diff options
Diffstat (limited to 'openssl-1.1.0h/crypto/camellia/asm')
| -rw-r--r-- | openssl-1.1.0h/crypto/camellia/asm/cmll-x86.pl | 1150 | ||||
| -rw-r--r-- | openssl-1.1.0h/crypto/camellia/asm/cmll-x86_64.pl | 1088 | ||||
| -rw-r--r-- | openssl-1.1.0h/crypto/camellia/asm/cmllt4-sparcv9.pl | 939 |
3 files changed, 3177 insertions, 0 deletions
diff --git a/openssl-1.1.0h/crypto/camellia/asm/cmll-x86.pl b/openssl-1.1.0h/crypto/camellia/asm/cmll-x86.pl new file mode 100644 index 0000000..59f9ed9 --- /dev/null +++ b/openssl-1.1.0h/crypto/camellia/asm/cmll-x86.pl @@ -0,0 +1,1150 @@ +#! /usr/bin/env perl +# Copyright 2008-2016 The OpenSSL Project Authors. All Rights Reserved. +# +# Licensed under the OpenSSL license (the "License"). You may not use +# this file except in compliance with the License. You can obtain a copy +# in the file LICENSE in the source distribution or at +# https://www.openssl.org/source/license.html + + +# ==================================================================== +# Copyright (c) 2008 Andy Polyakov <appro@openssl.org> +# +# This module may be used under the terms of either the GNU General +# Public License version 2 or later, the GNU Lesser General Public +# License version 2.1 or later, the Mozilla Public License version +# 1.1 or the BSD License. The exact terms of either license are +# distributed along with this module. For further details see +# http://www.openssl.org/~appro/camellia/. +# ==================================================================== + +# Performance in cycles per processed byte (less is better) in +# 'openssl speed ...' benchmark: +# +# AMD K8 Core2 PIII P4 +# -evp camellia-128-ecb 21.5 22.8 27.0 28.9 +# + over gcc 3.4.6 +90/11% +70/10% +53/4% +160/64% +# + over icc 8.0 +48/19% +21/15% +21/17% +55/37% +# +# camellia-128-cbc 17.3 21.1 23.9 25.9 +# +# 128-bit key setup 196 280 256 240 cycles/key +# + over gcc 3.4.6 +30/0% +17/11% +11/0% +63/40% +# + over icc 8.0 +18/3% +10/0% +10/3% +21/10% +# +# Pairs of numbers in "+" rows represent performance improvement over +# compiler generated position-independent code, PIC, and non-PIC +# respectively. PIC results are of greater relevance, as this module +# is position-independent, i.e. suitable for a shared library or PIE. +# Position independence "costs" one register, which is why compilers +# are so close with non-PIC results, they have an extra register to +# spare. CBC results are better than ECB ones thanks to "zero-copy" +# private _x86_* interface, and are ~30-40% better than with compiler +# generated cmll_cbc.o, and reach ~80-90% of x86_64 performance on +# same CPU (where applicable). + +$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1; +push(@INC,"${dir}","${dir}../../perlasm"); +require "x86asm.pl"; + +$OPENSSL=1; + +$output = pop; +open STDOUT,">$output"; + +&asm_init($ARGV[0],"cmll-586.pl",$ARGV[$#ARGV] eq "386"); + +@T=("eax","ebx","ecx","edx"); +$idx="esi"; +$key="edi"; +$Tbl="ebp"; + +# stack frame layout in _x86_Camellia_* routines, frame is allocated +# by caller +$__ra=&DWP(0,"esp"); # return address +$__s0=&DWP(4,"esp"); # s0 backing store +$__s1=&DWP(8,"esp"); # s1 backing store +$__s2=&DWP(12,"esp"); # s2 backing store +$__s3=&DWP(16,"esp"); # s3 backing store +$__end=&DWP(20,"esp"); # pointer to end/start of key schedule + +# stack frame layout in Camellia_[en|crypt] routines, which differs from +# above by 4 and overlaps by pointer to end/start of key schedule +$_end=&DWP(16,"esp"); +$_esp=&DWP(20,"esp"); + +# const unsigned int Camellia_SBOX[4][256]; +# Well, sort of... Camellia_SBOX[0][] is interleaved with [1][], +# and [2][] - with [3][]. This is done to optimize code size. +$SBOX1_1110=0; # Camellia_SBOX[0] +$SBOX4_4404=4; # Camellia_SBOX[1] +$SBOX2_0222=2048; # Camellia_SBOX[2] +$SBOX3_3033=2052; # Camellia_SBOX[3] +&static_label("Camellia_SIGMA"); +&static_label("Camellia_SBOX"); + +sub Camellia_Feistel { +my $i=@_[0]; +my $seed=defined(@_[1])?@_[1]:0; +my $scale=$seed<0?-8:8; +my $frame=defined(@_[2])?@_[2]:0; +my $j=($i&1)*2; +my $t0=@T[($j)%4],$t1=@T[($j+1)%4],$t2=@T[($j+2)%4],$t3=@T[($j+3)%4]; + + &xor ($t0,$idx); # t0^=key[0] + &xor ($t1,&DWP($seed+$i*$scale+4,$key)); # t1^=key[1] + &movz ($idx,&HB($t0)); # (t0>>8)&0xff + &mov ($t3,&DWP($SBOX3_3033,$Tbl,$idx,8)); # t3=SBOX3_3033[0] + &movz ($idx,&LB($t0)); # (t0>>0)&0xff + &xor ($t3,&DWP($SBOX4_4404,$Tbl,$idx,8)); # t3^=SBOX4_4404[0] + &shr ($t0,16); + &movz ($idx,&LB($t1)); # (t1>>0)&0xff + &mov ($t2,&DWP($SBOX1_1110,$Tbl,$idx,8)); # t2=SBOX1_1110[1] + &movz ($idx,&HB($t0)); # (t0>>24)&0xff + &xor ($t3,&DWP($SBOX1_1110,$Tbl,$idx,8)); # t3^=SBOX1_1110[0] + &movz ($idx,&HB($t1)); # (t1>>8)&0xff + &xor ($t2,&DWP($SBOX4_4404,$Tbl,$idx,8)); # t2^=SBOX4_4404[1] + &shr ($t1,16); + &movz ($t0,&LB($t0)); # (t0>>16)&0xff + &xor ($t3,&DWP($SBOX2_0222,$Tbl,$t0,8)); # t3^=SBOX2_0222[0] + &movz ($idx,&HB($t1)); # (t1>>24)&0xff + &mov ($t0,&DWP($frame+4*(($j+3)%4),"esp")); # prefetch "s3" + &xor ($t2,$t3); # t2^=t3 + &rotr ($t3,8); # t3=RightRotate(t3,8) + &xor ($t2,&DWP($SBOX2_0222,$Tbl,$idx,8)); # t2^=SBOX2_0222[1] + &movz ($idx,&LB($t1)); # (t1>>16)&0xff + &mov ($t1,&DWP($frame+4*(($j+2)%4),"esp")); # prefetch "s2" + &xor ($t3,$t0); # t3^=s3 + &xor ($t2,&DWP($SBOX3_3033,$Tbl,$idx,8)); # t2^=SBOX3_3033[1] + &mov ($idx,&DWP($seed+($i+1)*$scale,$key)); # prefetch key[i+1] + &xor ($t3,$t2); # t3^=t2 + &mov (&DWP($frame+4*(($j+3)%4),"esp"),$t3); # s3=t3 + &xor ($t2,$t1); # t2^=s2 + &mov (&DWP($frame+4*(($j+2)%4),"esp"),$t2); # s2=t2 +} + +# void Camellia_EncryptBlock_Rounds( +# int grandRounds, +# const Byte plaintext[], +# const KEY_TABLE_TYPE keyTable, +# Byte ciphertext[]) +&function_begin("Camellia_EncryptBlock_Rounds"); + &mov ("eax",&wparam(0)); # load grandRounds + &mov ($idx,&wparam(1)); # load plaintext pointer + &mov ($key,&wparam(2)); # load key schedule pointer + + &mov ("ebx","esp"); + &sub ("esp",7*4); # place for s[0-3],keyEnd,esp and ra + &and ("esp",-64); + + # place stack frame just "above mod 1024" the key schedule + # this ensures that cache associativity of 2 suffices + &lea ("ecx",&DWP(-64-63,$key)); + &sub ("ecx","esp"); + &neg ("ecx"); + &and ("ecx",0x3C0); # modulo 1024, but aligned to cache-line + &sub ("esp","ecx"); + &add ("esp",4); # 4 is reserved for callee's return address + + &shl ("eax",6); + &lea ("eax",&DWP(0,$key,"eax")); + &mov ($_esp,"ebx"); # save %esp + &mov ($_end,"eax"); # save keyEnd + + &call (&label("pic_point")); + &set_label("pic_point"); + &blindpop($Tbl); + &lea ($Tbl,&DWP(&label("Camellia_SBOX")."-".&label("pic_point"),$Tbl)); + + &mov (@T[0],&DWP(0,$idx)); # load plaintext + &mov (@T[1],&DWP(4,$idx)); + &mov (@T[2],&DWP(8,$idx)); + &bswap (@T[0]); + &mov (@T[3],&DWP(12,$idx)); + &bswap (@T[1]); + &bswap (@T[2]); + &bswap (@T[3]); + + &call ("_x86_Camellia_encrypt"); + + &mov ("esp",$_esp); + &bswap (@T[0]); + &mov ($idx,&wparam(3)); # load ciphertext pointer + &bswap (@T[1]); + &bswap (@T[2]); + &bswap (@T[3]); + &mov (&DWP(0,$idx),@T[0]); # write ciphertext + &mov (&DWP(4,$idx),@T[1]); + &mov (&DWP(8,$idx),@T[2]); + &mov (&DWP(12,$idx),@T[3]); +&function_end("Camellia_EncryptBlock_Rounds"); +# V1.x API +&function_begin_B("Camellia_EncryptBlock"); + &mov ("eax",128); + &sub ("eax",&wparam(0)); # load keyBitLength + &mov ("eax",3); + &adc ("eax",0); # keyBitLength==128?3:4 + &mov (&wparam(0),"eax"); + &jmp (&label("Camellia_EncryptBlock_Rounds")); +&function_end_B("Camellia_EncryptBlock"); + +if ($OPENSSL) { +# void Camellia_encrypt( +# const unsigned char *in, +# unsigned char *out, +# const CAMELLIA_KEY *key) +&function_begin("Camellia_encrypt"); + &mov ($idx,&wparam(0)); # load plaintext pointer + &mov ($key,&wparam(2)); # load key schedule pointer + + &mov ("ebx","esp"); + &sub ("esp",7*4); # place for s[0-3],keyEnd,esp and ra + &and ("esp",-64); + &mov ("eax",&DWP(272,$key)); # load grandRounds counter + + # place stack frame just "above mod 1024" the key schedule + # this ensures that cache associativity of 2 suffices + &lea ("ecx",&DWP(-64-63,$key)); + &sub ("ecx","esp"); + &neg ("ecx"); + &and ("ecx",0x3C0); # modulo 1024, but aligned to cache-line + &sub ("esp","ecx"); + &add ("esp",4); # 4 is reserved for callee's return address + + &shl ("eax",6); + &lea ("eax",&DWP(0,$key,"eax")); + &mov ($_esp,"ebx"); # save %esp + &mov ($_end,"eax"); # save keyEnd + + &call (&label("pic_point")); + &set_label("pic_point"); + &blindpop($Tbl); + &lea ($Tbl,&DWP(&label("Camellia_SBOX")."-".&label("pic_point"),$Tbl)); + + &mov (@T[0],&DWP(0,$idx)); # load plaintext + &mov (@T[1],&DWP(4,$idx)); + &mov (@T[2],&DWP(8,$idx)); + &bswap (@T[0]); + &mov (@T[3],&DWP(12,$idx)); + &bswap (@T[1]); + &bswap (@T[2]); + &bswap (@T[3]); + + &call ("_x86_Camellia_encrypt"); + + &mov ("esp",$_esp); + &bswap (@T[0]); + &mov ($idx,&wparam(1)); # load ciphertext pointer + &bswap (@T[1]); + &bswap (@T[2]); + &bswap (@T[3]); + &mov (&DWP(0,$idx),@T[0]); # write ciphertext + &mov (&DWP(4,$idx),@T[1]); + &mov (&DWP(8,$idx),@T[2]); + &mov (&DWP(12,$idx),@T[3]); +&function_end("Camellia_encrypt"); +} + +&function_begin_B("_x86_Camellia_encrypt"); + &xor (@T[0],&DWP(0,$key)); # ^=key[0-3] + &xor (@T[1],&DWP(4,$key)); + &xor (@T[2],&DWP(8,$key)); + &xor (@T[3],&DWP(12,$key)); + &mov ($idx,&DWP(16,$key)); # prefetch key[4] + + &mov ($__s0,@T[0]); # save s[0-3] + &mov ($__s1,@T[1]); + &mov ($__s2,@T[2]); + &mov ($__s3,@T[3]); + +&set_label("loop",16); + for ($i=0;$i<6;$i++) { Camellia_Feistel($i,16,4); } + + &add ($key,16*4); + &cmp ($key,$__end); + &je (&label("done")); + + # @T[0-1] are preloaded, $idx is preloaded with key[0] + &and ($idx,@T[0]); + &mov (@T[3],$__s3); + &rotl ($idx,1); + &mov (@T[2],@T[3]); + &xor (@T[1],$idx); + &or (@T[2],&DWP(12,$key)); + &mov ($__s1,@T[1]); # s1^=LeftRotate(s0&key[0],1); + &xor (@T[2],$__s2); + + &mov ($idx,&DWP(4,$key)); + &mov ($__s2,@T[2]); # s2^=s3|key[3]; + &or ($idx,@T[1]); + &and (@T[2],&DWP(8,$key)); + &xor (@T[0],$idx); + &rotl (@T[2],1); + &mov ($__s0,@T[0]); # s0^=s1|key[1]; + &xor (@T[3],@T[2]); + &mov ($idx,&DWP(16,$key)); # prefetch key[4] + &mov ($__s3,@T[3]); # s3^=LeftRotate(s2&key[2],1); + &jmp (&label("loop")); + +&set_label("done",8); + &mov (@T[2],@T[0]); # SwapHalf + &mov (@T[3],@T[1]); + &mov (@T[0],$__s2); + &mov (@T[1],$__s3); + &xor (@T[0],$idx); # $idx is preloaded with key[0] + &xor (@T[1],&DWP(4,$key)); + &xor (@T[2],&DWP(8,$key)); + &xor (@T[3],&DWP(12,$key)); + &ret (); +&function_end_B("_x86_Camellia_encrypt"); + +# void Camellia_DecryptBlock_Rounds( +# int grandRounds, +# const Byte ciphertext[], +# const KEY_TABLE_TYPE keyTable, +# Byte plaintext[]) +&function_begin("Camellia_DecryptBlock_Rounds"); + &mov ("eax",&wparam(0)); # load grandRounds + &mov ($idx,&wparam(1)); # load ciphertext pointer + &mov ($key,&wparam(2)); # load key schedule pointer + + &mov ("ebx","esp"); + &sub ("esp",7*4); # place for s[0-3],keyEnd,esp and ra + &and ("esp",-64); + + # place stack frame just "above mod 1024" the key schedule + # this ensures that cache associativity of 2 suffices + &lea ("ecx",&DWP(-64-63,$key)); + &sub ("ecx","esp"); + &neg ("ecx"); + &and ("ecx",0x3C0); # modulo 1024, but aligned to cache-line + &sub ("esp","ecx"); + &add ("esp",4); # 4 is reserved for callee's return address + + &shl ("eax",6); + &mov (&DWP(4*4,"esp"),$key); # save keyStart + &lea ($key,&DWP(0,$key,"eax")); + &mov (&DWP(5*4,"esp"),"ebx");# save %esp + + &call (&label("pic_point")); + &set_label("pic_point"); + &blindpop($Tbl); + &lea ($Tbl,&DWP(&label("Camellia_SBOX")."