xref: /OK3568_Linux_fs/kernel/arch/arm/lib/xor-neon.c (revision 4882a59341e53eb6f0b4789bf948001014eff981)
1*4882a593Smuzhiyun // SPDX-License-Identifier: GPL-2.0-only
2*4882a593Smuzhiyun /*
3*4882a593Smuzhiyun  * linux/arch/arm/lib/xor-neon.c
4*4882a593Smuzhiyun  *
5*4882a593Smuzhiyun  * Copyright (C) 2013 Linaro Ltd <ard.biesheuvel@linaro.org>
6*4882a593Smuzhiyun  */
7*4882a593Smuzhiyun 
8*4882a593Smuzhiyun #include <linux/raid/xor.h>
9*4882a593Smuzhiyun #include <linux/module.h>
10*4882a593Smuzhiyun 
11*4882a593Smuzhiyun MODULE_LICENSE("GPL");
12*4882a593Smuzhiyun 
13*4882a593Smuzhiyun #ifndef __ARM_NEON__
14*4882a593Smuzhiyun #error You should compile this file with '-march=armv7-a -mfloat-abi=softfp -mfpu=neon'
15*4882a593Smuzhiyun #endif
16*4882a593Smuzhiyun 
17*4882a593Smuzhiyun /*
18*4882a593Smuzhiyun  * Pull in the reference implementations while instructing GCC (through
19*4882a593Smuzhiyun  * -ftree-vectorize) to attempt to exploit implicit parallelism and emit
20*4882a593Smuzhiyun  * NEON instructions.
21*4882a593Smuzhiyun  */
22*4882a593Smuzhiyun #if __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 6)
23*4882a593Smuzhiyun #pragma GCC optimize "tree-vectorize"
24*4882a593Smuzhiyun #else
25*4882a593Smuzhiyun /*
26*4882a593Smuzhiyun  * While older versions of GCC do not generate incorrect code, they fail to
27*4882a593Smuzhiyun  * recognize the parallel nature of these functions, and emit plain ARM code,
28*4882a593Smuzhiyun  * which is known to be slower than the optimized ARM code in asm-arm/xor.h.
29*4882a593Smuzhiyun  *
30*4882a593Smuzhiyun  * #warning This code requires at least version 4.6 of GCC
31*4882a593Smuzhiyun  */
32*4882a593Smuzhiyun #endif
33*4882a593Smuzhiyun 
34*4882a593Smuzhiyun #pragma GCC diagnostic ignored "-Wunused-variable"
35*4882a593Smuzhiyun #include <asm-generic/xor.h>
36*4882a593Smuzhiyun 
37*4882a593Smuzhiyun struct xor_block_template const xor_block_neon_inner = {
38*4882a593Smuzhiyun 	.name	= "__inner_neon__",
39*4882a593Smuzhiyun 	.do_2	= xor_8regs_2,
40*4882a593Smuzhiyun 	.do_3	= xor_8regs_3,
41*4882a593Smuzhiyun 	.do_4	= xor_8regs_4,
42*4882a593Smuzhiyun 	.do_5	= xor_8regs_5,
43*4882a593Smuzhiyun };
44*4882a593Smuzhiyun EXPORT_SYMBOL(xor_block_neon_inner);
45