1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/security/nss/lib/freebl/mpi/mpi_x86_asm.c Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,535 @@
1.4 +/*
1.5 + * mpi_x86_asm.c - MSVC inline assembly implementation of s_mpv_ functions.
1.6 + *
1.7 + * This Source Code Form is subject to the terms of the Mozilla Public
1.8 + * License, v. 2.0. If a copy of the MPL was not distributed with this
1.9 + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
1.10 +
1.11 +#include "mpi-priv.h"
1.12 +
1.13 +static int is_sse = -1;
1.14 +extern unsigned long s_mpi_is_sse2();
1.15 +
1.16 +/*
1.17 + * ebp - 36: caller's esi
1.18 + * ebp - 32: caller's edi
1.19 + * ebp - 28:
1.20 + * ebp - 24:
1.21 + * ebp - 20:
1.22 + * ebp - 16:
1.23 + * ebp - 12:
1.24 + * ebp - 8:
1.25 + * ebp - 4:
1.26 + * ebp + 0: caller's ebp
1.27 + * ebp + 4: return address
1.28 + * ebp + 8: a argument
1.29 + * ebp + 12: a_len argument
1.30 + * ebp + 16: b argument
1.31 + * ebp + 20: c argument
1.32 + * registers:
1.33 + * eax:
1.34 + * ebx: carry
1.35 + * ecx: a_len
1.36 + * edx:
1.37 + * esi: a ptr
1.38 + * edi: c ptr
1.39 + */
1.40 +__declspec(naked) void
1.41 +s_mpv_mul_d(const mp_digit *a, mp_size a_len, mp_digit b, mp_digit *c)
1.42 +{
1.43 + __asm {
1.44 + mov eax, is_sse
1.45 + cmp eax, 0
1.46 + je s_mpv_mul_d_x86
1.47 + jg s_mpv_mul_d_sse2
1.48 + call s_mpi_is_sse2
1.49 + mov is_sse, eax
1.50 + cmp eax, 0
1.51 + jg s_mpv_mul_d_sse2
1.52 +s_mpv_mul_d_x86:
1.53 + push ebp
1.54 + mov ebp,esp
1.55 + sub esp,28
1.56 + push edi
1.57 + push esi
1.58 + push ebx
1.59 + mov ebx,0 ; carry = 0
1.60 + mov ecx,[ebp+12] ; ecx = a_len
1.61 + mov edi,[ebp+20]
1.62 + cmp ecx,0
1.63 + je L_2 ; jmp if a_len == 0
1.64 + mov esi,[ebp+8] ; esi = a
1.65 + cld
1.66 +L_1:
1.67 + lodsd ; eax = [ds:esi]; esi += 4
1.68 + mov edx,[ebp+16] ; edx = b
1.69 + mul edx ; edx:eax = Phi:Plo = a_i * b
1.70 +
1.71 + add eax,ebx ; add carry (ebx) to edx:eax
1.72 + adc edx,0
1.73 + mov ebx,edx ; high half of product becomes next carry
1.74 +
1.75 + stosd ; [es:edi] = ax; edi += 4;
1.76 + dec ecx ; --a_len
1.77 + jnz L_1 ; jmp if a_len != 0
1.78 +L_2:
1.79 + mov [edi],ebx ; *c = carry
1.80 + pop ebx
1.81 + pop esi
1.82 + pop edi
1.83 + leave
1.84 + ret
1.85 + nop
1.86 +s_mpv_mul_d_sse2:
1.87 + push ebp
1.88 + mov ebp, esp
1.89 + push edi
1.90 + push esi
1.91 + psubq mm2, mm2 ; carry = 0
1.92 + mov ecx, [ebp+12] ; ecx = a_len
1.93 + movd mm1, [ebp+16] ; mm1 = b
1.94 + mov edi, [ebp+20]
1.95 + cmp ecx, 0
1.96 + je L_6 ; jmp if a_len == 0
