1160814Ssimon#!/usr/bin/env perl 2160814Ssimon# 3160814Ssimon# ==================================================================== 4160814Ssimon# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL 5238405Sjkim# project. The module is, however, dual licensed under OpenSSL and 6238405Sjkim# CRYPTOGAMS licenses depending on where you obtain it. For further 7238405Sjkim# details see http://www.openssl.org/~appro/cryptogams/. 8160814Ssimon# ==================================================================== 9160814Ssimon# 10238405Sjkim# Version 4.3. 11160814Ssimon# 12160814Ssimon# You might fail to appreciate this module performance from the first 13160814Ssimon# try. If compared to "vanilla" linux-ia32-icc target, i.e. considered 14160814Ssimon# to be *the* best Intel C compiler without -KPIC, performance appears 15160814Ssimon# to be virtually identical... But try to re-configure with shared 16160814Ssimon# library support... Aha! Intel compiler "suddenly" lags behind by 30% 17160814Ssimon# [on P4, more on others]:-) And if compared to position-independent 18160814Ssimon# code generated by GNU C, this code performs *more* than *twice* as 19160814Ssimon# fast! Yes, all this buzz about PIC means that unlike other hand- 20160814Ssimon# coded implementations, this one was explicitly designed to be safe 21160814Ssimon# to use even in shared library context... This also means that this 22160814Ssimon# code isn't necessarily absolutely fastest "ever," because in order 23160814Ssimon# to achieve position independence an extra register has to be 24160814Ssimon# off-loaded to stack, which affects the benchmark result. 25160814Ssimon# 26160814Ssimon# Special note about instruction choice. Do you recall RC4_INT code 27160814Ssimon# performing poorly on P4? It might be the time to figure out why. 28160814Ssimon# RC4_INT code implies effective address calculations in base+offset*4 29160814Ssimon# form. Trouble is that it seems that offset scaling turned to be 30160814Ssimon# critical path... At least eliminating scaling resulted in 2.8x RC4 31160814Ssimon# performance improvement [as you might recall]. As AES code is hungry 32160814Ssimon# for scaling too, I [try to] avoid the latter by favoring off-by-2 33160814Ssimon# shifts and masking the result with 0xFF<<2 instead of "boring" 0xFF. 34160814Ssimon# 35160814Ssimon# As was shown by Dean Gaudet <dean@arctic.org>, the above note turned 36160814Ssimon# void. Performance improvement with off-by-2 shifts was observed on 37160814Ssimon# intermediate implementation, which was spilling yet another register 38160814Ssimon# to stack... Final offset*4 code below runs just a tad faster on P4, 39160814Ssimon# but exhibits up to 10% improvement on other cores. 40160814Ssimon# 41160814Ssimon# Second version is "monolithic" replacement for aes_core.c, which in 42238405Sjkim# addition to AES_[de|en]crypt implements private_AES_set_[de|en]cryption_key. 43160814Ssimon# This made it possible to implement little-endian variant of the 44160814Ssimon# algorithm without modifying the base C code. Motivating factor for 45160814Ssimon# the undertaken effort was that it appeared that in tight IA-32 46160814Ssimon# register window little-endian flavor could achieve slightly higher 47160814Ssimon# Instruction Level Parallelism, and it indeed resulted in up to 15% 48291721Sjkim# better performance on most recent ��-archs... 49160814Ssimon# 50160814Ssimon# Third version adds AES_cbc_encrypt implementation, which resulted in 51160814Ssimon# up to 40% performance imrovement of CBC benchmark results. 40% was 52160814Ssimon# observed on P4 core, where "overall" imrovement coefficient, i.e. if 53160814Ssimon# compared to PIC generated by GCC and in CBC mode, was observed to be 54160814Ssimon# as large as 4x:-) CBC performance is virtually identical to ECB now 55160814Ssimon# and on some platforms even better, e.g. 17.6 "small" cycles/byte on 56160814Ssimon# Opteron, because certain function prologues and epilogues are 57160814Ssimon# effectively taken out of the loop... 58160814Ssimon# 59160814Ssimon# Version 3.2 implements compressed tables and prefetch of these tables 60160814Ssimon# in CBC[!] mode. Former means that 3/4 of table references are now 61160814Ssimon# misaligned, which unfortunately has negative impact on elder IA-32 62160814Ssimon# implementations, Pentium suffered 30% penalty, PIII - 10%. 63160814Ssimon# 64160814Ssimon# Version 3.3 avoids L1 cache aliasing between stack frame and 65160814Ssimon# S-boxes, and 3.4 - L1 cache aliasing even between key schedule. The 66160814Ssimon# latter is achieved by copying the key schedule to controlled place in 67160814Ssimon# stack. This unfortunately has rather strong impact on small block CBC 68160814Ssimon# performance, ~2x deterioration on 16-byte block if compared to 3.3. 69160814Ssimon# 70162911Ssimon# Version 3.5 checks if there is L1 cache aliasing between user-supplied 71162911Ssimon# key schedule and S-boxes and abstains from copying the former if 72162911Ssimon# there is no. This allows end-user to consciously retain small block 73162911Ssimon# performance by aligning key schedule in specific manner. 74162911Ssimon# 75162911Ssimon# Version 3.6 compresses Td4 to 256 bytes and prefetches it in ECB. 76162911Ssimon# 77160814Ssimon# Current ECB performance numbers for 128-bit key in CPU cycles per 78160814Ssimon# processed byte [measure commonly used by AES benchmarkers] are: 79160814Ssimon# 80160814Ssimon# small footprint fully unrolled 81160814Ssimon# P4 24 22 82160814Ssimon# AMD K8 20 19 83160814Ssimon# PIII 25 23 84160814Ssimon# Pentium 81 78 85238405Sjkim# 86238405Sjkim# Version 3.7 reimplements outer rounds as "compact." Meaning that 87238405Sjkim# first and last rounds reference compact 256 bytes S-box. This means 88238405Sjkim# that first round consumes a lot more CPU cycles and that encrypt 89238405Sjkim# and decrypt performance becomes asymmetric. Encrypt performance 90238405Sjkim# drops by 10-12%, while decrypt - by 20-25%:-( 256 bytes S-box is 91238405Sjkim# aggressively pre-fetched. 