1 /** 2 * \file bn_mul.h 3 * 4 * \brief Multi-precision integer library 5 */ 6 /* 7 * Copyright The Mbed TLS Contributors 8 * SPDX-License-Identifier: Apache-2.0 OR GPL-2.0-or-later 9 */ 10 /* 11 * Multiply source vector [s] with b, add result 12 * to destination vector [d] and set carry c. 13 * 14 * Currently supports: 15 * 16 * . IA-32 (386+) . AMD64 / EM64T 17 * . IA-32 (SSE2) . Motorola 68000 18 * . PowerPC, 32-bit . MicroBlaze 19 * . PowerPC, 64-bit . TriCore 20 * . SPARC v8 . ARM v3+ 21 * . Alpha . MIPS32 22 * . C, longlong . C, generic 23 */ 24 #ifndef MBEDTLS_BN_MUL_H 25 #define MBEDTLS_BN_MUL_H 26 27 #include "mbedtls/build_info.h" 28 29 #include "mbedtls/bignum.h" 30 31 32 /* 33 * Conversion macros for embedded constants: 34 * build lists of mbedtls_mpi_uint's from lists of unsigned char's grouped by 8, 4 or 2 35 */ 36 #if defined(MBEDTLS_HAVE_INT32) 37 38 #define MBEDTLS_BYTES_TO_T_UINT_4(a, b, c, d) \ 39 ((mbedtls_mpi_uint) (a) << 0) | \ 40 ((mbedtls_mpi_uint) (b) << 8) | \ 41 ((mbedtls_mpi_uint) (c) << 16) | \ 42 ((mbedtls_mpi_uint) (d) << 24) 43 44 #define MBEDTLS_BYTES_TO_T_UINT_2(a, b) \ 45 MBEDTLS_BYTES_TO_T_UINT_4(a, b, 0, 0) 46 47 #define MBEDTLS_BYTES_TO_T_UINT_8(a, b, c, d, e, f, g, h) \ 48 MBEDTLS_BYTES_TO_T_UINT_4(a, b, c, d), \ 49 MBEDTLS_BYTES_TO_T_UINT_4(e, f, g, h) 50 51 #else /* 64-bits */ 52 53 #define MBEDTLS_BYTES_TO_T_UINT_8(a, b, c, d, e, f, g, h) \ 54 ((mbedtls_mpi_uint) (a) << 0) | \ 55 ((mbedtls_mpi_uint) (b) << 8) | \ 56 ((mbedtls_mpi_uint) (c) << 16) | \ 57 ((mbedtls_mpi_uint) (d) << 24) | \ 58 ((mbedtls_mpi_uint) (e) << 32) | \ 59 ((mbedtls_mpi_uint) (f) << 40) | \ 60 ((mbedtls_mpi_uint) (g) << 48) | \ 61 ((mbedtls_mpi_uint) (h) << 56) 62 63 #define MBEDTLS_BYTES_TO_T_UINT_4(a, b, c, d) \ 64 MBEDTLS_BYTES_TO_T_UINT_8(a, b, c, d, 0, 0, 0, 0) 65 66 #define MBEDTLS_BYTES_TO_T_UINT_2(a, b) \ 67 MBEDTLS_BYTES_TO_T_UINT_8(a, b, 0, 0, 0, 0, 0, 0) 68 69 #endif /* bits in mbedtls_mpi_uint */ 70 71 /* *INDENT-OFF* */ 72 #if defined(MBEDTLS_HAVE_ASM) 73 74 /* armcc5 --gnu defines __GNUC__ but doesn't support GNU's extended asm */ 75 #if defined(__GNUC__) && \ 76 ( !defined(__ARMCC_VERSION) || __ARMCC_VERSION >= 6000000 ) 77 78 /* 79 * GCC < 5.0 treated the x86 ebx (which is used for the GOT) as a 80 * fixed reserved register when building as PIC, leading to errors 81 * like: bn_mul.h:46:13: error: PIC register clobbered by 'ebx' in 'asm' 82 * 83 * This is fixed by an improved register allocator in GCC 5+. From the 84 * release notes: 85 * Register allocation improvements: Reuse of the PIC hard register, 86 * instead of using a fixed register, was implemented on x86/x86-64 87 * targets. This improves generated PIC code performance as more hard 88 * registers can be used. 89 */ 90 #if defined(__GNUC__) && __GNUC__ < 5 && defined(__PIC__) 91 #define MULADDC_CANNOT_USE_EBX 92 #endif 93 94 /* 95 * Disable use of the i386 assembly code below if option -O0, to disable all 96 * compiler optimisations, is passed, detected with __OPTIMIZE__ 97 * This is done as the number of registers used in the assembly code doesn't 98 * work with the -O0 option. 99 */ 100 #if defined(__i386__) && defined(__OPTIMIZE__) && !defined(MULADDC_CANNOT_USE_EBX) 101 102 #define MULADDC_X1_INIT \ 103 { mbedtls_mpi_uint t; \ 104 asm( \ 105 "movl %%ebx, %0 \n\t" \ 106 "movl %5, %%esi \n\t" \ 107 "movl %6, %%edi \n\t" \ 108 "movl %7, %%ecx \n\t" \ 109 "movl %8, %%ebx \n\t" 110 111 #define MULADDC_X1_CORE \ 112 "lodsl \n\t" \ 113 "mull %%ebx \n\t" \ 114 "addl %%ecx, %%eax \n\t" \ 115 "adcl $0, %%edx \n\t" \ 116 "addl (%%edi), %%eax \n\t" \ 117 "adcl $0, %%edx \n\t" \ 118 "movl %%edx, %%ecx \n\t" \ 119 "stosl \n\t" 120 121 #define MULADDC_X1_STOP \ 122 "movl %4, %%ebx \n\t" \ 123 "movl %%ecx, %1 \n\t" \ 124 "movl %%edi, %2 \n\t" \ 125 "movl %%esi, %3 \n\t" \ 126 : "=m" (t), "=m" (c), "=m" (d), "=m" (s) \ 127 : "m" (t), "m" (s), "m" (d), "m" (c), "m" (b) \ 128 : "eax", "ebx", "ecx", "edx", "esi", "edi" \ 129 ); } 130 131 #if defined(MBEDTLS_HAVE_SSE2) 132 133 #define MULADDC_X8_INIT MULADDC_X1_INIT 134 135 #define MULADDC_X8_CORE \ 136 "movd %%ecx, %%mm1 \n\t" \ 137 "movd %%ebx, %%mm0 \n\t" \ 138 "movd (%%edi), %%mm3 \n\t" \ 139 "paddq %%mm3, %%mm1 \n\t" \ 140 "movd (%%esi), %%mm2 \n\t" \ 141 "pmuludq %%mm0, %%mm2 \n\t" \ 142 "movd 4(%%esi), %%mm4 \n\t" \ 143 "pmuludq %%mm0, %%mm4 \n\t" \ 144 "movd 8(%%esi), %%mm6 \n\t" \ 145 "pmuludq %%mm0, %%mm6 \n\t" \ 146 "movd 12(%%esi), %%mm7 \n\t" \ 147 "pmuludq %%mm0, %%mm7 \n\t" \ 148 "paddq %%mm2, %%mm1 \n\t" \ 149 "movd 4(%%edi), %%mm3 \n\t" \ 150 "paddq %%mm4, %%mm3 \n\t" \ 151 "movd 8(%%edi), %%mm5 \n\t" \ 152 "paddq %%mm6, %%mm5 \n\t" \ 153 "movd 12(%%edi), %%mm4 \n\t" \ 154 "paddq %%mm4, %%mm7 \n\t" \ 155 "movd %%mm1, (%%edi) \n\t" \ 156 "movd 16(%%esi), %%mm2 \n\t" \ 157 "pmuludq %%mm0, %%mm2 \n\t" \ 158 "psrlq $32, %%mm1 \n\t" \ 159 "movd 20(%%esi), %%mm4 \n\t" \ 160 "pmuludq %%mm0, %%mm4 \n\t" \ 161 "paddq %%mm3, %%mm1 \n\t" \ 162 "movd 24(%%esi), %%mm6 \n\t" \ 163 "pmuludq %%mm0, %%mm6 \n\t" \ 164 "movd %%mm1, 4(%%edi) \n\t" \ 165 "psrlq $32, %%mm1 \n\t" \ 166 "movd 28(%%esi), %%mm3 \n\t" \ 167 "pmuludq %%mm0, %%mm3 \n\t" \ 168 "paddq %%mm5, %%mm1 \n\t" \ 169 "movd 16(%%edi), %%mm5 \n\t" \ 170 "paddq %%mm5, %%mm2 \n\t" \ 171 "movd %%mm1, 8(%%edi) \n\t" \ 172 "psrlq $32, %%mm1 \n\t" \ 173 "paddq %%mm7, %%mm1 \n\t" \ 174 "movd 20(%%edi), %%mm5 \n\t" \ 175 "paddq %%mm5, %%mm4 \n\t" \ 176 "movd %%mm1, 12(%%edi) \n\t" \ 177 "psrlq $32, %%mm1 \n\t" \ 178 "paddq %%mm2, %%mm1 \n\t" \ 179 "movd 24(%%edi), %%mm5 \n\t" \ 180 "paddq %%mm5, %%mm6 \n\t" \ 181 "movd %%mm1, 16(%%edi) \n\t" \ 182 "psrlq $32, %%mm1 \n\t" \ 183 "paddq %%mm4, %%mm1 \n\t" \ 184 "movd 28(%%edi), %%mm5 \n\t" \ 185 "paddq %%mm5, %%mm3 \n\t" \ 186 "movd %%mm1, 20(%%edi) \n\t" \ 187 "psrlq $32, %%mm1 \n\t" \ 188 "paddq %%mm6, %%mm1 \n\t" \ 189 "movd %%mm1, 24(%%edi) \n\t" \ 190 "psrlq $32, %%mm1 \n\t" \ 191 "paddq %%mm3, %%mm1 \n\t" \ 192 "movd %%mm1, 28(%%edi) \n\t" \ 193 "addl $32, %%edi \n\t" \ 194 "addl $32, %%esi \n\t" \ 195 "psrlq $32, %%mm1 \n\t" \ 196 "movd %%mm1, %%ecx \n\t" 197 198 #define MULADDC_X8_STOP \ 199 "emms \n\t" \ 200 "movl %4, %%ebx \n\t" \ 201 "movl %%ecx, %1 \n\t" \ 202 "movl %%edi, %2 \n\t" \ 203 "movl %%esi, %3 \n\t" \ 204 : "=m" (t), "=m" (c), "=m" (d), "=m" (s) \ 205 : "m" (t), "m" (s), "m" (d), "m" (c), "m" (b) \ 206 : "eax", "ebx", "ecx", "edx", "esi", "edi" \ 207 ); } \ 208 209 #endif /* SSE2 */ 210 211 #endif /* i386 */ 212 213 #if defined(__amd64__) || defined (__x86_64__) 214 215 #define MULADDC_X1_INIT \ 216 asm( \ 217 "xorq %%r8, %%r8\n" 218 219 #define MULADDC_X1_CORE \ 220 "movq (%%rsi), %%rax\n" \ 221 "mulq %%rbx\n" \ 222 "addq $8, %%rsi\n" \ 223 "addq %%rcx, %%rax\n" \ 224 "movq %%r8, %%rcx\n" \ 225 "adcq $0, %%rdx\n" \ 226 "nop \n" \ 227 "addq %%rax, (%%rdi)\n" \ 228 "adcq %%rdx, %%rcx\n" \ 229 "addq $8, %%rdi\n" 230 231 #define MULADDC_X1_STOP \ 232 : "+c" (c), "+D" (d), "+S" (s), "+m" (*(uint64_t (*)[16]) d) \ 233 : "b" (b), "m" (*(const uint64_t (*)[16]) s) \ 234 : "rax", "rdx", "r8" \ 235 ); 236 237 #endif /* AMD64 */ 238 239 // The following assembly code assumes that a pointer will fit in a 64-bit register 240 // (including ILP32 __aarch64__ ABIs such as on watchOS, hence the 2^32 - 1) 241 #if defined(__aarch64__) && (UINTPTR_MAX == 0xfffffffful || UINTPTR_MAX == 0xfffffffffffffffful) 242 243 /* 244 * There are some issues around different compilers requiring different constraint 245 * syntax for updating pointers from assembly code (see notes for 246 * MBEDTLS_ASM_AARCH64_PTR_CONSTRAINT in common.h), especially on aarch64_32 (aka ILP32). 