[linux-2.6-block.git] / lib / raid6 / recov_avx2.c

// SPDX-License-Identifier: GPL-2.0-only
/*
 * Copyright (C) 2012 Intel Corporation
 * Author: Jim Kukunas <james.t.kukunas@linux.intel.com>
 */

#ifdef CONFIG_AS_AVX2

#include <linux/raid/pq.h>
#include "x86.h"

static int raid6_has_avx2(void)
{
	return boot_cpu_has(X86_FEATURE_AVX2) &&
		boot_cpu_has(X86_FEATURE_AVX);
}

static void raid6_2data_recov_avx2(int disks, size_t bytes, int faila,
		int failb, void **ptrs)
{
	u8 *p, *q, *dp, *dq;
	const u8 *pbmul;	/* P multiplier table for B data */
	const u8 *qmul;		/* Q multiplier table (for both) */
	const u8 x0f = 0x0f;

	p = (u8 *)ptrs[disks-2];
	q = (u8 *)ptrs[disks-1];

	/* Compute syndrome with zero for the missing data pages
	   Use the dead data pages as temporary storage for
	   delta p and delta q */
	dp = (u8 *)ptrs[faila];
	ptrs[faila] = (void *)raid6_empty_zero_page;
	ptrs[disks-2] = dp;
	dq = (u8 *)ptrs[failb];
	ptrs[failb] = (void *)raid6_empty_zero_page;
	ptrs[disks-1] = dq;

	raid6_call.gen_syndrome(disks, bytes, ptrs);

	/* Restore pointer table */
	ptrs[faila]   = dp;
	ptrs[failb]   = dq;
	ptrs[disks-2] = p;
	ptrs[disks-1] = q;

	/* Now, pick the proper data tables */
	pbmul = raid6_vgfmul[raid6_gfexi[failb-faila]];
	qmul  = raid6_vgfmul[raid6_gfinv[raid6_gfexp[faila] ^
		raid6_gfexp[failb]]];

	kernel_fpu_begin();

	/* ymm0 = x0f[16] */
	asm volatile("vpbroadcastb %0, %%ymm7" : : "m" (x0f));

	while (bytes) {
#ifdef CONFIG_X86_64
		asm volatile("vmovdqa %0, %%ymm1" : : "m" (q[0]));
		asm volatile("vmovdqa %0, %%ymm9" : : "m" (q[32]));
		asm volatile("vmovdqa %0, %%ymm0" : : "m" (p[0]));
		asm volatile("vmovdqa %0, %%ymm8" : : "m" (p[32]));
		asm volatile("vpxor %0, %%ymm1, %%ymm1" : : "m" (dq[0]));
		asm volatile("vpxor %0, %%ymm9, %%ymm9" : : "m" (dq[32]));
		asm volatile("vpxor %0, %%ymm0, %%ymm0" : : "m" (dp[0]));
		asm volatile("vpxor %0, %%ymm8, %%ymm8" : : "m" (dp[32]));

		/*
		 * 1 = dq[0]  ^ q[0]
		 * 9 = dq[32] ^ q[32]
		 * 0 = dp[0]  ^ p[0]
		 * 8 = dp[32] ^ p[32]
		 */

		asm volatile("vbroadcasti128 %0, %%ymm4" : : "m" (qmul[0]));
		asm volatile("vbroadcasti128 %0, %%ymm5" : : "m" (qmul[16]));

		asm volatile("vpsraw $4, %ymm1, %ymm3");
		asm volatile("vpsraw $4, %ymm9, %ymm12");
		asm volatile("vpand %ymm7, %ymm1, %ymm1");
		asm volatile("vpand %ymm7, %ymm9, %ymm9");
		asm volatile("vpand %ymm7, %ymm3, %ymm3");
		asm volatile("vpand %ymm7, %ymm12, %ymm12");
		asm volatile("vpshufb %ymm9, %ymm4, %ymm14");
		asm volatile("vpshufb %ymm1, %ymm4, %ymm4");
		asm volatile("vpshufb %ymm12, %ymm5, %ymm15");
		asm volatile("vpshufb %ymm3, %ymm5, %ymm5");
		asm volatile("vpxor %ymm14, %ymm15, %ymm15");
		asm volatile("vpxor %ymm4, %ymm5, %ymm5");

