1/*
2 * arch/alpha/lib/ev6-stxncpy.S
3 * 21264 version contributed by Rick Gorton <rick.gorton@api-networks.com>
4 *
5 * Copy no more than COUNT bytes of the null-terminated string from
6 * SRC to DST.
7 *
8 * This is an internal routine used by strncpy, stpncpy, and strncat.
9 * As such, it uses special linkage conventions to make implementation
10 * of these public functions more efficient.
11 *
12 * On input:
13 *	t9 = return address
14 *	a0 = DST
15 *	a1 = SRC
16 *	a2 = COUNT
17 *
18 * Furthermore, COUNT may not be zero.
19 *
20 * On output:
21 *	t0  = last word written
22 *	t10 = bitmask (with one bit set) indicating the byte position of
23 *	      the end of the range specified by COUNT
24 *	t12 = bitmask (with one bit set) indicating the last byte written
25 *	a0  = unaligned address of the last *word* written
26 *	a2  = the number of full words left in COUNT
27 *
28 * Furthermore, v0, a3-a5, t11, and $at are untouched.
29 *
30 * Much of the information about 21264 scheduling/coding comes from:
31 *	Compiler Writer's Guide for the Alpha 21264
32 *	abbreviated as 'CWG' in other comments here
33 *	ftp.digital.com/pub/Digital/info/semiconductor/literature/dsc-library.html
34 * Scheduling notation:
35 *	E	- either cluster
36 *	U	- upper subcluster; U0 - subcluster U0; U1 - subcluster U1
37 *	L	- lower subcluster; L0 - subcluster L0; L1 - subcluster L1
38 * Try not to change the actual algorithm if possible for consistency.
39 */
40
41#include <asm/regdef.h>
42
43	.set noat
44	.set noreorder
45
46	.text
47
48
49
50	.ent stxncpy_aligned
51	.align 4
52stxncpy_aligned:
53	.frame sp, 0, t9, 0
54	.prologue 0
55
56	/* On entry to this basic block:
57	   t0 == the first destination word for masking back in
58	   t1 == the first source word.  */
59
60	/* Create the 1st output word and detect 0's in the 1st input word.  */
61	lda	t2, -1		# E : build a mask against false zero
62	mskqh	t2, a1, t2	# U :   detection in the src word (stall)
63	mskqh	t1, a1, t3	# U :
64	ornot	t1, t2, t2	# E : (stall)
65
66	mskql	t0, a1, t0	# U : assemble the first output word
67	cmpbge	zero, t2, t8	# E : bits set iff null found
68	or	t0, t3, t0	# E : (stall)
69	beq	a2, $a_eoc	# U :
70
71	bne	t8, $a_eos	# U :
72	nop
73	nop
74	nop
75
76	/* On entry to this basic block:
77	   t0 == a source word not containing a null.  */
78
79	/*
80	 * nops here to:
81	 *	separate store quads from load quads
82	 *	limit of 1 bcond/quad to permit training
83	 */
84$a_loop:
85	stq_u	t0, 0(a0)	# L :
86	addq	a0, 8, a0	# E :
87	subq	a2, 1, a2	# E :
88	nop
89
90	ldq_u	t0, 0(a1)	# L :
91	addq	a1, 8, a1	# E :
92	cmpbge	zero, t0, t8	# E :
93	beq	a2, $a_eoc      # U :
94
95	beq	t8, $a_loop	# U :
96	nop
97	nop
98	nop
99
100	/* Take care of the final (partial) word store.  At this point
101	   the end-of-count bit is set in t8 iff it applies.
102
103	   On entry to this basic block we have:
104	   t0 == the source word containing the null
105	   t8 == the cmpbge mask that found it.  */
106
107$a_eos:
108	negq	t8, t12		# E : find low bit set
109	and	t8, t12, t12	# E : (stall)
110	/* For the sake of the cache, don't read a destination word
111	   if we're not going to need it.  */
112	and	t12, 0x80, t6	# E : (stall)
113	bne	t6, 1f		# U : (stall)
114
115	/* We're doing a partial word store and so need to combine
116	   our source and original destination words.  */
117	ldq_u	t1, 0(a0)	# L :
118	subq	t12, 1, t6	# E :
119	or	t12, t6, t8	# E : (stall)
120	zapnot	t0, t8, t0	# U : clear src bytes > null (stall)
121
122	zap	t1, t8, t1	# .. e1 : clear dst bytes <= null
123	or	t0, t1, t0	# e1    : (stall)
124	nop
125	nop
126
1271:	stq_u	t0, 0(a0)	# L :
128	ret	(t9)		# L0 : Latency=3
129	nop
130	nop
131
132	/* Add the end-of-count bit to the eos detection bitmask.  */
133$a_eoc:
134	or	t10, t8, t8	# E :
135	br	$a_eos		# L0 : Latency=3
136	nop
137	nop
138
139	.end stxncpy_aligned
140
141	.align 4
142	.ent __stxncpy
143	.globl __stxncpy
144__stxncpy:
145	.frame sp, 0, t9, 0
146	.prologue 0
147
148	/* Are source and destination co-aligned?  */
149	xor	a0, a1, t1	# E :
150	and	a0, 7, t0	# E : find dest misalignment
151	and	t1, 7, t1	# E : (stall)
152	addq	a2, t0, a2	# E : bias count by dest misalignment (stall)
153
154	subq	a2, 1, a2	# E :
155	and	a2, 7, t2	# E : (stall)
156	srl	a2, 3, a2	# U : a2 = loop counter = (count - 1)/8 (stall)
157	addq	zero, 1, t10	# E :
158
159	sll	t10, t2, t10	# U : t10 = bitmask of last count byte
160	bne	t1, $unaligned	# U :
161	/* We are co-aligned; take care of a partial first word.  */
162	ldq_u	t1, 0(a1)	# L : load first src word
163	addq	a1, 8, a1	# E :
164
165	beq	t0, stxncpy_aligned     # U : avoid loading dest word if not needed
166	ldq_u	t0, 0(a0)	# L :
167	nop
168	nop
169
170	br	stxncpy_aligned	# .. e1 :
171	nop
172	nop
173	nop
174
175
176
177/* The source and destination are not co-aligned.  Align the destination
178   and cope.  We have to be very careful about not reading too much and
179   causing a SEGV.  */
180
181	.align 4
182$u_head:
183	/* We know just enough now to be able to assemble the first
184	   full source word.  We can still find a zero at the end of it
185	   that prevents us from outputting the whole thing.
