1 // SPDX-License-Identifier: CDDL-1.0
2 /*
3 * CDDL HEADER START
4 *
5 * The contents of this file are subject to the terms of the
6 * Common Development and Distribution License (the "License").
7 * You may not use this file except in compliance with the License.
8 *
9 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10 * or https://opensource.org/licenses/CDDL-1.0.
11 * See the License for the specific language governing permissions
12 * and limitations under the License.
13 *
14 * When distributing Covered Code, include this CDDL HEADER in each
15 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16 * If applicable, add the following below this CDDL HEADER, with the
17 * fields enclosed by brackets "[]" replaced with your own identifying
18 * information: Portions Copyright [yyyy] [name of copyright owner]
19 *
20 * CDDL HEADER END
21 */
22 /*
23 * Copyright (C) 2016 Romain Dolbeau. All rights reserved.
24 * Copyright (C) 2016 Gvozden Nešković. All rights reserved.
25 */
26
27 #include <sys/isa_defs.h>
28
29 #if defined(__x86_64) && defined(HAVE_AVX512F)
30
31 #include <sys/types.h>
32 #include <sys/simd.h>
33 #include <sys/debug.h>
34
35 #ifdef __linux__
36 #define __asm __asm__ __volatile__
37 #endif
38
39 #define _REG_CNT(_0, _1, _2, _3, _4, _5, _6, _7, N, ...) N
40 #define REG_CNT(r...) _REG_CNT(r, 8, 7, 6, 5, 4, 3, 2, 1)
41
42 #define VR0_(REG, ...) "zmm"#REG
43 #define VR1_(_1, REG, ...) "zmm"#REG
44 #define VR2_(_1, _2, REG, ...) "zmm"#REG
45 #define VR3_(_1, _2, _3, REG, ...) "zmm"#REG
46 #define VR4_(_1, _2, _3, _4, REG, ...) "zmm"#REG
47 #define VR5_(_1, _2, _3, _4, _5, REG, ...) "zmm"#REG
48 #define VR6_(_1, _2, _3, _4, _5, _6, REG, ...) "zmm"#REG
49 #define VR7_(_1, _2, _3, _4, _5, _6, _7, REG, ...) "zmm"#REG
50
51 #define VR0(r...) VR0_(r)
52 #define VR1(r...) VR1_(r)
53 #define VR2(r...) VR2_(r, 1)
54 #define VR3(r...) VR3_(r, 1, 2)
55 #define VR4(r...) VR4_(r, 1, 2)
56 #define VR5(r...) VR5_(r, 1, 2, 3)
57 #define VR6(r...) VR6_(r, 1, 2, 3, 4)
58 #define VR7(r...) VR7_(r, 1, 2, 3, 4, 5)
59
60 #define VRy0_(REG, ...) "ymm"#REG
61 #define VRy1_(_1, REG, ...) "ymm"#REG
62 #define VRy2_(_1, _2, REG, ...) "ymm"#REG
63 #define VRy3_(_1, _2, _3, REG, ...) "ymm"#REG
64 #define VRy4_(_1, _2, _3, _4, REG, ...) "ymm"#REG
65 #define VRy5_(_1, _2, _3, _4, _5, REG, ...) "ymm"#REG
66 #define VRy6_(_1, _2, _3, _4, _5, _6, REG, ...) "ymm"#REG