		/*
		 * 5 = qx[0]
		 * 15 = qx[32]
		 */

		asm volatile("vbroadcasti128 %0, %%ymm4" : : "m" (pbmul[0]));
		asm volatile("vbroadcasti128 %0, %%ymm1" : : "m" (pbmul[16]));
		asm volatile("vpsraw $4, %ymm0, %ymm2");
		asm volatile("vpsraw $4, %ymm8, %ymm6");
		asm volatile("vpand %ymm7, %ymm0, %ymm3");
		asm volatile("vpand %ymm7, %ymm8, %ymm14");
		asm volatile("vpand %ymm7, %ymm2, %ymm2");
		asm volatile("vpand %ymm7, %ymm6, %ymm6");
		asm volatile("vpshufb %ymm14, %ymm4, %ymm12");
		asm volatile("vpshufb %ymm3, %ymm4, %ymm4");
		asm volatile("vpshufb %ymm6, %ymm1, %ymm13");
		asm volatile("vpshufb %ymm2, %ymm1, %ymm1");
		asm volatile("vpxor %ymm4, %ymm1, %ymm1");
		asm volatile("vpxor %ymm12, %ymm13, %ymm13");

		/*
		 * 1  = pbmul[px[0]]
		 * 13 = pbmul[px[32]]
		 */
		asm volatile("vpxor %ymm5, %ymm1, %ymm1");
		asm volatile("vpxor %ymm15, %ymm13, %ymm13");

		/*
		 * 1 = db = DQ
		 * 13 = db[32] = DQ[32]
		 */
		asm volatile("vmovdqa %%ymm1, %0" : "=m" (dq[0]));
		asm volatile("vmovdqa %%ymm13,%0" : "=m" (dq[32]));
		asm volatile("vpxor %ymm1, %ymm0, %ymm0");
		asm volatile("vpxor %ymm13, %ymm8, %ymm8");

		asm volatile("vmovdqa %%ymm0, %0" : "=m" (dp[0]));
		asm volatile("vmovdqa %%ymm8, %0" : "=m" (dp[32]));

		bytes -= 64;
		p += 64;
		q += 64;
		dp += 64;
		dq += 64;
#else
		asm volatile("vmovdqa %0, %%ymm1" : : "m" (*q));
		asm volatile("vmovdqa %0, %%ymm0" : : "m" (*p));
		asm volatile("vpxor %0, %%ymm1, %%ymm1" : : "m" (*dq));
		asm volatile("vpxor %0, %%ymm0, %%ymm0" : : "m" (*dp));

		/* 1 = dq ^ q;  0 = dp ^ p */

		asm volatile("vbroadcasti128 %0, %%ymm4" : : "m" (qmul[0]));
		asm volatile("vbroadcasti128 %0, %%ymm5" : : "m" (qmul[16]));

		/*
		 * 1 = dq ^ q
		 * 3 = dq ^ p >> 4
		 */
		asm volatile("vpsraw $4, %ymm1, %ymm3");
		asm volatile("vpand %ymm7, %ymm1, %ymm1");
		asm volatile("vpand %ymm7, %ymm3, %ymm3");
		asm volatile("vpshufb %ymm1, %ymm4, %ymm4");
		asm volatile("vpshufb %ymm3, %ymm5, %ymm5");
		asm volatile("vpxor %ymm4, %ymm5, %ymm5");

		/* 5 = qx */

		asm volatile("vbroadcasti128 %0, %%ymm4" : : "m" (pbmul[0]));
		asm volatile("vbroadcasti128 %0, %%ymm1" : : "m" (pbmul[16]));

		asm volatile("vpsraw $4, %ymm0, %ymm2");
		asm volatile("vpand %ymm7, %ymm0, %ymm3");
		asm volatile("vpand %ymm7, %ymm2, %ymm2");
		asm volatile("vpshufb %ymm3, %ymm4, %ymm4");
		asm volatile("vpshufb %ymm2, %ymm1, %ymm1");
		asm volatile("vpxor %ymm4, %ymm1, %ymm1");

