aboutsummaryrefslogtreecommitdiff
path: root/arch/ia64/lib/ip_fast_csum.S
blob: 620d9dc5220f377c9cd2da899a81795054a27ad0 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
/*
 * Optmized version of the ip_fast_csum() function
 * Used for calculating IP header checksum
 *
 * Return: 16bit checksum, complemented
 *
 * Inputs:
 *      in0: address of buffer to checksum (char *)
 *      in1: length of the buffer (int)
 *
 * Copyright (C) 2002, 2006 Intel Corp.
 * Copyright (C) 2002, 2006 Ken Chen <kenneth.w.chen@intel.com>
 */

#include <asm/asmmacro.h>

/*
 * Since we know that most likely this function is called with buf aligned
 * on 4-byte boundary and 20 bytes in length, we can execution rather quickly
 * versus calling generic version of do_csum, which has lots of overhead in
 * handling various alignments and sizes.  However, due to lack of constrains
 * put on the function input argument, cases with alignment not on 4-byte or
 * size not equal to 20 bytes will be handled by the generic do_csum function.
 */

#define in0	r32
#define in1	r33
#define in2	r34
#define in3	r35
#define in4	r36
#define ret0	r8

GLOBAL_ENTRY(ip_fast_csum)
	.prologue
	.body
	cmp.ne	p6,p7=5,in1	// size other than 20 byte?
	and	r14=3,in0	// is it aligned on 4-byte?
	add	r15=4,in0	// second source pointer
	;;
	cmp.ne.or.andcm p6,p7=r14,r0
	;;
(p7)	ld4	r20=[in0],8
(p7)	ld4	r21=[r15],8
(p6)	br.spnt	.generic
	;;
	ld4	r22=[in0],8
	ld4	r23=[r15],8
	;;
	ld4	r24=[in0]
	add	r20=r20,r21
	add	r22=r22,r23
	;;
	add	r20=r20,r22
	;;
	add	r20=r20,r24
	;;
	shr.u	ret0=r20,16	// now need to add the carry
	zxt2	r20=r20
	;;
	add	r20=ret0,r20
	;;
	shr.u	ret0=r20,16	// add carry again
	zxt2	r20=r20
	;;
	add	r20=ret0,r20
	;;
	shr.u	ret0=r20,16
	zxt2	r20=r20
	;;
	add	r20=ret0,r20
	mov	r9=0xffff
	;;
	andcm	ret0=r9,r20
	.restore sp		// reset frame state
	br.ret.sptk.many b0
	;;

.generic:
	.prologue
	.save ar.pfs, r35
	alloc	r35=ar.pfs,2,2,2,0
	.save rp, r34
	mov	r34=b0
	.body
	dep.z	out1=in1,2,30
	mov	out0=in0
	;;
	br.call.sptk.many b0=do_csum
	;;
	andcm	ret0=-1,ret0
	mov	ar.pfs=r35
	mov	b0=r34
	br.ret.sptk.many b0
END(ip_fast_csum)

GLOBAL_ENTRY(csum_ipv6_magic)
	ld4	r20=[in0],4
	ld4	r21=[in1],4
	zxt4	in2=in2
	;;
	ld4	r22=[in0],4
	ld4	r23=[in1],4
	dep	r15=in3,in2,32,16
	;;
	ld4	r24=[in0],4
	ld4	r25=[in1],4
	mux1	r15=r15,@rev
	add	r16=r20,r21
	add	r17=r22,r23
	zxt4	in4=in4
	;;
	ld4	r26=[in0],4
	ld4	r27=[in1],4
	shr.u	r15=r15,16
	add	r18=r24,r25
	add	r8=r16,r17
	;;
	add	r19=r26,r27
	add	r8=r8,r18
	;;
	add	r8=r8,r19
	add	r15=r15,in4
	;;
	add	r8=r8,r15
	;;
	shr.u	r10=r8,32	// now fold sum into short
	zxt4	r11=r8
	;;
	add	r8=r10,r11
	;;
	shr.u	r10=r8,16	// yeah, keep it rolling
	zxt2	r11=r8
	;;
	add	r8=r10,r11
	;;
	shr.u	r10=r8,16	// three times lucky
	zxt2	r11=r8
	;;
	add	r8=r10,r11
	mov	r9=0xffff
	;;
	andcm	r8=r9,r8
	br.ret.sptk.many b0
END(csum_ipv6_magic)