1.97 + mov esi, [ebp+8] ; esi = a
1.98 + cld
1.99 +L_5:
1.100 + movd mm0, [esi] ; mm0 = *a++
1.101 + add esi, 4
1.102 + pmuludq mm0, mm1 ; mm0 = b * *a++
1.103 + paddq mm2, mm0 ; add the carry
1.104 + movd [edi], mm2 ; store the 32bit result
1.105 + add edi, 4
1.106 + psrlq mm2, 32 ; save the carry
1.107 + dec ecx ; --a_len
1.108 + jnz L_5 ; jmp if a_len != 0
1.109 +L_6:
1.110 + movd [edi], mm2 ; *c = carry
1.111 + emms
1.112 + pop esi
1.113 + pop edi
1.114 + leave
1.115 + ret
1.116 + nop
1.117 + }
1.118 +}
1.119 +
1.120 +/*
1.121 + * ebp - 36: caller's esi
1.122 + * ebp - 32: caller's edi
1.123 + * ebp - 28:
1.124 + * ebp - 24:
1.125 + * ebp - 20:
1.126 + * ebp - 16:
1.127 + * ebp - 12:
1.128 + * ebp - 8:
1.129 + * ebp - 4:
1.130 + * ebp + 0: caller's ebp
1.131 + * ebp + 4: return address
1.132 + * ebp + 8: a argument
1.133 + * ebp + 12: a_len argument
1.134 + * ebp + 16: b argument
1.135 + * ebp + 20: c argument
1.136 + * registers:
1.137 + * eax:
1.138 + * ebx: carry
1.139 + * ecx: a_len
1.140 + * edx:
1.141 + * esi: a ptr
1.142 + * edi: c ptr
1.143 + */
1.144 +__declspec(naked) void
1.145 +s_mpv_mul_d_add(const mp_digit *a, mp_size a_len, mp_digit b, mp_digit *c)
1.146 +{
1.147 + __asm {
1.148 + mov eax, is_sse
1.149 + cmp eax, 0
1.150 + je s_mpv_mul_d_add_x86
1.151 + jg s_mpv_mul_d_add_sse2
1.152 + call s_mpi_is_sse2
1.153 + mov is_sse, eax
1.154 + cmp eax, 0
1.155 + jg s_mpv_mul_d_add_sse2
1.156 +s_mpv_mul_d_add_x86:
1.157 + push ebp
1.158 + mov ebp,esp
1.159 + sub esp,28
1.160 + push edi
1.161 + push esi
1.162 + push ebx
1.163 + mov ebx,0 ; carry = 0
1.164 + mov ecx,[ebp+12] ; ecx = a_len
1.165 + mov edi,[ebp+20]
1.166 + cmp ecx,0
1.167 + je L_11 ; jmp if a_len == 0
1.168 + mov esi,[ebp+8] ; esi = a
1.169 + cld
1.170 +L_10:
1.171 + lodsd ; eax = [ds:esi]; esi += 4
1.172 + mov edx,[ebp+16] ; edx = b
1.173 + mul edx ; edx:eax = Phi:Plo = a_i * b
1.174 +
1.175 + add eax,ebx ; add carry (ebx) to edx:eax
1.176 + adc edx,0
1.177 + mov ebx,[edi] ; add in current word from *c
1.178 + add eax,ebx
1.179 + adc edx,0
1.180 + mov ebx,edx ; high half of product becomes next carry
1.181 +
1.182 + stosd ; [es:edi] = ax; edi += 4;
1.183 + dec ecx ; --a_len
1.184 + jnz L_10 ; jmp if a_len != 0
1.185 +L_11:
1.186 + mov [edi],ebx ; *c = carry
1.187 + pop ebx
1.188 + pop esi
1.189 + pop edi
1.190 + leave
1.191 + ret
1.192 + nop
1.193 +s_mpv_mul_d_add_sse2:
1.194 + push ebp
1.195 + mov ebp, esp
1.196 + push edi
1.197 + push esi
1.198 + psubq mm2, mm2 ; carry = 0