-".&label("pic_point"),$Tbl)); + + &mov (@T[0],&DWP(0,$idx)); # load ciphertext + &mov (@T[1],&DWP(4,$idx)); + &mov (@T[2],&DWP(8,$idx)); + &bswap (@T[0]); + &mov (@T[3],&DWP(12,$idx)); + &bswap (@T[1]); + &bswap (@T[2]); + &bswap (@T[3]); + + &call ("_x86_Camellia_decrypt"); + + &mov ("esp",&DWP(5*4,"esp")); + &bswap (@T[0]); + &mov ($idx,&wparam(3)); # load plaintext pointer + &bswap (@T[1]); + &bswap (@T[2]); + &bswap (@T[3]); + &mov (&DWP(0,$idx),@T[0]); # write plaintext + &mov (&DWP(4,$idx),@T[1]); + &mov (&DWP(8,$idx),@T[2]); + &mov (&DWP(12,$idx),@T[3]); +&function_end("Camellia_DecryptBlock_Rounds"); +# V1.x API +&function_begin_B("Camellia_DecryptBlock"); + &mov ("eax",128); + &sub ("eax",&wparam(0)); # load keyBitLength + &mov ("eax",3); + &adc ("eax",0); # keyBitLength==128?3:4 + &mov (&wparam(0),"eax"); + &jmp (&label("Camellia_DecryptBlock_Rounds")); +&function_end_B("Camellia_DecryptBlock"); + +if ($OPENSSL) { +# void Camellia_decrypt( +# const unsigned char *in, +# unsigned char *out, +# const CAMELLIA_KEY *key) +&function_begin("Camellia_decrypt"); + &mov ($idx,&wparam(0)); # load ciphertext pointer + &mov ($key,&wparam(2)); # load key schedule pointer + + &mov ("ebx","esp"); + &sub ("esp",7*4); # place for s[0-3],keyEnd,esp and ra + &and ("esp",-64); + &mov ("eax",&DWP(272,$key)); # load grandRounds counter + + # place stack frame just "above mod 1024" the key schedule + # this ensures that cache associativity of 2 suffices + &lea ("ecx",&DWP(-64-63,$key)); + &sub ("ecx","esp"); + &neg ("ecx"); + &and ("ecx",0x3C0); # modulo 1024, but aligned to cache-line + &sub ("esp","ecx"); + &add ("esp",4); # 4 is reserved for callee's return address + + &shl ("eax",6); + &mov (&DWP(4*4,"esp"),$key); # save keyStart + &lea ($key,&DWP(0,$key,"eax")); + &mov (&DWP(5*4,"esp"),"ebx");# save %esp + + &call (&label("pic_point")); + &set_label("pic_point"); + &blindpop($Tbl); + &lea ($Tbl,&DWP(&label("Camellia_SBOX")."-".&label("pic_point"),$Tbl)); + + &mov (@T[0],&DWP(0,$idx)); # load ciphertext + &mov (@T[1],&DWP(4,$idx)); + &mov (@T[2],&DWP(8,$idx)); + &bswap (@T[0]); + &mov (@T[3],&DWP(12,$idx)); + &bswap (@T[1]); + &bswap (@T[2]); + &bswap (@T[3]); + + &call ("_x86_Camellia_decrypt"); + + &mov ("esp",&DWP(5*4,"esp")); + &bswap (@T[0]); + &mov ($idx,&wparam(1)); # load plaintext pointer + &bswap (@T[1]); + &bswap (@T[2]); + &bswap (@T[3]); + &mov (&DWP(0,$idx),@T[0]); # write plaintext + &mov (&DWP(4,$idx),@T[1]); + &mov (&DWP(8,$idx),@T[2]); + &mov (&DWP(12,$idx),@T[3]); +&function_end("Camellia_decrypt"); +} + +&function_begin_B("_x86_Camellia_decrypt"); + &xor (@T[0],&DWP(0,$key)); # ^=key[0-3] + &xor (@T[1],&DWP(4,$key)); + &xor (@T[2],&DWP(8,$key)); + &xor (@T[3],&DWP(12,$key)); + &mov ($idx,&DWP(-8,$key)); # prefetch key[-2] + + &mov ($__s0,@T[0]); # save s[0-3] + &mov ($__s1,@T[1]); + &mov ($__s2,@T[2]); + &mov ($__s3,@T[3]); + +&set_label("loop",16); + for ($i=0;$i<6;$i++) { Camellia_Feistel($i,-8,4); } + + &sub ($key,16*4); + &cmp ($key,$__end); + &je (&label("done")); + + # @T[0-1] are preloaded, $idx is preloaded with key[2] + &and ($idx,@T[0]); + &mov (@T[3],$__s3); + &rotl ($idx,1); + &mov (@T[2],@T[3]); + &xor (@T[1],$idx); + &or (@T[2],&DWP(4,$key)); + &mov ($__s1,@T[1]); # s1^=LeftRotate(s0&key[0],1); + &xor (@T[2],$__s2); + + &mov ($idx,&DWP(12,$key)); + &mov ($__s2,@T[2]); # s2^=s3|key[3]; + &or ($idx,@T[1]); + &and (@T[2],&DWP(0,$key)); + &xor (@T[0],$idx); + &rotl (@T[2],1); + &mov ($__s0,@T[0]); # s0^=s1|key[1]; + &xor (@T[3],@T[2]); + &mov ($idx,&DWP(-8,$key)); # prefetch key[4] + &mov ($__s3,@T[3]); # s3^=LeftRotate(s2&key[2],1); + &jmp (&label("loop")); + +&set_label("done",8); + &mov (@T[2],@T[0]); # SwapHalf + &mov (@T[3],@T[1]); + &mov (@T[0],$__s2); + &mov (@T[1],$__s3); + &xor (@T[2],$idx); # $idx is preloaded with key[2] + &xor (@T[3],&DWP(12,$key)); + &xor (@T[0],&DWP(0,$key)); + &xor (@T[1],&DWP(4,$key)); + &ret (); +&function_end_B("_x86_Camellia_decrypt"); + +# shld is very slow on Intel P4 family. Even on AMD it limits +# instruction decode rate [because it's VectorPath] and consequently +# performance. PIII, PM and Core[2] seem to be the only ones which +# execute this code ~7% faster... +sub __rotl128 { + my ($i0,$i1,$i2,$i3,$rot,$rnd,@T)=@_; + + $rnd *= 2; + if ($rot) { + &mov ($idx,$i0); + &shld ($i0,$i1,$rot); + &shld ($i1,$i2,$rot); + &shld ($i2,$i3,$rot); + &shld ($i3,$idx,$rot); + } + &mov (&DWP(-128+4*$rnd++,$key),shift(@T)) if ($i0 eq @T[0]); + &mov (&DWP(-128+4*$rnd++,$key),shift(@T)) if ($i1 eq @T[0]); + &mov (&DWP(-128+4*$rnd++,$key),shift(@T)) if ($i2 eq @T[0]); + &mov (&DWP(-128+4*$rnd++,$key),shift(@T)) if ($i3 eq @T[0]); +} + +# ... Implementing 128-bit rotate without shld gives >3x performance +# improvement on P4, only ~7% degradation on other Intel CPUs and +# not worse performance on AMD. This is therefore preferred. +sub _rotl128 { + my ($i0,$i1,$i2,$i3,$rot,$rnd,@T)=@_; + + $rnd *= 2; + if ($rot) { + &mov ($Tbl,$i0); + &shl ($i0,$rot); + &mov ($idx,$i1); + &shr ($idx,32-$rot); + &shl ($i1,$rot); + &or ($i0,$idx); + &mov ($idx,$i2); + &shl ($i2,$rot); + &mov (&DWP(-128+4*$rnd++,$key),shift(@T)) if ($i0 eq @T[0]); + &shr ($idx,32-$rot); + &or ($i1,$idx); + &shr ($Tbl,32-$rot); + &mov ($idx,$i3); + &shr ($idx,32-$rot); + &mov (&DWP(-128+4*$rnd++,$key),shift(@T)) if ($i1 eq @T[0]); + &shl ($i3,$rot); + &or ($i2,$idx); + &or ($i3,$Tbl); + &mov (&DWP(-128+4*$rnd++,$key),shift(@T)) if ($i2 eq @T[0]); + &mov (&DWP(-128+4*$rnd++,$key),shift(@T)) if ($i3 eq @T[0]); + } else { + &mov (&DWP(-128+4*$rnd++,$key),shift(@T)) if ($i0 eq @T[0]); + &mov (&DWP(-128+4*$rnd++,$key),shift(@T)) if ($i1 eq @T[0]); + &mov (&DWP(-128+4*$rnd++,$key),shift(@T)) if ($i2 eq @T[0]); + &mov (&DWP(-128+4*$rnd++,$key),shift(@T)) if ($i3 eq @T[0]); + } +} + +sub _saveround { +my ($rnd,$key,@T)=@_; +my $bias=int(@T[0])?shift(@T):0; + + &mov (&DWP($bias+$rnd*8+0,$key),@T[0]); + &mov (&DWP($bias+$rnd*8+4,$key),@T[1]) if ($#T>=1); + &mov (&DWP($bias+$rnd*8+8,$key),@T[2]) if ($#T>=2); + &mov (&DWP($bias+$rnd*8+12,$key),@T[3]) if ($#T>=3); +} + +sub _loadround { +my ($rnd,$key,@T)=@_; +my $bias=int(@T[0])?shift(@T):0; + + &mov (@T[0],&DWP($bias+$rnd*8+0,$key)); + &mov (@T[1],&DWP($bias+$rnd*8+4,$key)) if ($#T>=1); + &mov (@T[2],&DWP($bias+$rnd*8+8,$key)) if ($#T>=2); + &mov (@T[3],&DWP($bias+$rnd*8+12,$key)) if ($#T>=3); +} + +# void Camellia_Ekeygen( +# const int keyBitLength, +# const Byte *rawKey, +# KEY_TABLE_TYPE keyTable) +&function_begin("Camellia_Ekeygen"); +{ my $step=0; + + &stack_push(4); # place for s[0-3] + + &mov ($Tbl,&wparam(0)); # load arguments + &mov ($idx,&wparam(1)); + &mov ($key,&wparam(2)); + + &mov (@T[0],&DWP(0,$idx)); # load 0-127 bits + &mov (@T[1],&DWP(4,$idx)); + &mov (@T[2],&DWP(8,$idx)); + &mov (@T[3],&DWP(12,$idx)); + + &bswap (@T[0]); + &bswap (@T[1]); + &bswap (@T[2]); + &bswap (@T[3]); + + &_saveround (0,$key,@T); # KL<<<0 + + &cmp ($Tbl,128); + &je (&label("1st128")); + + &mov (@T[0],&DWP(16,$idx)); # load 128-191 bits + &mov (@T[1],&DWP(20,$idx)); + &cmp ($Tbl,192); + &je (&label("1st192")); + &mov (@T[2],&DWP(24,$idx)); # load 192-255 bits + &mov (@T[3],&DWP(28,$idx)); + &jmp (&label("1st256")); +&set_label("1st192",4); + &mov (@T[2],@T[0]); + &mov (@T[3],@T[1]); + ¬ (@T[2]); + ¬ (@T[3]); +&set_label("1st256",4); + &bswap (@T[0]); + &bswap (@T[1]); + &bswap (@T[2]); + &bswap (@T[3]); + + &_saveround (4,$key,@T); # temporary storage for KR! + + &xor (@T[0],&DWP(0*8+0,$key)); # KR^KL + &xor (@T[1],&DWP(0*8+4,$key)); + &xor (@T[2],&DWP(1*8+0,$key)); + &xor (@T[3],&DWP(1*8+4,$key)); + +&set_label("1st128",4); + &call (&label("pic_point")); + &set_label("pic_point"); + &blindpop($Tbl); + &lea ($Tbl,&DWP(&label("Camellia_SBOX")."-".&label("pic_point"),$Tbl)); + &lea ($key,&DWP(&label("Camellia_SIGMA")."-".&label("Camellia_SBOX"),$Tbl)); + + &mov ($idx,&DWP($step*8,$key)); # prefetch SIGMA[0] + &mov (&swtmp(0),@T[0]); # save s[0-3] + &mov (&swtmp(1),@T[1]); + &mov (&swtmp(2),@T[2]); + &mov (&swtmp(3),@T[3]); + &Camellia_Feistel($step++); + &Camellia_Feistel($step++); + &mov (@T[2],&swtmp(2)); + &mov (@T[3],&swtmp(3)); + + &mov ($idx,&wparam(2)); + &xor (@T[0],&DWP(0*8+0,$idx)); # ^KL + &xor (@T[1],&DWP(0*8+4,$idx)); + &xor (@T[2],&DWP(1*8+0,$idx)); + &xor (@T[3],&DWP(1*8+4,$idx)); + + &mov ($idx,&DWP($step*8,$key)); # prefetch SIGMA[4] + &mov (&swtmp(0),@T[0]); # save s[0-3] + &mov (&swtmp(1),@T[1]); + &mov (&swtmp(2),@T[2]); + &mov (&swtmp(3),@T[3]); + &Camellia_Feistel($step++); + &Camellia_Feistel($step++); + &mov (@T[2],&swtmp(2)); + &mov (@T[3],&swtmp(3)); + + &mov ($idx,&wparam(0)); + &cmp ($idx,128); + &jne (&label("2nd256")); + + &mov ($key,&wparam(2)); + &lea ($key,&DWP(128,$key)); # size optimization + + ####### process KA + &_saveround (2,$key,-128,@T); # KA<<<0 + &_rotl128 (@T,15,6,@T); # KA<<<15 + &_rotl128 (@T,15,8,@T); # KA<<<(15+15=30) + &_rotl128 (@T,15,12,@T[0],@T[1]); # KA<<<(30+15=45) + &_rotl128 (@T,15,14,@T); # KA<<<(45+15=60) + push (@T,shift(@T)); # rotl128(@T,32); + &_rotl128 (@T,2,20,@T); # KA<<<(60+32+2=94) + &_rotl128 (@T,17,24,@T); # KA<<<(94+17=111) + + ####### process KL + &_loadround (0,$key,-128,@T); # load KL + &_rotl128 (@T,15,4,@T); # KL<<<15 + &_rotl128 (@T,30,10,@T); # KL<<<(15+30=45) + &_rotl128 (@T,15,13,@T[2],@T[3]); # KL<<<(45+15=60) + &_rotl128 (@T,17,16,@T); # KL<<<(60+17=77) + &_rotl128 (@T,17,18,@T); # KL<<<(77+17=94) + &_rotl128 (@T,17,22,@T); # KL<<<(94+17=111) + + while (@T[0] ne "eax") # restore order + { unshift (@T,pop(@T)); } + + &mov ("eax",3); # 3 grandRounds + &jmp (&label("done")); + +&set_label("2nd256",16); + &mov ($idx,&wparam(2)); + &_saveround (6,$idx,@T); # temporary storage for KA! + + &xor (@T[0],&DWP(4*8+0,$idx)); # KA^KR + &xor (@T[1],&DWP(4*8+4,$idx)); + &xor (@T[2],&DWP(5*8+0,$idx)); + &xor (@T[3],&DWP(5*8+4,$idx)); + + &mov ($idx,&DWP($step*8,$key)); # prefetch SIGMA[8] + &mov (&swtmp(0),@T[0]); # save s[0-3] + &mov (&swtmp(1),@T[1]); + &mov (&swtmp(2),@T[2]); + &mov (&swtmp(3),@T[3]); + &Camellia_Feistel($step++); + &Camellia_Feistel($step++); + &mov (@T[2],&swtmp(2)); + &mov (@T[3],&swtmp(3)); + + &mov ($key,&wparam(2)); + &lea ($key,&DWP(128,$key)); # size optimization + + ####### process KB + &_saveround (2,$key,-128,@T); # KB<<<0 + &_rotl128 (@T,30,10,@T); # KB<<<30 + &_rotl128 (@T,30,20,@T); # KB<<<(30+30=60) + push (@T,shift(@T)); # rotl128(@T,32); + &_rotl128 (@T,19,32,@T); # KB<<<(60+32+19=111) + + ####### process KR + &_loadround (4,$key,-128,@T); # load KR + &_rotl128 (@T,15,4,@T); # KR<<<15 + &_rotl128 (@T,15,8,@T); # KR<<<(15+15=30) + &_rotl128 (@T,30,18,@T); # KR<<<(30+30=60) + push (@T,shift(@T)); # rotl128(@T,32); + &_rotl128 (@T,2,26,@T); # KR<<<(60+32+2=94) + + ####### process KA + &_loadround (6,$key,-128,@T); # load KA + &_rotl128 (@T,15,6,@T); # KA<<<15 + &_rotl128 (@T,30,14,@T); # KA<<<(15+30=45) + push (@T,shift(@T)); # rotl128(@T,32); + &_rotl128 (@T,0,24,@T); # KA<<<(45+32+0=77) + &_rotl128 (@T,17,28,@T); # KA<<<(77+17=94) + + ####### process KL + &_loadround (0,$key,-128,@T); # load KL + push (@T,shift(@T)); # rotl128(@T,32); + &_rotl128 (@T,13,12,@T); # KL<<<(32+13=45) + &_rotl128 (@T,15,16,@T); # KL<<<(45+15=60) + &_rotl128 (@T,17,22,@T); # KL<<<(60+17=77) + push (@T,shift(@T)); # rotl128(@T,32); + &_rotl128 (@T,2,30,@T); # KL<<<(77+32+2=111) + + while (@T[0] ne "eax") # restore order + { unshift (@T,pop(@T)); } + + &mov ("eax",4); # 4 grandRounds +&set_label("done"); + &lea ("edx",&DWP(272-128,$key)); # end of key schedule + &stack_pop(4); +} +&function_end("Camellia_Ekeygen"); + +if ($OPENSSL) { +# int Camellia_set_key ( +# const unsigned char *userKey, +# int bits, +# CAMELLIA_KEY *key) +&function_begin_B("Camellia_set_key"); + &push ("ebx"); + &mov ("ecx",&wparam(0)); # pull arguments + &mov ("ebx",&wparam(1)); + &mov ("edx",&wparam(2)); + + &mov ("eax",-1); + &test ("ecx","ecx"); + &jz (&label("done")); # userKey==NULL? + &test ("edx","edx"); + &jz (&label("done")); # key==NULL? + + &mov ("eax",-2); + &cmp ("ebx",256); + &je (&label("arg_ok")); # bits==256? + &cmp ("ebx",192); + &je (&label("arg_ok")); # bits==192? + &cmp ("ebx",128); + &jne (&label("done")); # bits!=128? +&set_label("arg_ok",4); + + &push ("edx"); # push arguments + &push ("ecx"); + &push ("ebx"); + &call ("Camellia_Ekeygen"); + &stack_pop(3); + + # eax holds grandRounds and edx points at where to put it + &mov (&DWP(0,"edx"),"eax"); + &xor ("eax","eax"); +&set_label("done",4); + &pop ("ebx"); + &ret (); +&function_end_B("Camellia_set_key"); +} + +@SBOX=( +112,130, 44,236,179, 39,192,229,228,133, 87, 53,234, 12,174, 65, + 35,239,107,147, 69, 25,165, 33,237, 14, 79, 78, 29,101,146,189, +134,184,175,143,124,235, 31,206, 62, 48,220, 95, 94,197, 11, 26, +166,225, 57,202,213, 71, 93, 61,217, 1, 90,214, 81, 86,108, 77, +139, 13,154,102,251,204,176, 45,116, 18, 43, 32,240,177,132,153, +223, 76,203,194, 52,126,118, 5,109,183,169, 49,209, 23, 4,215, + 20, 88, 58, 97,222, 27, 17, 28, 50, 15,156, 22, 83, 24,242, 34, +254, 68,207,178,195,181,122,145, 36, 8,232,168, 96,252,105, 80, +170,208,160,125,161,137, 98,151, 84, 91, 30,149,224,255,100,210, + 16,196, 0, 72,163,247,117,219,138, 3,230,218, 9, 63,221,148, +135, 92,131, 2,205, 74,144, 51,115,103,246,243,157,127,191,226, + 82,155,216, 38,200, 55,198, 59,129,150,111, 75, 19,190, 99, 46, +233,121,167,140,159,110,188,142, 41,245,249,182, 47,253,180, 89, +120,152, 6,106,231, 70,113,186,212, 37,171, 66,136,162,141,250, +114, 7,185, 85,248,238,172, 10, 54, 73, 42,104, 60, 56,241,164, + 64, 40,211,123,187,201, 67,193, 21,227,173,244,119,199,128,158); + +sub S1110 { my $i=shift; $i=@SBOX[$i]; return $i<<24|$i<<16|$i<<8; } +sub S4404 { my $i=shift; $i=($i<<1|$i>>7)&0xff; $i=@SBOX[$i]; return $i<<24|$i<<16|$i; } +sub S0222 { my $i=shift; $i=@SBOX[$i]; $i=($i<<1|$i>>7)&0xff; return $i<<16|$i<<8|$i; } +sub S3033 { my $i=shift; $i=@SBOX[$i]; $i=($i>>1|$i<<7)&0xff; return $i<<24|$i<<8|$i; } + +&set_label("Camellia_SIGMA",64); +&data_word( + 0xa09e667f, 0x3bcc908b, 0xb67ae858, 0x4caa73b2, + 0xc6ef372f, 0xe94f82be, 0x54ff53a5, 0xf1d36f1c, + 0x10e527fa, 0xde682d1d, 0xb05688c2, 0xb3e6c1fd, + 0, 0, 0, 0); +&set_label("Camellia_SBOX",64); +# tables are interleaved, remember? +for ($i=0;$i<256;$i++) { &data_word(&S1110($i),&S4404($i)); } +for ($i=0;$i<256;$i++) { &data_word(&S0222($i),&S3033($i)); } + +# void Camellia_cbc_encrypt (const void char *inp, unsigned char *out, +# size_t length, const CAMELLIA_KEY *key, +# unsigned char *ivp,const int enc); +{ +# stack frame layout +# -4(%esp) # return address 0(%esp) +# 0(%esp) # s0 4(%esp) +# 4(%esp) # s1 8(%esp) +# 8(%esp) # s2 12(%esp) +# 12(%esp) # s3 16(%esp) +# 16(%esp) # end of key schedule 20(%esp) +# 20(%esp) # %esp backup +my $_inp=&DWP(24,"esp"); #copy of wparam(0) +my $_out=&DWP(28,"esp"); #copy of wparam(1) +my $_len=&DWP(32,"esp"); #copy of wparam(2) +my $_key=&DWP(36,"esp"); #copy of wparam(3) +my $_ivp=&DWP(40,"esp"); #copy of wparam(4) +my $ivec=&DWP(44,"esp"); #ivec[16] +my $_tmp=&DWP(44,"esp"); #volatile variable [yes, aliases with ivec] +my ($s0,$s1,$s2,$s3) = @T; + +&function_begin("Camellia_cbc_encrypt"); + &mov ($s2 eq "ecx"? $s2 : "",&wparam(2)); # load len + &cmp ($s2,0); + &je (&label("enc_out")); + + &pushf (); + &cld (); + + &mov ($s0,&wparam(0)); # load inp + &mov ($s1,&wparam(1)); # load out + #&mov ($s2,&wparam(2)); # load len + &mov ($s3,&wparam(3)); # load key + &mov ($Tbl,&wparam(4)); # load ivp + + # allocate aligned stack frame... + &lea ($idx,&DWP(-64,"esp")); + &and ($idx,-64); + + # place stack frame just "above mod 1024" the key schedule + # this ensures that cache associativity of 2 suffices + &lea ($key,&DWP(-64-63,$s3)); + &sub ($key,$idx); + &neg ($key); + &and ($key,0x3C0); # modulo 1024, but aligned to cache-line + &sub ($idx,$key); + + &mov ($key,&wparam(5)); # load enc + + &exch ("esp",$idx); + &add ("esp",4); # reserve for return address! + &mov ($_esp,$idx); # save %esp + + &mov ($_inp,$s0); # save copy of inp + &mov ($_out,$s1); # save copy of out + &mov ($_len,$s2); # save copy of len + &mov ($_key,$s3); # save copy of key + &mov ($_ivp,$Tbl); # save copy of ivp + + &call (&label("pic_point")); # make it PIC! + &set_label("pic_point"); + &blindpop($Tbl); + &lea ($Tbl,&DWP(&label("Camellia_SBOX")."-".&label("pic_point"),$Tbl)); + + &mov ($idx,32); + &set_label("prefetch_sbox",4); + &mov ($s0,&DWP(0,$Tbl)); + &mov ($s1,&DWP(32,$Tbl)); + &mov ($s2,&DWP(64,$Tbl)); + &mov ($s3,&DWP(96,$Tbl)); + &lea ($Tbl,&DWP(128,$Tbl)); + &dec ($idx); + &jnz (&label("prefetch_sbox")); + &mov ($s0,$_key); + &sub ($Tbl,4096); + &mov ($idx,$_inp); + &mov ($s3,&DWP(272,$s0)); # load grandRounds + + &cmp ($key,0); + &je (&label("DECRYPT")); + + &mov ($s2,$_len); + &mov ($key,$_ivp); + &shl ($s3,6); + &lea ($s3,&DWP(0,$s0,$s3)); + &mov ($_end,$s3); + + &test ($s2,0xFFFFFFF0); + &jz (&label("enc_tail")); # short input... + + &mov ($s0,&DWP(0,$key)); # load iv + &mov ($s1,&DWP(4,$key)); + + &set_label("enc_loop",4); + &mov ($s2,&DWP(8,$key)); + &mov ($s3,&DWP(12,$key)); + + &xor ($s0,&DWP(0,$idx)); # xor input data + &xor ($s1,&DWP(4,$idx)); + &xor ($s2,&DWP(8,$idx)); + &bswap ($s0); + &xor ($s3,&DWP(12,$idx)); + &bswap ($s1); + &mov ($key,$_key); # load key + &bswap ($s2); + &bswap ($s3); + + &call ("_x86_Camellia_encrypt"); + + &mov ($idx,$_inp); # load inp + &mov ($key,$_out); # load out + + &bswap ($s0); + &bswap ($s1); + &bswap ($s2); + &mov (&DWP(0,$key),$s0); # save output data + &bswap ($s3); + &mov (&DWP(4,$key),$s1); + &mov (&DWP(8,$key),$s2); + &mov (&DWP(12,$key),$s3); + + &mov ($s2,$_len); # load len + + &lea ($idx,&DWP(16,$idx)); + &mov ($_inp,$idx); # save inp + + &lea ($s3,&DWP(16,$key)); + &mov ($_out,$s3); # save out + + &sub ($s2,16); + &test ($s2,0xFFFFFFF0); + &mov ($_len,$s2); # save len + &jnz (&label("enc_loop")); + &test ($s2,15); + &jnz (&label("enc_tail")); + &mov ($idx,$_ivp); # load ivp + &mov ($s2,&DWP(8,$key)); # restore last dwords + &mov ($s3,&DWP(12,$key)); + &mov (&DWP(0,$idx),$s0); # save ivec + &mov (&DWP(4,$idx),$s1); + &mov (&DWP(8,$idx),$s2); + &mov (&DWP(12,$idx),$s3); + + &mov ("esp",$_esp); + &popf (); + &set_label("enc_out"); + &function_end_A(); + &pushf (); # kludge, never executed + + &set_label("enc_tail",4); + &mov ($s0,$key eq "edi" ? $key : ""); + &mov ($key,$_out); # load out + &push ($s0); # push ivp + &mov ($s1,16); + &sub ($s1,$s2); + &cmp ($key,$idx); # compare with inp + &je (&label("enc_in_place")); + &align (4); + &data_word(0xA4F3F689); # rep movsb # copy input + &jmp (&label("enc_skip_in_place")); + &set_label("enc_in_place"); + &lea ($key,&DWP(0,$key,$s2)); + &set_label("enc_skip_in_place"); + &mov ($s2,$s1); + &xor ($s0,$s0); + &align (4); + &data_word(0xAAF3F689); # rep stosb # zero tail + &pop ($key); # pop ivp + + &mov ($idx,$_out); # output as input + &mov ($s0,&DWP(0,$key)); + &mov ($s1,&DWP(4,$key)); + &mov ($_len,16); # len=16 + &jmp (&label("enc_loop")); # one more spin... + +#----------------------------- DECRYPT -----------------------------# +&set_label("DECRYPT",16); + &shl ($s3,6); + &lea ($s3,&DWP(0,$s0,$s3)); + &mov ($_end,$s0); + &mov ($_key,$s3); + + &cmp ($idx,$_out); + &je (&label("dec_in_place")); # in-place processing... + + &mov ($key,$_ivp); # load ivp + &mov ($_tmp,$key); + + &set_label("dec_loop",4); + &mov ($s0,&DWP(0,$idx)); # read input + &mov ($s1,&DWP(4,$idx)); + &mov ($s2,&DWP(8,$idx)); + &bswap ($s0); + &mov ($s3,&DWP(12,$idx)); + &bswap ($s1); + &mov ($key,$_key); # load key + &bswap ($s2); + &bswap ($s3); + + &call ("_x86_Camellia_decrypt"); + + &mov ($key,$_tmp); # load ivp + &mov ($idx,$_len); # load len + + &bswap ($s0); + &bswap ($s1); + &bswap ($s2); + &xor ($s0,&DWP(0,$key)); # xor iv + &bswap ($s3); + &xor ($s1,&DWP(4,$key)); + &xor ($s2,&DWP(8,$key)); + &xor ($s3,&DWP(12,$key)); + + &sub ($idx,16); + &jc (&label("dec_partial")); + &mov ($_len,$idx); # save len + &mov ($idx,$_inp); # load inp + &mov ($key,$_out); # load out + + &mov (&DWP(0,$key),$s0); # write output + &mov (&DWP(4,$key),$s1); + &mov (&DWP(8,$key),$s2); + &mov (&DWP(12,$key),$s3); + + &mov ($_tmp,$idx); # save ivp + &lea ($idx,&DWP(16,$idx)); + &mov ($_inp,$idx); # save inp + + &lea ($key,&DWP(16,$key)); + &mov ($_out,$key); # save out + + &jnz (&label("dec_loop")); + &mov ($key,$_tmp); # load temp ivp + &set_label("dec_end"); + &mov ($idx,$_ivp); # load user ivp + &mov ($s0,&DWP(0,$key)); # load iv + &mov ($s1,&DWP(4,$key)); + &mov ($s2,&DWP(8,$key)); + &mov ($s3,&DWP(12,$key)); + &mov (&DWP(0,$idx),$s0); # copy back to user + &mov (&DWP(4,$idx),$s1); + &mov (&DWP(8,$idx),$s2); + &mov (&DWP(12,$idx),$s3); + &jmp (&label("dec_out")); + + &set_label("dec_partial",4); + &lea ($key,$ivec); + &mov (&DWP(0,$key),$s0); # dump output to stack + &mov (&DWP(4,$key),$s1); + &mov (&DWP(8,$key),$s2); + &mov (&DWP(12,$key),$s3); + &lea ($s2 eq "ecx" ? $s2 : "",&DWP(16,$idx)); + &mov ($idx eq "esi" ? $idx : "",$key); + &mov ($key eq "edi" ? $key : "",$_out); # load out + &data_word(0xA4F3F689); # rep movsb # copy output + &mov ($key,$_inp); # use inp as temp ivp + &jmp (&label("dec_end")); + + &set_label("dec_in_place",4); + &set_label("dec_in_place_loop"); + &lea ($key,$ivec); + &mov ($s0,&DWP(0,$idx)); # read input + &mov ($s1,&DWP(4,$idx)); + &mov ($s2,&DWP(8,$idx)); + &mov ($s3,&DWP(12,$idx)); + + &mov (&DWP(0,$key),$s0); # copy to temp + &mov (&DWP(4,$key),$s1); + &mov (&DWP(8,$key),$s2); + &bswap ($s0); + &mov (&DWP(12,$key),$s3); + &bswap ($s1); + &mov ($key,$_key); # load key + &bswap ($s2); + &bswap ($s3); + + &call ("_x86_Camellia_decrypt"); + + &mov ($key,$_ivp); # load ivp + &mov ($idx,$_out); # load out + + &bswap ($s0); + &bswap ($s1); + &bswap ($s2); + &xor ($s0,&DWP(0,$key)); # xor iv + &bswap ($s3); + &xor ($s1,&DWP(4,$key)); + &xor ($s2,&DWP(8,$key)); + &xor ($s3,&DWP(12,$key)); + + &mov (&DWP(0,$idx),$s0); # write output + &mov (&DWP(4,$idx),$s1); + &mov (&DWP(8,$idx),$s2); + &mov (&DWP(12,$idx),$s3); + + &lea ($idx,&DWP(16,$idx)); + &mov ($_out,$idx); # save out + + &lea ($idx,$ivec); + &mov ($s0,&DWP(0,$idx)); # read temp + &mov ($s1,&DWP(4,$idx)); + &mov ($s2,&DWP(8,$idx)); + &mov ($s3,&DWP(12,$idx)); + + &mov (&DWP(0,$key),$s0); # copy iv + &mov (&DWP(4,$key),$s1); + &mov (&DWP(8,$key),$s2); + &mov (&DWP(12,$key),$s3); + + &mov ($idx,$_inp); # load inp + + &lea ($idx,&DWP(16,$idx)); + &mov ($_inp,$idx); # save inp + + &mov ($s2,$_len); # load len + &sub ($s2,16); + &jc (&label("dec_in_place_partial")); + &mov ($_len,$s2); # save len + &jnz (&label("dec_in_place_loop")); + &jmp (&label("dec_out")); + + &set_label("dec_in_place_partial",4); + # one can argue if this is actually required... + &mov ($key eq "edi" ? $key : "",$_out); + &lea ($idx eq "esi" ? $idx : "",$ivec); + &lea ($key,&DWP(0,$key,$s2)); + &lea ($idx,&DWP(16,$idx,$s2)); + &neg ($s2 eq "ecx" ? $s2 : ""); + &data_word(0xA4F3F689); # rep movsb # restore tail + + &set_label("dec_out",4); + &mov ("esp",$_esp); + &popf (); +&function_end("Camellia_cbc_encrypt"); +} + +&asciz("Camellia for x86 by <appro\@openssl.org>"); + +&asm_finish(); + +close STDOUT; diff --git a/openssl-1.1.0h/crypto/camellia/asm/cmll-x86_64.pl b/openssl-1.1.0h/crypto/camellia/asm/cmll-x86_64.pl new file mode 100644 index 0000000..da5ad7b --- /dev/null +++ b/openssl-1.1.0h/crypto/camellia/asm/cmll-x86_64.pl @@ -0,0 +1,1088 @@ +#! /usr/bin/env perl +# Copyright 2008-2016 The OpenSSL Project Authors. All Rights Reserved. +# +# Licensed under the OpenSSL license (the "License"). You may not use +# this file except in compliance with the License. You can obtain a copy +# in the file LICENSE in the source distribution or at +# https://www.openssl.org/source/license.html + + +# ==================================================================== +# Copyright (c) 2008 Andy Polyakov <appro@openssl.org> +# +# This module may be used under the terms of either the GNU General +# Public License version 2 or later, the GNU Lesser General Public +# License version 2.1 or later, the Mozilla Public License version +# 1.1 or the BSD License. The exact terms of either license are +# distributed along with this module. For further details see +# http://www.openssl.org/~appro/camellia/. +# ==================================================================== + +# Performance in cycles per processed byte (less is better) in +# 'openssl speed ...' benchmark: +# +# AMD64 Core2 EM64T +# -evp camellia-128-ecb 16.7 21.0 22.7 +# + over gcc 3.4.6 +25% +5% 0% +# +# camellia-128-cbc 15.7 20.4 21.1 +# +# 128-bit key setup 128 216 205 cycles/key +# + over gcc 3.4.6 +54% +39% +15% +# +# Numbers in "+" rows represent performance improvement over compiler +# generated code. Key setup timings are impressive on AMD and Core2 +# thanks to 64-bit operations being covertly deployed. Improvement on +# EM64T, pre-Core2 Intel x86_64 CPU, is not as impressive, because it +# apparently emulates some of 64-bit operations in [32-bit] microcode. + +$flavour = shift; +$output = shift; +if ($flavour =~ /\./) { $output = $flavour; undef $flavour; } + +$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/); + +$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1; +( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or +( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or +die "can't locate x86_64-xlate.pl"; + +open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\""; +*STDOUT=*OUT; + +sub hi() { my $r=shift; $r =~ s/%[er]([a-d])x/%\1h/; $r; } +sub lo() { my $r=shift; $r =~ s/%[er]([a-d])x/%\1l/; + $r =~ s/%[er]([sd]i)/%\1l/; + $r =~ s/%(r[0-9]+)[d]?/%\1b/; $r; } + +$t0="%eax";$t1="%ebx";$t2="%ecx";$t3="%edx"; +@S=("%r8d","%r9d","%r10d","%r11d"); +$i0="%esi"; +$i1="%edi"; +$Tbl="%rbp"; # size optimization +$inp="%r12"; +$out="%r13"; +$key="%r14"; +$keyend="%r15"; +$arg0d=$win64?"%ecx":"%edi"; + +# const unsigned int Camellia_SBOX[4][256]; +# Well, sort of... Camellia_SBOX[0][] is interleaved with [1][], +# and [2][] - with [3][]. This is done to minimize code size. +$SBOX1_1110=0; # Camellia_SBOX[0] +$SBOX4_4404=4; # Camellia_SBOX[1] +$SBOX2_0222=2048; # Camellia_SBOX[2] +$SBOX3_3033=2052; # Camellia_SBOX[3] + +sub Camellia_Feistel { +my $i=@_[0]; +my $seed=defined(@_[1])?@_[1]:0; +my $scale=$seed<0?-8:8; +my $j=($i&1)*2; +my ($s0,$s1,$s2,$s3)=(@S[($j)%4],@S[($j+1)%4],@S[($j+2)%4],@S[($j+3)%4]); + +$code.=<<___; + xor $s0,$t0 # t0^=key[0] + xor $s1,$t1 # t1^=key[1] + movz `&hi("$t0")`,$i0 # (t0>>8)&0xff + movz `&lo("$t1")`,$i1 # (t1>>0)&0xff + mov $SBOX3_3033($Tbl,$i0,8),$t3 # t3=SBOX3_3033[0] + mov $SBOX1_1110($Tbl,$i1,8),$t2 # t2=SBOX1_1110[1] + movz `&lo("$t0")`,$i0 # (t0>>0)&0xff + shr \$16,$t0 + movz `&hi("$t1")`,$i1 # (t1>>8)&0xff + xor $SBOX4_4404($Tbl,$i0,8),$t3 # t3^=SBOX4_4404[0] + shr \$16,$t1 + xor $SBOX4_4404($Tbl,$i1,8),$t2 # t2^=SBOX4_4404[1] + movz `&hi("$t0")`,$i0 # (t0>>24)&0xff + movz `&lo("$t1")`,$i1 # (t1>>16)&0xff + xor $SBOX1_1110($Tbl,$i0,8),$t3 # t3^=SBOX1_1110[0] + xor $SBOX3_3033($Tbl,$i1,8),$t2 # t2^=SBOX3_3033[1] + movz `&lo("$t0")`,$i0 # (t0>>16)&0xff + movz `&hi("$t1")`,$i1 # (t1>>24)&0xff + xor $SBOX2_0222($Tbl,$i0,8),$t3 # t3^=SBOX2_0222[0] + xor $SBOX2_0222($Tbl,$i1,8),$t2 # t2^=SBOX2_0222[1] + mov `$seed+($i+1)*$scale`($key),$t1 # prefetch key[i+1] + mov `$seed+($i+1)*$scale+4`($key),$t0 + xor $t3,$t2 # t2^=t3 + ror \$8,$t3 # t3=RightRotate(t3,8) + xor $t2,$s2 + xor $t2,$s3 + xor $t3,$s3 +___ +} + +# void Camellia_EncryptBlock_Rounds( +# int grandRounds, +# const Byte plaintext[], +# const KEY_TABLE_TYPE keyTable, +# Byte ciphertext[]) +$code=<<___; +.text + +# V1.x API +.globl Camellia_EncryptBlock +.type Camellia_EncryptBlock,\@abi-omnipotent +.align 16 +Camellia_EncryptBlock: + movl \$128,%eax + subl $arg0d,%eax + movl \$3,$arg0d + adcl \$0,$arg0d # keyBitLength==128?3:4 + jmp .Lenc_rounds +.size Camellia_EncryptBlock,.-Camellia_EncryptBlock +# V2 +.globl Camellia_EncryptBlock_Rounds +.type Camellia_EncryptBlock_Rounds,\@function,4 +.align 16 +.Lenc_rounds: +Camellia_EncryptBlock_Rounds: + push %rbx + push %rbp + push %r13 + push %r14 + push %r15 +.Lenc_prologue: + + #mov %rsi,$inp # put away arguments + mov %rcx,$out + mov %rdx,$key + + shl \$6,%edi # process grandRounds + lea .LCamellia_SBOX(%rip),$Tbl + lea ($key,%rdi),$keyend + + mov 0(%rsi),@S[0] # load plaintext + mov 4(%rsi),@S[1] + mov 8(%rsi),@S[2] + bswap @S[0] + mov 12(%rsi),@S[3] + bswap @S[1] + bswap @S[2] + bswap @S[3] + + call _x86_64_Camellia_encrypt + + bswap @S[0] + bswap @S[1] + bswap @S[2] + mov @S[0],0($out) + bswap @S[3] + mov @S[1],4($out) + mov @S[2],8($out) + mov @S[3],12($out) + + mov 0(%rsp),%r15 + mov 8(%rsp),%r14 + mov 16(%rsp),%r13 + mov 24(%rsp),%rbp + mov 32(%rsp),%rbx + lea 40(%rsp),%rsp +.Lenc_epilogue: + ret +.size Camellia_EncryptBlock_Rounds,.-Camellia_EncryptBlock_Rounds + +.type _x86_64_Camellia_encrypt,\@abi-omnipotent +.align 16 +_x86_64_Camellia_encrypt: + xor 0($key),@S[1] + xor 4($key),@S[0] # ^=key[0-3] + xor 8($key),@S[3] + xor 12($key),@S[2] +.align 16 +.Leloop: + mov 16($key),$t1 # prefetch key[4-5] + mov 20($key),$t0 + +___ + for ($i=0;$i<6;$i++) { Camellia_Feistel($i,16); } +$code.=<<___; + lea 16*4($key),$key + cmp $keyend,$key + mov 8($key),$t3 # prefetch key[2-3] + mov 12($key),$t2 + je .Ledone + + and @S[0],$t0 + or @S[3],$t3 + rol \$1,$t0 + xor $t3,@S[2] # s2^=s3|key[3]; + xor $t0,@S[1] # s1^=LeftRotate(s0&key[0],1); + and @S[2],$t2 + or @S[1],$t1 + rol \$1,$t2 + xor $t1,@S[0] # s0^=s1|key[1]; + xor $t2,@S[3] # s3^=LeftRotate(s2&key[2],1); + jmp .Leloop + +.align 16 +.Ledone: + xor @S[2],$t0 # SwapHalf + xor @S[3],$t1 + xor @S[0],$t2 + xor @S[1],$t3 + + mov $t0,@S[0] + mov $t1,@S[1] + mov $t2,@S[2] + mov $t3,@S[3] + + .byte 0xf3,0xc3 # rep ret +.size _x86_64_Camellia_encrypt,.-_x86_64_Camellia_encrypt + +# V1.x API +.globl Camellia_DecryptBlock +.type Camellia_DecryptBlock,\@abi-omnipotent +.align 16 +Camellia_DecryptBlock: + movl \$128,%eax + subl $arg0d,%eax + movl \$3,$arg0d + adcl \$0,$arg0d # keyBitLength==128?3:4 + jmp .Ldec_rounds +.size Camellia_DecryptBlock,.-Camellia_DecryptBlock +# V2 +.globl Camellia_DecryptBlock_Rounds +.type Camellia_DecryptBlock_Rounds,\@function,4 +.align 16 +.Ldec_rounds: +Camellia_DecryptBlock_Rounds: + push %rbx + push %rbp + push %r13 + push %r14 + push %r15 +.Ldec_prologue: + + #mov %rsi,$inp # put away arguments + mov %rcx,$out + mov %rdx,$keyend + + shl \$6,%edi # process grandRounds + lea .LCamellia_SBOX(%rip),$Tbl + lea ($keyend,%rdi),$key + + mov 0(%rsi),@S[0] # load plaintext + mov 4(%rsi),@S[1] + mov 8(%rsi),@S[2] + bswap @S[0] + mov 12(%rsi),@S[3] + bswap @S[1] + bswap @S[2] + bswap @S[3] + + call _x86_64_Camellia_decrypt + + bswap @S[0] + bswap @S[1] + bswap @S[2] + mov @S[0],0($out) + bswap @S[3] + mov @S[1],4($out) + mov @S[2],8($out) + mov @S[3],12($out) + + mov 0(%rsp),%r15 + mov 8(%rsp),%r14 + mov 16(%rsp),%r13 + mov 24(%rsp),%rbp + mov 32(%rsp),%rbx + lea 40(%rsp),%rsp +.Ldec_epilogue: + ret +.size Camellia_DecryptBlock_Rounds,.-Camellia_DecryptBlock_Rounds + +.type _x86_64_Camellia_decrypt,\@abi-omnipotent +.align 16 +_x86_64_Camellia_decrypt: + xor 0($key),@S[1] + xor 4($key),@S[0] # ^=key[0-3] + xor 8($key),@S[3] + xor 12($key),@S[2] +.align 16 +.Ldloop: + mov -8($key),$t1 # prefetch key[4-5] + mov -4($key),$t0 + +___ + for ($i=0;$i<6;$i++) { Camellia_Feistel($i,-8); } +$code.