diff options
Diffstat (limited to 'crypto/aes/asm/aes-586.pl')
| -rwxr-xr-x | crypto/aes/asm/aes-586.pl | 3304 |
1 files changed, 2243 insertions, 1061 deletions
diff --git a/crypto/aes/asm/aes-586.pl b/crypto/aes/asm/aes-586.pl index ee02ded463..aab40e6f1c 100755 --- a/crypto/aes/asm/aes-586.pl +++ b/crypto/aes/asm/aes-586.pl @@ -2,11 +2,12 @@ # # ==================================================================== # Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL -# project. Rights for redistribution and usage in source and binary -# forms are granted according to the OpenSSL license. +# project. The module is, however, dual licensed under OpenSSL and +# CRYPTOGAMS licenses depending on where you obtain it. For further +# details see http://www.openssl.org/~appro/cryptogams/. # ==================================================================== # -# Version 3.1. +# Version 4.3. # # You might fail to appreciate this module performance from the first # try. If compared to "vanilla" linux-ia32-icc target, i.e. considered @@ -46,33 +47,152 @@ # Instruction Level Parallelism, and it indeed resulted in up to 15% # better performance on most recent µ-archs... # -# Current ECB performance numbers for 128-bit key in CPU cycles per -# processed byte [measure commonly used by AES benchmarkers] are: -# -# small footprint fully unrolled -# P4[-3] 23[24] 22[23] -# AMD K8 19 18 -# PIII 26 23 -# Pentium 63(*) 52 -# -# (*) Performance difference between small footprint code and fully -# unrolled in more commonly used CBC mode is not as big, 4% for -# for Pentium. PIII's ~13% difference [in both cases in 3rd -# version] is considered tolerable... -# # Third version adds AES_cbc_encrypt implementation, which resulted in # up to 40% performance imrovement of CBC benchmark results. 40% was # observed on P4 core, where "overall" imrovement coefficient, i.e. if # compared to PIC generated by GCC and in CBC mode, was observed to be # as large as 4x:-) CBC performance is virtually identical to ECB now -# and on some platforms even better, e.g. 56 "small" cycles/byte on -# senior Pentium, because certain function prologues and epilogues are +# and on some platforms even better, e.g. 17.6 "small" cycles/byte on +# Opteron, because certain function prologues and epilogues are # effectively taken out of the loop... - -push(@INC,"perlasm","../../perlasm"); +# +# Version 3.2 implements compressed tables and prefetch of these tables +# in CBC[!] mode. Former means that 3/4 of table references are now +# misaligned, which unfortunately has negative impact on elder IA-32 +# implementations, Pentium suffered 30% penalty, PIII - 10%. +# +# Version 3.3 avoids L1 cache aliasing between stack frame and +# S-boxes, and 3.4 - L1 cache aliasing even between key schedule. The +# latter is achieved by copying the key schedule to controlled place in +# stack. This unfortunately has rather strong impact on small block CBC +# performance, ~2x deterioration on 16-byte block if compared to 3.3. +# +# Version 3.5 checks if there is L1 cache aliasing between user-supplied +# key schedule and S-boxes and abstains from copying the former if +# there is no. This allows end-user to consciously retain small block +# performance by aligning key schedule in specific manner. +# +# Version 3.6 compresses Td4 to 256 bytes and prefetches it in ECB. +# +# Current ECB performance numbers for 128-bit key in CPU cycles per +# processed byte [measure commonly used by AES benchmarkers] are: +# +# small footprint fully unrolled +# P4 24 22 +# AMD K8 20 19 +# PIII 25 23 +# Pentium 81 78 +# +# Version 3.7 reimplements outer rounds as "compact." Meaning that +# first and last rounds reference compact 256 bytes S-box. This means +# that first round consumes a lot more CPU cycles and that encrypt +# and decrypt performance becomes asymmetric. Encrypt performance +# drops by 10-12%, while decrypt - by 20-25%:-( 256 bytes S-box is +# aggressively pre-fetched. +# +# Version 4.0 effectively rolls back to 3.6 and instead implements +# additional set of functions, _[x86|sse]_AES_[en|de]crypt_compact, +# which use exclusively 256 byte S-box. These functions are to be +# called in modes not concealing plain text, such as ECB, or when +# we're asked to process smaller amount of data [or unconditionally +# on hyper-threading CPU]. Currently it's called unconditionally from +# AES_[en|de]crypt, which affects all modes, but CBC. CBC routine +# still needs to be modified to switch between slower and faster +# mode when appropriate... But in either case benchmark landscape +# changes dramatically and below numbers are CPU cycles per processed +# byte for 128-bit key. +# +# ECB encrypt ECB decrypt CBC large chunk +# P4 56[60] 84[100] 23 +# AMD K8 48[44] 70[79] 18 +# PIII 41[50] 61[91] 24 +# Core 2 32[38] 45[70] 18.5 +# Pentium 120 160 77 +# +# Version 4.1 switches to compact S-box even in key schedule setup. +# +# Version 4.2 prefetches compact S-box in every SSE round or in other +# words every cache-line is *guaranteed* to be accessed within ~50 +# cycles window. Why just SSE? Because it's needed on hyper-threading +# CPU! Which is also why it's prefetched with 64 byte stride. Best +# part is that it has no negative effect on performance:-) +# +# Version 4.3 implements switch between compact and non-compact block +# functions in AES_cbc_encrypt depending on how much data was asked +# to be processed in one stroke. +# +###################################################################### +# Timing attacks are classified in two classes: synchronous when +# attacker consciously initiates cryptographic operation and collects +# timing data of various character afterwards, and asynchronous when +# malicious code is executed on same CPU simultaneously with AES, +# instruments itself and performs statistical analysis of this data. +# +# As far as synchronous attacks go the root to the AES timing +# vulnerability is twofold. Firstly, of 256 S-box elements at most 160 +# are referred to in single 128-bit block operation. Well, in C +# implementation with 4 distinct tables it's actually as little as 40 +# references per 256 elements table, but anyway... Secondly, even +# though S-box elements are clustered into smaller amount of cache- +# lines, smaller than 160 and even 40, it turned out that for certain +# plain-text pattern[s] or simply put chosen plain-text and given key +# few cache-lines remain unaccessed during block operation. Now, if +# attacker can figure out this access pattern, he can deduct the key +# [or at least part of it]. The natural way to mitigate this kind of +# attacks is to minimize the amount of cache-lines in S-box and/or +# prefetch them to ensure that every one is accessed for more uniform +# timing. But note that *if* plain-text was concealed in such way that +# input to block function is distributed *uniformly*, then attack +# wouldn't apply. Now note that some encryption modes, most notably +# CBC, do mask the plain-text in this exact way [secure cipher output +# is distributed uniformly]. Yes, one still might find input that +# would reveal the information about given key, but if amount of +# candidate inputs to be tried is larger than amount of possible key +# combinations then attack becomes infeasible. This is why revised +# AES_cbc_encrypt "dares" to switch to larger S-box when larger chunk +# of data is to be processed in one stroke. The current size limit of +# 512 bytes is chosen to provide same [diminishigly low] probability +# for cache-line to remain untouched in large chunk operation with +# large S-box as for single block operation with compact S-box and +# surely needs more careful consideration... +# +# As for asynchronous attacks. There are two flavours: attacker code +# being interleaved with AES on hyper-threading CPU at *instruction* +# level, and two processes time sharing single core. As for latter. +# Two vectors. 1. Given that attacker process has higher priority, +# yield execution to process performing AES just before timer fires +# off the scheduler, immediately regain control of CPU and analyze the +# cache state. For this attack to be efficient attacker would have to +# effectively slow down the operation by several *orders* of magnitute, +# by ratio of time slice to duration of handful of AES rounds, which +# unlikely to remain unnoticed. Not to mention that this also means +# that he would spend correspondigly more time to collect enough +# statistical data to mount the attack. It's probably appropriate to +# say that if adeversary reckons that this attack is beneficial and +# risks to be noticed, you probably have larger problems having him +# mere opportunity. In other words suggested code design expects you +# to preclude/mitigate this attack by overall system security design. +# 2. Attacker manages to make his code interrupt driven. In order for +# this kind of attack to be feasible, interrupt rate has to be high +# enough, again comparable to duration of handful of AES rounds. But +# is there interrupt source of such rate? Hardly, not even 1Gbps NIC +# generates interrupts at such raging rate... +# +# And now back to the former, hyper-threading CPU or more specifically +# Intel P4. Recall that asynchronous attack implies that malicious +# code instruments itself. And naturally instrumentation granularity +# has be noticeably lower than duration of codepath accessing S-box. +# Given that all cache-lines are accessed during that time that is. +# Current implementation accesses *all* cache-lines within ~50 cycles +# window, which is actually *less* than RDTSC latency on Intel P4! + +$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1; +push(@INC,"${dir}","${dir}../../perlasm"); require "x86asm.pl"; -&asm_init($ARGV[0],"aes-586.pl",$ARGV[$#ARGV] eq "386"); +&asm_init($ARGV[0],"aes-586.pl",$x86only = $ARGV[$#ARGV] eq "386"); +&static_label("AES_Te"); +&static_label("AES_Td"); $s0="eax"; $s1="ebx"; @@ -80,15 +200,36 @@ $s2="ecx"; $s3="edx"; $key="edi"; $acc="esi"; - +$tbl="ebp"; + +# stack frame layout in _[x86|sse]_AES_* 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 +$__s3=&DWP(16,"esp"); # s3 backing store +$__key=&DWP(20,"esp"); # pointer to key schedule +$__end=&DWP(24,"esp"); # pointer to end of key schedule +$__tbl=&DWP(28,"esp"); # %ebp backing store + +# stack frame layout in AES_[en|crypt] routines, which differs from +# above by 4 and overlaps by %ebp backing store +$_tbl=&DWP(24,"esp"); +$_esp=&DWP(28,"esp"); + +sub _data_word() { my $i; while(defined($i=shift)) { &data_word($i,$i); } } + +$speed_limit=512; # chunks smaller than $speed_limit are + # processed with compact routine in CBC mode $small_footprint=1; # $small_footprint=1 code is ~5% slower [on # recent µ-archs], but ~5 times smaller! # I favor compact code to minimize cache # contention and in hope to "collect" 5% back # in real-life applications... + $vertical_spin=0; # shift "verticaly" defaults to 0, because of # its proof-of-concept status... - # Note that there is no decvert(), as well as last encryption round is # performed with "horizontal" shifts. This is because this "vertical" # implementation [one which groups shifts on a given $s[i] to form a @@ -104,87 +245,554 @@ sub encvert() my $v0 = $acc, $v1 = $key; &mov ($v0,$s[3]); # copy s3 - &mov (&DWP(0,"esp"),$s[2]); # save s2 + &mov (&DWP(4,"esp"),$s[2]); # save s2 &mov ($v1,$s[0]); # copy s0 - &mov (&DWP(4,"esp"),$s[1]); # save s1 + &mov (&DWP(8,"esp"),$s[1]); # save s1 &movz ($s[2],&HB($s[0])); &and ($s[0],0xFF); - &mov ($s[0],&DWP(1024*0,$te,$s[0],4)); # s0>>0 + &mov ($s[0],&DWP(0,$te,$s[0],8)); # s0>>0 &shr ($v1,16); - &mov ($s[3],&DWP(1024*1,$te,$s[2],4)); # s0>>8 + &mov ($s[3],&DWP(3,$te,$s[2],8)); # s0>>8 &movz ($s[1],&HB($v1)); &and ($v1,0xFF); - &mov ($s[2],&DWP(1024*2,$te,$v1,4)); # s0>>16 + &mov ($s[2],&DWP(2,$te,$v1,8)); # s0>>16 &mov ($v1,$v0); - &mov ($s[1],&DWP(1024*3,$te,$s[1],4)); # s0>>24 + &mov ($s[1],&DWP(1,$te,$s[1],8)); # s0>>24 &and ($v0,0xFF); - &xor ($s[3],&DWP(1024*0,$te,$v0,4)); # s3>>0 + &xor ($s[3],&DWP(0,$te,$v0,8)); # s3>>0 &movz ($v0,&HB($v1)); &shr ($v1,16); - &xor ($s[2],&DWP(1024*1,$te,$v0,4)); # s3>>8 + &xor ($s[2],&DWP(3,$te,$v0,8)); # s3>>8 &movz ($v0,&HB($v1)); &and ($v1,0xFF); - &xor ($s[1],&DWP(1024*2,$te,$v1,4)); # s3>>16 - &mov ($v1,&DWP(0,"esp")); # restore s2 - &xor ($s[0],&DWP(1024*3,$te,$v0,4)); # s3>>24 + &xor ($s[1],&DWP(2,$te,$v1,8)); # s3>>16 + &mov ($v1,&DWP(4,"esp")); # restore s2 + &xor ($s[0],&DWP(1,$te,$v0,8)); # s3>>24 &mov ($v0,$v1); &and ($v1,0xFF); - &xor ($s[2],&DWP(1024*0,$te,$v1,4)); # s2>>0 + &xor ($s[2],&DWP(0,$te,$v1,8)); # s2>>0 &movz ($v1,&HB($v0)); &shr ($v0,16); - &xor ($s[1],&DWP(1024*1,$te,$v1,4)); # s2>>8 + &xor ($s[1],&DWP(3,$te,$v1,8)); # s2>>8 &movz ($v1,&HB($v0)); &and ($v0,0xFF); - &xor ($s[0],&DWP(1024*2,$te,$v0,4)); # s2>>16 - &mov ($v0,&DWP(4,"esp")); # restore s1 - &xor ($s[3],&DWP(1024*3,$te,$v1,4)); # s2>>24 + &xor ($s[0],&DWP(2,$te,$v0,8)); # s2>>16 + &mov ($v0,&DWP(8,"esp")); # restore s1 + &xor ($s[3],&DWP(1,$te,$v1,8)); # s2>>24 &mov ($v1,$v0); &and ($v0,0xFF); - &xor ($s[1],&DWP(1024*0,$te,$v0,4)); # s1>>0 + &xor ($s[1],&DWP(0,$te,$v0,8)); # s1>>0 &movz ($v0,&HB($v1)); &shr ($v1,16); - &xor ($s[0],&DWP(1024*1,$te,$v0,4)); # s1>>8 + &xor ($s[0],&DWP(3,$te,$v0,8)); # s1>>8 &movz ($v0,&HB($v1)); &and ($v1,0xFF); - &xor ($s[3],&DWP(1024*2,$te,$v1,4)); # s1>>16 - &mov ($key,&DWP(12,"esp")); # reincarnate v1 as key - &xor ($s[2],&DWP(1024*3,$te,$v0,4)); # s1>>24 + &xor ($s[3],&DWP(2,$te,$v1,8)); # s1>>16 + &mov ($key,$__key); # reincarnate v1 as key + &xor ($s[2],&DWP(1,$te,$v0,8)); # s1>>24 +} + +# Another experimental routine, which features "horizontal spin," but +# eliminates one reference to stack. Strangely enough runs slower... +sub enchoriz() +{ my $v0 = $key, $v1 = $acc; + + &movz ($v0,&LB($s0)); # 3, 2, 1, 0* + &rotr ($s2,8); # 8,11,10, 9 + &mov ($v1,&DWP(0,$te,$v0,8)); # 0 + &movz ($v0,&HB($s1)); # 7, 6, 5*, 4 + &rotr ($s3,16); # 13,12,15,14 + &xor ($v1,&DWP(3,$te,$v0,8)); # 5 + &movz ($v0,&HB($s2)); # 8,11,10*, 9 + &rotr ($s0,16); # 1, 0, 3, 2 + &xor ($v1,&DWP(2,$te,$v0,8)); # 10 + &movz ($v0,&HB($s3)); # 13,12,15*,14 + &xor ($v1,&DWP(1,$te,$v0,8)); # 15, t[0] collected + &mov ($__s0,$v1); # t[0] saved + + &movz ($v0,&LB($s1)); # 7, 6, 5, 4* + &shr ($s1,16); # -, -, 7, 6 + &mov ($v1,&DWP(0,$te,$v0,8)); # 4 + &movz ($v0,&LB($s3)); # 13,12,15,14* + &xor ($v1,&DWP(2,$te,$v0,8)); # 14 + &movz ($v0,&HB($s0)); # 1, 0, 3*, 2 + &and ($s3,0xffff0000); # 13,12, -, - + &xor ($v1,&DWP(1,$te,$v0,8)); # 3 + &movz ($v0,&LB($s2)); # 8,11,10, 9* + &or ($s3,$s1); # 13,12, 7, 6 + &xor ($v1,&DWP(3,$te,$v0,8)); # 9, t[1] collected + &mov ($s1,$v1); # s[1]=t[1] + + &movz ($v0,&LB($s0)); # 1, 0, 3, 2* + &shr ($s2,16); # -, -, 8,11 + &mov ($v1,&DWP(2,$te,$v0,8)); # 2 + &movz ($v0,&HB($s3)); # 13,12, 7*, 6 + &xor ($v1,&DWP(1,$te,$v0,8)); # 7 + &movz ($v0,&HB($s2)); # -, -, 8*,11 + &xor ($v1,&DWP(0,$te,$v0,8)); # 8 + &mov ($v0,$s3); + &shr ($v0,24); # 13 + &xor ($v1,&DWP(3,$te,$v0,8)); # 13, t[2] collected + + &movz ($v0,&LB($s2)); # -, -, 8,11* + &shr ($s0,24); # 1* + &mov ($s2,&DWP(1,$te,$v0,8)); # 11 + &xor ($s2,&DWP(3,$te,$s0,8)); # 1 + &mov ($s0,$__s0); # s[0]=t[0] + &movz ($v0,&LB($s3)); # 13,12, 7, 6* + &shr ($s3,16); # , ,13,12 + &xor ($s2,&DWP(2,$te,$v0,8)); # 6 + &mov ($key,$__key); # reincarnate v0 as key + &and ($s3,0xff); # , ,13,12* + &mov ($s3,&DWP(0,$te,$s3,8)); # 12 + &xor ($s3,$s2); # s[2]=t[3] collected + &mov ($s2,$v1); # s[2]=t[2] +} + +# More experimental code... SSE one... Even though this one eliminates +# *all* references to stack, it's not faster... +sub sse_encbody() +{ + &movz ($acc,&LB("eax")); # 0 + &mov ("ecx",&DWP(0,$tbl,$acc,8)); # 0 + &pshufw ("mm2","mm0",0x0d); # 7, 6, 3, 2 + &movz ("edx",&HB("eax")); # 1 + &mov ("edx",&DWP(3,$tbl,"edx",8)); # 1 + &shr ("eax",16); # 5, 4 + + &movz ($acc,&LB("ebx")); # 10 + &xor ("ecx",&DWP(2,$tbl,$acc,8)); # 10 + &pshufw ("mm6","mm4",0x08); # 13,12, 9, 8 + &movz ($acc,&HB("ebx")); # 11 + &xor ("edx",&DWP(1,$tbl,$acc,8)); # 11 + &shr ("ebx",16); # 15,14 + + &movz ($acc,&HB("eax")); # 5 + &xor ("ecx",&DWP(3,$tbl,$acc,8)); # 5 + &movq ("mm3",QWP(16,$key)); + &movz ($acc,&HB("ebx")); # 15 + &xor ("ecx",&DWP(1,$tbl,$acc,8)); # 15 + &movd ("mm0","ecx"); # t[0] collected + + &movz ($acc,&LB("eax")); # 4 + &mov ("ecx",&DWP(0,$tbl,$acc,8)); # 4 + &movd ("eax","mm2"); # 7, 6, 3, 2 + &movz ($acc,&LB("ebx")); # 14 + &xor ("ecx",&DWP(2,$tbl,$acc,8)); # 14 + &movd ("ebx","mm6"); # 13,12, 9, 8 + + &movz ($acc,&HB("eax")); # 3 + &xor ("ecx",&DWP(1,$tbl,$acc,8)); # 3 + &movz ($acc,&HB("ebx")); # 9 + &xor ("ecx",&DWP(3,$tbl,$acc,8)); # 9 + &movd ("mm1","ecx"); # t[1] collected + + &movz ($acc,&LB("eax")); # 2 + &mov ("ecx",&DWP(2,$tbl,$acc,8)); # 2 + &shr ("eax",16); # 7, 6 + &punpckldq ("mm0","mm1"); # t[0,1] collected + &movz ($acc,&LB("ebx")); # 8 + &xor ("ecx",&DWP(0,$tbl,$acc,8)); # 8 + &shr ("ebx",16); # 13,12 + + &movz ($acc,&HB("eax")); # 7 + &xor ("ecx",&DWP(1,$tbl,$acc,8)); # 7 + &pxor ("mm0","mm3"); + &movz ("eax",&LB("eax")); # 6 + &xor ("edx",&DWP(2,$tbl,"eax",8)); # 6 + &pshufw ("mm1","mm0",0x08); # 5, 4, 1, 0 + &movz ($acc,&HB("ebx")); # 13 + &xor ("ecx",&DWP(3,$tbl,$acc,8)); # 13 + &xor ("ecx",&DWP(24,$key)); # t[2] + &movd ("mm4","ecx"); # t[2] collected + &movz ("ebx",&LB("ebx")); # 12 + &xor ("edx",&DWP(0,$tbl,"ebx",8)); # 12 + &shr ("ecx",16); + &movd ("eax","mm1"); # 5, 4, 1, 0 + &mov ("ebx",&DWP(28,$key)); # t[3] + &xor ("ebx","edx"); + &movd ("mm5","ebx"); # t[3] collected + &and ("ebx",0xffff0000); + &or ("ebx","ecx"); + + &punpckldq ("mm4","mm5"); # t[2,3] collected +} + +###################################################################### +# "Compact" block function +###################################################################### + +sub enccompact() +{ my $Fn = mov; + while ($#_>5) { pop(@_); $Fn=sub{}; } + my ($i,$te,@s)=@_; + my $tmp = $key; + my $out = $i==3?