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 Gvozden Nešković. All rights reserved. 24 */ 25 26 #include <sys/isa_defs.h> 27 28 #if defined(__x86_64) && defined(HAVE_SSE2) 29 30 #include <sys/types.h> 31 #include <sys/simd.h> 32 #include <sys/debug.h> 33 34 #ifdef __linux__ 35 #define __asm __asm__ __volatile__ 36 #endif 37 38 #define _REG_CNT(_0, _1, _2, _3, _4, _5, _6, _7, N, ...) N 39 #define REG_CNT(r...) _REG_CNT(r, 8, 7, 6, 5, 4, 3, 2, 1) 40 41 #define VR0_(REG, ...) "xmm"#REG 42 #define VR1_(_1, REG, ...) "xmm"#REG 43 #define VR2_(_1, _2, REG, ...) "xmm"#REG 44 #define VR3_(_1, _2, _3, REG, ...) "xmm"#REG 45 #define VR4_(_1, _2, _3, _4, REG, ...) "xmm"#REG 46 #define VR5_(_1, _2, _3, _4, _5, REG, ...) "xmm"#REG 47 #define VR6_(_1, _2, _3, _4, _5, _6, REG, ...) "xmm"#REG 48 #define VR7_(_1, _2, _3, _4, _5, _6, _7, REG, ...) "xmm"#REG 49 50 #define VR0(r...) VR0_(r, 1, 2, 3, 4, 5, 6) 51 #define VR1(r...) VR1_(r, 1, 2, 3, 4, 5, 6) 52 #define VR2(r...) VR2_(r, 1, 2, 3, 4, 5, 6) 53 #define VR3(r...) VR3_(r, 1, 2, 3, 4, 5, 6) 54 #define VR4(r...) VR4_(r, 1, 2, 3, 4, 5, 6) 55 #define VR5(r...) VR5_(r, 1, 2, 3, 4, 5, 6) 56 #define VR6(r...) VR6_(r, 1, 2, 3, 4, 5, 6) 57 #define VR7(r...) VR7_(r, 1, 2, 3, 4, 5, 6) 58 59 #define ELEM_SIZE 16 60 61 typedef struct v { 62 uint8_t b[ELEM_SIZE] __attribute__((aligned(ELEM_SIZE))); 63 } v_t; 64 65 #define XOR_ACC(src, r...) \ 66 { \ 67 switch (REG_CNT(r)) { \ 68 case 4: \ 69 __asm( \ 70 "pxor 0x00(%[SRC]), %%" VR0(r) "\n" \ 71 "pxor 0x10(%[SRC]), %%" VR1(r) "\n" \ 72 "pxor 0x20(%[SRC]), %%" VR2(r) "\n" \ 73 "pxor 0x30(%[SRC]), %%" VR3(r) "\n" \ 74 : : [SRC] "r" (src)); \ 75 break; \ 76 case 2: \ 77 __asm( \ 78 "pxor 0x00(%[SRC]), %%" VR0(r) "\n" \ 79 "pxor 0x10(%[SRC]), %%" VR1(r) "\n" \ 80 : : [SRC] "r" (src)); \ 81 break; \ 82 case 1: \ 83 __asm("pxor 0x00(%[SRC]), %%" VR0(r) "\n" \ 84 : : [SRC] "r" (src)); \ 85 break; \ 86 } \ 87 } 88 89 #define XOR(r...) \ 90 { \ 91 switch (REG_CNT(r)) { \ 92 case 8: \ 93 __asm( \ 94 "pxor %" VR0(r) ", %" VR4(r) "\n" \ 95 "pxor %" VR1(r) ", %" VR5(r) "\n" \ 96 "pxor %" VR2(r) ", %" VR6(r) "\n" \ 97 "pxor %" VR3(r) ", %" VR7(r)); \ 98 break; \ 99 case 4: \ 100 __asm( \ 101 "pxor %" VR0(r) ", %" VR2(r) "\n" \ 102 "pxor %" VR1(r) ", %" VR3(r)); \ 103 break; \ 104 case 2: \ 105 __asm( \ 106 "pxor %" VR0(r) ", %" VR1(r)); \ 107 break; \ 108 } \ 109 } 110 111 #define ZERO(r...) XOR(r, r) 112 113 #define COPY(r...) \ 114 { \ 115 switch (REG_CNT(r)) { \ 116 case 8: \ 117 __asm( \ 118 "movdqa %" VR0(r) ", %" VR4(r) "\n" \ 119 "movdqa %" VR1(r) ", %" VR5(r) "\n" \ 120 "movdqa %" VR2(r) ", %" VR6(r) "\n" \ 121 "movdqa %" VR3(r) ", %" VR7(r)); \ 122 break; \ 123 case 4: \ 124 __asm( \ 125 "movdqa %" VR0(r) ", %" VR2(r) "\n" \ 126 "movdqa %" VR1(r) ", %" VR3(r)); \ 127 break; \ 128 case 2: \ 129 __asm( \ 130 "movdqa %" VR0(r) ", %" VR1(r)); \ 131 break; \ 132 default: \ 133 VERIFY(0); \ 134 } \ 135 } 136 137 #define LOAD(src, r...) \ 138 { \ 139 switch (REG_CNT(r)) { \ 140 case 4: \ 141 __asm( \ 142 "movdqa 0x00(%[SRC]), %%" VR0(r) "\n" \ 143 "movdqa 0x10(%[SRC]), %%" VR1(r) "\n" \ 144 "movdqa 0x20(%[SRC]), %%" VR2(r) "\n" \ 145 "movdqa 0x30(%[SRC]), %%" VR3(r) "\n" \ 146 : : [SRC] "r" (src)); \ 147 break; \ 148 case 2: \ 149 __asm( \ 150 "movdqa 