=<<___; + lea -16*4($key),$key + cmp $keyend,$key + mov 0($key),$t3 # prefetch key[2-3] + mov 4($key),$t2 + je .Lddone + + and @S[0],$t0 + or @S[3],$t3 + rol \$1,$t0 + xor $t3,@S[2] # s2^=s3|key[3]; + xor $t0,@S[1] # s1^=LeftRotate(s0&key[0],1); + and @S[2],$t2 + or @S[1],$t1 + rol \$1,$t2 + xor $t1,@S[0] # s0^=s1|key[1]; + xor $t2,@S[3] # s3^=LeftRotate(s2&key[2],1); + + jmp .Ldloop + +.align 16 +.Lddone: + xor @S[2],$t2 + xor @S[3],$t3 + xor @S[0],$t0 + xor @S[1],$t1 + + mov $t2,@S[0] # SwapHalf + mov $t3,@S[1] + mov $t0,@S[2] + mov $t1,@S[3] + + .byte 0xf3,0xc3 # rep ret +.size _x86_64_Camellia_decrypt,.-_x86_64_Camellia_decrypt +___ + +sub _saveround { +my ($rnd,$key,@T)=@_; +my $bias=int(@T[0])?shift(@T):0; + + if ($#T==3) { + $code.=<<___; + mov @T[1],`$bias+$rnd*8+0`($key) + mov @T[0],`$bias+$rnd*8+4`($key) + mov @T[3],`$bias+$rnd*8+8`($key) + mov @T[2],`$bias+$rnd*8+12`($key) +___ + } else { + $code.=" mov @T[0],`$bias+$rnd*8+0`($key)\n"; + $code.=" mov @T[1],`$bias+$rnd*8+8`($key)\n" if ($#T>=1); + } +} + +sub _loadround { +my ($rnd,$key,@T)=@_; +my $bias=int(@T[0])?shift(@T):0; + +$code.=" mov `$bias+$rnd*8+0`($key),@T[0]\n"; +$code.=" mov `$bias+$rnd*8+8`($key),@T[1]\n" if ($#T>=1); +} + +# shld is very slow on Intel EM64T family. Even on AMD it limits +# instruction decode rate [because it's VectorPath] and consequently +# performance... +sub __rotl128 { +my ($i0,$i1,$rot)=@_; + + if ($rot) { + $code.=<<___; + mov $i0,%r11 + shld \$$rot,$i1,$i0 + shld \$$rot,%r11,$i1 +___ + } +} + +# ... Implementing 128-bit rotate without shld gives 80% better +# performance EM64T, +15% on AMD64 and only ~7% degradation on +# Core2. This is therefore preferred. +sub _rotl128 { +my ($i0,$i1,$rot)=@_; + + if ($rot) { + $code.=<<___; + mov $i0,%r11 + shl \$$rot,$i0 + mov $i1,%r9 + shr \$`64-$rot`,%r9 + shr \$`64-$rot`,%r11 + or %r9,$i0 + shl \$$rot,$i1 + or %r11,$i1 +___ + } +} + +{ my $step=0; + +$code.=<<___; +.globl Camellia_Ekeygen +.type Camellia_Ekeygen,\@function,3 +.align 16 +Camellia_Ekeygen: + push %rbx + push %rbp + push %r13 + push %r14 + push %r15 +.Lkey_prologue: + + mov %edi,${keyend}d # put away arguments, keyBitLength + mov %rdx,$out # keyTable + + mov 0(%rsi),@S[0] # load 0-127 bits + mov 4(%rsi),@S[1] + mov 8(%rsi),@S[2] + mov 12(%rsi),@S[3] + + bswap @S[0] + bswap @S[1] + bswap @S[2] + bswap @S[3] +___ + &_saveround (0,$out,@S); # KL<<<0 +$code.=<<___; + cmp \$128,$keyend # check keyBitLength + je .L1st128 + + mov 16(%rsi),@S[0] # load 128-191 bits + mov 20(%rsi),@S[1] + cmp \$192,$keyend + je .L1st192 + mov 24(%rsi),@S[2] # load 192-255 bits + mov 28(%rsi),@S[3] + jmp .L1st256 +.L1st192: + mov @S[0],@S[2] + mov @S[1],@S[3] + not @S[2] + not @S[3] +.L1st256: + bswap @S[0] + bswap @S[1] + bswap @S[2] + bswap @S[3] +___ + &_saveround (4,$out,@S); # temp storage for KR! +$code.=<<___; + xor 0($out),@S[1] # KR^KL + xor 4($out),@S[0] + xor 8($out),@S[3] + xor 12($out),@S[2] + +.L1st128: + lea .LCamellia_SIGMA(%rip),$key + lea .LCamellia_SBOX(%rip),$Tbl + + mov 0($key),$t1 + mov 4($key),$t0 +___ + &Camellia_Feistel($step++); + &Camellia_Feistel($step++); +$code.=<<___; + xor 0($out),@S[1] # ^KL + xor 4($out),@S[0] + xor 8($out),@S[3] + xor 12($out),@S[2] +___ + &Camellia_Feistel($step++); + &Camellia_Feistel($step++); +$code.=<<___; + cmp \$128,$keyend + jne .L2nd256 + + lea 128($out),$out # size optimization + shl \$32,%r8 # @S[0]|| + shl \$32,%r10 # @S[2]|| + or %r9,%r8 # ||@S[1] + or %r11,%r10 # ||@S[3] +___ + &_loadround (0,$out,-128,"%rax","%rbx"); # KL + &_saveround (2,$out,-128,"%r8","%r10"); # KA<<<0 + &_rotl128 ("%rax","%rbx",15); + &_saveround (4,$out,-128,"%rax","%rbx"); # KL<<<15 + &_rotl128 ("%r8","%r10",15); + &_saveround (6,$out,-128,"%r8","%r10"); # KA<<<15 + &_rotl128 ("%r8","%r10",15); # 15+15=30 + &_saveround (8,$out,-128,"%r8","%r10"); # KA<<<30 + &_rotl128 ("%rax","%rbx",30); # 15+30=45 + &_saveround (10,$out,-128,"%rax","%rbx"); # KL<<<45 + &_rotl128 ("%r8","%r10",15); # 30+15=45 + &_saveround (12,$out,-128,"%r8"); # KA<<<45 + &_rotl128 ("%rax","%rbx",15); # 45+15=60 + &_saveround (13,$out,-128,"%rbx"); # KL<<<60 + &_rotl128 ("%r8","%r10",15); # 45+15=60 + &_saveround (14,$out,-128,"%r8","%r10"); # KA<<<60 + &_rotl128 ("%rax","%rbx",17); # 60+17=77 + &_saveround (16,$out,-128,"%rax","%rbx"); # KL<<<77 + &_rotl128 ("%rax","%rbx",17); # 77+17=94 + &_saveround (18,$out,-128,"%rax","%rbx"); # KL<<<94 + &_rotl128 ("%r8","%r10",34); # 60+34=94 + &_saveround (20,$out,-128,"%r8","%r10"); # KA<<<94 + &_rotl128 ("%rax","%rbx",17); # 94+17=111 + &_saveround (22,$out,-128,"%rax","%rbx"); # KL<<<111 + &_rotl128 ("%r8","%r10",17); # 94+17=111 + &_saveround (24,$out,-128,"%r8","%r10"); # KA<<<111 +$code.=<<___; + mov \$3,%eax + jmp .Ldone +.align 16 +.L2nd256: +___ + &_saveround (6,$out,@S); # temp storage for KA! +$code.=<<___; + xor `4*8+0`($out),@S[1] # KA^KR + xor `4*8+4`($out),@S[0] + xor `5*8+0`($out),@S[3] + xor `5*8+4`($out),@S[2] +___ + &Camellia_Feistel($step++); + &Camellia_Feistel($step++); + + &_loadround (0,$out,"%rax","%rbx"); # KL + &_loadround (4,$out,"%rcx","%rdx"); # KR + &_loadround (6,$out,"%r14","%r15"); # KA +$code.=<<___; + lea 128($out),$out # size optimization + shl \$32,%r8 # @S[0]|| + shl \$32,%r10 # @S[2]|| + or %r9,%r8 # ||@S[1] + or %r11,%r10 # ||@S[3] +___ + &_saveround (2,$out,-128,"%r8","%r10"); # KB<<<0 + &_rotl128 ("%rcx","%rdx",15); + &_saveround (4,$out,-128,"%rcx","%rdx"); # KR<<<15 + &_rotl128 ("%r14","%r15",15); + &_saveround (6,$out,-128,"%r14","%r15"); # KA<<<15 + &_rotl128 ("%rcx","%rdx",15); # 15+15=30 + &_saveround (8,$out,-128,"%rcx","%rdx"); # KR<<<30 + &_rotl128 ("%r8","%r10",30); + &_saveround (10,$out,-128,"%r8","%r10"); # KB<<<30 + &_rotl128 ("%rax","%rbx",45); + &_saveround (12,$out,-128,"%rax","%rbx"); # KL<<<45 + &_rotl128 ("%r14","%r15",30); # 15+30=45 + &_saveround (14,$out,-128,"%r14","%r15"); # KA<<<45 + &_rotl128 ("%rax","%rbx",15); # 45+15=60 + &_saveround (16,$out,-128,"%rax","%rbx"); # KL<<<60 + &_rotl128 ("%rcx","%rdx",30); # 30+30=60 + &_saveround (18,$out,-128,"%rcx","%rdx"); # KR<<<60 + &_rotl128 ("%r8","%r10",30); # 30+30=60 + &_saveround (20,$out,-128,"%r8","%r10"); # KB<<<60 + &_rotl128 ("%rax","%rbx",17); # 60+17=77 + &_saveround (22,$out,-128,"%rax","%rbx"); # KL<<<77 + &_rotl128 ("%r14","%r15",32); # 45+32=77 + &_saveround (24,$out,-128,"%r14","%r15"); # KA<<<77 + &_rotl128 ("%rcx","%rdx",34); # 60+34=94 + &_saveround (26,$out,-128,"%rcx","%rdx"); # KR<<<94 + &_rotl128 ("%r14","%r15",17); # 77+17=94 + &_saveround (28,$out,-128,"%r14","%r15"); # KA<<<77 + &_rotl128 ("%rax","%rbx",34); # 77+34=111 + &_saveround (30,$out,-128,"%rax","%rbx"); # KL<<<111 + &_rotl128 ("%r8","%r10",51); # 60+51=111 + &_saveround (32,$out,-128,"%r8","%r10"); # KB<<<111 +$code.=<<___; + mov \$4,%eax +.Ldone: + mov 0(%rsp),%r15 + mov 8(%rsp),%r14 + mov 16(%rsp),%r13 + mov 24(%rsp),%rbp + mov 32(%rsp),%rbx + lea 40(%rsp),%rsp +.Lkey_epilogue: + ret +.size Camellia_Ekeygen,.-Camellia_Ekeygen +___ +} + +@SBOX=( +112,130, 44,236,179, 39,192,229,228,133, 87, 53,234, 12,174, 65, + 35,239,107,147, 69, 25,165, 33,237, 14, 79, 78, 29,101,146,189, +134,184,175,143,124,235, 31,206, 62, 48,220, 95, 94,197, 11, 26, +166,225, 57,202,213, 71, 93, 61,217, 1, 90,214, 81, 86,108, 77, +139, 13,154,102,251,204,176, 45,116, 18, 43, 32,240,177,132,153, +223, 76,203,194, 52,126,118, 5,109,183,169, 49,209, 23, 4,215, + 20, 88, 58, 97,222, 27, 17, 28, 50, 15,156, 22, 83, 24,242, 34, +254, 68,207,178,195,181,122,145, 36, 8,232,168, 96,252,105, 80, +170,208,160,125,161,137, 98,151, 84, 91, 30,149,224,255,100,210, + 16,196, 0, 72,163,247,117,219,138, 3,230,218, 9, 63,221,148, +135, 92,131, 2,205, 74,144, 51,115,103,246,243,157,127,191,226, + 82,155,216, 38,200, 55,198, 59,129,150,111, 75, 19,190, 99, 46, +233,121,167,140,159,110,188,142, 41,245,249,182, 47,253,180, 89, +120,152, 6,106,231, 70,113,186,212, 37,171, 66,136,162,141,250, +114, 7,185, 85,248,238,172, 10, 54, 73, 42,104, 60, 56,241,164, + 64, 40,211,123,187,201, 67,193, 21,227,173,244,119,199,128,158); + +sub S1110 { my $i=shift; $i=@SBOX[$i]; $i=$i<<24|$i<<16|$i<<8; sprintf("0x%08x",$i); } +sub S4404 { my $i=shift; $i=($i<<1|$i>>7)&0xff; $i=@SBOX[$i]; $i=$i<<24|$i<<16|$i; sprintf("0x%08x",$i); } +sub S0222 { my $i=shift; $i=@SBOX[$i]; $i=($i<<1|$i>>7)&0xff; $i=$i<<16|$i<<8|$i; sprintf("0x%08x",$i); } +sub S3033 { my $i=shift; $i=@SBOX[$i]; $i=($i>>1|$i<<7)&0xff; $i=$i<<24|$i<<8|$i; sprintf("0x%08x",$i); } + +$code.=<<___; +.align 64 +.LCamellia_SIGMA: +.long 0x3bcc908b, 0xa09e667f, 0x4caa73b2, 0xb67ae858 +.long 0xe94f82be, 0xc6ef372f, 0xf1d36f1c, 0x54ff53a5 +.long 0xde682d1d, 0x10e527fa, 0xb3e6c1fd, 0xb05688c2 +.long 0, 0, 0, 0 +.LCamellia_SBOX: +___ +# tables are interleaved, remember? +sub data_word { $code.=".long\t".join(',',@_)."\n"; } +for ($i=0;$i<256;$i++) { &data_word(&S1110($i),&S4404($i)); } +for ($i=0;$i<256;$i++) { &data_word(&S0222($i),&S3033($i)); } + +# void Camellia_cbc_encrypt (const void char *inp, unsigned char *out, +# size_t length, const CAMELLIA_KEY *key, +# unsigned char *ivp,const int enc); +{ +$_key="0(%rsp)"; +$_end="8(%rsp)"; # inp+len&~15 +$_res="16(%rsp)"; # len&15 +$ivec="24(%rsp)"; +$_ivp="40(%rsp)"; +$_rsp="48(%rsp)"; + +$code.=<<___; +.globl Camellia_cbc_encrypt +.type Camellia_cbc_encrypt,\@function,6 +.align 16 +Camellia_cbc_encrypt: + cmp \$0,%rdx + je .Lcbc_abort + push %rbx + push %rbp + push %r12 + push %r13 + push %r14 + push %r15 +.Lcbc_prologue: + + mov %rsp,%rbp + sub \$64,%rsp + and \$-64,%rsp + + # place stack frame just "above mod 1024" the key schedule, + # this ensures that cache associativity suffices + lea -64-63(%rcx),%r10 + sub %rsp,%r10 + neg %r10 + and \$0x3C0,%r10 + sub %r10,%rsp + #add \$8,%rsp # 8 is reserved for callee's ra + + mov %rdi,$inp # inp argument + mov %rsi,$out # out argument + mov %r8,%rbx # ivp argument + mov %rcx,$key # key argument + mov 272(%rcx),${keyend}d # grandRounds + + mov %r8,$_ivp + mov %rbp,$_rsp + +.Lcbc_body: + lea .LCamellia_SBOX(%rip),$Tbl + + mov \$32,%ecx +.align 4 +.Lcbc_prefetch_sbox: + mov 0($Tbl),%rax + mov 32($Tbl),%rsi + mov 64($Tbl),%rdi + mov 96($Tbl),%r11 + lea 128($Tbl),$Tbl + loop .Lcbc_prefetch_sbox + sub \$4096,$Tbl + shl \$6,$keyend + mov %rdx,%rcx # len argument + lea ($key,$keyend),$keyend + + cmp \$0,%r9d # enc argument + je .LCBC_DECRYPT + + and \$-16,%rdx + and \$15,%rcx # length residue + lea ($inp,%rdx),%rdx + mov $key,$_key + mov %rdx,$_end + mov %rcx,$_res + + cmp $inp,%rdx + mov 0(%rbx),@S[0] # load IV + mov 4(%rbx),@S[1] + mov 8(%rbx),@S[2] + mov 12(%rbx),@S[3] + je .Lcbc_enc_tail + jmp .Lcbc_eloop + +.align 16 +.Lcbc_eloop: + xor 0($inp),@S[0] + xor 4($inp),@S[1] + xor 8($inp),@S[2] + bswap @S[0] + xor 12($inp),@S[3] + bswap @S[1] + bswap @S[2] + bswap @S[3] + + call _x86_64_Camellia_encrypt + + mov $_key,$key # "rewind" the key + bswap @S[0] + mov $_end,%rdx + bswap @S[1] + mov $_res,%rcx + bswap @S[2] + mov @S[0],0($out) + bswap @S[3] + mov @S[1],4($out) + mov @S[2],8($out) + lea 16($inp),$inp + mov @S[3],12($out) + cmp %rdx,$inp + lea 