$s[0]:$acc; + + # $Fn is used in first compact round and its purpose is to + # void restoration of some values from stack, so that after + # 4xenccompact with extra argument $key value is left there... + if ($i==3) { &$Fn ($key,$__key); }##%edx + else { &mov ($out,$s[0]); } + &and ($out,0xFF); + if ($i==1) { &shr ($s[0],16); }#%ebx[1] + if ($i==2) { &shr ($s[0],24); }#%ecx[2] + &movz ($out,&BP(-128,$te,$out,1)); + + if ($i==3) { $tmp=$s[1]; }##%eax + &movz ($tmp,&HB($s[1])); + &movz ($tmp,&BP(-128,$te,$tmp,1)); + &shl ($tmp,8); + &xor ($out,$tmp); + + if ($i==3) { $tmp=$s[2]; &mov ($s[1],$__s0); }##%ebx + else { &mov ($tmp,$s[2]); + &shr ($tmp,16); } + if ($i==2) { &and ($s[1],0xFF); }#%edx[2] + &and ($tmp,0xFF); + &movz ($tmp,&BP(-128,$te,$tmp,1)); + &shl ($tmp,16); + &xor ($out,$tmp); + + if ($i==3) { $tmp=$s[3]; &mov ($s[2],$__s1); }##%ecx + elsif($i==2){ &movz ($tmp,&HB($s[3])); }#%ebx[2] + else { &mov ($tmp,$s[3]); + &shr ($tmp,24); } + &movz ($tmp,&BP(-128,$te,$tmp,1)); + &shl ($tmp,24); + &xor ($out,$tmp); + if ($i<2) { &mov (&DWP(4+4*$i,"esp"),$out); } + if ($i==3) { &mov ($s[3],$acc); } + &comment(); +} + +sub enctransform() +{ my @s = ($s0,$s1,$s2,$s3); + my $i = shift; + my $tmp = $tbl; + my $r2 = $key ; + + &mov ($acc,$s[$i]); + &and ($acc,0x80808080); + &mov ($tmp,$acc); + &shr ($tmp,7); + &lea ($r2,&DWP(0,$s[$i],$s[$i])); + &sub ($acc,$tmp); + &and ($r2,0xfefefefe); + &and ($acc,0x1b1b1b1b); + &mov ($tmp,$s[$i]); + &xor ($acc,$r2); # r2 + + &xor ($s[$i],$acc); # r0 ^ r2 + &rotl ($s[$i],24); + &xor ($s[$i],$acc) # ROTATE(r2^r0,24) ^ r2 + &rotr ($tmp,16); + &xor ($s[$i],$tmp); + &rotr ($tmp,8); + &xor ($s[$i],$tmp); +} + +&function_begin_B("_x86_AES_encrypt_compact"); + # note that caller is expected to allocate stack frame for me! + &mov ($__key,$key); # save key + + &xor ($s0,&DWP(0,$key)); # xor with key + &xor ($s1,&DWP(4,$key)); + &xor ($s2,&DWP(8,$key)); + &xor ($s3,&DWP(12,$key)); + + &mov ($acc,&DWP(240,$key)); # load key->rounds + &lea ($acc,&DWP(-2,$acc,$acc)); + &lea ($acc,&DWP(0,$key,$acc,8)); + &mov ($__end,$acc); # end of key schedule + + # prefetch Te4 + &mov ($key,&DWP(0-128,$tbl)); + &mov ($acc,&DWP(32-128,$tbl)); + &mov ($key,&DWP(64-128,$tbl)); + &mov ($acc,&DWP(96-128,$tbl)); + &mov ($key,&DWP(128-128,$tbl)); + &mov ($acc,&DWP(160-128,$tbl)); + &mov ($key,&DWP(192-128,$tbl)); + &mov ($acc,&DWP(224-128,$tbl)); + + &set_label("loop",16); + + &enccompact(0,$tbl,$s0,$s1,$s2,$s3,1); + &enccompact(1,$tbl,$s1,$s2,$s3,$s0,1); + &enccompact(2,$tbl,$s2,$s3,$s0,$s1,1); + &enccompact(3,$tbl,$s3,$s0,$s1,$s2,1); + &enctransform(2); + &enctransform(3); + &enctransform(0); + &enctransform(1); + &mov ($key,$__key); + &mov ($tbl,$__tbl); + &add ($key,16); # advance rd_key + &xor ($s0,&DWP(0,$key)); + &xor ($s1,&DWP(4,$key)); + &xor ($s2,&DWP(8,$key)); + &xor ($s3,&DWP(12,$key)); + + &cmp ($key,$__end); + &mov ($__key,$key); + &jb (&label("loop")); + + &enccompact(0,$tbl,$s0,$s1,$s2,$s3); + &enccompact(1,$tbl,$s1,$s2,$s3,$s0); + &enccompact(2,$tbl,$s2,$s3,$s0,$s1); + &enccompact(3,$tbl,$s3,$s0,$s1,$s2); + + &xor ($s0,&DWP(16,$key)); + &xor ($s1,&DWP(20,$key)); + &xor ($s2,&DWP(24,$key)); + &xor ($s3,&DWP(28,$key)); + + &ret (); +&function_end_B("_x86_AES_encrypt_compact"); + +###################################################################### +# "Compact" SSE block function. +###################################################################### +# +# Performance is not actually extraordinary in comparison to pure +# x86 code. In particular encrypt performance is virtually the same. +# Decrypt performance on the other hand is 15-20% better on newer +# µ-archs [but we're thankful for *any* improvement here], and ~50% +# better on PIII:-) And additionally on the pros side this code +# eliminates redundant references to stack and thus relieves/ +# minimizes the pressure on the memory bus. +# +# MMX register layout lsb +# +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ +# | mm4 | mm0 | +# +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ +# | s3 | s2 | s1 | s0 | +# +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ +# |15|14|13|12|11|10| 9| 8| 7| 6| 5| 4| 3| 2| 1| 0| +# +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ +# +# Indexes translate as s[N/4]>>(8*(N%4)), e.g. 5 means s1>>8. +# In this terms encryption and decryption "compact" permutation +# matrices can be depicted as following: +# +# encryption lsb # decryption lsb +# +----++----+----+----+----+ # +----++----+----+----+----+ +# | t0 || 15 | 10 | 5 | 0 | # | t0 || 7 | 10 | 13 | 0 | +# +----++----+----+----+----+ # +----++----+----+----+----+ +# | t1 || 3 | 14 | 9 | 4 | # | t1 || 11 | 14 | 1 | 4 | +# +----++----+----+----+----+ # +----++----+----+----+----+ +# | t2 || 7 | 2 | 13 | 8 | # | t2 || 15 | 2 | 5 | 8 | +# +----++----+----+----+----+ # +----++----+----+----+----+ +# | t3 || 11 | 6 | 1 | 12 | # | t3 || 3 | 6 | 9 | 12 | +# +----++----+----+----+----+ # +----++----+----+----+----+ +# +###################################################################### +# Why not xmm registers? Short answer. It was actually tested and +# was not any faster, but *contrary*, most notably on Intel CPUs. +# Longer answer. Main advantage of using mm registers is that movd +# latency is lower, especially on Intel P4. While arithmetic +# instructions are twice as many, they can be scheduled every cycle +# and not every second one when they are operating on xmm register, +# so that "arithmetic throughput" remains virtually the same. And +# finally the code can be executed even on elder SSE-only CPUs:-) + +sub sse_enccompact() +{ + &pshufw ("mm1","mm0",0x08); # 5, 4, 1, 0 + &pshufw ("mm5","mm4",0x0d); # 15,14,11,10 + &movd ("eax","mm1"); # 5, 4, 1, 0 + &movd ("ebx","mm5"); # 15,14,11,10 + + &movz ($acc,&LB("eax")); # 0 + &movz ("ecx",&BP(-128,$tbl,$acc,1)); # 0 + &pshufw ("mm2","mm0",0x0d); # 7, 6, 3, 2 + &movz ("edx",&HB("eax")); # 1 + &movz ("edx",&BP(-128,$tbl,"edx",1)); # 1 + &shl ("edx",8); # 1 + &shr ("eax",16); # 5, 4 + + &movz ($acc,&LB("ebx")); # 10 + &movz ($acc,&BP(-128,$tbl,$acc,1)); # 10 + &shl ($acc,16); # 10 + &or ("ecx",$acc); # 10 + &pshufw ("mm6","mm4",0x08); # 13,12, 9, 8 + &movz ($acc,&HB("ebx")); # 11 + &movz ($acc,&BP(-128,$tbl,$acc,1)); # 11 + &shl ($acc,24); # 11 + &or ("edx",$acc); # 11 + &shr ("ebx",16); # 15,14 + + &movz ($acc,&HB("eax")); # 5 + &movz ($acc,&BP(-128,$tbl,$acc,1)); # 5 + &shl ($acc,8); # 5 + &or ("ecx",$acc); # 5 + &movz ($acc,&HB("ebx")); # 15 + &movz ($acc,&BP(-128,$tbl,$acc,1)); # 15 + &shl ($acc,24); # 15 + &or ("ecx",$acc); # 15 + &movd ("mm0","ecx"); # t[0] collected + + &movz ($acc,&LB("eax")); # 4 + &movz ("ecx",&BP(-128,$tbl,$acc,1)); # 4 + &movd ("eax","mm2"); # 7, 6, 3, 2 + &movz ($acc,&LB("ebx")); # 14 + &movz ($acc,&BP(-128,$tbl,$acc,1)); # 14 + &shl ($acc,16); # 14 + &or ("ecx",$acc); # 14 + + &movd ("ebx","mm6"); # 13,12, 9, 8 + &movz ($acc,&HB("eax")); # 3 + &movz ($acc,&BP(-128,$tbl,$acc,1)); # 3 + &shl ($acc,24); # 3 + &or ("ecx",$acc); # 3 + &movz ($acc,&HB("ebx")); # 9 + &movz ($acc,&BP(-128,$tbl,$acc,1)); # 9 + &shl ($acc,8); # 9 + &or ("ecx",$acc); # 9 + &movd ("mm1","ecx"); # t[1] collected + + &movz ($acc,&LB("ebx")); # 8 + &movz ("ecx",&BP(-128,$tbl,$acc,1)); # 8 + &shr ("ebx",16); # 13,12 + &movz ($acc,&LB("eax")); # 2 + &movz ($acc,&BP(-128,$tbl,$acc,1)); # 2 + &shl ($acc,16); # 2 + &or ("ecx",$acc); # 2 + &shr ("eax",16); # 7, 6 + + &punpckldq ("mm0","mm1"); # t[0,1] collected + + &movz ($acc,&HB("eax")); # 7 + &movz ($acc,&BP(-128,$tbl,$acc,1)); # 7 + &shl ($acc,24); # 7 + &or ("ecx",$acc); # 7 + &and ("eax",0xff); # 6 + &movz ("eax",&BP(-128,$tbl,"eax",1)); # 6 + &shl ("eax",16); # 6 + &or ("edx","eax"); # 6 + &movz ($acc,&HB("ebx")); # 13 + &movz ($acc,&BP(-128,$tbl,$acc,1)); # 13 + &shl ($acc,8); # 13 + &or ("ecx",$acc); # 13 + &movd ("mm4","ecx"); # t[2] collected + &and ("ebx",0xff); # 12 + &movz ("ebx",&BP(-128,$tbl,"ebx",1)); # 12 + &or ("edx","ebx"); # 12 + &movd ("mm5","edx"); # t[3] collected + + &punpckldq ("mm4","mm5"); # t[2,3] collected } + if (!$x86only) { +&function_begin_B("_sse_AES_encrypt_compact"); + &pxor ("mm0",&QWP(0,$key)); # 7, 6, 5, 4, 3, 2, 1, 0 + &pxor ("mm4",&QWP(8,$key)); # 15,14,13,12,11,10, 9, 8 + + # note that caller is expected to allocate stack frame for me! + &mov ($acc,&DWP(240,$key)); # load key->rounds + &lea ($acc,&DWP(-2,$acc,$acc)); + &lea ($acc,&DWP(0,$key,$acc,8)); + &mov ($__end,$acc); # end of key schedule + + &mov ($s0,0x1b1b1b1b); # magic constant + &mov (&DWP(8,"esp"),$s0); + &mov (&DWP(12,"esp"),$s0); + + # prefetch Te4 + &mov ($s0,&DWP(0-128,$tbl)); + &mov ($s1,&DWP(32-128,$tbl)); + &mov ($s2,&DWP(64-128,$tbl)); + &mov ($s3,&DWP(96-128,$tbl)); + &mov ($s0,&DWP(128-128,$tbl)); + &mov ($s1,&DWP(160-128,$tbl)); + &mov ($s2,&DWP(192-128,$tbl)); + &mov ($s3,&DWP(224-128,$tbl)); + + &set_label("loop",16); + &sse_enccompact(); + &add ($key,16); + &cmp ($key,$__end); + &ja (&label("out")); + + &movq ("mm2",&QWP(8,"esp")); + &pxor ("mm3","mm3"); &pxor ("mm7","mm7"); + &movq ("mm1","mm0"); &movq ("mm5","mm4"); # r0 + &pcmpgtb("mm3","mm0"); &pcmpgtb("mm7","mm4"); + &pand ("mm3","mm2"); &pand ("mm7","mm2"); + &pshufw ("mm2","mm0",0xb1); &pshufw ("mm6","mm4",0xb1);# ROTATE(r0,16) + &paddb ("mm0","mm0"); &paddb ("mm4","mm4"); + &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # = r2 + &pshufw ("mm3","mm2",0xb1); &pshufw ("mm7","mm6",0xb1);# r0 + &pxor ("mm1","mm0"); &pxor ("mm5","mm4"); # r0^r2 + &pxor ("mm0","mm2"); &pxor ("mm4","mm6"); # ^= ROTATE(r0,16) + + &movq ("mm2","mm3"); &movq ("mm6","mm7"); + &pslld ("mm3",8); &pslld ("mm7",8); + &psrld ("mm2",24); &psrld ("mm6",24); + &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= r0<<8 + &pxor ("mm0","mm2"); &pxor ("mm4","mm6"); # ^= r0>>24 + + &movq ("mm3","mm1"); &movq ("mm7","mm5"); + &movq ("mm2",&QWP(0,$key)); &movq ("mm6",&QWP(8,$key)); + &psrld ("mm1",8); &psrld ("mm5",8); + &mov ($s0,&DWP(0-128,$tbl)); + &pslld ("mm3",24); &pslld ("mm7",24); + &mov ($s1,&DWP(64-128,$tbl)); + &pxor ("mm0","mm1"); &pxor ("mm4","mm5"); # ^= (r2^r0)<<8 + &mov ($s2,&DWP(128-128,$tbl)); + &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= (r2^r0)>>24 + &mov ($s3,&DWP(192-128,$tbl)); + + &pxor ("mm0","mm2"); &pxor ("mm4","mm6"); + &jmp (&label("loop")); + + &set_label("out",16); + &pxor ("mm0",&QWP(0,$key)); + &pxor ("mm4",&QWP(8,$key)); + + &ret (); +&function_end_B("_sse_AES_encrypt_compact"); + } + +###################################################################### +# Vanilla block function. +###################################################################### + sub encstep() { my ($i,$te,@s) = @_; my $tmp = $key; my $out = $i==3?$s[0]:$acc; # lines marked with #%e?x[i] denote "reordered" instructions... - if ($i==3) { &mov ($key,&DWP(12,"esp")); }##%edx + if ($i==3) { &mov ($key,$__key); }##%edx else { &mov ($out,$s[0]); &and ($out,0xFF); } if ($i==1) { &shr ($s[0],16); }#%ebx[1] if ($i==2) { &shr ($s[0],24); }#%ecx[2] - &mov ($out,&DWP(1024*0,$te,$out,4)); + &mov ($out,&DWP(0,$te,$out,8)); if ($i==3) { $tmp=$s[1]; }##%eax &movz ($tmp,&HB($s[1])); - &xor ($out,&DWP(1024*1,$te,$tmp,4)); + &xor ($out,&DWP(3,$te,$tmp,8)); - if ($i==3) { $tmp=$s[2]; &mov ($s[1],&DWP(0,"esp")); }##%ebx + if ($i==3) { $tmp=$s[2]; &mov ($s[1],$__s0); }##%ebx else { &mov ($tmp,$s[2]); &shr ($tmp,16); } if ($i==2) { &and ($s[1],0xFF); }#%edx[2] &and ($tmp,0xFF); - &xor ($out,&DWP(1024*2,$te,$tmp,4)); + &xor ($out,&DWP(2,$te,$tmp,8)); - if ($i==3) { $tmp=$s[3]; &mov ($s[2],&DWP(4,"esp")); }##%ecx + if ($i==3) { $tmp=$s[3]; &mov ($s[2],$__s1); }##%ecx elsif($i==2){ &movz ($tmp,&HB($s[3])); }#%ebx[2] else { &mov ($tmp,$s[3]); &shr ($tmp,24) } - &xor ($out,&DWP(1024*3,$te,$tmp,4)); - if ($i<2) { &mov (&DWP(4*$i,"esp"),$out); } + &xor ($out,&DWP(1,$te,$tmp,8)); + if ($i<2) { &mov (&DWP(4+4*$i,"esp"),$out); } if ($i==3) { &mov ($s[3],$acc); } &comment(); } @@ -194,41 +802,40 @@ sub enclast() my $tmp = $key; my $out = $i==3?$s[0]:$acc; - if ($i==3) { &mov ($key,&DWP(12,"esp")); }##%edx + if ($i==3) { &mov ($key,$__key); }##%edx else { &mov ($out,$s[0]); } &and ($out,0xFF); if ($i==1) { &shr ($s[0],16); }#%ebx[1] if ($i==2) { &shr ($s[0],24); }#%ecx[2] - &mov ($out,&DWP(1024*4,$te,$out,4)); + &mov ($out,&DWP(2,$te,$out,8)); &and ($out,0x000000ff); if ($i==3) { $tmp=$s[1]; }##%eax &movz ($tmp,&HB($s[1])); - &mov ($tmp,&DWP(1024*4,$te,$tmp,4)); + &mov ($tmp,&DWP(0,$te,$tmp,8)); &and ($tmp,0x0000ff00); &xor ($out,$tmp); - if ($i==3) { $tmp=$s[2]; &mov ($s[1],&DWP(0,"esp")); }##%ebx - else { mov ($tmp,$s[2]); + if ($i==3) { $tmp=$s[2]; &mov ($s[1],$__s0); }##%ebx + else { &mov ($tmp,$s[2]); &shr ($tmp,16); } if ($i==2) { &and ($s[1],0xFF); }#%edx[2] &and ($tmp,0xFF); - &mov ($tmp,&DWP(1024*4,$te,$tmp,4)); + &mov ($tmp,&DWP(0,$te,$tmp,8)); &and ($tmp,0x00ff0000); &xor ($out,$tmp); - if ($i==3) { $tmp=$s[3]; &mov ($s[2],&DWP(4,"esp")); }##%ecx + if ($i==3) { $tmp=$s[3]; &mov ($s[2],$__s1); }##%ecx elsif($i==2){ &movz ($tmp,&HB($s[3])); }#%ebx[2] else { &mov ($tmp,$s[3]); &shr ($tmp,24); } - &mov ($tmp,&DWP(1024*4,$te,$tmp,4)); + &mov ($tmp,&DWP(2,$te,$tmp,8)); &and ($tmp,0xff000000); &xor ($out,$tmp); - if ($i<2) { &mov (&DWP(4*$i,"esp"),$out); } + if ($i<2) { &mov (&DWP(4+4*$i,"esp"),$out); } if ($i==3) { &mov ($s[3],$acc); } } -&public_label("AES_Te"); &function_begin_B("_x86_AES_encrypt"); if ($vertical_spin) { # I need high parts of volatile registers to be accessible... @@ -236,13 +843,8 @@ sub enclast() &mov ($s2="esi",$acc="ecx"); } - # allocate aligned stack frame - &mov ($acc,"esp"); - &sub ("esp",20); - &and ("esp",-16); - - &mov (&DWP(12,"esp"),$key); # save key - &mov (&DWP(16,"esp"),$acc); # save %esp + # note that caller is expected to allocate stack frame for me! + &mov ($__key,$key); # save key &xor ($s0,&DWP(0,$key)); # xor with key &xor ($s1,&DWP(4,$key)); @@ -254,24 +856,24 @@ sub enclast() if ($small_footprint) { &lea ($acc,&DWP(-2,$acc,$acc)); &lea ($acc,&DWP(0,$key,$acc,8)); - &mov (&DWP(8,"esp"),$acc); # end of key schedule - &align (4); - &set_label("loop"); + &mov ($__end,$acc); # end of key schedule + + &set_label("loop",16); if ($vertical_spin) { - &encvert("ebp",$s0,$s1,$s2,$s3); + &encvert($tbl,$s0,$s1,$s2,$s3); } else { - &encstep(0,"ebp",$s0,$s1,$s2,$s3); - &encstep(1,"ebp",$s1,$s2,$s3,$s0); - &encstep(2,"ebp",$s2,$s3,$s0,$s1); - &encstep(3,"ebp",$s3,$s0,$s1,$s2); + &encstep(0,$tbl,$s0,$s1,$s2,$s3); + &encstep(1,$tbl,$s1,$s2,$s3,$s0); + &encstep(2,$tbl,$s2,$s3,$s0,$s1); + &encstep(3,$tbl,$s3,$s0,$s1,$s2); } - &add ($key,16); # advance rd_key + &add ($key,16); # advance rd_key &xor ($s0,&DWP(0,$key)); &xor ($s1,&DWP(4,$key)); &xor ($s2,&DWP(8,$key)); &xor ($s3,&DWP(12,$key)); - &cmp ($key,&DWP(8,"esp")); - &mov (&DWP(12,"esp"),$key); + &cmp ($key,$__end); + &mov ($__key,$key); &jb (&label("loop")); } else { @@ -280,15 +882,15 @@ sub enclast() &cmp ($acc,12); &jle (&label("12rounds")); - &set_label("14rounds"); + &set_label("14rounds",4); for ($i=1;$i<3;$i++) { if ($vertical_spin) { - &encvert("ebp",$s0,$s1,$s2,$s3); + &encvert($tbl,$s0,$s1,$s2,$s3); } else { - &encstep(0,"ebp",$s0,$s1,$s2,$s3); - &encstep(1,"ebp",$s1,$s2,$s3,$s0); - &encstep(2,"ebp",$s2,$s3,$s0,$s1); - &encstep(3,"ebp",$s3,$s0,$s1,$s2); + &encstep(0,$tbl,$s0,$s1,$s2,$s3); + &encstep(1,$tbl,$s1,$s2,$s3,$s0); + &encstep(2,$tbl,$s2,$s3,$s0,$s1); + &encstep(3,$tbl,$s3,$s0,$s1,$s2); } &xor ($s0,&DWP(16*$i+0,$key)); &xor ($s1,&DWP(16*$i+4,$key)); @@ -296,16 +898,16 @@ sub enclast() &xor ($s3,&DWP(16*$i+12,$key)); } &add ($key,32); - &mov (&DWP(12,"esp"),$key); # advance rd_key - &set_label("12rounds"); + &mov ($__key,$key); # advance rd_key + &set_label("12rounds",4); for ($i=1;$i<3;$i++) { if ($vertical_spin) { - &encvert("ebp",$s0,$s1,$s2,$s3); + &encvert($tbl,$s0,$s1,$s2,$s3); } else { - &encstep(0,"ebp",$s0,$s1,$s2,$s3); - &encstep(1,"ebp",$s1,$s2,$s3,$s0); - &encstep(2,"ebp",$s2,$s3,$s0,$s1); - &encstep(3,"ebp",$s3,$s0,$s1,$s2); + &encstep(0,$tbl,$s0,$s1,$s2,$s3); + &encstep(1,$tbl,$s1,$s2,$s3,$s0); + &encstep(2,$tbl,$s2,$s3,$s0,$s1); + &encstep(3,$tbl,$s3,$s0,$s1,$s2); } &xor ($s0,&DWP(16*$i+0,$key)); &xor ($s1,&DWP(16*$i+4,$key)); @@ -313,16 +915,16 @@ sub enclast() &xor ($s3,&DWP(16*$i+12,$key)); } &add ($key,32); - &mov (&DWP(12,"esp"),$key); # advance rd_key - &set_label("10rounds"); + &mov ($__key,$key); # advance rd_key + &set_label("10rounds",4); for ($i=1;$i<10;$i++) { if ($vertical_spin) { - &encvert("ebp",$s0,$s1,$s2,$s3); + &encvert($tbl,$s0,$s1,$s2,$s3); } else { - &encstep(0,"ebp",$s0,$s1,$s2,$s3); - &encstep(1,"ebp",$s1,$s2,$s3,$s0); - &encstep(2,"ebp",$s2,$s3,$s0,$s1); - &encstep(3,"ebp",$s3,$s0,$s1,$s2); + &encstep(0,$tbl,$s0,$s1,$s2,$s3); + &encstep(1,$tbl,$s1,$s2,$s3,$s0); + &encstep(2,$tbl,$s2,$s3,$s0,$s1); + &encstep(3,$tbl,$s3,$s0,$s1,$s2); } &xor ($s0,&DWP(16*$i+0,$key)); &xor ($s1,&DWP(16*$i+4,$key)); @@ -336,12 +938,11 @@ sub enclast() &mov ($s1="ebx",$key="edi"); &mov ($s2="ecx",$acc="esi"); } - &enclast(0,"ebp",$s0,$s1,$s2,$s3); - &enclast(1,"ebp",$s1,$s2,$s3,$s0); - &enclast(2,"ebp",$s2,$s3,$s0,$s1); - &enclast(3,"ebp",$s3,$s0,$s1,$s2); + &enclast(0,$tbl,$s0,$s1,$s2,$s3); + &enclast(1,$tbl,$s1,$s2,$s3,$s0); + &enclast(2,$tbl,$s2,$s3,$s0,$s1); + &enclast(3,$tbl,$s3,$s0,$s1,$s2); - &mov ("esp",&DWP(16,"esp")); # restore %esp &add ($key,$small_footprint?16:160); &xor ($s0,&DWP(0,$key)); &xor ($s1,&DWP(4,$key)); @@ -351,354 +952,262 @@ sub enclast() &ret (); &set_label("AES_Te",64); # Yes! I keep it in the code segment! - &data_word(0xa56363c6, 0x847c7cf8, 0x997777ee, 0x8d7b7bf6); - &data_word(0x0df2f2ff, 0xbd6b6bd6, 0xb16f6fde, 0x54c5c591); - &data_word(0x50303060, 0x03010102, 0xa96767ce, 0x7d2b2b56); - &data_word(0x19fefee7, 0x62d7d7b5, 0xe6abab4d, 0x9a7676ec); - &data_word(0x45caca8f, 0x9d82821f, 0x40c9c989, 0x877d7dfa); - &data_word(0x15fafaef, 0xeb5959b2, 0xc947478e, 0x0bf0f0fb); - &data_word(0xecadad41, 0x67d4d4b3, 0xfda2a25f, 0xeaafaf45); - &data_word(0xbf9c9c23, 0xf7a4a453, 0x967272e4, 0x5bc0c09b); - &data_word(0xc2b7b775, 0x1cfdfde1, 0xae93933d, 0x6a26264c); - &data_word(0x5a36366c, 0x413f3f7e, 0x02f7f7f5, 0x4fcccc83); - &data_word(0x5c343468, 0xf4a5a551, 0x34e5e5d1, 0x08f1f1f9); - &data_word(0x937171e2, 0x73d8d8ab, 0x53313162, 0x3f15152a); - &data_word(0x0c040408, 0x52c7c795, 0x65232346, 0x5ec3c39d); - &data_word(0x28181830, 0xa1969637, 0x0f05050a, 0xb59a9a2f); - &data_word(0x0907070e, 0x36121224, 0x9b80801b, 0x3de2e2df); - &data_word(0x26ebebcd, 0x6927274e, 0xcdb2b27f, 0x9f7575ea); - &data_word(0x1b090912, 0x9e83831d, 0x742c2c58, 0x2e1a1a34); - &data_word(0x2d1b1b36, 0xb26e6edc, 0xee5a5ab4, 0xfba0a05b); - &data_word(0xf65252a4, 0x4d3b3b76, 0x61d6d6b7, 0xceb3b37d); - &data_word(0x7b292952, 0x3ee3e3dd, 0x712f2f5e, 0x97848413); - &data_word(0xf55353a6, 0x68d1d1b9, 0x00000000, 0x2cededc1); - &data_word(0x60202040, 0x1ffcfce3, 0xc8b1b179, 0xed5b5bb6); - &data_word(0xbe6a6ad4, 0x46cbcb8d, 0xd9bebe67, 0x4b393972); - &data_word(0xde4a4a94, 0xd44c4c98, 0xe85858b0, 0x4acfcf85); - &data_word(0x6bd0d0bb, 0x2aefefc5, 0xe5aaaa4f, 0x16fbfbed); - &data_word(0xc5434386, 0xd74d4d9a, 0x55333366, 0x94858511); - &data_word(0xcf45458a, 0x10f9f9e9, 0x06020204, 0x817f7ffe); - &data_word(0xf05050a0, 0x443c3c78, 0xba9f9f25, 0xe3a8a84b); - &data_word(0xf35151a2, 0xfea3a35d, 0xc0404080, 0x8a8f8f05); - &data_word(0xad92923f, 0xbc9d9d21, 0x48383870, 0x04f5f5f1); - &data_word(0xdfbcbc63, 0xc1b6b677, 0x75dadaaf, 0x63212142); - &data_word(0x30101020, 0x1affffe5, 0x0ef3f3fd, 0x6dd2d2bf); - &data_word(0x4ccdcd81, 0x140c0c18, 0x35131326, 0x2fececc3); - &data_word(0xe15f5fbe, 0xa2979735, 0xcc444488, 0x3917172e); - &data_word(0x57c4c493, 0xf2a7a755, 0x827e7efc, 