1.199 + mov ecx, [ebp+12] ; ecx = a_len
1.200 + movd mm1, [ebp+16] ; mm1 = b
1.201 + mov edi, [ebp+20]
1.202 + cmp ecx, 0
1.203 + je L_16 ; jmp if a_len == 0
1.204 + mov esi, [ebp+8] ; esi = a
1.205 + cld
1.206 +L_15:
1.207 + movd mm0, [esi] ; mm0 = *a++
1.208 + add esi, 4
1.209 + pmuludq mm0, mm1 ; mm0 = b * *a++
1.210 + paddq mm2, mm0 ; add the carry
1.211 + movd mm0, [edi]
1.212 + paddq mm2, mm0 ; add the carry
1.213 + movd [edi], mm2 ; store the 32bit result
1.214 + add edi, 4
1.215 + psrlq mm2, 32 ; save the carry
1.216 + dec ecx ; --a_len
1.217 + jnz L_15 ; jmp if a_len != 0
1.218 +L_16:
1.219 + movd [edi], mm2 ; *c = carry
1.220 + emms
1.221 + pop esi
1.222 + pop edi
1.223 + leave
1.224 + ret
1.225 + nop
1.226 + }
1.227 +}
1.228 +
1.229 +/*
1.230 + * ebp - 36: caller's esi
1.231 + * ebp - 32: caller's edi
1.232 + * ebp - 28:
1.233 + * ebp - 24:
1.234 + * ebp - 20:
1.235 + * ebp - 16:
1.236 + * ebp - 12:
1.237 + * ebp - 8:
1.238 + * ebp - 4:
1.239 + * ebp + 0: caller's ebp
1.240 + * ebp + 4: return address
1.241 + * ebp + 8: a argument
1.242 + * ebp + 12: a_len argument
1.243 + * ebp + 16: b argument
1.244 + * ebp + 20: c argument
1.245 + * registers:
1.246 + * eax:
1.247 + * ebx: carry
1.248 + * ecx: a_len
1.249 + * edx:
1.250 + * esi: a ptr
1.251 + * edi: c ptr
1.252 + */
1.253 +__declspec(naked) void
1.254 +s_mpv_mul_d_add_prop(const mp_digit *a, mp_size a_len, mp_digit b, mp_digit *c)
1.255 +{
1.256 + __asm {
1.257 + mov eax, is_sse
1.258 + cmp eax, 0
1.259 + je s_mpv_mul_d_add_prop_x86
1.260 + jg s_mpv_mul_d_add_prop_sse2
1.261 + call s_mpi_is_sse2
1.262 + mov is_sse, eax
1.263 + cmp eax, 0
1.264 + jg s_mpv_mul_d_add_prop_sse2
1.265 +s_mpv_mul_d_add_prop_x86:
1.266 + push ebp
1.267 + mov ebp,esp
1.268 + sub esp,28
1.269 + push edi
1.270 + push esi
1.271 + push ebx
1.272 + mov ebx,0 ; carry = 0
1.273 + mov ecx,[ebp+12] ; ecx = a_len
1.274 + mov edi,[ebp+20]
1.275 + cmp ecx,0
1.276 + je L_21 ; jmp if a_len == 0
1.277 + cld
1.278 + mov esi,[ebp+8] ; esi = a
1.279 +L_20:
1.280 + lodsd ; eax = [ds:esi]; esi += 4
1.281 + mov edx,[ebp+16] ; edx = b
1.282 + mul edx ; edx:eax = Phi:Plo = a_i * b
1.283 +
1.284 + add eax,ebx ; add carry (ebx) to edx:eax
1.285 + adc edx,0
1.286 + mov ebx,[edi] ; add in current word from *c
1.287 + add eax,ebx
1.288 + adc edx,0
1.289 + mov ebx,edx ; high half of product becomes next carry
1.290 +
1.291 + stosd ; [es:edi] = ax; edi += 4;
1.292 + dec ecx ; --a_len
1.293 + jnz L_20 ; jmp if a_len != 0
1.294 +L_21:
1.295 + cmp ebx,0 ; is carry zero?