16($out),$out + jne .Lcbc_eloop + + cmp \$0,%rcx + jne .Lcbc_enc_tail + + mov $_ivp,$out + mov @S[0],0($out) # write out IV residue + mov @S[1],4($out) + mov @S[2],8($out) + mov @S[3],12($out) + jmp .Lcbc_done + +.align 16 +.Lcbc_enc_tail: + xor %rax,%rax + mov %rax,0+$ivec + mov %rax,8+$ivec + mov %rax,$_res + +.Lcbc_enc_pushf: + pushfq + cld + mov $inp,%rsi + lea 8+$ivec,%rdi + .long 0x9066A4F3 # rep movsb + popfq +.Lcbc_enc_popf: + + lea $ivec,$inp + lea 16+$ivec,%rax + mov %rax,$_end + jmp .Lcbc_eloop # one more time + +.align 16 +.LCBC_DECRYPT: + xchg $key,$keyend + add \$15,%rdx + and \$15,%rcx # length residue + and \$-16,%rdx + mov $key,$_key + lea ($inp,%rdx),%rdx + mov %rdx,$_end + mov %rcx,$_res + + mov (%rbx),%rax # load IV + mov 8(%rbx),%rbx + jmp .Lcbc_dloop +.align 16 +.Lcbc_dloop: + mov 0($inp),@S[0] + mov 4($inp),@S[1] + mov 8($inp),@S[2] + bswap @S[0] + mov 12($inp),@S[3] + bswap @S[1] + mov %rax,0+$ivec # save IV to temporary storage + bswap @S[2] + mov %rbx,8+$ivec + bswap @S[3] + + call _x86_64_Camellia_decrypt + + mov $_key,$key # "rewind" the key + mov $_end,%rdx + mov $_res,%rcx + + bswap @S[0] + mov ($inp),%rax # load IV for next iteration + bswap @S[1] + mov 8($inp),%rbx + bswap @S[2] + xor 0+$ivec,@S[0] + bswap @S[3] + xor 4+$ivec,@S[1] + xor 8+$ivec,@S[2] + lea 16($inp),$inp + xor 12+$ivec,@S[3] + cmp %rdx,$inp + je .Lcbc_ddone + + mov @S[0],0($out) + mov @S[1],4($out) + mov @S[2],8($out) + mov @S[3],12($out) + + lea 16($out),$out + jmp .Lcbc_dloop + +.align 16 +.Lcbc_ddone: + mov $_ivp,%rdx + cmp \$0,%rcx + jne .Lcbc_dec_tail + + mov @S[0],0($out) + mov @S[1],4($out) + mov @S[2],8($out) + mov @S[3],12($out) + + mov %rax,(%rdx) # write out IV residue + mov %rbx,8(%rdx) + jmp .Lcbc_done +.align 16 +.Lcbc_dec_tail: + mov @S[0],0+$ivec + mov @S[1],4+$ivec + mov @S[2],8+$ivec + mov @S[3],12+$ivec + +.Lcbc_dec_pushf: + pushfq + cld + lea 8+$ivec,%rsi + lea ($out),%rdi + .long 0x9066A4F3 # rep movsb + popfq +.Lcbc_dec_popf: + + mov %rax,(%rdx) # write out IV residue + mov %rbx,8(%rdx) + jmp .Lcbc_done + +.align 16 +.Lcbc_done: + mov $_rsp,%rcx + mov 0(%rcx),%r15 + mov 8(%rcx),%r14 + mov 16(%rcx),%r13 + mov 24(%rcx),%r12 + mov 32(%rcx),%rbp + mov 40(%rcx),%rbx + lea 48(%rcx),%rsp +.Lcbc_abort: + ret +.size Camellia_cbc_encrypt,.-Camellia_cbc_encrypt + +.asciz "Camellia for x86_64 by <appro\@openssl.org>" +___ +} + +# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame, +# CONTEXT *context,DISPATCHER_CONTEXT *disp) +if ($win64) { +$rec="%rcx"; +$frame="%rdx"; +$context="%r8"; +$disp="%r9"; + +$code.=<<___; +.extern __imp_RtlVirtualUnwind +.type common_se_handler,\@abi-omnipotent +.align 16 +common_se_handler: + push %rsi + push %rdi + push %rbx + push %rbp + push %r12 + push %r13 + push %r14 + push %r15 + pushfq + lea -64(%rsp),%rsp + + mov 120($context),%rax # pull context->Rax + mov 248($context),%rbx # pull context->Rip + + mov 8($disp),%rsi # disp->ImageBase + mov 56($disp),%r11 # disp->HandlerData + + mov 0(%r11),%r10d # HandlerData[0] + lea (%rsi,%r10),%r10 # prologue label + cmp %r10,%rbx # context->Rip<prologue label + jb .Lin_prologue + + mov 152($context),%rax # pull context->Rsp + + mov 4(%r11),%r10d # HandlerData[1] + lea (%rsi,%r10),%r10 # epilogue label + cmp %r10,%rbx # context->Rip>=epilogue label + jae .Lin_prologue + + lea 40(%rax),%rax + mov -8(%rax),%rbx + mov -16(%rax),%rbp + mov -24(%rax),%r13 + mov -32(%rax),%r14 + mov -40(%rax),%r15 + mov %rbx,144($context) # restore context->Rbx + mov %rbp,160($context) # restore context->Rbp + mov %r13,224($context) # restore context->R13 + mov %r14,232($context) # restore context->R14 + mov %r15,240($context) # restore context->R15 + +.Lin_prologue: + mov 8(%rax),%rdi + mov 16(%rax),%rsi + mov %rax,152($context) # restore context->Rsp + mov %rsi,168($context) # restore context->Rsi + mov %rdi,176($context) # restore context->Rdi + + jmp .Lcommon_seh_exit +.size common_se_handler,.-common_se_handler + +.type cbc_se_handler,\@abi-omnipotent +.align 16 +cbc_se_handler: + push %rsi + push %rdi + push %rbx + push %rbp + push %r12 + push %r13 + push %r14 + push %r15 + pushfq + lea -64(%rsp),%rsp + + mov 120($context),%rax # pull context->Rax + mov 248($context),%rbx # pull context->Rip + + lea .Lcbc_prologue(%rip),%r10 + cmp %r10,%rbx # context->Rip<.Lcbc_prologue + jb .Lin_cbc_prologue + + lea .Lcbc_body(%rip),%r10 + cmp %r10,%rbx # context->Rip<.Lcbc_body + jb .Lin_cbc_frame_setup + + mov 152($context),%rax # pull context->Rsp + + lea .Lcbc_abort(%rip),%r10 + cmp %r10,%rbx # context->Rip>=.Lcbc_abort + jae .Lin_cbc_prologue + + # handle pushf/popf in Camellia_cbc_encrypt + lea .Lcbc_enc_pushf(%rip),%r10 + cmp %r10,%rbx # context->Rip<=.Lcbc_enc_pushf + jbe .Lin_cbc_no_flag + lea 8(%rax),%rax + lea .Lcbc_enc_popf(%rip),%r10 + cmp %r10,%rbx # context->Rip<.Lcbc_enc_popf + jb .Lin_cbc_no_flag + lea -8(%rax),%rax + lea .Lcbc_dec_pushf(%rip),%r10 + cmp %r10,%rbx # context->Rip<=.Lcbc_dec_pushf + jbe .Lin_cbc_no_flag + lea 8(%rax),%rax + lea .Lcbc_dec_popf(%rip),%r10 + cmp %r10,%rbx # context->Rip<.Lcbc_dec_popf + jb .Lin_cbc_no_flag + lea -8(%rax),%rax + +.Lin_cbc_no_flag: + mov 48(%rax),%rax # $_rsp + lea 48(%rax),%rax + +.Lin_cbc_frame_setup: + mov -8(%rax),%rbx + mov -16(%rax),%rbp + mov -24(%rax),%r12 + mov -32(%rax),%r13 + mov -40(%rax),%r14 + mov -48(%rax),%r15 + mov %rbx,144($context) # restore context->Rbx + mov %rbp,160($context) # restore context->Rbp + mov %r12,216($context) # restore context->R12 + mov %r13,224($context) # restore context->R13 + mov %r14,232($context) # restore context->R14 + mov %r15,240($context) # restore context->R15 + +.Lin_cbc_prologue: + mov 8(%rax),%rdi + mov 16(%rax),%rsi + mov %rax,152($context) # restore context->Rsp + mov %rsi,168($context) # restore context->Rsi + mov %rdi,176($context) # restore context->Rdi + +.align 4 +.Lcommon_seh_exit: + + mov 40($disp),%rdi # disp->ContextRecord + mov $context,%rsi # context + mov \$`1232/8`,%ecx # sizeof(CONTEXT) + .long 0xa548f3fc # cld; rep movsq + + mov $disp,%rsi + xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER + mov 8(%rsi),%rdx # arg2, disp->ImageBase + mov 0(%rsi),%r8 # arg3, disp->ControlPc + mov 16(%rsi),%r9 # arg4, disp->FunctionEntry + mov 40(%rsi),%r10 # disp->ContextRecord + lea 56(%rsi),%r11 # &disp->HandlerData + lea 24(%rsi),%r12 # &disp->EstablisherFrame + mov %r10,32(%rsp) # arg5 + mov %r11,40(%rsp) # arg6 + mov %r12,48(%rsp) # arg7 + mov %rcx,56(%rsp) # arg8, (NULL) + call *__imp_RtlVirtualUnwind(%rip) + + mov \$1,%eax # ExceptionContinueSearch + lea 64(%rsp),%rsp + popfq + pop %r15 + pop %r14 + pop %r13 + pop %r12 + pop %rbp + pop %rbx + pop %rdi + pop %rsi + ret +.size cbc_se_handler,.-cbc_se_handler + +.section .pdata +.align 4 + .rva .LSEH_begin_Camellia_EncryptBlock_Rounds + .rva .LSEH_end_Camellia_EncryptBlock_Rounds + .rva .LSEH_info_Camellia_EncryptBlock_Rounds + + .rva .LSEH_begin_Camellia_DecryptBlock_Rounds + .rva .LSEH_end_Camellia_DecryptBlock_Rounds + .rva .LSEH_info_Camellia_DecryptBlock_Rounds + + .rva .LSEH_begin_Camellia_Ekeygen + .rva .LSEH_end_Camellia_Ekeygen + .rva .LSEH_info_Camellia_Ekeygen + + .rva .LSEH_begin_Camellia_cbc_encrypt + .rva .LSEH_end_Camellia_cbc_encrypt + .rva .LSEH_info_Camellia_cbc_encrypt + +.section .xdata +.align 8 +.LSEH_info_Camellia_EncryptBlock_Rounds: + .byte 9,0,0,0 + .rva common_se_handler + .rva .Lenc_prologue,.Lenc_epilogue # HandlerData[] +.LSEH_info_Camellia_DecryptBlock_Rounds: + .byte 9,0,0,0 + .rva common_se_handler + .rva .Ldec_prologue,.Ldec_epilogue # HandlerData[] +.LSEH_info_Camellia_Ekeygen: + .byte 9,0,0,0 + .rva common_se_handler + .rva .Lkey_prologue,.Lkey_epilogue # HandlerData[] +.LSEH_info_Camellia_cbc_encrypt: + .byte 9,0,0,0 + .rva cbc_se_handler +___ +} + +$code =~ s/\`([^\`]*)\`/eval $1/gem; +print $code; +close STDOUT; diff --git a/openssl-1.1.0h/crypto/camellia/asm/cmllt4-sparcv9.pl b/openssl-1.1.0h/crypto/camellia/asm/cmllt4-sparcv9.pl new file mode 100644 index 0000000..ffe4a7d --- /dev/null +++ b/openssl-1.1.0h/crypto/camellia/asm/cmllt4-sparcv9.pl @@ -0,0 +1,939 @@ +#! /usr/bin/env perl +# Copyright 2012-2016 The OpenSSL Project Authors. All Rights Reserved. +# +# Licensed under the OpenSSL license (the "License"). You may not use +# this file except in compliance with the License. You can obtain a copy +# in the file LICENSE in the source distribution or at +# https://www.openssl.org/source/license.html + + +# ==================================================================== +# Written by David S. Miller <davem@devemloft.net> and Andy Polyakov +# <appro@openssl.org>. The module is licensed under 2-clause BSD +# license. October 2012. All rights reserved. +# ==================================================================== + +###################################################################### +# Camellia for SPARC T4. +# +# As with AES below results [for aligned data] are virtually identical +# to critical path lenths for 3-cycle instruction latency: +# +# 128-bit key 192/256- +# CBC encrypt 4.14/4.21(*) 5.46/5.52 +# (*) numbers after slash are for +# misaligned data. +# +# As with Intel AES-NI, question is if it's possible to improve +# performance of parallelizeable modes by interleaving round +# instructions. In Camellia every instruction is dependent on +# previous, which means that there is place for 2 additional ones +# in between two dependent. Can we expect 3x performance improvement? +# At least one can argue that it should be possible to break 2x +# barrier... For some reason not even 2x appears to be possible: +# +# 128-bit key 192/256- +# CBC decrypt 2.21/2.74 2.99/3.40 +# CTR 2.15/2.68(*) 2.93/3.34 +# (*) numbers after slash are for +# misaligned data. +# +# This is for 2x interleave. But compared to 1x interleave CBC decrypt +# improved by ... 