blob: 2c40aeab3eaae8cb038a283b6fa2dc422d744d08 [file] [log] [blame]
Ard Biesheuvel01956592013-05-17 18:51:23 +02001/*
2 * linux/arch/arm/lib/xor-neon.c
3 *
4 * Copyright (C) 2013 Linaro Ltd <ard.biesheuvel@linaro.org>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 as
8 * published by the Free Software Foundation.
9 */
10
11#include <linux/raid/xor.h>
Ard Biesheuvel93192062013-09-09 15:08:38 +010012#include <linux/module.h>
13
14MODULE_LICENSE("GPL");
Ard Biesheuvel01956592013-05-17 18:51:23 +020015
16#ifndef __ARM_NEON__
17#error You should compile this file with '-mfloat-abi=softfp -mfpu=neon'
18#endif
19
20/*
21 * Pull in the reference implementations while instructing GCC (through
22 * -ftree-vectorize) to attempt to exploit implicit parallelism and emit
23 * NEON instructions.
24 */
25#if __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 6)
26#pragma GCC optimize "tree-vectorize"
27#else
28/*
29 * While older versions of GCC do not generate incorrect code, they fail to
30 * recognize the parallel nature of these functions, and emit plain ARM code,
31 * which is known to be slower than the optimized ARM code in asm-arm/xor.h.
32 */
33#warning This code requires at least version 4.6 of GCC
34#endif
35
36#pragma GCC diagnostic ignored "-Wunused-variable"
37#include <asm-generic/xor.h>
38
39struct xor_block_template const xor_block_neon_inner = {
40 .name = "__inner_neon__",
41 .do_2 = xor_8regs_2,
42 .do_3 = xor_8regs_3,
43 .do_4 = xor_8regs_4,
44 .do_5 = xor_8regs_5,
45};
Ard Biesheuvel93192062013-09-09 15:08:38 +010046EXPORT_SYMBOL(xor_block_neon_inner);