1.296 + jz L_23
1.297 + mov eax,[edi] ; add in current word from *c
1.298 + add eax,ebx
1.299 + stosd ; [es:edi] = ax; edi += 4;
1.300 + jnc L_23
1.301 +L_22:
1.302 + mov eax,[edi] ; add in current word from *c
1.303 + adc eax,0
1.304 + stosd ; [es:edi] = ax; edi += 4;
1.305 + jc L_22
1.306 +L_23:
1.307 + pop ebx
1.308 + pop esi
1.309 + pop edi
1.310 + leave
1.311 + ret
1.312 + nop
1.313 +s_mpv_mul_d_add_prop_sse2:
1.314 + push ebp
1.315 + mov ebp, esp
1.316 + push edi
1.317 + push esi
1.318 + push ebx
1.319 + psubq mm2, mm2 ; carry = 0
1.320 + mov ecx, [ebp+12] ; ecx = a_len
1.321 + movd mm1, [ebp+16] ; mm1 = b
1.322 + mov edi, [ebp+20]
1.323 + cmp ecx, 0
1.324 + je L_26 ; jmp if a_len == 0
1.325 + mov esi, [ebp+8] ; esi = a
1.326 + cld
1.327 +L_25:
1.328 + movd mm0, [esi] ; mm0 = *a++
1.329 + movd mm3, [edi] ; fetch the sum
1.330 + add esi, 4
1.331 + pmuludq mm0, mm1 ; mm0 = b * *a++
1.332 + paddq mm2, mm0 ; add the carry
1.333 + paddq mm2, mm3 ; add *c++
1.334 + movd [edi], mm2 ; store the 32bit result
1.335 + add edi, 4
1.336 + psrlq mm2, 32 ; save the carry
1.337 + dec ecx ; --a_len
1.338 + jnz L_25 ; jmp if a_len != 0
1.339 +L_26:
1.340 + movd ebx, mm2
1.341 + cmp ebx, 0 ; is carry zero?
1.342 + jz L_28
1.343 + mov eax, [edi]
1.344 + add eax, ebx
1.345 + stosd
1.346 + jnc L_28
1.347 +L_27:
1.348 + mov eax, [edi] ; add in current word from *c
1.349 + adc eax, 0
1.350 + stosd ; [es:edi] = ax; edi += 4;
1.351 + jc L_27
1.352 +L_28:
1.353 + emms
1.354 + pop ebx
1.355 + pop esi
1.356 + pop edi
1.357 + leave
1.358 + ret
1.359 + nop
1.360 + }
1.361 +}
1.362 +
1.363 +/*
1.364 + * ebp - 20: caller's esi
1.365 + * ebp - 16: caller's edi
1.366 + * ebp - 12:
1.367 + * ebp - 8: carry
1.368 + * ebp - 4: a_len local
1.369 + * ebp + 0: caller's ebp
1.370 + * ebp + 4: return address
1.371 + * ebp + 8: pa argument
1.372 + * ebp + 12: a_len argument
1.373 + * ebp + 16: ps argument
1.374 + * ebp + 20:
1.375 + * registers:
1.376 + * eax:
1.377 + * ebx: carry
1.378 + * ecx: a_len
1.379 + * edx:
1.380 + * esi: a ptr
1.381 + * edi: c ptr
1.382 + */
1.383 +__declspec(naked) void
1.384 +s_mpv_sqr_add_prop(const mp_digit *a, mp_size a_len, mp_digit *sqrs)
1.385 +{
1.386 + __asm {
1.387 + mov eax, is_sse
1.388 + cmp eax, 0
1.389 + je s_mpv_sqr_add_prop_x86
1.390 + jg s_mpv_sqr_add_prop_sse2
1.391 + call s_mpi_is_sse2
1.392 + mov is_sse, eax
1.393 + cmp eax, 0
1.394 + jg s_mpv_sqr_add_prop_sse2
1.395 +s_mpv_sqr_add_prop_x86:
1.396 + push ebp
1.397 + mov ebp,esp
1.398 + sub esp,12
1.399 + push edi
1.400 + push esi
1.401 + push ebx
1.402 + mov ebx,0 ; carry = 0
1.403 + mov ecx,[ebp+12] ; a_len
1.404 + mov edi,[ebp+16] ; edi = ps
1.405 + cmp ecx,0
1.406 + je L_31 ; jump if a_len == 0
1.407 + cld
1.408 + mov esi,[ebp+8] ; esi = pa
1.409 +L_30:
1.410 + lodsd ; eax = [ds:si]; si += 4;
1.411 + mul eax
1.412 +
1.413 + add eax,ebx ; add "carry"
1.414 + adc edx,0
1.415 + mov ebx,[edi]
1.416 + add eax,ebx ; add low word from result
1.417 + mov ebx,[edi+4]
1.418 + stosd ; [es:di] = eax; di += 4;
1.419 + adc edx,ebx ; add high word from result
1.420 + mov ebx,0
1.421 + mov eax,edx
1.422 + adc ebx,0
1.423 + stosd ; [es:di] = eax; di += 4;
1.424 + dec ecx ; --a_len
1.425 + jnz L_30 ; jmp if a_len != 0
1.426 +L_31:
1.427 + cmp ebx,0 ; is carry zero?