0x473d3d7a); - &data_word(0xac6464c8, 0xe75d5dba, 0x2b191932, 0x957373e6); - &data_word(0xa06060c0, 0x98818119, 0xd14f4f9e, 0x7fdcdca3); - &data_word(0x66222244, 0x7e2a2a54, 0xab90903b, 0x8388880b); - &data_word(0xca46468c, 0x29eeeec7, 0xd3b8b86b, 0x3c141428); - &data_word(0x79dedea7, 0xe25e5ebc, 0x1d0b0b16, 0x76dbdbad); - &data_word(0x3be0e0db, 0x56323264, 0x4e3a3a74, 0x1e0a0a14); - &data_word(0xdb494992, 0x0a06060c, 0x6c242448, 0xe45c5cb8); - &data_word(0x5dc2c29f, 0x6ed3d3bd, 0xefacac43, 0xa66262c4); - &data_word(0xa8919139, 0xa4959531, 0x37e4e4d3, 0x8b7979f2); - &data_word(0x32e7e7d5, 0x43c8c88b, 0x5937376e, 0xb76d6dda); - &data_word(0x8c8d8d01, 0x64d5d5b1, 0xd24e4e9c, 0xe0a9a949); - &data_word(0xb46c6cd8, 0xfa5656ac, 0x07f4f4f3, 0x25eaeacf); - &data_word(0xaf6565ca, 0x8e7a7af4, 0xe9aeae47, 0x18080810); - &data_word(0xd5baba6f, 0x887878f0, 0x6f25254a, 0x722e2e5c); - &data_word(0x241c1c38, 0xf1a6a657, 0xc7b4b473, 0x51c6c697); - &data_word(0x23e8e8cb, 0x7cdddda1, 0x9c7474e8, 0x211f1f3e); - &data_word(0xdd4b4b96, 0xdcbdbd61, 0x868b8b0d, 0x858a8a0f); - &data_word(0x907070e0, 0x423e3e7c, 0xc4b5b571, 0xaa6666cc); - &data_word(0xd8484890, 0x05030306, 0x01f6f6f7, 0x120e0e1c); - &data_word(0xa36161c2, 0x5f35356a, 0xf95757ae, 0xd0b9b969); - &data_word(0x91868617, 0x58c1c199, 0x271d1d3a, 0xb99e9e27); - &data_word(0x38e1e1d9, 0x13f8f8eb, 0xb398982b, 0x33111122); - &data_word(0xbb6969d2, 0x70d9d9a9, 0x898e8e07, 0xa7949433); - &data_word(0xb69b9b2d, 0x221e1e3c, 0x92878715, 0x20e9e9c9); - &data_word(0x49cece87, 0xff5555aa, 0x78282850, 0x7adfdfa5); - &data_word(0x8f8c8c03, 0xf8a1a159, 0x80898909, 0x170d0d1a); - &data_word(0xdabfbf65, 0x31e6e6d7, 0xc6424284, 0xb86868d0); - &data_word(0xc3414182, 0xb0999929, 0x772d2d5a, 0x110f0f1e); - &data_word(0xcbb0b07b, 0xfc5454a8, 0xd6bbbb6d, 0x3a16162c); -#Te1: - &data_word(0x6363c6a5, 0x7c7cf884, 0x7777ee99, 0x7b7bf68d); - &data_word(0xf2f2ff0d, 0x6b6bd6bd, 0x6f6fdeb1, 0xc5c59154); - &data_word(0x30306050, 0x01010203, 0x6767cea9, 0x2b2b567d); - &data_word(0xfefee719, 0xd7d7b562, 0xabab4de6, 0x7676ec9a); - &data_word(0xcaca8f45, 0x82821f9d, 0xc9c98940, 0x7d7dfa87); - &data_word(0xfafaef15, 0x5959b2eb, 0x47478ec9, 0xf0f0fb0b); - &data_word(0xadad41ec, 0xd4d4b367, 0xa2a25ffd, 0xafaf45ea); - &data_word(0x9c9c23bf, 0xa4a453f7, 0x7272e496, 0xc0c09b5b); - &data_word(0xb7b775c2, 0xfdfde11c, 0x93933dae, 0x26264c6a); - &data_word(0x36366c5a, 0x3f3f7e41, 0xf7f7f502, 0xcccc834f); - &data_word(0x3434685c, 0xa5a551f4, 0xe5e5d134, 0xf1f1f908); - &data_word(0x7171e293, 0xd8d8ab73, 0x31316253, 0x15152a3f); - &data_word(0x0404080c, 0xc7c79552, 0x23234665, 0xc3c39d5e); - &data_word(0x18183028, 0x969637a1, 0x05050a0f, 0x9a9a2fb5); - &data_word(0x07070e09, 0x12122436, 0x80801b9b, 0xe2e2df3d); - &data_word(0xebebcd26, 0x27274e69, 0xb2b27fcd, 0x7575ea9f); - &data_word(0x0909121b, 0x83831d9e, 0x2c2c5874, 0x1a1a342e); - &data_word(0x1b1b362d, 0x6e6edcb2, 0x5a5ab4ee, 0xa0a05bfb); - &data_word(0x5252a4f6, 0x3b3b764d, 0xd6d6b761, 0xb3b37dce); - &data_word(0x2929527b, 0xe3e3dd3e, 0x2f2f5e71, 0x84841397); - &data_word(0x5353a6f5, 0xd1d1b968, 0x00000000, 0xededc12c); - &data_word(0x20204060, 0xfcfce31f, 0xb1b179c8, 0x5b5bb6ed); - &data_word(0x6a6ad4be, 0xcbcb8d46, 0xbebe67d9, 0x3939724b); - &data_word(0x4a4a94de, 0x4c4c98d4, 0x5858b0e8, 0xcfcf854a); - &data_word(0xd0d0bb6b, 0xefefc52a, 0xaaaa4fe5, 0xfbfbed16); - &data_word(0x434386c5, 0x4d4d9ad7, 0x33336655, 0x85851194); - &data_word(0x45458acf, 0xf9f9e910, 0x02020406, 0x7f7ffe81); - &data_word(0x5050a0f0, 0x3c3c7844, 0x9f9f25ba, 0xa8a84be3); - &data_word(0x5151a2f3, 0xa3a35dfe, 0x404080c0, 0x8f8f058a); - &data_word(0x92923fad, 0x9d9d21bc, 0x38387048, 0xf5f5f104); - &data_word(0xbcbc63df, 0xb6b677c1, 0xdadaaf75, 0x21214263); - &data_word(0x10102030, 0xffffe51a, 0xf3f3fd0e, 0xd2d2bf6d); - &data_word(0xcdcd814c, 0x0c0c1814, 0x13132635, 0xececc32f); - &data_word(0x5f5fbee1, 0x979735a2, 0x444488cc, 0x17172e39); - &data_word(0xc4c49357, 0xa7a755f2, 0x7e7efc82, 0x3d3d7a47); - &data_word(0x6464c8ac, 0x5d5dbae7, 0x1919322b, 0x7373e695); - &data_word(0x6060c0a0, 0x81811998, 0x4f4f9ed1, 0xdcdca37f); - &data_word(0x22224466, 0x2a2a547e, 0x90903bab, 0x88880b83); - &data_word(0x46468cca, 0xeeeec729, 0xb8b86bd3, 0x1414283c); - &data_word(0xdedea779, 0x5e5ebce2, 0x0b0b161d, 0xdbdbad76); - &data_word(0xe0e0db3b, 0x32326456, 0x3a3a744e, 0x0a0a141e); - &data_word(0x494992db, 0x06060c0a, 0x2424486c, 0x5c5cb8e4); - &data_word(0xc2c29f5d, 0xd3d3bd6e, 0xacac43ef, 0x6262c4a6); - &data_word(0x919139a8, 0x959531a4, 0xe4e4d337, 0x7979f28b); - &data_word(0xe7e7d532, 0xc8c88b43, 0x37376e59, 0x6d6ddab7); - &data_word(0x8d8d018c, 0xd5d5b164, 0x4e4e9cd2, 0xa9a949e0); - &data_word(0x6c6cd8b4, 0x5656acfa, 0xf4f4f307, 0xeaeacf25); - &data_word(0x6565caaf, 0x7a7af48e, 0xaeae47e9, 0x08081018); - &data_word(0xbaba6fd5, 0x7878f088, 0x25254a6f, 0x2e2e5c72); - &data_word(0x1c1c3824, 0xa6a657f1, 0xb4b473c7, 0xc6c69751); - &data_word(0xe8e8cb23, 0xdddda17c, 0x7474e89c, 0x1f1f3e21); - &data_word(0x4b4b96dd, 0xbdbd61dc, 0x8b8b0d86, 0x8a8a0f85); - &data_word(0x7070e090, 0x3e3e7c42, 0xb5b571c4, 0x6666ccaa); - &data_word(0x484890d8, 0x03030605, 0xf6f6f701, 0x0e0e1c12); - &data_word(0x6161c2a3, 0x35356a5f, 0x5757aef9, 0xb9b969d0); - &data_word(0x86861791, 0xc1c19958, 0x1d1d3a27, 0x9e9e27b9); - &data_word(0xe1e1d938, 0xf8f8eb13, 0x98982bb3, 0x11112233); - &data_word(0x6969d2bb, 0xd9d9a970, 0x8e8e0789, 0x949433a7); - &data_word(0x9b9b2db6, 0x1e1e3c22, 0x87871592, 0xe9e9c920); - &data_word(0xcece8749, 0x5555aaff, 0x28285078, 0xdfdfa57a); - &data_word(0x8c8c038f, 0xa1a159f8, 0x89890980, 0x0d0d1a17); - &data_word(0xbfbf65da, 0xe6e6d731, 0x424284c6, 0x6868d0b8); - &data_word(0x414182c3, 0x999929b0, 0x2d2d5a77, 0x0f0f1e11); - &data_word(0xb0b07bcb, 0x5454a8fc, 0xbbbb6dd6, 0x16162c3a); -#Te2: - &data_word(0x63c6a563, 0x7cf8847c, 0x77ee9977, 0x7bf68d7b); - &data_word(0xf2ff0df2, 0x6bd6bd6b, 0x6fdeb16f, 0xc59154c5); - &data_word(0x30605030, 0x01020301, 0x67cea967, 0x2b567d2b); - &data_word(0xfee719fe, 0xd7b562d7, 0xab4de6ab, 0x76ec9a76); - &data_word(0xca8f45ca, 0x821f9d82, 0xc98940c9, 0x7dfa877d); - &data_word(0xfaef15fa, 0x59b2eb59, 0x478ec947, 0xf0fb0bf0); - &data_word(0xad41ecad, 0xd4b367d4, 0xa25ffda2, 0xaf45eaaf); - &data_word(0x9c23bf9c, 0xa453f7a4, 0x72e49672, 0xc09b5bc0); - &data_word(0xb775c2b7, 0xfde11cfd, 0x933dae93, 0x264c6a26); - &data_word(0x366c5a36, 0x3f7e413f, 0xf7f502f7, 0xcc834fcc); - &data_word(0x34685c34, 0xa551f4a5, 0xe5d134e5, 0xf1f908f1); - &data_word(0x71e29371, 0xd8ab73d8, 0x31625331, 0x152a3f15); - &data_word(0x04080c04, 0xc79552c7, 0x23466523, 0xc39d5ec3); - &data_word(0x18302818, 0x9637a196, 0x050a0f05, 0x9a2fb59a); - &data_word(0x070e0907, 0x12243612, 0x801b9b80, 0xe2df3de2); - &data_word(0xebcd26eb, 0x274e6927, 0xb27fcdb2, 0x75ea9f75); - &data_word(0x09121b09, 0x831d9e83, 0x2c58742c, 0x1a342e1a); - &data_word(0x1b362d1b, 0x6edcb26e, 0x5ab4ee5a, 0xa05bfba0); - &data_word(0x52a4f652, 0x3b764d3b, 0xd6b761d6, 0xb37dceb3); - &data_word(0x29527b29, 0xe3dd3ee3, 0x2f5e712f, 0x84139784); - &data_word(0x53a6f553, 0xd1b968d1, 0x00000000, 0xedc12ced); - &data_word(0x20406020, 0xfce31ffc, 0xb179c8b1, 0x5bb6ed5b); - &data_word(0x6ad4be6a, 0xcb8d46cb, 0xbe67d9be, 0x39724b39); - &data_word(0x4a94de4a, 0x4c98d44c, 0x58b0e858, 0xcf854acf); - &data_word(0xd0bb6bd0, 0xefc52aef, 0xaa4fe5aa, 0xfbed16fb); - &data_word(0x4386c543, 0x4d9ad74d, 0x33665533, 0x85119485); - &data_word(0x458acf45, 0xf9e910f9, 0x02040602, 0x7ffe817f); - &data_word(0x50a0f050, 0x3c78443c, 0x9f25ba9f, 0xa84be3a8); - &data_word(0x51a2f351, 0xa35dfea3, 0x4080c040, 0x8f058a8f); - &data_word(0x923fad92, 0x9d21bc9d, 0x38704838, 0xf5f104f5); - &data_word(0xbc63dfbc, 0xb677c1b6, 0xdaaf75da, 0x21426321); - &data_word(0x10203010, 0xffe51aff, 0xf3fd0ef3, 0xd2bf6dd2); - &data_word(0xcd814ccd, 0x0c18140c, 0x13263513, 0xecc32fec); - &data_word(0x5fbee15f, 0x9735a297, 0x4488cc44, 0x172e3917); - &data_word(0xc49357c4, 0xa755f2a7, 0x7efc827e, 0x3d7a473d); - &data_word(0x64c8ac64, 0x5dbae75d, 0x19322b19, 0x73e69573); - &data_word(0x60c0a060, 0x81199881, 0x4f9ed14f, 0xdca37fdc); - &data_word(0x22446622, 0x2a547e2a, 0x903bab90, 0x880b8388); - &data_word(0x468cca46, 0xeec729ee, 0xb86bd3b8, 0x14283c14); - &data_word(0xdea779de, 0x5ebce25e, 0x0b161d0b, 0xdbad76db); - &data_word(0xe0db3be0, 0x32645632, 0x3a744e3a, 0x0a141e0a); - &data_word(0x4992db49, 0x060c0a06, 0x24486c24, 0x5cb8e45c); - &data_word(0xc29f5dc2, 0xd3bd6ed3, 0xac43efac, 0x62c4a662); - &data_word(0x9139a891, 0x9531a495, 0xe4d337e4, 0x79f28b79); - &data_word(0xe7d532e7, 0xc88b43c8, 0x376e5937, 0x6ddab76d); - &data_word(0x8d018c8d, 0xd5b164d5, 0x4e9cd24e, 0xa949e0a9); - &data_word(0x6cd8b46c, 0x56acfa56, 0xf4f307f4, 0xeacf25ea); - &data_word(0x65caaf65, 0x7af48e7a, 0xae47e9ae, 0x08101808); - &data_word(0xba6fd5ba, 0x78f08878, 0x254a6f25, 0x2e5c722e); - &data_word(0x1c38241c, 0xa657f1a6, 0xb473c7b4, 0xc69751c6); - &data_word(0xe8cb23e8, 0xdda17cdd, 0x74e89c74, 0x1f3e211f); - &data_word(0x4b96dd4b, 0xbd61dcbd, 0x8b0d868b, 0x8a0f858a); - &data_word(0x70e09070, 0x3e7c423e, 0xb571c4b5, 0x66ccaa66); - &data_word(0x4890d848, 0x03060503, 0xf6f701f6, 0x0e1c120e); - &data_word(0x61c2a361, 0x356a5f35, 0x57aef957, 0xb969d0b9); - &data_word(0x86179186, 0xc19958c1, 0x1d3a271d, 0x9e27b99e); - &data_word(0xe1d938e1, 0xf8eb13f8, 0x982bb398, 0x11223311); - &data_word(0x69d2bb69, 0xd9a970d9, 0x8e07898e, 0x9433a794); - &data_word(0x9b2db69b, 0x1e3c221e, 0x87159287, 0xe9c920e9); - &data_word(0xce8749ce, 0x55aaff55, 0x28507828, 0xdfa57adf); - &data_word(0x8c038f8c, 0xa159f8a1, 0x89098089, 0x0d1a170d); - &data_word(0xbf65dabf, 0xe6d731e6, 0x4284c642, 0x68d0b868); - &data_word(0x4182c341, 0x9929b099, 0x2d5a772d, 0x0f1e110f); - &data_word(0xb07bcbb0, 0x54a8fc54, 0xbb6dd6bb, 0x162c3a16); -#Te3: - &data_word(0xc6a56363, 0xf8847c7c, 0xee997777, 0xf68d7b7b); - &data_word(0xff0df2f2, 0xd6bd6b6b, 0xdeb16f6f, 0x9154c5c5); - &data_word(0x60503030, 0x02030101, 0xcea96767, 0x567d2b2b); - &data_word(0xe719fefe, 0xb562d7d7, 0x4de6abab, 0xec9a7676); - &data_word(0x8f45caca, 0x1f9d8282, 0x8940c9c9, 0xfa877d7d); - &data_word(0xef15fafa, 0xb2eb5959, 0x8ec94747, 0xfb0bf0f0); - &data_word(0x41ecadad, 0xb367d4d4, 0x5ffda2a2, 0x45eaafaf); - &data_word(0x23bf9c9c, 0x53f7a4a4, 0xe4967272, 0x9b5bc0c0); - &data_word(0x75c2b7b7, 0xe11cfdfd, 0x3dae9393, 0x4c6a2626); - &data_word(0x6c5a3636, 0x7e413f3f, 0xf502f7f7, 0x834fcccc); - &data_word(0x685c3434, 0x51f4a5a5, 0xd134e5e5, 0xf908f1f1); - &data_word(0xe2937171, 0xab73d8d8, 0x62533131, 0x2a3f1515); - &data_word(0x080c0404, 0x9552c7c7, 0x46652323, 0x9d5ec3c3); - &data_word(0x30281818, 0x37a19696, 0x0a0f0505, 0x2fb59a9a); - &data_word(0x0e090707, 0x24361212, 0x1b9b8080, 0xdf3de2e2); - &data_word(0xcd26ebeb, 0x4e692727, 0x7fcdb2b2, 0xea9f7575); - &data_word(0x121b0909, 0x1d9e8383, 0x58742c2c, 0x342e1a1a); - &data_word(0x362d1b1b, 0xdcb26e6e, 0xb4ee5a5a, 0x5bfba0a0); - &data_word(0xa4f65252, 0x764d3b3b, 0xb761d6d6, 0x7dceb3b3); - &data_word(0x527b2929, 0xdd3ee3e3, 0x5e712f2f, 0x13978484); - &data_word(0xa6f55353, 0xb968d1d1, 0x00000000, 0xc12ceded); - &data_word(0x40602020, 0xe31ffcfc, 0x79c8b1b1, 0xb6ed5b5b); - &data_word(0xd4be6a6a, 0x8d46cbcb, 0x67d9bebe, 0x724b3939); - &data_word(0x94de4a4a, 0x98d44c4c, 0xb0e85858, 0x854acfcf); - &data_word(0xbb6bd0d0, 0xc52aefef, 0x4fe5aaaa, 0xed16fbfb); - &data_word(0x86c54343, 0x9ad74d4d, 0x66553333, 0x11948585); - &data_word(0x8acf4545, 0xe910f9f9, 0x04060202, 0xfe817f7f); - &data_word(0xa0f05050, 0x78443c3c, 0x25ba9f9f, 0x4be3a8a8); - &data_word(0xa2f35151, 0x5dfea3a3, 0x80c04040, 0x058a8f8f); - &data_word(0x3fad9292, 0x21bc9d9d, 0x70483838, 0xf104f5f5); - &data_word(0x63dfbcbc, 0x77c1b6b6, 0xaf75dada, 0x42632121); - &data_word(0x20301010, 0xe51affff, 0xfd0ef3f3, 0xbf6dd2d2); - &data_word(0x814ccdcd, 0x18140c0c, 0x26351313, 0xc32fecec); - &data_word(0xbee15f5f, 0x35a29797, 0x88cc4444, 0x2e391717); - &data_word(0x9357c4c4, 0x55f2a7a7, 0xfc827e7e, 0x7a473d3d); - &data_word(0xc8ac6464, 0xbae75d5d, 0x322b1919, 0xe6957373); - &data_word(0xc0a06060, 0x19988181, 0x9ed14f4f, 0xa37fdcdc); - &data_word(0x44662222, 0x547e2a2a, 0x3bab9090, 0x0b838888); - &data_word(0x8cca4646, 0xc729eeee, 0x6bd3b8b8, 0x283c1414); - &data_word(0xa779dede, 0xbce25e5e, 0x161d0b0b, 0xad76dbdb); - &data_word(0xdb3be0e0, 0x64563232, 0x744e3a3a, 0x141e0a0a); - &data_word(0x92db4949, 0x0c0a0606, 0x486c2424, 0xb8e45c5c); - &data_word(0x9f5dc2c2, 0xbd6ed3d3, 0x43efacac, 0xc4a66262); - &data_word(0x39a89191, 0x31a49595, 0xd337e4e4, 0xf28b7979); - &data_word(0xd532e7e7, 0x8b43c8c8, 0x6e593737, 0xdab76d6d); - &data_word(0x018c8d8d, 0xb164d5d5, 0x9cd24e4e, 0x49e0a9a9); - &data_word(0xd8b46c6c, 0xacfa5656, 0xf307f4f4, 0xcf25eaea); - &data_word(0xcaaf6565, 0xf48e7a7a, 0x47e9aeae, 0x10180808); - &data_word(0x6fd5baba, 0xf0887878, 0x4a6f2525, 0x5c722e2e); - &data_word(0x38241c1c, 0x57f1a6a6, 0x73c7b4b4, 0x9751c6c6); - &data_word(0xcb23e8e8, 0xa17cdddd, 0xe89c7474, 0x3e211f1f); - &data_word(0x96dd4b4b, 0x61dcbdbd, 0x0d868b8b, 0x0f858a8a); - &data_word(0xe0907070, 0x7c423e3e, 0x71c4b5b5, 0xccaa6666); - &data_word(0x90d84848, 0x06050303, 0xf701f6f6, 0x1c120e0e); - &data_word(0xc2a36161, 0x6a5f3535, 0xaef95757, 0x69d0b9b9); - &data_word(0x17918686, 0x9958c1c1, 0x3a271d1d, 0x27b99e9e); - &data_word(0xd938e1e1, 0xeb13f8f8, 0x2bb39898, 0x22331111); - &data_word(0xd2bb6969, 0xa970d9d9, 0x07898e8e, 0x33a79494); - &data_word(0x2db69b9b, 0x3c221e1e, 0x15928787, 0xc920e9e9); - &data_word(0x8749cece, 0xaaff5555, 0x50782828, 0xa57adfdf); - &data_word(0x038f8c8c, 0x59f8a1a1, 0x09808989, 0x1a170d0d); - &data_word(0x65dabfbf, 0xd731e6e6, 0x84c64242, 0xd0b86868); - &data_word(0x82c34141, 0x29b09999, 0x5a772d2d, 0x1e110f0f); - &data_word(0x7bcbb0b0, 0xa8fc5454, 0x6dd6bbbb, 0x2c3a1616); -#Te4: - &data_word(0x63636363, 0x7c7c7c7c, 0x77777777, 0x7b7b7b7b); - &data_word(0xf2f2f2f2, 0x6b6b6b6b, 0x6f6f6f6f, 0xc5c5c5c5); - &data_word(0x30303030, 0x01010101, 0x67676767, 0x2b2b2b2b); - &data_word(0xfefefefe, 0xd7d7d7d7, 0xabababab, 0x76767676); - &data_word(0xcacacaca, 0x82828282, 0xc9c9c9c9, 0x7d7d7d7d); - &data_word(0xfafafafa, 0x59595959, 0x47474747, 0xf0f0f0f0); - &data_word(0xadadadad, 0xd4d4d4d4, 0xa2a2a2a2, 0xafafafaf); - &data_word(0x9c9c9c9c, 0xa4a4a4a4, 0x72727272, 0xc0c0c0c0); - &data_word(0xb7b7b7b7, 0xfdfdfdfd, 0x93939393, 0x26262626); - &data_word(0x36363636, 0x3f3f3f3f, 0xf7f7f7f7, 0xcccccccc); - &data_word(0x34343434, 0xa5a5a5a5, 0xe5e5e5e5, 0xf1f1f1f1); - &data_word(0x71717171, 0xd8d8d8d8, 0x31313131, 0x15151515); - &data_word(0x04040404, 0xc7c7c7c7, 0x23232323, 0xc3c3c3c3); - &data_word(0x18181818, 0x96969696, 0x05050505, 0x9a9a9a9a); - &data_word(0x07070707, 0x12121212, 0x80808080, 0xe2e2e2e2); - &data_word(0xebebebeb, 0x27272727, 0xb2b2b2b2, 0x75757575); - &data_word(0x09090909, 0x83838383, 0x2c2c2c2c, 0x1a1a1a1a); - &data_word(0x1b1b1b1b, 0x6e6e6e6e, 0x5a5a5a5a, 0xa0a0a0a0); - &data_word(0x52525252, 0x3b3b3b3b, 0xd6d6d6d6, 0xb3b3b3b3); - &data_word(0x29292929, 0xe3e3e3e3, 0x2f2f2f2f, 0x84848484); - &data_word(0x53535353, 0xd1d1d1d1, 0x00000000, 0xedededed); - &data_word(0x20202020, 0xfcfcfcfc, 0xb1b1b1b1, 0x5b5b5b5b); - &data_word(0x6a6a6a6a, 0xcbcbcbcb, 0xbebebebe, 0x39393939); - &data_word(0x4a4a4a4a, 0x4c4c4c4c, 0x58585858, 0xcfcfcfcf); - &data_word(0xd0d0d0d0, 0xefefefef, 0xaaaaaaaa, 0xfbfbfbfb); - &data_word(0x43434343, 0x4d4d4d4d, 0x33333333, 0x85858585); - &data_word(0x45454545, 0xf9f9f9f9, 0x02020202, 0x7f7f7f7f); - &data_word(0x50505050, 0x3c3c3c3c, 0x9f9f9f9f, 0xa8a8a8a8); - &data_word(0x51515151, 0xa3a3a3a3, 0x40404040, 0x8f8f8f8f); - &data_word(0x92929292, 0x9d9d9d9d, 0x38383838, 0xf5f5f5f5); - &data_word(0xbcbcbcbc, 0xb6b6b6b6, 0xdadadada, 0x21212121); - &data_word(0x10101010, 0xffffffff, 0xf3f3f3f3, 0xd2d2d2d2); - &data_word(0xcdcdcdcd, 0x0c0c0c0c, 0x13131313, 0xecececec); - &data_word(0x5f5f5f5f, 0x97979797, 0x44444444, 0x17171717); - &data_word(0xc4c4c4c4, 0xa7a7a7a7, 0x7e7e7e7e, 0x3d3d3d3d); - &data_word(0x64646464, 0x5d5d5d5d, 0x19191919, 0x73737373); - &data_word(0x60606060, 0x81818181, 0x4f4f4f4f, 0xdcdcdcdc); - &data_word(0x22222222, 0x2a2a2a2a, 0x90909090, 0x88888888); - &data_word(0x46464646, 0xeeeeeeee, 0xb8b8b8b8, 0x14141414); - &data_word(0xdededede, 0x5e5e5e5e, 0x0b0b0b0b, 0xdbdbdbdb); - &data_word(0xe0e0e0e0, 0x32323232, 0x3a3a3a3a, 0x0a0a0a0a); - &data_word(0x49494949, 0x06060606, 0x24242424, 0x5c5c5c5c); - &data_word(0xc2c2c2c2, 0xd3d3d3d3, 0xacacacac, 0x62626262); - &data_word(0x91919191, 0x95959595, 0xe4e4e4e4, 0x79797979); - &data_word(0xe7e7e7e7, 0xc8c8c8c8, 0x37373737, 0x6d6d6d6d); - &data_word(0x8d8d8d8d, 0xd5d5d5d5, 0x4e4e4e4e, 0xa9a9a9a9); - &data_word(0x6c6c6c6c, 0x56565656, 0xf4f4f4f4, 0xeaeaeaea); - &data_word(0x65656565, 0x7a7a7a7a, 0xaeaeaeae, 0x08080808); - &data_word(0xbabababa, 0x78787878, 0x25252525, 0x2e2e2e2e); - &data_word(0x1c1c1c1c, 0xa6a6a6a6, 0xb4b4b4b4, 0xc6c6c6c6); - &data_word(0xe8e8e8e8, 0xdddddddd, 0x74747474, 0x1f1f1f1f); - &data_word(0x4b4b4b4b, 0xbdbdbdbd, 0x8b8b8b8b, 0x8a8a8a8a); - &data_word(0x70707070, 0x3e3e3e3e, 0xb5b5b5b5, 0x66666666); - &data_word(0x48484848, 0x03030303, 0xf6f6f6f6, 0x0e0e0e0e); - &data_word(0x61616161, 0x35353535, 0x57575757, 0xb9b9b9b9); - &data_word(0x86868686, 0xc1c1c1c1, 0x1d1d1d1d, 0x9e9e9e9e); - &data_word(0xe1e1e1e1, 0xf8f8f8f8, 0x98989898, 0x11111111); - &data_word(0x69696969, 0xd9d9d9d9, 0x8e8e8e8e, 0x94949494); - &data_word(0x9b9b9b9b, 0x1e1e1e1e, 0x87878787, 0xe9e9e9e9); - &data_word(0xcececece, 0x55555555, 0x28282828, 0xdfdfdfdf); - &data_word(0x8c8c8c8c, 0xa1a1a1a1, 0x89898989, 0x0d0d0d0d); - &data_word(0xbfbfbfbf, 0xe6e6e6e6, 0x42424242, 0x68686868); - &data_word(0x41414141, 0x99999999, 0x2d2d2d2d, 0x0f0f0f0f); - &data_word(0xb0b0b0b0, 0x54545454, 0xbbbbbbbb, 0x16161616); + &_data_word(0xa56363c6, 0x847c7cf8, 0x997777ee, 0x8d7b7bf6); + &_data_word(0x0df2f2ff, 0xbd6b6bd6, 0xb16f6fde, 0x54c5c591); + &_data_word(0x50303060, 0x03010102, 0xa96767ce, 0x7d2b2b56); + &_data_word(0x19fefee7, 0x62d7d7b5, 0xe6abab4d, 0x9a7676ec); + &_data_word(0x45caca8f, 0x9d82821f, 0x40c9c989, 0x877d7dfa); + &_data_word(0x15fafaef, 0xeb5959b2, 0xc947478e, 0x0bf0f0fb); + &_data_word(0xecadad41, 0x67d4d4b3, 0xfda2a25f, 0xeaafaf45); + &_data_word(0xbf9c9c23, 0xf7a4a453, 0x967272e4, 0x5bc0c09b); + &_data_word(0xc2b7b775, 0x1cfdfde1, 0xae93933d, 0x6a26264c); + &_data_word(0x5a36366c, 0x413f3f7e, 0x02f7f7f5, 0x4fcccc83); + &_data_word(0x5c343468, 0xf4a5a551, 0x34e5e5d1, 0x08f1f1f9); + &_data_word(0x937171e2, 0x73d8d8ab, 0x53313162, 0x3f15152a); + &_data_word(0x0c040408, 0x52c7c795, 0x65232346, 0x5ec3c39d); + &_data_word(0x28181830, 0xa1969637, 0x0f05050a, 0xb59a9a2f); + &_data_word(0x0907070e, 0x36121224, 0x9b80801b, 0x3de2e2df); + &_data_word(0x26ebebcd, 0x6927274e, 0xcdb2b27f, 0x9f7575ea); + &_data_word(0x1b090912, 0x9e83831d, 0x742c2c58, 0x2e1a1a34); + &_data_word(0x2d1b1b36, 