Lines Matching +full:- +full:e
1 /* SPDX-License-Identifier: GPL-2.0 */
3 * arch/alpha/lib/ev6-memchr.S
5 * 21264 version contributed by Rick Gorton <rick.gorton@alpha-processor.com>
9 * - memory accessed as aligned quadwords only
10 * - uses cmpbge to compare 8 bytes in parallel
11 * - does binary search to find 0 byte in last
18 * - only minimum number of quadwords may be accessed
19 * - the third argument is an unsigned long
24 * ftp.digital.com/pub/Digital/info/semiconductor/literature/dsc-library.html
26 * E - either cluster
27 * U - upper subcluster; U0 - subcluster U0; U1 - subcluster U1
28 * L - lower subcluster; L0 - subcluster L0; L1 - subcluster L1
42 # Hack -- if someone passes in (size_t)-1, hoping to just
45 # that we will never have a 56-bit address space, cropping
50 and $17, 0xff, $17 # E : L L U U : 00000000000000ch
53 cmpult $18, 9, $4 # E : small (< 1 quad) string?
54 or $2, $17, $17 # E : 000000000000chch
55 lda $3, -1($31) # E : U L L U
58 addq $16, $5, $5 # E : Max search address
59 or $2, $17, $17 # E : 00000000chchchch
62 or $2, $17, $17 # E : chchchchchchchch
65 ldq_u $6, -1($5) # L : L U U L : eight or less bytes to search Latency=3
68 mov $16, $0 # E :
69 nop # E :
70 or $7, $6, $1 # E : L U L U $1 = quadword starting at $16
73 # in $1. E.g.:
77 negq $18, $6 # E :
78 xor $17, $1, $1 # E :
80 cmpbge $31, $1, $2 # E : L U L U
84 and $2, $6, $2 # E :
94 addq $0, $3, $0 # E : All done
95 nop # E :
102 negq $2, $3 # E :
103 and $2, $3, $2 # E :
104 and $2, 0x0f, $1 # E :
105 addq $0, 4, $3 # E :
107 cmoveq $1, $3, $0 # E : Latency 2, extra map cycle
108 nop # E : keep with cmov
109 and $2, 0x33, $1 # E :
110 addq $0, 2, $3 # E : U L U L : 2 cycle stall on $0
112 cmoveq $1, $3, $0 # E : Latency 2, extra map cycle
113 nop # E : keep with cmov
114 and $2, 0x55, $1 # E :
115 addq $0, 1, $3 # E : U L U L : 2 cycle stall on $0
117 cmoveq $1, $3, $0 # E : Latency 2, extra map cycle
127 andnot $16, 0x7, $0 # E :
129 xor $1, $17, $1 # E :
130 or $1, $2, $1 # E : U L U L $1 = ====ffffffffffff
132 cmpbge $31, $1, $2 # E :
136 subq $5, 1, $18 # E : U L U L
138 addq $0, 8, $0 # E :
141 andnot $18, 0x7, $18 # E :
142 cmpult $0, $18, $2 # E :
147 subq $18, $0, $4 # E : $4 <- nr quads to be processed
148 and $4, 8, $4 # E : odd number of quads?
151 mov $1, $4 # E : L U L U : move prefetched value to correct reg
156 xor $17, $4, $2 # E :
157 cmpbge $31, $2, $2 # E :
160 addq $0, 8, $0 # E :
161 nop # E :
162 nop # E :
163 nop # E :
166 xor $17, $1, $2 # E :
168 cmpbge $31, $2, $2 # E :
169 addq $0, 8, $6 # E :
172 cmpult $6, $18, $6 # E :
173 addq $0, 8, $0 # E :
174 nop # E :
177 mov $4, $1 # E : move prefetched value into $1
178 nop # E :
179 nop # E :
181 $final: subq $5, $0, $18 # E : $18 <- number of bytes left to do
182 nop # E :
183 nop # E :
187 mov $31, $0 # E :
188 nop # E :
189 nop # E :