0x00(%[SRC]), %%" VR0(r) "\n" \ 151 "movdqa 0x10(%[SRC]), %%" VR1(r) "\n" \ 152 : : [SRC] "r" (src)); \ 153 break; \ 154 case 1: \ 155 __asm( \ 156 "movdqa 0x00(%[SRC]), %%" VR0(r) "\n" \ 157 : : [SRC] "r" (src)); \ 158 break; \ 159 } \ 160 } 161 162 #define STORE(dst, r...) \ 163 { \ 164 switch (REG_CNT(r)) { \ 165 case 4: \ 166 __asm( \ 167 "movdqa %%" VR0(r)", 0x00(%[DST])\n" \ 168 "movdqa %%" VR1(r)", 0x10(%[DST])\n" \ 169 "movdqa %%" VR2(r)", 0x20(%[DST])\n" \ 170 "movdqa %%" VR3(r)", 0x30(%[DST])\n" \ 171 : : [DST] "r" (dst)); \ 172 break; \ 173 case 2: \ 174 __asm( \ 175 "movdqa %%" VR0(r)", 0x00(%[DST])\n" \ 176 "movdqa %%" VR1(r)", 0x10(%[DST])\n" \ 177 : : [DST] "r" (dst)); \ 178 break; \ 179 case 1: \ 180 __asm( \ 181 "movdqa %%" VR0(r)", 0x00(%[DST])\n" \ 182 : : [DST] "r" (dst)); \ 183 break; \ 184 default: \ 185 VERIFY(0); \ 186 } \ 187 } 188 189 #define MUL2_SETUP() \ 190 { \ 191 __asm( \ 192 "movd %[mask], %%xmm15\n" \ 193 "pshufd $0x0, %%xmm15, %%xmm15\n" \ 194 : : [mask] "r" (0x1d1d1d1d)); \ 195 } 196 197 #define _MUL2_x1(a0) \ 198 { \ 199 __asm( \ 200 "pxor %xmm14, %xmm14\n" \ 201 "pcmpgtb %" a0", %xmm14\n" \ 202 "pand %xmm15, %xmm14\n" \ 203 "paddb %" a0", %" a0 "\n" \ 204 "pxor %xmm14, %" a0); \ 205 } 206 207 #define _MUL2_x2(a0, a1) \ 208 { \ 209 __asm( \ 210 "pxor %xmm14, %xmm14\n" \ 211 "pxor %xmm13, %xmm13\n" \ 212 "pcmpgtb %" a0", %xmm14\n" \ 213 "pcmpgtb %" a1", %xmm13\n" \ 214 "pand %xmm15, %xmm14\n" \ 215 "pand %xmm15, %xmm13\n" \ 216 "paddb %" a0", %" a0 "\n" \ 217 "paddb %" a1", %" a1 "\n" \ 218 "pxor %xmm14, %" a0 "\n" \ 219 "pxor %xmm13, %" a1); \ 220 } 221 222 #define MUL2(r...) \ 223 { \ 224 switch (REG_CNT(r)) { \ 225 case 4: \ 226 _MUL2_x2(VR0(r), VR1(r)); \ 227 _MUL2_x2(VR2(r), VR3(r)); \ 228 break; \ 229 case 2: \ 230 _MUL2_x2(VR0(r), VR1(r)); \ 231 break; \ 232 case 1: \ 233 _MUL2_x1(VR0(r)); \ 234 break; \ 235 } \ 236 } 237 238 #define MUL4(r...) \ 239 { \ 240 MUL2(r); \ 241 MUL2(r); \ 242 } 243 244 /* General multiplication by adding powers of two */ 245 246 #define _MUL_PARAM(x, in, acc) \ 247 { \ 248 if (x & 0x01) { COPY(in, acc); } else { ZERO(acc); } \ 249 if (x & 0xfe) { MUL2(in); } \ 250 if (x & 0x02) { XOR(in, acc); } \ 251 if (x & 0xfc) { MUL2(in); } \ 252 if (x & 0x04) { XOR(in, acc); } \ 253 if (x & 0xf8) { MUL2(in); } \ 254 if (x & 0x08) { XOR(in, acc); } \ 255 if (x & 0xf0) { MUL2(in); } \ 256 if (x & 0x10) { XOR(in, acc); } \ 257 if (x & 0xe0) { MUL2(in); } \ 258 if (x & 0x20) { XOR(in, acc); } \ 259 if (x & 0xc0) { MUL2(in); } \ 260 if (x & 0x40) { XOR(in, acc); } \ 261 if (x & 0x80) { MUL2(in); XOR(in, acc); } \ 262 } 263 264 #define _mul_x1_in 11 265 #define _mul_x1_acc 12 266 267 #define MUL_x1_DEFINE(x) \ 268 static void \ 269 mul_x1_ ## x(void) { _MUL_PARAM(x, _mul_x1_in, _mul_x1_acc); } 270 271 #define _mul_x2_in 9, 10 272 #define _mul_x2_acc 11, 12 273 274 #define MUL_x2_DEFINE(x) \ 275 static void \ 276 mul_x2_ ## x(void) { _MUL_PARAM(x, _mul_x2_in, _mul_x2_acc); } 277 278 MUL_x1_DEFINE(0); MUL_x1_DEFINE(1); MUL_x1_DEFINE(2); MUL_x1_DEFINE(3); 279 MUL_x1_DEFINE(4); MUL_x1_DEFINE(5); MUL_x1_DEFINE(6); MUL_x1_DEFINE(7); 280 MUL_x1_DEFINE(8); MUL_x1_DEFINE(9); MUL_x1_DEFINE(10); MUL_x1_DEFINE(11); 281 MUL_x1_DEFINE(12); MUL_x1_DEFINE(13); MUL_x1_DEFINE(14); MUL_x1_DEFINE(15); 