0xb26e6edc, 0xee5a5ab4, 0xfba0a05b); + &_data_word(0xf65252a4, 0x4d3b3b76, 0x61d6d6b7, 0xceb3b37d); + &_data_word(0x7b292952, 0x3ee3e3dd, 0x712f2f5e, 0x97848413); + &_data_word(0xf55353a6, 0x68d1d1b9, 0x00000000, 0x2cededc1); + &_data_word(0x60202040, 0x1ffcfce3, 0xc8b1b179, 0xed5b5bb6); + &_data_word(0xbe6a6ad4, 0x46cbcb8d, 0xd9bebe67, 0x4b393972); + &_data_word(0xde4a4a94, 0xd44c4c98, 0xe85858b0, 0x4acfcf85); + &_data_word(0x6bd0d0bb, 0x2aefefc5, 0xe5aaaa4f, 0x16fbfbed); + &_data_word(0xc5434386, 0xd74d4d9a, 0x55333366, 0x94858511); + &_data_word(0xcf45458a, 0x10f9f9e9, 0x06020204, 0x817f7ffe); + &_data_word(0xf05050a0, 0x443c3c78, 0xba9f9f25, 0xe3a8a84b); + &_data_word(0xf35151a2, 0xfea3a35d, 0xc0404080, 0x8a8f8f05); + &_data_word(0xad92923f, 0xbc9d9d21, 0x48383870, 0x04f5f5f1); + &_data_word(0xdfbcbc63, 0xc1b6b677, 0x75dadaaf, 0x63212142); + &_data_word(0x30101020, 0x1affffe5, 0x0ef3f3fd, 0x6dd2d2bf); + &_data_word(0x4ccdcd81, 0x140c0c18, 0x35131326, 0x2fececc3); + &_data_word(0xe15f5fbe, 0xa2979735, 0xcc444488, 0x3917172e); + &_data_word(0x57c4c493, 0xf2a7a755, 0x827e7efc, 0x473d3d7a); + &_data_word(0xac6464c8, 0xe75d5dba, 0x2b191932, 0x957373e6); + &_data_word(0xa06060c0, 0x98818119, 0xd14f4f9e, 0x7fdcdca3); + &_data_word(0x66222244, 0x7e2a2a54, 0xab90903b, 0x8388880b); + &_data_word(0xca46468c, 0x29eeeec7, 0xd3b8b86b, 0x3c141428); + &_data_word(0x79dedea7, 0xe25e5ebc, 0x1d0b0b16, 0x76dbdbad); + &_data_word(0x3be0e0db, 0x56323264, 0x4e3a3a74, 0x1e0a0a14); + &_data_word(0xdb494992, 0x0a06060c, 0x6c242448, 0xe45c5cb8); + &_data_word(0x5dc2c29f, 0x6ed3d3bd, 0xefacac43, 0xa66262c4); + &_data_word(0xa8919139, 0xa4959531, 0x37e4e4d3, 0x8b7979f2); + &_data_word(0x32e7e7d5, 0x43c8c88b, 0x5937376e, 0xb76d6dda); + &_data_word(0x8c8d8d01, 0x64d5d5b1, 0xd24e4e9c, 0xe0a9a949); + &_data_word(0xb46c6cd8, 0xfa5656ac, 0x07f4f4f3, 0x25eaeacf); + &_data_word(0xaf6565ca, 0x8e7a7af4, 0xe9aeae47, 0x18080810); + &_data_word(0xd5baba6f, 0x887878f0, 0x6f25254a, 0x722e2e5c); + &_data_word(0x241c1c38, 0xf1a6a657, 0xc7b4b473, 0x51c6c697); + &_data_word(0x23e8e8cb, 0x7cdddda1, 0x9c7474e8, 0x211f1f3e); + &_data_word(0xdd4b4b96, 0xdcbdbd61, 0x868b8b0d, 0x858a8a0f); + &_data_word(0x907070e0, 0x423e3e7c, 0xc4b5b571, 0xaa6666cc); + &_data_word(0xd8484890, 0x05030306, 0x01f6f6f7, 0x120e0e1c); + &_data_word(0xa36161c2, 0x5f35356a, 0xf95757ae, 0xd0b9b969); + &_data_word(0x91868617, 0x58c1c199, 0x271d1d3a, 0xb99e9e27); + &_data_word(0x38e1e1d9, 0x13f8f8eb, 0xb398982b, 0x33111122); + &_data_word(0xbb6969d2, 0x70d9d9a9, 0x898e8e07, 0xa7949433); + &_data_word(0xb69b9b2d, 0x221e1e3c, 0x92878715, 0x20e9e9c9); + &_data_word(0x49cece87, 0xff5555aa, 0x78282850, 0x7adfdfa5); + &_data_word(0x8f8c8c03, 0xf8a1a159, 0x80898909, 0x170d0d1a); + &_data_word(0xdabfbf65, 0x31e6e6d7, 0xc6424284, 0xb86868d0); + &_data_word(0xc3414182, 0xb0999929, 0x772d2d5a, 0x110f0f1e); + &_data_word(0xcbb0b07b, 0xfc5454a8, 0xd6bbbb6d, 0x3a16162c); + +#Te4 # four copies of Te4 to choose from to avoid L1 aliasing + &data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5); + &data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76); + &data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0); + &data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0); + &data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc); + &data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15); + &data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a); + &data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75); + &data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0); + &data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84); + &data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b); + &data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf); + &data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85); + &data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8); + &data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5); + &data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2); + &data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17); + &data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73); + &data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88); + &data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb); + &data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c); + &data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79); + &data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9); + &data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08); + &data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6); + &data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a); + &data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e); + &data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e); + &data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94); + &data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf); + &data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68); + &data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16); + + &data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5); + &data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76); + &data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0); + &data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0); + &data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc); + &data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15); + &data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a); + &data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75); + &data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0); + &data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84); + &data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b); + &data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf); + &data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85); + &data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8); + &data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5); + &data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2); + &data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17); + &data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73); + &data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88); + &data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb); + &data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c); + &data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79); + &data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9); + &data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08); + &data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6); + &data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a); + &data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e); + &data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e); + &data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94); + &data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf); + &data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68); + &data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16); + + &data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5); + &data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76); + &data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0); + &data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0); + &data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc); + &data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15); + &data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a); + &data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75); + &data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0); + &data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84); + &data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b); + &data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf); + &data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85); + &data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8); + &data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5); + &data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2); + &data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17); + &data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73); + &data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88); + &data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb); + &data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c); + &data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79); + &data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9); + &data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08); + &data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6); + &data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a); + &data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e); + &data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e); + &data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94); + &data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf); + &data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68); + &data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16); + + &data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5); + &data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76); + &data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0); + &data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0); + &data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc); + &data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15); + &data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a); + &data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75); + &data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0); + &data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84); + &data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b); + &data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf); + &data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85); + &data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8); + &data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5); + &data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2); + &data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17); + &data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73); + &data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88); + &data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb); + &data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c); + &data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79); + &data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9); + &data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08); + &data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6); + &data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a); + &data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e); + &data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e); + &data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94); + &data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf); + &data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68); + &data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16); #rcon: &data_word(0x00000001, 0x00000002, 0x00000004, 0x00000008); - &data_word(0x00000010, 0x00000020, 0x00000040, 0x00000080); - &data_word(0x0000001b, 0x00000036); + &data_word(0x00000010, 0x00000020, 0x00000040, 0x00000080); + &data_word(0x0000001b, 0x00000036, 0x00000000, 0x00000000); + &data_word(0x00000000, 0x00000000, 0x00000000, 0x00000000); &function_end_B("_x86_AES_encrypt"); # void AES_encrypt (const void *inp,void *out,const AES_KEY *key); -&public_label("AES_Te"); &function_begin("AES_encrypt"); &mov ($acc,&wparam(0)); # load inp &mov ($key,&wparam(2)); # load key + &mov ($s0,"esp"); + &sub ("esp",36); + &and ("esp",-64); # align to cache-line + + # place stack frame just "above" the key schedule + &lea ($s1,&DWP(-64-63,$key)); + &sub ($s1,"esp"); + &neg ($s1); + &and ($s1,0x3C0); # modulo 1024, but aligned to cache-line + &sub ("esp",$s1); + &add ("esp",4); # 4 is reserved for caller's return address + &mov ($_esp,$s0); # save stack pointer + &call (&label("pic_point")); # make it PIC! &set_label("pic_point"); - &blindpop("ebp"); - &lea ("ebp",&DWP(&label("AES_Te")."-".&label("pic_point"),"ebp")); - + &blindpop($tbl); + &picmeup($s0,"OPENSSL_ia32cap_P",$tbl,&label("pic_point")) if (!$x86only); + &lea ($tbl,&DWP(&label("AES_Te")."-".&label("pic_point"),$tbl)); + + # pick Te4 copy which can't "overlap" with stack frame or key schedule + &lea ($s1,&DWP(768-4,"esp")); + &sub ($s1,$tbl); + &and ($s1,0x300); + &lea ($tbl,&DWP(2048+128,$tbl,$s1)); + + if (!$x86only) { + &bt (&DWP(0,$s0),25); # check for SSE bit + &jnc (&label("x86")); + + &movq ("mm0",&QWP(0,$acc)); + &movq ("mm4",&QWP(8,$acc)); + &call ("_sse_AES_encrypt_compact"); + &mov ("esp",$_esp); # restore stack pointer + &mov ($acc,&wparam(1)); # load out + &movq (&QWP(0,$acc),"mm0"); # write output data + &movq (&QWP(8,$acc),"mm4"); + &emms (); + &function_end_A(); + } + &set_label("x86",16); + &mov ($_tbl,$tbl); &mov ($s0,&DWP(0,$acc)); # load input data &mov ($s1,&DWP(4,$acc)); &mov ($s2,&DWP(8,$acc)); &mov ($s3,&DWP(12,$acc)); - - &call ("_x86_AES_encrypt"); - + &call ("_x86_AES_encrypt_compact"); + &mov ("esp",$_esp); # restore stack pointer &mov ($acc,&wparam(1)); # load out &mov (&DWP(0,$acc),$s0); # write output data &mov (&DWP(4,$acc),$s1); @@ -706,7 +1215,370 @@ sub enclast() &mov (&DWP(12,$acc),$s3); &function_end("AES_encrypt"); -#------------------------------------------------------------------# +#--------------------------------------------------------------------# + +###################################################################### +# "Compact" block function +###################################################################### + +sub deccompact() +{ my $Fn = mov; + while ($#_>5) { pop(@_); $Fn=sub{}; } + my ($i,$td,@s)=@_; + my $tmp = $key; + my $out = $i==3?$s[0]:$acc; + + # $Fn is used in first compact round and its purpose is to + # void restoration of some values from stack, so that after + # 4xdeccompact with extra argument $key, $s0 and $s1 values + # are left there... + if($i==3) { &$Fn ($key,$__key); } + else { &mov ($out,$s[0]); } + &and ($out,0xFF); + &movz ($out,&BP(-128,$td,$out,1)); + + if ($i==3) { $tmp=$s[1]; } + &movz ($tmp,&HB($s[1])); + &movz ($tmp,&BP(-128,$td,$tmp,1)); + &shl ($tmp,8); + &xor ($out,$tmp); + + if ($i==3) { $tmp=$s[2]; &mov ($s[1],$acc); } + else { mov ($tmp,$s[2]); } + &shr ($tmp,16); + &and ($tmp,0xFF); + &movz ($tmp,&BP(-128,$td,$tmp,1)); + &shl ($tmp,16); + &xor ($out,$tmp); + + if ($i==3) { $tmp=$s[3]; &$Fn ($s[2],$__s1); } + else { &mov ($tmp,$s[3]); } + &shr ($tmp,24); + &movz ($tmp,&BP(-128,$td,$tmp,1)); + &shl ($tmp,24); + &xor ($out,$tmp); + if ($i<2) { &mov (&DWP(4+4*$i,"esp"),$out); } + if ($i==3) { &$Fn ($s[3],$__s0); } +} + +# must be called with 2,3,0,1 as argument sequence!!! +sub dectransform() +{ my @s = ($s0,$s1,$s2,$s3); + my $i = shift; + my $tmp = $key; + my $tp2 = @s[($i+2)%4]; $tp2 = @s[2] if ($i==1); + my $tp4 = @s[($i+3)%4]; $tp4 = @s[3] if ($i==1); + my $tp8 = $tbl; + + &mov ($acc,$s[$i]); + &and ($acc,0x80808080); + &mov ($tmp,$acc); + &shr ($tmp,7); + &lea ($tp2,&DWP(0,$s[$i],$s[$i])); + &sub ($acc,$tmp); + &and ($tp2,0xfefefefe); + &and ($acc,0x1b1b1b1b); + &xor ($acc,$tp2); + &mov ($tp2,$acc); + + &and ($acc,0x80808080); + &mov ($tmp,$acc); + &shr ($tmp,7); + &lea ($tp4,&DWP(0,$tp2,$tp2)); + &sub ($acc,$tmp); + &and ($tp4,0xfefefefe); + &and ($acc,0x1b1b1b1b); + &xor ($tp2,$s[$i]); # tp2^tp1 + &xor ($acc,$tp4); + &mov ($tp4,$acc); + + &and ($acc,0x80808080); + &mov ($tmp,$acc); + &shr ($tmp,7); + &lea ($tp8,&DWP(0,$tp4,$tp4)); + &sub ($acc,$tmp); + &and ($tp8,0xfefefefe); + &and ($acc,0x1b1b1b1b); + &xor ($tp4,$s[$i]); # tp4^tp1 + &rotl ($s[$i],8); # = ROTATE(tp1,8) + &xor ($tp8,$acc); + + &xor ($s[$i],$tp2); + &xor ($tp2,$tp8); + &rotl ($tp2,24); + &xor ($s[$i],$tp4); + &xor ($tp4,$tp8); + &rotl ($tp4,16); + &xor ($s[$i],$tp8); # ^= tp8^(tp4^tp1)^(tp2^tp1) + &rotl ($tp8,8); + &xor ($s[$i],$tp2); # ^= ROTATE(tp8^tp2^tp1,24) + &xor ($s[$i],$tp4); # ^= ROTATE(tp8^tp4^tp1,16) + &mov ($s[0],$__s0) if($i==2); #prefetch $s0 + &mov ($s[1],$__s1) if($i==3); #prefetch $s1 + &mov ($s[2],$__s2) if($i==1); + &xor ($s[$i],$tp8); # ^= ROTATE(tp8,8) + + &mov ($s[3],$__s3) if($i==1); + &mov (&DWP(4+4*$i,"esp"),$s[$i]) if($i>=2); +} + +&function_begin_B("_x86_AES_decrypt_compact"); + # note that caller is expected to allocate stack frame for me! + &mov ($__key,$key); # save key + + &xor ($s0,&DWP(0,$key)); # xor with key + &xor ($s1,&DWP(4,$key)); + &xor ($s2,&DWP(8,$key)); + &xor ($s3,&DWP(12,$key)); + + &mov ($acc,&DWP(240,$key)); # load key->rounds + + &lea ($acc,&DWP(-2,$acc,$acc)); + &lea ($acc,&DWP(0,$key,$acc,8)); + &mov ($__end,$acc); # end of key schedule + + # prefetch Td4 + &mov ($key,&DWP(0-128,$tbl)); + &mov ($acc,&DWP(32-128,$tbl)); + &mov ($key,&DWP(64-128,$tbl)); + &mov ($acc,&DWP(96-128,$tbl)); + &mov ($key,&DWP(128-128,$tbl)); + &mov ($acc,&DWP(160-128,$tbl)); + &mov ($key,&DWP(192-128,$tbl)); + &mov ($acc,&DWP(224-128,$tbl)); + + &set_label("loop",16); + + &deccompact(0,$tbl,$s0,$s3,$s2,$s1,1); + &deccompact(1,$tbl,$s1,$s0,$s3,$s2,1); + &deccompact(2,$tbl,$s2,$s1,$s0,$s3,1); + &deccompact(3,$tbl,$s3,$s2,$s1,$s0,1); + &dectransform(2); + &dectransform(3); + &dectransform(0); + &dectransform(1); + &mov ($key,$__key); + &mov ($tbl,$__tbl); + &add ($key,16); # advance rd_key + &xor ($s0,&DWP(0,$key)); + &xor ($s1,&DWP(4,$key)); + &xor ($s2,&DWP(8,$key)); + &xor ($s3,&DWP(12,$key)); + + &cmp ($key,$__end); + &mov ($__key,$key); + &jb (&label("loop")); + + &deccompact(0,$tbl,$s0,$s3,$s2,$s1); + &deccompact(1,$tbl,$s1,$s0,$s3,$s2); + &deccompact(2,$tbl,$s2,$s1,$s0,$s3); + &deccompact(3,$tbl,$s3,$s2,$s1,$s0); + + &xor ($s0,&DWP(16,$key)); + &xor ($s1,&DWP(20,$key)); + &xor ($s2,&DWP(24,$key)); + &xor ($s3,&DWP(28,$key)); + + &ret (); +&function_end_B("_x86_AES_decrypt_compact"); + +###################################################################### +# "Compact" SSE block function. +###################################################################### + +sub sse_deccompact() +{ + &pshufw ("mm1","mm0",0x0c); # 7, 6, 1, 0 + &movd ("eax","mm1"); # 7, 6, 1, 0 + + &pshufw ("mm5","mm4",0x09); # 13,12,11,10 + &movz ($acc,&LB("eax")); # 0 + &movz ("ecx",&BP(-128,$tbl,$acc,1)); # 0 + &movd ("ebx","mm5"); # 13,12,11,10 + &movz ("edx",&HB("eax")); # 1 + &movz ("edx",&BP(-128,$tbl,"edx",1)); # 1 + &shl ("edx",8); # 1 + + &pshufw ("mm2","mm0",0x06); # 3, 2, 5, 4 + &movz ($acc,&LB("ebx")); # 10 + &movz ($acc,&BP(-128,$tbl,$acc,1)); # 10 + &shl ($acc,16); # 10 + &or ("ecx",$acc); # 10 + &shr ("eax",16); # 7, 6 + &movz ($acc,&HB("ebx")); # 11 + &movz ($acc,&BP(-128,$tbl,$acc,1)); # 11 + &shl ($acc,24); # 11 + &or ("edx",$acc); # 11 + &shr ("ebx",16); # 13,12 + + &pshufw ("mm6","mm4",0x03); # 9, 8,15,14 + &movz ($acc,&HB("eax")); # 7 + &movz ($acc,&BP(-128,$tbl,$acc,1)); # 7 + &shl ($acc,24); # 7 + &or ("ecx",$acc); # 7 + &movz ($acc,&HB("ebx")); # 13 + &movz ($acc,&BP(-128,$tbl,$acc,1)); # 13 + &shl ($acc,8); # 13 + &or ("ecx",$acc); # 13 + &movd ("mm0","ecx"); # t[0] collected + + &movz ($acc,&LB("eax")); # 6 + &movd ("eax","mm2"); # 3, 2, 5, 4 + &movz ("ecx",&BP(-128,$tbl,$acc,1)); # 6 + &shl ("ecx",16); # 6 + &movz ($acc,&LB("ebx")); # 12 + &movd ("ebx","mm6"); # 9, 8,15,14 + &movz ($acc,&BP(-128,$tbl,$acc,1)); # 12 + &or ("ecx",$acc); # 12 + + &movz ($acc,&LB("eax")); # 4 + &movz ($acc,&BP(-128,$tbl,$acc,1)); # 4 + &or ("edx",$acc); # 4 + &movz ($acc,&LB("ebx")); # 14 + &movz ($acc,&BP(-128,$tbl,$acc,1)); # 14 + &shl ($acc,16); # 14 + &or ("edx",$acc); # 14 + &movd ("mm1","edx"); # t[1] collected + + &movz ($acc,&HB("eax")); # 5 + &movz ("edx",&BP(-128,$tbl,$acc,1)); # 5 + &shl ("edx",8); # 5 + &movz ($acc,&HB("ebx")); # 15 + &shr ("eax",16); # 3, 2 + &movz ($acc,&BP(-128,$tbl,$acc,1)); # 15 + &shl ($acc,24); # 15 + &or ("edx",$acc); # 15 + &shr ("ebx",16); # 9, 8 + + &punpckldq ("mm0","mm1"); # t[0,1] collected + + &movz ($acc,&HB("ebx")); # 9 + &movz ($acc,&BP(-128,$tbl,$acc,1)); # 9 + &shl ($acc,8); # 9 + &or ("ecx",$acc); # 9 + &and ("ebx",0xff); # 8 + &movz ("ebx",&BP(-128,$tbl,"ebx",1)); # 8 + &or ("edx","ebx"); # 8 + &movz ($acc,&LB("eax")); # 2 + &movz ($acc,&BP(-128,$tbl,$acc,1)); # 2 + &shl ($acc,16); # 2 + &or ("edx",$acc); # 2 + &movd ("mm4","edx"); # t[2] collected + &movz ("eax",&HB("eax")); # 3 + &movz ("eax",&BP(-128,$tbl,"eax",1)); # 3 + &shl ("eax",24); # 3 + &or ("ecx","eax"); # 3 + &movd ("mm5","ecx"); # t[3] collected + + &punpckldq ("mm4","mm5"); # t[2,3] collected +} + + if (!