	 * to the byte copier unless we're 			! \
358	 * copying at least 8 bytes.				! \
359	 */							! \
360	comib,>>,n	8, count, _LABEL(_do1)			! \
361								! \
362	/*							! \
363	 * Once we 4-byte align the source offset, 		! \
364	 * figure out how many bytes from the region		! \
365	 * will be in the first 4-byte word we read.		! \
366	 * Ditto for writing the destination offset.		! \
367	 */							! \
368	extru,<>	src_off, 31, 2, %t1			! \
369	ldi		4, %t1					! \
370	extru,<>	dst_off, 31, 2, %t2			! \
371	ldi		4, %t2					! \
372								! \
373	/*							! \
374	 * Calculate the byte shift required.  A 		! \
375	 * positive value means a source 4-byte 		! \
376	 * word has to be shifted to the right to 		! \
377	 * line up as a destination 4-byte word.		! \
378	 */							! \
379	sub		%t2, %t1, %t1				! \
380								! \
381	/*							! \
382	 * 4-byte align src_off, leaving it pointing 		! \
383	 * to the 4-byte word *after* the next word 		! \
384	 * we intend to load.					! \
385	 *							! \
386	 * It's somewhat important to note that this		! \
387	 * code thinks of count as "the number of bytes		! \
388	 * that haven't been stored yet", as opposed to		! \
389	 * "the number of bytes that haven't been copied	! \
390	 * yet".  The distinction is subtle, but becomes	! \
391	 * apparent at the end of the shifting code, where	! \
392	 * we "back up" src_off to correspond to count,		! \
393	 * as opposed to flushing the FIFO.			! \
394	 *							! \
395	 * We calculated above how many bytes our first		! \
396	 * store will store, so update count now.		! \
397	 *							! \
398	 * If the shift is zero, we use a copy loop that	! \
399	 * does no shifting.  NB: unlike the forward case,	! \
400	 * this is NOT strictly an optimization.  If the	! \
401	 * SAR is zero the vshds do NOT do the right thing.	! \
402	 * This is another assymetry more or less the "fault"	! \
403	 * of vshd.						! \
404	 */							! \
405	addi		3, src_off, src_off			! \
406	sub		count, %t2, count			! \
407	comb,<>		%r0, %t1, _LABEL(_shifting)		! \
408	depi		0, 31, 2, src_off			! \
409								! \
410	/* Load and store the first word. */			! \
411	ldws,mb		-4(src_spc, src_off), %t4		! \
412	_STBYS_E_M(%t4, dst_spc, dst_off)			! \
413								! \
414	/* Do the rest of the copy. */				! \
415	_COPY(src_spc,src_off,dst_spc,dst_off,count,mb,-1)	! \
416								! \
417.label _LABEL(_shifting)					! \
418								! \
419	/*							! \
420	 * If shift < 0, we need to shift words to the		! \
421	 * left.  Since we can't do this directly, we		! \
422	 * adjust the shift so it's a shift to the right	! \
423	 * and load a zero in to the low word of the FIFO.	! \
424	 * Otherwise, we load the first word into the		! \
425	 * low word of the FIFO.				! \
426	 *							! \
427	 * Note the nullification trickery here.  We 		! \
428	 * assume that we're shifting to the left, and		! \
429	 * load zero into the low word of the FIFO.  Then	! \
430	 * we nullify the addi if we're shifting to the		! \
431	 * right.  If the addi is not nullified, we are		! \
432 	 * shifting to the left, so we nullify the load.	! \
433	 * we branch if we're shifting to the 			! \
434	 */							! \
435	copy		%r0, %t3				! \
436	comb,<=,n	%r0, %t1, 0				! \
437	addi,tr		4, %t1, %t1				! \
438	ldws,mb		-4(src_spc, src_off), %t3		! \
439								! \
440	/*							! \
441	 * Turn the shift byte count into a bit count,		! \
442	 * load the next word, set the Shift Amount 		! \
443	 * Register, and form and store the first word.		! \
444	 */							! \
445	sh3add		%t1, %r0, %t1				! \
446	ldws,mb		-4(src_spc, src_off), %t4		! \
447	mtctl		%t1, %cr11				! \
448	vshd		%t4, %t3, %r1				! \
449	_STBYS_E_M(%r1, dst_spc, dst_off)			! \
450								! \
451	/* Do the rest of the copy. */				! \
452	_COPYS(src_spc,src_off,dst_spc,dst_off,count,mb,-1)
453
454/*
455 * For paranoia, when things aren't going well, enable this
456 * code to assemble byte-at-a-time-only copying.
457 */
458#if 1
459#undef _COPY_FORWARD
460#define _COPY_FORWARD(src_spc, src_off, dst_spc, dst_off, count)  \
461	comb,=,n	%r0, count, _LABEL(_done)		! \
462	ldbs,ma		1(src_spc, src_off), %r1		! \
463	addib,<>	-1, count, -12				! \
464	stbs,ma		%r1, 1(dst_spc, dst_off)		! \
465	b,n		_LABEL(_done)
466#undef _COPY_REVERSE
467#define _COPY_REVERSE(src_spc, src_off, dst_spc, dst_off, count)  \
468	comb,=		%r0, count, _LABEL(_done)		! \
469	add		src_off, count, src_off			! \
470	add		dst_off, count, dst_off			! \
471	ldbs,mb		-1(src_spc, src_off), %r1		! \
472	addib,<>	-1, count, -12				! \
473	stbs,mb		%r1, -1(dst_spc, dst_off)		! \
474	b,n		_LABEL(_done)
475#endif
476
477/*
478 * If none of the following are defined, define BCOPY.
