summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorLinus Nielsen Feltzing <linus@haxx.se>2006-07-25 11:16:03 +0000
committerLinus Nielsen Feltzing <linus@haxx.se>2006-07-25 11:16:03 +0000
commit5a0cb68e43eacc98d275baef08296f154f3ae6dd (patch)
tree07e93ecd803fc76b2605d0610a5c19d584d13623
parent9051afa8a061152941ed4cf99e0ceb9bc4a3caf1 (diff)
downloadrockbox-5a0cb68e43eacc98d275baef08296f154f3ae6dd.tar.gz
rockbox-5a0cb68e43eacc98d275baef08296f154f3ae6dd.zip
Moved some assembly optimizations to the target tree
git-svn-id: svn://svn.rockbox.org/rockbox/trunk@10318 a1c6a512-1295-4272-9138-f99709370657
Diffstat
-rw-r--r--firmware/SOURCES22
-rwxr-xr-xfirmware/target/arm/memset-arm.S96
-rwxr-xr-x[-rw-r--r--]firmware/target/coldfire/memcpy-coldfire.S (renamed from firmware/common/memcpy_a.S)198
-rwxr-xr-xfirmware/target/coldfire/memmove-coldfire.S (renamed from firmware/common/memmove_a.S)201
-rwxr-xr-x[-rw-r--r--]firmware/target/coldfire/memset-coldfire.S (renamed from firmware/common/memset_a.S)167
-rwxr-xr-xfirmware/target/coldfire/memset16-coldfire.S (renamed from firmware/common/memset16_a.S)2
-rwxr-xr-xfirmware/target/sh/memcpy-sh.S217
-rwxr-xr-xfirmware/target/sh/memmove-sh.S220
-rwxr-xr-xfirmware/target/sh/memset-sh.S107
-rwxr-xr-xfirmware/target/sh/strlen-sh.S (renamed from firmware/common/strlen_a.S)0
10 files changed, 652 insertions, 578 deletions
diff --git a/firmware/SOURCES b/firmware/SOURCES
index 7d93edaca1..2979e33ccc 100644
--- a/firmware/SOURCES
+++ b/firmware/SOURCES
@@ -30,7 +30,7 @@ common/strchr.c
30common/strcmp.c 30common/strcmp.c
31common/strcpy.c 31common/strcpy.c
32#if (CONFIG_CPU == SH7034) && !defined(SIMULATOR) 32#if (CONFIG_CPU == SH7034) && !defined(SIMULATOR)
33common/strlen_a.S 33target/sh/strlen-sh.S
34#else 34#else
35common/strlen.c 35common/strlen.c
36#endif 36#endif
@@ -39,22 +39,24 @@ common/strncpy.c
39common/strrchr.c 39common/strrchr.c
40common/strtok.c 40common/strtok.c
41common/timefuncs.c 41common/timefuncs.c
42#if (CONFIG_CPU == SH7034) || defined(CPU_COLDFIRE) 42
43common/memcpy_a.S 43#ifdef CPU_COLDFIRE
44common/memmove_a.S 44target/coldfire/memcpy-coldfire.S
45common/memset_a.S 45target/coldfire/memmove-coldfire.S
46target/coldfire/memset-coldfire.S
47target/coldfire/memset16-coldfire.S
48#elif (CONFIG_CPU == SH7034)
49target/sh/memcpy-sh.S
50target/sh/memmove-sh.S
51target/sh/memset-sh.S
46#elif defined(CPU_ARM) 52#elif defined(CPU_ARM)
47common/memcpy.c 53common/memcpy.c
48common/memmove.c 54common/memmove.c
49common/memset_a.S 55target/arm/memset-arm.S
50#else 56#else
51common/memcpy.c 57common/memcpy.c
52common/memmove.c 58common/memmove.c
53common/memset.c 59common/memset.c
54#endif
55#ifdef CPU_COLDFIRE
56common/memset16_a.S
57#else
58common/memset16.c 60common/memset16.c
59#endif 61#endif
60#ifdef HAVE_LCD_CHARCELLS 62#ifdef HAVE_LCD_CHARCELLS
diff --git a/firmware/target/arm/memset-arm.S b/firmware/target/arm/memset-arm.S
new file mode 100755
index 0000000000..b3faafcb37
--- /dev/null
+++ b/firmware/target/arm/memset-arm.S
@@ -0,0 +1,96 @@
1/***************************************************************************
2 * __________ __ ___.
3 * Open \______ \ ____ ____ | | _\_ |__ _______ ___
4 * Source | _// _ \_/ ___\| |/ /| __ \ / _ \ \/ /
5 * Jukebox | | ( <_> ) \___| < | \_\ ( <_> > < <
6 * Firmware |____|_ /\____/ \___ >__|_ \|___ /\____/__/\_ \
7 * \/ \/ \/ \/ \/
8 * $Id$
9 *
10 * Copyright (C) 2004 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#include "config.h"
20
21 .section .icode,"ax",%progbits
22
23 .align 2
24
25/* The following code is based on code found in Linux kernel version 2.6.15.3
26 * linux/arch/arm/lib/memset.S
27 *
28 * Copyright (C) 1995-2000 Russell King
29 */
30
31/* This code will align a pointer for memset, if needed */
321: cmp r2, #4 @ 1 do we have enough
33 blt 5f @ 1 bytes to align with?
34 cmp r3, #2 @ 1
35 strgtb r1, [r0, #-1]! @ 1
36 strgeb r1, [r0, #-1]! @ 1
37 strb r1, [r0, #-1]! @ 1
38 sub r2, r2, r3 @ 1 r2 = r2 - r3
39 b 2f
40
41 .global memset
42 .type memset,%function
43memset:
44 add r0, r0, r2 @ we'll write backwards in memory
45 ands r3, r0, #3 @ 1 unaligned?
46 bne 1b @ 1
472:
48/*
49 * we know that the pointer in r0 is aligned to a word boundary.
50 */
51 orr r1, r1, r1, lsl #8
52 orr r1, r1, r1, lsl #16
53 mov r3, r1
54 cmp r2, #16
55 blt 5f
56/*
57 * We need an extra register for this loop - save the return address and
58 * use the LR
59 */
60 str lr, [sp, #-4]!
61 mov ip, r1
62 mov lr, r1
63
643: subs r2, r2, #64
65 stmgedb r0!, {r1, r3, ip, lr} @ 64 bytes at a time.
66 stmgedb r0!, {r1, r3, ip, lr}
67 stmgedb r0!, {r1, r3, ip, lr}
68 stmgedb r0!, {r1, r3, ip, lr}
69 bgt 3b
70 ldmeqfd sp!, {pc} @ Now <64 bytes to go.
71/*
72 * No need to correct the count; we're only testing bits from now on
73 */
74 tst r2, #32
75 stmnedb r0!, {r1, r3, ip, lr}
76 stmnedb r0!, {r1, r3, ip, lr}
77 tst r2, #16
78 stmnedb r0!, {r1, r3, ip, lr}
79 ldr lr, [sp], #4
80
815: tst r2, #8
82 stmnedb r0!, {r1, r3}
83 tst r2, #4
84 strne r1, [r0, #-4]!