		/* 1 = pbmul[px] */
		asm volatile("vpxor %ymm5, %ymm1, %ymm1");
		/* 1 = db = DQ */
		asm volatile("vmovdqa %%ymm1, %0" : "=m" (dq[0]));

		asm volatile("vpxor %ymm1, %ymm0, %ymm0");
		asm volatile("vmovdqa %%ymm0, %0" : "=m" (dp[0]));

		bytes -= 32;
		p += 32;
		q += 32;
		dp += 32;
		dq += 32;
#endif
	}

	kernel_fpu_end();
}

static void raid6_datap_recov_avx2(int disks, size_t bytes, int faila,
		void **ptrs)
{
	u8 *p, *q, *dq;
	const u8 *qmul;		/* Q multiplier table */
	const u8 x0f = 0x0f;

	p = (u8 *)ptrs[disks-2];
	q = (u8 *)ptrs[disks-1];

	/* Compute syndrome with zero for the missing data page
	   Use the dead data page as temporary storage for delta q */
	dq = (u8 *)ptrs[faila];
	ptrs[faila] = (void *)raid6_empty_zero_page;
	ptrs[disks-1] = dq;

	raid6_call.gen_syndrome(disks, bytes, ptrs);

	/* Restore pointer table */
	ptrs[faila]   = dq;
	ptrs[disks-1] = q;

	/* Now, pick the proper data tables */
	qmul  = raid6_vgfmul[raid6_gfinv[raid6_gfexp[faila]]];

	kernel_fpu_begin();

	asm volatile("vpbroadcastb %0, %%ymm7" : : "m" (x0f));

	while (bytes) {
#ifdef CONFIG_X86_64
		asm volatile("vmovdqa %0, %%ymm3" : : "m" (dq[0]));
		asm volatile("vmovdqa %0, %%ymm8" : : "m" (dq[32]));
		asm volatile("vpxor %0, %%ymm3, %%ymm3" : : "m" (q[0]));
		asm volatile("vpxor %0, %%ymm8, %%ymm8" : : "m" (q[32]));

		/*
		 * 3 = q[0] ^ dq[0]
		 * 8 = q[32] ^ dq[32]
		 */
		asm volatile("vbroadcasti128 %0, %%ymm0" : : "m" (qmul[0]));
		asm volatile("vmovapd %ymm0, %ymm13");
		asm volatile("vbroadcasti128 %0, %%ymm1" : : "m" (qmul[16]));
		asm volatile("vmovapd %ymm1, %ymm14");

		asm volatile("vpsraw $4, %ymm3, %ymm6");
		asm volatile("vpsraw $4, %ymm8, %ymm12");
		asm volatile("vpand %ymm7, %ymm3, %ymm3");
		asm volatile("vpand %ymm7, %ymm8, %ymm8");
		asm volatile("vpand %ymm7, %ymm6, %ymm6");
		asm volatile("vpand %ymm7, %ymm12, %ymm12");
		asm volatile("vpshufb %ymm3, %ymm0, %ymm0");
		asm volatile("vpshufb %ymm8, %ymm13, %ymm13");
		asm volatile("vpshufb %ymm6, %ymm1, %ymm1");
		asm volatile("vpshufb %ymm12, %ymm14, %ymm14");
		asm volatile("vpxor %ymm0, %ymm1, %ymm1");
		asm volatile("vpxor %ymm13, %ymm14, %ymm14");

		/*
		 * 1  = qmul[q[0]  ^ dq[0]]
		 * 14 = qmul[q[32] ^ dq[32]]
		 */
		asm volatile("vmovdqa %0, %%ymm2" : : "m" (p[0]));
		asm volatile("vmovdqa %0, %%ymm12" : : "m" (p[32]));
		asm volatile("vpxor %ymm1, %ymm2, %ymm2");
		asm volatile("vpxor %ymm14, %ymm12, %ymm12");

		/*
		 * 2  = p[0]  ^ qmul[q[0]  ^ dq[0]]
		 * 12 = p[32] ^ qmul[q[32] ^ dq[32]]
		 */

		asm volatile("vmovdqa %%ymm1, %0" : "=m" (dq[0]));
		asm volatile("vmovdqa %%ymm14, %0" : "=m" (dq[32]));
		asm volatile("vmovdqa %%ymm2, %0" : "=m" (p[0]));
		asm volatile("vmovdqa %%ymm12,%0" : "=m" (p[32]));

		bytes -= 64;
		p += 64;
		q += 64;
		dq += 64;
#else
		asm volatile("vmovdqa %0, %%ymm3" : : "m" (dq[0]));
		asm volatile("vpxor %0, %%ymm3, %%ymm3" : : "m" (q[0]));

