Lines Matching +full:4 +full:e
16 * E - either cluster
27 .align 4
34 mov $16, $0 # E : copy dest to return
36 xor $16, $17, $1 # E : are source and dest alignments the same?
37 and $1, 7, $1 # E : are they the same mod 8?
41 and $16, 7, $1 # E : Are both 0mod8?
43 nop # E :
53 subq $18, 1, $18 # E : count--
54 addq $17, 1, $17 # E : src++
56 addq $16, 1, $16 # E : dest++
57 and $16, 7, $1 # E : Are we at 0mod8 yet?
62 cmple $18, 127, $1 # E : Can we unroll the loop?
64 and $16, 63, $1 # E : get mod64 alignment
69 subq $18, 8, $18 # E : count -= 8
70 addq $17, 8, $17 # E : src += 8
71 nop # E :
74 addq $16, 8, $16 # E : dest += 8
75 and $16, 63, $1 # E : get mod64 alignment
79 addq $16, 64, $7 # E : Initial (+1 trip) wh64 address
80 cmple $18, 127, $1 # E : Can we go through the unrolled loop?
82 nop # E :
88 nop # E :
89 nop # E :
91 ldq $4, 8($17) # L : bytes 8..15
93 addq $7, 64, $7 # E : Update next wh64 address
94 nop # E :
97 addq $16, 64, $1 # E : fallback value for wh64
98 nop # E :
99 nop # E :
101 addq $17, 32, $17 # E : src += 32 bytes
103 nop # E :
104 nop # E :
106 stq $4, 8($16) # L : bytes 8..15
108 subq $18, 192, $2 # E : At least two more trips to go?
109 nop # E :
112 addq $16, 32, $16 # E : dest += 32 bytes
113 nop # E :
114 nop # E :
117 ldq $4, 8($17) # L : bytes 8..15
118 cmovlt $2, $1, $7 # E : Latency 2, extra map slot - Use
120 nop # E :
124 addq $16, 32, $16 # E : dest += 32
125 subq $18, 64, $18 # E : count -= 64
127 addq $17, 32, $17 # E : src += 32
129 stq $4, -24($16) # L : bytes 8..15
130 cmple $18, 63, $1 # E : At least one more trip?
134 nop # E :
139 .align 4
140 subq $18, 8, $18 # E : At least a quad left?
142 nop # E :
143 nop # E :
147 subq $18, 8, $18 # E : count -= 8
148 addq $17, 8, $17 # E : src += 8
149 nop # E :
152 addq $16, 8, $16 # E : dest += 8
154 nop # E :
157 .align 4
158 addq $18, 8, $18 # E : add back for trailing bytes
160 nop # E :
161 nop # E :
165 subq $18, 1, $18 # E : count--
167 addq $17, 1, $17 # E : src++
168 nop # E :
171 addq $16, 1, $16 # E : dest++
173 nop # E :
177 nop # E :
178 nop # E :
179 nop # E :
182 mov $0, $4 # E : dest temp
183 and $0, 7, $1 # E : dest alignment mod8
190 subq $18, 1, $18 # E : count--
191 addq $17, 1, $17 # E : src++
193 stb $1, 0($4) # L : store it
194 addq $4, 1, $4 # E : dest++
195 and $4, 7, $1 # E : dest 0mod8 yet?
200 subq $18, 8, $18 # E : At least a quad left?
203 nop # E :
209 bis $3, $1, $1 # E : merged bytes to store
211 subq $18, 8, $18 # E : count -= 8
212 addq $17, 8, $17 # E : src += 8
213 stq $1, 0($4) # L : store 8 (aligned)
214 mov $16, $3 # E : "rotate" source data
216 addq $4, 8, $4 # E : dest += 8
222 addq $18, 8, $18 # E : account for tail stuff
229 subq $18, 1, $18 # E : count--
230 addq $17, 1, $17 # E : src++
231 nop # E :
233 stb $1, 0($4) # L : store
234 addq $4, 1, $4 # E : dest++
241 nop # E :
242 nop # E :
243 nop # E :