67 #define VRy7_(_1, _2, _3, _4, _5, _6, _7, REG, ...) "ymm"#REG
68
69 #define VRy0(r...) VRy0_(r)
70 #define VRy1(r...) VRy1_(r)
71 #define VRy2(r...) VRy2_(r, 1)
72 #define VRy3(r...) VRy3_(r, 1, 2)
73 #define VRy4(r...) VRy4_(r, 1, 2)
74 #define VRy5(r...) VRy5_(r, 1, 2, 3)
75 #define VRy6(r...) VRy6_(r, 1, 2, 3, 4)
76 #define VRy7(r...) VRy7_(r, 1, 2, 3, 4, 5)
77
78 #define R_01(REG1, REG2, ...) REG1, REG2
79 #define _R_23(_0, _1, REG2, REG3, ...) REG2, REG3
80 #define R_23(REG...) _R_23(REG, 1, 2, 3)
81
82 #define ELEM_SIZE 64
83
84 typedef struct v {
85 uint8_t b[ELEM_SIZE] __attribute__((aligned(ELEM_SIZE)));
86 } v_t;
87
88
89 #define XOR_ACC(src, r...) \
90 { \
91 switch (REG_CNT(r)) { \
92 case 4: \
93 __asm( \
94 "vpxorq 0x00(%[SRC]), %%" VR0(r)", %%" VR0(r) "\n" \
95 "vpxorq 0x40(%[SRC]), %%" VR1(r)", %%" VR1(r) "\n" \
96 "vpxorq 0x80(%[SRC]), %%" VR2(r)", %%" VR2(r) "\n" \
97 "vpxorq 0xc0(%[SRC]), %%" VR3(r)", %%" VR3(r) "\n" \
98 : : [SRC] "r" (src)); \
99 break; \
100 } \
101 }
102
103 #define XOR(r...) \
104 { \
105 switch (REG_CNT(r)) { \
106 case 8: \
107 __asm( \
108 "vpxorq %" VR0(r) ", %" VR4(r)", %" VR4(r) "\n" \
109 "vpxorq %" VR1(r) ", %" VR5(r)", %" VR5(r) "\n" \
110 "vpxorq %" VR2(r) ", %" VR6(r)", %" VR6(r) "\n" \
111 "vpxorq %" VR3(r) ", %" VR7(r)", %" VR7(r)); \
112 break; \
113 case 4: \
114 __asm( \
115 "vpxorq %" VR0(r) ", %" VR2(r)", %" VR2(r) "\n" \
116 "vpxorq %" VR1(r) ", %" VR3(r)", %" VR3(r)); \
117 break; \
118 } \
119 }
120
121
122 #define ZERO(r...) XOR(r, r)
123
124
125 #define COPY(r...) \
126 { \
127 switch (REG_CNT(r)) { \
128 case 8: \
129 __asm( \
130 "vmovdqa64 %" VR0(r) ", %" VR4(r) "\n" \
131 "vmovdqa64 %" VR1(r) ", %" VR5(r) "\n" \
132 "vmovdqa64 %" VR2(r) ", %" VR6(r) "\n" \
133 "vmovdqa64 %" VR3(r) ", %" VR7(r)); \
134 break; \
135 case 4: \
136 __asm( \
137 "vmovdqa64 %" VR0(r) ", %" VR2(r) "\n" \
138 "vmovdqa64 %" VR1(r) ", %" VR3(r)); \
139 break; \
140 } \
141 }
142
143 #define LOAD(src, r...) \
144 { \
145 switch (REG_CNT(r)) { \
146 case 4: \
147 __asm( \
148 "vmovdqa64 0x00(%[SRC]), %%" VR0(r) "\n" \
149 "vmovdqa64 0x40(%[SRC]), %%" VR1(r) "\n" \
150 "vmovdqa64 0x80(%[SRC]), %%" VR2(r) "\n" \
151 "vmovdqa64 0xc0(%[SRC]), %%" VR3(r) "\n" \
152 : : [SRC] "r" (src)); \
153 break; \
154 } \
155 }
156
157 #define STORE(dst, r...) \
158 { \
159 switch (REG_CNT(r)) { \