85/*
86 * When we get here, we've got less than 4 bytes to zero. We
87 * may have an unaligned pointer as well.
88 */
896: tst r2, #2
90 strneb r1, [r0, #-1]!
91 strneb r1, [r0, #-1]!
92 tst r2, #1
93 strneb r1, [r0, #-1]!
94 mov pc, lr
95end:
96 .size memset,.end-memset
diff --git a/firmware/common/memcpy_a.S b/firmware/target/coldfire/memcpy-coldfire.S
index 9f6c813be3..523e1f5ed9 100644..100755
--- a/firmware/common/memcpy_a.S
+++ b/firmware/target/coldfire/memcpy-coldfire.S
@@ -20,203 +20,6 @@
20 20
21 .section .icode,"ax",@progbits 21 .section .icode,"ax",@progbits
22 22
23#if CONFIG_CPU == SH7034
24 .align 2
25 .global _memcpy
26 .global ___memcpy_fwd_entry
27 .type _memcpy,@function
28
29/* Copies <length> bytes of data in memory from <source> to <dest>
30 * This version is optimized for speed
31 *
32 * arguments:
33 * r4 - destination address
34 * r5 - source address
35 * r6 - length
36 *
37 * return value:
38 * r0 - destination address (like ANSI version)
39 *
40 * register usage:
41 * r0 - data / scratch
42 * r1 - 2nd data / scratch
43 * r2 - scratch
44 * r3 - first long bound / adjusted end address (only if >= 11 bytes)
45 * r4 - current dest address
46 * r5 - current source address
47 * r6 - source end address
48 * r7 - stored dest start address
49 *
50 * The instruction order is devised in a way to utilize the pipelining
51 * of the SH1 to the max. The routine also tries to utilize fast page mode.
52 */
53
54_memcpy:
55 mov r4,r7 /* store dest for returning */
56___memcpy_fwd_entry:
57 add #-8,r4 /* offset for early increment (max. 2 longs) */
58 mov #11,r0
59 cmp/hs r0,r6 /* at least 11 bytes to copy? (ensures 2 aligned longs) */
60 add r5,r6 /* r6 = source_end */
61 bf .start_b2 /* no: jump directly to byte loop */
62
63 mov #3,r0
64 neg r5,r3
65 and r0,r3 /* r3 = (4 - align_offset) % 4 */
66 tst r3,r3 /* already aligned? */
67 bt .end_b1 /* yes: skip leading byte loop */
68
69 add r5,r3 /* r3 = first source long bound */
70
71 /* leading byte loop: copies 0..3 bytes */
72.loop_b1:
73 mov.b @r5+,r0 /* load byte & increment source addr */
74 add #1,r4 /* increment dest addr */
75 mov.b r0,@(7,r4) /* store byte */
76 cmp/hi r5,r3 /* runs r5 up to first long bound */
77 bt .loop_b1
78 /* now r5 is always at a long boundary */
79 /* -> memory reading is done in longs for all dest alignments */
80
81 /* selector for main copy loop */
82.end_b1:
83 mov #3,r1
84 and r4,r1 /* r1 = dest alignment offset */
85 mova .jmptab,r0
86 mov.b @(r0,r1),r1 /* select appropriate main loop */
87 add r0,r1
88 mov r6,r3 /* move end address to r3 */
89 jmp @r1 /* and jump to it */
90 add #-7,r3 /* adjust end addr for main loops doing 2 longs/pass */
91
92 /** main loops, copying 2 longs per pass to profit from fast page mode **/
93
94 /* long aligned destination (fastest) */
95 .align 2
96.loop_do0:
97 mov.l @r5+,r1 /* load first long & increment source addr */
98 add #16,r4 /* increment dest addr & account for decrementing stores */
99 mov.l @r5+,r0 /* load second long & increment source addr */
100 cmp/hi r5,r3 /* runs r5 up to last or second last long bound */
101 mov.l r0,@-r4 /* store second long */
102 mov.l r1,@-r4 /* store first long; NOT ALIGNED - no speed loss here! */
103 bt .loop_do0
104
105 add #4,r3 /* readjust end address */
106 cmp/hi r5,r3 /* one long left? */
107 bf .start_b2 /* no, jump to trailing byte loop */
108
109 mov.l @r5+,r0 /* load last long & increment source addr */
110 add #4,r4 /* increment dest addr */
111 bra .start_b2 /* jump to trailing byte loop */
112 mov.l r0,@(4,r4) /* store last long */
113
114 /* word aligned destination (long + 2) */
115 .align 2
116.loop_do2:
117 mov.l @r5+,r1 /* load first long & increment source addr */
118 add #16,r4 /* increment dest addr */
119 mov.l @r5+,r0 /* load second long & increment source addr */
120 cmp/hi r5,r3 /* runs r5 up to last or second last long bound */
121 mov.w r0,@-r4 /* store low word of second long */
122 xtrct r1,r0 /* extract low word of first long & high word of second long */
123 mov.l r0,@-r4 /* and store as long */
124 swap.w r1,r0 /* get high word of first long */
125 mov.w r0,@-r4 /* and store it */
126 bt .loop_do2
127
128 add #4,r3 /* readjust end address */
129 cmp/hi r5,r3 /* one long left? */
130 bf .start_b2 /* no, jump to trailing byte loop */
131
132 mov.l @r5+,r0 /* load last long & increment source addr */
133 add #4,r4 /* increment dest addr */
134 mov.w r0,@(6,r4) /* store low word */
135 shlr16 r0 /* get high word */
136 bra .start_b2 /* jump to trailing byte loop */
137 mov.w r0,@(4,r4) /* and store it */
138
139 /* jumptable for loop selector */
140 .align 2
141.jmptab:
142 .byte .loop_do0 - .jmptab /* placed in the middle because the SH1 */
143 .byte .loop_do1 - .jmptab /* loads bytes sign-extended. Otherwise */
144 .byte .loop_do2 - .jmptab /* the last loop would be out of reach */
145 .byte .loop_do3 - .jmptab /* of the offset range. */
146
147 /* byte aligned destination (long + 1) */
148 .align 2
149.loop_do1:
150 mov.l @r5+,r1 /* load first long & increment source addr */
151 add #16,r4 /* increment dest addr */
152 mov.l @r5+,r0 /* load second long & increment source addr */
153 cmp/hi r5,r3 /* runs r5 up to last or second last long bound */
154 mov.b r0,@-r4 /* store low byte of second long */
155 shlr8 r0 /* get upper 3 bytes */
156 mov r1,r2 /* copy first long */
157 shll16 r2 /* move low byte of first long all the way up, .. */
158 shll8 r2
159 or r2,r0 /* ..combine with the 3 bytes of second long.. */
160 mov.l r0,@-r4 /* ..and store as long */
161 shlr8 r1 /* get middle 2 bytes */