92238405Sjkim# 93238405Sjkim# Version 4.0 effectively rolls back to 3.6 and instead implements 94238405Sjkim# additional set of functions, _[x86|sse]_AES_[en|de]crypt_compact, 95238405Sjkim# which use exclusively 256 byte S-box. These functions are to be 96238405Sjkim# called in modes not concealing plain text, such as ECB, or when 97238405Sjkim# we're asked to process smaller amount of data [or unconditionally 98238405Sjkim# on hyper-threading CPU]. Currently it's called unconditionally from 99238405Sjkim# AES_[en|de]crypt, which affects all modes, but CBC. CBC routine 100238405Sjkim# still needs to be modified to switch between slower and faster 101238405Sjkim# mode when appropriate... But in either case benchmark landscape 102238405Sjkim# changes dramatically and below numbers are CPU cycles per processed 103238405Sjkim# byte for 128-bit key. 104238405Sjkim# 105238405Sjkim# ECB encrypt ECB decrypt CBC large chunk 106238405Sjkim# P4 56[60] 84[100] 23 107238405Sjkim# AMD K8 48[44] 70[79] 18 108238405Sjkim# PIII 41[50] 61[91] 24 109238405Sjkim# Core 2 32[38] 45[70] 18.5 110238405Sjkim# Pentium 120 160 77 111238405Sjkim# 112238405Sjkim# Version 4.1 switches to compact S-box even in key schedule setup. 113238405Sjkim# 114238405Sjkim# Version 4.2 prefetches compact S-box in every SSE round or in other 115238405Sjkim# words every cache-line is *guaranteed* to be accessed within ~50 116238405Sjkim# cycles window. Why just SSE? Because it's needed on hyper-threading 117238405Sjkim# CPU! Which is also why it's prefetched with 64 byte stride. Best 118238405Sjkim# part is that it has no negative effect on performance:-) 119238405Sjkim# 120238405Sjkim# Version 4.3 implements switch between compact and non-compact block 121238405Sjkim# functions in AES_cbc_encrypt depending on how much data was asked 122238405Sjkim# to be processed in one stroke. 123238405Sjkim# 124238405Sjkim###################################################################### 125238405Sjkim# Timing attacks are classified in two classes: synchronous when 126238405Sjkim# attacker consciously initiates cryptographic operation and collects 127238405Sjkim# timing data of various character afterwards, and asynchronous when 128238405Sjkim# malicious code is executed on same CPU simultaneously with AES, 129238405Sjkim# instruments itself and performs statistical analysis of this data. 130238405Sjkim# 131238405Sjkim# As far as synchronous attacks go the root to the AES timing 132238405Sjkim# vulnerability is twofold. Firstly, of 256 S-box elements at most 160 133238405Sjkim# are referred to in single 128-bit block operation. Well, in C 134238405Sjkim# implementation with 4 distinct tables it's actually as little as 40 135238405Sjkim# references per 256 elements table, but anyway... Secondly, even 136238405Sjkim# though S-box elements are clustered into smaller amount of cache- 137238405Sjkim# lines, smaller than 160 and even 40, it turned out that for certain 138238405Sjkim# plain-text pattern[s] or simply put chosen plain-text and given key 139238405Sjkim# few cache-lines remain unaccessed during block operation. Now, if 140238405Sjkim# attacker can figure out this access pattern, he can deduct the key 141238405Sjkim# [or at least part of it]. The natural way to mitigate this kind of 142238405Sjkim# attacks is to minimize the amount of cache-lines in S-box and/or 143238405Sjkim# prefetch them to ensure that every one is accessed for more uniform 144238405Sjkim# timing. But note that *if* plain-text was concealed in such way that 145238405Sjkim# input to block function is distributed *uniformly*, then attack 146238405Sjkim# wouldn't apply. Now note that some encryption modes, most notably 147238405Sjkim# CBC, do mask the plain-text in this exact way [secure cipher output 148238405Sjkim# is distributed uniformly]. Yes, one still might find input that 149238405Sjkim# would reveal the information about given key, but if amount of 150238405Sjkim# candidate inputs to be tried is larger than amount of possible key 151238405Sjkim# combinations then attack becomes infeasible. This is why revised 152238405Sjkim# AES_cbc_encrypt "dares" to switch to larger S-box when larger chunk 153238405Sjkim# of data is to be processed in one stroke. The current size limit of 154238405Sjkim# 512 bytes is chosen to provide same [diminishigly low] probability 155238405Sjkim# for cache-line to remain untouched in large chunk operation with 156238405Sjkim# large S-box as for single block operation with compact S-box and 157238405Sjkim# surely needs more careful consideration... 158238405Sjkim# 159238405Sjkim# As for asynchronous attacks. There are two flavours: attacker code 160238405Sjkim# being interleaved with AES on hyper-threading CPU at *instruction* 161238405Sjkim# level, and two processes time sharing single core. As for latter. 162238405Sjkim# Two vectors. 1. Given that attacker process has higher priority, 163238405Sjkim# yield execution to process performing AES just before timer fires 164238405Sjkim# off the scheduler, immediately regain control of CPU and analyze the 165238405Sjkim# cache state. For this attack to be efficient attacker would have to 166238405Sjkim# effectively slow down the operation by several *orders* of magnitute, 167238405Sjkim# by ratio of time slice to duration of handful of AES rounds, which 168238405Sjkim# unlikely to remain unnoticed. Not to mention that this also means 169238405Sjkim# that he would spend correspondigly more time to collect enough 170238405Sjkim# statistical data to mount the attack. It's probably appropriate to 171238405Sjkim# say that if adeversary reckons that this attack is beneficial and 172238405Sjkim# risks to be noticed, you probably have larger problems having him 173238405Sjkim# mere opportunity. In other words suggested code design expects you 174238405Sjkim# to preclude/mitigate this attack by overall system security design. 