282 MUL_x1_DEFINE(16); MUL_x1_DEFINE(17); MUL_x1_DEFINE(18); MUL_x1_DEFINE(19); 283 MUL_x1_DEFINE(20); MUL_x1_DEFINE(21); MUL_x1_DEFINE(22); MUL_x1_DEFINE(23); 284 MUL_x1_DEFINE(24); MUL_x1_DEFINE(25); MUL_x1_DEFINE(26); MUL_x1_DEFINE(27); 285 MUL_x1_DEFINE(28); MUL_x1_DEFINE(29); MUL_x1_DEFINE(30); MUL_x1_DEFINE(31); 286 MUL_x1_DEFINE(32); MUL_x1_DEFINE(33); MUL_x1_DEFINE(34); MUL_x1_DEFINE(35); 287 MUL_x1_DEFINE(36); MUL_x1_DEFINE(37); MUL_x1_DEFINE(38); MUL_x1_DEFINE(39); 288 MUL_x1_DEFINE(40); MUL_x1_DEFINE(41); MUL_x1_DEFINE(42); MUL_x1_DEFINE(43); 289 MUL_x1_DEFINE(44); MUL_x1_DEFINE(45); MUL_x1_DEFINE(46); MUL_x1_DEFINE(47); 290 MUL_x1_DEFINE(48); MUL_x1_DEFINE(49); MUL_x1_DEFINE(50); MUL_x1_DEFINE(51); 291 MUL_x1_DEFINE(52); MUL_x1_DEFINE(53); MUL_x1_DEFINE(54); MUL_x1_DEFINE(55); 292 MUL_x1_DEFINE(56); MUL_x1_DEFINE(57); MUL_x1_DEFINE(58); MUL_x1_DEFINE(59); 293 MUL_x1_DEFINE(60); MUL_x1_DEFINE(61); MUL_x1_DEFINE(62); MUL_x1_DEFINE(63); 294 MUL_x1_DEFINE(64); MUL_x1_DEFINE(65); MUL_x1_DEFINE(66); MUL_x1_DEFINE(67); 295 MUL_x1_DEFINE(68); MUL_x1_DEFINE(69); MUL_x1_DEFINE(70); MUL_x1_DEFINE(71); 296 MUL_x1_DEFINE(72); MUL_x1_DEFINE(73); MUL_x1_DEFINE(74); MUL_x1_DEFINE(75); 297 MUL_x1_DEFINE(76); MUL_x1_DEFINE(77); MUL_x1_DEFINE(78); MUL_x1_DEFINE(79); 298 MUL_x1_DEFINE(80); MUL_x1_DEFINE(81); MUL_x1_DEFINE(82); MUL_x1_DEFINE(83); 299 MUL_x1_DEFINE(84); MUL_x1_DEFINE(85); MUL_x1_DEFINE(86); MUL_x1_DEFINE(87); 300 MUL_x1_DEFINE(88); MUL_x1_DEFINE(89); MUL_x1_DEFINE(90); MUL_x1_DEFINE(91); 301 MUL_x1_DEFINE(92); MUL_x1_DEFINE(93); MUL_x1_DEFINE(94); MUL_x1_DEFINE(95); 302 MUL_x1_DEFINE(96); MUL_x1_DEFINE(97); MUL_x1_DEFINE(98); MUL_x1_DEFINE(99); 303 MUL_x1_DEFINE(100); MUL_x1_DEFINE(101); MUL_x1_DEFINE(102); MUL_x1_DEFINE(103); 304 MUL_x1_DEFINE(104); MUL_x1_DEFINE(105); MUL_x1_DEFINE(106); MUL_x1_DEFINE(107); 305 MUL_x1_DEFINE(108); MUL_x1_DEFINE(109); MUL_x1_DEFINE(110); MUL_x1_DEFINE(111); 306 MUL_x1_DEFINE(112); MUL_x1_DEFINE(113); MUL_x1_DEFINE(114); MUL_x1_DEFINE(115); 307 MUL_x1_DEFINE(116); MUL_x1_DEFINE(117); MUL_x1_DEFINE(118); MUL_x1_DEFINE(119); 308 MUL_x1_DEFINE(120); MUL_x1_DEFINE(121); MUL_x1_DEFINE(122); MUL_x1_DEFINE(123); 309 MUL_x1_DEFINE(124); MUL_x1_DEFINE(125); MUL_x1_DEFINE(126); MUL_x1_DEFINE(127); 310 MUL_x1_DEFINE(128); MUL_x1_DEFINE(129); MUL_x1_DEFINE(130); MUL_x1_DEFINE(131); 311 MUL_x1_DEFINE(132); MUL_x1_DEFINE(133); MUL_x1_DEFINE(134); MUL_x1_DEFINE(135); 312 MUL_x1_DEFINE(136); MUL_x1_DEFINE(137); MUL_x1_DEFINE(138); MUL_x1_DEFINE(139); 313 MUL_x1_DEFINE(140); MUL_x1_DEFINE(141); MUL_x1_DEFINE(142); MUL_x1_DEFINE(143); 314 MUL_x1_DEFINE(144); MUL_x1_DEFINE(145); MUL_x1_DEFINE(146); MUL_x1_DEFINE(147); 315 MUL_x1_DEFINE(148); MUL_x1_DEFINE(149); MUL_x1_DEFINE(150); MUL_x1_DEFINE(151); 316 MUL_x1_DEFINE(152); MUL_x1_DEFINE(153); MUL_x1_DEFINE(154); MUL_x1_DEFINE(155); 317 MUL_x1_DEFINE(156); MUL_x1_DEFINE(157); MUL_x1_DEFINE(158); MUL_x1_DEFINE(159); 318 MUL_x1_DEFINE(160); MUL_x1_DEFINE(161); MUL_x1_DEFINE(162); MUL_x1_DEFINE(163); 319 MUL_x1_DEFINE(164); MUL_x1_DEFINE(165); MUL_x1_DEFINE(166); MUL_x1_DEFINE(167); 320 MUL_x1_DEFINE(168); MUL_x1_DEFINE(169); MUL_x1_DEFINE(170); MUL_x1_DEFINE(171); 321 MUL_x1_DEFINE(172); MUL_x1_DEFINE(173); MUL_x1_DEFINE(174); MUL_x1_DEFINE(175); 322 