diff options
Diffstat (limited to 'firmware/target/sh/archos/ata-as-archos.S')
-rwxr-xr-x | firmware/target/sh/archos/ata-as-archos.S | 231 |
1 files changed, 231 insertions, 0 deletions
diff --git a/firmware/target/sh/archos/ata-as-archos.S b/firmware/target/sh/archos/ata-as-archos.S new file mode 100755 index 0000000000..4a4e7e4b94 --- /dev/null +++ b/firmware/target/sh/archos/ata-as-archos.S | |||
@@ -0,0 +1,231 @@ | |||
1 | /*************************************************************************** | ||
2 | * __________ __ ___. | ||
3 | * Open \______ \ ____ ____ | | _\_ |__ _______ ___ | ||
4 | * Source | _// _ \_/ ___\| |/ /| __ \ / _ \ \/ / | ||
5 | * Jukebox | | ( <_> ) \___| < | \_\ ( <_> > < < | ||
6 | * Firmware |____|_ /\____/ \___ >__|_ \|___ /\____/__/\_ \ | ||
7 | * \/ \/ \/ \/ \/ | ||
8 | * $Id$ | ||
9 | * | ||
10 | * Copyright (C) 2004-2006 by Jens Arnold | ||
11 | * | ||
12 | * All files in this archive are subject to the GNU General Public License. | ||
13 | * See the file COPYING in the source tree root for full license agreement. | ||
14 | * | ||
15 | * This software is distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY | ||
16 | * KIND, either express or implied. | ||
17 | * | ||
18 | ****************************************************************************/ | ||
19 | |||
20 | .section .icode,"ax",@progbits | ||
21 | |||
22 | .align 2 | ||
23 | .global _copy_read_sectors | ||
24 | .type _copy_read_sectors,@function | ||
25 | |||
26 | /* Read a number of words from the ATA data port | ||
27 | * | ||
28 | * Assumes wordcount to be a multiple of 4 | ||
29 | * | ||
30 | * Arguments: | ||
31 | * r4 - buffer address | ||
32 | * r5 - word count | ||
33 | * | ||
34 | * Register usage: | ||
35 | * r0 - scratch | ||
36 | * r1/r2 - read buffers | ||
37 | * r3 - mask (if unaligned) | ||
38 | * r4 - current address | ||
39 | * r5 - end address | ||
40 | * r6 - ata port | ||
41 | */ | ||
42 | |||
43 | _copy_read_sectors: | ||
44 | add r5, r5 /* words -> bytes */ | ||
45 | add r4, r5 /* bytes -> end address */ | ||
46 | add #-12, r5 /* adjust for offsets */ | ||
47 | mov.l .ata_data, r6 | ||
48 | |||
49 | mov r4, r0 | ||
50 | tst #1, r0 /* 16-bit aligned ? */ | ||
51 | bt .r_aligned /* yes, do word copy */ | ||
52 | |||
53 | /* not 16-bit aligned */ | ||
54 | mov #-1, r3 /* prepare a bit mask for high byte */ | ||
55 | shll8 r3 /* r3 = 0xFFFFFF00 */ | ||
56 | |||
57 | mov.w @r6, r2 /* read first word (1st round) */ | ||
58 | mov.b r2, @r4 /* store low byte of first word */ | ||
59 | bra .r_start_b /* jump into loop after next instr. */ | ||
60 | add #-5, r4 /* adjust for dest. offsets; now even */ | ||
61 | |||
62 | .align 2 | ||
63 | .r_loop_b: /* main loop: copy 4 words in a row */ | ||
64 | mov.w @r6, r2 /* read first word (2+ round) */ | ||
65 | and r3, r1 /* get high byte of fourth word (2+ round) */ | ||
66 | extu.b r2, r0 /* get low byte of first word (2+ round) */ | ||
67 | or r1, r0 /* combine with high byte of fourth word */ | ||
68 | mov.w r0, @(4, r4) /* store at buf[4] */ | ||
69 | nop /* maintain alignment */ | ||
70 | .r_start_b: | ||
71 | mov.w @r6, r1 /* read second word */ | ||
72 | and r3, r2 /* get high byte of first word */ | ||
73 | extu.b r1, r0 /* get low byte of second word */ | ||