186
187	   On entry to this basic block:
188	   t0 == the first dest word, unmasked
189	   t1 == the shifted low bits of the first source word
190	   t6 == bytemask that is -1 in dest word bytes */
191
192	ldq_u	t2, 8(a1)	# L : Latency=3 load second src word
193	addq	a1, 8, a1	# E :
194	mskql	t0, a0, t0	# U : mask trailing garbage in dst
195	extqh	t2, a1, t4	# U : (3 cycle stall on t2)
196
197	or	t1, t4, t1	# E : first aligned src word complete (stall)
198	mskqh	t1, a0, t1	# U : mask leading garbage in src (stall)
199	or	t0, t1, t0	# E : first output word complete (stall)
200	or	t0, t6, t6	# E : mask original data for zero test (stall)
201
202	cmpbge	zero, t6, t8	# E :
203	beq	a2, $u_eocfin	# U :
204	lda	t6, -1		# E :
205	nop
206
207	bne	t8, $u_final	# U :
208	mskql	t6, a1, t6	# U : mask out bits already seen
209	stq_u	t0, 0(a0)	# L : store first output word
210	or      t6, t2, t2	# E : (stall)
211
212	cmpbge	zero, t2, t8	# E : find nulls in second partial
213	addq	a0, 8, a0	# E :
214	subq	a2, 1, a2	# E :
215	bne	t8, $u_late_head_exit	# U :
216
217	/* Finally, we've got all the stupid leading edge cases taken care
218	   of and we can set up to enter the main loop.  */
219	extql	t2, a1, t1	# U : position hi-bits of lo word
220	beq	a2, $u_eoc	# U :
221	ldq_u	t2, 8(a1)	# L : read next high-order source word
222	addq	a1, 8, a1	# E :
223
224	extqh	t2, a1, t0	# U : position lo-bits of hi word (stall)
225	cmpbge	zero, t2, t8	# E :
226	nop
227	bne	t8, $u_eos	# U :
228
229	/* Unaligned copy main loop.  In order to avoid reading too much,
230	   the loop is structured to detect zeros in aligned source words.
231	   This has, unfortunately, effectively pulled half of a loop
232	   iteration out into the head and half into the tail, but it does
233	   prevent nastiness from accumulating in the very thing we want
234	   to run as fast as possible.
235
236	   On entry to this basic block:
237	   t0 == the shifted low-order bits from the current source word
238	   t1 == the shifted high-order bits from the previous source word
239	   t2 == the unshifted current source word
240
241	   We further know that t2 does not contain a null terminator.  */
242
243	.align 4
244$u_loop:
245	or	t0, t1, t0	# E : current dst word now complete
246	subq	a2, 1, a2	# E : decrement word count
247	extql	t2, a1, t1	# U : extract low bits for next time
248	addq	a0, 8, a0	# E :
249
250	stq_u	t0, -8(a0)	# U : save the current word
251	beq	a2, $u_eoc	# U :
252	ldq_u	t2, 8(a1)	# U : Latency=3 load high word for next time
253	addq	a1, 8, a1	# E :
254
255	extqh	t2, a1, t0	# U : extract low bits (2 cycle stall)
256	cmpbge	zero, t2, t8	# E : test new word for eos
257	nop
258	beq	t8, $u_loop	# U :
259
260	/* We've found a zero somewhere in the source word we just read.
261	   If it resides in the lower half, we have one (probably partial)
262	   word to write out, and if it resides in the upper half, we
263	   have one full and one partial word left to write out.