162 mov.w r1,@-r4 /* store as word */
163 shlr16 r1 /* get upper byte */
164 mov.b r1,@-r4 /* and store */
165 bt .loop_do1
166
167 add #4,r3 /* readjust end address */
168.last_do13:
169 cmp/hi r5,r3 /* one long left? */
170 bf .start_b2 /* no, jump to trailing byte loop */
171
172 mov.l @r5+,r0 /* load last long & increment source addr */
173 add #12,r4 /* increment dest addr */
174 mov.b r0,@-r4 /* store low byte */
175 shlr8 r0 /* get middle 2 bytes */
176 mov.w r0,@-r4 /* store as word */
177 shlr16 r0 /* get upper byte */
178 mov.b r0,@-r4 /* and store */
179 bra .start_b2 /* jump to trailing byte loop */
180 add #-4,r4 /* readjust destination */
181
182 /* byte aligned destination (long + 3) */
183 .align 2
184.loop_do3:
185 mov.l @r5+,r1 /* load first long & increment source addr */
186 add #16,r4 /* increment dest addr */
187 mov.l @r5+,r0 /* load second long & increment source addr */
188 mov r1,r2 /* copy first long */
189 mov.b r0,@-r4 /* store low byte of second long */
190 shlr8 r0 /* get middle 2 bytes */
191 mov.w r0,@-r4 /* store as word */
192 shlr16 r0 /* get upper byte */
193 shll8 r2 /* move lower 3 bytes of first long one up.. */
194 or r2,r0 /* ..combine with the 1 byte of second long.. */
195 mov.l r0,@-r4 /* ..and store as long */
196 shlr16 r1 /* get upper byte of first long.. */
197 shlr8 r1
198 cmp/hi r5,r3 /* runs r5 up to last or second last long bound */
199 mov.b r1,@-r4 /* ..and store */
200 bt .loop_do3
201
202 bra .last_do13 /* handle last longword: reuse routine for (long + 1) */
203 add #4,r3 /* readjust end address */
204
205 /* trailing byte loop: copies 0..3 bytes (or all for < 11 in total) */
206 .align 2
207.loop_b2:
208 mov.b @r5+,r0 /* load byte & increment source addr */
209 add #1,r4 /* increment dest addr */
210 mov.b r0,@(7,r4) /* store byte */
211.start_b2:
212 cmp/hi r5,r6 /* runs r5 up to end address */
213 bt .loop_b2
214
215 rts
216 mov r7,r0 /* return dest start address */
217.end:
218 .size _memcpy,.end-_memcpy
219#elif defined(CPU_COLDFIRE)
220#define FULLSPEED /* use burst writing for word aligned destinations */ 23#define FULLSPEED /* use burst writing for word aligned destinations */
221 .align 2 24 .align 2
222 .global memcpy 25 .global memcpy
@@ -875,4 +678,3 @@ __memcpy_fwd_entry:
875 678
876.end: 679.end:
877 .size memcpy,.end-memcpy 680 .size memcpy,.end-memcpy
878#endif
diff --git a/firmware/common/memmove_a.S b/firmware/target/coldfire/memmove-coldfire.S
index d7421333df..bdd2e2e206 100755
--- a/firmware/common/memmove_a.S
+++ b/firmware/target/coldfire/memmove-coldfire.S
@@ -20,206 +20,6 @@
20 20
21 .section .icode,"ax",@progbits 21 .section .icode,"ax",@progbits
22 22
23#if CONFIG_CPU == SH7034
24 .align 2
25 .global _memmove
26 .type _memmove,@function
27
28/* Moves <length> bytes of data in memory from <source> to <dest>
29 * Regions may overlap.
30 * This version is optimized for speed, and needs the corresponding memcpy
31 * implementation for the forward copy branch.
32 *
33 * arguments:
34 * r4 - destination address
35 * r5 - source address
36 * r6 - length
37 *
38 * return value:
39 * r0 - destination address (like ANSI version)
40 *
41 * register usage:
42 * r0 - data / scratch
43 * r1 - 2nd data / scratch
44 * r2 - scratch
45 * r3 - last long bound / adjusted start address (only if >= 11 bytes)
46 * r4 - current dest address
47 * r5 - source start address
48 * r6 - current source address
49 *
50 * The instruction order is devised in a way to utilize the pipelining
51 * of the SH1 to the max. The routine also tries to utilize fast page mode.
52 */
53
54_memmove:
55 cmp/hi r4,r5 /* source > destination */
56 bf .backward /* no: backward copy */
57 mov.l .memcpy_fwd,r0
58 jmp @r0
59 mov r4,r7 /* store dest for returning */
60
61 .align 2
62.memcpy_fwd:
63 .long ___memcpy_fwd_entry
64
65.backward:
66 add r6,r4 /* r4 = destination end */
67 mov #11,r0
68 cmp/hs r0,r6 /* at least 11 bytes to copy? (ensures 2 aligned longs) */
69 add #-8,r5 /* adjust for late decrement (max. 2 longs) */
70 add r5,r6 /* r6 = source end - 8 */
71 bf .start_b2r /* no: jump directly to byte loop */
72
73 mov #-4,r3 /* r3 = 0xfffffffc */
74 and r6,r3 /* r3 = last source long bound */
75 cmp/hi r3,r6 /* already aligned? */
76 bf .end_b1r /* yes: skip leading byte loop */
77
78.loop_b1r:
79 mov.b @(7,r6),r0 /* load byte */
80 add #-1,r6 /* decrement source addr */
81 mov.b r0,@-r4 /* store byte */
82 cmp/hi r3,r6 /* runs r6 down to last long bound */
83 bt .loop_b1r
84
85.end_b1r:
86 mov #3,r1
87 and r4,r1 /* r1 = dest alignment offset */
88 mova .jmptab_r,r0
89 mov.b @(r0,r1),r1 /* select appropriate main loop.. */
90 add r0,r1
91 mov r5,r3 /* copy start adress to r3 */
92 jmp @r1 /* ..and jump to it */
93 add #7,r3 /* adjust end addr for main loops doing 2 longs/pass */
94
95 /** main loops, copying 2 longs per pass to profit from fast page mode **/
96
97 /* long aligned destination (fastest) */
98 .align 2
99.loop_do0r:
100 mov.l @r6,r1 /* load first long */
101 add #-8,r6 /* decrement source addr */
102 mov.l @(12,r6),r0 /* load second long */
103 cmp/hi r3,r6 /* runs r6 down to first or second long bound */
104 mov.l r0,@-r4 /* store second long */
105 mov.l r1,@-r4 /* store first long; NOT ALIGNED - no speed loss here! */
106 bt .loop_do0r
107
108 add #-4,r3 /* readjust end address */
109 cmp/hi r3,r6 /* first long left? */
110 bf .start_b2r /* no, jump to trailing byte loop */
111
112 mov.l @(4,r6),r0 /* load first long */
113 add #-4,r6 /* decrement source addr */
114 bra .start_b2r /* jump to trailing byte loop */
115 mov.l r0,@-r4 /* store first long */
