bn/asm/*-mont.pl: harmonize with BN_from_montgomery_word.

[openssl.git] / crypto / bn / asm / x86_64-mont5.pl

diff --git a/crypto/bn/asm/x86_64-mont5.pl b/crypto/bn/asm/x86_64-mont5.pl
index 5779059ea268eefd9fea04985b9a0ec4e4785ea1..31c60d96348419cf300dedc68eaf7024eaf3472d 100755 (executable)
--- a/crypto/bn/asm/x86_64-mont5.pl
+++ b/crypto/bn/asm/x86_64-mont5.pl
@@ -414,18 +414,19 @@ $code.=<<___;
        jnz     .Lsub
 
        sbb     \$0,%rax                # handle upmost overflow bit
+       mov     \$-1,%rbx
+       xor     %rax,%rbx
        xor     $i,$i
-       and     %rax,$ap
-       not     %rax
-       mov     $rp,$np
-       and     %rax,$np
        mov     $num,$j                 # j=num
-       or      $np,$ap                 # ap=borrow?tp:rp
-.align 16
-.Lcopy:                                        # copy or in-place refresh
-       mov     ($ap,$i,8),%rax
+
+.Lcopy:                                        # conditional copy
+       mov     ($rp,$i,8),%rcx
+       mov     (%rsp,$i,8),%rdx
+       and     %rbx,%rcx
+       and     %rax,%rdx
        mov     $i,(%rsp,$i,8)          # zap temporary vector
-       mov     %rax,($rp,$i,8)         # rp[i]=tp[i]
+       or      %rcx,%rdx
+       mov     %rdx,($rp,$i,8)         # rp[i]=tp[i]
        lea     1($i),$i
        sub     \$1,$j
        jnz     .Lcopy