1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/security/nss/lib/freebl/mpi/mpi_x86_os2.s Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,538 @@
1.4 +#
1.5 +# This Source Code Form is subject to the terms of the Mozilla Public
1.6 +# License, v. 2.0. If a copy of the MPL was not distributed with this
1.7 +# file, You can obtain one at http://mozilla.org/MPL/2.0/.
1.8 +
1.9 +.data
1.10 +.align 4
1.11 + #
1.12 + # -1 means to call _s_mpi_is_sse to determine if we support sse
1.13 + # instructions.
1.14 + # 0 means to use x86 instructions
1.15 + # 1 means to use sse2 instructions
1.16 +.type is_sse,@object
1.17 +.size is_sse,4
1.18 +is_sse: .long -1
1.19 +
1.20 +#
1.21 +# sigh, handle the difference between -fPIC and not PIC
1.22 +# default to pic, since this file seems to be exclusively
1.23 +# linux right now (solaris uses mpi_i86pc.s and windows uses
1.24 +# mpi_x86_asm.c)
1.25 +#
1.26 +#.ifndef NO_PIC
1.27 +#.macro GET var,reg
1.28 +# movl \var@GOTOFF(%ebx),\reg
1.29 +#.endm
1.30 +#.macro PUT reg,var
1.31 +# movl \reg,\var@GOTOFF(%ebx)
1.32 +#.endm
1.33 +#.else
1.34 +.macro GET var,reg
1.35 + movl \var,\reg
1.36 +.endm
1.37 +.macro PUT reg,var
1.38 + movl \reg,\var
1.39 +.endm
1.40 +#.endif
1.41 +
1.42 +.text
1.43 +
1.44 +
1.45 + # ebp - 36: caller's esi
1.46 + # ebp - 32: caller's edi
1.47 + # ebp - 28:
1.48 + # ebp - 24:
1.49 + # ebp - 20:
1.50 + # ebp - 16:
1.51 + # ebp - 12:
1.52 + # ebp - 8:
1.53 + # ebp - 4:
1.54 + # ebp + 0: caller's ebp
1.55 + # ebp + 4: return address
1.56 + # ebp + 8: a argument
1.57 + # ebp + 12: a_len argument
1.58 + # ebp + 16: b argument
1.59 + # ebp + 20: c argument
1.60 + # registers:
1.61 + # eax:
1.62 + # ebx: carry
1.63 + # ecx: a_len
1.64 + # edx:
1.65 + # esi: a ptr
1.66 + # edi: c ptr
1.67 +.globl _s_mpv_mul_d
1.68 +.type _s_mpv_mul_d,@function
1.69 +_s_mpv_mul_d:
1.70 + GET is_sse,%eax
1.71 + cmp $0,%eax
1.72 + je _s_mpv_mul_d_x86
1.73 + jg _s_mpv_mul_d_sse2
1.74 + call _s_mpi_is_sse2
1.75 + PUT %eax,is_sse
1.76 + cmp $0,%eax
1.77 + jg _s_mpv_mul_d_sse2
1.78 +_s_mpv_mul_d_x86:
1.79 + push %ebp
1.80 + mov %esp,%ebp
1.81 + sub $28,%esp
1.82 + push %edi
1.83 + push %esi
1.84 + push %ebx
1.85 + movl $0,%ebx # carry = 0
1.86 + mov 12(%ebp),%ecx # ecx = a_len
1.87 + mov 20(%ebp),%edi
1.88 + cmp $0,%ecx
1.89 + je 2f # jmp if a_len == 0
1.90 + mov 8(%ebp),%esi # esi = a
1.91 + cld
1.92 +1:
1.93 + lodsl # eax = [ds:esi]; esi += 4
1.94 + mov 16(%ebp),%edx # edx = b
1.95 + mull %edx # edx:eax = Phi:Plo = a_i * b
1.96 +
1.97 + add %ebx,%eax # add carry (%ebx) to edx:eax
1.98 + adc $0,%edx
1.99 + mov %edx,%ebx # high half of product becomes next carry
1.100 +
1.101 + stosl # [es:edi] = ax; edi += 4;