1.428 + jz L_34
1.429 + mov eax,[edi] ; add in current word from *c
1.430 + add eax,ebx
1.431 + stosd ; [es:edi] = ax; edi += 4;
1.432 + jnc L_34
1.433 +L_32:
1.434 + mov eax,[edi] ; add in current word from *c
1.435 + adc eax,0
1.436 + stosd ; [es:edi] = ax; edi += 4;
1.437 + jc L_32
1.438 +L_34:
1.439 + pop ebx
1.440 + pop esi
1.441 + pop edi
1.442 + leave
1.443 + ret
1.444 + nop
1.445 +s_mpv_sqr_add_prop_sse2:
1.446 + push ebp
1.447 + mov ebp, esp
1.448 + push edi
1.449 + push esi
1.450 + push ebx
1.451 + psubq mm2, mm2 ; carry = 0
1.452 + mov ecx, [ebp+12] ; ecx = a_len
1.453 + mov edi, [ebp+16]
1.454 + cmp ecx, 0
1.455 + je L_36 ; jmp if a_len == 0
1.456 + mov esi, [ebp+8] ; esi = a
1.457 + cld
1.458 +L_35:
1.459 + movd mm0, [esi] ; mm0 = *a
1.460 + movd mm3, [edi] ; fetch the sum
1.461 + add esi, 4
1.462 + pmuludq mm0, mm0 ; mm0 = sqr(a)
1.463 + paddq mm2, mm0 ; add the carry
1.464 + paddq mm2, mm3 ; add the low word
1.465 + movd mm3, [edi+4]
1.466 + movd [edi], mm2 ; store the 32bit result
1.467 + psrlq mm2, 32
1.468 + paddq mm2, mm3 ; add the high word
1.469 + movd [edi+4], mm2 ; store the 32bit result
1.470 + psrlq mm2, 32 ; save the carry.
1.471 + add edi, 8
1.472 + dec ecx ; --a_len
1.473 + jnz L_35 ; jmp if a_len != 0
1.474 +L_36:
1.475 + movd ebx, mm2
1.476 + cmp ebx, 0 ; is carry zero?
1.477 + jz L_38
1.478 + mov eax, [edi]
1.479 + add eax, ebx
1.480 + stosd
1.481 + jnc L_38
1.482 +L_37:
1.483 + mov eax, [edi] ; add in current word from *c
1.484 + adc eax, 0
1.485 + stosd ; [es:edi] = ax; edi += 4;
1.486 + jc L_37
1.487 +L_38:
1.488 + emms
1.489 + pop ebx
1.490 + pop esi
1.491 + pop edi
1.492 + leave
1.493 + ret
1.494 + nop
1.495 + }
1.496 +}
1.497 +
1.498 +/*
1.499 + * Divide 64-bit (Nhi,Nlo) by 32-bit divisor, which must be normalized
1.500 + * so its high bit is 1. This code is from NSPR.
1.501 + *
1.502 + * Dump of assembler code for function s_mpv_div_2dx1d:
1.503 + *
1.504 + * esp + 0: Caller's ebx
1.505 + * esp + 4: return address
1.506 + * esp + 8: Nhi argument
1.507 + * esp + 12: Nlo argument
1.508 + * esp + 16: divisor argument
1.509 + * esp + 20: qp argument
1.510 + * esp + 24: rp argument
1.511 + * registers:
1.512 + * eax:
1.513 + * ebx: carry
1.514 + * ecx: a_len
1.515 + * edx:
1.516 + * esi: a ptr
1.517 + * edi: c ptr
1.518 + */
1.519 +__declspec(naked) mp_err
1.520 +s_mpv_div_2dx1d(mp_digit Nhi, mp_digit Nlo, mp_digit divisor,
1.521 + mp_digit *qp, mp_digit *rp)
1.522 +{
1.523 + __asm {
1.524 + push ebx
1.525 + mov edx,[esp+8]
1.526 + mov eax,[esp+12]
1.527 + mov ebx,[esp+16]
1.528 + div ebx
1.529 + mov ebx,[esp+20]
1.530 + mov [ebx],eax
1.531 + mov ebx,[esp+24]
1.532 + mov [ebx],edx
1.533 + xor eax,eax ; return zero
1.534 + pop ebx
1.535 + ret
1.536 + nop
1.537 + }
1.538 +}