74 | or r2, r0 /* combine with high byte of first word */ | ||
75 | mov.w r0, @(6, r4) /* store at buf[6] */ | ||
76 | add #8, r4 /* buf += 8 */ | ||
77 | mov.w @r6, r2 /* read third word */ | ||
78 | and r3, r1 /* get high byte of second word */ | ||
79 | extu.b r2, r0 /* get low byte of third word */ | ||
80 | or r1, r0 /* combine with high byte of second word */ | ||
81 | mov.w r0, @r4 /* store at buf[0] */ | ||
82 | cmp/hi r4, r5 /* check for end */ | ||
83 | mov.w @r6, r1 /* read fourth word */ | ||
84 | and r3, r2 /* get high byte of third word */ | ||
85 | extu.b r1, r0 /* get low byte of fourth word */ | ||
86 | or r2, r0 /* combine with high byte of third word */ | ||
87 | mov.w r0, @(2, r4) /* store at buf[2] */ | ||
88 | bt .r_loop_b | ||
89 | /* 24 instructions for 4 copies, takes 30 clock cycles (4 wait) */ | ||
90 | /* avg. 7.5 cycles per word */ | ||
91 | |||
92 | swap.b r1, r0 /* get high byte of last word */ | ||
93 | rts | ||
94 | mov.b r0, @(4, r4) /* and store it */ | ||
95 | |||
96 | /* 16-bit aligned, loop(read and store word) */ | ||
97 | .r_aligned: | ||
98 | mov.w @r6, r2 /* read first word (1st round) */ | ||
99 | bra .r_start_w /* jump into loop after next instr. */ | ||
100 | add #-6, r4 /* adjust for destination offsets */ | ||
101 | |||
102 | .align 2 | ||
103 | .r_loop_w: /* main loop: copy 4 words in a row */ | ||
104 | mov.w @r6, r2 /* read first word (2+ round) */ | ||
105 | swap.b r1, r0 /* swap fourth word (2+ round) */ | ||
106 | mov.w r0, @(4, r4) /* store fourth word (2+ round) */ | ||
107 | nop /* maintain alignment */ | ||
108 | .r_start_w: | ||
109 | mov.w @r6, r1 /* read second word */ | ||
110 | swap.b r2, r0 /* swap first word */ | ||
111 | mov.w r0, @(6, r4) /* store first word in buf[6] */ | ||
112 | add #8, r4 /* buf += 8 */ | ||
113 | mov.w @r6, r2 /* read third word */ | ||
114 | swap.b r1, r0 /* swap second word */ | ||
115 | mov.w r0, @r4 /* store second word in buf[0] */ | ||
116 | cmp/hi r4, r5 /* check for end */ | ||
117 | mov.w @r6, r1 /* read fourth word */ | ||
118 | swap.b r2, r0 /* swap third word */ | ||
119 | mov.w r0, @(2, r4) /* store third word */ | ||
120 | bt .r_loop_w | ||
121 | /* 16 instructions for 4 copies, takes 22 clock cycles (4 wait) */ | ||
122 | /* avg. 5.5 cycles per word */ | ||
123 | |||
124 | swap.b r1, r0 /* swap fourth word (last round) */ | ||
125 | rts | ||
126 | mov.w r0, @(4, r4) /* and store it */ | ||
127 | |||
128 | .r_end: | ||
129 | .size _copy_read_sectors,.r_end-_copy_read_sectors | ||
130 | |||
131 | .align 2 | ||
132 | .global _copy_write_sectors | ||
133 | .type _copy_write_sectors,@function | ||
134 | |||
135 | /* Write a number of words to the ATA data port | ||
136 | * | ||
137 | * Assumes wordcount to be a multiple of 2. | ||
138 | * Writing is not unrolled as much as reading, for several reasons: | ||
139 | * | ||
140 | * - a similar instruction sequence is faster for writing than for reading | ||
141 | * because the auto-incrementing load inctructions can be used | ||
142 | * - writing profits from warp mode | ||
143 | * | ||
144 | * Both of these add up to have writing faster than the more unrolled reading. | ||
145 | * | ||
146 | * Arguments: | ||
147 | * r4 - buffer address | ||
148 | * r5 - word count | ||
149 | * | ||
