2 # Copyright 2010-2016 The OpenSSL Project Authors. All Rights Reserved.
4 # Licensed under the OpenSSL license (the "License"). You may not use
5 # this file except in compliance with the License. You can obtain a copy
6 # in the file LICENSE in the source distribution or at
7 # https://www.openssl.org/source/license.html
10 # ====================================================================
11 # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
12 # project. The module is, however, dual licensed under OpenSSL and
13 # CRYPTOGAMS licenses depending on where you obtain it. For further
14 # details see http://www.openssl.org/~appro/cryptogams/.
15 # ====================================================================
17 # This module doesn't present direct interest for OpenSSL, because it
18 # doesn't provide better performance for longer keys, at least not on
19 # in-order-execution cores. While 512-bit RSA sign operations can be
20 # 65% faster in 64-bit mode, 1024-bit ones are only 15% faster, and
21 # 4096-bit ones are up to 15% slower. In 32-bit mode it varies from
22 # 16% improvement for 512-bit RSA sign to -33% for 4096-bit RSA
23 # verify:-( All comparisons are against bn_mul_mont-free assembler.
24 # The module might be of interest to embedded system developers, as
25 # the code is smaller than 1KB, yet offers >3x improvement on MIPS64
26 # and 75-30% [less for longer keys] on MIPS32 over compiler-generated
29 ######################################################################
30 # There is a number of MIPS ABI in use, O32 and N32/64 are most
31 # widely used. Then there is a new contender: NUBI. It appears that if
32 # one picks the latter, it's possible to arrange code in ABI neutral
33 # manner. Therefore let's stick to NUBI register layout:
35 ($zero,$at,$t0,$t1,$t2)=map("\$$_",(0..2,24,25));
36 ($a0,$a1,$a2,$a3,$a4,$a5,$a6,$a7)=map("\$$_",(4..11));
37 ($s0,$s1,$s2,$s3,$s4,$s5,$s6,$s7,$s8,$s9,$s10,$s11)=map("\$$_",(12..23));
38 ($gp,$tp,$sp,$fp,$ra)=map("\$$_",(3,28..31));
40 # The return value is placed in $a0. Following coding rules facilitate
43 # - never ever touch $tp, "thread pointer", former $gp;
44 # - copy return value to $t0, former $v0 [or to $a0 if you're adapting
46 # - on O32 populate $a4-$a7 with 'lw $aN,4*N($sp)' if necessary;
48 # For reference here is register layout for N32/64 MIPS ABIs:
50 # ($zero,$at,$v0,$v1)=map("\$$_",(0..3));
51 # ($a0,$a1,$a2,$a3,$a4,$a5,$a6,$a7)=map("\$$_",(4..11));
52 # ($t0,$t1,$t2,$t3,$t8,$t9)=map("\$$_",(12..15,24,25));
53 # ($s0,$s1,$s2,$s3,$s4,$s5,$s6,$s7)=map("\$$_",(16..23));
54 # ($gp,$sp,$fp,$ra)=map("\$$_",(28..31));
56 $flavour = shift || "o32"; # supported flavours are o32,n32,64,nubi32,nubi64
