1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 * Copyright (C) 2012 Intel Corporation
  4 * Copyright (C) 2017 Linaro Ltd. <ard.biesheuvel@linaro.org>
  5 */
  6
  7#include <arm_neon.h>
  8
  9#ifdef CONFIG_ARM
 10/*
 11 * AArch32 does not provide this intrinsic natively because it does not
 12 * implement the underlying instruction. AArch32 only provides a 64-bit
 13 * wide vtbl.8 instruction, so use that instead.
 14 */
 15static uint8x16_t vqtbl1q_u8(uint8x16_t a, uint8x16_t b)
 16{
 17	union {
 18		uint8x16_t	val;
 19		uint8x8x2_t	pair;
 20	} __a = { a };
 21
 22	return vcombine_u8(vtbl2_u8(__a.pair, vget_low_u8(b)),
 23			   vtbl2_u8(__a.pair, vget_high_u8(b)));
 24}
 25#endif
 26
 27void __raid6_2data_recov_neon(int bytes, uint8_t *p, uint8_t *q, uint8_t *dp,
 28			      uint8_t *dq, const uint8_t *pbmul,
 29			      const uint8_t *qmul)
 30{
 31	uint8x16_t pm0 = vld1q_u8(pbmul);
 32	uint8x16_t pm1 = vld1q_u8(pbmul + 16);
 33	uint8x16_t qm0 = vld1q_u8(qmul);
 34	uint8x16_t qm1 = vld1q_u8(qmul + 16);
 35	uint8x16_t x0f = vdupq_n_u8(0x0f);
 36
 37	/*
 38	 * while ( bytes-- ) {
 39	 *	uint8_t px, qx, db;
 40	 *
 41	 *	px    = *p ^ *dp;
 42	 *	qx    = qmul[*q ^ *dq];
 43	 *	*dq++ = db = pbmul[px] ^ qx;
 44	 *	*dp++ = db ^ px;
 45	 *	p++; q++;
 46	 * }
 47	 */
 48
 49	while (bytes) {
 50		uint8x16_t vx, vy, px, qx, db;
 51
 52		px = veorq_u8(vld1q_u8(p), vld1q_u8(dp));
 53		vx = veorq_u8(vld1q_u8(q), vld1q_u8(dq));
 54
 55		vy = vshrq_n_u8(vx, 4);
 56		vx = vqtbl1q_u8(qm0, vandq_u8(vx, x0f));
 57		vy = vqtbl1q_u8(qm1, vy);
 58		qx = veorq_u8(vx, vy);
 59
 60		vy = vshrq_n_u8(px, 4);
 61		vx = vqtbl1q_u8(pm0, vandq_u8(px, x0f));
 62		vy = vqtbl1q_u8(pm1, vy);
 63		vx = veorq_u8(vx, vy);
 64		db = veorq_u8(vx, qx);
 65
 66		vst1q_u8(dq, db);
 67		vst1q_u8(dp, veorq_u8(db, px));
 68
 69		bytes -= 16;
 70		p += 16;
 71		q += 16;
 72		dp += 16;
 73		dq += 16;
 74	}
 75}
 76
 77void __raid6_datap_recov_neon(int bytes, uint8_t *p, uint8_t *q, uint8_t *dq,
 78			      const uint8_t *qmul)
 79{
 80	uint8x16_t qm0 = vld1q_u8(qmul);
 81	uint8x16_t qm1 = vld1q_u8(qmul + 16);
 82	uint8x16_t x0f = vdupq_n_u8(0x0f);
 83
 84	/*
 85	 * while (bytes--) {
 86	 *	*p++ ^= *dq = qmul[*q ^ *dq];
 87	 *	q++; dq++;
 88	 * }
 89	 */
 90
 91	while (bytes) {
 92		uint8x16_t vx, vy;
 93
 94		vx = veorq_u8(vld1q_u8(q), vld1q_u8(dq));
 95
 96		vy = vshrq_n_u8(vx, 4);
 97		vx = vqtbl1q_u8(qm0, vandq_u8(vx, x0f));
 98		vy = vqtbl1q_u8(qm1, vy);
 99		vx = veorq_u8(vx, vy);
100		vy = veorq_u8(vx, vld1q_u8(p));
101
102		vst1q_u8(dq, vx);
103		vst1q_u8(p, vy);
104
105		bytes -= 16;
106		p += 16;
107		q += 16;
108		dq += 16;
109	}
110}