1.102 + dec %ecx # --a_len
1.103 + jnz 1b # jmp if a_len != 0
1.104 +2:
1.105 + mov %ebx,0(%edi) # *c = carry
1.106 + pop %ebx
1.107 + pop %esi
1.108 + pop %edi
1.109 + leave
1.110 + ret
1.111 + nop
1.112 +_s_mpv_mul_d_sse2:
1.113 + push %ebp
1.114 + mov %esp,%ebp
1.115 + push %edi
1.116 + push %esi
1.117 + psubq %mm2,%mm2 # carry = 0
1.118 + mov 12(%ebp),%ecx # ecx = a_len
1.119 + movd 16(%ebp),%mm1 # mm1 = b
1.120 + mov 20(%ebp),%edi
1.121 + cmp $0,%ecx
1.122 + je 6f # jmp if a_len == 0
1.123 + mov 8(%ebp),%esi # esi = a
1.124 + cld
1.125 +5:
1.126 + movd 0(%esi),%mm0 # mm0 = *a++
1.127 + add $4,%esi
1.128 + pmuludq %mm1,%mm0 # mm0 = b * *a++
1.129 + paddq %mm0,%mm2 # add the carry
1.130 + movd %mm2,0(%edi) # store the 32bit result
1.131 + add $4,%edi
1.132 + psrlq $32, %mm2 # save the carry
1.133 + dec %ecx # --a_len
1.134 + jnz 5b # jmp if a_len != 0
1.135 +6:
1.136 + movd %mm2,0(%edi) # *c = carry
1.137 + emms
1.138 + pop %esi
1.139 + pop %edi
1.140 + leave
1.141 + ret
1.142 + nop
1.143 +
1.144 + # ebp - 36: caller's esi
1.145 + # ebp - 32: caller's edi
1.146 + # ebp - 28:
1.147 + # ebp - 24:
1.148 + # ebp - 20:
1.149 + # ebp - 16:
1.150 + # ebp - 12:
1.151 + # ebp - 8:
1.152 + # ebp - 4:
1.153 + # ebp + 0: caller's ebp
1.154 + # ebp + 4: return address
1.155 + # ebp + 8: a argument
1.156 + # ebp + 12: a_len argument
1.157 + # ebp + 16: b argument
1.158 + # ebp + 20: c argument
1.159 + # registers:
1.160 + # eax:
1.161 + # ebx: carry
1.162 + # ecx: a_len
1.163 + # edx:
1.164 + # esi: a ptr
1.165 + # edi: c ptr
1.166 +.globl _s_mpv_mul_d_add
1.167 +.type _s_mpv_mul_d_add,@function
1.168 +_s_mpv_mul_d_add:
1.169 + GET is_sse,%eax
1.170 + cmp $0,%eax
1.171 + je _s_mpv_mul_d_add_x86
1.172 + jg _s_mpv_mul_d_add_sse2
1.173 + call _s_mpi_is_sse2
1.174 + PUT %eax,is_sse
1.175 + cmp $0,%eax
1.176 + jg _s_mpv_mul_d_add_sse2
1.177 +_s_mpv_mul_d_add_x86:
1.178 + push %ebp
1.179 + mov %esp,%ebp
1.180 + sub $28,%esp
1.181 + push %edi
1.182 + push %esi
1.183 + push %ebx
1.184 + movl $0,%ebx # carry = 0
1.185 + mov 12(%ebp),%ecx # ecx = a_len
1.186 + mov 20(%ebp),%edi
1.187 + cmp $0,%ecx
1.188 + je 11f # jmp if a_len == 0
1.189 + mov 8(%ebp),%esi # esi = a
1.190 + cld
1.191 +10:
1.192 + lodsl # eax = [ds:esi]; esi += 4
1.193 + mov 16(%ebp),%edx # edx = b
1.194 + mull %edx # edx:eax = Phi:Plo = a_i * b
1.195 +
1.196 + add %ebx,%eax # add carry (%ebx) to edx:eax
1.197 + adc $0,%edx
1.198 + mov 0(%edi),%ebx # add in current word from *c
1.199 + add %ebx,%eax
1.200 + adc $0,%edx
1.201 + mov %edx,%ebx # high half of product becomes next carry
1.202 +
1.203 + stosl # [es:edi] = ax; edi += 4;
1.204 + dec %ecx # --a_len