175238405Sjkim# 2. Attacker manages to make his code interrupt driven. In order for 176238405Sjkim# this kind of attack to be feasible, interrupt rate has to be high 177238405Sjkim# enough, again comparable to duration of handful of AES rounds. But 178238405Sjkim# is there interrupt source of such rate? Hardly, not even 1Gbps NIC 179238405Sjkim# generates interrupts at such raging rate... 180238405Sjkim# 181238405Sjkim# And now back to the former, hyper-threading CPU or more specifically 182238405Sjkim# Intel P4. Recall that asynchronous attack implies that malicious 183238405Sjkim# code instruments itself. And naturally instrumentation granularity 184238405Sjkim# has be noticeably lower than duration of codepath accessing S-box. 185238405Sjkim# Given that all cache-lines are accessed during that time that is. 186238405Sjkim# Current implementation accesses *all* cache-lines within ~50 cycles 187238405Sjkim# window, which is actually *less* than RDTSC latency on Intel P4! 188160814Ssimon 189238405Sjkim$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1; 190238405Sjkimpush(@INC,"${dir}","${dir}../../perlasm"); 191160814Ssimonrequire "x86asm.pl"; 192160814Ssimon 193238405Sjkim&asm_init($ARGV[0],"aes-586.pl",$x86only = $ARGV[$#ARGV] eq "386"); 194238405Sjkim&static_label("AES_Te"); 195238405Sjkim&static_label("AES_Td"); 196160814Ssimon 197160814Ssimon$s0="eax"; 198160814Ssimon$s1="ebx"; 199160814Ssimon$s2="ecx"; 200160814Ssimon$s3="edx"; 201160814Ssimon$key="edi"; 202160814Ssimon$acc="esi"; 203238405Sjkim$tbl="ebp"; 204160814Ssimon 205238405Sjkim# stack frame layout in _[x86|sse]_AES_* routines, frame is allocated 206238405Sjkim# by caller 207238405Sjkim$__ra=&DWP(0,"esp"); # return address 208238405Sjkim$__s0=&DWP(4,"esp"); # s0 backing store 209238405Sjkim$__s1=&DWP(8,"esp"); # s1 backing store 210238405Sjkim$__s2=&DWP(12,"esp"); # s2 backing store 211238405Sjkim$__s3=&DWP(16,"esp"); # s3 backing store 212238405Sjkim$__key=&DWP(20,"esp"); # pointer to key schedule 213238405Sjkim$__end=&DWP(24,"esp"); # pointer to end of key schedule 214238405Sjkim$__tbl=&DWP(28,"esp"); # %ebp backing store 215238405Sjkim 216238405Sjkim# stack frame layout in AES_[en|crypt] routines, which differs from 217238405Sjkim# above by 4 and overlaps by %ebp backing store 218238405Sjkim$_tbl=&DWP(24,"esp"); 219238405Sjkim$_esp=&DWP(28,"esp"); 220238405Sjkim 221238405Sjkimsub _data_word() { my $i; while(defined($i=shift)) { &data_word($i,$i); } } 222238405Sjkim 223238405Sjkim$speed_limit=512; # chunks smaller than $speed_limit are 224238405Sjkim # processed with compact routine in CBC mode 225160814Ssimon$small_footprint=1; # $small_footprint=1 code is ~5% slower [on 226291721Sjkim # recent ��-archs], but ~5 times smaller! 227160814Ssimon # I favor compact code to minimize cache 228160814Ssimon # contention and in hope to "collect" 5% back 229160814Ssimon # in real-life applications... 230238405Sjkim 231160814Ssimon$vertical_spin=0; # shift "verticaly" defaults to 0, because of 232160814Ssimon # its proof-of-concept status... 233160814Ssimon# Note that there is no decvert(), as well as last encryption round is 234160814Ssimon# performed with "horizontal" shifts. This is because this "vertical" 235160814Ssimon# implementation [one which groups shifts on a given $s[i] to form a 236160814Ssimon# "column," unlike "horizontal" one, which groups shifts on different 237160814Ssimon# $s[i] to form a "row"] is work in progress. It was observed to run 238160814Ssimon# few percents faster on Intel cores, but not AMD. On AMD K8 core it's 239160814Ssimon# whole 12% slower:-( So we face a trade-off... Shall it be resolved 240160814Ssimon# some day? Till then the code is considered experimental and by 241160814Ssimon# default remains dormant... 242160814Ssimon 243160814Ssimonsub encvert() 244160814Ssimon{ my ($te,@s) = @_; 245160814Ssimon my $v0 = $acc, $v1 = $key; 246160814Ssimon 247160814Ssimon &mov ($v0,$s[3]); # copy s3 248160814Ssimon &mov (&DWP(4,"esp"),$s[2]); # save s2 249160814Ssimon &mov ($v1,$s[0]); # copy s0 250160814Ssimon &mov (&DWP(8,"esp"),$s[1]); # save s1 251160814Ssimon 252160814Ssimon &movz ($s[2],&HB($s[0])); 253160814Ssimon &and ($s[0],0xFF); 254160814Ssimon &mov ($s[0],&DWP(0,$te,$s[0],8)); # s0>>0 255160814Ssimon &shr ($v1,16); 256160814Ssimon &mov ($s[3],&DWP(3,$te,$s[2],8)); # s0>>8 257160814Ssimon &movz ($s[1],&HB($v1)); 258160814Ssimon &and ($v1,0xFF); 259160814Ssimon &mov ($s[2],&DWP(2,$te,$v1,8)); # s0>>16 260160814Ssimon &mov ($v1,$v0); 261160814Ssimon &mov ($s[1],&DWP(1,$te,$s[1],8)); # s0>>24 262160814Ssimon 263160814Ssimon &and ($v0,0xFF); 264160814Ssimon &xor ($s[3],&DWP(0,$te,$v0,8)); # s3>>0 265160814Ssimon &movz ($v0,&HB($v1)); 266160814Ssimon &shr ($v1,16); 267160814Ssimon &xor ($s[2],&DWP(3,$te,$v0,8)); # s3>>8 268160814Ssimon &movz ($v0,&HB($v1)); 269160814Ssimon &and ($v1,0xFF); 270160814Ssimon &xor ($s[1],&DWP(2,$te,$v1,8)); # s3>>16 271160814Ssimon &mov ($v1,&DWP(4,"esp")); # restore s2 272160814Ssimon &xor ($s[0],&DWP(1,$te,$v0,8)); # s3>>24 273160814Ssimon 274160814Ssimon &mov ($v0,$v1); 275160814Ssimon &and ($v1,0xFF); 276160814Ssimon &xor ($s[2],&DWP(0,$te,$v1,8)); # s2>>0 277160814Ssimon &movz ($v1,&HB($v0)); 278160814Ssimon &shr ($v0,16); 279160814Ssimon &xor ($s[1],&DWP(3,$te,$v1,8)); # s2>>8 280160814Ssimon &movz ($v1,&HB($v0)); 281160814Ssimon &and ($v0,0xFF); 282160814Ssimon &xor ($s[0],&DWP(2,$te,$v0,8)); # s2>>16 283160814Ssimon &mov ($v0,&DWP(8,"esp")); # restore s1 284160814Ssimon &xor ($s[3],&DWP(1,$te,$v1,8)); # s2>>24 285160814Ssimon 286160814Ssimon &mov ($v1,$v0); 287160814Ssimon &and ($v0,0xFF); 288160814Ssimon &xor ($s[1],&DWP(0,$te,$v0,8)); # s1>>0 289160814Ssimon &movz ($v0,&HB($v1)); 290160814Ssimon &shr ($v1,16); 291160814Ssimon &xor ($s[0],&DWP(3,$te,$v0,8)); # s1>>8 292160814Ssimon &movz ($v0,&HB($v1)); 293160814Ssimon &and ($v1,0xFF); 294160814Ssimon &xor ($s[3],&DWP(2,$te,$v1,8)); # s1>>16 295238405Sjkim &mov ($key,$__key); # reincarnate v1 as key 296160814Ssimon &xor ($s[2],&DWP(1,$te,$v0,8)); # s1>>24 297160814Ssimon} 298160814Ssimon 299238405Sjkim# Another experimental routine, which features "horizontal spin," but 300238405Sjkim# eliminates one reference to stack. Strangely enough runs slower... 