Lines Matching +full:0 +full:- +full:7 +full:a +full:- +full:e
1 /* SPDX-License-Identifier: GPL-2.0 */
3 * arch/alpha/lib/ev6-copy_user.S
5 * 21264 version contributed by Rick Gorton <rick.gorton@alpha-processor.com>
10 * This is essentially the same as "memcpy()", but with a few twists.
11 * Notably, we have to make sure that $0 is always up-to-date and
13 * only _after_ a successful copy). There is also some rather minor
19 * ftp.digital.com/pub/Digital/info/semiconductor/literature/dsc-library.html
21 * E - either cluster
22 * U - upper subcluster; U0 - subcluster U0; U1 - subcluster U1
23 * L - lower subcluster; L0 - subcluster L0; L1 - subcluster L1
30 .section __ex_table,"a"; \
31 .long 99b - .; \
32 lda $31, $exitin-99b($31); \
37 .section __ex_table,"a"; \
38 .long 99b - .; \
39 lda $31, $exitout-99b($31); \
48 .prologue 0
49 mov $18, $0 # .. .. .. E
50 subq $18, 32, $1 # .. .. E. .. : Is this going to be a small copy?
51 nop # .. E .. ..
54 and $16,7,$3 # .. .. .. E : is leading dest misalignment
57 subq $3, 8, $3 # E .. .. .. : L U U L : trip counter
59 * The fetcher stall also hides the 1 cycle cross-cluster stall for $3 (L --> U)
60 * This loop aligns the destination a byte at a time
64 EXI( ldbu $1,0($17) ) # .. .. .. L : Keep loads separate from stores
65 addq $16,1,$16 # .. .. E .. : Section 3.8 in the CWG
66 addq $3,1,$3 # .. E .. .. :
67 nop # E .. .. .. : U L U L
70 * the -1 is to compensate for the inc($16) done in a previous quadpack
73 EXO( stb $1,-1($16) ) # .. .. .. L :
74 addq $17,1,$17 # .. .. E .. : Section 3.8 in the CWG
75 subq $0,1,$0 # .. E .. .. :
79 * If we fell through into here, we have a minimum of 33 - 7 bytes
80 * If we arrived via branch, we have a minimum of 32 bytes
83 and $17,7,$1 # .. .. .. E : Check _current_ source alignment
84 bic $0,7,$4 # .. .. E .. : number bytes as a quadword loop
85 EXI( ldq_u $3,0($17) ) # .. L .. .. : Forward fetch for fallthrough code
89 * In the worst case, we've just executed an ldq_u here from 0($17)
93 /* Misaligned quadword loop - not unrolled. Leave it that way. */
96 subq $4,8,$4 # .. .. E .. :
100 bis $3,$1,$1 # .. .. .. E :
101 EXO( stq $1,0($16) ) # .. .. L .. :
102 addq $17,8,$17 # .. E .. .. :
103 subq $0,8,$0 # E .. .. .. : U L L U
105 addq $16,8,$16 # .. .. .. E :
106 bis $2,$2,$3 # .. .. E .. :
107 nop # .. E .. .. :
110 nop # .. .. .. E
111 nop # .. .. E ..
112 nop # .. E .. ..
113 beq $0,$zerolength # U .. .. .. : U L U L
116 EXI ( ldbu $2,0($17) ) # .. .. .. L : No loads in the same quad
117 addq $16,1,$16 # .. .. E .. : as the store (Section 3.8 in CWG)
118 nop # .. E .. .. :
123 * A minimum of (33 - 7) bytes to do a quad at a time.
125 * $0 - number of bytes to be moved
126 * $4 - number of bytes to move as quadwords
131 subq $4, 32, $2 # .. .. .. E : do not unroll for small stuff
132 nop # .. .. E ..
133 nop # .. E .. ..
137 * There is a significant assumption here that the source and destination
138 * addresses differ by more than 32 bytes. In this particular case, a
139 * sparsity of registers further bounds this to be a minimum of 8 bytes.
141 * Furthermore, due to a lack of available registers, we really can't
146 EXI( ldq $1,0($17) ) # .. .. .. L
148 subq $4,32,$4 # .. E .. ..
149 nop # E .. .. .. : U U L L
151 addq $17,16,$17 # .. .. .. E
152 EXO( stq $1,0($16) ) # .. .. L ..
154 subq $0,16,$0 # E .. .. .. : U L L U
156 addq $16,16,$16 # .. .. .. E
157 EXI( ldq $1,0($17) ) # .. .. L ..
159 subq $4, 32, $3 # E .. .. .. : U U L L : is there enough for another trip?
161 EXO( stq $1,0($16) ) # .. .. .. L
163 subq $0,16,$0 # .. E .. ..
164 addq $17,16,$17 # E .. .. .. : U L L U
166 nop # .. .. .. E
167 nop # .. .. E ..
168 addq $16,16,$16 # .. E .. ..
177 EXI( ldq $1,0($17) )
182 EXO( stq $1,0($16) )
183 subq $0,8,$0
191 beq $0,$zerolength
194 * For small copies (or the tail of a larger copy), do a very simple byte loop.
195 * There's no point in doing a lot of complex alignment calculations to try to
196 * to quadword stuff for a small amount of data.
197 * $0 - remaining number of bytes left to copy
198 * $16 - current dest addr
199 * $17 - current source addr
203 EXI ( ldbu $2,0($17) ) # .. .. .. L : No loads in the same quad
204 addq $16,1,$16 # .. .. E .. : as the store (Section 3.8 in CWG)
205 nop # .. E .. .. :
206 nop # E .. .. .. : U L U L
210 * the -1 is to compensate for the inc($16) done in a previous quadpack
213 EXO ( stb $2,-1($16) ) # .. .. .. L :
214 addq $17,1,$17 # .. .. E .. : quadpack as the load
215 subq $0,1,$0 # .. E .. .. : change count _after_ copy
216 bgt $0,$onebyteloop # U .. .. .. : U L U L
221 nop # .. .. .. E
222 nop # .. .. E ..
223 nop # .. E .. ..