0% for 128-bit key, and 11% for 192/256-bit one. +# So that out-of-order execution logic can take non-interleaved code +# to 1.87x, but can't take 2x interleaved one any further. There +# surely is some explanation... As result 3x interleave was not even +# attempted. Instead an effort was made to share specific modes +# implementations with AES module (therefore sparct4_modes.pl). +# +# To anchor to something else, software C implementation processes +# one byte in 38 cycles with 128-bit key on same processor. + +$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1; +push(@INC,"${dir}","${dir}../../perlasm"); +require "sparcv9_modes.pl"; + +$output = pop; +open STDOUT,">$output"; + +$::evp=1; # if $evp is set to 0, script generates module with +# Camellia_[en|de]crypt, Camellia_set_key and Camellia_cbc_encrypt +# entry points. These are fully compatible with openssl/camellia.h. + +###################################################################### +# single-round subroutines +# +{ +my ($inp,$out,$key,$rounds,$tmp,$mask)=map("%o$_",(0..5)); + +$code=<<___; +#include "sparc_arch.h" + +.text + +.globl cmll_t4_encrypt +.align 32 +cmll_t4_encrypt: + andcc $inp, 7, %g1 ! is input aligned? + andn $inp, 7, $inp + + ldx [$key + 0], %g4 + ldx [$key + 8], %g5 + + ldx [$inp + 0], %o4 + bz,pt %icc, 1f + ldx [$inp + 8], %o5 + ldx [$inp + 16], $inp + sll %g1, 3, %g1 + sub %g0, %g1, %o3 + sllx %o4, %g1, %o4 + sllx %o5, %g1, %g1 + srlx %o5, %o3, %o5 + srlx $inp, %o3, %o3 + or %o5, %o4, %o4 + or %o3, %g1, %o5 +1: + ld [$key + 272], $rounds ! grandRounds, 3 or 4 + ldd [$key + 16], %f12 + ldd [$key + 24], %f14 + xor %g4, %o4, %o4 + xor %g5, %o5, %o5 + ldd [$key + 32], %f16 + ldd [$key + 40], %f18 + movxtod %o4, %f0 + movxtod %o5, %f2 + ldd [$key + 48], %f20 + ldd [$key + 56], %f22 + sub $rounds, 1, $rounds + ldd [$key + 64], %f24 + ldd [$key + 72], %f26 + add $key, 80, $key + +.Lenc: + camellia_f %f12, %f2, %f0, %f2 + ldd [$key + 0], %f12 + sub $rounds,1,$rounds + camellia_f %f14, %f0, %f2, %f0 + ldd [$key + 8], %f14 + camellia_f %f16, %f2, %f0, %f2 + ldd [$key + 16], %f16 + camellia_f %f18, %f0, %f2, %f0 + ldd [$key + 24], %f18 + camellia_f %f20, %f2, %f0, %f2 + ldd [$key + 32], %f20 + camellia_f %f22, %f0, %f2, %f0 + ldd [$key + 40], %f22 + camellia_fl %f24, %f0, %f0 + ldd [$key + 48], %f24 + camellia_fli %f26, %f2, %f2 + ldd [$key + 56], %f26 + brnz,pt $rounds, .Lenc + add $key, 64, $key + + andcc $out, 7, $tmp ! is output aligned? + camellia_f %f12, %f2, %f0, %f2 + camellia_f %f14, %f0, %f2, %f0 + camellia_f %f16, %f2, %f0, %f2 + camellia_f %f18, %f0, %f2, %f0 + camellia_f %f20, %f2, %f0, %f4 + camellia_f %f22, %f0, %f4, %f2 + fxor %f24, %f4, %f0 + fxor %f26, %f2, %f2 + + bnz,pn %icc, 2f + nop + + std %f0, [$out + 0] + retl + std %f2, [$out + 8] + +2: alignaddrl $out, %g0, $out + mov 0xff, $mask + srl $mask, $tmp, $mask + + faligndata %f0, %f0, %f4 + faligndata %f0, %f2, %f6 + faligndata %f2, %f2, %f8 + + stda %f4, [$out + $mask]0xc0 ! partial store + std %f6, [$out + 8] + add $out, 16, $out + orn %g0, $mask, $mask + retl + stda %f8, [$out + $mask]0xc0 ! partial store +.type cmll_t4_encrypt,#function +.size cmll_t4_encrypt,.-cmll_t4_encrypt + +.globl cmll_t4_decrypt +.align 32 +cmll_t4_decrypt: + ld [$key + 272], $rounds ! grandRounds, 3 or 4 + andcc $inp, 7, %g1 ! is input aligned? + andn $inp, 7, $inp + + sll $rounds, 6, $rounds + add $rounds, $key, $key + + ldx [$inp + 0], %o4 + bz,pt %icc, 1f + ldx [$inp + 8], %o5 + ldx [$inp + 16], $inp + sll %g1, 3, %g1 + sub %g0, %g1, %g4 + sllx %o4, %g1, %o4 + sllx %o5, %g1, %g1 + srlx %o5, %g4, %o5 + srlx $inp, %g4, %g4 + or %o5, %o4, %o4 + or %g4, %g1, %o5 +1: + ldx [$key + 0], %g4 + ldx [$key + 8], %g5 + ldd [$key - 8], %f12 + ldd [$key - 16], %f14 + xor %g4, %o4, %o4 + xor %g5, %o5, %o5 + ldd [$key - 24], %f16 + ldd [$key - 32], %f18 + movxtod %o4, %f0 + movxtod %o5, %f2 + ldd [$key - 40], %f20 + ldd [$key - 48], %f22 + sub $rounds, 64, $rounds + ldd [$key - 56], %f24 + ldd [$key - 64], %f26 + sub $key, 64, $key + +.Ldec: + camellia_f %f12, %f2, %f0, %f2 + ldd [$key - 8], %f12 + sub $rounds, 64, $rounds + camellia_f %f14, %f0, %f2, %f0 + ldd [$key - 16], %f14 + camellia_f %f16, %f2, %f0, %f2 + ldd [$key - 24], %f16 + camellia_f %f18, %f0, %f2, %f0 + ldd [$key - 32], %f18 + camellia_f %f20, %f2, %f0, %f2 + ldd [$key - 40], %f20 + camellia_f %f22, %f0, %f2, %f0 + ldd [$key - 48], %f22 + camellia_fl %f24, %f0, %f0 + ldd [$key - 56], %f24 + camellia_fli %f26, %f2, %f2 + ldd [$key - 64], %f26 + brnz,pt $rounds, .Ldec + sub $key, 64, $key + + andcc $out, 7, $tmp ! is output aligned? + camellia_f %f12, %f2, %f0, %f2 + camellia_f %f14, %f0, %f2, %f0 + camellia_f %f16, %f2, %f0, %f2 + camellia_f %f18, %f0, %f2, %f0 + camellia_f %f20, %f2, %f0, %f4 + camellia_f %f22, %f0, %f4, %f2 + fxor %f26, %f4, %f0 + fxor %f24, %f2, %f2 + + bnz,pn %icc, 2f + nop + + std %f0, [$out + 0] + retl + std %f2, [$out + 8] + +2: alignaddrl $out, %g0, $out + mov 0xff, $mask + srl $mask, $tmp, $mask + + faligndata %f0, %f0, %f4 + faligndata %f0, %f2, %f6 + faligndata %f2, %f2, %f8 + + stda %f4, [$out + $mask]0xc0 ! partial store + std %f6, [$out + 8] + add $out, 16, $out + orn %g0, $mask, $mask + retl + stda %f8, [$out + $mask]0xc0 ! partial store +.type cmll_t4_decrypt,#function +.size cmll_t4_decrypt,.-cmll_t4_decrypt +___ +} + +###################################################################### +# key setup subroutines +# +{ +sub ROTL128 { + my $rot = shift; + + "srlx %o4, 64-$rot, %g4\n\t". + "sllx %o4, $rot, %o4\n\t". + "srlx %o5, 64-$rot, %g5\n\t". + "sllx %o5, $rot, %o5\n\t". + "or %o4, %g5, %o4\n\t". + "or %o5, %g4, %o5"; +} + +my ($inp,$bits,$out,$tmp)=map("%o$_",(0..5)); +$code.=<<___; +.globl cmll_t4_set_key +.align 32 +cmll_t4_set_key: + and $inp, 7, $tmp + alignaddr $inp, %g0, $inp + cmp $bits, 192 + ldd [$inp + 0], %f0 + bl,pt %icc,.L128 + ldd [$inp + 8], %f2 + + be,pt %icc,.L192 + ldd [$inp + 16], %f4 + + brz,pt $tmp, .L256aligned + ldd [$inp + 24], %f6 + + ldd [$inp + 32], %f8 + faligndata %f0, %f2, %f0 + faligndata %f2, %f4, %f2 + faligndata %f4, %f6, %f4 + b .L256aligned + faligndata %f6, %f8, %f6 + +.align 16 +.L192: + brz,a,pt $tmp, .L256aligned + fnot2 %f4, %f6 + + ldd [$inp + 24], %f6 + nop + faligndata %f0, %f2, %f0 + faligndata %f2, %f4, %f2 + faligndata %f4, %f6, %f4 + fnot2 %f4, %f6 + +.L256aligned: + std %f0, [$out + 0] ! k[0, 1] + fsrc2 %f0, %f28 + std %f2, [$out + 8] ! k[2, 3] + fsrc2 %f2, %f30 + fxor %f4, %f0, %f0 + b .L128key + fxor %f6, %f2, %f2 + +.align 16 +.L128: + brz,pt $tmp, .L128aligned + nop + + ldd [$inp + 16], %f4 + nop + faligndata %f0, %f2, %f0 + faligndata %f2, %f4, %f2 + +.L128aligned: + std %f0, [$out + 0] ! k[0, 1] + fsrc2 %f0, %f28 + std %f2, [$out + 8] ! k[2, 3] + fsrc2 %f2, %f30 + +.L128key: + mov %o7, %o5 +1: call .+8 + add %o7, SIGMA-1b, %o4 + mov %o5, %o7 + + ldd [%o4 + 0], %f16 + ldd [%o4 + 8], %f18 + ldd [%o4 + 16], %f20 + ldd [%o4 + 24], %f22 + + camellia_f %f16, %f2, %f0, %f2 + camellia_f %f18, %f0, %f2, %f0 + fxor %f28, %f0, %f0 + fxor %f30, %f2, %f2 + camellia_f %f20, %f2, %f0, %f2 + camellia_f %f22, %f0, %f2, %f0 + + bge,pn %icc, .L256key + nop + std %f0, [$out + 0x10] ! k[ 4, 5] + std %f2, [$out + 0x18] ! k[ 6, 7] + + movdtox %f0, %o4 + movdtox %f2, %o5 + `&ROTL128(15)` + stx %o4, [$out + 0x30] ! k[12, 13] + stx %o5, [$out + 0x38] ! k[14, 15] + `&ROTL128(15)` + stx %o4, [$out + 0x40] ! k[16, 17] + stx %o5, [$out + 0x48] ! k[18, 19] + `&ROTL128(15)` + stx %o4, [$out + 0x60] ! k[24, 25] + `&ROTL128(15)` + stx %o4, [$out + 0x70] ! k[28, 29] + stx %o5, [$out + 0x78] ! k[30, 31] + `&ROTL128(34)` + stx %o4, [$out + 0xa0] ! k[40, 41] + stx %o5, [$out + 0xa8] ! k[42, 43] + `&ROTL128(17)` + stx %o4, [$out + 0xc0] ! k[48, 49] + stx %o5, [$out + 0xc8] ! k[50, 51] + + movdtox %f28, %o4 ! k[ 0, 1] + movdtox %f30, %o5 ! k[ 2, 3] + `&ROTL128(15)` + stx %o4, [$out + 0x20] ! k[ 8, 9] + stx %o5, [$out + 0x28] ! k[10, 11] + `&ROTL128(30)` + stx %o4, [$out + 0x50] ! k[20, 21] + stx %o5, [$out + 0x58] ! k[22, 23] + `&ROTL128(15)` + stx %o5, [$out + 0x68] ! k[26, 27] + `&ROTL128(17)` + stx %o4, [$out + 0x80] ! k[32, 33] + stx %o5, [$out + 0x88] ! k[34, 35] + `&ROTL128(17)` + stx %o4, [$out + 0x90] ! k[36, 37] + stx %o5, [$out + 0x98] ! k[38, 39] + `&ROTL128(17)` + stx %o4, [$out + 0xb0] ! k[44, 45] + stx %o5, [$out + 0xb8] ! k[46, 47] + + mov 3, $tmp + st $tmp, [$out + 0x110] + retl + xor %o0, %o0, %o0 + +.align 16 +.L256key: + ldd [%o4 + 32], %f24 + ldd [%o4 + 40], %f26 + + std %f0, [$out + 0x30] ! k[12, 13] + std %f2, [$out + 0x38] ! k[14, 15] + + fxor %f4, %f0, %f0 + fxor %f6, %f2, %f2 + camellia_f %f24, %f2, %f0, %f2 + camellia_f %f26, %f0, %f2, %f0 + + std %f0, [$out + 0x10] ! k[ 4, 5] + std %f2, [$out + 0x18] ! k[ 6, 7] + + movdtox %f0, %o4 + movdtox %f2, %o5 + `&ROTL128(30)` + stx %o4, [$out + 0x50] ! k[20, 21] + stx %o5, [$out + 0x58] ! k[22, 23] + `&ROTL128(30)` + stx %o4, [$out + 0xa0] ! k[40, 41] + stx %o5, [$out + 0xa8] ! k[42, 43] + `&ROTL128(51)` + stx %o4, [$out + 0x100] ! k[64, 65] + stx %o5, [$out + 0x108] ! k[66, 67] + + movdtox %f4, %o4 ! k[ 8, 9] + movdtox %f6, %o5 ! k[10, 11] + `&ROTL128(15)` + stx %o4, [$out + 0x20] ! k[ 8, 9] + stx %o5, [$out + 0x28] ! k[10, 11] + `&ROTL128(15)` + stx %o4, [$out + 0x40] ! k[16, 17] + stx %o5, [$out + 0x48] ! k[18, 19] + `&ROTL128(30)` + stx %o4, [$out + 0x90] ! k[36, 37] + stx %o5, [$out + 0x98] ! k[38, 39] + `&ROTL128(34)` + stx %o4, [$out + 0xd0] ! k[52, 53] + stx %o5, [$out + 0xd8] ! k[54, 55] + ldx [$out + 0x30], %o4 ! k[12, 13] + ldx [$out + 0x38], %o5 ! k[14, 15] + `&ROTL128(15)` + stx %o4, [$out + 0x30] ! k[12, 13] + stx %o5, [$out + 0x38] ! k[14, 15] + `&ROTL128(30)` + stx %o4, [$out + 0x70] ! k[28, 29] + stx %o5, [$out + 0x78] ! k[30, 31] + srlx %o4, 32, %g4 + srlx %o5, 32, %g5 + st %o4, [$out + 0xc0] ! k[48] + st %g5, [$out + 0xc4] ! k[49] + st %o5, [$out + 0xc8] ! k[50] + st %g4, [$out + 0xcc] ! k[51] + `&ROTL128(49)` + stx %o4, [$out + 0xe0] ! k[56, 57] + stx %o5, [$out + 0xe8] ! k[58, 59] + + movdtox %f28, %o4 ! k[ 0, 1] + movdtox %f30, %o5 ! k[ 2, 3] + `&ROTL128(45)` + stx %o4, [$out + 0x60] ! k[24, 25] + stx %o5, [$out + 0x68] ! k[26, 27] + `&ROTL128(15)` + stx %o4, [$out + 0x80] ! k[32, 33] + stx %o5, [$out + 0x88] ! k[34, 35] + `&ROTL128(17)` + stx %o4, [$out + 0xb0] ! k[44, 45] + stx %o5, [$out + 0xb8] ! k[46, 47] + `&ROTL128(34)` + stx %o4, [$out + 0xf0] ! k[60, 61] + stx %o5, [$out + 0xf8] ! k[62, 63] + + mov 4, $tmp + st $tmp, [$out + 0x110] + retl + xor %o0, %o0, %o0 +.type cmll_t4_set_key,#function +.size cmll_t4_set_key,.-cmll_t4_set_key +.align 32 +SIGMA: + .long 0xa09e667f, 0x3bcc908b, 0xb67ae858, 0x4caa73b2 + .long 0xc6ef372f, 0xe94f82be, 0x54ff53a5, 0xf1d36f1c + .long 0x10e527fa, 0xde682d1d, 0xb05688c2, 0xb3e6c1fd +.type SIGMA,#object +.size SIGMA,.-SIGMA +.asciz "Camellia for SPARC T4, David S. Miller, Andy Polyakov" +___ +} + +{{{ +my ($inp,$out,$len,$key,$ivec,$enc)=map("%i$_",(0..5)); +my ($ileft,$iright,$ooff,$omask,$ivoff)=map("%l$_",(1..7)); + +$code.=<<___; +.align 32 +_cmll128_load_enckey: + ldx [$key + 0], %g4 + ldx [$key + 8], %g5 +___ +for ($i=2; $i<26;$i++) { # load key schedule + $code.=<<___; + ldd [$key + `8*$i`], %f`12+2*$i` +___ +} +$code.=<<___; + retl + nop +.type _cmll128_load_enckey,#function +.size _cmll128_load_enckey,.-_cmll128_load_enckey +_cmll256_load_enckey=_cmll128_load_enckey + +.align 32 +_cmll256_load_deckey: + ldd [$key + 64], %f62 + ldd [$key + 72], %f60 + b .Load_deckey + add $key, 64, $key +_cmll128_load_deckey: + ldd [$key + 0], %f60 + ldd [$key + 8], %f62 +.Load_deckey: +___ +for ($i=2; $i<24;$i++) { # load key schedule + $code.=<<___; + ldd [$key + `8*$i`], %f`62-2*$i` +___ +} +$code.