160 case 4: \
161 __asm( \
162 "vmovdqa64 %%" VR0(r) ", 0x00(%[DST])\n" \
163 "vmovdqa64 %%" VR1(r) ", 0x40(%[DST])\n" \
164 "vmovdqa64 %%" VR2(r) ", 0x80(%[DST])\n" \
165 "vmovdqa64 %%" VR3(r) ", 0xc0(%[DST])\n" \
166 : : [DST] "r" (dst)); \
167 break; \
168 } \
169 }
170
171 #define MUL2_SETUP() \
172 { \
173 __asm("vmovq %0, %%xmm31" :: "r"(0x1d1d1d1d1d1d1d1d)); \
174 __asm("vpbroadcastq %xmm31, %zmm31"); \
175 __asm("vmovq %0, %%xmm30" :: "r"(0x8080808080808080)); \
176 __asm("vpbroadcastq %xmm30, %zmm30"); \
177 __asm("vmovq %0, %%xmm29" :: "r"(0xfefefefefefefefe)); \
178 __asm("vpbroadcastq %xmm29, %zmm29"); \
179 }
180
181 #define _MUL2(r...) \
182 { \
183 switch (REG_CNT(r)) { \
184 case 2: \
185 __asm( \
186 "vpandq %" VR0(r)", %zmm30, %zmm26\n" \
187 "vpandq %" VR1(r)", %zmm30, %zmm25\n" \
188 "vpsrlq $7, %zmm26, %zmm28\n" \
189 "vpsrlq $7, %zmm25, %zmm27\n" \
190 "vpsllq $1, %zmm26, %zmm26\n" \
191 "vpsllq $1, %zmm25, %zmm25\n" \
192 "vpsubq %zmm28, %zmm26, %zmm26\n" \
193 "vpsubq %zmm27, %zmm25, %zmm25\n" \
194 "vpsllq $1, %" VR0(r)", %" VR0(r) "\n" \
195 "vpsllq $1, %" VR1(r)", %" VR1(r) "\n" \
196 "vpandq %zmm26, %zmm31, %zmm26\n" \
197 "vpandq %zmm25, %zmm31, %zmm25\n" \
198 "vpternlogd $0x6c,%zmm29, %zmm26, %" VR0(r) "\n" \
199 "vpternlogd $0x6c,%zmm29, %zmm25, %" VR1(r)); \
200 break; \
201 default: \
202 VERIFY(0); \
203 } \
204 }
205
206 #define MUL2(r...) \
207 { \
208 switch (REG_CNT(r)) { \
209 case 4: \
210 _MUL2(R_01(r)); \
211 _MUL2(R_23(r)); \
212 break; \
213 case 2: \
214 _MUL2(r); \
215 break; \
216 } \
217 }
218
219 #define MUL4(r...) \
220 { \
221 MUL2(r); \
222 MUL2(r); \
223 }
224
225
226 /* General multiplication by adding powers of two */
227
228 #define _mul_x2_in 21, 22
229 #define _mul_x2_acc 23, 24
230
231 #define _MUL_PARAM(x, in, acc) \
232 { \
233 if (x & 0x01) { COPY(in, acc); } else { ZERO(acc); } \
234 if (x & 0xfe) { MUL2(in); } \
235 if (x & 0x02) { XOR(in, acc); } \
236 if (x & 0xfc) { MUL2(in); } \
237 if (x & 0x04) { XOR(in, acc); } \
238 if (x & 0xf8) { MUL2(in); } \
239 if (x & 0x08) { XOR(in, acc); } \
240 if (x & 0xf0) { MUL2(in); } \
241 if (x & 0x10) { XOR(in, acc); } \
242 if (x & 0xe0) { MUL2(in); } \
243 if (x & 0x20) { XOR(in, acc); } \
244 if (x & 0xc0) { MUL2(in); } \
245 if (x & 0x40) { XOR(in, acc); } \
246 if (x & 0x80) { MUL2(in); XOR(in, acc); } \
247 }