MUL_x1_DEFINE(176); MUL_x1_DEFINE(177); MUL_x1_DEFINE(178); MUL_x1_DEFINE(179); 323 MUL_x1_DEFINE(180); MUL_x1_DEFINE(181); MUL_x1_DEFINE(182); MUL_x1_DEFINE(183); 324 MUL_x1_DEFINE(184); MUL_x1_DEFINE(185); MUL_x1_DEFINE(186); MUL_x1_DEFINE(187); 325 MUL_x1_DEFINE(188); MUL_x1_DEFINE(189); MUL_x1_DEFINE(190); MUL_x1_DEFINE(191); 326 MUL_x1_DEFINE(192); MUL_x1_DEFINE(193); MUL_x1_DEFINE(194); MUL_x1_DEFINE(195); 327 MUL_x1_DEFINE(196); MUL_x1_DEFINE(197); MUL_x1_DEFINE(198); MUL_x1_DEFINE(199); 328 MUL_x1_DEFINE(200); MUL_x1_DEFINE(201); MUL_x1_DEFINE(202); MUL_x1_DEFINE(203); 329 MUL_x1_DEFINE(204); MUL_x1_DEFINE(205); MUL_x1_DEFINE(206); MUL_x1_DEFINE(207); 330 MUL_x1_DEFINE(208); MUL_x1_DEFINE(209); MUL_x1_DEFINE(210); MUL_x1_DEFINE(211); 331 MUL_x1_DEFINE(212); MUL_x1_DEFINE(213); MUL_x1_DEFINE(214); MUL_x1_DEFINE(215); 332 MUL_x1_DEFINE(216); MUL_x1_DEFINE(217); MUL_x1_DEFINE(218); MUL_x1_DEFINE(219); 333 MUL_x1_DEFINE(220); MUL_x1_DEFINE(221); MUL_x1_DEFINE(222); MUL_x1_DEFINE(223); 334 MUL_x1_DEFINE(224); MUL_x1_DEFINE(225); MUL_x1_DEFINE(226); MUL_x1_DEFINE(227); 335 MUL_x1_DEFINE(228); MUL_x1_DEFINE(229); MUL_x1_DEFINE(230); MUL_x1_DEFINE(231); 336 MUL_x1_DEFINE(232); MUL_x1_DEFINE(233); MUL_x1_DEFINE(234); MUL_x1_DEFINE(235); 337 MUL_x1_DEFINE(236); MUL_x1_DEFINE(237); MUL_x1_DEFINE(238); MUL_x1_DEFINE(239); 338 MUL_x1_DEFINE(240); MUL_x1_DEFINE(241); MUL_x1_DEFINE(242); MUL_x1_DEFINE(243); 339 MUL_x1_DEFINE(244); MUL_x1_DEFINE(245); MUL_x1_DEFINE(246); MUL_x1_DEFINE(247); 340 MUL_x1_DEFINE(248); MUL_x1_DEFINE(249); MUL_x1_DEFINE(250); MUL_x1_DEFINE(251); 341 MUL_x1_DEFINE(252); MUL_x1_DEFINE(253); MUL_x1_DEFINE(254); MUL_x1_DEFINE(255); 342 343 MUL_x2_DEFINE(0); MUL_x2_DEFINE(1); MUL_x2_DEFINE(2); MUL_x2_DEFINE(3); 344 MUL_x2_DEFINE(4); MUL_x2_DEFINE(5); MUL_x2_DEFINE(6); MUL_x2_DEFINE(7); 345 MUL_x2_DEFINE(8); MUL_x2_DEFINE(9); MUL_x2_DEFINE(10); MUL_x2_DEFINE(11); 346 MUL_x2_DEFINE(12); MUL_x2_DEFINE(13); MUL_x2_DEFINE(14); MUL_x2_DEFINE(15); 347 MUL_x2_DEFINE(16); MUL_x2_DEFINE(17); MUL_x2_DEFINE(18); MUL_x2_DEFINE(19); 348 MUL_x2_DEFINE(20); MUL_x2_DEFINE(21); MUL_x2_DEFINE(22); MUL_x2_DEFINE(23); 349 MUL_x2_DEFINE(24); MUL_x2_DEFINE(25); MUL_x2_DEFINE(26); MUL_x2_DEFINE(27); 350 MUL_x2_DEFINE(28); MUL_x2_DEFINE(29); MUL_x2_DEFINE(30); MUL_x2_DEFINE(31); 351 MUL_x2_DEFINE(32); MUL_x2_DEFINE(33); MUL_x2_DEFINE(34); MUL_x2_DEFINE(35); 352 MUL_x2_DEFINE(36); MUL_x2_DEFINE(37); MUL_x2_DEFINE(38); MUL_x2_DEFINE(39); 353 MUL_x2_DEFINE(40); MUL_x2_DEFINE(41); MUL_x2_DEFINE(42); MUL_x2_DEFINE(43); 354 MUL_x2_DEFINE(44); MUL_x2_DEFINE(45); MUL_x2_DEFINE(46); MUL_x2_DEFINE(47); 355 MUL_x2_DEFINE(48); MUL_x2_DEFINE(49); MUL_x2_DEFINE(50); MUL_x2_DEFINE(51); 356 MUL_x2_DEFINE(52); MUL_x2_DEFINE(53); MUL_x2_DEFINE(54); MUL_x2_DEFINE(55); 357 MUL_x2_DEFINE(56); MUL_x2_DEFINE(57); MUL_x2_DEFINE(58); MUL_x2_DEFINE(59); 358 MUL_x2_DEFINE(60); MUL_x2_DEFINE(61); MUL_x2_DEFINE(62); MUL_x2_DEFINE(63); 359 MUL_x2_DEFINE(64); MUL_x2_DEFINE(65); MUL_x2_DEFINE(66); MUL_x2_DEFINE(67); 360 MUL_x2_DEFINE(68); MUL_x2_DEFINE(69); MUL_x2_DEFINE(70); MUL_x2_DEFINE(71); 361 MUL_x2_DEFINE(72); MUL_x2_DEFINE(73); MUL_x2_DEFINE(74); MUL_x2_DEFINE(75); 362 MUL_x2_DEFINE(76); MUL_x2_DEFINE(77); MUL_x2_DEFINE(78); MUL_x2_DEFINE(79); 363 