1.205 + jnz 10b # jmp if a_len != 0
1.206 +11:
1.207 + mov %ebx,0(%edi) # *c = carry
1.208 + pop %ebx
1.209 + pop %esi
1.210 + pop %edi
1.211 + leave
1.212 + ret
1.213 + nop
1.214 +_s_mpv_mul_d_add_sse2:
1.215 + push %ebp
1.216 + mov %esp,%ebp
1.217 + push %edi
1.218 + push %esi
1.219 + psubq %mm2,%mm2 # carry = 0
1.220 + mov 12(%ebp),%ecx # ecx = a_len
1.221 + movd 16(%ebp),%mm1 # mm1 = b
1.222 + mov 20(%ebp),%edi
1.223 + cmp $0,%ecx
1.224 + je 16f # jmp if a_len == 0
1.225 + mov 8(%ebp),%esi # esi = a
1.226 + cld
1.227 +15:
1.228 + movd 0(%esi),%mm0 # mm0 = *a++
1.229 + add $4,%esi
1.230 + pmuludq %mm1,%mm0 # mm0 = b * *a++
1.231 + paddq %mm0,%mm2 # add the carry
1.232 + movd 0(%edi),%mm0
1.233 + paddq %mm0,%mm2 # add the carry
1.234 + movd %mm2,0(%edi) # store the 32bit result
1.235 + add $4,%edi
1.236 + psrlq $32, %mm2 # save the carry
1.237 + dec %ecx # --a_len
1.238 + jnz 15b # jmp if a_len != 0
1.239 +16:
1.240 + movd %mm2,0(%edi) # *c = carry
1.241 + emms
1.242 + pop %esi
1.243 + pop %edi
1.244 + leave
1.245 + ret
1.246 + nop
1.247 +
1.248 + # ebp - 8: caller's esi
1.249 + # ebp - 4: caller's edi
1.250 + # ebp + 0: caller's ebp
1.251 + # ebp + 4: return address
1.252 + # ebp + 8: a argument
1.253 + # ebp + 12: a_len argument
1.254 + # ebp + 16: b argument
1.255 + # ebp + 20: c argument
1.256 + # registers:
1.257 + # eax:
1.258 + # ebx: carry
1.259 + # ecx: a_len
1.260 + # edx:
1.261 + # esi: a ptr
1.262 + # edi: c ptr
1.263 +.globl _s_mpv_mul_d_add_prop
1.264 +.type _s_mpv_mul_d_add_prop,@function
1.265 +_s_mpv_mul_d_add_prop:
1.266 + GET is_sse,%eax
1.267 + cmp $0,%eax
1.268 + je _s_mpv_mul_d_add_prop_x86
1.269 + jg _s_mpv_mul_d_add_prop_sse2
1.270 + call _s_mpi_is_sse2
1.271 + PUT %eax,is_sse
1.272 + cmp $0,%eax
1.273 + jg _s_mpv_mul_d_add_prop_sse2
1.274 +_s_mpv_mul_d_add_prop_x86:
1.275 + push %ebp
1.276 + mov %esp,%ebp
1.277 + sub $28,%esp
1.278 + push %edi
1.279 + push %esi
1.280 + push %ebx
1.281 + movl $0,%ebx # carry = 0
1.282 + mov 12(%ebp),%ecx # ecx = a_len
1.283 + mov 20(%ebp),%edi
1.284 + cmp $0,%ecx
1.285 + je 21f # jmp if a_len == 0
1.286 + cld
1.287 + mov 8(%ebp),%esi # esi = a
1.288 +20:
1.289 + lodsl # eax = [ds:esi]; esi += 4
1.290 + mov 16(%ebp),%edx # edx = b
1.291 + mull %edx # edx:eax = Phi:Plo = a_i * b
1.292 +
1.293 + add %ebx,%eax # add carry (%ebx) to edx:eax
1.294 + adc $0,%edx
1.295 + mov 0(%edi),%ebx # add in current word from *c
1.296 + add %ebx,%eax
1.297 + adc $0,%edx
1.298 + mov %edx,%ebx # high half of product becomes next carry
1.299 +
1.300 + stosl # [es:edi] = ax; edi += 4;
1.301 + dec %ecx # --a_len
1.302 + jnz 20b # jmp if a_len != 0
1.303 +21:
1.304 + cmp $0,%ebx # is carry zero?