116
117 /* word aligned destination (long + 2) */
118 .align 2
119.loop_do2r:
120 mov.l @r6,r1 /* load first long */
121 add #-8,r6 /* decrement source addr */
122 mov.l @(12,r6),r0 /* load second long */
123 cmp/hi r3,r6 /* runs r6 down to first or second long bound */
124 mov.w r0,@-r4 /* store low word of second long */
125 xtrct r1,r0 /* extract low word of first long & high word of second long */
126 mov.l r0,@-r4 /* and store as long */
127 shlr16 r1 /* get high word of first long */
128 mov.w r1,@-r4 /* and store it */
129 bt .loop_do2r
130
131 add #-4,r3 /* readjust end address */
132 cmp/hi r3,r6 /* first long left? */
133 bf .start_b2r /* no, jump to trailing byte loop */
134
135 mov.l @(4,r6),r0 /* load first long & decrement source addr */
136 add #-4,r6 /* decrement source addr */
137 mov.w r0,@-r4 /* store low word */
138 shlr16 r0 /* get high word */
139 bra .start_b2r /* jump to trailing byte loop */
140 mov.w r0,@-r4 /* and store it */
141
142 /* jumptable for loop selector */
143 .align 2
144.jmptab_r:
145 .byte .loop_do0r - .jmptab_r /* placed in the middle because the SH1 */
146 .byte .loop_do1r - .jmptab_r /* loads bytes sign-extended. Otherwise */
147 .byte .loop_do2r - .jmptab_r /* the last loop would be out of reach */
148 .byte .loop_do3r - .jmptab_r /* of the offset range. */
149
150 /* byte aligned destination (long + 1) */
151 .align 2
152.loop_do1r:
153 mov.l @r6,r1 /* load first long */
154 add #-8,r6 /* decrement source addr */
155 mov.l @(12,r6),r0 /* load second long */
156 cmp/hi r3,r6 /* runs r6 down to first or second long bound */
157 mov.b r0,@-r4 /* store low byte of second long */
158 shlr8 r0 /* get upper 3 bytes */
159 mov r1,r2 /* copy first long */
160 shll16 r2 /* move low byte of first long all the way up, .. */
161 shll8 r2
162 or r2,r0 /* ..combine with the 3 bytes of second long.. */
163 mov.l r0,@-r4 /* ..and store as long */
164 shlr8 r1 /* get middle 2 bytes */
165 mov.w r1,@-r4 /* store as word */
166 shlr16 r1 /* get upper byte */
167 mov.b r1,@-r4 /* and store */
168 bt .loop_do1r
169
170 add #-4,r3 /* readjust end address */
171.last_do13r:
172 cmp/hi r3,r6 /* first long left? */
173 bf .start_b2r /* no, jump to trailing byte loop */
174
175 nop /* alignment */
176 mov.l @(4,r6),r0 /* load first long */
177 add #-4,r6 /* decrement source addr */
178 mov.b r0,@-r4 /* store low byte */
179 shlr8 r0 /* get middle 2 bytes */
180 mov.w r0,@-r4 /* store as word */
181 shlr16 r0 /* get upper byte */
182 bra .start_b2r /* jump to trailing byte loop */
183 mov.b r0,@-r4 /* and store */
184
185 /* byte aligned destination (long + 3) */
186 .align 2
187.loop_do3r:
188 mov.l @r6,r1 /* load first long */
189 add #-8,r6 /* decrement source addr */
190 mov.l @(12,r6),r0 /* load second long */
191 mov r1,r2 /* copy first long */
192 mov.b r0,@-r4 /* store low byte of second long */
193 shlr8 r0 /* get middle 2 bytes */
194 mov.w r0,@-r4 /* store as word */
195 shlr16 r0 /* get upper byte */
196 shll8 r2 /* move lower 3 bytes of first long one up.. */
197 or r2,r0 /* ..combine with the 1 byte of second long.. */
198 mov.l r0,@-r4 /* ..and store as long */
199 shlr16 r1 /* get upper byte of first long */
200 shlr8 r1
201 cmp/hi r3,r6 /* runs r6 down to first or second long bound */
202 mov.b r1,@-r4 /* ..and store */
203 bt .loop_do3r
204
205 bra .last_do13r /* handle first longword: reuse routine for (long + 1) */
206 add #-4,r3 /* readjust end address */
207
208 /* trailing byte loop: copies 0..3 bytes (or all for < 11 in total) */
209 .align 2
210.loop_b2r:
211 mov.b @(7,r6),r0 /* load byte */
212 add #-1,r6 /* decrement source addr */
213 mov.b r0,@-r4 /* store byte */
214.start_b2r:
215 cmp/hi r5,r6 /* runs r6 down to start address */
216 bt .loop_b2r
217
218 rts
219 mov r4,r0 /* return dest start address */
220.end:
221 .size _memmove,.end-_memmove
222#elif defined(CPU_COLDFIRE)
223#define FULLSPEED /* use burst writing for word aligned destinations */ 23#define FULLSPEED /* use burst writing for word aligned destinations */
224 .align 2 24 .align 2
225 .global memmove 25 .global memmove
@@ -866,4 +666,3 @@ memmove:
866 666
867.end: 667.end:
868 .size memmove,.end-memmove 668 .size memmove,.end-memmove
869#endif
diff --git a/firmware/common/memset_a.S b/firmware/target/coldfire/memset-coldfire.S
index 6dbdab9595..7c9fe88463 100644..100755
--- a/firmware/common/memset_a.S
+++ b/firmware/target/coldfire/memset-coldfire.S
@@ -18,99 +18,9 @@
18 ****************************************************************************/ 18 ****************************************************************************/
19#include "config.h" 19#include "config.h"
20 20
21#ifdef CPU_ARM
22 .section .icode,"ax",%progbits
23#else
24 .section .icode,"ax",@progbits 21 .section .icode,"ax",@progbits
25#endif
26 22
27 .align 2 23 .align 2
28#if CONFIG_CPU == SH7034
29 .global _memset
30 .type _memset,@function
31
32/* Fills a memory region with specified byte value
33 * This version is optimized for speed
34 *
35 * arguments:
36 * r4 - start address
37 * r5 - data
38 * r6 - length
39 *
40 * return value:
41 * r0 - start address (like ANSI version)
42 *
43 * register usage:
44 * r0 - temporary
45 * r1 - start address +11 for main loop
46 * r4 - start address
47 * r5 - data (spread to all 4 bytes when using long stores)
48 * r6 - current address (runs down from end to start)
49 *
50 * The instruction order below is devised in a way to utilize the pipelining
51 * of the SH1 to the max. The routine fills memory from end to start in
52 * order to utilize the auto-decrementing store instructions.