MUL_x2_DEFINE(80); MUL_x2_DEFINE(81); MUL_x2_DEFINE(82); MUL_x2_DEFINE(83); 364 MUL_x2_DEFINE(84); MUL_x2_DEFINE(85); MUL_x2_DEFINE(86); MUL_x2_DEFINE(87); 365 MUL_x2_DEFINE(88); MUL_x2_DEFINE(89); MUL_x2_DEFINE(90); MUL_x2_DEFINE(91); 366 MUL_x2_DEFINE(92); MUL_x2_DEFINE(93); MUL_x2_DEFINE(94); MUL_x2_DEFINE(95); 367 MUL_x2_DEFINE(96); MUL_x2_DEFINE(97); MUL_x2_DEFINE(98); MUL_x2_DEFINE(99); 368 MUL_x2_DEFINE(100); MUL_x2_DEFINE(101); MUL_x2_DEFINE(102); MUL_x2_DEFINE(103); 369 MUL_x2_DEFINE(104); MUL_x2_DEFINE(105); MUL_x2_DEFINE(106); MUL_x2_DEFINE(107); 370 MUL_x2_DEFINE(108); MUL_x2_DEFINE(109); MUL_x2_DEFINE(110); MUL_x2_DEFINE(111); 371 MUL_x2_DEFINE(112); MUL_x2_DEFINE(113); MUL_x2_DEFINE(114); MUL_x2_DEFINE(115); 372 MUL_x2_DEFINE(116); MUL_x2_DEFINE(117); MUL_x2_DEFINE(118); MUL_x2_DEFINE(119); 373 MUL_x2_DEFINE(120); MUL_x2_DEFINE(121); MUL_x2_DEFINE(122); MUL_x2_DEFINE(123); 374 MUL_x2_DEFINE(124); MUL_x2_DEFINE(125); MUL_x2_DEFINE(126); MUL_x2_DEFINE(127); 375 MUL_x2_DEFINE(128); MUL_x2_DEFINE(129); MUL_x2_DEFINE(130); MUL_x2_DEFINE(131); 376 MUL_x2_DEFINE(132); MUL_x2_DEFINE(133); MUL_x2_DEFINE(134); MUL_x2_DEFINE(135); 377 MUL_x2_DEFINE(136); MUL_x2_DEFINE(137); MUL_x2_DEFINE(138); MUL_x2_DEFINE(139); 378 MUL_x2_DEFINE(140); MUL_x2_DEFINE(141); MUL_x2_DEFINE(142); MUL_x2_DEFINE(143); 379 MUL_x2_DEFINE(144); MUL_x2_DEFINE(145); MUL_x2_DEFINE(146); MUL_x2_DEFINE(147); 380 MUL_x2_DEFINE(148); MUL_x2_DEFINE(149); MUL_x2_DEFINE(150); MUL_x2_DEFINE(151); 381 MUL_x2_DEFINE(152); MUL_x2_DEFINE(153); MUL_x2_DEFINE(154); MUL_x2_DEFINE(155); 382 MUL_x2_DEFINE(156); MUL_x2_DEFINE(157); MUL_x2_DEFINE(158); MUL_x2_DEFINE(159); 383 MUL_x2_DEFINE(160); MUL_x2_DEFINE(161); MUL_x2_DEFINE(162); MUL_x2_DEFINE(163); 384 MUL_x2_DEFINE(164); MUL_x2_DEFINE(165); MUL_x2_DEFINE(166); MUL_x2_DEFINE(167); 385 MUL_x2_DEFINE(168); MUL_x2_DEFINE(169); MUL_x2_DEFINE(170); MUL_x2_DEFINE(171); 386 MUL_x2_DEFINE(172); MUL_x2_DEFINE(173); MUL_x2_DEFINE(174); MUL_x2_DEFINE(175); 387 MUL_x2_DEFINE(176); MUL_x2_DEFINE(177); MUL_x2_DEFINE(178); MUL_x2_DEFINE(179); 388 MUL_x2_DEFINE(180); MUL_x2_DEFINE(181); MUL_x2_DEFINE(182); MUL_x2_DEFINE(183); 389 MUL_x2_DEFINE(184); MUL_x2_DEFINE(185); MUL_x2_DEFINE(186); MUL_x2_DEFINE(187); 390 MUL_x2_DEFINE(188); MUL_x2_DEFINE(189); MUL_x2_DEFINE(190); MUL_x2_DEFINE(191); 391 MUL_x2_DEFINE(192); MUL_x2_DEFINE(193); MUL_x2_DEFINE(194); MUL_x2_DEFINE(195); 392 MUL_x2_DEFINE(196); MUL_x2_DEFINE(197); MUL_x2_DEFINE(198); MUL_x2_DEFINE(199); 393 MUL_x2_DEFINE(200); MUL_x2_DEFINE(201); MUL_x2_DEFINE(202); MUL_x2_DEFINE(203); 394 MUL_x2_DEFINE(204); MUL_x2_DEFINE(205); MUL_x2_DEFINE(206); MUL_x2_DEFINE(207); 395 MUL_x2_DEFINE(208); MUL_x2_DEFINE(209); MUL_x2_DEFINE(210); MUL_x2_DEFINE(211); 396 MUL_x2_DEFINE(212); MUL_x2_DEFINE(213); MUL_x2_DEFINE(214); MUL_x2_DEFINE(215); 397 MUL_x2_DEFINE(216); MUL_x2_DEFINE(217); MUL_x2_DEFINE(218); MUL_x2_DEFINE(219); 398 MUL_x2_DEFINE(220); MUL_x2_DEFINE(221); MUL_x2_DEFINE(222); MUL_x2_DEFINE(223); 399 MUL_x2_DEFINE(224); MUL_x2_DEFINE(225); MUL_x2_DEFINE(226); MUL_x2_DEFINE(227); 400 MUL_x2_DEFINE(228); MUL_x2_DEFINE(229); MUL_x2_DEFINE(230); MUL_x2_DEFINE(231); 401 MUL_x2_DEFINE(232); MUL_x2_DEFINE(233); MUL_x2_DEFINE(234); MUL_x2_DEFINE(235); 402 MUL_x2_DEFINE(236); MUL_x2_DEFINE(237); MUL_x2_DEFINE(238); MUL_x2_DEFINE(239); 