$x86only) { +&function_begin_B("_sse_AES_decrypt_compact"); + &pxor ("mm0",&QWP(0,$key)); # 7, 6, 5, 4, 3, 2, 1, 0 + &pxor ("mm4",&QWP(8,$key)); # 15,14,13,12,11,10, 9, 8 + + # note that caller is expected to allocate stack frame for me! + &mov ($acc,&DWP(240,$key)); # load key->rounds + &lea ($acc,&DWP(-2,$acc,$acc)); + &lea ($acc,&DWP(0,$key,$acc,8)); + &mov ($__end,$acc); # end of key schedule + + &mov ($s0,0x1b1b1b1b); # magic constant + &mov (&DWP(8,"esp"),$s0); + &mov (&DWP(12,"esp"),$s0); + + # prefetch Td4 + &mov ($s0,&DWP(0-128,$tbl)); + &mov ($s1,&DWP(32-128,$tbl)); + &mov ($s2,&DWP(64-128,$tbl)); + &mov ($s3,&DWP(96-128,$tbl)); + &mov ($s0,&DWP(128-128,$tbl)); + &mov ($s1,&DWP(160-128,$tbl)); + &mov ($s2,&DWP(192-128,$tbl)); + &mov ($s3,&DWP(224-128,$tbl)); + + &set_label("loop",16); + &sse_deccompact(); + &add ($key,16); + &cmp ($key,$__end); + &ja (&label("out")); + + # ROTATE(x^y,N) == ROTATE(x,N)^ROTATE(y,N) + &movq ("mm3","mm0"); &movq ("mm7","mm4"); + &movq ("mm2","mm0",1); &movq ("mm6","mm4",1); + &movq ("mm1","mm0"); &movq ("mm5","mm4"); + &pshufw ("mm0","mm0",0xb1); &pshufw ("mm4","mm4",0xb1);# = ROTATE(tp0,16) + &pslld ("mm2",8); &pslld ("mm6",8); + &psrld ("mm3",8); &psrld ("mm7",8); + &pxor ("mm0","mm2"); &pxor ("mm4","mm6"); # ^= tp0<<8 + &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= tp0>>8 + &pslld ("mm2",16); &pslld ("mm6",16); + &psrld ("mm3",16); &psrld ("mm7",16); + &pxor ("mm0","mm2"); &pxor ("mm4","mm6"); # ^= tp0<<24 + &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= tp0>>24 + + &movq ("mm3",&QWP(8,"esp")); + &pxor ("mm2","mm2"); &pxor ("mm6","mm6"); + &pcmpgtb("mm2","mm1"); &pcmpgtb("mm6","mm5"); + &pand ("mm2","mm3"); &pand ("mm6","mm3"); + &paddb ("mm1","mm1"); &paddb ("mm5","mm5"); + &pxor ("mm1","mm2"); &pxor ("mm5","mm6"); # tp2 + &movq ("mm3","mm1"); &movq ("mm7","mm5"); + &movq ("mm2","mm1"); &movq ("mm6","mm5"); + &pxor ("mm0","mm1"); &pxor ("mm4","mm5"); # ^= tp2 + &pslld ("mm3",24); &pslld ("mm7",24); + &psrld ("mm2",8); &psrld ("mm6",8); + &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= tp2<<24 + &pxor ("mm0","mm2"); &pxor ("mm4","mm6"); # ^= tp2>>8 + + &movq ("mm2",&QWP(8,"esp")); + &pxor ("mm3","mm3"); &pxor ("mm7","mm7"); + &pcmpgtb("mm3","mm1"); &pcmpgtb("mm7","mm5"); + &pand ("mm3","mm2"); &pand ("mm7","mm2"); + &paddb ("mm1","mm1"); &paddb ("mm5","mm5"); + &pxor ("mm1","mm3"); &pxor ("mm5","mm7"); # tp4 + &pshufw ("mm3","mm1",0xb1); &pshufw ("mm7","mm5",0xb1); + &pxor ("mm0","mm1"); &pxor ("mm4","mm5"); # ^= tp4 + &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= ROTATE(tp4,16) + + &pxor ("mm3","mm3"); &pxor ("mm7","mm7"); + &pcmpgtb("mm3","mm1"); &pcmpgtb("mm7","mm5"); + &pand ("mm3","mm2"); &pand ("mm7","mm2"); + &paddb ("mm1","mm1"); &paddb ("mm5","mm5"); + &pxor ("mm1","mm3"); &pxor ("mm5","mm7"); # tp8 + &pxor ("mm0","mm1"); &pxor ("mm4","mm5"); # ^= tp8 + &movq ("mm3","mm1"); &movq ("mm7","mm5"); + &pshufw ("mm2","mm1",0xb1); &pshufw ("mm6","mm5",0xb1); + &pxor ("mm0","mm2"); &pxor ("mm4","mm6"); # ^= ROTATE(tp8,16) + &pslld ("mm1",8); &pslld ("mm5",8); + &psrld ("mm3",8); &psrld ("mm7",8); + &movq ("mm2",&QWP(0,$key)); &movq ("mm6",&QWP(8,$key)); + &pxor ("mm0","mm1"); &pxor ("mm4","mm5"); # ^= tp8<<8 + &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= tp8>>8 + &mov ($s0,&DWP(0-128,$tbl)); + &pslld ("mm1",16); &pslld ("mm5",16); + &mov ($s1,&DWP(64-128,$tbl)); + &psrld ("mm3",16); &psrld ("mm7",16); + &mov ($s2,&DWP(128-128,$tbl)); + &pxor ("mm0","mm1"); &pxor ("mm4","mm5"); # ^= tp8<<24 + &mov ($s3,&DWP(192-128,$tbl)); + &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= tp8>>24 + + &pxor ("mm0","mm2"); &pxor ("mm4","mm6"); + &jmp (&label("loop")); + + &set_label("out",16); + &pxor ("mm0",&QWP(0,$key)); + &pxor ("mm4",&QWP(8,$key)); + + &ret (); +&function_end_B("_sse_AES_decrypt_compact"); + } + +###################################################################### +# Vanilla block function. +###################################################################### sub decstep() { my ($i,$td,@s) = @_; @@ -717,27 +1589,27 @@ sub decstep() # optimal... or rather that all attempts to reorder didn't # result in better performance [which by the way is not a # bit lower than ecryption]. - if($i==3) { &mov ($key,&DWP(12,"esp")); } + if($i==3) { &mov ($key,$__key); } else { &mov ($out,$s[0]); } &and ($out,0xFF); - &mov ($out,&DWP(1024*0,$td,$out,4)); + &mov ($out,&DWP(0,$td,$out,8)); if ($i==3) { $tmp=$s[1]; } &movz ($tmp,&HB($s[1])); - &xor ($out,&DWP(1024*1,$td,$tmp,4)); + &xor ($out,&DWP(3,$td,$tmp,8)); if ($i==3) { $tmp=$s[2]; &mov ($s[1],$acc); } else { &mov ($tmp,$s[2]); } &shr ($tmp,16); &and ($tmp,0xFF); - &xor ($out,&DWP(1024*2,$td,$tmp,4)); + &xor ($out,&DWP(2,$td,$tmp,8)); - if ($i==3) { $tmp=$s[3]; &mov ($s[2],&DWP(4,"esp")); } + if ($i==3) { $tmp=$s[3]; &mov ($s[2],$__s1); } else { &mov ($tmp,$s[3]); } &shr ($tmp,24); - &xor ($out,&DWP(1024*3,$td,$tmp,4)); - if ($i<2) { &mov (&DWP(4*$i,"esp"),$out); } - if ($i==3) { &mov ($s[3],&DWP(0,"esp")); } + &xor ($out,&DWP(1,$td,$tmp,8)); + if ($i<2) { &mov (&DWP(4+4*$i,"esp"),$out); } + if ($i==3) { &mov ($s[3],$__s0); } &comment(); } @@ -746,45 +1618,49 @@ sub declast() my $tmp = $key; my $out = $i==3?$s[0]:$acc; - if($i==3) { &mov ($key,&DWP(12,"esp")); } + if($i==0) { &lea ($td,&DWP(2048+128,$td)); + &mov ($tmp,&DWP(0-128,$td)); + &mov ($acc,&DWP(32-128,$td)); + &mov ($tmp,&DWP(64-128,$td)); + &mov ($acc,&DWP(96-128,$td)); + &mov ($tmp,&DWP(128-128,$td)); + &mov ($acc,&DWP(160-128,$td)); + &mov ($tmp,&DWP(192-128,$td)); + &mov ($acc,&DWP(224-128,$td)); + &lea ($td,&DWP(-128,$td)); } + if($i==3) { &mov ($key,$__key); } else { &mov ($out,$s[0]); } &and ($out,0xFF); - &mov ($out,&DWP(1024*4,$td,$out,4)); - &and ($out,0x000000ff); + &movz ($out,&BP(0,$td,$out,1)); if ($i==3) { $tmp=$s[1]; } &movz ($tmp,&HB($s[1])); - &mov ($tmp,&DWP(1024*4,$td,$tmp,4)); - &and ($tmp,0x0000ff00); + &movz ($tmp,&BP(0,$td,$tmp,1)); + &shl ($tmp,8); &xor ($out,$tmp); if ($i==3) { $tmp=$s[2]; &mov ($s[1],$acc); } else { mov ($tmp,$s[2]); } &shr ($tmp,16); &and ($tmp,0xFF); - &mov ($tmp,&DWP(1024*4,$td,$tmp,4)); - &and ($tmp,0x00ff0000); + &movz ($tmp,&BP(0,$td,$tmp,1)); + &shl ($tmp,16); &xor ($out,$tmp); - if ($i==3) { $tmp=$s[3]; &mov ($s[2],&DWP(4,"esp")); } + if ($i==3) { $tmp=$s[3]; &mov ($s[2],$__s1); } else { &mov ($tmp,$s[3]); } &shr ($tmp,24); - &mov ($tmp,&DWP(1024*4,$td,$tmp,4)); - &and ($tmp,0xff000000); + &movz ($tmp,&BP(0,$td,$tmp,1)); + &shl ($tmp,24); &xor ($out,$tmp); - if ($i<2) { &mov (&DWP(4*$i,"esp"),$out); } - if ($i==3) { &mov ($s[3],&DWP(0,"esp")); } + if ($i<2) { &mov (&DWP(4+4*$i,"esp"),$out); } + if ($i==3) { &mov ($s[3],$__s0); + &lea ($td,&DWP(-2048,$td)); } } -&public_label("AES_Td"); &function_begin_B("_x86_AES_decrypt"); - # allocate aligned stack frame - &mov ($acc,"esp"); - &sub ("esp",20); - &and ("esp",-16); - - &mov (&DWP(12,"esp"),$key); # save key - &mov (&DWP(16,"esp"),$acc); # save %esp + # note that caller is expected to allocate stack frame for me! + &mov ($__key,$key); # save key &xor ($s0,&DWP(0,$key)); # xor with key &xor ($s1,&DWP(4,$key)); @@ -796,20 +1672,19 @@ sub declast() if ($small_footprint) { &lea ($acc,&DWP(-2,$acc,$acc)); &lea ($acc,&DWP(0,$key,$acc,8)); - &mov (&DWP(8,"esp"),$acc); # end of key schedule - &align (4); - &set_label("loop"); - &decstep(0,"ebp",$s0,$s3,$s2,$s1); - &decstep(1,"ebp",$s1,$s0,$s3,$s2); - &decstep(2,"ebp",$s2,$s1,$s0,$s3); - &decstep(3,"ebp",$s3,$s2,$s1,$s0); - &add ($key,16); # advance rd_key + &mov ($__end,$acc); # end of key schedule + &set_label("loop",16); + &decstep(0,$tbl,$s0,$s3,$s2,$s1); + &decstep(1,$tbl,$s1,$s0,$s3,$s2); + &decstep(2,$tbl,$s2,$s1,$s0,$s3); + &decstep(3,$tbl,$s3,$s2,$s1,$s0); + &add ($key,16); # advance rd_key &xor ($s0,&DWP(0,$key)); &xor ($s1,&DWP(4,$key)); &xor ($s2,&DWP(8,$key)); &xor ($s3,&DWP(12,$key)); - &cmp ($key,&DWP(8,"esp")); - &mov (&DWP(12,"esp"),$key); + &cmp ($key,$__end); + &mov ($__key,$key); &jb (&label("loop")); } else { @@ -818,38 +1693,38 @@ sub declast() &cmp ($acc,12); &jle (&label("12rounds")); - &set_label("14rounds"); + &set_label("14rounds",4); for ($i=1;$i<3;$i++) { - &decstep(0,"ebp",$s0,$s3,$s2,$s1); - &decstep(1,"ebp",$s1,$s0,$s3,$s2); - &decstep(2,"ebp",$s2,$s1,$s0,$s3); - &decstep(3,"ebp",$s3,$s2,$s1,$s0); + &decstep(0,$tbl,$s0,$s3,$s2,$s1); + &decstep(1,$tbl,$s1,$s0,$s3,$s2); + &decstep(2,$tbl,$s2,$s1,$s0,$s3); + &decstep(3,$tbl,$s3,$s2,$s1,$s0); &xor ($s0,&DWP(16*$i+0,$key)); &xor ($s1,&DWP(16*$i+4,$key)); &xor ($s2,&DWP(16*$i+8,$key)); &xor ($s3,&DWP(16*$i+12,$key)); } &add ($key,32); - &mov (&DWP(12,"esp"),$key); # advance rd_key - &set_label("12rounds"); + &mov ($__key,$key); # advance rd_key + &set_label("12rounds",4); for ($i=1;$i<3;$i++) { - &decstep(0,"ebp",$s0,$s3,$s2,$s1); - &decstep(1,"ebp",$s1,$s0,$s3,$s2); - &decstep(2,"ebp",$s2,$s1,$s0,$s3); - &decstep(3,"ebp",$s3,$s2,$s1,$s0); + &decstep(0,$tbl,$s0,$s3,$s2,$s1); + &decstep(1,$tbl,$s1,$s0,$s3,$s2); + &decstep(2,$tbl,$s2,$s1,$s0,$s3); + &decstep(3,$tbl,$s3,$s2,$s1,$s0); &xor ($s0,&DWP(16*$i+0,$key)); &xor ($s1,&DWP(16*$i+4,$key)); &xor ($s2,&DWP(16*$i+8,$key)); &xor ($s3,&DWP(16*$i+12,$key)); } &add ($key,32); - &mov (&DWP(12,"esp"),$key); # advance rd_key - &set_label("10rounds"); + &mov ($__key,$key); # advance rd_key + &set_label("10rounds",4); for ($i=1;$i<10;$i++) { - &decstep(0,"ebp",$s0,$s3,$s2,$s1); - &decstep(1,"ebp",$s1,$s0,$s3,$s2); - &decstep(2,"ebp",$s2,$s1,$s0,$s3); - &decstep(3,"ebp",$s3,$s2,$s1,$s0); + &decstep(0,$tbl,$s0,$s3,$s2,$s1); + &decstep(1,$tbl,$s1,$s0,$s3,$s2); + &decstep(2,$tbl,$s2,$s1,$s0,$s3); + &decstep(3,$tbl,$s3,$s2,$s1,$s0); &xor ($s0,&DWP(16*$i+0,$key)); &xor ($s1,&DWP(16*$i+4,$key)); &xor ($s2,&DWP(16*$i+8,$key)); @@ -857,12 +1732,11 @@ sub declast() } } - &declast(0,"ebp",$s0,$s3,$s2,$s1); - &declast(1,"ebp",$s1,$s0,$s3,$s2); - &declast(2,"ebp",$s2,$s1,$s0,$s3); - &declast(3,"ebp",$s3,$s2,$s1,$s0); + &declast(0,$tbl,$s0,$s3,$s2,$s1); + &declast(1,$tbl,$s1,$s0,$s3,$s2); + &declast(2,$tbl,$s2,$s1,$s0,$s3); + &declast(3,$tbl,$s3,$s2,$s1,$s0); - &mov ("esp",&DWP(16,"esp")); # restore %esp &add ($key,$small_footprint?16:160); &xor ($s0,&DWP(0,$key)); &xor ($s1,&DWP(4,$key)); @@ -872,350 +1746,257 @@ sub declast() &ret (); &set_label("AES_Td",64); # Yes! I keep it in the code segment! - &data_word(0x50a7f451, 0x5365417e, 0xc3a4171a, 0x965e273a); - &data_word(0xcb6bab3b, 0xf1459d1f, 0xab58faac, 0x9303e34b); - &data_word(0x55fa3020, 0xf66d76ad, 0x9176cc88, 0x254c02f5); - &data_word(0xfcd7e54f, 0xd7cb2ac5, 0x80443526, 0x8fa362b5); - &data_word(0x495ab1de, 0x671bba25, 0x980eea45, 0xe1c0fe5d); - &data_word(0x02752fc3, 0x12f04c81, 0xa397468d, 0xc6f9d36b); - &data_word(0xe75f8f03, 0x959c9215, 0xeb7a6dbf, 0xda595295); - &data_word(0x2d83bed4, 0xd3217458, 0x2969e049, 0x44c8c98e); - &data_word(0x6a89c275, 0x78798ef4, 0x6b3e5899, 0xdd71b927); - &data_word(0xb64fe1be, 0x17ad88f0, 0x66ac20c9, 0xb43ace7d); - &data_word(0x184adf63, 0x82311ae5, 0x60335197, 0x457f5362); - &data_word(0xe07764b1, 0x84ae6bbb, 0x1ca081fe, 0x942b08f9); - &data_word(0x58684870, 0x19fd458f, 0x876cde94, 0xb7f87b52); - &data_word(0x23d373ab, 0xe2024b72, 0x578f1fe3, 0x2aab5566); - &data_word(0x0728ebb2, 0x03c2b52f, 0x9a7bc586, 0xa50837d3); - &data_word(0xf2872830, 0xb2a5bf23, 0xba6a0302, 0x5c8216ed); - &data_word(0x2b1ccf8a, 0x92b479a7, 0xf0f207f3, 0xa1e2694e); - &data_word(0xcdf4da65, 0xd5be0506, 0x1f6234d1, 0x8afea6c4); - &data_word(0x9d532e34, 0xa055f3a2, 0x32e18a05, 0x75ebf6a4); - &data_word(0x39ec830b, 0xaaef6040, 0x069f715e, 0x51106ebd); - &data_word(0xf98a213e, 0x3d06dd96, 0xae053edd, 0x46bde64d); - &data_word(0xb58d5491, 0x055dc471, 0x6fd40604, 0xff155060); - &data_word(0x24fb9819, 0x97e9bdd6, 0xcc434089, 0x779ed967); - &data_word(0xbd42e8b0, 0x888b8907, 0x385b19e7, 0xdbeec879); - &data_word(0x470a7ca1, 0xe90f427c, 0xc91e84f8, 0x00000000); - &data_word(0x83868009, 0x48ed2b32, 0xac70111e, 0x4e725a6c); - &data_word(0xfbff0efd, 0x5638850f, 0x1ed5ae3d, 0x27392d36); - &data_word(0x64d90f0a, 0x21a65c68, 0xd1545b9b, 0x3a2e3624); - &data_word(0xb1670a0c, 0x0fe75793, 0xd296eeb4, 0x9e919b1b); - &data_word(0x4fc5c080, 0xa220dc61, 0x694b775a, 0x161a121c); - &data_word(0x0aba93e2, 0xe52aa0c0, 0x43e0223c, 0x1d171b12); - &data_word(0x0b0d090e, 0xadc78bf2, 0xb9a8b62d, 0xc8a91e14); - &data_word(0x8519f157, 0x4c0775af, 0xbbdd99ee, 0xfd607fa3); - &data_word(0x9f2601f7, 0xbcf5725c, 0xc53b6644, 0x347efb5b); - &data_word(0x7629438b, 0xdcc623cb, 0x68fcedb6, 0x63f1e4b8); - &data_word(0xcadc31d7, 0x10856342, 0x40229713, 0x2011c684); - &data_word(0x7d244a85, 0xf83dbbd2, 0x1132f9ae, 0x6da129c7); - &data_word(0x4b2f9e1d, 0xf330b2dc, 0xec52860d, 0xd0e3c177); - &data_word(0x6c16b32b, 0x99b970a9, 0xfa489411, 0x2264e947); - &data_word(0xc48cfca8, 0x1a3ff0a0, 0xd82c7d56, 0xef903322); - &data_word(0xc74e4987, 0xc1d138d9, 0xfea2ca8c, 0x360bd498); - &data_word(0xcf81f5a6, 0x28de7aa5, 0x268eb7da, 0xa4bfad3f); - &data_word(0xe49d3a2c, 0x0d927850, 0x9bcc5f6a, 0x62467e54); - &data_word(0xc2138df6, 0xe8b8d890, 0x5ef7392e, 0xf5afc382); - &data_word(0xbe805d9f, 0x7c93d069, 0xa92dd56f, 0xb31225cf); - &data_word(0x3b99acc8, 0xa77d1810, 0x6e639ce8, 0x7bbb3bdb); - &data_word(0x097826cd, 0xf418596e, 0x01b79aec, 0xa89a4f83); - &data_word(0x656e95e6, 0x7ee6ffaa, 0x08cfbc21, 0xe6e815ef); - &data_word(0xd99be7ba, 0xce366f4a, 0xd4099fea, 0xd67cb029); - &data_word(0xafb2a431, 0x31233f2a, 0x3094a5c6, 0xc066a235); - &data_word(0x37bc4e74, 0xa6ca82fc, 0xb0d090e0, 0x15d8a733); - &data_word(0x4a9804f1, 0xf7daec41, 0x0e50cd7f, 0x2ff69117); - &data_word(0x8dd64d76, 0x4db0ef43, 0x544daacc, 0xdf0496e4); - &data_word(0xe3b5d19e, 0x1b886a4c, 0xb81f2cc1, 0x7f516546); - &data_word(0x04ea5e9d, 0x5d358c01, 0x737487fa, 0x2e410bfb); - &data_word(0x5a1d67b3, 0x52d2db92, 0x335610e9, 0x1347d66d); - &data_word(0x8c61d79a, 0x7a0ca137, 0x8e14f859, 0x893c13eb); - &data_word(0xee27a9ce, 0x35c961b7, 0xede51ce1, 0x3cb1477a); - &data_word(0x59dfd29c, 0x3f73f255, 0x79ce1418, 0xbf37c773); - &data_word(0xeacdf753, 0x5baafd5f, 0x146f3ddf, 0x86db4478); - &data_word(0x81f3afca, 0x3ec468b9, 0x2c342438, 0x5f40a3c2); - &data_word(0x72c31d16, 0x0c25e2bc, 0x8b493c28, 0x41950dff); - &data_word(0x7101a839, 0xdeb30c08, 0x9ce4b4d8, 0x90c15664); - &data_word(0x6184cb7b, 0x70b632d5, 0x745c6c48, 0x4257b8d0); -#Td1: - &data_word(0xa7f45150, 0x65417e53, 0xa4171ac3, 0x5e273a96); - &data_word(0x6bab3bcb, 0x459d1ff1, 0x58faacab, 0x03e34b93); - &data_word(0xfa302055, 0x6d76adf6, 0x76cc8891, 0x4c02f525); - &data_word(0xd7e54ffc, 0xcb2ac5d7, 0x44352680, 0xa362b58f); - &data_word(0x5ab1de49, 0x1bba2567, 0x0eea4598, 0xc0fe5de1); - &data_word(0x752fc302, 0xf04c8112, 0x97468da3, 0xf9d36bc6); - &data_word(0x5f8f03e7, 0x9c921595, 0x7a6dbfeb, 0x595295da); - &data_word(0x83bed42d, 0x217458d3, 0x69e04929, 0xc8c98e44); - &data_word(0x89c2756a, 0x798ef478, 0x3e58996b, 0x71b927dd); - &data_word(0x4fe1beb6, 0xad88f017, 0xac20c966, 0x3ace7db4); - &data_word(0x4adf6318, 0x311ae582, 0x33519760, 0x7f536245); - &data_word(0x7764b1e0, 0xae6bbb84, 0xa081fe1c, 0x2b08f994); - &data_word(0x68487058, 0xfd458f19, 0x6cde9487, 0xf87b52b7); - &data_word(0xd373ab23, 0x024b72e2, 0x8f1fe357, 0xab55662a); - &data_word(0x28ebb207, 0xc2b52f03, 0x7bc5869a, 0x0837d3a5); - &data_word(0x872830f2, 0xa5bf23b2, 0x6a0302ba, 0x8216ed5c); - &data_word(0x1ccf8a2b, 0xb479a792, 0xf207f3f0, 0xe2694ea1); - &data_word(0xf4da65cd, 0xbe0506d5, 0x6234d11f, 0xfea6c48a); - &data_word(0x532e349d, 0x55f3a2a0, 0xe18a0532, 0xebf6a475); - &data_word(0xec830b39, 0xef6040aa, 0x9f715e06, 0x106ebd51); - &data_word(0x8a213ef9, 0x06dd963d, 0x053eddae, 0xbde64d46); - &data_word(0x8d5491b5, 0x5dc47105, 0xd406046f, 0x155060ff); - &data_word(0xfb981924, 0xe9bdd697, 0x434089cc, 0x9ed96777); - &data_word(0x42e8b0bd, 0x8b890788, 0x5b19e738, 0xeec879db); - &data_word(0x0a7ca147, 0x0f427ce9, 0x1e84f8c9, 0x00000000); - &data_word(0x86800983, 0xed2b3248, 0x70111eac, 0x725a6c4e); - &data_word(0xff0efdfb, 0x38850f56, 0xd5ae3d1e, 0x392d3627); - &data_word(0xd90f0a64, 0xa65c6821, 0x545b9bd1, 0x2e36243a); - &data_word(0x670a0cb1, 0xe757930f, 0x96eeb4d2, 0x919b1b9e); - &data_word(0xc5c0804f, 0x20dc61a2, 0x4b775a69, 0x1a121c16); - &data_word(0xba93e20a, 0x2aa0c0e5, 0xe0223c43, 0x171b121d); - &data_word(0x0d090e0b, 0xc78bf2ad, 0xa8b62db9, 0xa91e14c8); - &data_word(0x19f15785, 0x0775af4c, 0xdd99eebb, 0x607fa3fd); - &data_word(0x2601f79f, 0xf5725cbc, 0x3b6644c5, 0x7efb5b34); - &data_word(0x29438b76, 0xc623cbdc, 0xfcedb668, 0xf1e4b863); - &data_word(0xdc31d7ca, 0x85634210, 0x22971340, 0x11c68420); - &data_word(0x244a857d, 0x3dbbd2f8, 0x32f9ae11, 0xa129c76d); - &data_word(0x2f9e1d4b, 0x30b2dcf3, 0x52860dec, 0xe3c177d0); - &data_word(0x16b32b6c, 0xb970a999, 0x489411fa, 0x64e94722); - &data_word(0x8cfca8c4, 0x3ff0a01a, 0x2c7d56d8, 0x903322ef); - &data_word(0x4e4987c7, 