301238405Sjkimsub enchoriz() 302238405Sjkim{ my $v0 = $key, $v1 = $acc; 303238405Sjkim 304238405Sjkim &movz ($v0,&LB($s0)); # 3, 2, 1, 0* 305238405Sjkim &rotr ($s2,8); # 8,11,10, 9 306238405Sjkim &mov ($v1,&DWP(0,$te,$v0,8)); # 0 307238405Sjkim &movz ($v0,&HB($s1)); # 7, 6, 5*, 4 308238405Sjkim &rotr ($s3,16); # 13,12,15,14 309238405Sjkim &xor ($v1,&DWP(3,$te,$v0,8)); # 5 310238405Sjkim &movz ($v0,&HB($s2)); # 8,11,10*, 9 311238405Sjkim &rotr ($s0,16); # 1, 0, 3, 2 312238405Sjkim &xor ($v1,&DWP(2,$te,$v0,8)); # 10 313238405Sjkim &movz ($v0,&HB($s3)); # 13,12,15*,14 314238405Sjkim &xor ($v1,&DWP(1,$te,$v0,8)); # 15, t[0] collected 315238405Sjkim &mov ($__s0,$v1); # t[0] saved 316238405Sjkim 317238405Sjkim &movz ($v0,&LB($s1)); # 7, 6, 5, 4* 318238405Sjkim &shr ($s1,16); # -, -, 7, 6 319238405Sjkim &mov ($v1,&DWP(0,$te,$v0,8)); # 4 320238405Sjkim &movz ($v0,&LB($s3)); # 13,12,15,14* 321238405Sjkim &xor ($v1,&DWP(2,$te,$v0,8)); # 14 322238405Sjkim &movz ($v0,&HB($s0)); # 1, 0, 3*, 2 323238405Sjkim &and ($s3,0xffff0000); # 13,12, -, - 324238405Sjkim &xor ($v1,&DWP(1,$te,$v0,8)); # 3 325238405Sjkim &movz ($v0,&LB($s2)); # 8,11,10, 9* 326238405Sjkim &or ($s3,$s1); # 13,12, 7, 6 327238405Sjkim &xor ($v1,&DWP(3,$te,$v0,8)); # 9, t[1] collected 328238405Sjkim &mov ($s1,$v1); # s[1]=t[1] 329238405Sjkim 330238405Sjkim &movz ($v0,&LB($s0)); # 1, 0, 3, 2* 331238405Sjkim &shr ($s2,16); # -, -, 8,11 332238405Sjkim &mov ($v1,&DWP(2,$te,$v0,8)); # 2 333238405Sjkim &movz ($v0,&HB($s3)); # 13,12, 7*, 6 334238405Sjkim &xor ($v1,&DWP(1,$te,$v0,8)); # 7 335238405Sjkim &movz ($v0,&HB($s2)); # -, -, 8*,11 336238405Sjkim &xor ($v1,&DWP(0,$te,$v0,8)); # 8 337238405Sjkim &mov ($v0,$s3); 338238405Sjkim &shr ($v0,24); # 13 339238405Sjkim &xor ($v1,&DWP(3,$te,$v0,8)); # 13, t[2] collected 340238405Sjkim 341238405Sjkim &movz ($v0,&LB($s2)); # -, -, 8,11* 342238405Sjkim &shr ($s0,24); # 1* 343238405Sjkim &mov ($s2,&DWP(1,$te,$v0,8)); # 11 344238405Sjkim &xor ($s2,&DWP(3,$te,$s0,8)); # 1 345238405Sjkim &mov ($s0,$__s0); # s[0]=t[0] 346238405Sjkim &movz ($v0,&LB($s3)); # 13,12, 7, 6* 347238405Sjkim &shr ($s3,16); # , ,13,12 348238405Sjkim &xor ($s2,&DWP(2,$te,$v0,8)); # 6 349238405Sjkim &mov ($key,$__key); # reincarnate v0 as key 350238405Sjkim &and ($s3,0xff); # , ,13,12* 351238405Sjkim &mov ($s3,&DWP(0,$te,$s3,8)); # 12 352238405Sjkim &xor ($s3,$s2); # s[2]=t[3] collected 353238405Sjkim &mov ($s2,$v1); # s[2]=t[2] 354238405Sjkim} 355238405Sjkim 356238405Sjkim# More experimental code... SSE one... Even though this one eliminates 357238405Sjkim# *all* references to stack, it's not faster... 358238405Sjkimsub sse_encbody() 359238405Sjkim{ 360238405Sjkim &movz ($acc,&LB("eax")); # 0 361238405Sjkim &mov ("ecx",&DWP(0,$tbl,$acc,8)); # 0 362238405Sjkim &pshufw ("mm2","mm0",0x0d); # 7, 6, 3, 2 363238405Sjkim &movz ("edx",&HB("eax")); # 1 364238405Sjkim &mov ("edx",&DWP(3,$tbl,"edx",8)); # 1 365238405Sjkim &shr ("eax",16); # 5, 4 366238405Sjkim 367238405Sjkim &movz ($acc,&LB("ebx")); # 10 368238405Sjkim &xor ("ecx",&DWP(2,$tbl,$acc,8)); # 10 369238405Sjkim &pshufw ("mm6","mm4",0x08); # 13,12, 9, 8 370238405Sjkim &movz ($acc,&HB("ebx")); # 11 371238405Sjkim &xor ("edx",&DWP(1,$tbl,$acc,8)); # 11 372238405Sjkim &shr ("ebx",16); # 15,14 373238405Sjkim 374238405Sjkim &movz ($acc,&HB("eax")); # 5 375238405Sjkim &xor ("ecx",&DWP(3,$tbl,$acc,8)); # 5 376238405Sjkim &movq ("mm3",QWP(16,$key)); 377238405Sjkim &movz ($acc,&HB("ebx")); # 15 378238405Sjkim &xor ("ecx",&DWP(1,$tbl,$acc,8)); # 15 379238405Sjkim &movd ("mm0","ecx"); # t[0] collected 380238405Sjkim 381238405Sjkim &movz ($acc,&LB("eax")); # 4 382238405Sjkim &mov ("ecx",&DWP(0,$tbl,$acc,8)); # 4 383238405Sjkim &movd ("eax","mm2"); # 7, 6, 3, 2 384238405Sjkim &movz ($acc,&LB("ebx")); # 14 385238405Sjkim &xor ("ecx",&DWP(2,$tbl,$acc,8)); # 14 386238405Sjkim &movd ("ebx","mm6"); # 13,12, 9, 8 387238405Sjkim 388238405Sjkim &movz ($acc,&HB("eax")); # 3 389238405Sjkim &xor ("ecx",&DWP(1,$tbl,$acc,8)); # 3 390238405Sjkim &movz ($acc,&HB("ebx")); # 9 391238405Sjkim &xor ("ecx",&DWP(3,$tbl,$acc,8)); # 9 392238405Sjkim &movd ("mm1","ecx"); # t[1] collected 393238405Sjkim 394238405Sjkim &movz ($acc,&LB("eax")); # 2 395238405Sjkim &mov ("ecx",&DWP(2,$tbl,$acc,8)); # 2 396238405Sjkim &shr ("eax",16); # 7, 6 397238405Sjkim &punpckldq ("mm0","mm1"); # t[0,1] collected 398238405Sjkim &movz ($acc,&LB("ebx")); # 8 399238405Sjkim &xor ("ecx",&DWP(0,$tbl,$acc,8)); # 8 400238405Sjkim &shr ("ebx",16); # 13,12 401238405Sjkim 402238405Sjkim &movz ($acc,&HB("eax")); # 7 403238405Sjkim &xor ("ecx",&DWP(1,$tbl,$acc,8)); # 7 404238405Sjkim &pxor ("mm0","mm3"); 405238405Sjkim &movz ("eax",&LB("eax")); # 6 406238405Sjkim &xor ("edx",&DWP(2,$tbl,"eax",8)); # 6 407238405Sjkim &pshufw ("mm1","mm0",0x08); # 5, 4, 1, 0 408238405Sjkim &movz ($acc,&HB("ebx")); # 13 409238405Sjkim &xor ("ecx",&DWP(3,$tbl,$acc,8)); # 13 410238405Sjkim &xor ("ecx",&DWP(24,$key)); # t[2] 411238405Sjkim &movd ("mm4","ecx"); # t[2] collected 412238405Sjkim &movz ("ebx",&LB("ebx")); # 12 413238405Sjkim &xor ("edx",&DWP(0,$tbl,"ebx",8)); # 12 414238405Sjkim &shr ("ecx",16); 415238405Sjkim &movd ("eax","mm1"); # 5, 4, 1, 0 416238405Sjkim &mov ("ebx",&DWP(28,$key)); # t[3] 417238405Sjkim &xor ("ebx","edx"); 418238405Sjkim &movd ("mm5","ebx"); # t[3] collected 419238405Sjkim &and ("ebx",0xffff0000); 420238405Sjkim &or ("ebx","ecx"); 421238405Sjkim 422238405Sjkim &punpckldq ("mm4","mm5"); # t[2,3] collected 423238405Sjkim} 424238405Sjkim 425238405Sjkim###################################################################### 426238405Sjkim# "Compact" block function 427238405Sjkim###################################################################### 428238405Sjkim 429238405Sjkimsub enccompact() 430238405Sjkim{ my $Fn = mov; 431238405Sjkim while ($#_>5) { pop(@_); $Fn=sub{}; } 432238405Sjkim my ($i,$te,@s)=@_; 433238405Sjkim my $tmp = $key; 434238405Sjkim my $out = $i==3?