403 MUL_x2_DEFINE(240); MUL_x2_DEFINE(241); MUL_x2_DEFINE(242); MUL_x2_DEFINE(243); 404 MUL_x2_DEFINE(244); MUL_x2_DEFINE(245); MUL_x2_DEFINE(246); MUL_x2_DEFINE(247); 405 MUL_x2_DEFINE(248); MUL_x2_DEFINE(249); MUL_x2_DEFINE(250); MUL_x2_DEFINE(251); 406 MUL_x2_DEFINE(252); MUL_x2_DEFINE(253); MUL_x2_DEFINE(254); MUL_x2_DEFINE(255); 407 408 409 410 typedef void (*mul_fn_ptr_t)(void); 411 412 static const mul_fn_ptr_t __attribute__((aligned(256))) 413 gf_x1_mul_fns[256] = { 414 mul_x1_0, mul_x1_1, mul_x1_2, mul_x1_3, mul_x1_4, mul_x1_5, 415 mul_x1_6, mul_x1_7, mul_x1_8, mul_x1_9, mul_x1_10, mul_x1_11, 416 mul_x1_12, mul_x1_13, mul_x1_14, mul_x1_15, mul_x1_16, mul_x1_17, 417 mul_x1_18, mul_x1_19, mul_x1_20, mul_x1_21, mul_x1_22, mul_x1_23, 418 mul_x1_24, mul_x1_25, mul_x1_26, mul_x1_27, mul_x1_28, mul_x1_29, 419 mul_x1_30, mul_x1_31, mul_x1_32, mul_x1_33, mul_x1_34, mul_x1_35, 420 mul_x1_36, mul_x1_37, mul_x1_38, mul_x1_39, mul_x1_40, mul_x1_41, 421 mul_x1_42, mul_x1_43, mul_x1_44, mul_x1_45, mul_x1_46, mul_x1_47, 422 mul_x1_48, mul_x1_49, mul_x1_50, mul_x1_51, mul_x1_52, mul_x1_53, 423 mul_x1_54, mul_x1_55, mul_x1_56, mul_x1_57, mul_x1_58, mul_x1_59, 424 mul_x1_60, mul_x1_61, mul_x1_62, mul_x1_63, mul_x1_64, mul_x1_65, 425 mul_x1_66, mul_x1_67, mul_x1_68, mul_x1_69, mul_x1_70, mul_x1_71, 426 mul_x1_72, mul_x1_73, mul_x1_74, mul_x1_75, mul_x1_76, mul_x1_77, 427 mul_x1_78, mul_x1_79, mul_x1_80, mul_x1_81, mul_x1_82, mul_x1_83, 428 mul_x1_84, mul_x1_85, mul_x1_86, mul_x1_87, mul_x1_88, mul_x1_89, 429 mul_x1_90, mul_x1_91, mul_x1_92, mul_x1_93, mul_x1_94, mul_x1_95, 430 mul_x1_96, mul_x1_97, mul_x1_98, mul_x1_99, mul_x1_100, mul_x1_101, 431 mul_x1_102, mul_x1_103, mul_x1_104, mul_x1_105, mul_x1_106, mul_x1_107, 432 mul_x1_108, mul_x1_109, mul_x1_110, mul_x1_111, mul_x1_112, mul_x1_113, 433 mul_x1_114, mul_x1_115, mul_x1_116, mul_x1_117, mul_x1_118, mul_x1_119, 434 mul_x1_120, mul_x1_121, mul_x1_122, mul_x1_123, mul_x1_124, mul_x1_125, 435 mul_x1_126, mul_x1_127, mul_x1_128, mul_x1_129, mul_x1_130, mul_x1_131, 436 mul_x1_132, mul_x1_133, mul_x1_134, mul_x1_135, mul_x1_136, mul_x1_137, 437 mul_x1_138, mul_x1_139, mul_x1_140, mul_x1_141, mul_x1_142, mul_x1_143, 438 mul_x1_144, mul_x1_145, mul_x1_146, mul_x1_147, mul_x1_148, mul_x1_149, 439 mul_x1_150, mul_x1_151, mul_x1_152, mul_x1_153, mul_x1_154, mul_x1_155, 440 mul_x1_156, mul_x1_157, mul_x1_158, mul_x1_159, mul_x1_160, mul_x1_161, 441 mul_x1_162, mul_x1_163, mul_x1_164, mul_x1_165, mul_x1_166, mul_x1_167, 442 mul_x1_168, mul_x1_169, mul_x1_170, mul_x1_171, mul_x1_172, mul_x1_173, 443 mul_x1_174, mul_x1_175, mul_x1_176, mul_x1_177, mul_x1_178, mul_x1_179, 444 mul_x1_180, mul_x1_181, mul_x1_182, mul_x1_183, mul_x1_184, mul_x1_185, 445 mul_x1_186, mul_x1_187, mul_x1_188, mul_x1_189, mul_x1_190, mul_x1_191, 446 mul_x1_192, mul_x1_193, mul_x1_194, mul_x1_195, mul_x1_196, mul_x1_197, 447 mul_x1_198, mul_x1_199, mul_x1_200, mul_x1_201, mul_x1_202, mul_x1_203, 448 mul_x1_204, mul_x1_205, mul_x1_206, mul_x1_207, mul_x1_208, mul_x1_209, 449 mul_x1_210, mul_x1_211, mul_x1_212, mul_x1_213, mul_x1_214, mul_x1_215, 450 mul_x1_216, mul_x1_217, mul_x1_218, mul_x1_219, mul_x1_220, mul_x1_221, 451 mul_x1_222, mul_x1_223, mul_x1_224, mul_x1_225, mul_x1_226, mul_x1_227, 