248
249 #define MUL_x2_DEFINE(x) \
250 static void \
251 mul_x2_ ## x(void) { _MUL_PARAM(x, _mul_x2_in, _mul_x2_acc); }
252
253
254 MUL_x2_DEFINE(0); MUL_x2_DEFINE(1); MUL_x2_DEFINE(2); MUL_x2_DEFINE(3);
255 MUL_x2_DEFINE(4); MUL_x2_DEFINE(5); MUL_x2_DEFINE(6); MUL_x2_DEFINE(7);
256 MUL_x2_DEFINE(8); MUL_x2_DEFINE(9); MUL_x2_DEFINE(10); MUL_x2_DEFINE(11);
257 MUL_x2_DEFINE(12); MUL_x2_DEFINE(13); MUL_x2_DEFINE(14); MUL_x2_DEFINE(15);
258 MUL_x2_DEFINE(16); MUL_x2_DEFINE(17); MUL_x2_DEFINE(18); MUL_x2_DEFINE(19);
259 MUL_x2_DEFINE(20); MUL_x2_DEFINE(21); MUL_x2_DEFINE(22); MUL_x2_DEFINE(23);
260 MUL_x2_DEFINE(24); MUL_x2_DEFINE(25); MUL_x2_DEFINE(26); MUL_x2_DEFINE(27);
261 MUL_x2_DEFINE(28); MUL_x2_DEFINE(29); MUL_x2_DEFINE(30); MUL_x2_DEFINE(31);
262 MUL_x2_DEFINE(32); MUL_x2_DEFINE(33); MUL_x2_DEFINE(34); MUL_x2_DEFINE(35);
263 MUL_x2_DEFINE(36); MUL_x2_DEFINE(37); MUL_x2_DEFINE(38); MUL_x2_DEFINE(39);
264 MUL_x2_DEFINE(40); MUL_x2_DEFINE(41); MUL_x2_DEFINE(42); MUL_x2_DEFINE(43);
265 MUL_x2_DEFINE(44); MUL_x2_DEFINE(45); MUL_x2_DEFINE(46); MUL_x2_DEFINE(47);
266 MUL_x2_DEFINE(48); MUL_x2_DEFINE(49); MUL_x2_DEFINE(50); MUL_x2_DEFINE(51);
267 MUL_x2_DEFINE(52); MUL_x2_DEFINE(53); MUL_x2_DEFINE(54); MUL_x2_DEFINE(55);
268 MUL_x2_DEFINE(56); MUL_x2_DEFINE(57); MUL_x2_DEFINE(58); MUL_x2_DEFINE(59);
269 MUL_x2_DEFINE(60); MUL_x2_DEFINE(61); MUL_x2_DEFINE(62); MUL_x2_DEFINE(63);
270 MUL_x2_DEFINE(64); MUL_x2_DEFINE(65); MUL_x2_DEFINE(66); MUL_x2_DEFINE(67);
271 MUL_x2_DEFINE(68); MUL_x2_DEFINE(69); MUL_x2_DEFINE(70); MUL_x2_DEFINE(71);
272 MUL_x2_DEFINE(72); MUL_x2_DEFINE(73); MUL_x2_DEFINE(74); MUL_x2_DEFINE(75);
273 MUL_x2_DEFINE(76); MUL_x2_DEFINE(77); MUL_x2_DEFINE(78); MUL_x2_DEFINE(79);
274 MUL_x2_DEFINE(80); MUL_x2_DEFINE(81); MUL_x2_DEFINE(82); MUL_x2_DEFINE(83);
275 MUL_x2_DEFINE(84); MUL_x2_DEFINE(85); MUL_x2_DEFINE(86); MUL_x2_DEFINE(87);
276 MUL_x2_DEFINE(88); MUL_x2_DEFINE(89); MUL_x2_DEFINE(90); MUL_x2_DEFINE(91);
277 MUL_x2_DEFINE(92); MUL_x2_DEFINE(93); MUL_x2_DEFINE(94); MUL_x2_DEFINE(95);
278 MUL_x2_DEFINE(96); MUL_x2_DEFINE(97); MUL_x2_DEFINE(98); MUL_x2_DEFINE(99);
279 MUL_x2_DEFINE(100); MUL_x2_DEFINE(101); MUL_x2_DEFINE(102); MUL_x2_DEFINE(103);
280 MUL_x2_DEFINE(104); MUL_x2_DEFINE(105); MUL_x2_DEFINE(106); MUL_x2_DEFINE(107);
281 MUL_x2_DEFINE(108); MUL_x2_DEFINE(109); MUL_x2_DEFINE(110); MUL_x2_DEFINE(111);
