/security/nss/lib/freebl/mpi/mpi_x86_asm.c
C | 568 lines | 402 code | 14 blank | 152 comment | 32 complexity | 2e0c9172c15d1aa207e1fa77c0aa6003 MD5 | raw file
1/* 2 * mpi_x86.c - MSVC inline assembly implementation of s_mpv_ functions. 3 * 4 * ***** BEGIN LICENSE BLOCK ***** 5 * Version: MPL 1.1/GPL 2.0/LGPL 2.1 6 * 7 * The contents of this file are subject to the Mozilla Public License Version 8 * 1.1 (the "License"); you may not use this file except in compliance with 9 * the License. You may obtain a copy of the License at 10 * http://www.mozilla.org/MPL/ 11 * 12 * Software distributed under the License is distributed on an "AS IS" basis, 13 * WITHOUT WARRANTY OF ANY KIND, either express or implied. See the License 14 * for the specific language governing rights and limitations under the 15 * License. 16 * 17 * The Original Code is the Netscape security libraries. 18 * 19 * The Initial Developer of the Original Code is 20 * Netscape Communications Corporation. 21 * Portions created by the Initial Developer are Copyright (C) 2000 22 * the Initial Developer. All Rights Reserved. 23 * 24 * Contributor(s): 25 * Benjamin Smedberg <benjamin@smedbergs.us> 26 * 27 * Alternatively, the contents of this file may be used under the terms of 28 * either the GNU General Public License Version 2 or later (the "GPL"), or 29 * the GNU Lesser General Public License Version 2.1 or later (the "LGPL"), 30 * in which case the provisions of the GPL or the LGPL are applicable instead 31 * of those above. If you wish to allow use of your version of this file only 32 * under the terms of either the GPL or the LGPL, and not to allow others to 33 * use your version of this file under the terms of the MPL, indicate your 34 * decision by deleting the provisions above and replace them with the notice 35 * and other provisions required by the GPL or the LGPL. If you do not delete 36 * the provisions above, a recipient may use your version of this file under 37 * the terms of any one of the MPL, the GPL or the LGPL. 38 * 39 * ***** END LICENSE BLOCK ***** */ 40 41#include "mpi-priv.h" 42 43static int is_sse = -1; 44extern unsigned long s_mpi_is_sse2(); 45 46/* 47 * ebp - 36: caller's esi 48 * ebp - 32: caller's edi 49 * ebp - 28: 50 * ebp - 24: 51 * ebp - 20: 52 * ebp - 16: 53 * ebp - 12: 54 * ebp - 8: 55 * ebp - 4: 56 * ebp + 0: caller's ebp 57 * ebp + 4: return address 58 * ebp + 8: a argument 59 * ebp + 12: a_len argument 60 * ebp + 16: b argument 61 * ebp + 20: c argument 62 * registers: 63 * eax: 64 * ebx: carry 65 * ecx: a_len 66 * edx: 67 * esi: a ptr 68 * edi: c ptr 69 */ 70__declspec(naked) void 71s_mpv_mul_d(const mp_digit *a, mp_size a_len, mp_digit b, mp_digit *c) 72{ 73 __asm { 74 mov eax, is_sse 75 cmp eax, 0 76 je s_mpv_mul_d_x86 77 jg s_mpv_mul_d_sse2 78 call s_mpi_is_sse2 79 mov is_sse, eax 80 cmp eax, 0 81 jg s_mpv_mul_d_sse2 82s_mpv_mul_d_x86: 83 push ebp 84 mov ebp,esp 85 sub esp,28 86 push edi 87 push esi 88 push ebx 89 mov ebx,0 ; carry = 0 90 mov ecx,[ebp+12] ; ecx = a_len 91 mov edi,[ebp+20] 92 cmp ecx,0 93 je L_2 ; jmp if a_len == 0 94 mov esi,[ebp+8] ; esi = a 95 cld 96L_1: 97 lodsd ; eax = [ds:esi]; esi += 4 98 mov edx,[ebp+16] ; edx = b 99 mul edx ; edx:eax = Phi:Plo = a_i * b 100 101 add eax,ebx ; add carry (ebx) to edx:eax 102 adc edx,0 103 mov ebx,edx ; high half of product becomes next carry 104 105 stosd ; [es:edi] = ax; edi += 4; 106 dec ecx ; --a_len 107 jnz L_1 ; jmp if a_len != 0 108L_2: 109 mov [edi],ebx ; *c = carry 110 pop ebx 111 pop esi 112 pop edi 113 leave 114 ret 115 nop 116s_mpv_mul_d_sse2: 117 push ebp 118 mov ebp, esp 119 push edi 120 push esi 121 psubq mm2, mm2 ; carry = 0 122 mov ecx, [ebp+12] ; ecx = a_len 123 movd mm1, [ebp+16] ; mm1 = b 124 mov edi, [ebp+20] 125 cmp ecx, 0 126 je L_6 ; jmp if a_len == 0 127 mov esi, [ebp+8] ; esi = a 128 cld 129L_5: 130 movd mm0, [esi] ; mm0 = *a++ 131 add esi, 4 132 pmuludq mm0, mm1 ; mm0 = b * *a++ 133 paddq mm2, mm0 ; add the carry 134 movd [edi], mm2 ; store the 32bit result 135 add edi, 4 136 psrlq mm2, 32 ; save the carry 137 dec ecx ; --a_len 138 jnz L_5 ; jmp if a_len != 0 139L_6: 140 movd [edi], mm2 ; *c = carry 141 emms 142 pop esi 143 pop edi 144 leave 145 ret 146 nop 147 } 148} 149 150/* 151 * ebp - 36: caller's esi 152 * ebp - 32: caller's edi 153 * ebp - 28: 154 * ebp - 24: 155 * ebp - 20: 156 * ebp - 16: 157 * ebp - 12: 158 * ebp - 8: 159 * ebp - 4: 160 * ebp + 0: caller's ebp 161 * ebp + 4: return address 162 * ebp + 8: a argument 163 * ebp + 12: a_len argument 164 * ebp + 16: b argument 165 * ebp + 20: c argument 166 * registers: 167 * eax: 168 * ebx: carry 169 * ecx: a_len 170 * edx: 171 * esi: a ptr 172 * edi: c ptr 173 */ 174__declspec(naked) void 175s_mpv_mul_d_add(const mp_digit *a, mp_size a_len, mp_digit b, mp_digit *c) 176{ 177 __asm { 178 mov eax, is_sse 179 cmp eax, 0 180 je s_mpv_mul_d_add_x86 181 jg s_mpv_mul_d_add_sse2 182 call s_mpi_is_sse2 183 mov is_sse, eax 184 cmp eax, 0 185 jg s_mpv_mul_d_add_sse2 186s_mpv_mul_d_add_x86: 187 push ebp 188 mov ebp,esp 189 sub esp,28 190 push edi 191 push esi 192 push ebx 193 mov ebx,0 ; carry = 0 194 mov ecx,[ebp+12] ; ecx = a_len 195 mov edi,[ebp+20] 196 cmp ecx,0 197 je L_11 ; jmp if a_len == 0 198 mov esi,[ebp+8] ; esi = a 199 cld 200L_10: 201 lodsd ; eax = [ds:esi]; esi += 4 202 mov edx,[ebp+16] ; edx = b 203 mul edx ; edx:eax = Phi:Plo = a_i * b 204 205 add eax,ebx ; add carry (ebx) to edx:eax 206 adc edx,0 207 mov ebx,[edi] ; add in current word from *c 208 add eax,ebx 209 adc edx,0 210 mov ebx,edx ; high half of product becomes next carry 211 212 stosd ; [es:edi] = ax; edi += 4; 213 dec ecx ; --a_len 214 jnz L_10 ; jmp if a_len != 0 215L_11: 216 mov [edi],ebx ; *c = carry 217 pop ebx 218 pop esi 219 pop edi 220 leave 221 ret 222 nop 223s_mpv_mul_d_add_sse2: 224 push ebp 225 mov ebp, esp 226 push edi 227 push esi 228 psubq mm2, mm2 ; carry = 0 229 mov ecx, [ebp+12] ; ecx = a_len 230 movd mm1, [ebp+16] ; mm1 = b 231 mov edi, [ebp+20] 232 cmp ecx, 0 233 je L_16 ; jmp if a_len == 0 234 mov esi, [ebp+8] ; esi = a 235 cld 236L_15: 237 movd mm0, [esi] ; mm0 = *a++ 238 add esi, 4 239 pmuludq mm0, mm1 ; mm0 = b * *a++ 240 paddq mm2, mm0 ; add the carry 241 movd mm0, [edi] 242 paddq mm2, mm0 ; add the carry 243 movd [edi], mm2 ; store the 32bit result 244 add edi, 4 245 psrlq mm2, 32 ; save the carry 246 dec ecx ; --a_len 247 jnz L_15 ; jmp if a_len != 0 248L_16: 249 movd [edi], mm2 ; *c = carry 250 emms 251 pop esi 252 pop edi 253 leave 254 ret 255 nop 256 } 257} 258 259/* 260 * ebp - 36: caller's esi 261 * ebp - 32: caller's edi 262 * ebp - 28: 263 * ebp - 24: 264 * ebp - 20: 265 * ebp - 16: 266 * ebp - 12: 267 * ebp - 8: 268 * ebp - 4: 269 * ebp + 0: caller's ebp 270 * ebp + 4: return address 271 * ebp + 8: a argument 272 * ebp + 12: a_len argument 273 * ebp + 16: b argument 274 * ebp + 20: c argument 275 * registers: 276 * eax: 277 * ebx: carry 278 * ecx: a_len 279 * edx: 280 * esi: a ptr 281 * edi: c ptr 282 */ 283__declspec(naked) void 284s_mpv_mul_d_add_prop(const mp_digit *a, mp_size a_len, mp_digit b, mp_digit *c) 285{ 286 __asm { 287 mov eax, is_sse 288 cmp eax, 0 289 je s_mpv_mul_d_add_prop_x86 290 jg s_mpv_mul_d_add_prop_sse2 291 call s_mpi_is_sse2 292 mov is_sse, eax 293 cmp eax, 0 294 jg s_mpv_mul_d_add_prop_sse2 295s_mpv_mul_d_add_prop_x86: 296 push ebp 297 mov ebp,esp 298 sub esp,28 299 push edi 300 push esi 301 push ebx 302 mov ebx,0 ; carry = 0 303 mov ecx,[ebp+12] ; ecx = a_len 304 mov edi,[ebp+20] 305 cmp ecx,0 306 je L_21 ; jmp if a_len == 0 307 cld 308 mov esi,[ebp+8] ; esi = a 309L_20: 310 lodsd ; eax = [ds:esi]; esi += 4 311 mov edx,[ebp+16] ; edx = b 312 mul edx ; edx:eax = Phi:Plo = a_i * b 313 314 add eax,ebx ; add carry (ebx) to edx:eax 315 adc edx,0 316 mov ebx,[edi] ; add in current word from *c 317 add eax,ebx 318 adc edx,0 319 mov ebx,edx ; high half of product becomes next carry 320 321 stosd ; [es:edi] = ax; edi += 4; 322 dec ecx ; --a_len 323 jnz L_20 ; jmp if a_len != 0 324L_21: 325 cmp ebx,0 ; is carry zero? 326 jz L_23 327 mov eax,[edi] ; add in current word from *c 328 add eax,ebx 329 stosd ; [es:edi] = ax; edi += 4; 330 jnc L_23 331L_22: 332 mov eax,[edi] ; add in current word from *c 333 adc eax,0 334 stosd ; [es:edi] = ax; edi += 4; 335 jc L_22 336L_23: 337 pop ebx 338 pop esi 339 pop edi 340 leave 341 ret 342 nop 343s_mpv_mul_d_add_prop_sse2: 344 push ebp 345 mov ebp, esp 346 push edi 347 push esi 348 push ebx 349 psubq mm2, mm2 ; carry = 0 350 mov ecx, [ebp+12] ; ecx = a_len 351 movd mm1, [ebp+16] ; mm1 = b 352 mov edi, [ebp+20] 353 cmp ecx, 0 354 je L_26 ; jmp if a_len == 0 355 mov esi, [ebp+8] ; esi = a 356 cld 357L_25: 358 movd mm0, [esi] ; mm0 = *a++ 359 movd mm3, [edi] ; fetch the sum 360 add esi, 4 361 pmuludq mm0, mm1 ; mm0 = b * *a++ 362 paddq mm2, mm0 ; add the carry 363 paddq mm2, mm3 ; add *c++ 364 movd [edi], mm2 ; store the 32bit result 365 add edi, 4 366 psrlq mm2, 32 ; save the carry 367 dec ecx ; --a_len 368 jnz L_25 ; jmp if a_len != 0 369L_26: 370 movd ebx, mm2 371 cmp ebx, 0 ; is carry zero? 372 jz L_28 373 mov eax, [edi] 374 add eax, ebx 375 stosd 376 jnc L_28 377L_27: 378 mov eax, [edi] ; add in current word from *c 379 adc eax, 0 380 stosd ; [es:edi] = ax; edi += 4; 381 jc L_27 382L_28: 383 emms 384 pop ebx 385 pop esi 386 pop edi 387 leave 388 ret 389 nop 390 } 391} 392 393/* 394 * ebp - 20: caller's esi 395 * ebp - 16: caller's edi 396 * ebp - 12: 397 * ebp - 8: carry 398 * ebp - 4: a_len local 399 * ebp + 0: caller's ebp 400 * ebp + 4: return address 401 * ebp + 8: pa argument 402 * ebp + 12: a_len argument 403 * ebp + 16: ps argument 404 * ebp + 20: 405 * registers: 406 * eax: 407 * ebx: carry 408 * ecx: a_len 409 * edx: 410 * esi: a ptr 411 * edi: c ptr 412 */ 413__declspec(naked) void 414s_mpv_sqr_add_prop(const mp_digit *a, mp_size a_len, mp_digit *sqrs) 415{ 416 __asm { 417 mov eax, is_sse 418 cmp eax, 0 419 je s_mpv_sqr_add_prop_x86 420 jg s_mpv_sqr_add_prop_sse2 421 call s_mpi_is_sse2 422 mov is_sse, eax 423 cmp eax, 0 424 jg s_mpv_sqr_add_prop_sse2 425s_mpv_sqr_add_prop_x86: 426 push ebp 427 mov ebp,esp 428 sub esp,12 429 push edi 430 push esi 431 push ebx 432 mov ebx,0 ; carry = 0 433 mov ecx,[ebp+12] ; a_len 434 mov edi,[ebp+16] ; edi = ps 435 cmp ecx,0 436 je L_31 ; jump if a_len == 0 437 cld 438 mov esi,[ebp+8] ; esi = pa 439L_30: 440 lodsd ; eax = [ds:si]; si += 4; 441 mul eax 442 443 add eax,ebx ; add "carry" 444 adc edx,0 445 mov ebx,[edi] 446 add eax,ebx ; add low word from result 447 mov ebx,[edi+4] 448 stosd ; [es:di] = eax; di += 4; 449 adc edx,ebx ; add high word from result 450 mov ebx,0 451 mov eax,edx 452 adc ebx,0 453 stosd ; [es:di] = eax; di += 4; 454 dec ecx ; --a_len 455 jnz L_30 ; jmp if a_len != 0 456L_31: 457 cmp ebx,0 ; is carry zero? 458 jz L_34 459 mov eax,[edi] ; add in current word from *c 460 add eax,ebx 461 stosd ; [es:edi] = ax; edi += 4; 462 jnc L_34 463L_32: 464 mov eax,[edi] ; add in current word from *c 465 adc eax,0 466 stosd ; [es:edi] = ax; edi += 4; 467 jc L_32 468L_34: 469 pop ebx 470 pop esi 471 pop edi 472 leave 473 ret 474 nop 475s_mpv_sqr_add_prop_sse2: 476 push ebp 477 mov ebp, esp 478 push edi 479 push esi 480 push ebx 481 psubq mm2, mm2 ; carry = 0 482 mov ecx, [ebp+12] ; ecx = a_len 483 mov edi, [ebp+16] 484 cmp ecx, 0 485 je L_36 ; jmp if a_len == 0 486 mov esi, [ebp+8] ; esi = a 487 cld 488L_35: 489 movd mm0, [esi] ; mm0 = *a 490 movd mm3, [edi] ; fetch the sum 491 add esi, 4 492 pmuludq mm0, mm0 ; mm0 = sqr(a) 493 paddq mm2, mm0 ; add the carry 494 paddq mm2, mm3 ; add the low word 495 movd mm3, [edi+4] 496 movd [edi], mm2 ; store the 32bit result 497 psrlq mm2, 32 498 paddq mm2, mm3 ; add the high word 499 movd [edi+4], mm2 ; store the 32bit result 500 psrlq mm2, 32 ; save the carry. 501 add edi, 8 502 dec ecx ; --a_len 503 jnz L_35 ; jmp if a_len != 0 504L_36: 505 movd ebx, mm2 506 cmp ebx, 0 ; is carry zero? 507 jz L_38 508 mov eax, [edi] 509 add eax, ebx 510 stosd 511 jnc L_38 512L_37: 513 mov eax, [edi] ; add in current word from *c 514 adc eax, 0 515 stosd ; [es:edi] = ax; edi += 4; 516 jc L_37 517L_38: 518 emms 519 pop ebx 520 pop esi 521 pop edi 522 leave 523 ret 524 nop 525 } 526} 527 528/* 529 * Divide 64-bit (Nhi,Nlo) by 32-bit divisor, which must be normalized 530 * so its high bit is 1. This code is from NSPR. 531 * 532 * Dump of assembler code for function s_mpv_div_2dx1d: 533 * 534 * esp + 0: Caller's ebx 535 * esp + 4: return address 536 * esp + 8: Nhi argument 537 * esp + 12: Nlo argument 538 * esp + 16: divisor argument 539 * esp + 20: qp argument 540 * esp + 24: rp argument 541 * registers: 542 * eax: 543 * ebx: carry 544 * ecx: a_len 545 * edx: 546 * esi: a ptr 547 * edi: c ptr 548 */ 549__declspec(naked) mp_err 550s_mpv_div_2dx1d(mp_digit Nhi, mp_digit Nlo, mp_digit divisor, 551 mp_digit *qp, mp_digit *rp) 552{ 553 __asm { 554 push ebx 555 mov edx,[esp+8] 556 mov eax,[esp+12] 557 mov ebx,[esp+16] 558 div ebx 559 mov ebx,[esp+20] 560 mov [ebx],eax 561 mov ebx,[esp+24] 562 mov [ebx],edx 563 xor eax,eax ; return zero 564 pop ebx 565 ret 566 nop 567 } 568}