452 mul_x1_228, mul_x1_229, mul_x1_230, mul_x1_231, mul_x1_232, mul_x1_233, 453 mul_x1_234, mul_x1_235, mul_x1_236, mul_x1_237, mul_x1_238, mul_x1_239, 454 mul_x1_240, mul_x1_241, mul_x1_242, mul_x1_243, mul_x1_244, mul_x1_245, 455 mul_x1_246, mul_x1_247, mul_x1_248, mul_x1_249, mul_x1_250, mul_x1_251, 456 mul_x1_252, mul_x1_253, mul_x1_254, mul_x1_255 457 }; 458 459 static const mul_fn_ptr_t __attribute__((aligned(256))) 460 gf_x2_mul_fns[256] = { 461 mul_x2_0, mul_x2_1, mul_x2_2, mul_x2_3, mul_x2_4, mul_x2_5, 462 mul_x2_6, mul_x2_7, mul_x2_8, mul_x2_9, mul_x2_10, mul_x2_11, 463 mul_x2_12, mul_x2_13, mul_x2_14, mul_x2_15, mul_x2_16, mul_x2_17, 464 mul_x2_18, mul_x2_19, mul_x2_20, mul_x2_21, mul_x2_22, mul_x2_23, 465 mul_x2_24, mul_x2_25, mul_x2_26, mul_x2_27, mul_x2_28, mul_x2_29, 466 mul_x2_30, mul_x2_31, mul_x2_32, mul_x2_33, mul_x2_34, mul_x2_35, 467 mul_x2_36, mul_x2_37, mul_x2_38, mul_x2_39, mul_x2_40, mul_x2_41, 468 mul_x2_42, mul_x2_43, mul_x2_44, mul_x2_45, mul_x2_46, mul_x2_47, 469 mul_x2_48, mul_x2_49, mul_x2_50, mul_x2_51, mul_x2_52, mul_x2_53, 470 mul_x2_54, mul_x2_55, mul_x2_56, mul_x2_57, mul_x2_58, mul_x2_59, 471 mul_x2_60, mul_x2_61, mul_x2_62, mul_x2_63, mul_x2_64, mul_x2_65, 472 mul_x2_66, mul_x2_67, mul_x2_68, mul_x2_69, mul_x2_70, mul_x2_71, 473 mul_x2_72, mul_x2_73, mul_x2_74, mul_x2_75, mul_x2_76, mul_x2_77, 474 mul_x2_78, mul_x2_79, mul_x2_80, mul_x2_81, mul_x2_82, mul_x2_83, 475 mul_x2_84, mul_x2_85, mul_x2_86, mul_x2_87, mul_x2_88, mul_x2_89, 476 mul_x2_90, mul_x2_91, mul_x2_92, mul_x2_93, mul_x2_94, mul_x2_95, 477 mul_x2_96, mul_x2_97, mul_x2_98, mul_x2_99, mul_x2_100, mul_x2_101, 478 mul_x2_102, mul_x2_103, mul_x2_104, mul_x2_105, mul_x2_106, mul_x2_107, 479 mul_x2_108, mul_x2_109, mul_x2_110, mul_x2_111, mul_x2_112, mul_x2_113, 480 mul_x2_114, mul_x2_115, mul_x2_116, mul_x2_117, mul_x2_118, mul_x2_119, 481 mul_x2_120, mul_x2_121, mul_x2_122, mul_x2_123, mul_x2_124, mul_x2_125, 482 mul_x2_126, mul_x2_127, mul_x2_128, mul_x2_129, mul_x2_130, mul_x2_131, 483 mul_x2_132, mul_x2_133, mul_x2_134, mul_x2_135, mul_x2_136, mul_x2_137, 484 mul_x2_138, mul_x2_139, mul_x2_140, mul_x2_141, mul_x2_142, mul_x2_143, 485 mul_x2_144, mul_x2_145, mul_x2_146, mul_x2_147, mul_x2_148, mul_x2_149, 486 mul_x2_150, mul_x2_151, mul_x2_152, mul_x2_153, mul_x2_154, mul_x2_155, 487 mul_x2_156, mul_x2_157, mul_x2_158, mul_x2_159, mul_x2_160, mul_x2_161, 488 mul_x2_162, mul_x2_163, mul_x2_164, mul_x2_165, mul_x2_166, mul_x2_167, 489 mul_x2_168, mul_x2_169, mul_x2_170, mul_x2_171, mul_x2_172, mul_x2_173, 490 mul_x2_174, mul_x2_175, mul_x2_176, mul_x2_177, mul_x2_178, mul_x2_179, 491 mul_x2_180, mul_x2_181, mul_x2_182, mul_x2_183, mul_x2_184, mul_x2_185, 492 mul_x2_186, mul_x2_187, mul_x2_188, mul_x2_189, mul_x2_190, mul_x2_191, 493 mul_x2_192, mul_x2_193, mul_x2_194, mul_x2_195, mul_x2_196, mul_x2_197, 494 mul_x2_198, mul_x2_199, mul_x2_200, mul_x2_201, mul_x2_202, mul_x2_203, 495 mul_x2_204, mul_x2_205, mul_x2_206, mul_x2_207, mul_x2_208, mul_x2_209, 496 mul_x2_210, mul_x2_211, mul_x2_212, mul_x2_213, mul_x2_214, mul_x2_215, 497 mul_x2_216, mul_x2_217, mul_x2_218, mul_x2_219, mul_x2_220, mul_x2_221, 498 mul_x2_222, mul_x2_223, mul_x2_224, mul_x2_225, mul_x2_226, mul_x2_227, 499 mul_x2_228, mul_x2_229, mul_x2_230, mul_x2_231, mul_x2_232, mul_x2_233, 