1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Accelerated CRC64 (NVMe) using ARM NEON C intrinsics 4 */ 5 6 #include <linux/types.h> 7 #include <asm/neon-intrinsics.h> 8 9 #include "crc64-neon.h" 10 11 u64 crc64_nvme_neon(u64 crc, const u8 *p, size_t len); 12 13 /* x^191 mod G, x^127 mod G */ 14 static const u64 fold_consts_val[2] = { 0xeadc41fd2ba3d420ULL, 15 0x21e9761e252621acULL }; 16 /* floor(x^127 / G), (G - x^64) / x */ 17 static const u64 bconsts_val[2] = { 0x27ecfa329aef9f77ULL, 18 0x34d926535897936aULL }; 19 20 u64 crc64_nvme_neon(u64 crc, const u8 *p, size_t len) 21 { 22 uint64x2_t fold_consts = vld1q_u64(fold_consts_val); 23 uint64x2_t v0 = { crc, 0 }; 24 uint64x2_t zero = { }; 25 26 for (;;) { 27 v0 ^= vreinterpretq_u64_u8(vld1q_u8(p)); 28 29 p += 16; 30 len -= 16; 31 if (len < 16) 32 break; 33 34 v0 = pmull64(fold_consts, v0) ^ pmull64_high(fold_consts, v0); 35 } 36 37 /* Multiply the 128-bit value by x^64 and reduce it back to 128 bits. */ 38 v0 = vextq_u64(v0, zero, 1) ^ pmull64_hi_lo(fold_consts, v0); 39 40 /* Final Barrett reduction */ 41 uint64x2_t bconsts = vld1q_u64(bconsts_val); 42 uint64x2_t final = pmull64(bconsts, v0); 43 44 v0 ^= vextq_u64(zero, final, 1) ^ pmull64_hi_lo(bconsts, final); 45 46 return vgetq_lane_u64(v0, 1); 47 } 48