1.305 + jz 23f
1.306 + mov 0(%edi),%eax # add in current word from *c
1.307 + add %ebx,%eax
1.308 + stosl # [es:edi] = ax; edi += 4;
1.309 + jnc 23f
1.310 +22:
1.311 + mov 0(%edi),%eax # add in current word from *c
1.312 + adc $0,%eax
1.313 + stosl # [es:edi] = ax; edi += 4;
1.314 + jc 22b
1.315 +23:
1.316 + pop %ebx
1.317 + pop %esi
1.318 + pop %edi
1.319 + leave
1.320 + ret
1.321 + nop
1.322 +_s_mpv_mul_d_add_prop_sse2:
1.323 + push %ebp
1.324 + mov %esp,%ebp
1.325 + push %edi
1.326 + push %esi
1.327 + push %ebx
1.328 + psubq %mm2,%mm2 # carry = 0
1.329 + mov 12(%ebp),%ecx # ecx = a_len
1.330 + movd 16(%ebp),%mm1 # mm1 = b
1.331 + mov 20(%ebp),%edi
1.332 + cmp $0,%ecx
1.333 + je 26f # jmp if a_len == 0
1.334 + mov 8(%ebp),%esi # esi = a
1.335 + cld
1.336 +25:
1.337 + movd 0(%esi),%mm0 # mm0 = *a++
1.338 + movd 0(%edi),%mm3 # fetch the sum
1.339 + add $4,%esi
1.340 + pmuludq %mm1,%mm0 # mm0 = b * *a++
1.341 + paddq %mm0,%mm2 # add the carry
1.342 + paddq %mm3,%mm2 # add *c++
1.343 + movd %mm2,0(%edi) # store the 32bit result
1.344 + add $4,%edi
1.345 + psrlq $32, %mm2 # save the carry
1.346 + dec %ecx # --a_len
1.347 + jnz 25b # jmp if a_len != 0
1.348 +26:
1.349 + movd %mm2,%ebx
1.350 + cmp $0,%ebx # is carry zero?
1.351 + jz 28f
1.352 + mov 0(%edi),%eax
1.353 + add %ebx, %eax
1.354 + stosl
1.355 + jnc 28f
1.356 +27:
1.357 + mov 0(%edi),%eax # add in current word from *c
1.358 + adc $0,%eax
1.359 + stosl # [es:edi] = ax; edi += 4;
1.360 + jc 27b
1.361 +28:
1.362 + emms
1.363 + pop %ebx
1.364 + pop %esi
1.365 + pop %edi
1.366 + leave
1.367 + ret
1.368 + nop
1.369 +
1.370 +
1.371 + # ebp - 20: caller's esi
1.372 + # ebp - 16: caller's edi
1.373 + # ebp - 12:
1.374 + # ebp - 8: carry
1.375 + # ebp - 4: a_len local
1.376 + # ebp + 0: caller's ebp
1.377 + # ebp + 4: return address
1.378 + # ebp + 8: pa argument
1.379 + # ebp + 12: a_len argument
1.380 + # ebp + 16: ps argument
1.381 + # ebp + 20:
1.382 + # registers:
1.383 + # eax:
1.384 + # ebx: carry
1.385 + # ecx: a_len
1.386 + # edx:
1.387 + # esi: a ptr
1.388 + # edi: c ptr
1.389 +
1.390 +.globl _s_mpv_sqr_add_prop
1.391 +.type _s_mpv_sqr_add_prop,@function
1.392 +_s_mpv_sqr_add_prop:
1.393 + GET is_sse,%eax
1.394 + cmp $0,%eax
1.395 + je _s_mpv_sqr_add_prop_x86
1.396 + jg _s_mpv_sqr_add_prop_sse2
1.397 + call _s_mpi_is_sse2
1.398 + PUT %eax,is_sse
1.399 + cmp $0,%eax
1.400 + jg _s_mpv_sqr_add_prop_sse2
1.401 +_s_mpv_sqr_add_prop_x86:
1.402 + push %ebp
1.403 + mov %esp,%ebp
1.404 + sub $12,%esp
1.405 + push %edi
1.406 + push %esi
1.407 + push %ebx
1.408 + movl $0,%ebx # carry = 0
1.409 + mov 12(%ebp),%ecx # a_len
1.410 + mov 16(%ebp),%edi # edi = ps
1.411 + cmp $0,%ecx
1.412 + je 31f # jump if a_len == 0
1.413 + cld
1.414 + mov 8(%ebp),%esi # esi = pa
1.415 +30:
1.416 + lodsl # %eax = [ds:si]; si += 4;
1.417 + mull %eax
1.418 +
1.419 + add %ebx,%eax # add "carry"
1.420 + adc $0,%edx
1.421 + mov 0(%edi),%ebx
1.422 + add %ebx,%eax # add low word from result
1.423 + mov 4(%edi),%ebx
1.424 + stosl # [es:di] = %eax; di += 4;
1.425 + adc %ebx,%edx # add high word from result
1.426 + movl $0,%ebx
1.427 + mov %edx,%eax
1.428 + adc $0,%ebx
1.429 + stosl # [es:di] = %eax; di += 4;
1.430 + dec %ecx # --a_len
1.431 + jnz 30b # jmp if a_len != 0
1.432 +31:
1.433 + cmp $0,%ebx # is carry zero?
