xref: /linux/lib/raid6/altivec.uc (revision cc04a46f11ea046ed53e2c832ae29e4790f7e35f)
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 * $#-way unrolled portable integer math RAID-6 instruction set
17 *
18 * This file is postprocessed using unroll.awk
19 *
20 * <benh> hpa: in process,
21 * you can just "steal" the vec unit with enable_kernel_altivec() (but
22 * bracked this with preempt_disable/enable or in a lock)
23 */
24
25#include <linux/raid/pq.h>
26
27#include <altivec.h>
28#ifdef __KERNEL__
29# include <asm/cputable.h>
30# include <asm/switch_to.h>
31
32/*
33 * This is the C data type to use.  We use a vector of
34 * signed char so vec_cmpgt() will generate the right
35 * instruction.
36 */
37
38typedef vector signed char unative_t;
39
40#define NBYTES(x) ((vector signed char) {x,x,x,x, x,x,x,x, x,x,x,x, x,x,x,x})
41#define NSIZE	sizeof(unative_t)
42
43/*
44 * The SHLBYTE() operation shifts each byte left by 1, *not*
45 * rolling over into the next byte
46 */
47static inline __attribute_const__ unative_t SHLBYTE(unative_t v)
48{
49	return vec_add(v,v);
50}
51
52/*
53 * The MASK() operation returns 0xFF in any byte for which the high
54 * bit is 1, 0x00 for any byte for which the high bit is 0.
55 */
56static inline __attribute_const__ unative_t MASK(unative_t v)
57{
58	unative_t zv = NBYTES(0);
59
60	/* vec_cmpgt returns a vector bool char; thus the need for the cast */
61	return (unative_t)vec_cmpgt(zv, v);
62}
63
64
65/* This is noinline to make damned sure that gcc doesn't move any of the
66   Altivec code around the enable/disable code */
67static void noinline
68raid6_altivec$#_gen_syndrome_real(int disks, size_t bytes, void **ptrs)
69{
70	u8 **dptr = (u8 **)ptrs;
71	u8 *p, *q;
72	int d, z, z0;
73
74	unative_t wd$$, wq$$, wp$$, w1$$, w2$$;
75	unative_t x1d = NBYTES(0x1d);
76
77	z0 = disks - 3;		/* Highest data disk */
78	p = dptr[z0+1];		/* XOR parity */
79	q = dptr[z0+2];		/* RS syndrome */
80
81	for ( d = 0 ; d < bytes ; d += NSIZE*$# ) {
82		wq$$ = wp$$ = *(unative_t *)&dptr[z0][d+$$*NSIZE];
83		for ( z = z0-1 ; z >= 0 ; z-- ) {
84			wd$$ = *(unative_t *)&dptr[z][d+$$*NSIZE];
85			wp$$ = vec_xor(wp$$, wd$$);
86			w2$$ = MASK(wq$$);
87			w1$$ = SHLBYTE(wq$$);
88			w2$$ = vec_and(w2$$, x1d);
89			w1$$ = vec_xor(w1$$, w2$$);
90			wq$$ = vec_xor(w1$$, wd$$);
91		}
92		*(unative_t *)&p[d+NSIZE*$$] = wp$$;
93		*(unative_t *)&q[d+NSIZE*$$] = wq$$;
94	}
95}
96
97static void raid6_altivec$#_gen_syndrome(int disks, size_t bytes, void **ptrs)
98{
99	preempt_disable();
100	enable_kernel_altivec();
101
102	raid6_altivec$#_gen_syndrome_real(disks, bytes, ptrs);
103
104	preempt_enable();
105}
106
107int raid6_have_altivec(void);
108#if $# == 1
109int raid6_have_altivec(void)
110{
111	/* This assumes either all CPUs have Altivec or none does */
112# ifdef __KERNEL__
113	return cpu_has_feature(CPU_FTR_ALTIVEC);
114# else
115	return 1;
116# endif
117}
118#endif
119
120const struct raid6_calls raid6_altivec$# = {
121	raid6_altivec$#_gen_syndrome,
122	NULL,			/* XOR not yet implemented */
123	raid6_have_altivec,
124	"altivecx$#",
125	0
126};
127
128#endif /* CONFIG_ALTIVEC */
129