xref: /linux/arch/arm/lib/xor-neon.c (revision 3a39d672e7f48b8d6b91a09afa4b55352773b4b5)
1d2912cb1SThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only
201956597SArd Biesheuvel /*
301956597SArd Biesheuvel  * linux/arch/arm/lib/xor-neon.c
401956597SArd Biesheuvel  *
501956597SArd Biesheuvel  * Copyright (C) 2013 Linaro Ltd <ard.biesheuvel@linaro.org>
601956597SArd Biesheuvel  */
701956597SArd Biesheuvel 
801956597SArd Biesheuvel #include <linux/raid/xor.h>
99319206dSArd Biesheuvel #include <linux/module.h>
109319206dSArd Biesheuvel 
11*f6fc302dSJeff Johnson MODULE_DESCRIPTION("NEON accelerated XOR implementation");
129319206dSArd Biesheuvel MODULE_LICENSE("GPL");
1301956597SArd Biesheuvel 
1401956597SArd Biesheuvel #ifndef __ARM_NEON__
15de9c0d49SNathan Chancellor #error You should compile this file with '-march=armv7-a -mfloat-abi=softfp -mfpu=neon'
1601956597SArd Biesheuvel #endif
1701956597SArd Biesheuvel 
1801956597SArd Biesheuvel /*
1901956597SArd Biesheuvel  * Pull in the reference implementations while instructing GCC (through
2001956597SArd Biesheuvel  * -ftree-vectorize) to attempt to exploit implicit parallelism and emit
21a69cb445SArd Biesheuvel  * NEON instructions. Clang does this by default at O2 so no pragma is
22a69cb445SArd Biesheuvel  * needed.
2301956597SArd Biesheuvel  */
24a69cb445SArd Biesheuvel #ifdef CONFIG_CC_IS_GCC
2501956597SArd Biesheuvel #pragma GCC optimize "tree-vectorize"
2601956597SArd Biesheuvel #endif
2701956597SArd Biesheuvel 
2801956597SArd Biesheuvel #pragma GCC diagnostic ignored "-Wunused-variable"
2901956597SArd Biesheuvel #include <asm-generic/xor.h>
3001956597SArd Biesheuvel 
3101956597SArd Biesheuvel struct xor_block_template const xor_block_neon_inner = {
3201956597SArd Biesheuvel 	.name	= "__inner_neon__",
3301956597SArd Biesheuvel 	.do_2	= xor_8regs_2,
3401956597SArd Biesheuvel 	.do_3	= xor_8regs_3,
3501956597SArd Biesheuvel 	.do_4	= xor_8regs_4,
3601956597SArd Biesheuvel 	.do_5	= xor_8regs_5,
3701956597SArd Biesheuvel };
389319206dSArd Biesheuvel EXPORT_SYMBOL(xor_block_neon_inner);
39