282 MUL_x2_DEFINE(112); MUL_x2_DEFINE(113); MUL_x2_DEFINE(114); MUL_x2_DEFINE(115);
283 MUL_x2_DEFINE(116); MUL_x2_DEFINE(117); MUL_x2_DEFINE(118); MUL_x2_DEFINE(119);
284 MUL_x2_DEFINE(120); MUL_x2_DEFINE(121); MUL_x2_DEFINE(122); MUL_x2_DEFINE(123);
285 MUL_x2_DEFINE(124); MUL_x2_DEFINE(125); MUL_x2_DEFINE(126); MUL_x2_DEFINE(127);
286 MUL_x2_DEFINE(128); MUL_x2_DEFINE(129); MUL_x2_DEFINE(130); MUL_x2_DEFINE(131);
287 MUL_x2_DEFINE(132); MUL_x2_DEFINE(133); MUL_x2_DEFINE(134); MUL_x2_DEFINE(135);
288 MUL_x2_DEFINE(136); MUL_x2_DEFINE(137); MUL_x2_DEFINE(138); MUL_x2_DEFINE(139);
289 MUL_x2_DEFINE(140); MUL_x2_DEFINE(141); MUL_x2_DEFINE(142); MUL_x2_DEFINE(143);
290 MUL_x2_DEFINE(144); MUL_x2_DEFINE(145); MUL_x2_DEFINE(146); MUL_x2_DEFINE(147);
291 MUL_x2_DEFINE(148); MUL_x2_DEFINE(149); MUL_x2_DEFINE(150); MUL_x2_DEFINE(151);
292 MUL_x2_DEFINE(152); MUL_x2_DEFINE(153); MUL_x2_DEFINE(154); MUL_x2_DEFINE(155);
293 MUL_x2_DEFINE(156); MUL_x2_DEFINE(157); MUL_x2_DEFINE(158); MUL_x2_DEFINE(159);
294 MUL_x2_DEFINE(160); MUL_x2_DEFINE(161); MUL_x2_DEFINE(162); MUL_x2_DEFINE(163);
295 MUL_x2_DEFINE(164); MUL_x2_DEFINE(165); MUL_x2_DEFINE(166); MUL_x2_DEFINE(167);
296 MUL_x2_DEFINE(168); MUL_x2_DEFINE(169); MUL_x2_DEFINE(170); MUL_x2_DEFINE(171);
297 MUL_x2_DEFINE(172); MUL_x2_DEFINE(173); MUL_x2_DEFINE(174); MUL_x2_DEFINE(175);
298 MUL_x2_DEFINE(176); MUL_x2_DEFINE(177); MUL_x2_DEFINE(178); MUL_x2_DEFINE(179);
299 MUL_x2_DEFINE(180); MUL_x2_DEFINE(181); MUL_x2_DEFINE(182); MUL_x2_DEFINE(183);
300 MUL_x2_DEFINE(184); MUL_x2_DEFINE(185); MUL_x2_DEFINE(186); MUL_x2_DEFINE(187);
301 MUL_x2_DEFINE(188); MUL_x2_DEFINE(189); MUL_x2_DEFINE(190); MUL_x2_DEFINE(191);
302 MUL_x2_DEFINE(192); MUL_x2_DEFINE(193); MUL_x2_DEFINE(194); MUL_x2_DEFINE(195);
303 MUL_x2_DEFINE(196); MUL_x2_DEFINE(197); MUL_x2_DEFINE(198); MUL_x2_DEFINE(199);
304 MUL_x2_DEFINE(200); MUL_x2_DEFINE(201); MUL_x2_DEFINE(202); MUL_x2_DEFINE(203);
305 MUL_x2_DEFINE(204); MUL_x2_DEFINE(205); MUL_x2_DEFINE(206); MUL_x2_DEFINE(207);
306 MUL_x2_DEFINE(208); MUL_x2_DEFINE(209); MUL_x2_DEFINE(210); MUL_x2_DEFINE(211);
307 MUL_x2_DEFINE(212); MUL_x2_DEFINE(213); MUL_x2_DEFINE(214); MUL_x2_DEFINE(215);
308 MUL_x2_DEFINE(216); MUL_x2_DEFINE(217); MUL_x2_DEFINE(218); MUL_x2_DEFINE(219);
309 MUL_x2_DEFINE(220); MUL_x2_DEFINE(221); MUL_x2_DEFINE(222); MUL_x2_DEFINE(223);
