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#include "neon.h"
  9
 10#ifdef CONFIG_ARM
 11/*
 12 * AArch32 does not provide this intrinsic natively because it does not
 13 * implement the underlying instruction. AArch32 only provides a 64-bit
 14 * wide vtbl.8 instruction, so use that instead.
 15 */
 16static uint8x16_t vqtbl1q_u8(uint8x16_t a, uint8x16_t b)
 17{
 18	union {
 19		uint8x16_t	val;
 20		uint8x8x2_t	pair;
 21	} __a = { a };
 22
 23	return vcombine_u8(vtbl2_u8(__a.pair, vget_low_u8(b)),
 24			   vtbl2_u8(__a.pair, vget_high_u8(b)));
 25}
 26#endif
 27
 28void __raid6_2data_recov_neon(int bytes, uint8_t *p, uint8_t *q, uint8_t *dp,
 29			      uint8_t *dq, const uint8_t *pbmul,
 30			      const uint8_t *qmul)
 31{
 32	uint8x16_t pm0 = vld1q_u8(pbmul);
 33	uint8x16_t pm1 = vld1q_u8(pbmul + 16);
 34	uint8x16_t qm0 = vld1q_u8(qmul);
 35	uint8x16_t qm1 = vld1q_u8(qmul + 16);
 36	uint8x16_t x0f = vdupq_n_u8(0x0f);
 37
 38	/*
 39	 * while ( bytes-- ) {
 40	 *	uint8_t px, qx, db;
 41	 *
 42	 *	px    = *p ^ *dp;
 43	 *	qx    = qmul[*q ^ *dq];
 44	 *	*dq++ = db = pbmul[px] ^ qx;
 45	 *	*dp++ = db ^ px;
 46	 *	p++; q++;
 47	 * }
 48	 */
 49
 50	while (bytes) {
 51		uint8x16_t vx, vy, px, qx, db;
 52
 53		px = veorq_u8(vld1q_u8(p), vld1q_u8(dp));
 54		vx = veorq_u8(vld1q_u8(q), vld1q_u8(dq));
 55
 56		vy = vshrq_n_u8(vx, 4);
 57		vx = vqtbl1q_u8(qm0, vandq_u8(vx, x0f));
 58		vy = vqtbl1q_u8(qm1, vy);
 59		qx = veorq_u8(vx, vy);
 60
 61		vy = vshrq_n_u8(px, 4);
 62		vx = vqtbl1q_u8(pm0, vandq_u8(px, x0f));
 63		vy = vqtbl1q_u8(pm1, vy);
 64		vx = veorq_u8(vx, vy);
 65		db = veorq_u8(vx, qx);
 66
 67		vst1q_u8(dq, db);
 68		vst1q_u8(dp, veorq_u8(db, px));
 69
 70		bytes -= 16;
 71		p += 16;
 72		q += 16;
 73		dp += 16;
 74		dq += 16;
 75	}
 76}
 77
 78void __raid6_datap_recov_neon(int bytes, uint8_t *p, uint8_t *q, uint8_t *dq,
 79			      const uint8_t *qmul)
 80{
 81	uint8x16_t qm0 = vld1q_u8(qmul);
 82	uint8x16_t qm1 = vld1q_u8(qmul + 16);
 83	uint8x16_t x0f = vdupq_n_u8(0x0f);
 84
 85	/*
 86	 * while (bytes--) {
 87	 *	*p++ ^= *dq = qmul[*q ^ *dq];
 88	 *	q++; dq++;
 89	 * }
 90	 */
 91
 92	while (bytes) {
 93		uint8x16_t vx, vy;
 94
 95		vx = veorq_u8(vld1q_u8(q), vld1q_u8(dq));
 96
 97		vy = vshrq_n_u8(vx, 4);
 98		vx = vqtbl1q_u8(qm0, vandq_u8(vx, x0f));
 99		vy = vqtbl1q_u8(qm1, vy);
100		vx = veorq_u8(vx, vy);
101		vy = veorq_u8(vx, vld1q_u8(p));
102
103		vst1q_u8(dq, vx);
104		vst1q_u8(p, vy);
105
106		bytes -= 16;
107		p += 16;
108		q += 16;
109		dq += 16;
110	}
111}