blob: 1b9ed793342d757e09d19e2e5cd168e65655c0e7 [file] [log] [blame]
Ard Biesheuvel7d119652013-05-16 17:20:32 +02001/* -----------------------------------------------------------------------
2 *
3 * neon.uc - RAID-6 syndrome calculation using ARM NEON instructions
4 *
5 * Copyright (C) 2012 Rob Herring
6 *
7 * Based on altivec.uc:
8 * Copyright 2002-2004 H. Peter Anvin - All Rights Reserved
9 *
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License as published by
12 * the Free Software Foundation, Inc., 53 Temple Place Ste 330,
13 * Boston MA 02111-1307, USA; either version 2 of the License, or
14 * (at your option) any later version; incorporated herein by reference.
15 *
16 * ----------------------------------------------------------------------- */
17
18/*
19 * neon$#.c
20 *
21 * $#-way unrolled NEON intrinsics math RAID-6 instruction set
22 *
23 * This file is postprocessed using unroll.awk
24 */
25
26#include <arm_neon.h>
27
28typedef uint8x16_t unative_t;
29
30#define NBYTES(x) ((unative_t){x,x,x,x, x,x,x,x, x,x,x,x, x,x,x,x})
31#define NSIZE sizeof(unative_t)
32
33/*
34 * The SHLBYTE() operation shifts each byte left by 1, *not*
35 * rolling over into the next byte
36 */
37static inline unative_t SHLBYTE(unative_t v)
38{
39 return vshlq_n_u8(v, 1);
40}
41
42/*
43 * The MASK() operation returns 0xFF in any byte for which the high
44 * bit is 1, 0x00 for any byte for which the high bit is 0.
45 */
46static inline unative_t MASK(unative_t v)
47{
48 const uint8x16_t temp = NBYTES(0);
49 return (unative_t)vcltq_s8((int8x16_t)v, (int8x16_t)temp);
50}
51
52void raid6_neon$#_gen_syndrome_real(int disks, unsigned long bytes, void **ptrs)
53{
54 uint8_t **dptr = (uint8_t **)ptrs;
55 uint8_t *p, *q;
56 int d, z, z0;
57
58 register unative_t wd$$, wq$$, wp$$, w1$$, w2$$;
59 const unative_t x1d = NBYTES(0x1d);
60
61 z0 = disks - 3; /* Highest data disk */
62 p = dptr[z0+1]; /* XOR parity */
63 q = dptr[z0+2]; /* RS syndrome */
64
65 for ( d = 0 ; d < bytes ; d += NSIZE*$# ) {
66 wq$$ = wp$$ = vld1q_u8(&dptr[z0][d+$$*NSIZE]);
67 for ( z = z0-1 ; z >= 0 ; z-- ) {
68 wd$$ = vld1q_u8(&dptr[z][d+$$*NSIZE]);
69 wp$$ = veorq_u8(wp$$, wd$$);
70 w2$$ = MASK(wq$$);
71 w1$$ = SHLBYTE(wq$$);
72
73 w2$$ = vandq_u8(w2$$, x1d);
74 w1$$ = veorq_u8(w1$$, w2$$);
75 wq$$ = veorq_u8(w1$$, wd$$);
76 }
77 vst1q_u8(&p[d+NSIZE*$$], wp$$);
78 vst1q_u8(&q[d+NSIZE*$$], wq$$);
79 }
80}