		/* 3 = q ^ dq */

		asm volatile("vbroadcasti128 %0, %%ymm0" : : "m" (qmul[0]));
		asm volatile("vbroadcasti128 %0, %%ymm1" : : "m" (qmul[16]));

		asm volatile("vpsraw $4, %ymm3, %ymm6");
		asm volatile("vpand %ymm7, %ymm3, %ymm3");
		asm volatile("vpand %ymm7, %ymm6, %ymm6");
		asm volatile("vpshufb %ymm3, %ymm0, %ymm0");
		asm volatile("vpshufb %ymm6, %ymm1, %ymm1");
		asm volatile("vpxor %ymm0, %ymm1, %ymm1");

		/* 1 = qmul[q ^ dq] */

		asm volatile("vmovdqa %0, %%ymm2" : : "m" (p[0]));
		asm volatile("vpxor %ymm1, %ymm2, %ymm2");

		/* 2 = p ^ qmul[q ^ dq] */

		asm volatile("vmovdqa %%ymm1, %0" : "=m" (dq[0]));
		asm volatile("vmovdqa %%ymm2, %0" : "=m" (p[0]));

		bytes -= 32;
		p += 32;
		q += 32;
		dq += 32;
#endif
	}

	kernel_fpu_end();
}

const struct raid6_recov_calls raid6_recov_avx2 = {
	.data2 = raid6_2data_recov_avx2,
	.datap = raid6_datap_recov_avx2,
	.valid = raid6_has_avx2,
#ifdef CONFIG_X86_64
	.name = "avx2x2",
#else
	.name = "avx2x1",
#endif
	.priority = 2,
};