0xd138d9c1, 0xa2ca8cfe, 0x0bd49836); - &data_word(0x81f5a6cf, 0xde7aa528, 0x8eb7da26, 0xbfad3fa4); - &data_word(0x9d3a2ce4, 0x9278500d, 0xcc5f6a9b, 0x467e5462); - &data_word(0x138df6c2, 0xb8d890e8, 0xf7392e5e, 0xafc382f5); - &data_word(0x805d9fbe, 0x93d0697c, 0x2dd56fa9, 0x1225cfb3); - &data_word(0x99acc83b, 0x7d1810a7, 0x639ce86e, 0xbb3bdb7b); - &data_word(0x7826cd09, 0x18596ef4, 0xb79aec01, 0x9a4f83a8); - &data_word(0x6e95e665, 0xe6ffaa7e, 0xcfbc2108, 0xe815efe6); - &data_word(0x9be7bad9, 0x366f4ace, 0x099fead4, 0x7cb029d6); - &data_word(0xb2a431af, 0x233f2a31, 0x94a5c630, 0x66a235c0); - &data_word(0xbc4e7437, 0xca82fca6, 0xd090e0b0, 0xd8a73315); - &data_word(0x9804f14a, 0xdaec41f7, 0x50cd7f0e, 0xf691172f); - &data_word(0xd64d768d, 0xb0ef434d, 0x4daacc54, 0x0496e4df); - &data_word(0xb5d19ee3, 0x886a4c1b, 0x1f2cc1b8, 0x5165467f); - &data_word(0xea5e9d04, 0x358c015d, 0x7487fa73, 0x410bfb2e); - &data_word(0x1d67b35a, 0xd2db9252, 0x5610e933, 0x47d66d13); - &data_word(0x61d79a8c, 0x0ca1377a, 0x14f8598e, 0x3c13eb89); - &data_word(0x27a9ceee, 0xc961b735, 0xe51ce1ed, 0xb1477a3c); - &data_word(0xdfd29c59, 0x73f2553f, 0xce141879, 0x37c773bf); - &data_word(0xcdf753ea, 0xaafd5f5b, 0x6f3ddf14, 0xdb447886); - &data_word(0xf3afca81, 0xc468b93e, 0x3424382c, 0x40a3c25f); - &data_word(0xc31d1672, 0x25e2bc0c, 0x493c288b, 0x950dff41); - &data_word(0x01a83971, 0xb30c08de, 0xe4b4d89c, 0xc1566490); - &data_word(0x84cb7b61, 0xb632d570, 0x5c6c4874, 0x57b8d042); -#Td2: - &data_word(0xf45150a7, 0x417e5365, 0x171ac3a4, 0x273a965e); - &data_word(0xab3bcb6b, 0x9d1ff145, 0xfaacab58, 0xe34b9303); - &data_word(0x302055fa, 0x76adf66d, 0xcc889176, 0x02f5254c); - &data_word(0xe54ffcd7, 0x2ac5d7cb, 0x35268044, 0x62b58fa3); - &data_word(0xb1de495a, 0xba25671b, 0xea45980e, 0xfe5de1c0); - &data_word(0x2fc30275, 0x4c8112f0, 0x468da397, 0xd36bc6f9); - &data_word(0x8f03e75f, 0x9215959c, 0x6dbfeb7a, 0x5295da59); - &data_word(0xbed42d83, 0x7458d321, 0xe0492969, 0xc98e44c8); - &data_word(0xc2756a89, 0x8ef47879, 0x58996b3e, 0xb927dd71); - &data_word(0xe1beb64f, 0x88f017ad, 0x20c966ac, 0xce7db43a); - &data_word(0xdf63184a, 0x1ae58231, 0x51976033, 0x5362457f); - &data_word(0x64b1e077, 0x6bbb84ae, 0x81fe1ca0, 0x08f9942b); - &data_word(0x48705868, 0x458f19fd, 0xde94876c, 0x7b52b7f8); - &data_word(0x73ab23d3, 0x4b72e202, 0x1fe3578f, 0x55662aab); - &data_word(0xebb20728, 0xb52f03c2, 0xc5869a7b, 0x37d3a508); - &data_word(0x2830f287, 0xbf23b2a5, 0x0302ba6a, 0x16ed5c82); - &data_word(0xcf8a2b1c, 0x79a792b4, 0x07f3f0f2, 0x694ea1e2); - &data_word(0xda65cdf4, 0x0506d5be, 0x34d11f62, 0xa6c48afe); - &data_word(0x2e349d53, 0xf3a2a055, 0x8a0532e1, 0xf6a475eb); - &data_word(0x830b39ec, 0x6040aaef, 0x715e069f, 0x6ebd5110); - &data_word(0x213ef98a, 0xdd963d06, 0x3eddae05, 0xe64d46bd); - &data_word(0x5491b58d, 0xc471055d, 0x06046fd4, 0x5060ff15); - &data_word(0x981924fb, 0xbdd697e9, 0x4089cc43, 0xd967779e); - &data_word(0xe8b0bd42, 0x8907888b, 0x19e7385b, 0xc879dbee); - &data_word(0x7ca1470a, 0x427ce90f, 0x84f8c91e, 0x00000000); - &data_word(0x80098386, 0x2b3248ed, 0x111eac70, 0x5a6c4e72); - &data_word(0x0efdfbff, 0x850f5638, 0xae3d1ed5, 0x2d362739); - &data_word(0x0f0a64d9, 0x5c6821a6, 0x5b9bd154, 0x36243a2e); - &data_word(0x0a0cb167, 0x57930fe7, 0xeeb4d296, 0x9b1b9e91); - &data_word(0xc0804fc5, 0xdc61a220, 0x775a694b, 0x121c161a); - &data_word(0x93e20aba, 0xa0c0e52a, 0x223c43e0, 0x1b121d17); - &data_word(0x090e0b0d, 0x8bf2adc7, 0xb62db9a8, 0x1e14c8a9); - &data_word(0xf1578519, 0x75af4c07, 0x99eebbdd, 0x7fa3fd60); - &data_word(0x01f79f26, 0x725cbcf5, 0x6644c53b, 0xfb5b347e); - &data_word(0x438b7629, 0x23cbdcc6, 0xedb668fc, 0xe4b863f1); - &data_word(0x31d7cadc, 0x63421085, 0x97134022, 0xc6842011); - &data_word(0x4a857d24, 0xbbd2f83d, 0xf9ae1132, 0x29c76da1); - &data_word(0x9e1d4b2f, 0xb2dcf330, 0x860dec52, 0xc177d0e3); - &data_word(0xb32b6c16, 0x70a999b9, 0x9411fa48, 0xe9472264); - &data_word(0xfca8c48c, 0xf0a01a3f, 0x7d56d82c, 0x3322ef90); - &data_word(0x4987c74e, 0x38d9c1d1, 0xca8cfea2, 0xd498360b); - &data_word(0xf5a6cf81, 0x7aa528de, 0xb7da268e, 0xad3fa4bf); - &data_word(0x3a2ce49d, 0x78500d92, 0x5f6a9bcc, 0x7e546246); - &data_word(0x8df6c213, 0xd890e8b8, 0x392e5ef7, 0xc382f5af); - &data_word(0x5d9fbe80, 0xd0697c93, 0xd56fa92d, 0x25cfb312); - &data_word(0xacc83b99, 0x1810a77d, 0x9ce86e63, 0x3bdb7bbb); - &data_word(0x26cd0978, 0x596ef418, 0x9aec01b7, 0x4f83a89a); - &data_word(0x95e6656e, 0xffaa7ee6, 0xbc2108cf, 0x15efe6e8); - &data_word(0xe7bad99b, 0x6f4ace36, 0x9fead409, 0xb029d67c); - &data_word(0xa431afb2, 0x3f2a3123, 0xa5c63094, 0xa235c066); - &data_word(0x4e7437bc, 0x82fca6ca, 0x90e0b0d0, 0xa73315d8); - &data_word(0x04f14a98, 0xec41f7da, 0xcd7f0e50, 0x91172ff6); - &data_word(0x4d768dd6, 0xef434db0, 0xaacc544d, 0x96e4df04); - &data_word(0xd19ee3b5, 0x6a4c1b88, 0x2cc1b81f, 0x65467f51); - &data_word(0x5e9d04ea, 0x8c015d35, 0x87fa7374, 0x0bfb2e41); - &data_word(0x67b35a1d, 0xdb9252d2, 0x10e93356, 0xd66d1347); - &data_word(0xd79a8c61, 0xa1377a0c, 0xf8598e14, 0x13eb893c); - &data_word(0xa9ceee27, 0x61b735c9, 0x1ce1ede5, 0x477a3cb1); - &data_word(0xd29c59df, 0xf2553f73, 0x141879ce, 0xc773bf37); - &data_word(0xf753eacd, 0xfd5f5baa, 0x3ddf146f, 0x447886db); - &data_word(0xafca81f3, 0x68b93ec4, 0x24382c34, 0xa3c25f40); - &data_word(0x1d1672c3, 0xe2bc0c25, 0x3c288b49, 0x0dff4195); - &data_word(0xa8397101, 0x0c08deb3, 0xb4d89ce4, 0x566490c1); - &data_word(0xcb7b6184, 0x32d570b6, 0x6c48745c, 0xb8d04257); -#Td3: - &data_word(0x5150a7f4, 0x7e536541, 0x1ac3a417, 0x3a965e27); - &data_word(0x3bcb6bab, 0x1ff1459d, 0xacab58fa, 0x4b9303e3); - &data_word(0x2055fa30, 0xadf66d76, 0x889176cc, 0xf5254c02); - &data_word(0x4ffcd7e5, 0xc5d7cb2a, 0x26804435, 0xb58fa362); - &data_word(0xde495ab1, 0x25671bba, 0x45980eea, 0x5de1c0fe); - &data_word(0xc302752f, 0x8112f04c, 0x8da39746, 0x6bc6f9d3); - &data_word(0x03e75f8f, 0x15959c92, 0xbfeb7a6d, 0x95da5952); - &data_word(0xd42d83be, 0x58d32174, 0x492969e0, 0x8e44c8c9); - &data_word(0x756a89c2, 0xf478798e, 0x996b3e58, 0x27dd71b9); - &data_word(0xbeb64fe1, 0xf017ad88, 0xc966ac20, 0x7db43ace); - &data_word(0x63184adf, 0xe582311a, 0x97603351, 0x62457f53); - &data_word(0xb1e07764, 0xbb84ae6b, 0xfe1ca081, 0xf9942b08); - &data_word(0x70586848, 0x8f19fd45, 0x94876cde, 0x52b7f87b); - &data_word(0xab23d373, 0x72e2024b, 0xe3578f1f, 0x662aab55); - &data_word(0xb20728eb, 0x2f03c2b5, 0x869a7bc5, 0xd3a50837); - &data_word(0x30f28728, 0x23b2a5bf, 0x02ba6a03, 0xed5c8216); - &data_word(0x8a2b1ccf, 0xa792b479, 0xf3f0f207, 0x4ea1e269); - &data_word(0x65cdf4da, 0x06d5be05, 0xd11f6234, 0xc48afea6); - &data_word(0x349d532e, 0xa2a055f3, 0x0532e18a, 0xa475ebf6); - &data_word(0x0b39ec83, 0x40aaef60, 0x5e069f71, 0xbd51106e); - &data_word(0x3ef98a21, 0x963d06dd, 0xddae053e, 0x4d46bde6); - &data_word(0x91b58d54, 0x71055dc4, 0x046fd406, 0x60ff1550); - &data_word(0x1924fb98, 0xd697e9bd, 0x89cc4340, 0x67779ed9); - &data_word(0xb0bd42e8, 0x07888b89, 0xe7385b19, 0x79dbeec8); - &data_word(0xa1470a7c, 0x7ce90f42, 0xf8c91e84, 0x00000000); - &data_word(0x09838680, 0x3248ed2b, 0x1eac7011, 0x6c4e725a); - &data_word(0xfdfbff0e, 0x0f563885, 0x3d1ed5ae, 0x3627392d); - &data_word(0x0a64d90f, 0x6821a65c, 0x9bd1545b, 0x243a2e36); - &data_word(0x0cb1670a, 0x930fe757, 0xb4d296ee, 0x1b9e919b); - &data_word(0x804fc5c0, 0x61a220dc, 0x5a694b77, 0x1c161a12); - &data_word(0xe20aba93, 0xc0e52aa0, 0x3c43e022, 0x121d171b); - &data_word(0x0e0b0d09, 0xf2adc78b, 0x2db9a8b6, 0x14c8a91e); - &data_word(0x578519f1, 0xaf4c0775, 0xeebbdd99, 0xa3fd607f); - &data_word(0xf79f2601, 0x5cbcf572, 0x44c53b66, 0x5b347efb); - &data_word(0x8b762943, 0xcbdcc623, 0xb668fced, 0xb863f1e4); - &data_word(0xd7cadc31, 0x42108563, 0x13402297, 0x842011c6); - &data_word(0x857d244a, 0xd2f83dbb, 0xae1132f9, 0xc76da129); - &data_word(0x1d4b2f9e, 0xdcf330b2, 0x0dec5286, 0x77d0e3c1); - &data_word(0x2b6c16b3, 0xa999b970, 0x11fa4894, 0x472264e9); - &data_word(0xa8c48cfc, 0xa01a3ff0, 0x56d82c7d, 0x22ef9033); - &data_word(0x87c74e49, 0xd9c1d138, 0x8cfea2ca, 0x98360bd4); - &data_word(0xa6cf81f5, 0xa528de7a, 0xda268eb7, 0x3fa4bfad); - &data_word(0x2ce49d3a, 0x500d9278, 0x6a9bcc5f, 0x5462467e); - &data_word(0xf6c2138d, 0x90e8b8d8, 0x2e5ef739, 0x82f5afc3); - &data_word(0x9fbe805d, 0x697c93d0, 0x6fa92dd5, 0xcfb31225); - &data_word(0xc83b99ac, 0x10a77d18, 0xe86e639c, 0xdb7bbb3b); - &data_word(0xcd097826, 0x6ef41859, 0xec01b79a, 0x83a89a4f); - &data_word(0xe6656e95, 0xaa7ee6ff, 0x2108cfbc, 0xefe6e815); - &data_word(0xbad99be7, 0x4ace366f, 0xead4099f, 0x29d67cb0); - &data_word(0x31afb2a4, 0x2a31233f, 0xc63094a5, 0x35c066a2); - &data_word(0x7437bc4e, 0xfca6ca82, 0xe0b0d090, 0x3315d8a7); - &data_word(0xf14a9804, 0x41f7daec, 0x7f0e50cd, 0x172ff691); - &data_word(0x768dd64d, 0x434db0ef, 0xcc544daa, 0xe4df0496); - &data_word(0x9ee3b5d1, 0x4c1b886a, 0xc1b81f2c, 0x467f5165); - &data_word(0x9d04ea5e, 0x015d358c, 0xfa737487, 0xfb2e410b); - &data_word(0xb35a1d67, 0x9252d2db, 0xe9335610, 0x6d1347d6); - &data_word(0x9a8c61d7, 0x377a0ca1, 0x598e14f8, 0xeb893c13); - &data_word(0xceee27a9, 0xb735c961, 0xe1ede51c, 0x7a3cb147); - &data_word(0x9c59dfd2, 0x553f73f2, 0x1879ce14, 0x73bf37c7); - &data_word(0x53eacdf7, 0x5f5baafd, 0xdf146f3d, 0x7886db44); - &data_word(0xca81f3af, 0xb93ec468, 0x382c3424, 0xc25f40a3); - &data_word(0x1672c31d, 0xbc0c25e2, 0x288b493c, 0xff41950d); - &data_word(0x397101a8, 0x08deb30c, 0xd89ce4b4, 0x6490c156); - &data_word(0x7b6184cb, 0xd570b632, 0x48745c6c, 0xd04257b8); -#Td4: - &data_word(0x52525252, 0x09090909, 0x6a6a6a6a, 0xd5d5d5d5); - &data_word(0x30303030, 0x36363636, 0xa5a5a5a5, 0x38383838); - &data_word(0xbfbfbfbf, 0x40404040, 0xa3a3a3a3, 0x9e9e9e9e); - &data_word(0x81818181, 0xf3f3f3f3, 0xd7d7d7d7, 0xfbfbfbfb); - &data_word(0x7c7c7c7c, 0xe3e3e3e3, 0x39393939, 0x82828282); - &data_word(0x9b9b9b9b, 0x2f2f2f2f, 0xffffffff, 0x87878787); - &data_word(0x34343434, 0x8e8e8e8e, 0x43434343, 0x44444444); - &data_word(0xc4c4c4c4, 0xdededede, 0xe9e9e9e9, 0xcbcbcbcb); - &data_word(0x54545454, 0x7b7b7b7b, 0x94949494, 0x32323232); - &data_word(0xa6a6a6a6, 0xc2c2c2c2, 0x23232323, 0x3d3d3d3d); - &data_word(0xeeeeeeee, 0x4c4c4c4c, 0x95959595, 0x0b0b0b0b); - &data_word(0x42424242, 0xfafafafa, 0xc3c3c3c3, 0x4e4e4e4e); - &data_word(0x08080808, 0x2e2e2e2e, 0xa1a1a1a1, 0x66666666); - &data_word(0x28282828, 0xd9d9d9d9, 0x24242424, 0xb2b2b2b2); - &data_word(0x76767676, 0x5b5b5b5b, 0xa2a2a2a2, 0x49494949); - &data_word(0x6d6d6d6d, 0x8b8b8b8b, 0xd1d1d1d1, 0x25252525); - &data_word(0x72727272, 0xf8f8f8f8, 0xf6f6f6f6, 0x64646464); - &data_word(0x86868686, 0x68686868, 0x98989898, 0x16161616); - &data_word(0xd4d4d4d4, 0xa4a4a4a4, 0x5c5c5c5c, 0xcccccccc); - &data_word(0x5d5d5d5d, 0x65656565, 0xb6b6b6b6, 0x92929292); - &data_word(0x6c6c6c6c, 0x70707070, 0x48484848, 0x50505050); - &data_word(0xfdfdfdfd, 0xedededed, 0xb9b9b9b9, 0xdadadada); - &data_word(0x5e5e5e5e, 0x15151515, 0x46464646, 0x57575757); - &data_word(0xa7a7a7a7, 0x8d8d8d8d, 0x9d9d9d9d, 0x84848484); - &data_word(0x90909090, 0xd8d8d8d8, 0xabababab, 0x00000000); - &data_word(0x8c8c8c8c, 0xbcbcbcbc, 0xd3d3d3d3, 0x0a0a0a0a); - &data_word(0xf7f7f7f7, 0xe4e4e4e4, 0x58585858, 0x05050505); - &data_word(0xb8b8b8b8, 0xb3b3b3b3, 0x45454545, 0x06060606); - &data_word(0xd0d0d0d0, 0x2c2c2c2c, 0x1e1e1e1e, 0x8f8f8f8f); - &data_word(0xcacacaca, 0x3f3f3f3f, 0x0f0f0f0f, 0x02020202); - &data_word(0xc1c1c1c1, 0xafafafaf, 0xbdbdbdbd, 0x03030303); - &data_word(0x01010101, 0x13131313, 0x8a8a8a8a, 0x6b6b6b6b); - &data_word(0x3a3a3a3a, 0x91919191, 0x11111111, 0x41414141); - &data_word(0x4f4f4f4f, 0x67676767, 0xdcdcdcdc, 0xeaeaeaea); - &data_word(0x97979797, 0xf2f2f2f2, 0xcfcfcfcf, 0xcececece); - &data_word(0xf0f0f0f0, 0xb4b4b4b4, 0xe6e6e6e6, 0x73737373); - &data_word(0x96969696, 0xacacacac, 0x74747474, 0x22222222); - &data_word(0xe7e7e7e7, 0xadadadad, 0x35353535, 0x85858585); - &data_word(0xe2e2e2e2, 0xf9f9f9f9, 0x37373737, 0xe8e8e8e8); - &data_word(0x1c1c1c1c, 0x75757575, 0xdfdfdfdf, 0x6e6e6e6e); - &data_word(0x47474747, 0xf1f1f1f1, 0x1a1a1a1a, 0x71717171); - &data_word(0x1d1d1d1d, 0x29292929, 0xc5c5c5c5, 0x89898989); - &data_word(0x6f6f6f6f, 0xb7b7b7b7, 0x62626262, 0x0e0e0e0e); - &data_word(0xaaaaaaaa, 0x18181818, 0xbebebebe, 0x1b1b1b1b); - &data_word(0xfcfcfcfc, 0x56565656, 0x3e3e3e3e, 0x4b4b4b4b); - &data_word(0xc6c6c6c6, 0xd2d2d2d2, 0x79797979, 0x20202020); - &data_word(0x9a9a9a9a, 0xdbdbdbdb, 0xc0c0c0c0, 0xfefefefe); - &data_word(0x78787878, 0xcdcdcdcd, 0x5a5a5a5a, 0xf4f4f4f4); - &data_word(0x1f1f1f1f, 0xdddddddd, 0xa8a8a8a8, 0x33333333); - &data_word(0x88888888, 0x07070707, 0xc7c7c7c7, 0x31313131); - &data_word(0xb1b1b1b1, 0x12121212, 0x10101010, 0x59595959); - &data_word(0x27272727, 0x80808080, 0xecececec, 0x5f5f5f5f); - &data_word(0x60606060, 0x51515151, 0x7f7f7f7f, 0xa9a9a9a9); - &data_word(0x19191919, 0xb5b5b5b5, 0x4a4a4a4a, 0x0d0d0d0d); - &data_word(0x2d2d2d2d, 0xe5e5e5e5, 0x7a7a7a7a, 0x9f9f9f9f); - &data_word(0x93939393, 0xc9c9c9c9, 0x9c9c9c9c, 0xefefefef); - &data_word(0xa0a0a0a0, 0xe0e0e0e0, 0x3b3b3b3b, 0x4d4d4d4d); - &data_word(0xaeaeaeae, 0x2a2a2a2a, 0xf5f5f5f5, 0xb0b0b0b0); - &data_word(0xc8c8c8c8, 0xebebebeb, 0xbbbbbbbb, 0x3c3c3c3c); - &data_word(0x83838383, 0x53535353, 0x99999999, 0x61616161); - &data_word(0x17171717, 0x2b2b2b2b, 0x04040404, 0x7e7e7e7e); - &data_word(0xbabababa, 0x77777777, 0xd6d6d6d6, 0x26262626); - &data_word(0xe1e1e1e1, 0x69696969, 0x14141414, 0x63636363); - &data_word(0x55555555, 0x21212121, 0x0c0c0c0c, 0x7d7d7d7d); + &_data_word(0x50a7f451, 0x5365417e, 0xc3a4171a, 0x965e273a); + &_data_word(0xcb6bab3b, 0xf1459d1f, 0xab58faac, 0x9303e34b); + &_data_word(0x55fa3020, 0xf66d76ad, 0x9176cc88, 0x254c02f5); + &_data_word(0xfcd7e54f, 0xd7cb2ac5, 0x80443526, 0x8fa362b5); + &_data_word(0x495ab1de, 0x671bba25, 0x980eea45, 0xe1c0fe5d); + &_data_word(0x02752fc3, 0x12f04c81, 0xa397468d, 0xc6f9d36b); + &_data_word(0xe75f8f03, 0x959c9215, 0xeb7a6dbf, 0xda595295); + &_data_word(0x2d83bed4, 0xd3217458, 0x2969e049, 0x44c8c98e); + &_data_word(0x6a89c275, 0x78798ef4, 0x6b3e5899, 0xdd71b927); + &_data_word(0xb64fe1be, 0x17ad88f0, 0x66ac20c9, 0xb43ace7d); + &_data_word(0x184adf63, 0x82311ae5, 0x60335197, 0x457f5362); + &_data_word(0xe07764b1, 0x84ae6bbb, 0x1ca081fe, 0x942b08f9); + &_data_word(0x58684870, 0x19fd458f, 0x876cde94, 0xb7f87b52); + &_data_word(0x23d373ab, 0xe2024b72, 0x578f1fe3, 0x2aab5566); + &_data_word(0x0728ebb2, 0x03c2b52f, 0x9a7bc586, 0xa50837d3); + &_data_word(0xf2872830, 0xb2a5bf23, 0xba6a0302, 0x5c8216ed); + &_data_word(0x2b1ccf8a, 0x92b479a7, 0xf0f207f3, 0xa1e2694e); + &_data_word(0xcdf4da65, 0xd5be0506, 0x1f6234d1, 0x8afea6c4); + &_data_word(0x9d532e34, 0xa055f3a2, 0x32e18a05, 0x75ebf6a4); + &_data_word(0x39ec830b, 0xaaef6040, 0x069f715e, 0x51106ebd); + &_data_word(0xf98a213e, 0x3d06dd96, 0xae053edd, 0x46bde64d); + &_data_word(0xb58d5491, 0x055dc471, 0x6fd40604, 0xff155060); + &_data_word(0x24fb9819, 0x97e9bdd6, 0xcc434089, 0x779ed967); + &_data_word(0xbd42e8b0, 0x888b8907, 0x385b19e7, 0xdbeec879); + &_data_word(0x470a7ca1, 0xe90f427c, 0xc91e84f8, 0x00000000); + &_data_word(0x83868009, 0x48ed2b32, 0xac70111e, 0x4e725a6c); + &_data_word(0xfbff0efd, 0x5638850f, 0x1ed5ae3d, 0x27392d36); + &_data_word(0x64d90f0a, 0x21a65c68, 0xd1545b9b, 0x3a2e3624); + &_data_word(0xb1670a0c, 0x0fe75793, 0xd296eeb4, 0x9e919b1b); + &_data_word(0x4fc5c080, 0xa220dc61, 0x694b775a, 0x161a121c); + &_data_word(0x0aba93e2, 0xe52aa0c0, 0x43e0223c, 0x1d171b12); + &_data_word(0x0b0d090e, 0xadc78bf2, 0xb9a8b62d, 0xc8a91e14); + &_data_word(0x8519f157, 0x4c0775af, 0xbbdd99ee, 0xfd607fa3); + &_data_word(0x9f2601f7, 0xbcf5725c, 0xc53b6644, 0x347efb5b); + &_data_word(0x7629438b, 0xdcc623cb, 0x68fcedb6, 0x63f1e4b8); + &_data_word(0xcadc31d7, 0x10856342, 0x40229713, 0x2011c684); + &_data_word(0x7d244a85, 0xf83dbbd2, 0x1132f9ae, 0x6da129c7); + &_data_word(0x4b2f9e1d, 0xf330b2dc, 0xec52860d, 0xd0e3c177); + &_data_word(0x6c16b32b, 0x99b970a9, 0xfa489411, 0x2264e947); + &_data_word(0xc48cfca8, 0x1a3ff0a0, 0xd82c7d56, 0xef903322); + &_data_word(0xc74e4987, 0xc1d138d9, 0xfea2ca8c, 0x360bd498); + &_data_word(0xcf81f5a6, 0x28de7aa5, 0x268eb7da, 0xa4bfad3f); + &_data_word(0xe49d3a2c, 0x0d927850, 0x9bcc5f6a, 0x62467e54); + &_data_word(0xc2138df6, 0xe8b8d890, 0x5ef7392e, 0xf5afc382); + &_data_word(0xbe805d9f, 0x7c93d069, 0xa92dd56f, 0xb31225cf); + &_data_word(0x3b99acc8, 0xa77d1810, 0x6e639ce8, 0x7bbb3bdb); + &_data_word(0x097826cd, 0xf418596e, 0x01b79aec, 0xa89a4f83); + &_data_word(0x656e95e6, 0x7ee6ffaa, 0x08cfbc21, 0xe6e815ef); + &_data_word(0xd99be7ba, 0xce366f4a, 0xd4099fea, 0xd67cb029); + &_data_word(0xafb2a431, 0x31233f2a, 0x3094a5c6, 0xc066a235); + &_data_word(0x37bc4e74, 0xa6ca82fc, 0xb0d090e0, 0x15d8a733); + &_data_word(0x4a9804f1, 0xf7daec41, 0x0e50cd7f, 0x2ff69117); + &_data_word(0x8dd64d76, 0x4db0ef43, 0x544daacc, 0xdf0496e4); + &_data_word(0xe3b5d19e, 0x1b886a4c, 0xb81f2cc1, 0x7f516546); + &_data_word(0x04ea5e9d, 0x5d358c01, 0x737487fa, 0x2e410bfb); + &_data_word(0x5a1d67b3, 0x52d2db92, 0x335610e9, 0x1347d66d); + &_data_word(0x8c61d79a, 0x7a0ca137, 0x8e14f859, 0x893c13eb); + &_data_word(0xee27a9ce, 0x35c961b7, 0xede51ce1, 0x3cb1477a); + &_data_word(0x59dfd29c, 0x3f73f255, 0x79ce1418, 0xbf37c773); + &_data_word(0xeacdf753, 0x5baafd5f, 0x146f3ddf, 0x86db4478); + &_data_word(0x81f3afca, 0x3ec468b9, 0x2c342438, 0x5f40a3c2); + &_data_word(0x72c31d16, 0x0c25e2bc, 0x8b493c28, 0x41950dff); + &_data_word(0x7101a839, 0xdeb30c08, 0x9ce4b4d8, 0x90c15664); + &_data_word(0x6184cb7b, 0x70b632d5, 0x745c6c48, 0x4257b8d0); + +#Td4: # four copies of Td4 to choose from to avoid L1 aliasing + &data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38); + &data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb); + &data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87); + &data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb); + &data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d); + &data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e); + &data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2); + &data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25); + &data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16); + &data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92); + &data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda); + &data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84); + &data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a); + &data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06); + &data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02); + &data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b); + &data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea); + &data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73); + &data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85); + &data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e); + &data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89); + &data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b); + &data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20); + &data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4); + &data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31); + &data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f); + &data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d); + &data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef); + &data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0); + &data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61); + &data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26); + &data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d); + + &data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38); + &data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb); + &data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87); + &data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb); + &data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d); + &data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e); + &data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2); + &data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25); + &data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16); + &data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92); + &data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda); + &data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84); + &data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a); + &data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06); + &data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02); + &data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b); + &data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea); + &data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73); + &data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85); + &data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e); + &data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89); + &data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b); + &data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20); + &data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4); + &data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31); + &data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f); + &data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d); + &data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef); + &data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0); + &data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61); + &data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26); + &data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d); + + &data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38); + &data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb); + &data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87); + &data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb); + &data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d); + &data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e); + &data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2); + &data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25); + &data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16); + &data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92); + &data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda); + &data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84); + &data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a); + &data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06); + &data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02); + &data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b); + &data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea); + &data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73); + &data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85); + &data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e); + &data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89); + &data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b); + &data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20); + &data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4); + &data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31); + &data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f); + &data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d); + &data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef); + &data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0); + &data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61); + &data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26); + &data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d); + + &data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38); + &data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb); + &data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87); + &data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb); + &data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d); + &data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e); + &data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2); + &data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25); + &data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16); + &data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92); + &data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda); + &data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84); + &data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a); + &data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06); + &data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02); + &data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b); + &data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea); + &data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73); + &data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85); + &data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e); + &data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89); + &data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b); + &data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20); + &data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4); + &data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31); + &data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f); + &data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d); + &data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef); + &data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0); + &data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61); + &data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26); + &data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d); &function_end_B("_x86_AES_decrypt"); # void AES_decrypt (const void *inp,void *out,const AES_KEY *key); -&public_label("AES_Td"); &function_begin("AES_decrypt"); &mov ($acc,&wparam(0)); # load inp &mov ($key,&wparam(2)); # load key + &mov ($s0,"esp"); + &sub ("esp",36); + &and ("esp",-64); # align to cache-line + + # place stack frame just "above" the key schedule + &lea ($s1,&DWP(-64-63,$key)); + &sub ($s1,"esp"); + &neg ($s1); + &and ($s1,0x3C0); # modulo 1024, but aligned to cache-line + &sub ("esp",$s1); + &add ("esp",4); # 4 is reserved for caller's return address + &mov ($_esp,$s0); # save stack pointer + &call (&label("pic_point")); # make it PIC! &set_label("pic_point"); - &blindpop("ebp"); - &lea ("ebp",&DWP(&label("AES_Td")."-".&label("pic_point"),"ebp")); - + &blindpop($tbl); + &picmeup($s0,"OPENSSL_ia32cap_P",$tbl,&label("pic_point")) if(!$x86only); + &lea ($tbl,&DWP(&label("AES_Td")."-".&label("pic_point"),$tbl)); + + # pick Td4 copy which can't "overlap" with stack frame or key schedule + &lea ($s1,&DWP(768-4,"esp")); + &sub ($s1,$tbl); + &and ($s1,0x300); + &lea ($tbl,&DWP(2048+128,$tbl,$s1)); + + if (!$x86only) { + &bt (&DWP(0,$s0),25); # check for SSE bit + &jnc (&label("x86")); + + &movq ("mm0",&QWP(0,$acc)); + &movq ("mm4",&QWP(8,$acc)); + &call ("_sse_AES_decrypt_compact"); + &mov ("esp",$_esp); # restore stack pointer + &mov ($acc,&wparam(1)); # load out + &movq (&QWP(0,$acc),"mm0"); # write output data + &movq (&QWP(8,$acc),"mm4"); + &emms (); + &function_end_A(); + } + &set_label("x86",16); + &mov ($_tbl,$tbl); &mov ($s0,&DWP(0,$acc)); # load input data &mov ($s1,&DWP(4,$acc)); &mov ($s2,&DWP(8,$acc)); &mov ($s3,&DWP(12,$acc)); - - &call ("_x86_AES_decrypt"); - + &call ("_x86_AES_decrypt_compact"); + &mov ("esp",$_esp); # restore stack pointer &mov ($acc,&wparam(1)); # load out &mov (&DWP(0,$acc),$s0); # write output data &mov (&DWP(4,$acc),$s1); @@ -1225,156 +2006,233 @@ sub declast() # void AES_cbc_encrypt (const void char *inp, unsigned char *out, # size_t length, const AES_KEY *key, -# unsigned char *ivp,const int enc); -&public_label("AES_Te"); -&public_label("AES_Td"); +# unsigned char *ivp,const int enc); +{ +# stack frame layout +# -4(%esp) # return address 0(%esp) +# 0(%esp) # s0 backing store 4(%esp) +# 4(%esp) # s1 backing store 8(%esp) +# 8(%esp) # s2 backing store 12(%esp) +# 12(%esp) # s3 backing store 16(%esp) +# 16(%esp) # key backup 20(%esp) +# 20(%esp) # end of key schedule 24(%esp) +# 24(%esp) # %ebp backup 28(%esp) +# 28(%esp) # %esp backup +my $_inp=&DWP(32,"esp"); # copy of wparam(0) +my $_out=&DWP(36,"esp"); # copy of wparam(1) +my $_len=&DWP(40,"esp"); # copy of wparam(2) +my $_key=&DWP(44,"esp"); # copy of wparam(3) +my $_ivp=&DWP(48,"esp"); # copy of wparam(4) +my $_tmp=&DWP(52,"esp"); # volatile variable +# +my $ivec=&DWP(60,"esp"); # ivec[16] +my $aes_key=&DWP(76,"esp"); # copy of aes_key +my $mark=&DWP(76+240,"esp"); # copy of aes_key->rounds + &function_begin("AES_cbc_encrypt"); &mov ($s2 eq "ecx"? $s2 : "",&wparam(2)); # load len &cmp ($s2,0); - &je (&label("enc_out")); + &je (&label("drop_out")); - &call (&label("pic_point")); # make it PIC! + &call (&label("pic_point")); # make it PIC! &set_label("pic_point"); - &blindpop("ebp"); + &blindpop($tbl); + &picmeup($s0,"OPENSSL_ia32cap_P",$tbl,&label("pic_point")) if(!$x86only); &cmp (&wparam(5),0); - &je (&label("DECRYPT")); + &lea ($tbl,&DWP(&label("AES_Te")."-".&label("pic_point"),$tbl)); + &jne (&label("picked_te")); + &lea ($tbl,&DWP(&label("AES_Td")."-".&label("AES_Te"),$tbl)); + &set_label("picked_te"); - &lea ("ebp",&DWP(&label("AES_Te")."-".&label("pic_point"),"ebp")); - - &mov ($acc,&wparam(0)); # load inp - &mov ($key,&wparam(4)); # load ivp - - &test ($s2,~15); - &jz (&label("enc_tail")); # short input... + # one can argue if this is required + &pushf (); + &cld (); + &cmp ($s2,$speed_limit); + &jb (&label("slow_way")); + &test ($s2,15); + &jnz (&label("slow_way")); + if (!$x86only) { + &bt (&DWP(0,$s0),28); # check for hyper-threading bit + &jc (&label("slow_way")); + } + # pre-allocate aligned stack frame... + &lea ($acc,&DWP(-80-244,"esp")); + &and ($acc,-64); + + # ... and make sure it doesn't alias with $tbl modulo 4096 + &mov ($s0,$tbl); + &lea ($s1,&DWP(2048+256,$tbl)); + &mov ($s3,$acc); + &and ($s0,0xfff); # s = %ebp&0xfff + &and ($s1,0xfff); # e = (%ebp+2048+256)&0xfff + &and ($s3,0xfff); # p = %esp&0xfff + + &cmp ($s3,$s1); # if (p>=e) %esp =- (p-e); + &jb (&label("tbl_break_out")); + &sub ($s3,$s1); + &sub ($acc,$s3); + &jmp (&label("tbl_ok")); + &set_label("tbl_break_out",4); # else %esp -= (p-s)&0xfff + framesz; + &sub ($s3,$s0); + &and ($s3,0xfff); + &add ($s3,384); + &sub ($acc,$s3); + &set_label("tbl_ok",4); + + &lea ($s3,&wparam(0)); # obtain pointer to parameter block + &exch ("esp",$acc); # allocate stack frame + &add ("esp",4); # reserve for return address! + &mov ($_tbl,$tbl); # save %ebp + &mov ($_esp,$acc); # save %esp + + &mov ($s0,&DWP(0,$s3)); # load inp + &mov ($s1,&DWP(4,$s3)); # load out + #&mov ($s2,&DWP(8,$s3)); # load len + &mov ($key,&DWP(12,$s3)); # load key + &mov ($acc,&DWP(16,$s3)); # load ivp + &mov ($s3,&DWP(20,$s3)); # load enc flag + + &mov ($_inp,$s0); # save copy of inp + &mov ($_out,$s1); # save copy of out + &mov ($_len,$s2); # save copy of len + &mov ($_key,$key); # save copy of key + &mov ($_ivp,$acc); # save copy of ivp + + &mov ($mark,0); # copy of aes_key->rounds = 0; + # do we copy key schedule to stack? + &mov ($s1 eq "ebx" ? $s1 : "",$key); + &mov ($s2 eq "ecx" ? $s2 : "",244/4); + &sub ($s1,$tbl); + &mov ("esi",$key); + &and ($s1,0xfff); + &lea ("edi",$aes_key); + &cmp ($s1,2048+256); + &jb (&label("do_copy")); + &cmp ($s1,4096-244); + &jb (&label("skip_copy")); + &set_label("do_copy",4); + &mov ($_key,"edi"); + &data_word(0xA5F3F689); # rep movsd + &set_label("skip_copy"); + + &mov ($key,16); + &set_label("prefetch_tbl",4); + &mov ($s0,&DWP(0,$tbl)); + &mov ($s1,&DWP(32,$tbl)); + &mov ($s2,&DWP(64,$tbl)); + &mov ($acc,&DWP(96,$tbl)); + &lea ($tbl,&DWP(128,$tbl)); + &sub ($key,1); + &jnz (&label("prefetch_tbl")); + &sub ($tbl,2048); + + &mov ($acc,$_inp); + &mov ($key,$_ivp); + + &cmp ($s3,0); + &je (&label("fast_decrypt")); + +#----------------------------- ENCRYPT -----------------------------# &mov ($s0,&DWP(0,$key)); # load iv &mov ($s1,&DWP(4,$key)); - &align (4); - &set_label("enc_loop"); + &set_label("fast_enc_loop",16); &mov ($s2,&DWP(8,$key)); &mov ($s3,&DWP(12,$key)); - &xor ($s0,&DWP(0,$acc)); # xor input data + &xor ($s0,&DWP(0,$acc)); # xor input data &xor ($s1,&DWP(4,$acc)); &xor ($s2,&DWP(8,$acc)); &xor ($s3,&DWP(12,$acc)); - &mov ($key,&wparam(3)); # load key + &mov ($key,$_key); # load key &call ("_x86_AES_encrypt"); - &mov ($acc,&wparam(0)); # load inp - &mov ($key,&wparam(1)); # load out + &mov ($acc,$_inp); # load inp + &mov ($key,$_out); # load out - &mov (&DWP(0,$key),$s0); # save output data + &mov (&DWP(0,$key),$s0); # save output data &mov (&DWP(4,$key),$s1); &mov (&DWP(8,$key),$s2); &mov (&DWP(12,$key),$s3); - &mov ($s2,&wparam(2)); # load len - - &lea ($acc,&DWP(16,$acc)); - &mov (&wparam(0),$acc); # save inp - - &lea ($s3,&DWP(16,$key)); - &mov (&wparam(1),$s3); # save out - - &sub ($s2,16); - &test ($s2,~15); - &mov (&wparam(2),$s2); # save len - &jnz (&label("enc_loop")); - &test ($s2,15); - &jnz (&label("enc_tail")); - &mov ($acc,&wparam(4)); # load ivp - &mov ($s2,&DWP(8,$key)); # restore last dwords + &lea ($acc,&DWP(16,$acc)); # advance inp + &mov ($s2,$_len); # load len + &mov ($_inp,$acc); # save inp + &lea ($s3,&DWP(16,$key)); # advance out + &mov ($_out,$s3); # save out + &sub ($s2,16); # decrease len + &mov ($_len,$s2); # save len + &jnz (&label("fast_enc_loop")); + &mov ($acc,$_ivp); # load ivp + &mov ($s2,&DWP(8,$key)); # restore last 2 dwords &mov ($s3,&DWP(12,$key)); - &mov (&DWP(0,$acc),$s0); # save iv + &mov (&DWP(0,$acc),$s0); # save ivec &mov (&DWP(4,$acc),$s1); &mov (&DWP(8,$acc),$s2); &mov (&DWP(12,$acc),$s3); - &set_label("enc_out"); - &function_end_A(); - &align (4); - &set_label("enc_tail"); - &push ($key eq "edi" ? $key : ""); # push ivp - &pushf (); - &mov ($key,&wparam(1)); # load out - &mov ($s1,16); - &sub ($s1,$s2); - &cmp ($key,$acc); # compare with inp - &je (&label("enc_in_place")); - &data_word(0x90A4F3FC); # cld; rep movsb; nop # copy input - &jmp (&label("enc_skip_in_place")); - &set_label("enc_in_place"); - &lea ($key,&DWP(0,$key,$s2)); - &set_label("enc_skip_in_place"); - &mov ($s2,$s1); - &xor ($s0,$s0); - &data_word(0x90AAF3FC); # cld; rep stosb; nop # zero tail + &cmp ($mark,0); # was the key schedule copied? + &mov ("edi",$_key); + &je (&label("skip_ezero")); + # zero copy of key schedule + &mov ("ecx",240/4); + &xor ("eax","eax"); + &align (4); + &data_word(0xABF3F689); # rep stosd + &set_label("skip_ezero") + &mov ("esp",$_esp); &popf (); - &pop ($key); # pop ivp - - &mov ($acc,&wparam(1)); # output as input - &mov ($s0,&DWP(0,$key)); - &mov ($s1,&DWP(4,$key)); - &mov (&wparam(2),16); # len=16 - &jmp (&label("enc_loop")); # one more spin... + &set_label("drop_out"); + &function_end_A(); + &pushf (); # kludge, never executed #----------------------------- DECRYPT -----------------------------# -&align (4); -&set_label("DECRYPT"); - &stack_push(5); # allocate temp + ivp +&set_label("fast_decrypt",16); - &lea ("ebp",&DWP(&label("AES_Td")."-".