1 /* Copyright (C) 1996, 1997, 2003 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 /* Bytewise compare two null-terminated strings. */
39 ldq_u t0, 0(a0) # e0 : give cache time to catch up
40 xor a0, a1, t2 # .. e1 : are s1 and s2 co-aligned?
41 ldq_u t1, 0(a1) # e0 :
42 and t2, 7, t2 # .. e1 :
44 bne t2, $unaligned # .. e1 :
46 /* On entry to this basic block:
47 t0 == the first destination word for masking back in
48 t1 == the first source word.
52 mskqh t3, a0, t3 # e0 :
54 ornot t1, t3, t1 # e0 :
55 ornot t0, t3, t0 # .. e1 :
56 cmpbge zero, t1, t7 # e0 : bits set iff null found
57 bne t7, $eos # e1 (zdb)
59 /* Aligned compare main loop.
60 On entry to this basic block:
62 t1 == an s2 word not containing a null. */
66 bne t2, $wordcmp # .. e1 (zdb)
67 ldq_u t1, 8(a1) # e0 :
68 ldq_u t0, 8(a0) # .. e1 :
70 addq a0, 8, a0 # .. e1 :
71 cmpbge zero, t1, t7 # e0 :
72 beq t7, $a_loop # .. e1 (zdb)
75 /* The two strings are not co-aligned. Align s1 and cope. */
78 and a0, 7, t4 # e0 : find s1 misalignment
79 and a1, 7, t5 # .. e1 : find s2 misalignment
80 subq a1, t4, a1 # e0 :
82 /* If s2 misalignment is larger than s2 misalignment, we need
83 extra startup checks to avoid SEGV. */
85 cmplt t4, t5, t8 # .. e1 :
86 beq t8, $u_head # e1 :
88 mskqh t3, t5, t3 # e0 :
89 ornot t1, t3, t3 # e0 :
90 cmpbge zero, t3, t7 # e1 : is there a zero?
91 beq t7, $u_head # e1 :
93 /* We've found a zero in the first partial word of s2. Align
94 our current s1 and s2 words and compare what we've got. */
96 extql t1, t5, t1 # e0 :
97 extql t0, a0, t0 # e0 :
98 cmpbge zero, t1, t7 # .. e1 : find that zero again
99 br $eos # e1 : and finish up
103 /* We know just enough now to be able to assemble the first
104 full word of s2. We can still find a zero at the end of it.
106 On entry to this basic block:
107 t0 == first word of s1
108 t1 == first partial word of s2. */
110 ldq_u t2, 8(a1) # e0 : load second partial s2 word
111 lda t3, -1 # .. e1 : create leading garbage mask
112 extql t1, a1, t1 # e0 : create first s2 word
113 mskqh t3, a0, t3 # e0 :
114 extqh t2, a1, t4 # e0 :
115 ornot t0, t3, t0 # .. e1 : kill s1 garbage
116 or t1, t4, t1 # e0 : s2 word now complete
117 cmpbge zero, t0, t7 # .. e1 : find zero in first s1 word
118 ornot t1, t3, t1 # e0 : kill s2 garbage
120 mskql t3, a1, t3 # e0 : mask for s2[1] bits we have seen
121 bne t7, $eos # .. e1 :
122 xor t0, t1, t4 # e0 : compare aligned words
123 bne t4, $wordcmp # .. e1 (zdb)
125 cmpbge zero, t3, t7 # e1 :
126 bne t7, $u_final # e1 :
128 /* Unaligned copy main loop. In order to avoid reading too much,
129 the loop is structured to detect zeros in aligned words from s2.
130 This has, unfortunately, effectively pulled half of a loop
131 iteration out into the head and half into the tail, but it does
132 prevent nastiness from accumulating in the very thing we want
133 to run as fast as possible.
135 On entry to this basic block:
136 t2 == the unshifted low-bits from the next s2 word. */
140 extql t2, a1, t3 # e0 :
141 ldq_u t2, 16(a1) # .. e1 : load next s2 high bits
142 ldq_u t0, 8(a0) # e0 : load next s1 word
143 addq a1, 8, a1 # .. e1 :
144 addq a0, 8, a0 # e0 :
146 extqh t2, a1, t1 # e0 :
147 cmpbge zero, t0, t7 # .. e1 : find zero in current s1 word
149 bne t7, $eos # .. e1 :
150 xor t0, t1, t4 # e0 : compare the words
151 bne t4, $wordcmp # .. e1 (zdb)
152 cmpbge zero, t2, t4 # e0 : find zero in next low bits
153 beq t4, $u_loop # .. e1 (zdb)
155 /* We've found a zero in the low bits of the last s2 word. Get
156 the next s1 word and align them. */
158 ldq_u t0, 8(a0) # e1 :
159 extql t2, a1, t1 # .. e0 :
160 cmpbge zero, t1, t7 # e0 :
162 /* We've found a zero somewhere in a word we just read.
163 On entry to this basic block:
166 t7 == cmpbge mask containing the zero. */
170 negq t7, t6 # e0 : create bytemask of valid data
171 and t6, t7, t8 # e1 :
172 subq t8, 1, t6 # e0 :
174 zapnot t0, t7, t0 # e0 : kill the garbage
175 zapnot t1, t7, t1 # .. e1 :
176 xor t0, t1, v0 # e0 : and compare
177 beq v0, $done # .. e1 :
179 /* Here we have two differing co-aligned words in t0 & t1.
180 Bytewise compare them and return (t0 > t1 ? 1 : -1). */
182 cmpbge t0, t1, t2 # e0 : comparison yields bit mask of ge
183 cmpbge t1, t0, t3 # .. e1 :
184 xor t2, t3, t0 # e0 : bits set iff t0/t1 bytes differ
185 negq t0, t1 # e1 : clear all but least bit
186 and t0, t1, t0 # e0 :
188 and t0, t2, t1 # e0 : was bit set in t0 > t1?
189 cmovne t1, 1, v0 # .. e1 (zdb)
195 libc_hidden_builtin_def (strcmp)