150 | * Register usage: | ||
151 | * r0/r1 - scratch | ||
152 | * r2/r3 - write buffers | ||
153 | * r4 - current address | ||
154 | * r5 - end address | ||
155 | * r6 - mask (if unaligned) | ||
156 | * r7 - ata port | ||
157 | */ | ||
158 | |||
159 | _copy_write_sectors: | ||
160 | add r5, r5 /* words -> bytes */ | ||
161 | add r4, r5 /* bytes -> end address */ | ||
162 | add #-4, r5 /* adjust for offsets */ | ||
163 | mov.l .ata_data, r7 | ||
164 | |||
165 | mov r4, r0 | ||
166 | tst #1, r0 /* 16-bit aligned ? */ | ||
167 | bt .w_aligned /* yes, do word copy */ | ||
168 | |||
169 | /* not 16-bit aligned */ | ||
170 | mov #-1, r6 /* prepare a bit mask for high byte */ | ||
171 | shll8 r6 /* r6 = 0xFFFFFF00 */ | ||
172 | |||
173 | mov.b @r4+, r2 /* load (initial old second) first byte */ | ||
174 | mov.w @r4+, r3 /* load (initial) first word */ | ||
175 | bra .w_start_b | ||
176 | extu.b r2, r0 /* extend unsigned */ | ||
177 | |||
178 | .align 2 | ||
179 | .w_loop_b: /* main loop: copy 2 words in a row */ | ||
180 | mov.w @r4+, r3 /* load first word (2+ round) */ | ||
181 | extu.b r2, r0 /* put away low byte of second word (2+ round) */ | ||
182 | and r6, r2 /* get high byte of second word (2+ round) */ | ||
183 | or r1, r2 /* combine with low byte of old first word */ | ||
184 | mov.w r2, @r7 /* write that */ | ||
185 | .w_start_b: | ||
186 | cmp/hi r4, r5 /* check for end */ | ||
187 | mov.w @r4+, r2 /* load second word */ | ||
188 | extu.b r3, r1 /* put away low byte of first word */ | ||
189 | and r6, r3 /* get high byte of first word */ | ||
190 | or r0, r3 /* combine with high byte of old second word */ | ||
191 | mov.w r3, @r7 /* write that */ | ||
192 | bt .w_loop_b | ||
193 | /* 12 instructions for 2 copies, takes 14 clock cycles */ | ||
194 | /* avg. 7 cycles per word */ | ||
195 | |||
196 | /* the loop "overreads" 1 byte past the buffer end, however, the last */ | ||
197 | /* byte is not written to disk */ | ||
198 | and r6, r2 /* get high byte of last word */ | ||
199 | or r1, r2 /* combine with low byte of old first word */ | ||
200 | rts | ||
201 | mov.w r2, @r7 /* write last word */ | ||
202 | |||
203 | /* 16-bit aligned, loop(load and write word) */ | ||
204 | .w_aligned: | ||
205 | bra .w_start_w /* jump into loop after next instr. */ | ||
206 | mov.w @r4+, r2 /* load first word (1st round) */ | ||
207 | |||
208 | .align 2 | ||
209 | .w_loop_w: /* main loop: copy 2 words in a row */ | ||
210 | mov.w @r4+, r2 /* load first word (2+ round) */ | ||
211 | swap.b r1, r0 /* swap second word (2+ round) */ | ||
212 | mov.w r0, @r7 /* write second word (2+ round) */ | ||
213 | .w_start_w: | ||
214 | cmp/hi r4, r5 /* check for end */ | ||
215 | mov.w @r4+, r1 /* load second word */ | ||
216 | swap.b r2, r0 /* swap first word */ | ||
217 | mov.w r0, @r7 /* write first word */ | ||
218 | bt .w_loop_w | ||
219 | /* 8 instructions for 2 copies, takes 10 clock cycles */ | ||
220 | /* avg. 5 cycles per word */ | ||
221 | |||
222 | swap.b r1, r0 /* swap second word (last round) */ | ||
223 | rts | ||
224 | mov.w r0, @r7 /* and write it */ | ||
225 | |||
226 | .w_end: | ||
227 | .size _copy_write_sectors,.w_end-_copy_write_sectors | ||
228 | |||
229 | .align 2 | ||
230 | .ata_data: | ||
231 | .long 0x06104100 /* ATA data port */ | ||