53 */
54
55_memset:
56 neg r4,r0
57 and #3,r0 /* r0 = (4 - align_offset) % 4 */
58 add #4,r0
59 cmp/hs r0,r6 /* at least one aligned longword to fill? */
60 add r4,r6 /* r6 = end_address */
61 bf .no_longs /* no, jump directly to byte loop */
62
63 extu.b r5,r5 /* start: spread data to all 4 bytes */
64 swap.b r5,r0
65 or r0,r5 /* data now in 2 lower bytes of r5 */
66 swap.w r5,r0
67 or r0,r5 /* data now in all 4 bytes of r5 */
68
69 mov r6,r0
70 tst #3,r0 /* r0 already long aligned? */
71 bt .end_b1 /* yes: skip loop */
72
73 /* leading byte loop: sets 0..3 bytes */
74.loop_b1:
75 mov.b r5,@-r0 /* store byte */
76 tst #3,r0 /* r0 long aligned? */
77 bf .loop_b1 /* runs r0 down until long aligned */
78
79 mov r0,r6 /* r6 = last long bound */
80 nop /* keep alignment */
81
82.end_b1:
83 mov r4,r1 /* r1 = start_address... */
84 add #11,r1 /* ... + 11, combined for rounding and offset */
85 xor r1,r0
86 tst #4,r0 /* bit 2 tells whether an even or odd number of */
87 bf .loop_odd /* longwords to set */
88
89 /* main loop: set 2 longs per pass */
90.loop_2l:
91 mov.l r5,@-r6 /* store first long */
92.loop_odd:
93 cmp/hi r1,r6 /* runs r6 down to first long bound */
94 mov.l r5,@-r6 /* store second long */
95 bt .loop_2l
96
97.no_longs:
98 cmp/hi r4,r6 /* any bytes left? */
99 bf .end_b2 /* no: skip loop */
100
101 /* trailing byte loop */
102.loop_b2:
103 mov.b r5,@-r6 /* store byte */
104 cmp/hi r4,r6 /* runs r6 down to the start address */
105 bt .loop_b2
106
107.end_b2:
108 rts
109 mov r4,r0 /* return start address */
110
111.end:
112 .size _memset,.end-_memset
113#elif defined(CPU_COLDFIRE)
114 .global memset 24 .global memset
115 .type memset,@function 25 .type memset,@function
116 26
@@ -238,80 +148,3 @@ memset:
238 148
239.end: 149.end:
240 .size memset,.end-memset 150 .size memset,.end-memset
241
242#elif defined(CPU_ARM)
243
244/* The following code is based on code found in Linux kernel version 2.6.15.3
245 * linux/arch/arm/lib/memset.S
246 *
247 * Copyright (C) 1995-2000 Russell King
248 */
249
250/* This code will align a pointer for memset, if needed */
2511: cmp r2, #4 @ 1 do we have enough
252 blt 5f @ 1 bytes to align with?
253 cmp r3, #2 @ 1
254 strgtb r1, [r0, #-1]! @ 1
255 strgeb r1, [r0, #-1]! @ 1
256 strb r1, [r0, #-1]! @ 1
257 sub r2, r2, r3 @ 1 r2 = r2 - r3
258 b 2f
259
260 .global memset
261 .type memset,%function
262memset:
263 add r0, r0, r2 @ we'll write backwards in memory
264 ands r3, r0, #3 @ 1 unaligned?
265 bne 1b @ 1
2662:
267/*
268 * we know that the pointer in r0 is aligned to a word boundary.
269 */
270 orr r1, r1, r1, lsl #8
271 orr r1, r1, r1, lsl #16
272 mov r3, r1
273 cmp r2, #16
274 blt 5f
275/*
276 * We need an extra register for this loop - save the return address and
277 * use the LR
278 */
279 str lr, [sp, #-4]!
280 mov ip, r1
281 mov lr, r1
282
2833: subs r2, r2, #64
284 stmgedb r0!, {r1, r3, ip, lr} @ 64 bytes at a time.
285 stmgedb r0!, {r1, r3, ip, lr}
286 stmgedb r0!, {r1, r3, ip, lr}
287 stmgedb r0!, {r1, r3, ip, lr}
288 bgt 3b
289 ldmeqfd sp!, {pc} @ Now <64 bytes to go.
290/*
291 * No need to correct the count; we're only testing bits from now on
292 */
293 tst r2, #32
294 stmnedb r0!, {r1, r3, ip, lr}
295 stmnedb r0!, {r1, r3, ip, lr}
296 tst r2, #16
297 stmnedb r0!, {r1, r3, ip, lr}
298 ldr lr, [sp], #4
299
3005: tst r2, #8
301 stmnedb r0!, {r1, r3}
302 tst r2, #4
303 strne r1, [r0, #-4]!
304/*
305 * When we get here, we've got less than 4 bytes to zero. We
306 * may have an unaligned pointer as well.
307 */
3086: tst r2, #2
309 strneb r1, [r0, #-1]!
310 strneb r1, [r0, #-1]!
311 tst r2, #1
312 strneb r1, [r0, #-1]!
313 mov pc, lr
314end:
315 .size memset,.end-memset
316#endif
317
diff --git a/firmware/common/memset16_a.S b/firmware/target/coldfire/memset16-coldfire.S
index 9ab1bdcb5b..d9f72f683f 100755
--- a/firmware/common/memset16_a.S
+++ b/firmware/target/coldfire/memset16-coldfire.S
@@ -20,7 +20,6 @@
20 20
21 .section .icode,"ax",@progbits 21 .section .icode,"ax",@progbits
22 22
23#ifdef CPU_COLDFIRE
24 .global memset16 23 .global memset16
25 .type memset16,@function 24 .type memset16,@function
26 25
@@ -143,4 +142,3 @@ memset16:
143 142
144.end: 143.end:
145 .size memset16,.end-memset16 144 .size memset16,.end-memset16
146#endif
diff --git a/firmware/target/sh/memcpy-sh.S b/firmware/target/sh/memcpy-sh.S
new file mode 100755
index 0000000000..0b5e086be9
--- /dev/null
+++ b/firmware/target/sh/memcpy-sh.S
@@ -0,0 +1,217 @@
1/***************************************************************************
2 * __________ __ ___.
3 * Open \______ \ ____ ____ | | _\_ |__ _______ ___
4 * Source | _// _ \_/ ___\| |/ /| __ \ / _ \ \/ /
5 * Jukebox | | ( <_> ) \___| < | \_\ ( <_> > < <
6 * Firmware |____|_ /\____/ \___ >__|_ \|___ /\____/__/\_ \
7 * \/ \/ \/ \/ \/
8 * $Id$
9 *
10 * Copyright (C) 2004-2005 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#include "config.h"
20
21 .section .icode,"ax",@progbits
22
23 .align 2
24 .global _memcpy
25 .global ___memcpy_fwd_entry
26 .type _memcpy,@function
27
28/* Copies <length> bytes of data in memory from <source> to <dest>
29 * This version is optimized for speed
30 *
31 * arguments:
32 * r4 - destination address
33 * r5 - source address
34 * r6 - length
35 *
36 * return value:
37 * r0 - destination address (like ANSI version)
38 *
39 * register usage:
40 * r0 - data / scratch
41 * r1 - 2nd data / scratch
42 * r2 - scratch
43 * r3 - first long bound / adjusted end address (only if >= 11 bytes)
44 * r4 - current dest address
45 * r5 - current source address
46 * r6 - source end address
47 * r7 - stored dest start address
48 *
49 * The instruction order is devised in a way to utilize the pipelining
50 * of the SH1 to the max. The routine also tries to utilize fast page mode.