#else
#warning "your version of binutils lacks AVX2 support"
#endif
Commit	Line	Data
b886d83c	1	// SPDX-License-Identifier: GPL-2.0-only
7056741f JK	2	/*
	3	* Copyright (C) 2012 Intel Corporation
	4	* Author: Jim Kukunas <james.t.kukunas@linux.intel.com>
7056741f JK	5	*/
7056741f JK	6
75aaf4c3	7	#ifdef CONFIG_AS_AVX2
7056741f JK	8
	9	#include <linux/raid/pq.h>
	10	#include "x86.h"
	11
	12	static int raid6_has_avx2(void)
	13	{
	14	return boot_cpu_has(X86_FEATURE_AVX2) &&
	15	boot_cpu_has(X86_FEATURE_AVX);
	16	}
	17
	18	static void raid6_2data_recov_avx2(int disks, size_t bytes, int faila,
	19	int failb, void **ptrs)
	20	{
	21	u8 p, q, dp, dq;
	22	const u8 pbmul; / P multiplier table for B data */
	23	const u8 qmul; / Q multiplier table (for both) */
	24	const u8 x0f = 0x0f;
	25
	26	p = (u8 *)ptrs[disks-2];
	27	q = (u8 *)ptrs[disks-1];
	28
	29	/* Compute syndrome with zero for the missing data pages
	30	Use the dead data pages as temporary storage for
	31	delta p and delta q */
	32	dp = (u8 *)ptrs[faila];
	33	ptrs[faila] = (void *)raid6_empty_zero_page;
	34	ptrs[disks-2] = dp;
	35	dq = (u8 *)ptrs[failb];
	36	ptrs[failb] = (void *)raid6_empty_zero_page;
	37	ptrs[disks-1] = dq;
	38
	39	raid6_call.gen_syndrome(disks, bytes, ptrs);
	40
	41	/* Restore pointer table */
	42	ptrs[faila] = dp;
	43	ptrs[failb] = dq;
	44	ptrs[disks-2] = p;
	45	ptrs[disks-1] = q;
	46
	47	/* Now, pick the proper data tables */
	48	pbmul = raid6_vgfmul[raid6_gfexi[failb-faila]];
	49	qmul = raid6_vgfmul[raid6_gfinv[raid6_gfexp[faila] ^
	50	raid6_gfexp[failb]]];
	51
	52	kernel_fpu_begin();
	53
	54	/* ymm0 = x0f[16] */
	55	asm volatile("vpbroadcastb %0, %%ymm7" : : "m" (x0f));
	56
	57	while (bytes) {
	58	#ifdef CONFIG_X86_64
	59	asm volatile("vmovdqa %0, %%ymm1" : : "m" (q[0]));
	60	asm volatile("vmovdqa %0, %%ymm9" : : "m" (q[32]));
	61	asm volatile("vmovdqa %0, %%ymm0" : : "m" (p[0]));
	62	asm volatile("vmovdqa %0, %%ymm8" : : "m" (p[32]));
	63	asm volatile("vpxor %0, %%ymm1, %%ymm1" : : "m" (dq[0]));
	64	asm volatile("vpxor %0, %%ymm9, %%ymm9" : : "m" (dq[32]));
	65	asm volatile("vpxor %0, %%ymm0, %%ymm0" : : "m" (dp[0]));
	66	asm volatile("vpxor %0, %%ymm8, %%ymm8" : : "m" (dp[32]));
	67
	68	/*
	69	* 1 = dq[0] ^ q[0]
	70	* 9 = dq[32] ^ q[32]
	71	* 0 = dp[0] ^ p[0]
72	* 8 = dp[32] ^ p[32]
73	*/
74
75	asm volatile("vbroadcasti128 %0, %%ymm4" : : "m" (qmul[0]));
76	asm volatile("vbroadcasti128 %0, %%ymm5" : : "m" (qmul[16]));
77
78	asm volatile("vpsraw $4, %ymm1, %ymm3");
79	asm volatile("vpsraw $4, %ymm9, %ymm12");
80	asm volatile("vpand %ymm7, %ymm1, %ymm1");
81	asm volatile("vpand %ymm7, %ymm9, %ymm9");
82	asm volatile("vpand %ymm7, %ymm3, %ymm3");
83	asm volatile("vpand %ymm7, %ymm12, %ymm12");
84	asm volatile("vpshufb %ymm9, %ymm4, %ymm14");
85	asm volatile("vpshufb %ymm1, %ymm4, %ymm4");
86	asm volatile("vpshufb %ymm12, %ymm5, %ymm15");
87	asm volatile("vpshufb %ymm3, %ymm5, %ymm5");
88	asm volatile("vpxor %ymm14, %ymm15, %ymm15");
89	asm volatile("vpxor %ymm4, %ymm5, %ymm5");
90
91	/*
92	* 5 = qx[0]
93	* 15 = qx[32]
94	*/
95
96	asm volatile("vbroadcasti128 %0, %%ymm4" : : "m" (pbmul[0]));
97	asm volatile("vbroadcasti128 %0, %%ymm1" : : "m" (pbmul[16]));
98	asm volatile("vpsraw $4, %ymm0, %ymm2");
99	asm volatile("vpsraw $4, %ymm8, %ymm6");
100	asm volatile("vpand %ymm7, %ymm0, %ymm3");
