1 /* Copyright (C) 1996, 1997 Free Software Foundation, Inc.
2 Contributed by Richard Henderson (rth@tamu.edu)
3 This file is part of the GNU C Library.
5 The GNU C Library is free software; you can redistribute it and/or
6 modify it under the terms of the GNU Lesser General Public
7 License as published by the Free Software Foundation; either
8 version 2.1 of the License, or (at your option) any later version.
10 The GNU C Library is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 Lesser General Public License for more details.
15 You should have received a copy of the GNU Lesser General Public
16 License along with the GNU C Library; if not, write to the Free
17 Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
20 /* Copy a null-terminated string from SRC to DST.
22 This is an internal routine used by strcpy, stpcpy, and strcat.
23 As such, it uses special linkage conventions to make implementation
24 of these public functions more efficient.
32 t8 = bitmask (with one bit set) indicating the last byte written
33 a0 = unaligned address of the last *word* written
35 Furthermore, v0, a3-a5, t11, and t12 are untouched.
38 /* This is generally scheduled for the EV5, but should still be pretty
39 good for the EV4 too. */
48 /* There is a problem with either gdb (as of 4.16) or gas (as of 2.7) that
49 doesn't like putting the entry point for a procedure somewhere in the
50 middle of the procedure descriptor. Work around this by putting the
51 aligned copy in its own procedure descriptor */
59 /* On entry to this basic block:
60 t0 == the first destination word for masking back in
61 t1 == the first source word. */
63 /* Create the 1st output word and detect 0's in the 1st input word. */
64 lda t2, -1 # e1 : build a mask against false zero
65 mskqh t2, a1, t2 # e0 : detection in the src word
66 mskqh t1, a1, t3 # e0 :
67 ornot t1, t2, t2 # .. e1 :
68 mskql t0, a1, t0 # e0 : assemble the first output word
69 cmpbge zero, t2, t7 # .. e1 : bits set iff null found
71 bne t7, $a_eos # .. e1 :
73 /* On entry to this basic block:
74 t0 == the first destination word for masking back in
75 t1 == a source word not containing a null. */
78 stq_u t1, 0(a0) # e0 :
79 addq a0, 8, a0 # .. e1 :
80 ldq_u t1, 0(a1) # e0 :
81 addq a1, 8, a1 # .. e1 :
82 cmpbge zero, t1, t7 # e0 (stall)
83 beq t7, $a_loop # .. e1 (zdb)
85 /* Take care of the final (partial) word store.
86 On entry to this basic block we have:
87 t1 == the source word containing the null
88 t7 == the cmpbge mask that found it. */
90 negq t7, t6 # e0 : find low bit set
91 and t7, t6, t8 # e1 (stall)
93 /* For the sake of the cache, don't read a destination word
94 if we're not going to need it. */
95 and t8, 0x80, t6 # e0 :
96 bne t6, 1f # .. e1 (zdb)
98 /* We're doing a partial word store and so need to combine
99 our source and original destination words. */
100 ldq_u t0, 0(a0) # e0 :
101 subq t8, 1, t6 # .. e1 :
102 zapnot t1, t6, t1 # e0 : clear src bytes >= null
103 or t8, t6, t7 # .. e1 :
104 zap t0, t7, t0 # e0 : clear dst bytes <= null
107 1: stq_u t1, 0(a0) # e0 :
119 /* Are source and destination co-aligned? */
120 xor a0, a1, t0 # e0 :
123 bne t0, $unaligned # .. e1 :
125 /* We are co-aligned; take care of a partial first word. */
126 ldq_u t1, 0(a1) # e0 : load first src word
127 and a0, 7, t0 # .. e1 : take care not to load a word ...
128 addq a1, 8, a1 # e0 :
129 beq t0, stxcpy_aligned # .. e1 : ... if we wont need it
130 ldq_u t0, 0(a0) # e0 :
131 br stxcpy_aligned # .. e1 :
134 /* The source and destination are not co-aligned. Align the destination
135 and cope. We have to be very careful about not reading too much and
140 /* We know just enough now to be able to assemble the first
141 full source word. We can still find a zero at the end of it
142 that prevents us from outputting the whole thing.
144 On entry to this basic block:
145 t0 == the first dest word, for masking back in, if needed else 0
146 t1 == the low bits of the first source word
147 t6 == bytemask that is -1 in dest word bytes */
149 ldq_u t2, 8(a1) # e0 :
150 addq a1, 8, a1 # .. e1 :
152 extql t1, a1, t1 # e0 :
153 extqh t2, a1, t4 # e0 :
154 mskql t0, a0, t0 # e0 :
155 or t1, t4, t1 # .. e1 :
156 mskqh t1, a0, t1 # e0 :
160 cmpbge zero, t6, t7 # .. e1 :
161 lda t6, -1 # e0 : for masking just below
162 bne t7, $u_final # .. e1 :
164 mskql t6, a1, t6 # e0 : mask out the bits we have
165 or t6, t2, t2 # e1 : already extracted before
166 cmpbge zero, t2, t7 # e0 : testing eos
167 bne t7, $u_late_head_exit # .. e1 (zdb)
169 /* Finally, we've got all the stupid leading edge cases taken care
170 of and we can set up to enter the main loop. */
172 stq_u t1, 0(a0) # e0 : store first output word
173 addq a0, 8, a0 # .. e1 :
174 extql t2, a1, t0 # e0 : position ho-bits of lo word
175 ldq_u t2, 8(a1) # .. e1 : read next high-order source word
176 addq a1, 8, a1 # e0 :
177 cmpbge zero, t2, t7 # .. e1 :
179 bne t7, $u_eos # .. e1 :
181 /* Unaligned copy main loop. In order to avoid reading too much,
182 the loop is structured to detect zeros in aligned source words.