&label("pic_point"),"ebp")); + &cmp ($acc,$_out); + &je (&label("fast_dec_in_place")); # in-place processing... - &mov ($acc,&wparam(0)); # load inp - &cmp ($acc,&wparam(1)); - &je (&label("dec_in_place")); # in-place processing... - - &mov ($key,&wparam(4)); # load ivp - &mov (&swtmp(4),$key); + &mov ($_tmp,$key); &align (4); - &set_label("dec_loop"); - &mov ($s0,&DWP(0,$acc)); # read input + &set_label("fast_dec_loop",16); + &mov ($s0,&DWP(0,$acc)); # read input &mov ($s1,&DWP(4,$acc)); &mov ($s2,&DWP(8,$acc)); &mov ($s3,&DWP(12,$acc)); - &mov ($key,&wparam(3)); # load key + &mov ($key,$_key); # load key &call ("_x86_AES_decrypt"); - &mov ($key,&swtmp(4)); # load ivp - &mov ($acc,&wparam(2)); # load len - &xor ($s0,&DWP(0,$key)); # xor iv + &mov ($key,$_tmp); # load ivp + &mov ($acc,$_len); # load len + &xor ($s0,&DWP(0,$key)); # xor iv &xor ($s1,&DWP(4,$key)); &xor ($s2,&DWP(8,$key)); &xor ($s3,&DWP(12,$key)); - &sub ($acc,16); - &jc (&label("dec_partial")); - &mov (&wparam(2),$acc); # save len - &mov ($acc,&wparam(0)); # load inp - &mov ($key,&wparam(1)); # load out + &mov ($key,$_out); # load out + &mov ($acc,$_inp); # load inp - &mov (&DWP(0,$key),$s0); # write output + &mov (&DWP(0,$key),$s0); # write output &mov (&DWP(4,$key),$s1); &mov (&DWP(8,$key),$s2); &mov (&DWP(12,$key),$s3); - &mov (&swtmp(4),$acc); # save ivp - &lea ($acc,&DWP(16,$acc)); - &mov (&wparam(0),$acc); # save inp - - &lea ($key,&DWP(16,$key)); - &mov (&wparam(1),$key); # save out - - &jnz (&label("dec_loop")); - &mov ($key,&swtmp(4)); # load temp ivp - &set_label("dec_end"); - &mov ($acc,&wparam(4)); # load user ivp + &mov ($s2,$_len); # load len + &mov ($_tmp,$acc); # save ivp + &lea ($acc,&DWP(16,$acc)); # advance inp + &mov ($_inp,$acc); # save inp + &lea ($key,&DWP(16,$key)); # advance out + &mov ($_out,$key); # save out + &sub ($s2,16); # decrease len + &mov ($_len,$s2); # save len + &jnz (&label("fast_dec_loop")); + &mov ($key,$_tmp); # load temp ivp + &mov ($acc,$_ivp); # load user ivp &mov ($s0,&DWP(0,$key)); # load iv &mov ($s1,&DWP(4,$key)); &mov ($s2,&DWP(8,$key)); @@ -1383,130 +2241,418 @@ sub declast() &mov (&DWP(4,$acc),$s1); &mov (&DWP(8,$acc),$s2); &mov (&DWP(12,$acc),$s3); - &jmp (&label("dec_out")); + &jmp (&label("fast_dec_out")); - &align (4); - &set_label("dec_partial"); - &lea ($key,&swtmp(0)); - &mov (&DWP(0,$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" ? $s2 : "",&DWP(16,$acc)); - &mov ($acc eq "esi" ? $acc : "",$key); - &mov ($key eq "edi" ? $key : "",&wparam(1)); - &pushf (); - &data_word(0x90A4F3FC); # cld; rep movsb; nop # copy output - &popf (); - &mov ($key,&wparam(0)); # load temp ivp - &jmp (&label("dec_end")); - - &align (4); - &set_label("dec_in_place"); - &set_label("dec_in_place_loop"); - &lea ($key,&swtmp(0)); - &mov ($s0,&DWP(0,$acc)); # read input + &set_label("fast_dec_in_place",16); + &set_label("fast_dec_in_place_loop"); + &mov ($s0,&DWP(0,$acc)); # read input &mov ($s1,&DWP(4,$acc)); &mov ($s2,&DWP(8,$acc)); &mov ($s3,&DWP(12,$acc)); - &mov (&DWP(0,$key),$s0); # copy to temp + &lea ($key,$ivec); + &mov (&DWP(0,$key),$s0); # copy to temp &mov (&DWP(4,$key),$s1); &mov (&DWP(8,$key),$s2); &mov (&DWP(12,$key),$s3); - &mov ($key,&wparam(3)); # load key + &mov ($key,$_key); # load key &call ("_x86_AES_decrypt"); - &mov ($key,&wparam(4)); # load ivp - &mov ($acc,&wparam(1)); # load out - &xor ($s0,&DWP(0,$key)); # xor iv + &mov ($key,$_ivp); # load ivp + &mov ($acc,$_out); # load out + &xor ($s0,&DWP(0,$key)); # xor iv &xor ($s1,&DWP(4,$key)); &xor ($s2,&DWP(8,$key)); &xor ($s3,&DWP(12,$key)); - &mov (&DWP(0,$acc),$s0); # write output + &mov (&DWP(0,$acc),$s0); # write output &mov (&DWP(4,$acc),$s1); &mov (&DWP(8,$acc),$s2); &mov (&DWP(12,$acc),$s3); - &lea ($acc,&DWP(16,$acc)); - &mov (&wparam(1),$acc); # save out + &lea ($acc,&DWP(16,$acc)); # advance out + &mov ($_out,$acc); # save out - &lea ($acc,&swtmp(0)); - &mov ($s0,&DWP(0,$acc)); # read temp + &lea ($acc,$ivec); + &mov ($s0,&DWP(0,$acc)); # read temp &mov ($s1,&DWP(4,$acc)); &mov ($s2,&DWP(8,$acc)); &mov ($s3,&DWP(12,$acc)); - &mov (&DWP(0,$key),$s0); # copy iv + &mov (&DWP(0,$key),$s0); # copy iv &mov (&DWP(4,$key),$s1); &mov (&DWP(8,$key),$s2); &mov (&DWP(12,$key),$s3); - &mov ($acc,&wparam(0)); # load inp + &mov ($acc,$_inp); # load inp + &mov ($s2,$_len); # load len + &lea ($acc,&DWP(16,$acc)); # advance inp + &mov ($_inp,$acc); # save inp + &sub ($s2,16); # decrease len + &mov ($_len,$s2); # save len + &jnz (&label("fast_dec_in_place_loop")); + + &set_label("fast_dec_out",4); + &cmp ($mark,0); # was the key schedule copied? + &mov ("edi",$_key); + &je (&label("skip_dzero")); + # zero copy of key schedule + &mov ("ecx",240/4); + &xor ("eax","eax"); + &align (4); + &data_word(0xABF3F689); # rep stosd + &set_label("skip_dzero") + &mov ("esp",$_esp); + &popf (); + &function_end_A(); + &pushf (); # kludge, never executed + +#--------------------------- SLOW ROUTINE ---------------------------# +&set_label("slow_way",16); + + &mov ($s0,&DWP(0,$s0)) if (!$x86only);# load OPENSSL_ia32cap + &mov ($key,&wparam(3)); # load key + + # pre-allocate aligned stack frame... + &lea ($acc,&DWP(-80,"esp")); + &and ($acc,-64); + + # ... and make sure it doesn't alias with $key modulo 1024 + &lea ($s1,&DWP(-80-63,$key)); + &sub ($s1,$acc); + &neg ($s1); + &and ($s1,0x3C0); # modulo 1024, but aligned to cache-line + &sub ($acc,$s1); + + # pick S-box copy which can't overlap with stack frame or $key + &lea ($s1,&DWP(768,$acc)); + &sub ($s1,$tbl); + &and ($s1,0x300); + &lea ($tbl,&DWP(2048+128,$tbl,$s1)); + + &lea ($s3,&wparam(0)); # pointer to parameter block + + &exch ("esp",$acc); + &add ("esp",4); # reserve for return address! + &mov ($_tbl,$tbl); # save %ebp + &mov ($_esp,$acc); # save %esp + &mov ($_tmp,$s0); # save OPENSSL_ia32cap + + &mov ($s0,&DWP(0,$s3)); # load inp + &mov ($s1,&DWP(4,$s3)); # load out + #&mov ($s2,&DWP(8,$s3)); # load len + #&mov ($key,&DWP(12,$s3)); # load key + &mov ($acc,&DWP(16,$s3)); # load ivp + &mov ($s3,&DWP(20,$s3)); # load enc flag + + &mov ($_inp,$s0); # save copy of inp + &mov ($_out,$s1); # save copy of out + &mov ($_len,$s2); # save copy of len + &mov ($_key,$key); # save copy of key + &mov ($_ivp,$acc); # save copy of ivp + + &mov ($key,$acc); + &mov ($acc,$s0); + + &cmp ($s3,0); + &je (&label("slow_decrypt")); + +#--------------------------- SLOW ENCRYPT ---------------------------# + &cmp ($s2,16); + &mov ($s3,$s1); + &jb (&label("slow_enc_tail")); + + if (!$x86only) { + &bt ($_tmp,25); # check for SSE bit + &jnc (&label("slow_enc_x86")); + + &movq ("mm0",&QWP(0,$key)); # load iv + &movq ("mm4",&QWP(8,$key)); + + &set_label("slow_enc_loop_sse",16); + &pxor ("mm0",&QWP(0,$acc)); # xor input data + &pxor ("mm4",&QWP(8,$acc)); + + &mov ($key,$_key); + &call ("_sse_AES_encrypt_compact"); + + &mov ($acc,$_inp); # load inp + &mov ($key,$_out); # load out + &mov ($s2,$_len); # load len + + &movq (&QWP(0,$key),"mm0"); # save output data + &movq (&QWP(8,$key),"mm4"); + + &lea ($acc,&DWP(16,$acc)); # advance inp + &mov ($_inp,$acc); # save inp + &lea ($s3,&DWP(16,$key)); # advance out + &mov ($_out,$s3); # save out + &sub ($s2,16); # decrease len + &cmp ($s2,16); + &mov ($_len,$s2); # save len + &jae (&label("slow_enc_loop_sse")); + &test ($s2,15); + &jnz (&label("slow_enc_tail")); + &mov ($acc,$_ivp); # load ivp + &movq (&QWP(0,$acc),"mm0"); # save ivec + &movq (&QWP(8,$acc),"mm4"); + &emms (); + &mov ("esp",$_esp); + &popf (); + &function_end_A(); + &pushf (); # kludge, never executed + } + &set_label("slow_enc_x86",16); + &mov ($s0,&DWP(0,$key)); # load iv + &mov ($s1,&DWP(4,$key)); + + &set_label("slow_enc_loop_x86",4); + &mov ($s2,&DWP(8,$key)); + &mov ($s3,&DWP(12,$key)); + + &xor ($s0,&DWP(0,$acc)); # xor input data + &xor ($s1,&DWP(4,$acc)); + &xor ($s2,&DWP(8,$acc)); + &xor ($s3,&DWP(12,$acc)); + + &mov ($key,$_key); # load key + &call ("_x86_AES_encrypt_compact"); + + &mov ($acc,$_inp); # load inp + &mov ($key,$_out); # load out - &lea ($acc,&DWP(16,$acc)); - &mov (&wparam(0),$acc); # save inp + &mov (&DWP(0,$key),$s0); # save output data + &mov (&DWP(4,$key),$s1); + &mov (&DWP(8,$key),$s2); + &mov (&DWP(12,$key),$s3); + + &mov ($s2,$_len); # load len + &lea ($acc,&DWP(16,$acc)); # advance inp + &mov ($_inp,$acc); # save inp + &lea ($s3,&DWP(16,$key)); # advance out + &mov ($_out,$s3); # save out + &sub ($s2,16); # decrease len + &cmp ($s2,16); + &mov ($_len,$s2); # save len + &jae (&label("slow_enc_loop_x86")); + &test ($s2,15); + &jnz (&label("slow_enc_tail")); + &mov ($acc,$_ivp); # load ivp + &mov ($s2,&DWP(8,$key)); # restore last dwords + &mov ($s3,&DWP(12,$key)); + &mov (&DWP(0,$acc),$s0); # save ivec + &mov (&DWP(4,$acc),$s1); + &mov (&DWP(8,$acc),$s2); + &mov (&DWP(12,$acc),$s3); - &mov ($s2,&wparam(2)); # load len - &sub ($s2,16); - &jc (&label("dec_in_place_partial")); - &mov (&wparam(2),$s2); # save len - &jnz (&label("dec_in_place_loop")); - &jmp (&label("dec_out")); + &mov ("esp",$_esp); + &popf (); + &function_end_A(); + &pushf (); # kludge, never executed - &align (4); - &set_label("dec_in_place_partial"); - # one can argue if this is actually required... - &mov ($key eq "edi" ? $key : "",&wparam(1)); - &lea ($acc eq "esi" ? $acc : "",&swtmp(0)); + &set_label("slow_enc_tail",16); + &emms () if (!$x86only); + &mov ($key eq "edi"? $key:"",$s3); # load out to edi + &mov ($s1,16); + &sub ($s1,$s2); + &cmp ($key,$acc eq "esi"? $acc:""); # compare with inp + &je (&label("enc_in_place")); + &align (4); + &data_word(0xA4F3F689); # rep movsb # copy input + &jmp (&label("enc_skip_in_place")); + &set_label("enc_in_place"); &lea ($key,&DWP(0,$key,$s2)); - &lea ($acc,&DWP(16,$acc,$s2)); - &neg ($s2 eq "ecx" ? $s2 : ""); - &pushf (); - &data_word(0x90A4F3FC); # cld; rep movsb; nop # restore tail + &set_label("enc_skip_in_place"); + &mov ($s2,$s1); + &xor ($s0,$s0); + &align (4); + &data_word(0xAAF3F689); # rep stosb # zero tail + + &mov ($key,$_ivp); # restore ivp + &mov ($acc,$s3); # output as input + &mov ($s0,&DWP(0,$key)); + &mov ($s1,&DWP(4,$key)); + &mov ($_len,16); # len=16 + &jmp (&label("slow_enc_loop_x86")); # one more spin... + +#--------------------------- SLOW DECRYPT ---------------------------# +&set_label("slow_decrypt",16); + if (!$x86only) { + &bt ($_tmp,25); # check for SSE bit + &jnc (&label("slow_dec_loop_x86")); + + &set_label("slow_dec_loop_sse",4); + &movq ("mm0",&QWP(0,$acc)); # read input + &movq ("mm4",&QWP(8,$acc)); + + &mov ($key,$_key); + &call ("_sse_AES_decrypt_compact"); + + &mov ($acc,$_inp); # load inp + &lea ($s0,$ivec); + &mov ($s1,$_out); # load out + &mov ($s2,$_len); # load len + &mov ($key,$_ivp); # load ivp + + &movq ("mm1",&QWP(0,$acc)); # re-read input + &movq ("mm5",&QWP(8,$acc)); + + &pxor ("mm0",&QWP(0,$key)); # xor iv + &pxor ("mm4",&QWP(8,$key)); + + &movq (&QWP(0,$key),"mm1"); # copy input to iv + &movq (&QWP(8,$key),"mm5"); + + &sub ($s2,16); # decrease len + &jc (&label("slow_dec_partial_sse")); + + &movq (&QWP(0,$s1),"mm0"); # write output + &movq (&QWP(8,$s1),"mm4"); + + &lea ($s1,&DWP(16,$s1)); # advance out + &mov ($_out,$s1); # save out + &lea ($acc,&DWP(16,$acc)); # advance inp + &mov ($_inp,$acc); # save inp + &mov ($_len,$s2); # save len + &jnz (&label("slow_dec_loop_sse")); + &emms (); + &mov ("esp",$_esp); &popf (); + &function_end_A(); + &pushf (); # kludge, never executed + + &set_label("slow_dec_partial_sse",16); + &movq (&QWP(0,$s0),"mm0"); # save output to temp + &movq (&QWP(8,$s0),"mm4"); + &emms (); + + &add ($s2 eq "ecx" ? "ecx":"",16); + &mov ("edi",$s1); # out + &mov ("esi",$s0); # temp + &align (4); + &data_word(0xA4F3F689); # rep movsb # copy partial output + + &mov ("esp",$_esp); + &popf (); + &function_end_A(); + &pushf (); # kludge, never executed + } + &set_label("slow_dec_loop_x86",16); + &mov ($s0,&DWP(0,$acc)); # read input + &mov ($s1,&DWP(4,$acc)); + &mov ($s2,&DWP(8,$acc)); + &mov ($s3,&DWP(12,$acc)); + + &lea ($key,$ivec); + &mov (&DWP(0,$key),$s0); # copy to temp + &mov (&DWP(4,$key),$s1); + &mov (&DWP(8,$key),$s2); + &mov (&DWP(12,$key),$s3); + + &mov ($key,$_key); # load key + &call ("_x86_AES_decrypt_compact"); + + &mov ($key,$_ivp); # load ivp + &mov ($acc,$_len); # load len + &xor ($s0,&DWP(0,$key)); # xor iv + &xor ($s1,&DWP(4,$key)); + &xor ($s2,&DWP(8,$key)); + &xor ($s3,&DWP(12,$key)); + + &sub ($acc,16); + &jc (&label("slow_dec_partial_x86")); + + &mov ($_len,$acc); # save len + &mov ($acc,$_out); # load out + + &mov (&DWP(0,$acc),$s0); # write output + &mov (&DWP(4,$acc),$s1); + &mov (&DWP(8,$acc),$s2); + &mov (&DWP(12,$acc),$s3); + + &lea ($acc,&DWP(16,$acc)); # advance out + &mov ($_out,$acc); # save out + + &lea ($acc,$ivec); + &mov ($s0,&DWP(0,$acc)); # read temp + &mov ($s1,&DWP(4,$acc)); + &mov ($s2,&DWP(8,$acc)); + &mov ($s3,&DWP(12,$acc)); - &align (4); - &set_label("dec_out"); - &stack_pop(5); + &mov (&DWP(0,$key),$s0); # copy it to iv + &mov (&DWP(4,$key),$s1); + &mov (&DWP(8,$key),$s2); + &mov (&DWP(12,$key),$s3); + + &mov ($acc,$_inp); # load inp + &lea ($acc,&DWP(16,$acc)); # advance inp + &mov ($_inp,$acc); # save inp + &jnz (&label("slow_dec_loop_x86")); + &mov ("esp",$_esp); + &popf (); + &function_end_A(); + &pushf (); # kludge, never executed + + &set_label("slow_dec_partial_x86",16); + &lea ($acc,$ivec); + &mov (&DWP(0,$acc),$s0); # save output to temp + &mov (&DWP(4,$acc),$s1); + &mov (&DWP(8,$acc),$s2); + &mov (&DWP(12,$acc),$s3); + + &mov ($acc,$_inp); + &mov ($s0,&DWP(0,$acc)); # re-read input + &mov ($s1,&DWP(4,$acc)); + &mov ($s2,&DWP(8,$acc)); + &mov ($s3,&DWP(12,$acc)); + + &mov (&DWP(0,$key),$s0); # copy it to iv + &mov (&DWP(4,$key),$s1); + &mov (&DWP(8,$key),$s2); + &mov (&DWP(12,$key),$s3); + + &mov ("ecx",$_len); + &mov ("edi",$_out); + &lea ("esi",$ivec); + &align (4); + &data_word(0xA4F3F689); # rep movsb # copy partial output + + &mov ("esp",$_esp); + &popf (); &function_end("AES_cbc_encrypt"); +} #------------------------------------------------------------------# sub enckey() { &movz ("esi",&LB("edx")); # rk[i]>>0 - &mov ("ebx",&DWP(0,"ebp","esi",4)); + &movz ("ebx",&BP(-128,$tbl,"esi",1)); &movz ("esi",&HB("edx")); # rk[i]>>8 - &and ("ebx",0xFF000000); + &shl ("ebx",24); &xor ("eax","ebx"); - &mov ("ebx",&DWP(0,"ebp","esi",4)); + &movz ("ebx",&BP(-128,$tbl,"esi",1)); &shr ("edx",16); - &and ("ebx",0x000000FF); &movz ("esi",&LB("edx")); # rk[i]>>16 &xor ("eax","ebx"); - &mov ("ebx",&DWP(0,"ebp","esi",4)); + &movz ("ebx",&BP(-128,$tbl,"esi",1)); &movz ("esi",&HB("edx")); # rk[i]>>24 - &and ("ebx",0x0000FF00); + &shl ("ebx",8); &xor ("eax","ebx"); - &mov ("ebx",&DWP(0,"ebp","esi",4)); - &and ("ebx",0x00FF0000); + &movz ("ebx",&BP(-128,$tbl,"esi",1)); + &shl ("ebx",16); &xor ("eax","ebx"); - &xor ("eax",&DWP(1024,"ebp","ecx",4)); # rcon + &xor ("eax",&DWP(1024-128,$tbl,"ecx",4)); # rcon } -# int AES_set_encrypt_key(const unsigned char *userKey, const int bits, -# AES_KEY *key) -&public_label("AES_Te"); -&function_begin("AES_set_encrypt_key"); - &mov ("esi",&wparam(0)); # user supplied key - &mov ("edi",&wparam(2)); # private key schedule +&function_begin("_x86_AES_set_encrypt_key"); + &mov ("esi",&wparam(1)); # user supplied key + &mov ("edi",&wparam(3)); # private key schedule &test ("esi",-1); &jz (&label("badpointer")); @@ -1515,11 +2661,21 @@ sub enckey() &call (&label("pic_point")); &set_label("pic_point"); - &blindpop("ebp"); - &lea ("ebp",&DWP(&label("AES_Te")."-".&label("pic_point"),"ebp")); - &add ("ebp",1024*4); # skip to Te4 - - &mov ("ecx",&wparam(1)); # number of bits in key + &blindpop($tbl); + &lea ($tbl,&DWP(&label("AES_Te")."-".&label("pic_point"),$tbl)); + &lea ($tbl,&DWP(2048+128,$tbl)); + + # prefetch Te4 + &mov ("eax",&DWP(0-128,$tbl)); + &mov ("ebx",&DWP(32-128,$tbl)); + &mov ("ecx",&DWP(64-128,$tbl)); + &mov ("edx",&DWP(96-128,$tbl)); + &mov ("eax",&DWP(128-128,$tbl)); + &mov ("ebx",&DWP(160-128,$tbl)); + &mov ("ecx",&DWP(192-128,$tbl)); + &mov ("edx",&DWP(224-128,$tbl)); + + &mov ("ecx",&wparam(2)); # number of bits in key &cmp ("ecx",128); &je (&label("10rounds")); &cmp ("ecx",192); @@ -1658,24 +2814,23 @@ sub enckey() &mov ("edx","eax"); &mov ("eax",&DWP(16,"edi")); # rk[4] &movz ("esi",&LB("edx")); # rk[11]>>0 - &mov ("ebx",&DWP(0,"ebp","esi",4)); + &movz ("ebx",&BP(-128,$tbl,"esi",1)); &movz ("esi",&HB("edx")); # rk[11]>>8 - &and ("ebx",0x000000FF); &xor ("eax","ebx"); - &mov ("ebx",&DWP(0,"ebp","esi",4)); + &movz ("ebx",&BP(-128,$tbl,"esi",1)); &shr ("edx",16); - &and ("ebx",0x0000FF00); + &shl ("ebx",8); &movz ("esi",&LB("edx")); # rk[11]>>16 &xor ("eax","ebx"); - &mov ("ebx",&DWP(0,"ebp","esi",4)); + &movz ("ebx",&BP(-128,$tbl,"esi",1)); &movz ("esi",&HB("edx")); # rk[11]>>24 - &and ("ebx",0x00FF0000); + &shl ("ebx",16); &xor ("eax","ebx"); - &mov ("ebx",&DWP(0,"ebp","esi",4)); - &and ("ebx",0xFF000000); + &movz ("ebx",&BP(-128,$tbl,"esi",1)); + &shl ("ebx",24); &xor ("eax","ebx"); &mov (&DWP(48,"edi"),"eax"); # rk[12] @@ -1697,43 +2852,74 @@ sub enckey() &set_label("badpointer"); &mov ("eax",-1); &set_label("exit"); -&function_end("AES_set_encrypt_key"); +&function_end("_x86_AES_set_encrypt_key"); -sub deckey() -{ my ($i,$ptr,$te4,$td) = @_; +# int AES_set_encrypt_key(const unsigned char *userKey, const int bits, +# AES_KEY *key) +&function_begin_B("AES_set_encrypt_key"); + &call ("_x86_AES_set_encrypt_key"); + &ret (); +&function_end_B("AES_set_encrypt_key"); - &mov ("eax",&DWP($i,$ptr)); - &mov ("edx","eax"); - &movz ("ebx",&HB("eax")); - &shr ("edx",16); - &and ("eax",0xFF); - &movz ("eax",&BP(0,$te4,"eax",4)); - &movz ("ebx",&BP(0,$te4,"ebx",4)); - &mov ("eax",&DWP(1024*0,$td,"eax",4)); - &xor ("eax",&DWP(1024*1,$td,"ebx",4)); - &movz ("ebx",&HB("edx")); - &and ("edx",0xFF); - &movz ("edx",&BP(0,$te4,"edx",4)); - &movz ("ebx",&BP(0,$te4,"ebx",4)); - &xor ("eax",&DWP(1024*2,$td,"edx",4)); - &xor ("eax",&DWP(1024*3,$td,"ebx",4)); - &mov (&DWP($i,$ptr),"eax"); +sub deckey() +{ my ($i,$key,$tp1,$tp2,$tp4,$tp8) = @_; + my $tmp = $tbl; + + &mov ($acc,$tp1); + &and ($acc,0x80808080); + &mov ($tmp,$acc); + &shr ($tmp,7); + &lea ($tp2,&DWP(0,$tp1,$tp1)); + &sub ($acc,$tmp); + &and ($tp2,0xfefefefe); + &and ($acc,0x1b1b1b1b); + &xor ($acc,$tp2); + &mov ($tp2,$acc); + + &and ($acc,0x80808080); + &mov ($tmp,$acc); + &shr ($tmp,7); + &lea ($tp4,&DWP(0,$tp2,$tp2)); + &sub ($acc,$tmp); + &and ($tp4,0xfefefefe); + &and ($acc,0x1b1b1b1b); + &xor ($tp2,$tp1); # tp2^tp1 + &xor ($acc,$tp4); + &mov ($tp4,$acc); + + &and ($acc,0x80808080); + &mov ($tmp,$acc); + &shr ($tmp,7); + &lea ($tp8,&DWP(0,$tp4,$tp4)); + &xor ($tp4,$tp1); # tp4^tp1 + &sub ($acc,$tmp); + &and ($tp8,0xfefefefe); + &and ($acc,0x1b1b1b1b); + &rotl ($tp1,8); # = ROTATE(tp1,8) + &xor ($tp8,$acc); + + &mov ($tmp,&DWP(4*($i+1),$key)); # modulo-scheduled load + + &xor ($tp1,$tp2); + &xor ($tp2,$tp8); + &xor ($tp1,$tp4); + &rotl ($tp2,24); + &xor ($tp4,$tp8); + &xor ($tp1,$tp8); # ^= tp8^(tp4^tp1)^(tp2^tp1) + &rotl ($tp4,16); + &xor ($tp1,$tp2); # ^= ROTATE(tp8^tp2^tp1,24) + &rotl ($tp8,8); + &xor ($tp1,$tp4); # ^= ROTATE(tp8^tp4^tp1,16) + &mov ($tp2,$tmp); + &xor ($tp1,$tp8); # ^= ROTATE(tp8,8) + + &mov (&DWP(4*$i,$key),$tp1); } # int AES_set_decrypt_key(const unsigned char *userKey, const int bits, # AES_KEY *key) -&public_label("AES_Td"); -&public_label("AES_Te"); &function_begin_B("AES_set_decrypt_key"); - &mov ("eax",&wparam(0)); - &mov ("ecx",&wparam(1)); - &mov ("edx",&wparam(2)); - &sub ("esp",12); - &mov (&DWP(0,"esp"),"eax"); - &mov (&DWP(4,"esp"),"ecx"); - &mov (&DWP(8,"esp"),"edx"); - &call ("AES_set_encrypt_key"); - &add ("esp",12); + &call ("_x86_AES_set_encrypt_key"); &cmp ("eax",0); &je (&label("proceed")); &ret (); @@ -1749,8 +2935,7 @@ sub deckey() &lea ("ecx",&DWP(0,"","ecx",4)); &lea ("edi",&DWP(0,"esi","ecx",4)); # pointer to last chunk - &align (4); - &set_label("invert"); # invert order of chunks + &set_label("invert",4); # invert order of chunks &mov ("eax",&DWP(0,"esi")); &mov ("ebx",&DWP(4,"esi")); &mov ("ecx",&DWP(0,"edi")); @@ -1772,27 +2957,24 @@ sub deckey() &cmp ("esi","edi"); &jne (&label("invert")); - &call (&label("pic_point")); - &set_label("pic_point"); - blindpop("ebp"); - &lea ("edi",&DWP(&label("AES_Td")."-".&label("pic_point"),"ebp")); - &lea ("ebp",&DWP(&label("AES_Te")."-".&label("pic_point"),"ebp")); - &add ("ebp",1024*4); # skip to Te4 - - &mov ("esi",&wparam(2)); - &mov ("ecx",&DWP(240,"esi")); # pull number of rounds - &dec ("ecx"); - &align (4); - &set_label("permute"); # permute the key schedule - &add ("esi",16); - &deckey (0,"esi","ebp","edi"); - &deckey (4,"esi","ebp","edi"); - &deckey (8,"esi","ebp","edi"); - &deckey (12,"esi","ebp","edi"); - &dec ("ecx"); - &jnz (&label("permute")); + &mov ($key,&wparam(2)); + &mov ($acc,&DWP(240,$key)); # pull number of rounds + &lea ($acc,&DWP(-2,$acc,$acc)); + &lea ($acc,&DWP(0,$key,$acc,8)); + &mov (&wparam(2),$acc); + + &mov ($s0,&DWP(16,$key)); # modulo-scheduled load + &set_label("permute",4); # permute the key schedule + &add ($key,16); + &deckey (0,$key,$s0,$s1,$s2,$s3); + &deckey (1,$key,$s1,$s2,$s3,$s0); + &deckey (2,$key,$s2,$s3,$s0,$s1); + &deckey (3,$key,$s3,$s0,$s1,$s2); + &cmp ($key,&wparam(2)); + &jb (&label("permute")); &xor ("eax","eax"); # return success &function_end("AES_set_decrypt_key"); +&asciz("AES for x86, CRYPTOGAMS by <appro\@openssl.org>"); &asm_finish(); |