1.434 + jz 34f
1.435 + mov 0(%edi),%eax # add in current word from *c
1.436 + add %ebx,%eax
1.437 + stosl # [es:edi] = ax; edi += 4;
1.438 + jnc 34f
1.439 +32:
1.440 + mov 0(%edi),%eax # add in current word from *c
1.441 + adc $0,%eax
1.442 + stosl # [es:edi] = ax; edi += 4;
1.443 + jc 32b
1.444 +34:
1.445 + pop %ebx
1.446 + pop %esi
1.447 + pop %edi
1.448 + leave
1.449 + ret
1.450 + nop
1.451 +_s_mpv_sqr_add_prop_sse2:
1.452 + push %ebp
1.453 + mov %esp,%ebp
1.454 + push %edi
1.455 + push %esi
1.456 + push %ebx
1.457 + psubq %mm2,%mm2 # carry = 0
1.458 + mov 12(%ebp),%ecx # ecx = a_len
1.459 + mov 16(%ebp),%edi
1.460 + cmp $0,%ecx
1.461 + je 36f # jmp if a_len == 0
1.462 + mov 8(%ebp),%esi # esi = a
1.463 + cld
1.464 +35:
1.465 + movd 0(%esi),%mm0 # mm0 = *a
1.466 + movd 0(%edi),%mm3 # fetch the sum
1.467 + add $4,%esi
1.468 + pmuludq %mm0,%mm0 # mm0 = sqr(a)
1.469 + paddq %mm0,%mm2 # add the carry
1.470 + paddq %mm3,%mm2 # add the low word
1.471 + movd 4(%edi),%mm3
1.472 + movd %mm2,0(%edi) # store the 32bit result
1.473 + psrlq $32, %mm2
1.474 + paddq %mm3,%mm2 # add the high word
1.475 + movd %mm2,4(%edi) # store the 32bit result
1.476 + psrlq $32, %mm2 # save the carry.
1.477 + add $8,%edi
1.478 + dec %ecx # --a_len
1.479 + jnz 35b # jmp if a_len != 0
1.480 +36:
1.481 + movd %mm2,%ebx
1.482 + cmp $0,%ebx # is carry zero?
1.483 + jz 38f
1.484 + mov 0(%edi),%eax
1.485 + add %ebx, %eax
1.486 + stosl
1.487 + jnc 38f
1.488 +37:
1.489 + mov 0(%edi),%eax # add in current word from *c
1.490 + adc $0,%eax
1.491 + stosl # [es:edi] = ax; edi += 4;
1.492 + jc 37b
1.493 +38:
1.494 + emms
1.495 + pop %ebx
1.496 + pop %esi
1.497 + pop %edi
1.498 + leave
1.499 + ret
1.500 + nop
1.501 +
1.502 + #
1.503 + # Divide 64-bit (Nhi,Nlo) by 32-bit divisor, which must be normalized
1.504 + # so its high bit is 1. This code is from NSPR.
1.505 + #
1.506 + # mp_err _s_mpv_div_2dx1d(mp_digit Nhi, mp_digit Nlo, mp_digit divisor,
1.507 + # mp_digit *qp, mp_digit *rp)
1.508 +
1.509 + # esp + 0: Caller's ebx
1.510 + # esp + 4: return address
1.511 + # esp + 8: Nhi argument
1.512 + # esp + 12: Nlo argument
1.513 + # esp + 16: divisor argument
1.514 + # esp + 20: qp argument
1.515 + # esp + 24: rp argument
1.516 + # registers:
1.517 + # eax:
1.518 + # ebx: carry
1.519 + # ecx: a_len
1.520 + # edx:
1.521 + # esi: a ptr
1.522 + # edi: c ptr
1.523 + #
1.524 +
1.525 +.globl _s_mpv_div_2dx1d
1.526 +.type _s_mpv_div_2dx1d,@function
1.527 +_s_mpv_div_2dx1d:
1.528 + push %ebx
1.529 + mov 8(%esp),%edx
1.530 + mov 12(%esp),%eax
1.531 + mov 16(%esp),%ebx
1.532 + div %ebx
1.533 + mov 20(%esp),%ebx
1.534 + mov %eax,0(%ebx)
1.535 + mov 24(%esp),%ebx
1.536 + mov %edx,0(%ebx)
1.537 + xor %eax,%eax # return zero
1.538 + pop %ebx
1.539 + ret
1.540 + nop
1.541 +