500 mul_x2_234, mul_x2_235, mul_x2_236, mul_x2_237, mul_x2_238, mul_x2_239, 501 mul_x2_240, mul_x2_241, mul_x2_242, mul_x2_243, mul_x2_244, mul_x2_245, 502 mul_x2_246, mul_x2_247, mul_x2_248, mul_x2_249, mul_x2_250, mul_x2_251, 503 mul_x2_252, mul_x2_253, mul_x2_254, mul_x2_255 504 }; 505 506 #define MUL(c, r...) \ 507 { \ 508 switch (REG_CNT(r)) { \ 509 case 2: \ 510 COPY(r, _mul_x2_in); \ 511 gf_x2_mul_fns[c](); \ 512 COPY(_mul_x2_acc, r); \ 513 break; \ 514 case 1: \ 515 COPY(r, _mul_x1_in); \ 516 gf_x1_mul_fns[c](); \ 517 COPY(_mul_x1_acc, r); \ 518 break; \ 519 default: \ 520 VERIFY(0); \ 521 } \ 522 } 523 524 525 #define raidz_math_begin() kfpu_begin() 526 #define raidz_math_end() kfpu_end() 527 528 #define SYN_STRIDE 4 529 530 #define ZERO_STRIDE 4 531 #define ZERO_DEFINE() {} 532 #define ZERO_D 0, 1, 2, 3 533 534 #define COPY_STRIDE 4 535 #define COPY_DEFINE() {} 536 #define COPY_D 0, 1, 2, 3 537 538 #define ADD_STRIDE 4 539 #define ADD_DEFINE() {} 540 #define ADD_D 0, 1, 2, 3 541 542 #define MUL_STRIDE 2 543 #define MUL_DEFINE() MUL2_SETUP() 544 #define MUL_D 0, 1 545 546 #define GEN_P_STRIDE 4 547 #define GEN_P_DEFINE() {} 548 #define GEN_P_P 0, 1, 2, 3 549 550 #define GEN_PQ_STRIDE 4 551 #define GEN_PQ_DEFINE() {} 552 #define GEN_PQ_D 0, 1, 2, 3 553 #define GEN_PQ_C 4, 5, 6, 7 554 555 #define GEN_PQR_STRIDE 4 556 #define GEN_PQR_DEFINE() {} 557 #define GEN_PQR_D 0, 1, 2, 3 558 #define GEN_PQR_C 4, 5, 6, 7 559 560 #define SYN_Q_DEFINE() {} 561 #define SYN_Q_D 0, 1, 2, 3 562 #define SYN_Q_X 4, 5, 6, 7 563 564 #define SYN_R_DEFINE() {} 565 #define SYN_R_D 0, 1, 2, 3 566 #define SYN_R_X 4, 5, 6, 7 567 568 #define SYN_PQ_DEFINE() {} 569 #define SYN_PQ_D 0, 1, 2, 3 570 #define SYN_PQ_X 4, 5, 6, 7 571 572 #define REC_PQ_STRIDE 2 573 #define REC_PQ_DEFINE() MUL2_SETUP() 574 #define REC_PQ_X 0, 1 575 #define REC_PQ_Y 2, 3 576 #define REC_PQ_T 4, 5 577 578 #define SYN_PR_DEFINE() {} 579 #define SYN_PR_D 0, 1, 2, 3 580 #define SYN_PR_X 4, 5, 6, 7 581 582 #define REC_PR_STRIDE 2 583 #define REC_PR_DEFINE() MUL2_SETUP() 584 #define REC_PR_X 0, 1 585 #define REC_PR_Y 2, 3 586 #define REC_PR_T 4, 5 587 588 #define SYN_QR_DEFINE() {} 589 #define SYN_QR_D 0, 1, 2, 3 590 #define SYN_QR_X 4, 5, 6, 7 591 592 #define REC_QR_STRIDE 2 593 #define REC_QR_DEFINE() MUL2_SETUP() 594 #define REC_QR_X 0, 1 595 #define REC_QR_Y 2, 3 596 #define REC_QR_T 4, 5 597 598 #define SYN_PQR_DEFINE() {} 599 #define SYN_PQR_D 0, 1, 2, 3 600 #define SYN_PQR_X 4, 5, 6, 7 601 602 #define REC_PQR_STRIDE 1 603 #define REC_PQR_DEFINE() MUL2_SETUP() 604 #define REC_PQR_X 0 605 #define REC_PQR_Y 1 606 #define REC_PQR_Z 2 607 #define REC_PQR_XS 3 608 #define REC_PQR_YS 4 609 610 611 #include <sys/vdev_raidz_impl.h> 612 #include "vdev_raidz_math_impl.h" 613 614 DEFINE_GEN_METHODS(sse2); 615 DEFINE_REC_METHODS(sse2); 616 617 static boolean_t 618 raidz_will_sse2_work(void) 619 { 620 return (kfpu_allowed() && zfs_sse_available() && zfs_sse2_available()); 621 } 622 623 const raidz_impl_ops_t vdev_raidz_sse2_impl = { 624 .init = NULL, 625 .fini = NULL, 626 .gen = RAIDZ_GEN_METHODS(sse2), 627 .rec = RAIDZ_REC_METHODS(sse2), 628 .is_supported = &raidz_will_sse2_work, 629 .name = "sse2" 630 }; 631 632 #endif /* defined(__x86_64) && defined(HAVE_SSE2) */ 633