$s[0]:$acc; 435238405Sjkim 436238405Sjkim # $Fn is used in first compact round and its purpose is to 437238405Sjkim # void restoration of some values from stack, so that after 438238405Sjkim # 4xenccompact with extra argument $key value is left there... 439238405Sjkim if ($i==3) { &$Fn ($key,$__key); }##%edx 440238405Sjkim else { &mov ($out,$s[0]); } 441238405Sjkim &and ($out,0xFF); 442238405Sjkim if ($i==1) { &shr ($s[0],16); }#%ebx[1] 443238405Sjkim if ($i==2) { &shr ($s[0],24); }#%ecx[2] 444238405Sjkim &movz ($out,&BP(-128,$te,$out,1)); 445238405Sjkim 446238405Sjkim if ($i==3) { $tmp=$s[1]; }##%eax 447238405Sjkim &movz ($tmp,&HB($s[1])); 448238405Sjkim &movz ($tmp,&BP(-128,$te,$tmp,1)); 449238405Sjkim &shl ($tmp,8); 450238405Sjkim &xor ($out,$tmp); 451238405Sjkim 452238405Sjkim if ($i==3) { $tmp=$s[2]; &mov ($s[1],$__s0); }##%ebx 453238405Sjkim else { &mov ($tmp,$s[2]); 454238405Sjkim &shr ($tmp,16); } 455238405Sjkim if ($i==2) { &and ($s[1],0xFF); }#%edx[2] 456238405Sjkim &and ($tmp,0xFF); 457238405Sjkim &movz ($tmp,&BP(-128,$te,$tmp,1)); 458238405Sjkim &shl ($tmp,16); 459238405Sjkim &xor ($out,$tmp); 460238405Sjkim 461238405Sjkim if ($i==3) { $tmp=$s[3]; &mov ($s[2],$__s1); }##%ecx 462238405Sjkim elsif($i==2){ &movz ($tmp,&HB($s[3])); }#%ebx[2] 463238405Sjkim else { &mov ($tmp,$s[3]); 464238405Sjkim &shr ($tmp,24); } 465238405Sjkim &movz ($tmp,&BP(-128,$te,$tmp,1)); 466238405Sjkim &shl ($tmp,24); 467238405Sjkim &xor ($out,$tmp); 468238405Sjkim if ($i<2) { &mov (&DWP(4+4*$i,"esp"),$out); } 469238405Sjkim if ($i==3) { &mov ($s[3],$acc); } 470238405Sjkim &comment(); 471238405Sjkim} 472238405Sjkim 473238405Sjkimsub enctransform() 474238405Sjkim{ my @s = ($s0,$s1,$s2,$s3); 475238405Sjkim my $i = shift; 476238405Sjkim my $tmp = $tbl; 477238405Sjkim my $r2 = $key ; 478238405Sjkim 479238405Sjkim &mov ($acc,$s[$i]); 480238405Sjkim &and ($acc,0x80808080); 481238405Sjkim &mov ($tmp,$acc); 482238405Sjkim &shr ($tmp,7); 483238405Sjkim &lea ($r2,&DWP(0,$s[$i],$s[$i])); 484238405Sjkim &sub ($acc,$tmp); 485238405Sjkim &and ($r2,0xfefefefe); 486238405Sjkim &and ($acc,0x1b1b1b1b); 487238405Sjkim &mov ($tmp,$s[$i]); 488238405Sjkim &xor ($acc,$r2); # r2 489238405Sjkim 490238405Sjkim &xor ($s[$i],$acc); # r0 ^ r2 491238405Sjkim &rotl ($s[$i],24); 492238405Sjkim &xor ($s[$i],$acc) # ROTATE(r2^r0,24) ^ r2 493238405Sjkim &rotr ($tmp,16); 494238405Sjkim &xor ($s[$i],$tmp); 495238405Sjkim &rotr ($tmp,8); 496238405Sjkim &xor ($s[$i],$tmp); 497238405Sjkim} 498238405Sjkim 499238405Sjkim&function_begin_B("_x86_AES_encrypt_compact"); 500238405Sjkim # note that caller is expected to allocate stack frame for me! 501238405Sjkim &mov ($__key,$key); # save key 502238405Sjkim 503238405Sjkim &xor ($s0,&DWP(0,$key)); # xor with key 504238405Sjkim &xor ($s1,&DWP(4,$key)); 505238405Sjkim &xor ($s2,&DWP(8,$key)); 506238405Sjkim &xor ($s3,&DWP(12,$key)); 507238405Sjkim 508238405Sjkim &mov ($acc,&DWP(240,$key)); # load key->rounds 509238405Sjkim &lea ($acc,&DWP(-2,$acc,$acc)); 510238405Sjkim &lea ($acc,&DWP(0,$key,$acc,8)); 511238405Sjkim &mov ($__end,$acc); # end of key schedule 512238405Sjkim 513238405Sjkim # prefetch Te4 514238405Sjkim &mov ($key,&DWP(0-128,$tbl)); 515238405Sjkim &mov ($acc,&DWP(32-128,$tbl)); 516238405Sjkim &mov ($key,&DWP(64-128,$tbl)); 517238405Sjkim &mov ($acc,&DWP(96-128,$tbl)); 518238405Sjkim &mov ($key,&DWP(128-128,$tbl)); 519238405Sjkim &mov ($acc,&DWP(160-128,$tbl)); 520238405Sjkim &mov ($key,&DWP(192-128,$tbl)); 521238405Sjkim &mov ($acc,&DWP(224-128,$tbl)); 522238405Sjkim 523238405Sjkim &set_label("loop",16); 524238405Sjkim 525238405Sjkim &enccompact(0,$tbl,$s0,$s1,$s2,$s3,1); 526238405Sjkim &enccompact(1,$tbl,$s1,$s2,$s3,$s0,1); 527238405Sjkim &enccompact(2,$tbl,$s2,$s3,$s0,$s1,1); 528238405Sjkim &enccompact(3,$tbl,$s3,$s0,$s1,$s2,1); 529238405Sjkim &enctransform(2); 530238405Sjkim &enctransform(3); 531238405Sjkim &enctransform(0); 532238405Sjkim &enctransform(1); 533238405Sjkim &mov ($key,$__key); 534238405Sjkim &mov ($tbl,$__tbl); 535238405Sjkim &add ($key,16); # advance rd_key 536238405Sjkim &xor ($s0,&DWP(0,$key)); 537238405Sjkim &xor ($s1,&DWP(4,$key)); 538238405Sjkim &xor ($s2,&DWP(8,$key)); 539238405Sjkim &xor ($s3,&DWP(12,$key)); 540238405Sjkim 541238405Sjkim &cmp ($key,$__end); 542238405Sjkim &mov ($__key,$key); 543238405Sjkim &jb (&label("loop")); 544238405Sjkim 545238405Sjkim &enccompact(0,$tbl,$s0,$s1,$s2,$s3); 546238405Sjkim &enccompact(1,$tbl,$s1,$s2,$s3,$s0); 547238405Sjkim &enccompact(2,$tbl,$s2,$s3,$s0,$s1); 548238405Sjkim &enccompact(3,$tbl,$s3,$s0,$s1,$s2); 549238405Sjkim 550238405Sjkim &xor ($s0,&DWP(16,$key)); 551238405Sjkim &xor ($s1,&DWP(20,$key)); 552238405Sjkim &xor ($s2,&DWP(24,$key)); 553238405Sjkim &xor ($s3,&DWP(28,$key)); 554238405Sjkim 555238405Sjkim &ret (); 556238405Sjkim&function_end_B("_x86_AES_encrypt_compact"); 557238405Sjkim 558238405Sjkim###################################################################### 559238405Sjkim# "Compact" SSE block function. 560238405Sjkim###################################################################### 561238405Sjkim# 562238405Sjkim# Performance is not actually extraordinary in comparison to pure 563238405Sjkim# x86 code. In particular encrypt performance is virtually the same. 564238405Sjkim# Decrypt performance on the other hand is 15-20% better on newer 565291721Sjkim# ��-archs [but we're thankful for *any* improvement here], and ~50% 566238405Sjkim# better on PIII:-) And additionally on the pros side this code 567238405Sjkim# eliminates redundant references to stack and thus relieves/ 568238405Sjkim# minimizes the pressure on the memory bus. 569238405Sjkim# 570238405Sjkim# MMX register layout lsb 571238405Sjkim# +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ 572238405Sjkim# | mm4 | mm0 | 573238405Sjkim# +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ 574238405Sjkim# | s3 | s2 | s1 | s0 | 575238405Sjkim# +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ 576238405Sjkim# |15|14|13|12|11|10| 9| 8| 7| 6| 5| 4| 3| 2| 1| 0| 577238405Sjkim# +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ 578238405Sjkim# 579238405Sjkim# Indexes translate as s[N/4]>>(8*(N%4)), e.g. 5 means s1>>8. 