264
265	   On entry to this basic block:
266	   t0 == the shifted low-order bits from the current source word
267	   t1 == the shifted high-order bits from the previous source word
268	   t2 == the unshifted current source word.  */
269$u_eos:
270	or	t0, t1, t0	# E : first (partial) source word complete
271	nop
272	cmpbge	zero, t0, t8	# E : is the null in this first bit? (stall)
273	bne	t8, $u_final	# U : (stall)
274
275	stq_u	t0, 0(a0)	# L : the null was in the high-order bits
276	addq	a0, 8, a0	# E :
277	subq	a2, 1, a2	# E :
278	nop
279
280$u_late_head_exit:
281	extql	t2, a1, t0	# U :
282	cmpbge	zero, t0, t8	# E :
283	or	t8, t10, t6	# E : (stall)
284	cmoveq	a2, t6, t8	# E : Latency=2, extra map slot (stall)
285
286	/* Take care of a final (probably partial) result word.
287	   On entry to this basic block:
288	   t0 == assembled source word
289	   t8 == cmpbge mask that found the null.  */
290$u_final:
291	negq	t8, t6		# E : isolate low bit set
292	and	t6, t8, t12	# E : (stall)
293	and	t12, 0x80, t6	# E : avoid dest word load if we can (stall)
294	bne	t6, 1f		# U : (stall)
295
296	ldq_u	t1, 0(a0)	# L :
297	subq	t12, 1, t6	# E :
298	or	t6, t12, t8	# E : (stall)
299	zapnot	t0, t8, t0	# U : kill source bytes > null
300
301	zap	t1, t8, t1	# U : kill dest bytes <= null
302	or	t0, t1, t0	# E : (stall)
303	nop
304	nop
305
3061:	stq_u	t0, 0(a0)	# L :
307	ret	(t9)		# L0 : Latency=3
308
309	  /* Got to end-of-count before end of string.
310	     On entry to this basic block:
311	     t1 == the shifted high-order bits from the previous source word  */
312$u_eoc:
313	and	a1, 7, t6	# E : avoid final load if possible
314	sll	t10, t6, t6	# U : (stall)
315	and	t6, 0xff, t6	# E : (stall)
316	bne	t6, 1f		# U : (stall)
317
318	ldq_u	t2, 8(a1)	# L : load final src word
319	nop
320	extqh	t2, a1, t0	# U : extract low bits for last word (stall)
321	or	t1, t0, t1	# E : (stall)
322
3231:	cmpbge	zero, t1, t8	# E :
324	mov	t1, t0		# E :
325
326$u_eocfin:			# end-of-count, final word
327	or	t10, t8, t8	# E :
328	br	$u_final	# L0 : Latency=3
329
330	/* Unaligned copy entry point.  */
331	.align 4
332$unaligned:
333
334	ldq_u	t1, 0(a1)	# L : load first source word
335	and	a0, 7, t4	# E : find dest misalignment
336	and	a1, 7, t5	# E : find src misalignment
337	/* Conditionally load the first destination word and a bytemask
338	   with 0xff indicating that the destination byte is sacrosanct.  */
339	mov	zero, t0	# E :
340
341	mov	zero, t6	# E :
342	beq	t4, 1f		# U :
343	ldq_u	t0, 0(a0)	# L :
344	lda	t6, -1		# E :
345
346	mskql	t6, a0, t6	# U :
347	nop
348	nop
349	subq	a1, t4, a1	# E : sub dest misalignment from src addr
350
351	/* If source misalignment is larger than dest misalignment, we need
352	   extra startup checks to avoid SEGV.  */
353
3541:	cmplt	t4, t5, t12	# E :
355	extql	t1, a1, t1	# U : shift src into place
356	lda	t2, -1		# E : for creating masks later
357	beq	t12, $u_head	# U : (stall)
358
359	extql	t2, a1, t2	# U :
360	cmpbge	zero, t1, t8	# E : is there a zero?
361	andnot	t2, t6, t12	# E : dest mask for a single word copy
362	or	t8, t10, t5	# E : test for end-of-count too
363
364	cmpbge	zero, t12, t3	# E :
365	cmoveq	a2, t5, t8	# E : Latency=2, extra map slot
366	nop			# E : keep with cmoveq
367	andnot	t8, t3, t8	# E : (stall)
368
369	beq	t8, $u_head	# U :
370	/* At this point we've found a zero in the first partial word of
371	   the source.  We need to isolate the valid source data and mask
372	   it into the original destination data.  (Incidentally, we know
373	   that we'll need at least one byte of that original dest word.) */
374	ldq_u	t0, 0(a0)	# L :
375	negq	t8, t6		# E : build bitmask of bytes <= zero
376	mskqh	t1, t4, t1	# U :
377
378	and	t6, t8, t2	# E :
379	subq	t2, 1, t6	# E : (stall)
380	or	t6, t2, t8	# E : (stall)
381	zapnot	t12, t8, t12	# U : prepare source word; mirror changes (stall)
382
383	zapnot	t1, t8, t1	# U : to source validity mask
384	andnot	t0, t12, t0	# E : zero place for source to reside
385	or	t0, t1, t0	# E : and put it there (stall both t0, t1)
386	stq_u	t0, 0(a0)	# L : (stall)
387
388	ret	(t9)		# L0 : Latency=3
389	nop
390	nop
391	nop
392
393	.end __stxncpy
394