310 MUL_x2_DEFINE(224); MUL_x2_DEFINE(225); MUL_x2_DEFINE(226); MUL_x2_DEFINE(227);
311 MUL_x2_DEFINE(228); MUL_x2_DEFINE(229); MUL_x2_DEFINE(230); MUL_x2_DEFINE(231);
312 MUL_x2_DEFINE(232); MUL_x2_DEFINE(233); MUL_x2_DEFINE(234); MUL_x2_DEFINE(235);
313 MUL_x2_DEFINE(236); MUL_x2_DEFINE(237); MUL_x2_DEFINE(238); MUL_x2_DEFINE(239);
314 MUL_x2_DEFINE(240); MUL_x2_DEFINE(241); MUL_x2_DEFINE(242); MUL_x2_DEFINE(243);
315 MUL_x2_DEFINE(244); MUL_x2_DEFINE(245); MUL_x2_DEFINE(246); MUL_x2_DEFINE(247);
316 MUL_x2_DEFINE(248); MUL_x2_DEFINE(249); MUL_x2_DEFINE(250); MUL_x2_DEFINE(251);
317 MUL_x2_DEFINE(252); MUL_x2_DEFINE(253); MUL_x2_DEFINE(254); MUL_x2_DEFINE(255);
318
319
320 typedef void (*mul_fn_ptr_t)(void);
321
322 static const mul_fn_ptr_t __attribute__((aligned(256)))
323 gf_x2_mul_fns[256] = {
324 mul_x2_0, mul_x2_1, mul_x2_2, mul_x2_3, mul_x2_4, mul_x2_5,
325 mul_x2_6, mul_x2_7, mul_x2_8, mul_x2_9, mul_x2_10, mul_x2_11,
326 mul_x2_12, mul_x2_13, mul_x2_14, mul_x2_15, mul_x2_16, mul_x2_17,
327 mul_x2_18, mul_x2_19, mul_x2_20, mul_x2_21, mul_x2_22, mul_x2_23,
328 mul_x2_24, mul_x2_25, mul_x2_26, mul_x2_27, mul_x2_28, mul_x2_29,
329 mul_x2_30, mul_x2_31, mul_x2_32, mul_x2_33, mul_x2_34, mul_x2_35,
330 mul_x2_36, mul_x2_37, mul_x2_38, mul_x2_39, mul_x2_40, mul_x2_41,
331 mul_x2_42, mul_x2_43, mul_x2_44, mul_x2_45, mul_x2_46, mul_x2_47,
332 mul_x2_48, mul_x2_49, mul_x2_50, mul_x2_51, mul_x2_52, mul_x2_53,
333 mul_x2_54, mul_x2_55, mul_x2_56, mul_x2_57, mul_x2_58, mul_x2_59,
334 mul_x2_60, mul_x2_61, mul_x2_62, mul_x2_63, mul_x2_64, mul_x2_65,
335 mul_x2_66, mul_x2_67, mul_x2_68, mul_x2_69, mul_x2_70, mul_x2_71,
336 mul_x2_72, mul_x2_73, mul_x2_74, mul_x2_75, mul_x2_76, mul_x2_77,
337 mul_x2_78, mul_x2_79, mul_x2_80, mul_x2_81, mul_x2_82, mul_x2_83,
338 mul_x2_84, mul_x2_85, mul_x2_86, mul_x2_87, mul_x2_88, mul_x2_89,
339 mul_x2_90, mul_x2_91, mul_x2_92, mul_x2_93, mul_x2_94, mul_x2_95,
340 mul_x2_96, mul_x2_97, mul_x2_98, mul_x2_99, mul_x2_100, mul_x2_101,
341 mul_x2_102, mul_x2_103, mul_x2_104, mul_x2_105, mul_x2_106, mul_x2_107,
342 mul_x2_108, mul_x2_109, mul_x2_110, mul_x2_111, mul_x2_112, mul_x2_113,
343 mul_x2_114, mul_x2_115, mul_x2_116, mul_x2_117, mul_x2_118, mul_x2_119,
344 mul_x2_120, mul_x2_121, mul_x2_122, mul_x2_123, mul_x2_124, mul_x2_125,
345 mul_x2_126, mul_x2_127, mul_x2_128, mul_x2_129, mul_x2_130, mul_x2_131,