479 */
480#if !(defined(SPCOPY) || defined(MEMCPY) || defined(MEMMOVE))
481#define BCOPY
482#endif
483
484#if defined(SPCOPY) && !defined(_STANDALONE)
485#include <sys/errno.h>
486#include "assym.h"
487
488/*
489 * int spcopy(pa_space_t ssp, const void *src, pa_space_t dsp, void *dst,
490 * 	size_t len)
491 *
492 * We assume that the regions do not overlap.
493 */
494LEAF_ENTRY(spcopy)
495
496        /*
497	 * Setup the fault handler, and load %ret0
498	 * with EFAULT, assuming the copy will fail.
499	 */
500	.import	curlwp, data
501	ldil    L%curlwp, %r31
502	ldw     R%curlwp(%r31), %r31
503#ifdef	DIAGNOSTIC
504	comb,<>,n %r0, %r31, Lspcopy_curlwp_ok
505	ldil	L%panic, %r1
506	ldil	L%Lspcopy_curlwp_bad, %arg0
507	ldo	R%panic(%r1), %r1
508	ldo	R%Lspcopy_curlwp_bad(%arg0), %arg0
509	.call
510	bv,n    %r0(%r1)
511	nop
512Lspcopy_curlwp_bad:
513	.asciz	"spcopy: curlwp == NULL\n"
514	.align	8
515Lspcopy_curlwp_ok:
516#endif /* DIAGNOSTIC */
517	ldil    L%spcopy_fault, %r1
518	ldw     L_ADDR(%r31), %r31
519	ldo     R%spcopy_fault(%r1), %r1
520	ldi	EFAULT, %ret0
521	stw     %r1, U_PCB+PCB_ONFAULT(%r31)
522
523	/* Setup the space registers. */
524	mfsp	%sr2, %ret1
525	mtsp	%arg0, %sr1
526	mtsp	%arg2, %sr2
527
528	/* Get the len argument and do the copy. */
529	ldw	HPPA_FRAME_ARG(4)(%sp), %arg0
530#define	_LABEL(l) __CONCAT(spcopy,l)
531	_COPY_FORWARD(%sr1,%arg1,%sr2,%arg3,%arg0)
532_LABEL(_done):
533
534	/* Return. */
535	copy	%r0, %ret0
536ALTENTRY(spcopy_fault)
537	stw     %r0, U_PCB+PCB_ONFAULT(%r31)
538	bv	%r0(%rp)
539	mtsp	%ret1, %sr2
540EXIT(spcopy)
541#endif /* SPCOPY && !_STANDALONE */
542
543#ifdef MEMCPY
544/*
545 * void *memcpy(void *restrict dst, const void *restrict src, size_t len);
546 *
547 * memcpy is specifically restricted to working on
548 * non-overlapping regions, so we can just copy forward.
549 */
550LEAF_ENTRY(memcpy)
551	copy	%arg0, %ret0
552#define	_LABEL(l) __CONCAT(memcpy,l)
553	_COPY_FORWARD(%sr0,%arg1,%sr0,%arg0,%arg2)
554_LABEL(_done):
555	bv,n	%r0(%rp)
556	nop
557EXIT(memcpy)
558#endif /* MEMCPY */
559
560#ifdef BCOPY
561/*
562 * void bcopy(const void *src, void *dst, size_t len);
563 */
564LEAF_ENTRY(bcopy)
565	copy	%arg0, %r1
566	copy	%arg1, %arg0
567	copy	%r1, %arg1
568	/* FALLTHROUGH */
569#define _LABEL_F(l) __CONCAT(bcopy_F,l)
570#define _LABEL_R(l) __CONCAT(bcopy_R,l)
571#endif
572
573#ifdef MEMMOVE
574/*
575 * void *memmove(void *dst, const void *src, size_t len);
576 */
577LEAF_ENTRY(memmove)
578#define _LABEL_F(l) __CONCAT(memmove_F,l)
579#define _LABEL_R(l) __CONCAT(memmove_R,l)
580	copy	%arg0, %ret0
581#endif /* MEMMOVE */
582
583#if defined(BCOPY) || defined(MEMMOVE)
584
585	/*
586	 * If src >= dst or src + len <= dst, we copy
587	 * forward, else we copy in reverse.
588	 */
589	add		%arg1, %arg2, %r1
590	comb,>>=,n	%arg1, %arg0, 0
591	comb,>>,n	%r1, %arg0, _LABEL_R(_go)
592
593#define _LABEL _LABEL_F
594	_COPY_FORWARD(%sr0,%arg1,%sr0,%arg0,%arg2)
595#undef _LABEL
596
597_LABEL_R(_go):
598#define _LABEL _LABEL_R
599	_COPY_REVERSE(%sr0,%arg1,%sr0,%arg0,%arg2)
600#undef _LABEL
601
602_LABEL_F(_done):
603_LABEL_R(_done):
604	bv,n	%r0(%rp)
605	nop
606#ifdef BCOPY
607EXIT(bcopy)
608#else
609EXIT(memmove)
610#endif
611#endif /* BCOPY || MEMMOVE */
612