58 if ($flavour =~ /64|n32/i) {
59 $PTR_ADD="dadd"; # incidentally works even on n32
60 $PTR_SUB="dsub"; # incidentally works even on n32
71 $SAVED_REGS_MASK = ($flavour =~ /nubi/i) ? 0x00fff000 : 0x00ff0000;
75 ######################################################################
77 while (($output=shift) && ($output!~/\w[\w\-]*\.\w+$/)) {}
78 open STDOUT,">$output";
80 if ($flavour =~ /64|n32/i) {
97 $rp=$a0; # BN_ULONG *rp,
98 $ap=$a1; # const BN_ULONG *ap,
99 $bp=$a2; # const BN_ULONG *bp,
100 $np=$a3; # const BN_ULONG *np,
101 $n0=$a4; # const BN_ULONG *n0,
102 $num=$a5; # int num);
134 $code.=<<___ if ($flavour =~ /o32/i);
142 slt $at,$num,17 # on in-order CPU
143 bnez $at,bn_mul_mont_internal
150 .ent bn_mul_mont_internal
151 bn_mul_mont_internal:
152 .frame $fp,$FRAMESIZE*$SZREG,$ra
153 .mask 0x40000000|$SAVED_REGS_MASK,-$SZREG
154 $PTR_SUB $sp,$FRAMESIZE*$SZREG
155 $REG_S $fp,($FRAMESIZE-1)*$SZREG($sp)
156 $REG_S $s11,($FRAMESIZE-2)*$SZREG($sp)
157 $REG_S $s10,($FRAMESIZE-3)*$SZREG($sp)
158 $REG_S $s9,($FRAMESIZE-4)*$SZREG($sp)
159 $REG_S $s8,($FRAMESIZE-5)*$SZREG($sp)
160 $REG_S $s7,($FRAMESIZE-6)*$SZREG($sp)
161 $REG_S $s6,($FRAMESIZE-7)*$SZREG($sp)
162 $REG_S $s5,($FRAMESIZE-8)*$SZREG($sp)
163 $REG_S $s4,($FRAMESIZE-9)*$SZREG($sp)
165 $code.=<<___ if ($flavour =~ /nubi/i);
166 $REG_S $s3,($FRAMESIZE-10)*$SZREG($sp)
167 $REG_S $s2,($FRAMESIZE-11)*$SZREG($sp)
168 $REG_S $s1,($FRAMESIZE-12)*$SZREG($sp)
169 $REG_S $s0,($FRAMESIZE-13)*$SZREG($sp)
176 $LD $bi,0($bp) # bp[0]
177 $LD $aj,0($ap) # ap[0]
178 $LD $nj,0($np) # np[0]
180 $PTR_SUB $sp,2*$BNSZ # place for two extra words
181 sll $num,`log($BNSZ)/log(2)`
355 $ST $hi1,2*$BNSZ($tp)
362 $PTR_ADD $tj,$sp,$num # &tp[num]
365 li $hi0,0 # clear borrow bit
368 .Lsub: $LD $lo0,($tp)
372 $SUBU $lo1,$lo0,$lo1 # tp[i]-np[i]
382 $SUBU $hi0,$hi1,$hi0 # handle upmost overflow bit
384 $PTR_SUB $rp,$num # restore rp
389 or $ap,$ap,$bp # ap=borrow?tp:rp
392 .Lcopy: $LD $aj,($ap)
406 $REG_L $fp,($FRAMESIZE-1)*$SZREG($sp)
407 $REG_L $s11,($FRAMESIZE-2)*$SZREG($sp)
408 $REG_L $s10,($FRAMESIZE-3)*$SZREG($sp)
409 $REG_L $s9,($FRAMESIZE-4)*$SZREG($sp)
410 $REG_L $s8,($FRAMESIZE-5)*$SZREG($sp)
411 $REG_L $s7,($FRAMESIZE-6)*$SZREG($sp)
412 $REG_L $s6,($FRAMESIZE-7)*$SZREG($sp)
413 $REG_L $s5,($FRAMESIZE-8)*$SZREG($sp)
414 $REG_L $s4,($FRAMESIZE-9)*$SZREG($sp)
416 $code.=<<___ if ($flavour =~ /nubi/i);
417 $REG_L $s3,($FRAMESIZE-10)*$SZREG($sp)
418 $REG_L $s2,($FRAMESIZE-11)*$SZREG($sp)
419 $REG_L $s1,($FRAMESIZE-12)*$SZREG($sp)
420 $REG_L $s0,($FRAMESIZE-13)*$SZREG($sp)
424 $PTR_ADD $sp,$FRAMESIZE*$SZREG
425 .end bn_mul_mont_internal
427 .asciiz "Montgomery Multiplication for MIPS, CRYPTOGAMS by <appro\@openssl.org>"
430 $code =~ s/\`([^\`]*)\`/eval $1/gem;