generic-library/openssl
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Minor, +10%, tune-up for x86_64-mont.pl.

Browse Source
This commit is contained in:
Andy Polyakov 2006-12-08 10:13:51 +00:00
parent 8583eba015
commit d28134b8f3
1 changed files with 16 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

31
crypto/bn/asm/x86_64-mont.pl
View File

@ -2,8 +2,9 @@
# ==================================================================== # ====================================================================
# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL # Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
# project. Rights for redistribution and usage in source and binary # project. The module is, however, dual licensed under OpenSSL and
# forms are granted according to the OpenSSL license. # CRYPTOGAMS licenses depending on where you obtain it. For further
# details see http://www.openssl.org/~appro/cryptogams/.
# ==================================================================== # ====================================================================
# October 2005. # October 2005.
@ -81,19 +82,18 @@ bn_mul_mont:
add $hi0,%rax add $hi0,%rax
adc \$0,%rdx adc \$0,%rdx
mov %rax,$lo0 mov %rax,$lo0
mov ($np,$j,8),%rax
mov %rdx,$hi0 mov %rdx,$hi0
mov ($np,$j,8),%rax
mulq $m1 # np[j]*m1 mulq $m1 # np[j]*m1
add $hi1,%rax add $hi1,%rax
lea 1($j),$j # j++
adc \$0,%rdx adc \$0,%rdx
add $lo0,%rax # np[j]*m1+ap[j]*bp[0] add $lo0,%rax # np[j]*m1+ap[j]*bp[0]
adc \$0,%rdx adc \$0,%rdx
mov %rax,-8(%rsp,$j,8) # tp[j-1] mov %rax,-16(%rsp,$j,8) # tp[j-1]
mov %rdx,$hi1
lea 1($j),$j # j++
cmp $num,$j cmp $num,$j
mov %rdx,$hi1
jl .L1st jl .L1st
xor %rdx,%rdx xor %rdx,%rdx
@ -120,6 +120,7 @@ bn_mul_mont:
mulq ($np,$j,8) # np[0]*m1 mulq ($np,$j,8) # np[0]*m1
add $lo0,%rax # discarded add $lo0,%rax # discarded
mov 8(%rsp),$lo0 # tp[1]
adc \$0,%rdx adc \$0,%rdx
mov %rdx,$hi1 mov %rdx,$hi1
@ -130,28 +131,27 @@ bn_mul_mont:
mulq $m0 # ap[j]*bp[i] mulq $m0 # ap[j]*bp[i]
add $hi0,%rax add $hi0,%rax
adc \$0,%rdx adc \$0,%rdx
add (%rsp,$j,8),%rax # ap[j]*bp[i]+tp[j] add %rax,$lo0 # ap[j]*bp[i]+tp[j]
mov ($np,$j,8),%rax
adc \$0,%rdx adc \$0,%rdx
mov %rax,$lo0
mov %rdx,$hi0 mov %rdx,$hi0
mov ($np,$j,8),%rax
mulq $m1 # np[j]*m1 mulq $m1 # np[j]*m1
add $hi1,%rax add $hi1,%rax
lea 1($j),$j # j++
adc \$0,%rdx adc \$0,%rdx
add $lo0,%rax # np[j]*m1+ap[j]*bp[i]+tp[j] add $lo0,%rax # np[j]*m1+ap[j]*bp[i]+tp[j]
adc \$0,%rdx adc \$0,%rdx
mov %rax,-8(%rsp,$j,8) # tp[j-1] mov (%rsp,$j,8),$lo0
mov %rdx,$hi1
lea 1($j),$j # j++
cmp $num,$j cmp $num,$j
mov %rax,-16(%rsp,$j,8) # tp[j-1]
mov %rdx,$hi1
jl .Linner jl .Linner
xor %rdx,%rdx xor %rdx,%rdx
add $hi0,$hi1 add $hi0,$hi1
adc \$0,%rdx adc \$0,%rdx
add (%rsp,$num,8),$hi1 # pull upmost overflow bit add $lo0,$hi1 # pull upmost overflow bit
adc \$0,%rdx adc \$0,%rdx
mov $hi1,-8(%rsp,$num,8) mov $hi1,-8(%rsp,$num,8)
mov %rdx,(%rsp,$num,8) # store upmost overflow bit mov %rdx,(%rsp,$num,8) # store upmost overflow bit
@ -202,6 +202,7 @@ bn_mul_mont:
jge .Lzap jge .Lzap
jmp .Lexit jmp .Lexit
.size bn_mul_mont,.-bn_mul_mont .size bn_mul_mont,.-bn_mul_mont
.asciz "Montgomery Multiplication for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
___ ___
print $code; print $code;