346 mul_x2_132, mul_x2_133, mul_x2_134, mul_x2_135, mul_x2_136, mul_x2_137,
347 mul_x2_138, mul_x2_139, mul_x2_140, mul_x2_141, mul_x2_142, mul_x2_143,
348 mul_x2_144, mul_x2_145, mul_x2_146, mul_x2_147, mul_x2_148, mul_x2_149,
349 mul_x2_150, mul_x2_151, mul_x2_152, mul_x2_153, mul_x2_154, mul_x2_155,
350 mul_x2_156, mul_x2_157, mul_x2_158, mul_x2_159, mul_x2_160, mul_x2_161,
351 mul_x2_162, mul_x2_163, mul_x2_164, mul_x2_165, mul_x2_166, mul_x2_167,
352 mul_x2_168, mul_x2_169, mul_x2_170, mul_x2_171, mul_x2_172, mul_x2_173,
353 mul_x2_174, mul_x2_175, mul_x2_176, mul_x2_177, mul_x2_178, mul_x2_179,
354 mul_x2_180, mul_x2_181, mul_x2_182, mul_x2_183, mul_x2_184, mul_x2_185,
355 mul_x2_186, mul_x2_187, mul_x2_188, mul_x2_189, mul_x2_190, mul_x2_191,
356 mul_x2_192, mul_x2_193, mul_x2_194, mul_x2_195, mul_x2_196, mul_x2_197,
357 mul_x2_198, mul_x2_199, mul_x2_200, mul_x2_201, mul_x2_202, mul_x2_203,
358 mul_x2_204, mul_x2_205, mul_x2_206, mul_x2_207, mul_x2_208, mul_x2_209,
359 mul_x2_210, mul_x2_211, mul_x2_212, mul_x2_213, mul_x2_214, mul_x2_215,
360 mul_x2_216, mul_x2_217, mul_x2_218, mul_x2_219, mul_x2_220, mul_x2_221,
361 mul_x2_222, mul_x2_223, mul_x2_224, mul_x2_225, mul_x2_226, mul_x2_227,
362 mul_x2_228, mul_x2_229, mul_x2_230, mul_x2_231, mul_x2_232, mul_x2_233,
363 mul_x2_234, mul_x2_235, mul_x2_236, mul_x2_237, mul_x2_238, mul_x2_239,
364 mul_x2_240, mul_x2_241, mul_x2_242, mul_x2_243, mul_x2_244, mul_x2_245,
365 mul_x2_246, mul_x2_247, mul_x2_248, mul_x2_249, mul_x2_250, mul_x2_251,
366 mul_x2_252, mul_x2_253, mul_x2_254, mul_x2_255
367 };
368
369 #define MUL(c, r...) \
370 { \
371 switch (REG_CNT(r)) { \
372 case 4: \
373 COPY(R_01(r), _mul_x2_in); \
374 gf_x2_mul_fns[c](); \
375 COPY(_mul_x2_acc, R_01(r)); \
376 COPY(R_23(r), _mul_x2_in); \
377 gf_x2_mul_fns[c](); \
378 COPY(_mul_x2_acc, R_23(r)); \
379 break; \
380 default: \
381 VERIFY(0); \
382 } \
383 }
384
385
386 #define raidz_math_begin() kfpu_begin()
387 #define raidz_math_end() kfpu_end()
388
389
390 #define SYN_STRIDE 4
391
392 #define ZERO_STRIDE 4
393 #define ZERO_DEFINE() {}
394 #define ZERO_D 0, 1, 2, 3
395
396 #define COPY_STRIDE 4
397 #define COPY_DEFINE() {}
398 #define COPY_D 0, 1, 2, 3
399
400 #define ADD_STRIDE 4
401 #define ADD_DEFINE() {}
402 #define ADD_D 0, 1, 2, 3
403
404 #define MUL_STRIDE 4
405 #define MUL_DEFINE() MUL2_SETUP()
406 #define MUL_D 0, 1, 2, 3
407
