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

Eliminate 64-bit alignment limitation in sparcv9a-mont.

Browse Source
This commit is contained in:
Andy Polyakov 2006-12-08 15:18:41 +00:00
parent 98c1509f34
commit 87d3af6475
1 changed files with 27 additions and 22 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

49
crypto/bn/asm/sparcv9a-mont.pl
View File

@ -37,8 +37,6 @@
# input arguments: # input arguments:
# - num may not be less than 4; # - num may not be less than 4;
# - num has to be even; # - num has to be even;
# - ap, bp, rp, np has to be 64-bit aligned [which is not a problem
# as long as BIGNUM.d are malloc-ated];
# Failure to meet either condition has no fatal effects, simply # Failure to meet either condition has no fatal effects, simply
# doesn't give any performance gain. # doesn't give any performance gain.
@ -183,12 +181,12 @@ $fname:
add $ap,$j,%o3 add $ap,$j,%o3
add $bp,$i,%o4 add $bp,$i,%o4
ldx [$bp+$i],%o0 ! bp[0] ld [%o3+4],%g1 ! bp[0]
ldx [$ap+$j],%o1 ! ap[0] ld [%o3+0],%o0
sllx %o0,32,%g1 ld [%o4+4],%g5 ! ap[0]
sllx %o1,32,%g5 sllx %g1,32,%g1
srlx %o0,32,%o0 ld [%o4+0],%o1
srlx %o1,32,%o1 sllx %g5,32,%g5
or %g1,%o0,%o0 or %g1,%o0,%o0
or %g5,%o1,%o1 or %g5,%o1,%o1
@ -510,14 +508,15 @@ $fname:
sub %g0,$num,$j ! j=-num sub %g0,$num,$j ! j=-num
add %sp,$bias+$frame+$locals,$tp add %sp,$bias+$frame+$locals,$tp
add $ap,$j,%o3
add $bp,$i,%o4 add $bp,$i,%o4
ldx [$bp+$i],%o0 ! bp[i] ld [%o3+4],%g1 ! bp[i]
ldx [$ap+$j],%o1 ! ap[0] ld [%o3+0],%o0
sllx %o0,32,%g1 ld [%o4+4],%g5 ! ap[0]
sllx %o1,32,%g5 sllx %g1,32,%g1
srlx %o0,32,%o0 ld [%o4+0],%o1
srlx %o1,32,%o1 sllx %g5,32,%g5
or %g1,%o0,%o0 or %g1,%o0,%o0
or %g5,%o1,%o1 or %g5,%o1,%o1
@ -818,14 +817,18 @@ $fname:
.align 32,0x1000000 .align 32,0x1000000
.Lsub: .Lsub:
ldd [$tp+%o7],%o0 ldx [$tp+%o7],%o0
ldd [$np+%o7],%o2 add $np,%o7,%g1
subccc %o1,%o2,%o2 ld [%g1+0],%o2
subccc %o0,%o3,%o3 ld [%g1+4],%o3
std %o2,[$rp+%o7] srlx %o0,32,%o1
subccc %o0,%o2,%o2
add $rp,%o7,%g1
subccc %o1,%o3,%o3
st %o2,[%g1+0]
add %o7,8,%o7 add %o7,8,%o7
brnz,pt %o7,.Lsub brnz,pt %o7,.Lsub
nop st %o3,[%g1+4]
subccc $carry,0,$carry subccc $carry,0,$carry
bcc,pt %icc,.Lzap bcc,pt %icc,.Lzap
sub %g0,$num,%o7 ! n=-num sub %g0,$num,%o7 ! n=-num
@ -834,10 +837,11 @@ $fname:
.Lcopy: .Lcopy:
ldx [$tp+%o7],%o0 ldx [$tp+%o7],%o0
srlx %o0,32,%o1 srlx %o0,32,%o1
std %o0,[$rp+%o7] add $rp,%o7,%g1
st %o0,[%g1+0]
add %o7,8,%o7 add %o7,8,%o7
brnz,pt %o7,.Lcopy brnz,pt %o7,.Lcopy
nop st %o1,[%g1+4]
ba .Lzap ba .Lzap
sub %g0,$num,%o7 ! n=-num sub %g0,$num,%o7 ! n=-num
@ -861,6 +865,7 @@ $fname:
restore restore
.type $fname,#function .type $fname,#function
.size $fname,(.-$fname) .size $fname,(.-$fname)
.asciz "Montgomery Multipltication for UltraSPARC, CRYPTOGAMS by <appro\@openssl.org>"
___ ___
$code =~ s/\`([^\`]*)\`/eval($1)/gem; $code =~ s/\`([^\`]*)\`/eval($1)/gem;