51 */
52
53_memcpy:
54 mov r4,r7 /* store dest for returning */
55___memcpy_fwd_entry:
56 add #-8,r4 /* offset for early increment (max. 2 longs) */
57 mov #11,r0
58 cmp/hs r0,r6 /* at least 11 bytes to copy? (ensures 2 aligned longs) */
59 add r5,r6 /* r6 = source_end */
60 bf .start_b2 /* no: jump directly to byte loop */
61
62 mov #3,r0
63 neg r5,r3
64 and r0,r3 /* r3 = (4 - align_offset) % 4 */
65 tst r3,r3 /* already aligned? */
66 bt .end_b1 /* yes: skip leading byte loop */
67
68 add r5,r3 /* r3 = first source long bound */
69
70 /* leading byte loop: copies 0..3 bytes */
71.loop_b1:
72 mov.b @r5+,r0 /* load byte & increment source addr */
73 add #1,r4 /* increment dest addr */
74 mov.b r0,@(7,r4) /* store byte */
75 cmp/hi r5,r3 /* runs r5 up to first long bound */
76 bt .loop_b1
77 /* now r5 is always at a long boundary */
78 /* -> memory reading is done in longs for all dest alignments */
79
80 /* selector for main copy loop */
81.end_b1:
82 mov #3,r1
83 and r4,r1 /* r1 = dest alignment offset */
84 mova .jmptab,r0
85 mov.b @(r0,r1),r1 /* select appropriate main loop */
86 add r0,r1
87 mov r6,r3 /* move end address to r3 */
88 jmp @r1 /* and jump to it */
89 add #-7,r3 /* adjust end addr for main loops doing 2 longs/pass */
90
91 /** main loops, copying 2 longs per pass to profit from fast page mode **/
92
93 /* long aligned destination (fastest) */
94 .align 2
95.loop_do0:
96 mov.l @r5+,r1 /* load first long & increment source addr */
97 add #16,r4 /* increment dest addr & account for decrementing stores */
98 mov.l @r5+,r0 /* load second long & increment source addr */
99 cmp/hi r5,r3 /* runs r5 up to last or second last long bound */
100 mov.l r0,@-r4 /* store second long */
101 mov.l r1,@-r4 /* store first long; NOT ALIGNED - no speed loss here! */
102 bt .loop_do0
103
104 add #4,r3 /* readjust end address */
105 cmp/hi r5,r3 /* one long left? */
106 bf .start_b2 /* no, jump to trailing byte loop */
107
108 mov.l @r5+,r0 /* load last long & increment source addr */
109 add #4,r4 /* increment dest addr */
110 bra .start_b2 /* jump to trailing byte loop */
111 mov.l r0,@(4,r4) /* store last long */
112
113 /* word aligned destination (long + 2) */
114 .align 2
115.loop_do2:
116 mov.l @r5+,r1 /* load first long & increment source addr */
117 add #16,r4 /* increment dest addr */
118 mov.l @r5+,r0 /* load second long & increment source addr */
119 cmp/hi r5,r3 /* runs r5 up to last or second last long bound */
120 mov.w r0,@-r4 /* store low word of second long */
121 xtrct r1,r0 /* extract low word of first long & high word of second long */
122 mov.l r0,@-r4 /* and store as long */
123 swap.w r1,r0 /* get high word of first long */
124 mov.w r0,@-r4 /* and store it */
125 bt .loop_do2
126
127 add #4,r3 /* readjust end address */
128 cmp/hi r5,r3 /* one long left? */
129 bf .start_b2 /* no, jump to trailing byte loop */
130
131 mov.l @r5+,r0 /* load last long & increment source addr */
132 add #4,r4 /* increment dest addr */
133 mov.w r0,@(6,r4) /* store low word */
134 shlr16 r0 /* get high word */
135 bra .start_b2 /* jump to trailing byte loop */
136 mov.w r0,@(4,r4) /* and store it */
137
138 /* jumptable for loop selector */
139 .align 2
140.jmptab:
141 .byte .loop_do0 - .jmptab /* placed in the middle because the SH1 */
142 .byte .loop_do1 - .jmptab /* loads bytes sign-extended. Otherwise */
143 .byte .loop_do2 - .jmptab /* the last loop would be out of reach */
144 .byte .loop_do3 - .jmptab /* of the offset range. */
145
146 /* byte aligned destination (long + 1) */
147 .align 2
148.loop_do1:
149 mov.l @r5+,r1 /* load first long & increment source addr */
150 add #16,r4 /* increment dest addr */
151 mov.l @r5+,r0 /* load second long & increment source addr */
152 cmp/hi r5,r3 /* runs r5 up to last or second last long bound */
153 mov.b r0,@-r4 /* store low byte of second long */
154 shlr8 r0 /* get upper 3 bytes */
155 mov r1,r2 /* copy first long */
156 shll16 r2 /* move low byte of first long all the way up, .. */
157 shll8 r2
158 or r2,r0 /* ..combine with the 3 bytes of second long.. */
159 mov.l r0,@-r4 /* ..and store as long */
160 shlr8 r1 /* get middle 2 bytes */
161 mov.w r1,@-r4 /* store as word */
162 shlr16 r1 /* get upper byte */
163 mov.b r1,@-r4 /* and store */
164 bt .loop_do1
165
166 add #4,r3 /* readjust end address */
167.last_do13:
168 cmp/hi r5,r3 /* one long left? */
169 bf .start_b2 /* no, jump to trailing byte loop */
170
171 mov.l @r5+,r0 /* load last long & increment source addr */
172 add #12,r4 /* increment dest addr */
173 mov.b r0,@-r4 /* store low byte */
174 shlr8 r0 /* get middle 2 bytes */
175 mov.w r0,@-r4 /* store as word */
176 shlr16 r0 /* get upper byte */
177 mov.b r0,@-r4 /* and store */
178 bra .start_b2 /* jump to trailing byte loop */
179 add #-4,r4 /* readjust destination */
180
181 /* byte aligned destination (long + 3) */
182 .align 2
183.loop_do3:
184 mov.l @r5+,r1 /* load first long & increment source addr */
185 add #16,r4 /* increment dest addr */
186 mov.l @r5+,r0 /* load second long & increment source addr */
187 mov r1,r2 /* copy first long */
188 mov.b r0,@-r4 /* store low byte of second long */
189 shlr8 r0 /* get middle 2 bytes */
190 mov.w r0,@-r4 /* store as word */
191 shlr16 r0 /* get upper byte */
192 shll8 r2 /* move lower 3 bytes of first long one up.. */
193 or r2,r0 /* ..combine with the 1 byte of second long.. */
194 mov.l r0,@-r4 /* ..and store as long */
195 shlr16 r1 /* get upper byte of first long.. */
196 shlr8 r1
197 cmp/hi r5,r3 /* runs r5 up to last or second last long bound */
198 mov.b r1,@-r4 /* ..and store */
199 bt .loop_do3
200
201 bra .last_do13 /* handle last longword: reuse routine for (long + 1) */
202 add #4,r3 /* readjust end address */
203
204 /* trailing byte loop: copies 0..3 bytes (or all for < 11 in total) */
205 .align 2
206.loop_b2:
207 mov.b @r5+,r0 /* load byte & increment source addr */
208 add #1,r4 /* increment dest addr */
209 mov.b r0,@(7,r4) /* store byte */
210.start_b2:
211 cmp/hi r5,r6 /* runs r5 up to end address */
212 bt .loop_b2
213
214 rts
215 mov r7,r0 /* return dest start address */
216.end:
217 .size _memcpy,.end-_memcpy
diff --git a/firmware/target/sh/memmove-sh.S b/firmware/target/sh/memmove-sh.S
new file mode 100755
index 0000000000..9ae9ae5fa2
--- /dev/null
+++ b/firmware/target/sh/memmove-sh.S
@@ -0,0 +1,220 @@
1/***************************************************************************
2 * __________ __ ___.
