32 lines
776 B
C
32 lines
776 B
C
// SPDX-License-Identifier: GPL-2.0-only
|
|
/*
|
|
* Copyright (C) 2013 Linaro Ltd <ard.biesheuvel@linaro.org>
|
|
*/
|
|
|
|
#include "xor_impl.h"
|
|
|
|
#ifndef __ARM_NEON__
|
|
#error You should compile this file with '-march=armv7-a -mfloat-abi=softfp -mfpu=neon'
|
|
#endif
|
|
|
|
/*
|
|
* Pull in the reference implementations while instructing GCC (through
|
|
* -ftree-vectorize) to attempt to exploit implicit parallelism and emit
|
|
* NEON instructions. Clang does this by default at O2 so no pragma is
|
|
* needed.
|
|
*/
|
|
#ifdef CONFIG_CC_IS_GCC
|
|
#pragma GCC optimize "tree-vectorize"
|
|
#endif
|
|
|
|
#define NO_TEMPLATE
|
|
#include "../xor-8regs.c"
|
|
|
|
struct xor_block_template const xor_block_neon_inner = {
|
|
.name = "__inner_neon__",
|
|
.do_2 = xor_8regs_2,
|
|
.do_3 = xor_8regs_3,
|
|
.do_4 = xor_8regs_4,
|
|
.do_5 = xor_8regs_5,
|
|
};
|