247 * 248 * For this reason we cast the pointers to/from uintptr_t here. 249 */ 250 #define MULADDC_X1_INIT \ 251 do { uintptr_t muladdc_d = (uintptr_t) d, muladdc_s = (uintptr_t) s; asm( 252 253 #define MULADDC_X1_CORE \ 254 "ldr x4, [%x2], #8 \n\t" \ 255 "ldr x5, [%x1] \n\t" \ 256 "mul x6, x4, %4 \n\t" \ 257 "umulh x7, x4, %4 \n\t" \ 258 "adds x5, x5, x6 \n\t" \ 259 "adc x7, x7, xzr \n\t" \ 260 "adds x5, x5, %0 \n\t" \ 261 "adc %0, x7, xzr \n\t" \ 262 "str x5, [%x1], #8 \n\t" 263 264 #define MULADDC_X1_STOP \ 265 : "+r" (c), \ 266 "+r" (muladdc_d), \ 267 "+r" (muladdc_s), \ 268 "+m" (*(uint64_t (*)[16]) d) \ 269 : "r" (b), "m" (*(const uint64_t (*)[16]) s) \ 270 : "x4", "x5", "x6", "x7", "cc" \ 271 ); d = (mbedtls_mpi_uint *)muladdc_d; s = (mbedtls_mpi_uint *)muladdc_s; } while (0); 272 273 #endif /* Aarch64 */ 274 275 #if defined(__mc68020__) || defined(__mcpu32__) 276 277 #define MULADDC_X1_INIT \ 278 asm( \ 279 "movl %3, %%a2 \n\t" \ 280 "movl %4, %%a3 \n\t" \ 281 "movl %5, %%d3 \n\t" \ 282 "movl %6, %%d2 \n\t" \ 283 "moveq #0, %%d0 \n\t" 284 285 #define MULADDC_X1_CORE \ 286 "movel %%a2@+, %%d1 \n\t" \ 287 "mulul %%d2, %%d4:%%d1 \n\t" \ 288 "addl %%d3, %%d1 \n\t" \ 289 "addxl %%d0, %%d4 \n\t" \ 290 "moveq #0, %%d3 \n\t" \ 291 "addl %%d1, %%a3@+ \n\t" \ 292 "addxl %%d4, %%d3 \n\t" 293 294 #define MULADDC_X1_STOP \ 295 "movl %%d3, %0 \n\t" \ 296 "movl %%a3, %1 \n\t" \ 297 "movl %%a2, %2 \n\t" \ 298 : "=m" (c), "=m" (d), "=m" (s) \ 299 : "m" (s), "m" (d), "m" (c), "m" (b) \ 300 : "d0", "d1", "d2", "d3", "d4", "a2", "a3" \ 301 ); 302 303 #define MULADDC_X8_INIT MULADDC_X1_INIT 304 305 #define MULADDC_X8_CORE \ 306 "movel %%a2@+, %%d1 \n\t" \ 307 "mulul %%d2, %%d4:%%d1 \n\t" \ 308 "addxl %%d3, %%d1 \n\t" \ 309 "addxl %%d0, %%d4 \n\t" \ 310 "addl %%d1, %%a3@+ \n\t" \ 311 "movel %%a2@+, %%d1 \n\t" \ 312 "mulul %%d2, %%d3:%%d1 \n\t" \ 313 "addxl %%d4, %%d1 \n\t" \ 314 "addxl %%d0, %%d3 \n\t" \ 315 "addl %%d1, %%a3@+ \n\t" \ 316 "movel %%a2@+, %%d1 \n\t" \ 317 "mulul %%d2, %%d4:%%d1 \n\t" \ 318 "addxl %%d3, %%d1 \n\t" \ 319 "addxl %%d0, %%d4 \n\t" \ 320 "addl %%d1, %%a3@+ \n\t" \ 321 "movel %%a2@+, %%d1 \n\t" \ 322 "mulul %%d2, %%d3:%%d1 \n\t" \ 323 "addxl %%d4, %%d1 \n\t" \ 324 "addxl %%d0, %%d3 \n\t" \ 325 "addl %%d1, %%a3@+ \n\t" \ 326 "movel %%a2@+, %%d1 \n\t" \ 327 "mulul %%d2, %%d4:%%d1 \n\t" \ 328 "addxl %%d3, %%d1 \n\t" \ 329 "addxl %%d0, %%d4 \n\t" \ 330 "addl %%d1, %%a3@+ \n\t" \ 331 "movel %%a2@+, %%d1 \n\t" \ 332 "mulul %%d2, %%d3:%%d1 \n\t" \ 333 "addxl %%d4, %%d1 \n\t" \ 334 "addxl %%d0, %%d3 \n\t" \ 335 "addl %%d1, %%a3@+ \n\t" \ 336 "movel %%a2@+, %%d1 \n\t" \ 337 "mulul %%d2, %%d4:%%d1 \n\t" \ 338 "addxl %%d3, %%d1 \n\t" \ 339 "addxl %%d0, %%d4 \n\t" \ 340 "addl %%d1, %%a3@+ \n\t" \ 341 "movel %%a2@+, %%d1 \n\t" \ 342 "mulul %%d2, %%d3:%%d1 \n\t" \ 343 "addxl %%d4, %%d1 \n\t" \ 344 "addxl %%d0, %%d3 \n\t" \ 345 "addl %%d1, %%a3@+ \n\t" \ 346 "addxl %%d0, %%d3 \n\t" 347 348 #define MULADDC_X8_STOP MULADDC_X1_STOP 349 350 #endif /* MC68000 */ 351 352 #if defined(__powerpc64__) || defined(__ppc64__) 353 354 #if defined(__MACH__) && defined(__APPLE__) 355 356 #define MULADDC_X1_INIT \ 357 asm( \ 358 "ld r3, %3 \n\t" \ 359 "ld r4, %4 \n\t" \ 360 "ld r5, %5 \n\t" \ 361 "ld r6, %6 \n\t" \ 362 "addi r3, r3, -8 \n\t" \ 363 "addi r4, r4, -8 \n\t" \ 364 "addic r5, r5, 0 \n\t" 365 366 #define MULADDC_X1_CORE \ 367 "ldu r7, 8(r3) \n\t" \ 368 "mulld r8, r7, r6 \n\t" \ 369 "mulhdu r9, r7, r6 \n\t" \ 370 "adde r8, r8, r5 \n\t" \ 371 "ld r7, 8(r4) \n\t" \ 372 "addze r5, r9 \n\t" \ 373 "addc r8, r8, r7 \n\t" \ 374 "stdu r8, 8(r4) \n\t" 375 376 #define MULADDC_X1_STOP \ 377 "addze r5, r5 \n\t" \ 378 "addi r4, r4, 8 \n\t" \ 379 "addi r3, r3, 8 \n\t" \ 380 "std r5, %0 \n\t" \ 381 "std r4, %1 \n\t" \ 382 "std r3, %2 \n\t" \ 383 : "=m" (c), "=m" (d), "=m" (s) \ 384 : "m" (s), "m" (d), "m" (c), "m" (b) \ 385 : "r3", "r4", "r5", "r6", "r7", "r8", "r9" \ 386 ); 387 388 389 #else /* __MACH__ && __APPLE__ */ 390 391 #define MULADDC_X1_INIT \ 392 asm( \ 393 "ld %%r3, %3 \n\t" \ 394 "ld %%r4, %4 \n\t" \ 395 "ld %%r5, %5 \n\t" \ 396 "ld %%r6, %6 \n\t" \ 397 "addi %%r3, %%r3, -8 \n\t" \ 398 "addi %%r4, %%r4, -8 \n\t" \ 