=<<___; + ldx [$key + 192], %g4 + retl + ldx [$key + 200], %g5 +.type _cmll256_load_deckey,#function +.size _cmll256_load_deckey,.-_cmll256_load_deckey + +.align 32 +_cmll128_encrypt_1x: +___ +for ($i=0; $i<3; $i++) { + $code.=<<___; + camellia_f %f`16+16*$i+0`, %f2, %f0, %f2 + camellia_f %f`16+16*$i+2`, %f0, %f2, %f0 + camellia_f %f`16+16*$i+4`, %f2, %f0, %f2 + camellia_f %f`16+16*$i+6`, %f0, %f2, %f0 +___ +$code.=<<___ if ($i<2); + camellia_f %f`16+16*$i+8`, %f2, %f0, %f2 + camellia_f %f`16+16*$i+10`, %f0, %f2, %f0 + camellia_fl %f`16+16*$i+12`, %f0, %f0 + camellia_fli %f`16+16*$i+14`, %f2, %f2 +___ +} +$code.=<<___; + camellia_f %f56, %f2, %f0, %f4 + camellia_f %f58, %f0, %f4, %f2 + fxor %f60, %f4, %f0 + retl + fxor %f62, %f2, %f2 +.type _cmll128_encrypt_1x,#function +.size _cmll128_encrypt_1x,.-_cmll128_encrypt_1x +_cmll128_decrypt_1x=_cmll128_encrypt_1x + +.align 32 +_cmll128_encrypt_2x: +___ +for ($i=0; $i<3; $i++) { + $code.=<<___; + camellia_f %f`16+16*$i+0`, %f2, %f0, %f2 + camellia_f %f`16+16*$i+0`, %f6, %f4, %f6 + camellia_f %f`16+16*$i+2`, %f0, %f2, %f0 + camellia_f %f`16+16*$i+2`, %f4, %f6, %f4 + camellia_f %f`16+16*$i+4`, %f2, %f0, %f2 + camellia_f %f`16+16*$i+4`, %f6, %f4, %f6 + camellia_f %f`16+16*$i+6`, %f0, %f2, %f0 + camellia_f %f`16+16*$i+6`, %f4, %f6, %f4 +___ +$code.=<<___ if ($i<2); + camellia_f %f`16+16*$i+8`, %f2, %f0, %f2 + camellia_f %f`16+16*$i+8`, %f6, %f4, %f6 + camellia_f %f`16+16*$i+10`, %f0, %f2, %f0 + camellia_f %f`16+16*$i+10`, %f4, %f6, %f4 + camellia_fl %f`16+16*$i+12`, %f0, %f0 + camellia_fl %f`16+16*$i+12`, %f4, %f4 + camellia_fli %f`16+16*$i+14`, %f2, %f2 + camellia_fli %f`16+16*$i+14`, %f6, %f6 +___ +} +$code.=<<___; + camellia_f %f56, %f2, %f0, %f8 + camellia_f %f56, %f6, %f4, %f10 + camellia_f %f58, %f0, %f8, %f2 + camellia_f %f58, %f4, %f10, %f6 + fxor %f60, %f8, %f0 + fxor %f60, %f10, %f4 + fxor %f62, %f2, %f2 + retl + fxor %f62, %f6, %f6 +.type _cmll128_encrypt_2x,#function +.size _cmll128_encrypt_2x,.-_cmll128_encrypt_2x +_cmll128_decrypt_2x=_cmll128_encrypt_2x + +.align 32 +_cmll256_encrypt_1x: + camellia_f %f16, %f2, %f0, %f2 + camellia_f %f18, %f0, %f2, %f0 + ldd [$key + 208], %f16 + ldd [$key + 216], %f18 + camellia_f %f20, %f2, %f0, %f2 + camellia_f %f22, %f0, %f2, %f0 + ldd [$key + 224], %f20 + ldd [$key + 232], %f22 + camellia_f %f24, %f2, %f0, %f2 + camellia_f %f26, %f0, %f2, %f0 + ldd [$key + 240], %f24 + ldd [$key + 248], %f26 + camellia_fl %f28, %f0, %f0 + camellia_fli %f30, %f2, %f2 + ldd [$key + 256], %f28 + ldd [$key + 264], %f30 +___ +for ($i=1; $i<3; $i++) { + $code.=<<___; + camellia_f %f`16+16*$i+0`, %f2, %f0, %f2 + camellia_f %f`16+16*$i+2`, %f0, %f2, %f0 + camellia_f %f`16+16*$i+4`, %f2, %f0, %f2 + camellia_f %f`16+16*$i+6`, %f0, %f2, %f0 + camellia_f %f`16+16*$i+8`, %f2, %f0, %f2 + camellia_f %f`16+16*$i+10`, %f0, %f2, %f0 + camellia_fl %f`16+16*$i+12`, %f0, %f0 + camellia_fli %f`16+16*$i+14`, %f2, %f2 +___ +} +$code.=<<___; + camellia_f %f16, %f2, %f0, %f2 + camellia_f %f18, %f0, %f2, %f0 + ldd [$key + 16], %f16 + ldd [$key + 24], %f18 + camellia_f %f20, %f2, %f0, %f2 + camellia_f %f22, %f0, %f2, %f0 + ldd [$key + 32], %f20 + ldd [$key + 40], %f22 + camellia_f %f24, %f2, %f0, %f4 + camellia_f %f26, %f0, %f4, %f2 + ldd [$key + 48], %f24 + ldd [$key + 56], %f26 + fxor %f28, %f4, %f0 + fxor %f30, %f2, %f2 + ldd [$key + 64], %f28 + retl + ldd [$key + 72], %f30 +.type _cmll256_encrypt_1x,#function +.size _cmll256_encrypt_1x,.-_cmll256_encrypt_1x + +.align 32 +_cmll256_encrypt_2x: + camellia_f %f16, %f2, %f0, %f2 + camellia_f %f16, %f6, %f4, %f6 + camellia_f %f18, %f0, %f2, %f0 + camellia_f %f18, %f4, %f6, %f4 + ldd [$key + 208], %f16 + ldd [$key + 216], %f18 + camellia_f %f20, %f2, %f0, %f2 + camellia_f %f20, %f6, %f4, %f6 + camellia_f %f22, %f0, %f2, %f0 + camellia_f %f22, %f4, %f6, %f4 + ldd [$key + 224], %f20 + ldd [$key + 232], %f22 + camellia_f %f24, %f2, %f0, %f2 + camellia_f %f24, %f6, %f4, %f6 + camellia_f %f26, %f0, %f2, %f0 + camellia_f %f26, %f4, %f6, %f4 + ldd [$key + 240], %f24 + ldd [$key + 248], %f26 + camellia_fl %f28, %f0, %f0 + camellia_fl %f28, %f4, %f4 + camellia_fli %f30, %f2, %f2 + camellia_fli %f30, %f6, %f6 + ldd [$key + 256], %f28 + ldd [$key + 264], %f30 +___ +for ($i=1; $i<3; $i++) { + $code.=<<___; + camellia_f %f`16+16*$i+0`, %f2, %f0, %f2 + camellia_f %f`16+16*$i+0`, %f6, %f4, %f6 + camellia_f %f`16+16*$i+2`, %f0, %f2, %f0 + camellia_f %f`16+16*$i+2`, %f4, %f6, %f4 + camellia_f %f`16+16*$i+4`, %f2, %f0, %f2 + camellia_f %f`16+16*$i+4`, %f6, %f4, %f6 + camellia_f %f`16+16*$i+6`, %f0, %f2, %f0 + camellia_f %f`16+16*$i+6`, %f4, %f6, %f4 + camellia_f %f`16+16*$i+8`, %f2, %f0, %f2 + camellia_f %f`16+16*$i+8`, %f6, %f4, %f6 + camellia_f %f`16+16*$i+10`, %f0, %f2, %f0 + camellia_f %f`16+16*$i+10`, %f4, %f6, %f4 + camellia_fl %f`16+16*$i+12`, %f0, %f0 + camellia_fl %f`16+16*$i+12`, %f4, %f4 + camellia_fli %f`16+16*$i+14`, %f2, %f2 + camellia_fli %f`16+16*$i+14`, %f6, %f6 +___ +} +$code.=<<___; + camellia_f %f16, %f2, %f0, %f2 + camellia_f %f16, %f6, %f4, %f6 + camellia_f %f18, %f0, %f2, %f0 + camellia_f %f18, %f4, %f6, %f4 + ldd [$key + 16], %f16 + ldd [$key + 24], %f18 + camellia_f %f20, %f2, %f0, %f2 + camellia_f %f20, %f6, %f4, %f6 + camellia_f %f22, %f0, %f2, %f0 + camellia_f %f22, %f4, %f6, %f4 + ldd [$key + 32], %f20 + ldd [$key + 40], %f22 + camellia_f %f24, %f2, %f0, %f8 + camellia_f %f24, %f6, %f4, %f10 + camellia_f %f26, %f0, %f8, %f2 + camellia_f %f26, %f4, %f10, %f6 + ldd [$key + 48], %f24 + ldd [$key + 56], %f26 + fxor %f28, %f8, %f0 + fxor %f28, %f10, %f4 + fxor %f30, %f2, %f2 + fxor %f30, %f6, %f6 + ldd [$key + 64], %f28 + retl + ldd [$key + 72], %f30 +.type _cmll256_encrypt_2x,#function +.size _cmll256_encrypt_2x,.-_cmll256_encrypt_2x + +.align 32 +_cmll256_decrypt_1x: + camellia_f %f16, %f2, %f0, %f2 + camellia_f %f18, %f0, %f2, %f0 + ldd [$key - 8], %f16 + ldd [$key - 16], %f18 + camellia_f %f20, %f2, %f0, %f2 + camellia_f %f22, %f0, %f2, %f0 + ldd [$key - 24], %f20 + ldd [$key - 32], %f22 + camellia_f %f24, %f2, %f0, %f2 + camellia_f %f26, %f0, %f2, %f0 + ldd [$key - 40], %f24 + ldd [$key - 48], %f26 + camellia_fl %f28, %f0, %f0 + camellia_fli %f30, %f2, %f2 + ldd [$key - 56], %f28 + ldd [$key - 64], %f30 +___ +for ($i=1; $i<3; $i++) { + $code.=<<___; + camellia_f %f`16+16*$i+0`, %f2, %f0, %f2 + camellia_f %f`16+16*$i+2`, %f0, %f2, %f0 + camellia_f %f`16+16*$i+4`, %f2, %f0, %f2 + camellia_f %f`16+16*$i+6`, %f0, %f2, %f0 + camellia_f %f`16+16*$i+8`, %f2, %f0, %f2 + camellia_f %f`16+16*$i+10`, %f0, %f2, %f0 + camellia_fl %f`16+16*$i+12`, %f0, %f0 + camellia_fli %f`16+16*$i+14`, %f2, %f2 +___ +} +$code.=<<___; + camellia_f %f16, %f2, %f0, %f2 + camellia_f %f18, %f0, %f2, %f0 + ldd [$key + 184], %f16 + ldd [$key + 176], %f18 + camellia_f %f20, %f2, %f0, %f2 + camellia_f %f22, %f0, %f2, %f0 + ldd [$key + 168], %f20 + ldd [$key + 160], %f22 + camellia_f %f24, %f2, %f0, %f4 + camellia_f %f26, %f0, %f4, %f2 + ldd [$key + 152], %f24 + ldd [$key + 144], %f26 + fxor %f30, %f4, %f0 + fxor %f28, %f2, %f2 + ldd [$key + 136], %f28 + retl + ldd [$key + 128], %f30 +.type _cmll256_decrypt_1x,#function +.size _cmll256_decrypt_1x,.-_cmll256_decrypt_1x + +.align 32 +_cmll256_decrypt_2x: + camellia_f %f16, %f2, %f0, %f2 + camellia_f %f16, %f6, %f4, %f6 + camellia_f %f18, %f0, %f2, %f0 + camellia_f %f18, %f4, %f6, %f4 + ldd [$key - 8], %f16 + ldd [$key - 16], %f18 + camellia_f %f20, %f2, %f0, %f2 + camellia_f %f20, %f6, %f4, %f6 + camellia_f %f22, %f0, %f2, %f0 + camellia_f %f22, %f4, %f6, %f4 + ldd [$key - 24], %f20 + ldd [$key - 32], %f22 + camellia_f %f24, %f2, %f0, %f2 + camellia_f %f24, %f6, %f4, %f6 + camellia_f %f26, %f0, %f2, %f0 + camellia_f %f26, %f4, %f6, %f4 + ldd [$key - 40], %f24 + ldd [$key - 48], %f26 + camellia_fl %f28, %f0, %f0 + camellia_fl %f28, %f4, %f4 + camellia_fli %f30, %f2, %f2 + camellia_fli %f30, %f6, %f6 + ldd [$key - 56], %f28 + ldd [$key - 64], %f30 +___ +for ($i=1; $i<3; $i++) { + $code.=<<___; + camellia_f %f`16+16*$i+0`, %f2, %f0, %f2 + camellia_f %f`16+16*$i+0`, %f6, %f4, %f6 + camellia_f %f`16+16*$i+2`, %f0, %f2, %f0 + camellia_f %f`16+16*$i+2`, %f4, %f6, %f4 + camellia_f %f`16+16*$i+4`, %f2, %f0, %f2 + camellia_f %f`16+16*$i+4`, %f6, %f4, %f6 + camellia_f %f`16+16*$i+6`, %f0, %f2, %f0 + camellia_f %f`16+16*$i+6`, %f4, %f6, %f4 + camellia_f %f`16+16*$i+8`, %f2, %f0, %f2 + camellia_f %f`16+16*$i+8`, %f6, %f4, %f6 + camellia_f %f`16+16*$i+10`, %f0, %f2, %f0 + camellia_f %f`16+16*$i+10`, %f4, %f6, %f4 + camellia_fl %f`16+16*$i+12`, %f0, %f0 + camellia_fl %f`16+16*$i+12`, %f4, %f4 + camellia_fli %f`16+16*$i+14`, %f2, %f2 + camellia_fli %f`16+16*$i+14`, %f6, %f6 +___ +} +$code.=<<___; + camellia_f %f16, %f2, %f0, %f2 + camellia_f %f16, %f6, %f4, %f6 + camellia_f %f18, %f0, %f2, %f0 + camellia_f %f18, %f4, %f6, %f4 + ldd [$key + 184], %f16 + ldd [$key + 176], %f18 + camellia_f %f20, %f2, %f0, %f2 + camellia_f %f20, %f6, %f4, %f6 + camellia_f %f22, %f0, %f2, %f0 + camellia_f %f22, %f4, %f6, %f4 + ldd [$key + 168], %f20 + ldd [$key + 160], %f22 + camellia_f %f24, %f2, %f0, %f8 + camellia_f %f24, %f6, %f4, %f10 + camellia_f %f26, %f0, %f8, %f2 + camellia_f %f26, %f4, %f10, %f6 + ldd [$key + 152], %f24 + ldd [$key + 144], %f26 + fxor %f30, %f8, %f0 + fxor %f30, %f10, %f4 + fxor %f28, %f2, %f2 + fxor %f28, %f6, %f6 + ldd [$key + 136], %f28 + retl + ldd [$key + 128], %f30 +.type _cmll256_decrypt_2x,#function +.size _cmll256_decrypt_2x,.-_cmll256_decrypt_2x +___ + +&alg_cbc_encrypt_implement("cmll",128); +&alg_cbc_encrypt_implement("cmll",256); + +&alg_cbc_decrypt_implement("cmll",128); +&alg_cbc_decrypt_implement("cmll",256); + +if ($::evp) { + &alg_ctr32_implement("cmll",128); + &alg_ctr32_implement("cmll",256); +} +}}} + +if (!$::evp) { +$code.=<<___; +.global Camellia_encrypt +Camellia_encrypt=cmll_t4_encrypt +.global Camellia_decrypt +Camellia_decrypt=cmll_t4_decrypt +.global Camellia_set_key +.align 32 +Camellia_set_key: + andcc %o2, 7, %g0 ! double-check alignment + bnz,a,pn %icc, 1f + mov -1, %o0 + brz,a,pn %o0, 1f + mov -1, %o0 + brz,a,pn %o2, 1f + mov -1, %o0 + andncc %o1, 0x1c0, %g0 + bnz,a,pn %icc, 1f + mov -2, %o0 + cmp %o1, 128 + bl,a,pn %icc, 1f + mov -2, %o0 + b cmll_t4_set_key + nop +1: retl + nop +.type Camellia_set_key,#function +.size Camellia_set_key,.-Camellia_set_key +___ + +my ($inp,$out,$len,$key,$ivec,$enc)=map("%o$_",(0..5)); + +$code.=<<___; +.globl Camellia_cbc_encrypt +.align 32 +Camellia_cbc_encrypt: + ld [$key + 272], %g1 + nop + brz $enc, .Lcbc_decrypt + cmp %g1, 3 + + be,pt %icc, cmll128_t4_cbc_encrypt + nop + ba cmll256_t4_cbc_encrypt + nop + +.Lcbc_decrypt: + be,pt %icc, cmll128_t4_cbc_decrypt + nop + ba cmll256_t4_cbc_decrypt + nop +.type Camellia_cbc_encrypt,#function +.size Camellia_cbc_encrypt,.-Camellia_cbc_encrypt +___ +} + +&emit_assembler(); + +close STDOUT; |