183 This has, unfortunately, effectively pulled half of a loop
184 iteration out into the head and half into the tail, but it does
185 prevent nastiness from accumulating in the very thing we want
186 to run as fast as possible.
188 On entry to this basic block:
189 t0 == the shifted high-order bits from the previous source word
190 t2 == the unshifted current source word
192 We further know that t2 does not contain a null terminator. */
196 extqh t2, a1, t1 # e0 : extract high bits for current word
197 addq a1, 8, a1 # .. e1 :
198 extql t2, a1, t3 # e0 : extract low bits for next time
199 addq a0, 8, a0 # .. e1 :
200 or t0, t1, t1 # e0 : current dst word now complete
201 ldq_u t2, 0(a1) # .. e1 : load high word for next time
202 stq_u t1, -8(a0) # e0 : save the current word
204 cmpbge zero, t2, t7 # e0 : test new word for eos
205 beq t7, $u_loop # .. e1 :
207 /* We've found a zero somewhere in the source word we just read.
208 If it resides in the lower half, we have one (probably partial)
209 word to write out, and if it resides in the upper half, we
210 have one full and one partial word left to write out.
212 On entry to this basic block:
213 t0 == the shifted high-order bits from the previous source word
214 t2 == the unshifted current source word. */
216 extqh t2, a1, t1 # e0 :
217 or t0, t1, t1 # e1 : first (partial) source word complete
219 cmpbge zero, t1, t7 # e0 : is the null in this first bit?
220 bne t7, $u_final # .. e1 (zdb)
223 stq_u t1, 0(a0) # e0 : the null was in the high-order bits
224 addq a0, 8, a0 # .. e1 :
225 extql t2, a1, t1 # e0 :
226 cmpbge zero, t1, t7 # .. e1 :
228 /* Take care of a final (probably partial) result word.
229 On entry to this basic block:
230 t1 == assembled source word
231 t7 == cmpbge mask that found the null. */
233 negq t7, t6 # e0 : isolate low bit set
234 and t6, t7, t8 # e1 :
236 and t8, 0x80, t6 # e0 : avoid dest word load if we can
237 bne t6, 1f # .. e1 (zdb)
239 ldq_u t0, 0(a0) # e0 :
240 subq t8, 1, t6 # .. e1 :
242 zapnot t1, t6, t1 # .. e1 : kill source bytes >= null
243 zap t0, t7, t0 # e0 : kill dest bytes <= null
246 1: stq_u t1, 0(a0) # e0 :
249 /* Unaligned copy entry point. */
253 ldq_u t1, 0(a1) # e0 : load first source word
255 and a0, 7, t4 # .. e1 : find dest misalignment
256 and a1, 7, t5 # e0 : find src misalignment
258 /* Conditionally load the first destination word and a bytemask
259 with 0xff indicating that the destination byte is sacrosanct. */
261 mov zero, t0 # .. e1 :
264 ldq_u t0, 0(a0) # e0 :
266 mskql t6, a0, t6 # e0 :
268 subq a1, t4, a1 # .. e1 : sub dest misalignment from src addr
270 /* If source misalignment is larger than dest misalignment, we need
271 extra startup checks to avoid SEGV. */
273 cmplt t4, t5, t8 # e0 :
274 beq t8, $u_head # .. e1 (zdb)
276 lda t2, -1 # e1 : mask out leading garbage in source
277 mskqh t2, t5, t2 # e0 :
279 ornot t1, t2, t3 # .. e1 :
280 cmpbge zero, t3, t7 # e0 : is there a zero?
281 beq t7, $u_head # .. e1 (zdb)
283 /* At this point we've found a zero in the first partial word of
284 the source. We need to isolate the valid source data and mask
285 it into the original destination data. (Incidentally, we know
286 that we'll need at least one byte of that original dest word.) */
288 ldq_u t0, 0(a0) # e0 :
290 negq t7, t6 # .. e1 : build bitmask of bytes <= zero
291 and t6, t7, t8 # e0 :
292 and a1, 7, t5 # .. e1 :
293 subq t8, 1, t6 # e0 :
295 srl t8, t5, t8 # e0 : adjust final null return value
297 zapnot t2, t7, t2 # .. e1 : prepare source word; mirror changes
298 and t1, t2, t1 # e1 : to source validity mask
299 extql t2, a1, t2 # .. e0 :
300 extql t1, a1, t1 # e0 :
302 andnot t0, t2, t0 # .. e1 : zero place for source to reside
303 or t0, t1, t1 # e1 : and put it there
304 stq_u t1, 0(a0) # .. e0 :