399 "addic %%r5, %%r5, 0 \n\t" 400 401 #define MULADDC_X1_CORE \ 402 "ldu %%r7, 8(%%r3) \n\t" \ 403 "mulld %%r8, %%r7, %%r6 \n\t" \ 404 "mulhdu %%r9, %%r7, %%r6 \n\t" \ 405 "adde %%r8, %%r8, %%r5 \n\t" \ 406 "ld %%r7, 8(%%r4) \n\t" \ 407 "addze %%r5, %%r9 \n\t" \ 408 "addc %%r8, %%r8, %%r7 \n\t" \ 409 "stdu %%r8, 8(%%r4) \n\t" 410 411 #define MULADDC_X1_STOP \ 412 "addze %%r5, %%r5 \n\t" \ 413 "addi %%r4, %%r4, 8 \n\t" \ 414 "addi %%r3, %%r3, 8 \n\t" \ 415 "std %%r5, %0 \n\t" \ 416 "std %%r4, %1 \n\t" \ 417 "std %%r3, %2 \n\t" \ 418 : "=m" (c), "=m" (d), "=m" (s) \ 419 : "m" (s), "m" (d), "m" (c), "m" (b) \ 420 : "r3", "r4", "r5", "r6", "r7", "r8", "r9" \ 421 ); 422 423 #endif /* __MACH__ && __APPLE__ */ 424 425 #elif defined(__powerpc__) || defined(__ppc__) /* end PPC64/begin PPC32 */ 426 427 #if defined(__MACH__) && defined(__APPLE__) 428 429 #define MULADDC_X1_INIT \ 430 asm( \ 431 "lwz r3, %3 \n\t" \ 432 "lwz r4, %4 \n\t" \ 433 "lwz r5, %5 \n\t" \ 434 "lwz r6, %6 \n\t" \ 435 "addi r3, r3, -4 \n\t" \ 436 "addi r4, r4, -4 \n\t" \ 437 "addic r5, r5, 0 \n\t" 438 439 #define MULADDC_X1_CORE \ 440 "lwzu r7, 4(r3) \n\t" \ 441 "mullw r8, r7, r6 \n\t" \ 442 "mulhwu r9, r7, r6 \n\t" \ 443 "adde r8, r8, r5 \n\t" \ 444 "lwz r7, 4(r4) \n\t" \ 445 "addze r5, r9 \n\t" \ 446 "addc r8, r8, r7 \n\t" \ 447 "stwu r8, 4(r4) \n\t" 448 449 #define MULADDC_X1_STOP \ 450 "addze r5, r5 \n\t" \ 451 "addi r4, r4, 4 \n\t" \ 452 "addi r3, r3, 4 \n\t" \ 453 "stw r5, %0 \n\t" \ 454 "stw r4, %1 \n\t" \ 455 "stw r3, %2 \n\t" \ 456 : "=m" (c), "=m" (d), "=m" (s) \ 457 : "m" (s), "m" (d), "m" (c), "m" (b) \ 458 : "r3", "r4", "r5", "r6", "r7", "r8", "r9" \ 459 ); 460 461 #else /* __MACH__ && __APPLE__ */ 462 463 #define MULADDC_X1_INIT \ 464 asm( \ 465 "lwz %%r3, %3 \n\t" \ 466 "lwz %%r4, %4 \n\t" \ 467 "lwz %%r5, %5 \n\t" \ 468 "lwz %%r6, %6 \n\t" \ 469 "addi %%r3, %%r3, -4 \n\t" \ 470 "addi %%r4, %%r4, -4 \n\t" \ 471 "addic %%r5, %%r5, 0 \n\t" 472 473 #define MULADDC_X1_CORE \ 474 "lwzu %%r7, 4(%%r3) \n\t" \ 475 "mullw %%r8, %%r7, %%r6 \n\t" \ 476 "mulhwu %%r9, %%r7, %%r6 \n\t" \ 477 "adde %%r8, %%r8, %%r5 \n\t" \ 478 "lwz %%r7, 4(%%r4) \n\t" \ 479 "addze %%r5, %%r9 \n\t" \ 480 "addc %%r8, %%r8, %%r7 \n\t" \ 481 "stwu %%r8, 4(%%r4) \n\t" 482 483 #define MULADDC_X1_STOP \ 484 "addze %%r5, %%r5 \n\t" \ 485 "addi %%r4, %%r4, 4 \n\t" \ 486 "addi %%r3, %%r3, 4 \n\t" \ 487 "stw %%r5, %0 \n\t" \ 488 "stw %%r4, %1 \n\t" \ 489 "stw %%r3, %2 \n\t" \ 490 : "=m" (c), "=m" (d), "=m" (s) \ 491 : "m" (s), "m" (d), "m" (c), "m" (b) \ 492 : "r3", "r4", "r5", "r6", "r7", "r8", "r9" \ 493 ); 494 495 #endif /* __MACH__ && __APPLE__ */ 496 497 #endif /* PPC32 */ 498 499 /* 500 * The Sparc(64) assembly is reported to be broken. 501 * Disable it for now, until we're able to fix it. 502 */ 503 #if 0 && defined(__sparc__) 504 #if defined(__sparc64__) 505 506 #define MULADDC_X1_INIT \ 507 asm( \ 508 "ldx %3, %%o0 \n\t" \ 509 "ldx %4, %%o1 \n\t" \ 510 "ld %5, %%o2 \n\t" \ 511 "ld %6, %%o3 \n\t" 512 513 #define MULADDC_X1_CORE \ 514 "ld [%%o0], %%o4 \n\t" \ 515 "inc 4, %%o0 \n\t" \ 516 "ld [%%o1], %%o5 \n\t" \ 517 "umul %%o3, %%o4, %%o4 \n\t" \ 518 "addcc %%o4, %%o2, %%o4 \n\t" \ 519 "rd %%y, %%g1 \n\t" \ 520 "addx %%g1, 0, %%g1 \n\t" \ 521 "addcc %%o4, %%o5, %%o4 \n\t" \ 522 "st %%o4, [%%o1] \n\t" \ 523 "addx %%g1, 0, %%o2 \n\t" \ 524 "inc 4, %%o1 \n\t" 525 526 #define MULADDC_X1_STOP \ 527 "st %%o2, %0 \n\t" \ 528 "stx %%o1, %1 \n\t" \ 529 "stx %%o0, %2 \n\t" \ 530 : "=m" (c), "=m" (d), "=m" (s) \ 531 : "m" (s), "m" (d), "m" (c), "m" (b) \ 532 : "g1", "o0", "o1", "o2", "o3", "o4", \ 533 "o5" \ 534 ); 535 536 #else /* __sparc64__ */ 537 538 #define MULADDC_X1_INIT \ 539 asm( \ 540 "ld %3, %%o0 \n\t" \ 541 "ld %4, %%o1 \n\t" \ 542 "ld %5, %%o2 \n\t" \ 543 "ld %6, %%o3 \n\t" 544 545 #define MULADDC_X1_CORE \ 546 "ld [%%o0], %%o4 \n\t" \ 547 "inc 4, %%o0 \n\t" \ 548 "ld [%%o1], %%o5 \n\t" \ 549 "umul %%o3, %%o4, %%o4 \n\t" \ 550 "addcc %%o4, %%o2, %%o4 \n\t" \ 551 "rd %%y, %%g1 \n\t" \ 552 "addx %%g1, 0, %%g1 \n\t" \ 553 "addcc %%o4, %%o5, %%o4 \n\t" \ 554 "st %%o4, [%%o1] \n\t" \ 555 "addx %%g1, 0, %%o2 \n\t" \ 556 "inc 4, %%o1 \n\t" 557 558 #define MULADDC_X1_STOP \ 559 "st %%o2, %0 \n\t" \ 560 "st %%o1, %1 \n\t" \ 561 "st %%o0, %2 \n\t" \ 562 : "=m" (c), "=m" (d), "=m" (s) \ 563 : "m" (s), "m" (d), "m" (c), "m" (b) \ 564 : "g1", "o0", "o1", "o2", "o3", "o4", \ 565 "o5" \ 566 ); 567 568 #endif /* __sparc64__ */ 569 #endif /* __sparc__ */ 570 571 #if defined(__microblaze__) || defined(microblaze) 572 573 #define MULADDC_X1_INIT \ 574 asm( \ 575 "lwi r3, %3 \n\t" \ 576 "lwi r4, %4 \n\t" \ 577 "lwi r5, %5 \n\t" \ 578 "lwi r6, %6 \n\t" \ 579 "andi r7, r6, 0xffff \n\t" \ 580 "bsrli r6, r6, 16 \n\t" 581 582 #if(__BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) 583 #define MULADDC_LHUI \ 584 "lhui r9, r3, 0 \n\t" \ 585 "addi r3, r3, 2 \n\t" \ 586 "lhui r8, r3, 0 \n\t" 587 #else 588 #define MULADDC_LHUI \ 589 "lhui r8, r3, 0 \n\t" \ 590 "addi r3, r3, 2 \n\t" \ 591 "lhui r9, r3, 0 \n\t" 592 #endif 593 594 #define MULADDC_X1_CORE \ 595 MULADDC_LHUI \ 596 "addi r3, r3, 2 \n\t" \ 597 "mul r10, r9, r6 \n\t" \ 598 "mul r11, r8, r7 \n\t" \ 599 "mul r12, r9, r7 \n\t" \ 600 "mul r13, r8, r6 \n\t" \ 601 "bsrli r8, r10, 16 \n\t" \ 602 "bsrli r9, r11, 16 \n\t" \ 603 "add r13, r13, r8 \n\t" \ 604 "add r13, r13, r9 \n\t" \ 605 "bslli r10, r10, 16 \n\t" \ 606 "bslli r11, r11, 16 \n\t" \ 607 "add r12, r12, r10 \n\t" \ 608 "addc r13, r13, r0 \n\t" \ 609 "add r12, r12, r11 \n\t" \ 610 "addc r13, r13, r0 \n\t" \ 611 "lwi r10, r4, 0 \n\t" \ 612 "add r12, r12, r10 \n\t" \ 613 "addc r13, r13, r0 \n\t" \ 614 "add r12, r12, r5 \n\t" \ 615 "addc r5, r13, r0 \n\t" \ 616 "swi r12, r4, 0 \n\t" \ 617 "addi r4, r4, 4 \n\t" 618 619 #define MULADDC_X1_STOP \ 620 "swi r5, %0 \n\t" \ 621 "swi r4, %1 \n\t" \ 622 "swi r3, %2 \n\t" \ 623 : "=m" (c), "=m" (d), "=m" (s) \ 624 : "m" (s), "m" (d), "m" (c), "m" (b) \ 625 : "r3", "r4", "r5", "r6", "r7", "r8", \ 626 "r9", "r10", "r11", "r12", "r13" \ 627 ); 628 629 #endif /* MicroBlaze */ 630 631 #if defined(__tricore__) 632 633 #define MULADDC_X1_INIT \ 634 asm( \ 635 "ld.a %%a2, %3 \n\t" \ 636 "ld.a %%a3, %4 \n\t" \ 637 "ld.w %%d4, %5 \n\t" \ 638 "ld.w %%d1, %6 \n\t" \ 639 "xor %%d5, %%d5 \n\t" 640 641 #define MULADDC_X1_CORE \ 642 "ld.w %%d0, [%%a2+] \n\t" \ 643 "madd.u %%e2, %%e4, %%d0, %%d1 \n\t" \ 644 "ld.w %%d0, [%%a3] \n\t" \ 645 "addx %%d2, %%d2, %%d0 \n\t" \ 646 "addc %%d3, %%d3, 0 \n\t" \ 647 "mov %%d4, %%d3 \n\t" \ 648 "st.w [%%a3+], %%d2 \n\t" 649 650 #define MULADDC_X1_STOP \ 651 "st.w %0, %%d4 \n\t" \ 652 "st.a %1, %%a3 \n\t" \ 653 "st.a %2, %%a2 \n\t" \ 654 : "=m" (c), "=m" (d), "=m" (s) \ 655 : "m" (s), "m" (d), "m" (c), "m" (b) \ 656 : "d0", "d1", "e2", "d4", "a2", "a3" \ 657 ); 658 659 #endif /* TriCore */ 660 661 #if defined(__arm__) 662 663 #if defined(__thumb__) && !defined(__thumb2__) 664 #if defined(MBEDTLS_COMPILER_IS_GCC) 665 /* 666 * Thumb 1 ISA. This code path has only been tested successfully on gcc; 667 * it does not compile on clang or armclang. 668 */ 669 670 #if !defined(__OPTIMIZE__) && defined(__GNUC__) 671 /* 672 * Note, gcc -O0 by default uses r7 for the frame pointer, so it complains about 673 * our use of r7 below, unless -fomit-frame-pointer is passed. 674 * 675 * On the other hand, -fomit-frame-pointer is implied by any -Ox options with 676 * x !=0, which we can detect using __OPTIMIZE__ (which is also defined by 677 * clang and armcc5 under the same conditions). 678 * 679 * If gcc needs to use r7, we use r1 as a scratch register and have a few extra 680 * instructions to preserve/restore it; otherwise, we can use r7 and avoid 681 * the preserve/restore overhead. 