1 /* Copyright (C) 1996 Free Software Foundation, Inc.
2 Contributed by Richard Henderson (rth@tamu.edu)
4 This file is part of the GNU C Library.
6 The GNU C Library is free software; you can redistribute it and/or
7 modify it under the terms of the GNU Library General Public License as
8 published by the Free Software Foundation; either version 2 of the
9 License, or (at your option) any later version.
11 The GNU C Library is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Library General Public License for more details.
16 You should have received a copy of the GNU Library General Public
17 License along with the GNU C Library; see the file COPYING.LIB. If
18 not, write to the Free Software Foundation, Inc., 675 Mass Ave,
19 Cambridge, MA 02139, USA. */
21 /* Bytewise compare two null-terminated strings. */
40 ldq_u t0, 0(a0) # e0 : give cache time to catch up
41 xor a0, a1, t2 # .. e1 : are s1 and s2 co-aligned?
42 ldq_u t1, 0(a1) # e0 :
43 and t2, 7, t2 # .. e1 :
45 bne t2, $unaligned # .. e1 :
47 /* On entry to this basic block:
48 t0 == the first destination word for masking back in
49 t1 == the first source word.
53 mskqh t3, a0, t3 # e0 :
55 ornot t1, t3, t1 # e0 :
56 ornot t0, t3, t0 # .. e1 :
57 cmpbge zero, t1, t7 # e0 : bits set iff null found
58 bne t7, $eos # e1 (zdb)
60 /* Aligned compare main loop.
61 On entry to this basic block:
63 t1 == an s2 word not containing a null. */
67 bne t2, $wordcmp # .. e1 (zdb)
68 ldq_u t1, 8(a1) # e0 :
69 ldq_u t0, 8(a0) # .. e1 :
71 addq a0, 8, a0 # .. e1 :
72 cmpbge zero, t1, t7 # e0 :
73 beq t7, $a_loop # .. e1 (zdb)
76 /* The two strings are not co-aligned. Align s1 and cope. */
79 and a0, 7, t4 # e0 : find s1 misalignment
80 and a1, 7, t5 # .. e1 : find s2 misalignment
81 subq a1, t4, a1 # e0 :
83 /* If s2 misalignment is larger than s2 misalignment, we need
84 extra startup checks to avoid SEGV. */
86 cmplt t4, t5, t8 # .. e1 :
87 beq t8, $u_head # e1 :
89 mskqh t3, t5, t3 # e0 :
90 ornot t1, t3, t3 # e0 :
91 cmpbge zero, t3, t7 # e1 : is there a zero?
92 beq t7, $u_head # e1 :
94 /* We've found a zero in the first partial word of s2. Align
95 our current s1 and s2 words and compare what we've got. */
97 extql t1, t5, t1 # e0 :
98 extql t0, a0, t0 # e0 :
99 cmpbge zero, t1, t7 # .. e1 : find that zero again
100 br $eos # e1 : and finish up
104 /* We know just enough now to be able to assemble the first
105 full word of s2. We can still find a zero at the end of it.
107 On entry to this basic block:
108 t0 == first word of s1
109 t1 == first partial word of s2. */
111 ldq_u t2, 8(a1) # e0 : load second partial s2 word
112 lda t3, -1 # .. e1 : create leading garbage mask
113 extql t1, a1, t1 # e0 : create first s2 word
114 mskqh t3, a0, t3 # e0 :
115 extqh t2, a1, t4 # e0 :
116 ornot t0, t3, t0 # .. e1 : kill s1 garbage
117 or t1, t4, t1 # e0 : s2 word now complete
118 cmpbge zero, t0, t7 # .. e1 : find zero in first s1 word
119 ornot t1, t3, t1 # e0 : kill s2 garbage
121 mskql t3, a1, t3 # e0 : mask for s2[1] bits we have seen
122 bne t7, $eos # .. e1 :
123 xor t0, t1, t4 # e0 : compare aligned words
124 bne t4, $wordcmp # .. e1 (zdb)
126 cmpbge zero, t3, t7 # e1 :
127 bne t7, $u_final # e1 :
129 /* Unaligned copy main loop. In order to avoid reading too much,
130 the loop is structured to detect zeros in aligned words from s2.
131 This has, unfortunately, effectively pulled half of a loop
132 iteration out into the head and half into the tail, but it does
133 prevent nastiness from accumulating in the very thing we want
134 to run as fast as possible.
136 On entry to this basic block:
137 t2 == the unshifted low-bits from the next s2 word. */
141 extql t2, a1, t3 # e0 :
142 ldq_u t2, 16(a1) # .. e1 : load next s2 high bits
143 ldq_u t0, 8(a0) # e0 : load next s1 word
144 addq a1, 8, a1 # .. e1 :
145 addq a0, 8, a0 # e0 :
147 extqh t2, a1, t1 # e0 :
148 cmpbge zero, t0, t7 # .. e1 : find zero in current s1 word
150 bne t7, $eos # .. e1 :
151 xor t0, t1, t4 # e0 : compare the words
152 bne t4, $wordcmp # .. e1 (zdb)
153 cmpbge zero, t2, t4 # e0 : find zero in next low bits
154 beq t4, $u_loop # .. e1 (zdb)
156 /* We've found a zero in the low bits of the last s2 word. Get
157 the next s1 word and align them. */
159 ldq_u t0, 8(a0) # e1 :
160 extql t2, a1, t1 # .. e0 :
161 cmpbge zero, t1, t7 # e0 :
163 /* We've found a zero somewhere in a word we just read.
164 On entry to this basic block:
167 t7 == cmpbge mask containing the zero. */
171 negq t7, t6 # e0 : create bytemask of valid data
172 and t6, t7, t8 # e1 :
173 subq t8, 1, t6 # e0 :
175 zapnot t0, t7, t0 # e0 : kill the garbage
176 zapnot t1, t7, t1 # .. e1 :
177 xor t0, t1, v0 # e0 : and compare
178 beq v0, $done # .. e1 :
180 /* Here we have two differing co-aligned words in t0 & t1.
181 Bytewise compare them and return (t0 > t1 ? 1 : -1). */
183 cmpbge t0, t1, t2 # e0 : comparison yields bit mask of ge
184 cmpbge t1, t0, t3 # .. e1 :
185 xor t2, t3, t0 # e0 : bits set iff t0/t1 bytes differ
186 negq t0, t1 # e1 : clear all but least bit
187 and t0, t1, t0 # e0 :
189 and t0, t2, t1 # e0 : was bit set in t0 > t1?
190 cmovne t1, 1, v0 # .. e1 (zdb)