580238405Sjkim# In this terms encryption and decryption "compact" permutation 581238405Sjkim# matrices can be depicted as following: 582238405Sjkim# 583238405Sjkim# encryption lsb # decryption lsb 584238405Sjkim# +----++----+----+----+----+ # +----++----+----+----+----+ 585238405Sjkim# | t0 || 15 | 10 | 5 | 0 | # | t0 || 7 | 10 | 13 | 0 | 586238405Sjkim# +----++----+----+----+----+ # +----++----+----+----+----+ 587238405Sjkim# | t1 || 3 | 14 | 9 | 4 | # | t1 || 11 | 14 | 1 | 4 | 588238405Sjkim# +----++----+----+----+----+ # +----++----+----+----+----+ 589238405Sjkim# | t2 || 7 | 2 | 13 | 8 | # | t2 || 15 | 2 | 5 | 8 | 590238405Sjkim# +----++----+----+----+----+ # +----++----+----+----+----+ 591238405Sjkim# | t3 || 11 | 6 | 1 | 12 | # | t3 || 3 | 6 | 9 | 12 | 592238405Sjkim# +----++----+----+----+----+ # +----++----+----+----+----+ 593238405Sjkim# 594238405Sjkim###################################################################### 595238405Sjkim# Why not xmm registers? Short answer. It was actually tested and 596238405Sjkim# was not any faster, but *contrary*, most notably on Intel CPUs. 597238405Sjkim# Longer answer. Main advantage of using mm registers is that movd 598238405Sjkim# latency is lower, especially on Intel P4. While arithmetic 599238405Sjkim# instructions are twice as many, they can be scheduled every cycle 600238405Sjkim# and not every second one when they are operating on xmm register, 601238405Sjkim# so that "arithmetic throughput" remains virtually the same. And 602238405Sjkim# finally the code can be executed even on elder SSE-only CPUs:-) 603238405Sjkim 604238405Sjkimsub sse_enccompact() 605238405Sjkim{ 606238405Sjkim &pshufw ("mm1","mm0",0x08); # 5, 4, 1, 0 607238405Sjkim &pshufw ("mm5","mm4",0x0d); # 15,14,11,10 608238405Sjkim &movd ("eax","mm1"); # 5, 4, 1, 0 609238405Sjkim &movd ("ebx","mm5"); # 15,14,11,10 610238405Sjkim 611238405Sjkim &movz ($acc,&LB("eax")); # 0 612238405Sjkim &movz ("ecx",&BP(-128,$tbl,$acc,1)); # 0 613238405Sjkim &pshufw ("mm2","mm0",0x0d); # 7, 6, 3, 2 614238405Sjkim &movz ("edx",&HB("eax")); # 1 615238405Sjkim &movz ("edx",&BP(-128,$tbl,"edx",1)); # 1 616238405Sjkim &shl ("edx",8); # 1 617238405Sjkim &shr ("eax",16); # 5, 4 618238405Sjkim 619238405Sjkim &movz ($acc,&LB("ebx")); # 10 620238405Sjkim &movz ($acc,&BP(-128,$tbl,$acc,1)); # 10 621238405Sjkim &shl ($acc,16); # 10 622238405Sjkim &or ("ecx",$acc); # 10 623238405Sjkim &pshufw ("mm6","mm4",0x08); # 13,12, 9, 8 624238405Sjkim &movz ($acc,&HB("ebx")); # 11 625238405Sjkim &movz ($acc,&BP(-128,$tbl,$acc,1)); # 11 626238405Sjkim &shl ($acc,24); # 11 627238405Sjkim &or ("edx",$acc); # 11 628238405Sjkim &shr ("ebx",16); # 15,14 629238405Sjkim 630238405Sjkim &movz ($acc,&HB("eax")); # 5 631238405Sjkim &movz ($acc,&BP(-128,$tbl,$acc,1)); # 5 632238405Sjkim &shl ($acc,8); # 5 633238405Sjkim &or ("ecx",$acc); # 5 634238405Sjkim &movz ($acc,&HB("ebx")); # 15 635238405Sjkim &movz ($acc,&BP(-128,$tbl,$acc,1)); # 15 636238405Sjkim &shl ($acc,24); # 15 637238405Sjkim &or ("ecx",$acc); # 15 638238405Sjkim &movd ("mm0","ecx"); # t[0] collected 639238405Sjkim 640238405Sjkim &movz ($acc,&LB("eax")); # 4 641238405Sjkim &movz ("ecx",&BP(-128,$tbl,$acc,1)); # 4 642238405Sjkim &movd ("eax","mm2"); # 7, 6, 3, 2 643238405Sjkim &movz ($acc,&LB("ebx")); # 14 644238405Sjkim &movz ($acc,&BP(-128,$tbl,$acc,1)); # 14 645238405Sjkim &shl ($acc,16); # 14 646238405Sjkim &or ("ecx",$acc); # 14 647238405Sjkim 648238405Sjkim &movd ("ebx","mm6"); # 13,12, 9, 8 649238405Sjkim &movz ($acc,&HB("eax")); # 3 650238405Sjkim &movz ($acc,&BP(-128,$tbl,$acc,1)); # 3 651238405Sjkim &shl ($acc,24); # 3 652238405Sjkim &or ("ecx",$acc); # 3 653238405Sjkim &movz ($acc,&HB("ebx")); # 9 654238405Sjkim &movz ($acc,&BP(-128,$tbl,$acc,1)); # 9 655238405Sjkim &shl ($acc,8); # 9 656238405Sjkim &or ("ecx",$acc); # 9 657238405Sjkim &movd ("mm1","ecx"); # t[1] collected 658238405Sjkim 659238405Sjkim &movz ($acc,&LB("ebx")); # 8 660238405Sjkim &movz ("ecx",&BP(-128,$tbl,$acc,1)); # 8 661238405Sjkim &shr ("ebx",16); # 13,12 662238405Sjkim &movz ($acc,&LB("eax")); # 2 663238405Sjkim &movz ($acc,&BP(-128,$tbl,$acc,1)); # 2 664238405Sjkim &shl ($acc,16); # 2 665238405Sjkim &or ("ecx",$acc); # 2 666238405Sjkim &shr ("eax",16); # 7, 6 667238405Sjkim 668238405Sjkim &punpckldq ("mm0","mm1"); # t[0,1] collected 669238405Sjkim 670238405Sjkim &movz ($acc,&HB("eax")); # 7 671238405Sjkim &movz ($acc,&BP(-128,$tbl,$acc,1)); # 7 672238405Sjkim &shl ($acc,24); # 7 673238405Sjkim &or ("ecx",$acc); # 7 674238405Sjkim &and ("eax",0xff); # 6 675238405Sjkim &movz ("eax",&BP(-128,$tbl,"eax",1)); # 6 676238405Sjkim &shl ("eax",16); # 6 677238405Sjkim &or ("edx","eax"); # 6 678238405Sjkim &movz ($acc,&HB("ebx")); # 13 679238405Sjkim &movz ($acc,&BP(-128,$tbl,$acc,1)); # 13 680238405Sjkim &shl ($acc,8); # 13 681238405Sjkim &or ("ecx",$acc); # 13 682238405Sjkim &movd ("mm4","ecx"); # t[2] collected 683238405Sjkim &and ("ebx",0xff); # 12 684238405Sjkim &movz ("ebx",&BP(-128,$tbl,"ebx",1)); # 12 685238405Sjkim &or ("edx","ebx"); # 12 686238405Sjkim &movd ("mm5","edx"); # t[3] collected 687238405Sjkim 688238405Sjkim &punpckldq ("mm4","mm5"); # t[2,3] collected 689238405Sjkim} 690238405Sjkim 691238405Sjkim if (!