682 */ 683 #define MULADDC_SCRATCH "RS .req r1 \n\t" 684 #define MULADDC_PRESERVE_SCRATCH "mov r10, r1 \n\t" 685 #define MULADDC_RESTORE_SCRATCH "mov r1, r10 \n\t" 686 #define MULADDC_SCRATCH_CLOBBER "r10" 687 #else /* !defined(__OPTIMIZE__) && defined(__GNUC__) */ 688 #define MULADDC_SCRATCH "RS .req r7 \n\t" 689 #define MULADDC_PRESERVE_SCRATCH "" 690 #define MULADDC_RESTORE_SCRATCH "" 691 #define MULADDC_SCRATCH_CLOBBER "r7" 692 #endif /* !defined(__OPTIMIZE__) && defined(__GNUC__) */ 693 694 #define MULADDC_X1_INIT \ 695 asm( \ 696 MULADDC_SCRATCH \ 697 "ldr r0, %3 \n\t" \ 698 "ldr r1, %4 \n\t" \ 699 "ldr r2, %5 \n\t" \ 700 "ldr r3, %6 \n\t" \ 701 "lsr r4, r3, #16 \n\t" \ 702 "mov r9, r4 \n\t" \ 703 "lsl r4, r3, #16 \n\t" \ 704 "lsr r4, r4, #16 \n\t" \ 705 "mov r8, r4 \n\t" \ 706 707 708 #define MULADDC_X1_CORE \ 709 MULADDC_PRESERVE_SCRATCH \ 710 "ldmia r0!, {r6} \n\t" \ 711 "lsr RS, r6, #16 \n\t" \ 712 "lsl r6, r6, #16 \n\t" \ 713 "lsr r6, r6, #16 \n\t" \ 714 "mov r4, r8 \n\t" \ 715 "mul r4, r6 \n\t" \ 716 "mov r3, r9 \n\t" \ 717 "mul r6, r3 \n\t" \ 718 "mov r5, r9 \n\t" \ 719 "mul r5, RS \n\t" \ 720 "mov r3, r8 \n\t" \ 721 "mul RS, r3 \n\t" \ 722 "lsr r3, r6, #16 \n\t" \ 723 "add r5, r5, r3 \n\t" \ 724 "lsr r3, RS, #16 \n\t" \ 725 "add r5, r5, r3 \n\t" \ 726 "add r4, r4, r2 \n\t" \ 727 "mov r2, #0 \n\t" \ 728 "adc r5, r2 \n\t" \ 729 "lsl r3, r6, #16 \n\t" \ 730 "add r4, r4, r3 \n\t" \ 731 "adc r5, r2 \n\t" \ 732 "lsl r3, RS, #16 \n\t" \ 733 "add r4, r4, r3 \n\t" \ 734 "adc r5, r2 \n\t" \ 735 MULADDC_RESTORE_SCRATCH \ 736 "ldr r3, [r1] \n\t" \ 737 "add r4, r4, r3 \n\t" \ 738 "adc r2, r5 \n\t" \ 739 "stmia r1!, {r4} \n\t" 740 741 #define MULADDC_X1_STOP \ 742 "str r2, %0 \n\t" \ 743 "str r1, %1 \n\t" \ 744 "str r0, %2 \n\t" \ 745 : "=m" (c), "=m" (d), "=m" (s) \ 746 : "m" (s), "m" (d), "m" (c), "m" (b) \ 747 : "r0", "r1", "r2", "r3", "r4", "r5", \ 748 "r6", MULADDC_SCRATCH_CLOBBER, "r8", "r9", "cc" \ 749 ); 750 #endif /* !defined(__ARMCC_VERSION) && !defined(__clang__) */ 751 752 #elif (__ARM_ARCH >= 6) && \ 753 defined (__ARM_FEATURE_DSP) && (__ARM_FEATURE_DSP == 1) 754 /* Armv6-M (or later) with DSP Instruction Set Extensions. 755 * Requires support for either Thumb 2 or Arm ISA. 756 */ 757 758 #define MULADDC_X1_INIT \ 759 { \ 760 mbedtls_mpi_uint tmp_a, tmp_b; \ 761 asm volatile ( 762 763 #define MULADDC_X1_CORE \ 764 ".p2align 2 \n\t" \ 765 "ldr %[a], [%[in]], #4 \n\t" \ 766 "ldr %[b], [%[acc]] \n\t" \ 767 "umaal %[b], %[carry], %[scalar], %[a] \n\t" \ 768 "str %[b], [%[acc]], #4 \n\t" 769 770 #define MULADDC_X1_STOP \ 771 : [a] "=&r" (tmp_a), \ 772 [b] "=&r" (tmp_b), \ 773 [in] "+r" (s), \ 774 [acc] "+r" (d), \ 775 [carry] "+l" (c) \ 776 : [scalar] "r" (b) \ 777 : "memory" \ 778 ); \ 779 } 780 781 #define MULADDC_X2_INIT \ 782 { \ 783 mbedtls_mpi_uint tmp_a0, tmp_b0; \ 784 mbedtls_mpi_uint tmp_a1, tmp_b1; \ 785 asm volatile ( 786 787 /* - Make sure loop is 4-byte aligned to avoid stalls 788 * upon repeated non-word aligned instructions in 789 * some microarchitectures. 790 * - Don't use ldm with post-increment or back-to-back 791 * loads with post-increment and same address register 792 * to avoid stalls on some microarchitectures. 793 * - Bunch loads and stores to reduce latency on some 794 * microarchitectures. E.g., on Cortex-M4, the first 795 * in a series of load/store operations has latency 796 * 2 cycles, while subsequent loads/stores are single-cycle. */ 797 #define MULADDC_X2_CORE \ 798 ".p2align 2 \n\t" \ 799 "ldr %[a0], [%[in]], #+8 \n\t" \ 800 "ldr %[b0], [%[acc]], #+8 \n\t" \ 801 "ldr %[a1], [%[in], #-4] \n\t" \ 802 "ldr %[b1], [%[acc], #-4] \n\t" \ 803 "umaal %[b0], %[carry], %[scalar], %[a0] \n\t" \ 804 "umaal %[b1], %[carry], %[scalar], %[a1] \n\t" \ 805 "str %[b0], [%[acc], #-8] \n\t" \ 806 "str %[b1], [%[acc], #-4] \n\t" 807 808 #define MULADDC_X2_STOP \ 809 : [a0] "=&r" (tmp_a0), \ 810 [b0] "=&r" (tmp_b0), \ 811 [a1] "=&r" (tmp_a1), \ 812 [b1] "=&r" (tmp_b1), \ 813 [in] "+r" (s), \ 814 [acc] "+r" (d), \ 815 [carry] "+l" (c) \ 816 : [scalar] "r" (b) \ 817 : "memory" \ 818 ); \ 819 } 820 821 #else /* Thumb 2 or Arm ISA, without DSP extensions */ 822 823 #define MULADDC_X1_INIT \ 824 asm( \ 825 "ldr r0, %3 \n\t" \ 826 "ldr r1, %4 \n\t" \ 827 "ldr r2, %5 \n\t" \ 828 "ldr r3, %6 \n\t" 829 830 #define MULADDC_X1_CORE \ 831 "ldr r4, [r0], #4 \n\t" \ 832 "mov r5, #0 \n\t" \ 833 "ldr r6, [r1] \n\t" \ 834 "umlal r2, r5, r3, r4 \n\t" \ 835 "adds r4, r6, r2 \n\t" \ 836 "adc r2, r5, #0 \n\t" \ 837 "str r4, [r1], #4 \n\t" 838 839 #define MULADDC_X1_STOP \ 840 "str r2, %0 \n\t" \ 841 "str r1, %1 \n\t" \ 842 "str r0, %2 \n\t" \ 843 : "=m" (c), "=m" (d), "=m" (s) \ 844 : "m" (s), "m" (d), "m" (c), "m" (b) \ 845 : "r0", "r1", "r2", "r3", "r4", "r5", \ 846 "r6", "cc" \ 847 ); 848 849 #endif /* ISA codepath selection */ 850 851 #endif /* defined(__arm__) */ 852 853 #if defined(__alpha__) 854 855 #define MULADDC_X1_INIT \ 856 asm( \ 857 "ldq $1, %3 \n\t" \ 858 "ldq $2, %4 \n\t" \ 859 "ldq $3, %5 \n\t" \ 860 "ldq $4, %6 \n\t" 861 862 #define MULADDC_X1_CORE \ 863 "ldq $6, 0($1) \n\t" \ 864 "addq $1, 8, $1 \n\t" \ 865 "mulq $6, $4, $7 \n\t" \ 866 "umulh $6, $4, $6 \n\t" \ 867 "addq $7, $3, $7 \n\t" \ 868 "cmpult $7, $3, $3 \n\t" \ 869 "ldq $5, 0($2) \n\t" \ 870 "addq $7, $5, $7 \n\t" \ 871 "cmpult $7, $5, $5 \n\t" \ 872 "stq $7, 0($2) \n\t" \ 873 "addq $2, 8, $2 \n\t" \ 874 "addq $6, $3, $3 \n\t" \ 875 "addq $5, $3, $3 \n\t" 876 877 #define MULADDC_X1_STOP \ 878 "stq $3, %0 \n\t" \ 879 "stq $2, %1 \n\t" \ 880 "stq $1, %2 \n\t" \ 881 : "=m" (c), "=m" (d), "=m" (s) \ 882 : "m" (s), "m" (d), "m" (c), "m" (b) \ 883 : "$1", "$2", "$3", "$4", "$5", "$6", "$7" \ 884 ); 885 #endif /* Alpha */ 886 887 #if defined(__mips__) && !defined(__mips64) 888 889 #define MULADDC_X1_INIT \ 890 asm( \ 891 "lw $10, %3 \n\t" \ 892 "lw $11, %4 \n\t" \ 893 "lw $12, %5 \n\t" \ 894 "lw $13, %6 \n\t" 895 896 #define MULADDC_X1_CORE \ 897 "lw $14, 0($10) \n\t" \ 898 "multu $13, $14 \n\t" \ 899 "addi $10, $10, 4 \n\t" \ 900 "mflo $14 \n\t" \ 901 "mfhi $9 \n\t" \ 902 "addu $14, $12, $14 \n\t" \ 903 "lw $15, 0($11) \n\t" \ 904 "sltu $12, $14, $12 \n\t" \ 905 "addu $15, $14, $15 \n\t" \ 906 "sltu $14, $15, $14 \n\t" \ 907 "addu $12, $12, $9 \n\t" \ 908 "sw $15, 0($11) \n\t" \ 909 "addu $12, $12, $14 \n\t" \ 910 "addi $11, $11, 4 \n\t" 911 912 #define MULADDC_X1_STOP \ 913 "sw $12, %0 \n\t" \ 914 "sw $11, %1 \n\t" \ 915 "sw $10, %2 \n\t" \ 916 : "=m" (c), "=m" (d), "=m" (s) \ 917 : "m" (s), "m" (d), "m" (c), "m" (b) \ 918 : "$9", "$10", "$11", "$12", "$13", "$14", "$15", "lo", "hi" \ 919 ); 920 921 #endif /* MIPS */ 922 #endif /* GNUC */ 923 924 #if (defined(_MSC_VER) && defined(_M_IX86)) || defined(__WATCOMC__) 925 926 #define MULADDC_X1_INIT \ 927 __asm mov esi, s \ 928 __asm mov edi, d \ 929 __asm mov ecx, c \ 930 __asm mov ebx, b 931 932 #define MULADDC_X1_CORE \ 933 __asm lodsd \ 934 __asm mul ebx \ 935 __asm add eax, ecx \ 936 __asm adc edx, 0 \ 937 __asm add eax, [edi] \ 938 __asm adc edx, 0 \ 939 __asm mov ecx, edx \ 940 __asm stosd 941 942 #define MULADDC_X1_STOP \ 943 __asm mov c, ecx \ 944 __asm mov d, edi \ 945 __asm mov s, esi 946 947 #if defined(MBEDTLS_HAVE_SSE2) 948 949 #define EMIT __asm _emit 950 951 #define MULADDC_X8_INIT MULADDC_X1_INIT 952 953 #define MULADDC_X8_CORE \ 954 EMIT 0x0F EMIT 0x6E EMIT 0xC9 \ 955 EMIT 0x0F EMIT 0x6E EMIT 0xC3 \ 956 EMIT 0x0F EMIT 0x6E EMIT 0x1F \ 957 EMIT 0x0F EMIT 0xD4 EMIT 0xCB \ 958 EMIT 0x0F EMIT 0x6E EMIT 0x16 \ 959 EMIT 0x0F EMIT 0xF4 EMIT 0xD0 \ 960 EMIT 0x0F EMIT 0x6E EMIT 0x66 EMIT 0x04 \ 961 EMIT 0x0F EMIT 0xF4 EMIT 0xE0 \ 962 EMIT 0x0F EMIT 0x6E EMIT 0x76 EMIT 0x08 \ 963 EMIT 0x0F EMIT 0xF4 EMIT 0xF0 \ 964 EMIT 0x0F EMIT 0x6E EMIT 0x7E EMIT 0x0C \ 965 EMIT 0x0F EMIT 0xF4 EMIT 0xF8 \ 966 EMIT 0x0F EMIT 0xD4 EMIT 0xCA \ 967 EMIT 0x0F EMIT 0x6E EMIT 0x5F EMIT 0x04 \ 968 EMIT 0x0F EMIT 0xD4 EMIT 0xDC \ 969 EMIT 0x0F EMIT 0x6E EMIT 0x6F EMIT 0x08 \ 970 EMIT 0x0F EMIT 0xD4 EMIT 0xEE \ 971 EMIT 0x0F EMIT 0x6E EMIT 0x67 EMIT 0x0C \ 972 EMIT 0x0F EMIT 0xD4 EMIT 