diff options
Diffstat (limited to 'arch/alpha/lib/ev67-strncat.S')
-rw-r--r-- | arch/alpha/lib/ev67-strncat.S | 94 |
1 files changed, 94 insertions, 0 deletions
diff --git a/arch/alpha/lib/ev67-strncat.S b/arch/alpha/lib/ev67-strncat.S new file mode 100644 index 000000000000..4ae716cd2bfb --- /dev/null +++ b/arch/alpha/lib/ev67-strncat.S | |||
@@ -0,0 +1,94 @@ | |||
1 | /* | ||
2 | * arch/alpha/lib/ev67-strncat.S | ||
3 | * 21264 version contributed by Rick Gorton <rick.gorton@api-networks.com> | ||
4 | * | ||
5 | * Append no more than COUNT characters from the null-terminated string SRC | ||
6 | * to the null-terminated string DST. Always null-terminate the new DST. | ||
7 | * | ||
8 | * This differs slightly from the semantics in libc in that we never write | ||
9 | * past count, whereas libc may write to count+1. This follows the generic | ||
10 | * implementation in lib/string.c and is, IMHO, more sensible. | ||
11 | * | ||
12 | * Much of the information about 21264 scheduling/coding comes from: | ||
13 | * Compiler Writer's Guide for the Alpha 21264 | ||
14 | * abbreviated as 'CWG' in other comments here | ||
15 | * ftp.digital.com/pub/Digital/info/semiconductor/literature/dsc-library.html | ||
16 | * Scheduling notation: | ||
17 | * E - either cluster | ||
18 | * U - upper subcluster; U0 - subcluster U0; U1 - subcluster U1 | ||
19 | * L - lower subcluster; L0 - subcluster L0; L1 - subcluster L1 | ||
20 | * Try not to change the actual algorithm if possible for consistency. | ||
21 | */ | ||
22 | |||
23 | |||
24 | .text | ||
25 | |||
26 | .align 4 | ||
27 | .globl strncat | ||
28 | .ent strncat | ||
29 | strncat: | ||
30 | .frame $30, 0, $26 | ||
31 | .prologue 0 | ||
32 | |||
33 | mov $16, $0 # set up return value | ||
34 | beq $18, $zerocount # U : | ||
35 | /* Find the end of the string. */ | ||
36 | ldq_u $1, 0($16) # L : load first quadword ($16 may be misaligned) | ||
37 | lda $2, -1($31) # E : | ||
38 | |||
39 | insqh $2, $0, $2 # U : | ||
40 | andnot $16, 7, $16 # E : | ||
41 | nop # E : | ||
42 | or $2, $1, $1 # E : | ||
43 | |||
44 | nop # E : | ||
45 | nop # E : | ||
46 | cmpbge $31, $1, $2 # E : bits set iff byte == 0 | ||
47 | bne $2, $found # U : | ||
48 | |||
49 | $loop: ldq $1, 8($16) # L : | ||
50 | addq $16, 8, $16 # E : | ||
51 | cmpbge $31, $1, $2 # E : | ||
52 | beq $2, $loop # U : | ||
53 | |||
54 | $found: cttz $2, $3 # U0 : | ||
55 | addq $16, $3, $16 # E : | ||
56 | nop # E : | ||
57 | bsr $23, __stxncpy # L0 :/* Now do the append. */ | ||
58 | |||
59 | /* Worry about the null termination. */ | ||
60 | |||
61 | zapnot $1, $27, $2 # U : was last byte a null? | ||
62 | cmplt $27, $24, $5 # E : did we fill the buffer completely? | ||
63 | bne $2, 0f # U : | ||
64 | ret # L0 : | ||
65 | |||
66 | 0: or $5, $18, $2 # E : | ||
67 | nop | ||
68 | bne $2, 2f # U : | ||
69 | and $24, 0x80, $3 # E : no zero next byte | ||
70 | |||
71 | nop # E : | ||
72 | bne $3, 1f # U : | ||
73 | /* Here there are bytes left in the current word. Clear one. */ | ||
74 | addq $24, $24, $24 # E : end-of-count bit <<= 1 | ||
75 | nop # E : | ||
76 | |||
77 | 2: zap $1, $24, $1 # U : | ||
78 | nop # E : | ||
79 | stq_u $1, 0($16) # L : | ||
80 | ret # L0 : | ||
81 | |||
82 | 1: /* Here we must clear the first byte of the next DST word */ | ||
83 | stb $31, 8($16) # L : | ||
84 | nop # E : | ||
85 | nop # E : | ||
86 | ret # L0 : | ||
87 | |||
88 | $zerocount: | ||
89 | nop # E : | ||
90 | nop # E : | ||
91 | nop # E : | ||
92 | ret # L0 : | ||
93 | |||
94 | .end strncat | ||