Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input
[linux-2.6] / arch / sh / lib64 / strcpy.S
1 /* Cloned and hacked for uClibc by Paul Mundt, December 2003 */
2 /* Modified by SuperH, Inc. September 2003 */
3 ! Entry: arg0: destination
4 !        arg1: source
5 ! Exit:  result: destination
6 !
7 ! SH5 code Copyright 2002 SuperH Ltd.
8
9 #if __BYTE_ORDER == __LITTLE_ENDIAN
10 #define SHHI shlld
11 #define SHLO shlrd
12 #else
13 #define SHHI shlrd
14 #define SHLO shlld
15 #endif
16
17         .section .text..SHmedia32,"ax"
18         .globl  strcpy
19         .type   strcpy, @function
20         .align 5
21
22 strcpy:
23
24         pta/l shortstring,tr1
25         ldlo.q r3,0,r4
26         ptabs r18,tr4
27         shlli r3,3,r7
28         addi r2, 8, r0
29         mcmpeq.b r4,r63,r6
30         SHHI r6,r7,r6
31         bnei/u r6,0,tr1 // shortstring
32         pta/l no_lddst, tr2
33         ori r3,-8,r23
34         sub r2, r23, r0
35         sub r3, r2, r21
36         addi r21, 8, r20
37         ldx.q r0, r21, r5
38         pta/l loop, tr0
39         ori r2,-8,r22
40         mcmpeq.b r5, r63, r6
41         bgt/u r22, r23, tr2 // no_lddst
42
43         // r22 < r23 :  Need to do a load from the destination.
44         // r22 == r23 : Doesn't actually need to load from destination,
45         //              but still can be handled here.
46         ldlo.q r2, 0, r9
47         movi -1, r8
48         SHLO r8, r7, r8
49         mcmv r4, r8, r9
50         stlo.q r2, 0, r9
51         beqi/l r6, 0, tr0 // loop
52
53         add r5, r63, r4
54         addi r0, 8, r0
55         blink tr1, r63 // shortstring
56 no_lddst:
57         // r22 > r23: note that for r22 == r23 the sthi.q would clobber
58         //            bytes before the destination region.
59         stlo.q r2, 0, r4
60         SHHI r4, r7, r4
61         sthi.q r0, -1, r4
62         beqi/l r6, 0, tr0 // loop
63
64         add r5, r63, r4
65         addi r0, 8, r0
66 shortstring:
67 #if __BYTE_ORDER != __LITTLE_ENDIAN
68         pta/l shortstring2,tr1
69         byterev r4,r4
70 #endif
71 shortstring2:
72         st.b r0,-8,r4
73         andi r4,0xff,r5
74         shlri r4,8,r4
75         addi r0,1,r0
76         bnei/l r5,0,tr1
77         blink tr4,r63 // return
78         
79         .balign 8
80 loop:
81         stlo.q r0, 0, r5
82         ldx.q r0, r20, r4
83         addi r0, 16, r0
84         sthi.q r0, -9, r5
85         mcmpeq.b r4, r63, r6
86         bnei/u r6, 0, tr1 // shortstring
87         ldx.q r0, r21, r5
88         stlo.q r0, -8, r4
89         sthi.q r0, -1, r4
90         mcmpeq.b r5, r63, r6
91         beqi/l r6, 0, tr0 // loop
92
93         add r5, r63, r4
94         addi r0, 8, r0
95         blink tr1, r63 // shortstring
96
97         .size   strcpy,.-strcpy