0xFC \ 973 EMIT 0x0F EMIT 0x7E EMIT 0x0F \ 974 EMIT 0x0F EMIT 0x6E EMIT 0x56 EMIT 0x10 \ 975 EMIT 0x0F EMIT 0xF4 EMIT 0xD0 \ 976 EMIT 0x0F EMIT 0x73 EMIT 0xD1 EMIT 0x20 \ 977 EMIT 0x0F EMIT 0x6E EMIT 0x66 EMIT 0x14 \ 978 EMIT 0x0F EMIT 0xF4 EMIT 0xE0 \ 979 EMIT 0x0F EMIT 0xD4 EMIT 0xCB \ 980 EMIT 0x0F EMIT 0x6E EMIT 0x76 EMIT 0x18 \ 981 EMIT 0x0F EMIT 0xF4 EMIT 0xF0 \ 982 EMIT 0x0F EMIT 0x7E EMIT 0x4F EMIT 0x04 \ 983 EMIT 0x0F EMIT 0x73 EMIT 0xD1 EMIT 0x20 \ 984 EMIT 0x0F EMIT 0x6E EMIT 0x5E EMIT 0x1C \ 985 EMIT 0x0F EMIT 0xF4 EMIT 0xD8 \ 986 EMIT 0x0F EMIT 0xD4 EMIT 0xCD \ 987 EMIT 0x0F EMIT 0x6E EMIT 0x6F EMIT 0x10 \ 988 EMIT 0x0F EMIT 0xD4 EMIT 0xD5 \ 989 EMIT 0x0F EMIT 0x7E EMIT 0x4F EMIT 0x08 \ 990 EMIT 0x0F EMIT 0x73 EMIT 0xD1 EMIT 0x20 \ 991 EMIT 0x0F EMIT 0xD4 EMIT 0xCF \ 992 EMIT 0x0F EMIT 0x6E EMIT 0x6F EMIT 0x14 \ 993 EMIT 0x0F EMIT 0xD4 EMIT 0xE5 \ 994 EMIT 0x0F EMIT 0x7E EMIT 0x4F EMIT 0x0C \ 995 EMIT 0x0F EMIT 0x73 EMIT 0xD1 EMIT 0x20 \ 996 EMIT 0x0F EMIT 0xD4 EMIT 0xCA \ 997 EMIT 0x0F EMIT 0x6E EMIT 0x6F EMIT 0x18 \ 998 EMIT 0x0F EMIT 0xD4 EMIT 0xF5 \ 999 EMIT 0x0F EMIT 0x7E EMIT 0x4F EMIT 0x10 \ 1000 EMIT 0x0F EMIT 0x73 EMIT 0xD1 EMIT 0x20 \ 1001 EMIT 0x0F EMIT 0xD4 EMIT 0xCC \ 1002 EMIT 0x0F EMIT 0x6E EMIT 0x6F EMIT 0x1C \ 1003 EMIT 0x0F EMIT 0xD4 EMIT 0xDD \ 1004 EMIT 0x0F EMIT 0x7E EMIT 0x4F EMIT 0x14 \ 1005 EMIT 0x0F EMIT 0x73 EMIT 0xD1 EMIT 0x20 \ 1006 EMIT 0x0F EMIT 0xD4 EMIT 0xCE \ 1007 EMIT 0x0F EMIT 0x7E EMIT 0x4F EMIT 0x18 \ 1008 EMIT 0x0F EMIT 0x73 EMIT 0xD1 EMIT 0x20 \ 1009 EMIT 0x0F EMIT 0xD4 EMIT 0xCB \ 1010 EMIT 0x0F EMIT 0x7E EMIT 0x4F EMIT 0x1C \ 1011 EMIT 0x83 EMIT 0xC7 EMIT 0x20 \ 1012 EMIT 0x83 EMIT 0xC6 EMIT 0x20 \ 1013 EMIT 0x0F EMIT 0x73 EMIT 0xD1 EMIT 0x20 \ 1014 EMIT 0x0F EMIT 0x7E EMIT 0xC9 1015 1016 #define MULADDC_X8_STOP \ 1017 EMIT 0x0F EMIT 0x77 \ 1018 __asm mov c, ecx \ 1019 __asm mov d, edi \ 1020 __asm mov s, esi 1021 1022 #endif /* SSE2 */ 1023 #endif /* MSVC */ 1024 1025 #endif /* MBEDTLS_HAVE_ASM */ 1026 1027 #if !defined(MULADDC_X1_CORE) 1028 #if defined(MBEDTLS_HAVE_UDBL) 1029 1030 #define MULADDC_X1_INIT \ 1031 { \ 1032 mbedtls_t_udbl r; \ 1033 mbedtls_mpi_uint r0, r1; 1034 1035 #define MULADDC_X1_CORE \ 1036 r = *(s++) * (mbedtls_t_udbl) b; \ 1037 r0 = (mbedtls_mpi_uint) r; \ 1038 r1 = (mbedtls_mpi_uint)( r >> biL ); \ 1039 r0 += c; r1 += (r0 < c); \ 1040 r0 += *d; r1 += (r0 < *d); \ 1041 c = r1; *(d++) = r0; 1042 1043 #define MULADDC_X1_STOP \ 1044 } 1045 1046 #else /* MBEDTLS_HAVE_UDBL */ 1047 1048 #define MULADDC_X1_INIT \ 1049 { \ 1050 mbedtls_mpi_uint s0, s1, b0, b1; \ 1051 mbedtls_mpi_uint r0, r1, rx, ry; \ 1052 b0 = ( b << biH ) >> biH; \ 1053 b1 = ( b >> biH ); 1054 1055 #define MULADDC_X1_CORE \ 1056 s0 = ( *s << biH ) >> biH; \ 1057 s1 = ( *s >> biH ); s++; \ 1058 rx = s0 * b1; r0 = s0 * b0; \ 1059 ry = s1 * b0; r1 = s1 * b1; \ 1060 r1 += ( rx >> biH ); \ 1061 r1 += ( ry >> biH ); \ 1062 rx <<= biH; ry <<= biH; \ 1063 r0 += rx; r1 += (r0 < rx); \ 1064 r0 += ry; r1 += (r0 < ry); \ 1065 r0 += c; r1 += (r0 < c); \ 1066 r0 += *d; r1 += (r0 < *d); \ 1067 c = r1; *(d++) = r0; 1068 1069 #define MULADDC_X1_STOP \ 1070 } 1071 1072 #endif /* C (longlong) */ 1073 #endif /* C (generic) */ 1074 1075 #if !defined(MULADDC_X2_CORE) 1076 #define MULADDC_X2_INIT MULADDC_X1_INIT 1077 #define MULADDC_X2_STOP MULADDC_X1_STOP 1078 #define MULADDC_X2_CORE MULADDC_X1_CORE MULADDC_X1_CORE 1079 #endif /* MULADDC_X2_CORE */ 1080 1081 #if !defined(MULADDC_X4_CORE) 1082 #define MULADDC_X4_INIT MULADDC_X2_INIT 1083 #define MULADDC_X4_STOP MULADDC_X2_STOP 1084 #define MULADDC_X4_CORE MULADDC_X2_CORE MULADDC_X2_CORE 1085 #endif /* MULADDC_X4_CORE */ 1086 1087 #if !defined(MULADDC_X8_CORE) 1088 #define MULADDC_X8_INIT MULADDC_X4_INIT 1089 #define MULADDC_X8_STOP MULADDC_X4_STOP 1090 #define MULADDC_X8_CORE MULADDC_X4_CORE MULADDC_X4_CORE 1091 #endif /* MULADDC_X8_CORE */ 1092 1093 /* *INDENT-ON* */ 1094 #endif /* bn_mul.h */ 1095