3 * Open \______ \ ____ ____ | | _\_ |__ _______ ___
4 * Source | _// _ \_/ ___\| |/ /| __ \ / _ \ \/ /
5 * Jukebox | | ( <_> ) \___| < | \_\ ( <_> > < <
6 * Firmware |____|_ /\____/ \___ >__|_ \|___ /\____/__/\_ \
7 * \/ \/ \/ \/ \/
8 * $Id$
9 *
10 * Copyright (C) 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#include "config.h"
20
21 .section .icode,"ax",@progbits
22
23 .align 2
24 .global _memmove
25 .type _memmove,@function
26
27/* Moves <length> bytes of data in memory from <source> to <dest>
28 * Regions may overlap.
29 * This version is optimized for speed, and needs the corresponding memcpy
30 * implementation for the forward copy branch.
31 *
32 * arguments:
33 * r4 - destination address
34 * r5 - source address
35 * r6 - length
36 *
37 * return value:
38 * r0 - destination address (like ANSI version)
39 *
40 * register usage:
41 * r0 - data / scratch
42 * r1 - 2nd data / scratch
43 * r2 - scratch
44 * r3 - last long bound / adjusted start address (only if >= 11 bytes)
45 * r4 - current dest address
46 * r5 - source start address
47 * r6 - current source address
48 *
49 * The instruction order is devised in a way to utilize the pipelining
50 * of the SH1 to the max. The routine also tries to utilize fast page mode.
51 */
52
53_memmove:
54 cmp/hi r4,r5 /* source > destination */
55 bf .backward /* no: backward copy */
56 mov.l .memcpy_fwd,r0
57 jmp @r0
58 mov r4,r7 /* store dest for returning */
59
60 .align 2
61.memcpy_fwd:
62 .long ___memcpy_fwd_entry
63
64.backward:
65 add r6,r4 /* r4 = destination end */
66 mov #11,r0
67 cmp/hs r0,r6 /* at least 11 bytes to copy? (ensures 2 aligned longs) */
68 add #-8,r5 /* adjust for late decrement (max. 2 longs) */
69 add r5,r6 /* r6 = source end - 8 */
70 bf .start_b2r /* no: jump directly to byte loop */
71
72 mov #-4,r3 /* r3 = 0xfffffffc */
73 and r6,r3 /* r3 = last source long bound */
74 cmp/hi r3,r6 /* already aligned? */
75 bf .end_b1r /* yes: skip leading byte loop */
76
77.loop_b1r:
78 mov.b @(7,r6),r0 /* load byte */
79 add #-1,r6 /* decrement source addr */
80 mov.b r0,@-r4 /* store byte */
81 cmp/hi r3,r6 /* runs r6 down to last long bound */
82 bt .loop_b1r
83
84.end_b1r:
85 mov #3,r1
86 and r4,r1 /* r1 = dest alignment offset */
87 mova .jmptab_r,r0
88 mov.b @(r0,r1),r1 /* select appropriate main loop.. */
89 add r0,r1
90 mov r5,r3 /* copy start adress to r3 */
91 jmp @r1 /* ..and jump to it */
92 add #7,r3 /* adjust end addr for main loops doing 2 longs/pass */
93
94 /** main loops, copying 2 longs per pass to profit from fast page mode **/
95
96 /* long aligned destination (fastest) */
97 .align 2
98.loop_do0r:
99 mov.l @r6,r1 /* load first long */
100 add #-8,r6 /* decrement source addr */
101 mov.l @(12,r6),r0 /* load second long */
102 cmp/hi r3,r6 /* runs r6 down to first or second long bound */
103 mov.l r0,@-r4 /* store second long */
104 mov.l r1,@-r4 /* store first long; NOT ALIGNED - no speed loss here! */
105 bt .loop_do0r
106
107 add #-4,r3 /* readjust end address */
108 cmp/hi r3,r6 /* first long left? */
109 bf .start_b2r /* no, jump to trailing byte loop */
110
111 mov.l @(4,r6),r0 /* load first long */
112 add #-4,r6 /* decrement source addr */
113 bra .start_b2r /* jump to trailing byte loop */
114 mov.l r0,@-r4 /* store first long */
115
116 /* word aligned destination (long + 2) */
117 .align 2
118.loop_do2r:
119 mov.l @r6,r1 /* load first long */
120 add #-8,r6 /* decrement source addr */
121 mov.l @(12,r6),r0 /* load second long */
122 cmp/hi r3,r6 /* runs r6 down to first or second long bound */
123 mov.w r0,@-r4 /* store low word of second long */
124 xtrct r1,r0 /* extract low word of first long & high word of second long */
125 mov.l r0,@-r4 /* and store as long */
126 shlr16 r1 /* get high word of first long */
127 mov.w r1,@-r4 /* and store it */
128 bt .loop_do2r
129
130 add #-4,r3 /* readjust end address */
131 cmp/hi r3,r6 /* first long left? */
132 bf .start_b2r /* no, jump to trailing byte loop */
133
134 mov.l @(4,r6),r0 /* load first long & decrement source addr */
135 add #-4,r6 /* decrement source addr */
136 mov.w r0,@-r4 /* store low word */
137 shlr16 r0 /* get high word */
138 bra .start_b2r /* jump to trailing byte loop */
139 mov.w r0,@-r4 /* and store it */
140
141 /* jumptable for loop selector */
142 .align 2
143.jmptab_r:
144 .byte .loop_do0r - .jmptab_r /* placed in the middle because the SH1 */
145 .byte .loop_do1r - .jmptab_r /* loads bytes sign-extended. Otherwise */
146 .byte .loop_do2r - .jmptab_r /* the last loop would be out of reach */
147 .byte .loop_do3r - .jmptab_r /* of the offset range. */
148
149 /* byte aligned destination (long + 1) */
150 .align 2
151.loop_do1r:
152 mov.l @r6,r1 /* load first long */
153 add #-8,r6 /* decrement source addr */
154 mov.l @(12,r6),r0 /* load second long */
155 cmp/hi r3,r6 /* runs r6 down to first or second long bound */
156 mov.b r0,@-r4 /* store low byte of second long */
157 shlr8 r0 /* get upper 3 bytes */
158 mov r1,r2 /* copy first long */
159 shll16 r2 /* move low byte of first long all the way up, .. */
160 shll8 r2
161 or r2,r0 /* ..combine with the 3 bytes of second long.. */
162 mov.l r0,@-r4 /* ..and store as long */