$x86only) { 692238405Sjkim&function_begin_B("_sse_AES_encrypt_compact"); 693238405Sjkim &pxor ("mm0",&QWP(0,$key)); # 7, 6, 5, 4, 3, 2, 1, 0 694238405Sjkim &pxor ("mm4",&QWP(8,$key)); # 15,14,13,12,11,10, 9, 8 695238405Sjkim 696238405Sjkim # note that caller is expected to allocate stack frame for me! 697238405Sjkim &mov ($acc,&DWP(240,$key)); # load key->rounds 698238405Sjkim &lea ($acc,&DWP(-2,$acc,$acc)); 699238405Sjkim &lea ($acc,&DWP(0,$key,$acc,8)); 700238405Sjkim &mov ($__end,$acc); # end of key schedule 701238405Sjkim 702238405Sjkim &mov ($s0,0x1b1b1b1b); # magic constant 703238405Sjkim &mov (&DWP(8,"esp"),$s0); 704238405Sjkim &mov (&DWP(12,"esp"),$s0); 705238405Sjkim 706238405Sjkim # prefetch Te4 707238405Sjkim &mov ($s0,&DWP(0-128,$tbl)); 708238405Sjkim &mov ($s1,&DWP(32-128,$tbl)); 709238405Sjkim &mov ($s2,&DWP(64-128,$tbl)); 710238405Sjkim &mov ($s3,&DWP(96-128,$tbl)); 711238405Sjkim &mov ($s0,&DWP(128-128,$tbl)); 712238405Sjkim &mov ($s1,&DWP(160-128,$tbl)); 713238405Sjkim &mov ($s2,&DWP(192-128,$tbl)); 714238405Sjkim &mov ($s3,&DWP(224-128,$tbl)); 715238405Sjkim 716238405Sjkim &set_label("loop",16); 717238405Sjkim &sse_enccompact(); 718238405Sjkim &add ($key,16); 719238405Sjkim &cmp ($key,$__end); 720238405Sjkim &ja (&label("out")); 721238405Sjkim 722238405Sjkim &movq ("mm2",&QWP(8,"esp")); 723238405Sjkim &pxor ("mm3","mm3"); &pxor ("mm7","mm7"); 724238405Sjkim &movq ("mm1","mm0"); &movq ("mm5","mm4"); # r0 725238405Sjkim &pcmpgtb("mm3","mm0"); &pcmpgtb("mm7","mm4"); 726238405Sjkim &pand ("mm3","mm2"); &pand ("mm7","mm2"); 727238405Sjkim &pshufw ("mm2","mm0",0xb1); &pshufw ("mm6","mm4",0xb1);# ROTATE(r0,16) 728238405Sjkim &paddb ("mm0","mm0"); &paddb ("mm4","mm4"); 729238405Sjkim &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # = r2 730238405Sjkim &pshufw ("mm3","mm2",0xb1); &pshufw ("mm7","mm6",0xb1);# r0 731238405Sjkim &pxor ("mm1","mm0"); &pxor ("mm5","mm4"); # r0^r2 732238405Sjkim &pxor ("mm0","mm2"); &pxor ("mm4","mm6"); # ^= ROTATE(r0,16) 733238405Sjkim 734238405Sjkim &movq ("mm2","mm3"); &movq ("mm6","mm7"); 735238405Sjkim &pslld ("mm3",8); &pslld ("mm7",8); 736238405Sjkim &psrld ("mm2",24); &psrld ("mm6",24); 737238405Sjkim &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= r0<<8 738238405Sjkim &pxor ("mm0","mm2"); &pxor ("mm4","mm6"); # ^= r0>>24 739238405Sjkim 740238405Sjkim &movq ("mm3","mm1"); &movq ("mm7","mm5"); 741238405Sjkim &movq ("mm2",&QWP(0,$key)); &movq ("mm6",&QWP(8,$key)); 742238405Sjkim &psrld ("mm1",8); &psrld ("mm5",8); 743238405Sjkim &mov ($s0,&DWP(0-128,$tbl)); 744238405Sjkim &pslld ("mm3",24); &pslld ("mm7",24); 745238405Sjkim &mov ($s1,&DWP(64-128,$tbl)); 746238405Sjkim &pxor ("mm0","mm1"); &pxor ("mm4","mm5"); # ^= (r2^r0)<<8 747238405Sjkim &mov ($s2,&DWP(128-128,$tbl)); 748238405Sjkim &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= (r2^r0)>>24 749238405Sjkim &mov ($s3,&DWP(192-128,$tbl)); 750238405Sjkim 751238405Sjkim &pxor ("mm0","mm2"); &pxor ("mm4","mm6"); 752238405Sjkim &jmp (&label("loop")); 753238405Sjkim 754238405Sjkim &set_label("out",16); 755238405Sjkim &pxor ("mm0",&QWP(0,$key)); 756238405Sjkim &pxor ("mm4",&QWP(8,$key)); 757238405Sjkim 758238405Sjkim &ret (); 759238405Sjkim&function_end_B("_sse_AES_encrypt_compact"); 760238405Sjkim } 761238405Sjkim 762238405Sjkim###################################################################### 763238405Sjkim# Vanilla block function. 764238405Sjkim###################################################################### 765238405Sjkim 766160814Ssimonsub encstep() 767160814Ssimon{ my ($i,$te,@s) = @_; 768160814Ssimon my $tmp = $key; 769160814Ssimon my $out = $i==3?$s[0]:$acc; 770160814Ssimon 771160814Ssimon # lines marked with #%e?x[i] denote "reordered" instructions... 772238405Sjkim if ($i==3) { &mov ($key,$__key); }##%edx 773160814Ssimon else { &mov ($out,$s[0]); 774160814Ssimon &and ($out,0xFF); } 775160814Ssimon if ($i==1) { &shr ($s[0],16); }#%ebx[1] 776160814Ssimon if ($i==2) { &shr ($s[0],24); }#%ecx[2] 777160814Ssimon &mov ($out,&DWP(0,$te,$out,8)); 778160814Ssimon 779160814Ssimon if ($i==3) { $tmp=$s[1]; }##%eax 780160814Ssimon &movz ($tmp,&HB($s[1])); 781160814Ssimon &xor ($out,&DWP(3,$te,$tmp,8)); 782160814Ssimon 783238405Sjkim if ($i==3) { $tmp=$s[2]; &mov ($s[1],$__s0); }##%ebx 784160814Ssimon else { &mov ($tmp,$s[2]); 785160814Ssimon &shr ($tmp,16); } 786160814Ssimon if ($i==2) { &and ($s[1],0xFF); }#%edx[2] 787160814Ssimon &and ($tmp,0xFF); 788160814Ssimon &xor ($out,&DWP(2,$te,$tmp,8)); 789160814Ssimon 790238405Sjkim if ($i==3) { $tmp=$s[3]; &mov ($s[2],$__s1); }##%ecx 791160814Ssimon elsif($i==2){ &movz ($tmp,&HB($s[3])); }#%ebx[2] 792160814Ssimon else { &mov ($tmp,$s[3]); 793160814Ssimon &shr ($tmp,24) } 794160814Ssimon &xor ($out,&DWP(1,$te,$tmp,8)); 795160814Ssimon if ($i<2) { &mov (&DWP(4+4*$i,"esp"),$out); } 796160814Ssimon if ($i==3) { &mov ($s[3],$acc); } 797160814Ssimon &comment(); 798160814Ssimon} 799160814Ssimon 800160814Ssimonsub enclast() 801160814Ssimon{ my ($i,$te,@s)=@_; 802160814Ssimon my $tmp = $key; 803160814Ssimon my $out = $i==3?$s[0]:$acc; 804160814Ssimon 805238405Sjkim if ($i==3) { &mov ($key,$__key); }##%edx 806160814Ssimon else { &mov ($out,$s[0]); } 807160814Ssimon &and ($out,0xFF); 808160814Ssimon if ($i==1) { &shr ($s[0],16); }#%ebx[1] 809160814Ssimon if ($i==2) { &shr ($s[0],24); }#%ecx[2] 810160814Ssimon &mov ($out,&DWP(2,$te,$out,8)); 811160814Ssimon &and ($out,0x000000ff); 812160814Ssimon 813160814Ssimon if ($i==3) { $tmp=$s[1]; }##%eax 814160814Ssimon &movz ($tmp,&HB($s[1])); 815160814Ssimon &mov ($tmp,&DWP(0,$te,$tmp,8)); 816160814Ssimon &and ($tmp,0x0000ff00); 817160814Ssimon &xor ($out,$tmp); 818160814Ssimon 819238405Sjkim if ($i==3) { $tmp=$s[2]; &mov ($s[1],$__s0); }##%ebx 820238405Sjkim else { &mov ($tmp,$s[2]); 821160814Ssimon &shr ($tmp,16); } 822160814Ssimon if ($i==2) { &and ($s[1],0xFF); }#%edx[2] 823160814Ssimon &and ($tmp,0xFF); 824160814Ssimon &mov ($tmp,&DWP(0,$te,$tmp,8)); 825160814Ssimon &and ($tmp,0x00ff0000); 826160814Ssimon &xor ($out,$tmp); 827160814Ssimon 828238405Sjkim if ($i==3) { $tmp=$s[3]; &mov ($s[2],$__s1); }##%ecx 829160814Ssimon elsif($i==2){ &movz ($tmp,&HB($s[3])); }#%ebx[2] 830160814Ssimon else { &mov ($tmp,$s[3]); 831160814Ssimon &shr ($tmp,24); } 832160814Ssimon &mov ($tmp,&DWP(2,$te,$tmp,8)); 833160814Ssimon &and ($tmp,0xff000000); 834160814Ssimon &xor ($out,$tmp); 835160814Ssimon if ($i<2) { &mov (&DWP(4+4*$i,"esp"),$out); } 836160814Ssimon if ($i==3) { &mov ($s[3],$acc); } 837160814Ssimon} 838160814Ssimon 839160814Ssimon&function_begin_B("_x86_AES_encrypt"); 840160814Ssimon if ($vertical_spin) { 841160814Ssimon # I need high parts of volatile registers to be accessible... 