xref: /linux/lib/crc/arm64/crc64-neon-inner.c (revision 0fc8f6200d2313278fbf4539bbab74677c685531)
163432fd6SDemian Shulhan // SPDX-License-Identifier: GPL-2.0-only
263432fd6SDemian Shulhan /*
363432fd6SDemian Shulhan  * Accelerated CRC64 (NVMe) using ARM NEON C intrinsics
463432fd6SDemian Shulhan  */
563432fd6SDemian Shulhan 
663432fd6SDemian Shulhan #include <linux/types.h>
763432fd6SDemian Shulhan #include <asm/neon-intrinsics.h>
863432fd6SDemian Shulhan 
963432fd6SDemian Shulhan u64 crc64_nvme_arm64_c(u64 crc, const u8 *p, size_t len);
1063432fd6SDemian Shulhan 
1163432fd6SDemian Shulhan /* x^191 mod G, x^127 mod G */
1263432fd6SDemian Shulhan static const u64 fold_consts_val[2] = { 0xeadc41fd2ba3d420ULL,
1363432fd6SDemian Shulhan 					0x21e9761e252621acULL };
1463432fd6SDemian Shulhan /* floor(x^127 / G), (G - x^64) / x */
1563432fd6SDemian Shulhan static const u64 bconsts_val[2] = { 0x27ecfa329aef9f77ULL,
1663432fd6SDemian Shulhan 				    0x34d926535897936aULL };
1763432fd6SDemian Shulhan 
18*8fdef85dSArd Biesheuvel static inline uint64x2_t pmull64(uint64x2_t a, uint64x2_t b)
19*8fdef85dSArd Biesheuvel {
20*8fdef85dSArd Biesheuvel 	return vreinterpretq_u64_p128(vmull_p64(vgetq_lane_u64(a, 0),
21*8fdef85dSArd Biesheuvel 						vgetq_lane_u64(b, 0)));
22*8fdef85dSArd Biesheuvel }
23*8fdef85dSArd Biesheuvel 
24*8fdef85dSArd Biesheuvel static inline uint64x2_t pmull64_high(uint64x2_t a, uint64x2_t b)
25*8fdef85dSArd Biesheuvel {
26*8fdef85dSArd Biesheuvel 	poly64x2_t l = vreinterpretq_p64_u64(a);
27*8fdef85dSArd Biesheuvel 	poly64x2_t m = vreinterpretq_p64_u64(b);
28*8fdef85dSArd Biesheuvel 
29*8fdef85dSArd Biesheuvel 	return vreinterpretq_u64_p128(vmull_high_p64(l, m));
30*8fdef85dSArd Biesheuvel }
31*8fdef85dSArd Biesheuvel 
32*8fdef85dSArd Biesheuvel static inline uint64x2_t pmull64_hi_lo(uint64x2_t a, uint64x2_t b)
33*8fdef85dSArd Biesheuvel {
34*8fdef85dSArd Biesheuvel 	return vreinterpretq_u64_p128(vmull_p64(vgetq_lane_u64(a, 1),
35*8fdef85dSArd Biesheuvel 						vgetq_lane_u64(b, 0)));
36*8fdef85dSArd Biesheuvel }
37*8fdef85dSArd Biesheuvel 
3863432fd6SDemian Shulhan u64 crc64_nvme_arm64_c(u64 crc, const u8 *p, size_t len)
3963432fd6SDemian Shulhan {
40*8fdef85dSArd Biesheuvel 	uint64x2_t fold_consts = vld1q_u64(fold_consts_val);
41*8fdef85dSArd Biesheuvel 	uint64x2_t v0 = { crc, 0 };
42*8fdef85dSArd Biesheuvel 	uint64x2_t zero = { };
4363432fd6SDemian Shulhan 
44*8fdef85dSArd Biesheuvel 	for (;;) {
45*8fdef85dSArd Biesheuvel 		v0 ^= vreinterpretq_u64_u8(vld1q_u8(p));
4663432fd6SDemian Shulhan 
4763432fd6SDemian Shulhan 		p += 16;
4863432fd6SDemian Shulhan 		len -= 16;
49*8fdef85dSArd Biesheuvel 		if (len < 16)
50*8fdef85dSArd Biesheuvel 			break;
51*8fdef85dSArd Biesheuvel 
52*8fdef85dSArd Biesheuvel 		v0 = pmull64(fold_consts, v0) ^ pmull64_high(fold_consts, v0);
53*8fdef85dSArd Biesheuvel 	}
5463432fd6SDemian Shulhan 
5563432fd6SDemian Shulhan 	/* Multiply the 128-bit value by x^64 and reduce it back to 128 bits. */
56*8fdef85dSArd Biesheuvel 	v0 = vextq_u64(v0, zero, 1) ^ pmull64_hi_lo(fold_consts, v0);
5763432fd6SDemian Shulhan 
5863432fd6SDemian Shulhan 	/* Final Barrett reduction */
59*8fdef85dSArd Biesheuvel 	uint64x2_t bconsts = vld1q_u64(bconsts_val);
60*8fdef85dSArd Biesheuvel 	uint64x2_t final = pmull64(bconsts, v0);
6163432fd6SDemian Shulhan 
62*8fdef85dSArd Biesheuvel 	v0 ^= vextq_u64(zero, final, 1) ^ pmull64_hi_lo(bconsts, final);
6363432fd6SDemian Shulhan 
64*8fdef85dSArd Biesheuvel 	return vgetq_lane_u64(v0, 1);
6563432fd6SDemian Shulhan }
66