xref: /linux/lib/raid6/altivec.uc (revision 24bce201d79807b668bf9d9e0aca801c5c0d5f78)
1/* -*- linux-c -*- ------------------------------------------------------- *
2 *
3 *   Copyright 2002-2004 H. Peter Anvin - All Rights Reserved
4 *
5 *   This program is free software; you can redistribute it and/or modify
6 *   it under the terms of the GNU General Public License as published by
7 *   the Free Software Foundation, Inc., 53 Temple Place Ste 330,
8 *   Boston MA 02111-1307, USA; either version 2 of the License, or
9 *   (at your option) any later version; incorporated herein by reference.
10 *
11 * ----------------------------------------------------------------------- */
12
13/*
14 * raid6altivec$#.c
15 *
16 * $#-way unrolled portable integer math RAID-6 instruction set
17 *
18 * This file is postprocessed using unroll.awk
19 *
20 * <benh> hpa: in process,
21 * you can just "steal" the vec unit with enable_kernel_altivec() (but
22 * bracked this with preempt_disable/enable or in a lock)
23 */
24
25#include <linux/raid/pq.h>
26
27#ifdef CONFIG_ALTIVEC
28
29#include <altivec.h>
30#ifdef __KERNEL__
31# include <asm/cputable.h>
32# include <asm/switch_to.h>
33#endif /* __KERNEL__ */
34
35/*
36 * This is the C data type to use.  We use a vector of
37 * signed char so vec_cmpgt() will generate the right
38 * instruction.
39 */
40
41typedef vector signed char unative_t;
42
43#define NBYTES(x) ((vector signed char) {x,x,x,x, x,x,x,x, x,x,x,x, x,x,x,x})
44#define NSIZE	sizeof(unative_t)
45
46/*
47 * The SHLBYTE() operation shifts each byte left by 1, *not*
48 * rolling over into the next byte
49 */
50static inline __attribute_const__ unative_t SHLBYTE(unative_t v)
51{
52	return vec_add(v,v);
53}
54
55/*
56 * The MASK() operation returns 0xFF in any byte for which the high
57 * bit is 1, 0x00 for any byte for which the high bit is 0.
58 */
59static inline __attribute_const__ unative_t MASK(unative_t v)
60{
61	unative_t zv = NBYTES(0);
62
63	/* vec_cmpgt returns a vector bool char; thus the need for the cast */
64	return (unative_t)vec_cmpgt(zv, v);
65}
66
67
68/* This is noinline to make damned sure that gcc doesn't move any of the
69   Altivec code around the enable/disable code */
70static void noinline
71raid6_altivec$#_gen_syndrome_real(int disks, size_t bytes, void **ptrs)
72{
73	u8 **dptr = (u8 **)ptrs;
74	u8 *p, *q;
75	int d, z, z0;
76
77	unative_t wd$$, wq$$, wp$$, w1$$, w2$$;
78	unative_t x1d = NBYTES(0x1d);
79
80	z0 = disks - 3;		/* Highest data disk */
81	p = dptr[z0+1];		/* XOR parity */
82	q = dptr[z0+2];		/* RS syndrome */
83
84	for ( d = 0 ; d < bytes ; d += NSIZE*$# ) {
85		wq$$ = wp$$ = *(unative_t *)&dptr[z0][d+$$*NSIZE];
86		for ( z = z0-1 ; z >= 0 ; z-- ) {
87			wd$$ = *(unative_t *)&dptr[z][d+$$*NSIZE];
88			wp$$ = vec_xor(wp$$, wd$$);
89			w2$$ = MASK(wq$$);
90			w1$$ = SHLBYTE(wq$$);
91			w2$$ = vec_and(w2$$, x1d);
92			w1$$ = vec_xor(w1$$, w2$$);
93			wq$$ = vec_xor(w1$$, wd$$);
94		}
95		*(unative_t *)&p[d+NSIZE*$$] = wp$$;
96		*(unative_t *)&q[d+NSIZE*$$] = wq$$;
97	}
98}
99
100static void raid6_altivec$#_gen_syndrome(int disks, size_t bytes, void **ptrs)
101{
102	preempt_disable();
103	enable_kernel_altivec();
104
105	raid6_altivec$#_gen_syndrome_real(disks, bytes, ptrs);
106
107	disable_kernel_altivec();
108	preempt_enable();
109}
110
111int raid6_have_altivec(void);
112#if $# == 1
113int raid6_have_altivec(void)
114{
115	/* This assumes either all CPUs have Altivec or none does */
116# ifdef __KERNEL__
117	return cpu_has_feature(CPU_FTR_ALTIVEC);
118# else
119	return 1;
120# endif
121}
122#endif
123
124const struct raid6_calls raid6_altivec$# = {
125	raid6_altivec$#_gen_syndrome,
126	NULL,			/* XOR not yet implemented */
127	raid6_have_altivec,
128	"altivecx$#",
129	0
130};
131
132#endif /* CONFIG_ALTIVEC */
133