408 #define GEN_P_STRIDE 4
409 #define GEN_P_DEFINE() {}
410 #define GEN_P_P 0, 1, 2, 3
411
412 #define GEN_PQ_STRIDE 4
413 #define GEN_PQ_DEFINE() {}
414 #define GEN_PQ_D 0, 1, 2, 3
415 #define GEN_PQ_C 4, 5, 6, 7
416
417 #define GEN_PQR_STRIDE 4
418 #define GEN_PQR_DEFINE() {}
419 #define GEN_PQR_D 0, 1, 2, 3
420 #define GEN_PQR_C 4, 5, 6, 7
421
422 #define SYN_Q_DEFINE() {}
423 #define SYN_Q_D 0, 1, 2, 3
424 #define SYN_Q_X 4, 5, 6, 7
425
426 #define SYN_R_DEFINE() {}
427 #define SYN_R_D 0, 1, 2, 3
428 #define SYN_R_X 4, 5, 6, 7
429
430 #define SYN_PQ_DEFINE() {}
431 #define SYN_PQ_D 0, 1, 2, 3
432 #define SYN_PQ_X 4, 5, 6, 7
433
434 #define REC_PQ_STRIDE 4
435 #define REC_PQ_DEFINE() MUL2_SETUP()
436 #define REC_PQ_X 0, 1, 2, 3
437 #define REC_PQ_Y 4, 5, 6, 7
438 #define REC_PQ_T 8, 9, 10, 11
439
440 #define SYN_PR_DEFINE() {}
441 #define SYN_PR_D 0, 1, 2, 3
442 #define SYN_PR_X 4, 5, 6, 7
443
444 #define REC_PR_STRIDE 4
445 #define REC_PR_DEFINE() MUL2_SETUP()
446 #define REC_PR_X 0, 1, 2, 3
447 #define REC_PR_Y 4, 5, 6, 7
448 #define REC_PR_T 8, 9, 10, 11
449
450 #define SYN_QR_DEFINE() {}
451 #define SYN_QR_D 0, 1, 2, 3
452 #define SYN_QR_X 4, 5, 6, 7
453
454 #define REC_QR_STRIDE 4
455 #define REC_QR_DEFINE() MUL2_SETUP()
456 #define REC_QR_X 0, 1, 2, 3
457 #define REC_QR_Y 4, 5, 6, 7
458 #define REC_QR_T 8, 9, 10, 11
459
460 #define SYN_PQR_DEFINE() {}
461 #define SYN_PQR_D 0, 1, 2, 3
462 #define SYN_PQR_X 4, 5, 6, 7
463
464 #define REC_PQR_STRIDE 4
465 #define REC_PQR_DEFINE() MUL2_SETUP()
466 #define REC_PQR_X 0, 1, 2, 3
467 #define REC_PQR_Y 4, 5, 6, 7
468 #define REC_PQR_Z 8, 9, 10, 11
469 #define REC_PQR_XS 12, 13, 14, 15
470 #define REC_PQR_YS 16, 17, 18, 19
471
472
473 #include <sys/vdev_raidz_impl.h>
474 #include "vdev_raidz_math_impl.h"
475
476 DEFINE_GEN_METHODS(avx512f);
477 DEFINE_REC_METHODS(avx512f);
478
479 static boolean_t
raidz_will_avx512f_work(void)480 raidz_will_avx512f_work(void)
481 {
482 return (kfpu_allowed() && zfs_avx_available() &&
483 zfs_avx2_available() && zfs_avx512f_available());
484 }
485
486 const raidz_impl_ops_t vdev_raidz_avx512f_impl = {
487 .init = NULL,
488 .fini = NULL,
489 .gen = RAIDZ_GEN_METHODS(avx512f),
490 .rec = RAIDZ_REC_METHODS(avx512f),
491 .is_supported = &raidz_will_avx512f_work,
492 .name = "avx512f"
493 };
494
495 #endif /* defined(__x86_64) && defined(HAVE_AVX512F) */
496