163 shlr8 r1 /* get middle 2 bytes */
164 mov.w r1,@-r4 /* store as word */
165 shlr16 r1 /* get upper byte */
166 mov.b r1,@-r4 /* and store */
167 bt .loop_do1r
168
169 add #-4,r3 /* readjust end address */
170.last_do13r:
171 cmp/hi r3,r6 /* first long left? */
172 bf .start_b2r /* no, jump to trailing byte loop */
173
174 nop /* alignment */
175 mov.l @(4,r6),r0 /* load first long */
176 add #-4,r6 /* decrement source addr */
177 mov.b r0,@-r4 /* store low byte */
178 shlr8 r0 /* get middle 2 bytes */
179 mov.w r0,@-r4 /* store as word */
180 shlr16 r0 /* get upper byte */
181 bra .start_b2r /* jump to trailing byte loop */
182 mov.b r0,@-r4 /* and store */
183
184 /* byte aligned destination (long + 3) */
185 .align 2
186.loop_do3r:
187 mov.l @r6,r1 /* load first long */
188 add #-8,r6 /* decrement source addr */
189 mov.l @(12,r6),r0 /* load second long */
190 mov r1,r2 /* copy first long */
191 mov.b r0,@-r4 /* store low byte of second long */
192 shlr8 r0 /* get middle 2 bytes */
193 mov.w r0,@-r4 /* store as word */
194 shlr16 r0 /* get upper byte */
195 shll8 r2 /* move lower 3 bytes of first long one up.. */
196 or r2,r0 /* ..combine with the 1 byte of second long.. */
197 mov.l r0,@-r4 /* ..and store as long */
198 shlr16 r1 /* get upper byte of first long */
199 shlr8 r1
200 cmp/hi r3,r6 /* runs r6 down to first or second long bound */
201 mov.b r1,@-r4 /* ..and store */
202 bt .loop_do3r
203
204 bra .last_do13r /* handle first longword: reuse routine for (long + 1) */
205 add #-4,r3 /* readjust end address */
206
207 /* trailing byte loop: copies 0..3 bytes (or all for < 11 in total) */
208 .align 2
209.loop_b2r:
210 mov.b @(7,r6),r0 /* load byte */
211 add #-1,r6 /* decrement source addr */
212 mov.b r0,@-r4 /* store byte */
213.start_b2r:
214 cmp/hi r5,r6 /* runs r6 down to start address */
215 bt .loop_b2r
216
217 rts
218 mov r4,r0 /* return dest start address */
219.end:
220 .size _memmove,.end-_memmove
diff --git a/firmware/target/sh/memset-sh.S b/firmware/target/sh/memset-sh.S
new file mode 100755
index 0000000000..9b96b93f27
--- /dev/null
+++ b/firmware/target/sh/memset-sh.S
@@ -0,0 +1,107 @@
1/***************************************************************************
2 * __________ __ ___.
3 * Open \______ \ ____ ____ | | _\_ |__ _______ ___
4 * Source | _// _ \_/ ___\| |/ /| __ \ / _ \ \/ /
5 * Jukebox | | ( <_> ) \___| < | \_\ ( <_> > < <
6 * Firmware |____|_ /\____/ \___ >__|_ \|___ /\____/__/\_ \
7 * \/ \/ \/ \/ \/
8 * $Id$
9 *
10 * Copyright (C) 2004 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#include "config.h"
20
21 .section .icode,"ax",@progbits
22
23 .align 2
24 .global _memset
25 .type _memset,@function
26
27/* Fills a memory region with specified byte value
28 * This version is optimized for speed
29 *
30 * arguments:
31 * r4 - start address
32 * r5 - data
33 * r6 - length
34 *
35 * return value:
36 * r0 - start address (like ANSI version)
37 *
38 * register usage:
39 * r0 - temporary
40 * r1 - start address +11 for main loop
41 * r4 - start address
42 * r5 - data (spread to all 4 bytes when using long stores)
43 * r6 - current address (runs down from end to start)
44 *
45 * The instruction order below is devised in a way to utilize the pipelining
46 * of the SH1 to the max. The routine fills memory from end to start in
47 * order to utilize the auto-decrementing store instructions.
48 */
49
50_memset:
51 neg r4,r0
52 and #3,r0 /* r0 = (4 - align_offset) % 4 */
53 add #4,r0
54 cmp/hs r0,r6 /* at least one aligned longword to fill? */
55 add r4,r6 /* r6 = end_address */
56 bf .no_longs /* no, jump directly to byte loop */
57
58 extu.b r5,r5 /* start: spread data to all 4 bytes */
59 swap.b r5,r0
60 or r0,r5 /* data now in 2 lower bytes of r5 */
61 swap.w r5,r0
62 or r0,r5 /* data now in all 4 bytes of r5 */
63
64 mov r6,r0
65 tst #3,r0 /* r0 already long aligned? */
66 bt .end_b1 /* yes: skip loop */
67
68 /* leading byte loop: sets 0..3 bytes */
69.loop_b1:
70 mov.b r5,@-r0 /* store byte */
71 tst #3,r0 /* r0 long aligned? */
72 bf .loop_b1 /* runs r0 down until long aligned */
73
74 mov r0,r6 /* r6 = last long bound */
75 nop /* keep alignment */
76
77.end_b1:
78 mov r4,r1 /* r1 = start_address... */
79 add #11,r1 /* ... + 11, combined for rounding and offset */
80 xor r1,r0
81 tst #4,r0 /* bit 2 tells whether an even or odd number of */
82 bf .loop_odd /* longwords to set */
83
84 /* main loop: set 2 longs per pass */
85.loop_2l:
86 mov.l r5,@-r6 /* store first long */
87.loop_odd:
88 cmp/hi r1,r6 /* runs r6 down to first long bound */
89 mov.l r5,@-r6 /* store second long */
90 bt .loop_2l
91
92.no_longs:
93 cmp/hi r4,r6 /* any bytes left? */
94 bf .end_b2 /* no: skip loop */
95
96 /* trailing byte loop */
97.loop_b2:
98 mov.b r5,@-r6 /* store byte */
99 cmp/hi r4,r6 /* runs r6 down to the start address */
100 bt .loop_b2
101
102.end_b2:
103 rts
104 mov r4,r0 /* return start address */
105
106.end:
107 .size _memset,.end-_memset
diff --git a/firmware/common/strlen_a.S b/firmware/target/sh/strlen-sh.S
index 34837605ac..34837605ac 100755
--- a/firmware/common/strlen_a.S
+++ b/firmware/target/sh/strlen-sh.S
generated by cgit v1.2.3 (git 2.39.1) at 2024-07-07 17:59:46 +0000