101	asm volatile("vpand %ymm7, %ymm8, %ymm14");
102	asm volatile("vpand %ymm7, %ymm2, %ymm2");
103	asm volatile("vpand %ymm7, %ymm6, %ymm6");
104	asm volatile("vpshufb %ymm14, %ymm4, %ymm12");
105	asm volatile("vpshufb %ymm3, %ymm4, %ymm4");
106	asm volatile("vpshufb %ymm6, %ymm1, %ymm13");
107	asm volatile("vpshufb %ymm2, %ymm1, %ymm1");
108	asm volatile("vpxor %ymm4, %ymm1, %ymm1");
109	asm volatile("vpxor %ymm12, %ymm13, %ymm13");
110
111	/*
112	* 1 = pbmul[px[0]]
113	* 13 = pbmul[px[32]]
114	*/
115	asm volatile("vpxor %ymm5, %ymm1, %ymm1");
116	asm volatile("vpxor %ymm15, %ymm13, %ymm13");
117
118	/*
119	* 1 = db = DQ
120	* 13 = db[32] = DQ[32]
121	*/
122	asm volatile("vmovdqa %%ymm1, %0" : "=m" (dq[0]));
123	asm volatile("vmovdqa %%ymm13,%0" : "=m" (dq[32]));
124	asm volatile("vpxor %ymm1, %ymm0, %ymm0");
125	asm volatile("vpxor %ymm13, %ymm8, %ymm8");
126
127	asm volatile("vmovdqa %%ymm0, %0" : "=m" (dp[0]));
128	asm volatile("vmovdqa %%ymm8, %0" : "=m" (dp[32]));
129
130	bytes -= 64;
131	p += 64;
132	q += 64;
133	dp += 64;
134	dq += 64;
135	#else
136	asm volatile("vmovdqa %0, %%ymm1" : : "m" (*q));
137	asm volatile("vmovdqa %0, %%ymm0" : : "m" (*p));
138	asm volatile("vpxor %0, %%ymm1, %%ymm1" : : "m" (*dq));
139	asm volatile("vpxor %0, %%ymm0, %%ymm0" : : "m" (*dp));
140
141	/* 1 = dq ^ q; 0 = dp ^ p */
142
143	asm volatile("vbroadcasti128 %0, %%ymm4" : : "m" (qmul[0]));
144	asm volatile("vbroadcasti128 %0, %%ymm5" : : "m" (qmul[16]));
145
146	/*
147	* 1 = dq ^ q
148	* 3 = dq ^ p >> 4
149	*/
150	asm volatile("vpsraw $4, %ymm1, %ymm3");
151	asm volatile("vpand %ymm7, %ymm1, %ymm1");
152	asm volatile("vpand %ymm7, %ymm3, %ymm3");
153	asm volatile("vpshufb %ymm1, %ymm4, %ymm4");
154	asm volatile("vpshufb %ymm3, %ymm5, %ymm5");
155	asm volatile("vpxor %ymm4, %ymm5, %ymm5");
156
157	/* 5 = qx */
158
159	asm volatile("vbroadcasti128 %0, %%ymm4" : : "m" (pbmul[0]));
160	asm volatile("vbroadcasti128 %0, %%ymm1" : : "m" (pbmul[16]));
161
162	asm volatile("vpsraw $4, %ymm0, %ymm2");
163	asm volatile("vpand %ymm7, %ymm0, %ymm3");
164	asm volatile("vpand %ymm7, %ymm2, %ymm2");
165	asm volatile("vpshufb %ymm3, %ymm4, %ymm4");
166	asm volatile("vpshufb %ymm2, %ymm1, %ymm1");
167	asm volatile("vpxor %ymm4, %ymm1, %ymm1");
168
169	/* 1 = pbmul[px] */
170	asm volatile("vpxor %ymm5, %ymm1, %ymm1");
171	/* 1 = db = DQ */
172	asm volatile("vmovdqa %%ymm1, %0" : "=m" (dq[0]));
173
174	asm volatile("vpxor %ymm1, %ymm0, %ymm0");
175	asm volatile("vmovdqa %%ymm0, %0" : "=m" (dp[0]));
176
177	bytes -= 32;
178	p += 32;
179	q += 32;
180	dp += 32;
181	dq += 32;
182	#endif
183	}
184
185	kernel_fpu_end();
186	}
187
188	static void raid6_datap_recov_avx2(int disks, size_t bytes, int faila,
189	void **ptrs)
190	{
191	u8 p, q, *dq;
192	const u8 qmul; / Q multiplier table */
193	const u8 x0f = 0x0f;
194
195	p = (u8 *)ptrs[disks-2];
196	q = (u8 *)ptrs[disks-1];
197
198	/* Compute syndrome with zero for the missing data page
199	Use the dead data page as temporary storage for delta q */
200	dq = (u8 *)ptrs[faila];
201	ptrs[faila] = (void *)raid6_empty_zero_page;
202	ptrs[disks-1] = dq;
203
204	raid6_call.gen_syndrome(disks, bytes, ptrs);
205
206	/* Restore pointer table */
207	ptrs[faila] = dq;
208	ptrs[disks-1] = q;
209
210	/* Now, pick the proper data tables */
