| 1 |
// SPDX-License-Identifier: GPL-2.0-only |
| 2 |
/* |
| 3 |
* linux/arch/arm/lib/xor-neon.c |
| 4 |
* |
| 5 |
* Copyright (C) 2013 Linaro Ltd <[email protected]> |
| 6 |
*/ |
| 7 |
|
| 8 |
#include <linux/raid/xor.h> |
| 9 |
#include <linux/module.h> |
| 10 |
|
| 11 |
MODULE_DESCRIPTION("NEON accelerated XOR implementation"); |
| 12 |
MODULE_LICENSE("GPL"); |
| 13 |
|
| 14 |
#ifndef __ARM_NEON__ |
| 15 |
#error You should compile this file with '-march=armv7-a -mfloat-abi=softfp -mfpu=neon' |
| 16 |
#endif |
| 17 |
|
| 18 |
/* |
| 19 |
* Pull in the reference implementations while instructing GCC (through |
| 20 |
* -ftree-vectorize) to attempt to exploit implicit parallelism and emit |
| 21 |
* NEON instructions. Clang does this by default at O2 so no pragma is |
| 22 |
* needed. |
| 23 |
*/ |
| 24 |
#ifdef CONFIG_CC_IS_GCC |
| 25 |
#pragma GCC optimize "tree-vectorize" |
| 26 |
#endif |
| 27 |
|
| 28 |
#pragma GCC diagnostic ignored "-Wunused-variable" |
| 29 |
#include <asm-generic/xor.h> |
| 30 |
|
| 31 |
struct xor_block_template const xor_block_neon_inner = { |
| 32 |
.name = "__inner_neon__", |
| 33 |
.do_2 = xor_8regs_2, |
| 34 |
.do_3 = xor_8regs_3, |
| 35 |
.do_4 = xor_8regs_4, |
| 36 |
.do_5 = xor_8regs_5, |
| 37 |
}; |
| 38 |
EXPORT_SYMBOL(xor_block_neon_inner); |