blob: b99dd8e1c93f17230476608778449ddbfbe82488 [file] [log] [blame]
Thomas Gleixnerd2912cb2019-06-04 10:11:33 +02001// SPDX-License-Identifier: GPL-2.0-only
Ard Biesheuvel01956592013-05-17 18:51:23 +02002/*
3 * linux/arch/arm/lib/xor-neon.c
4 *
5 * Copyright (C) 2013 Linaro Ltd <ard.biesheuvel@linaro.org>
Ard Biesheuvel01956592013-05-17 18:51:23 +02006 */
7
8#include <linux/raid/xor.h>
Ard Biesheuvel93192062013-09-09 15:08:38 +01009#include <linux/module.h>
10
11MODULE_LICENSE("GPL");
Ard Biesheuvel01956592013-05-17 18:51:23 +020012
13#ifndef __ARM_NEON__
Nathan Chancellorde9c0d42019-02-02 03:34:36 +010014#error You should compile this file with '-march=armv7-a -mfloat-abi=softfp -mfpu=neon'
Ard Biesheuvel01956592013-05-17 18:51:23 +020015#endif
16
17/*
18 * Pull in the reference implementations while instructing GCC (through
19 * -ftree-vectorize) to attempt to exploit implicit parallelism and emit
20 * NEON instructions.
21 */
22#if __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 6)
23#pragma GCC optimize "tree-vectorize"
24#else
25/*
26 * While older versions of GCC do not generate incorrect code, they fail to
27 * recognize the parallel nature of these functions, and emit plain ARM code,
28 * which is known to be slower than the optimized ARM code in asm-arm/xor.h.
29 */
30#warning This code requires at least version 4.6 of GCC
31#endif
32
33#pragma GCC diagnostic ignored "-Wunused-variable"
34#include <asm-generic/xor.h>
35
36struct xor_block_template const xor_block_neon_inner = {
37 .name = "__inner_neon__",
38 .do_2 = xor_8regs_2,
39 .do_3 = xor_8regs_3,
40 .do_4 = xor_8regs_4,
41 .do_5 = xor_8regs_5,
42};
Ard Biesheuvel93192062013-09-09 15:08:38 +010043EXPORT_SYMBOL(xor_block_neon_inner);