211	qmul = raid6_vgfmul[raid6_gfinv[raid6_gfexp[faila]]];
212
213	kernel_fpu_begin();
214
215	asm volatile("vpbroadcastb %0, %%ymm7" : : "m" (x0f));
216
217	while (bytes) {
218	#ifdef CONFIG_X86_64
219	asm volatile("vmovdqa %0, %%ymm3" : : "m" (dq[0]));
220	asm volatile("vmovdqa %0, %%ymm8" : : "m" (dq[32]));
221	asm volatile("vpxor %0, %%ymm3, %%ymm3" : : "m" (q[0]));
222	asm volatile("vpxor %0, %%ymm8, %%ymm8" : : "m" (q[32]));
223
224	/*
225	* 3 = q[0] ^ dq[0]
226	* 8 = q[32] ^ dq[32]
227	*/
228	asm volatile("vbroadcasti128 %0, %%ymm0" : : "m" (qmul[0]));
229	asm volatile("vmovapd %ymm0, %ymm13");
230	asm volatile("vbroadcasti128 %0, %%ymm1" : : "m" (qmul[16]));
231	asm volatile("vmovapd %ymm1, %ymm14");
232
233	asm volatile("vpsraw $4, %ymm3, %ymm6");
234	asm volatile("vpsraw $4, %ymm8, %ymm12");
235	asm volatile("vpand %ymm7, %ymm3, %ymm3");
236	asm volatile("vpand %ymm7, %ymm8, %ymm8");
237	asm volatile("vpand %ymm7, %ymm6, %ymm6");
238	asm volatile("vpand %ymm7, %ymm12, %ymm12");
239	asm volatile("vpshufb %ymm3, %ymm0, %ymm0");
240	asm volatile("vpshufb %ymm8, %ymm13, %ymm13");
241	asm volatile("vpshufb %ymm6, %ymm1, %ymm1");
242	asm volatile("vpshufb %ymm12, %ymm14, %ymm14");
243	asm volatile("vpxor %ymm0, %ymm1, %ymm1");
244	asm volatile("vpxor %ymm13, %ymm14, %ymm14");
245
246	/*
247	* 1 = qmul[q[0] ^ dq[0]]
248	* 14 = qmul[q[32] ^ dq[32]]
249	*/
250	asm volatile("vmovdqa %0, %%ymm2" : : "m" (p[0]));
251	asm volatile("vmovdqa %0, %%ymm12" : : "m" (p[32]));
252	asm volatile("vpxor %ymm1, %ymm2, %ymm2");
253	asm volatile("vpxor %ymm14, %ymm12, %ymm12");
254
255	/*
256	* 2 = p[0] ^ qmul[q[0] ^ dq[0]]
257	* 12 = p[32] ^ qmul[q[32] ^ dq[32]]
258	*/
259
260	asm volatile("vmovdqa %%ymm1, %0" : "=m" (dq[0]));
261	asm volatile("vmovdqa %%ymm14, %0" : "=m" (dq[32]));
262	asm volatile("vmovdqa %%ymm2, %0" : "=m" (p[0]));
263	asm volatile("vmovdqa %%ymm12,%0" : "=m" (p[32]));
264
265	bytes -= 64;
266	p += 64;
267	q += 64;
268	dq += 64;
269	#else
270	asm volatile("vmovdqa %0, %%ymm3" : : "m" (dq[0]));
271	asm volatile("vpxor %0, %%ymm3, %%ymm3" : : "m" (q[0]));
272
273	/* 3 = q ^ dq */
274
275	asm volatile("vbroadcasti128 %0, %%ymm0" : : "m" (qmul[0]));
276	asm volatile("vbroadcasti128 %0, %%ymm1" : : "m" (qmul[16]));
277
278	asm volatile("vpsraw $4, %ymm3, %ymm6");
279	asm volatile("vpand %ymm7, %ymm3, %ymm3");
280	asm volatile("vpand %ymm7, %ymm6, %ymm6");
281	asm volatile("vpshufb %ymm3, %ymm0, %ymm0");
282	asm volatile("vpshufb %ymm6, %ymm1, %ymm1");
283	asm volatile("vpxor %ymm0, %ymm1, %ymm1");
284
285	/* 1 = qmul[q ^ dq] */
286
287	asm volatile("vmovdqa %0, %%ymm2" : : "m" (p[0]));
288	asm volatile("vpxor %ymm1, %ymm2, %ymm2");
289
290	/* 2 = p ^ qmul[q ^ dq] */
291
292	asm volatile("vmovdqa %%ymm1, %0" : "=m" (dq[0]));
293	asm volatile("vmovdqa %%ymm2, %0" : "=m" (p[0]));
294
295	bytes -= 32;
296	p += 32;
297	q += 32;
298	dq += 32;
299	#endif
300	}
301
302	kernel_fpu_end();
303	}
304
305	const struct raid6_recov_calls raid6_recov_avx2 = {
306	.data2 = raid6_2data_recov_avx2,
307	.datap = raid6_datap_recov_avx2,
308	.valid = raid6_has_avx2,
309	#ifdef CONFIG_X86_64
310	.name = "avx2x2",
311	#else
312	.name = "avx2x1",
313	#endif
314	.priority = 2,
315	};
316
317	#else
318	#warning "your version of binutils lacks AVX2 support"
319	#endif