842160814Ssimon &exch ($s1="edi",$key="ebx"); 843160814Ssimon &mov ($s2="esi",$acc="ecx"); 844160814Ssimon } 845160814Ssimon 846160814Ssimon # note that caller is expected to allocate stack frame for me! 847238405Sjkim &mov ($__key,$key); # save key 848160814Ssimon 849160814Ssimon &xor ($s0,&DWP(0,$key)); # xor with key 850160814Ssimon &xor ($s1,&DWP(4,$key)); 851160814Ssimon &xor ($s2,&DWP(8,$key)); 852160814Ssimon &xor ($s3,&DWP(12,$key)); 853160814Ssimon 854160814Ssimon &mov ($acc,&DWP(240,$key)); # load key->rounds 855160814Ssimon 856160814Ssimon if ($small_footprint) { 857160814Ssimon &lea ($acc,&DWP(-2,$acc,$acc)); 858160814Ssimon &lea ($acc,&DWP(0,$key,$acc,8)); 859238405Sjkim &mov ($__end,$acc); # end of key schedule 860238405Sjkim 861238405Sjkim &set_label("loop",16); 862160814Ssimon if ($vertical_spin) { 863238405Sjkim &encvert($tbl,$s0,$s1,$s2,$s3); 864160814Ssimon } else { 865238405Sjkim &encstep(0,$tbl,$s0,$s1,$s2,$s3); 866238405Sjkim &encstep(1,$tbl,$s1,$s2,$s3,$s0); 867238405Sjkim &encstep(2,$tbl,$s2,$s3,$s0,$s1); 868238405Sjkim &encstep(3,$tbl,$s3,$s0,$s1,$s2); 869160814Ssimon } 870160814Ssimon &add ($key,16); # advance rd_key 871160814Ssimon &xor ($s0,&DWP(0,$key)); 872160814Ssimon &xor ($s1,&DWP(4,$key)); 873160814Ssimon &xor ($s2,&DWP(8,$key)); 874160814Ssimon &xor ($s3,&DWP(12,$key)); 875238405Sjkim &cmp ($key,$__end); 876238405Sjkim &mov ($__key,$key); 877160814Ssimon &jb (&label("loop")); 878160814Ssimon } 879160814Ssimon else { 880160814Ssimon &cmp ($acc,10); 881160814Ssimon &jle (&label("10rounds")); 882160814Ssimon &cmp ($acc,12); 883160814Ssimon &jle (&label("12rounds")); 884160814Ssimon 885238405Sjkim &set_label("14rounds",4); 886160814Ssimon for ($i=1;$i<3;$i++) { 887160814Ssimon if ($vertical_spin) { 888238405Sjkim &encvert($tbl,$s0,$s1,$s2,$s3); 889160814Ssimon } else { 890238405Sjkim &encstep(0,$tbl,$s0,$s1,$s2,$s3); 891238405Sjkim &encstep(1,$tbl,$s1,$s2,$s3,$s0); 892238405Sjkim &encstep(2,$tbl,$s2,$s3,$s0,$s1); 893238405Sjkim &encstep(3,$tbl,$s3,$s0,$s1,$s2); 894160814Ssimon } 895160814Ssimon &xor ($s0,&DWP(16*$i+0,$key)); 896160814Ssimon &xor ($s1,&DWP(16*$i+4,$key)); 897160814Ssimon &xor ($s2,&DWP(16*$i+8,$key)); 898160814Ssimon &xor ($s3,&DWP(16*$i+12,$key)); 899160814Ssimon } 900160814Ssimon &add ($key,32); 901238405Sjkim &mov ($__key,$key); # advance rd_key 902238405Sjkim &set_label("12rounds",4); 903160814Ssimon for ($i=1;$i<3;$i++) { 904160814Ssimon if ($vertical_spin) { 905238405Sjkim &encvert($tbl,$s0,$s1,$s2,$s3); 906160814Ssimon } else { 907238405Sjkim &encstep(0,$tbl,$s0,$s1,$s2,$s3); 908238405Sjkim &encstep(1,$tbl,$s1,$s2,$s3,$s0); 909238405Sjkim &encstep(2,$tbl,$s2,$s3,$s0,$s1); 910238405Sjkim &encstep(3,$tbl,$s3,$s0,$s1,$s2); 911160814Ssimon } 912160814Ssimon &xor ($s0,&DWP(16*$i+0,$key)); 913160814Ssimon &xor ($s1,&DWP(16*$i+4,$key)); 914160814Ssimon &xor ($s2,&DWP(16*$i+8,$key)); 915160814Ssimon &xor ($s3,&DWP(16*$i+12,$key)); 916160814Ssimon } 917160814Ssimon &add ($key,32); 918238405Sjkim &mov ($__key,$key); # advance rd_key 919238405Sjkim &set_label("10rounds",4); 920160814Ssimon for ($i=1;$i<10;$i++) { 921160814Ssimon if ($vertical_spin) { 922238405Sjkim &encvert($tbl,$s0,$s1,$s2,$s3); 923160814Ssimon } else { 924238405Sjkim &encstep(0,$tbl,$s0,$s1,$s2,$s3); 925238405Sjkim &encstep(1,$tbl,$s1,$s2,$s3,$s0); 926238405Sjkim &encstep(2,$tbl,$s2,$s3,$s0,$s1); 927238405Sjkim &encstep(3,$tbl,$s3,$s0,$s1,$s2); 928160814Ssimon } 929160814Ssimon &xor ($s0,&DWP(16*$i+0,$key)); 930160814Ssimon &xor ($s1,&DWP(16*$i+4,$key)); 931160814Ssimon &xor ($s2,&DWP(16*$i+8,$key)); 932160814Ssimon &xor ($s3,&DWP(16*$i+12,$key)); 933160814Ssimon } 934160814Ssimon } 935160814Ssimon 936160814Ssimon if ($vertical_spin) { 937160814Ssimon # "reincarnate" some registers for "horizontal" spin... 938160814Ssimon &mov ($s1="ebx",$key="edi"); 939160814Ssimon &mov ($s2="ecx",$acc="esi"); 940160814Ssimon } 941238405Sjkim &enclast(0,$tbl,$s0,$s1,$s2,$s3); 942238405Sjkim &enclast(1,$tbl,$s1,$s2,$s3,$s0); 943238405Sjkim &enclast(2,$tbl,$s2,$s3,$s0,$s1); 944238405Sjkim &enclast(3,$tbl,$s3,$s0,$s1,$s2); 945160814Ssimon 946160814Ssimon &add ($key,$small_footprint?16:160); 947160814Ssimon &xor ($s0,&DWP(0,$key)); 948160814Ssimon &xor ($s1,&DWP(4,$key)); 949160814Ssimon &xor ($s2,&DWP(8,$key)); 950160814Ssimon &xor ($s3,&DWP(12,$key)); 951160814Ssimon 952160814Ssimon &ret (); 953160814Ssimon 954160814Ssimon&set_label("AES_Te",64); # Yes! I keep it in the code segment! 955160814Ssimon &_data_word(0xa56363c6, 0x847c7cf8, 0x997777ee, 0x8d7b7bf6); 956160814Ssimon &_data_word(0x0df2f2ff, 0xbd6b6bd6, 0xb16f6fde, 0x54c5c591); 957160814Ssimon &_data_word(0x50303060, 0x03010102, 0xa96767ce, 0x7d2b2b56); 958160814Ssimon &_data_word(0x19fefee7, 0x62d7d7b5, 0xe6abab4d, 0x9a7676ec); 959160814Ssimon &_data_word(0x45caca8f, 0x9d82821f, 0x40c9c989, 0x877d7dfa); 960160814Ssimon &_data_word(0x15fafaef, 0xeb5959b2, 0xc947478e, 0x0bf0f0fb); 961160814Ssimon &_data_word(0xecadad41, 0x67d4d4b3, 0xfda2a25f, 0xeaafaf45); 962160814Ssimon &_data_word(0xbf9c9c23, 0xf7a4a453, 0x967272e4, 0x5bc0c09b); 963160814Ssimon &_data_word(0xc2b7b775, 0x1cfdfde1, 0xae93933d, 0x6a26264c); 964160814Ssimon &_data_word(0x5a36366c, 0x413f3f7e, 0x02f7f7f5, 0x4fcccc83); 965160814Ssimon &_data_word(0x5c343468, 0xf4a5a551, 0x34e5e5d1, 0x08f1f1f9); 966160814Ssimon &_data_word(0x937171e2, 0x73d8d8ab, 0x53313162, 0x3f15152a); 967160814Ssimon &_data_word(0x0c040408, 0x52c7c795, 0x65232346, 0x5ec3c39d); 968160814Ssimon &_data_word(0x28181830, 0xa1969637, 0x0f05050a, 0xb59a9a2f); 969160814Ssimon &_data_word(0x0907070e, 0x36121224, 0x9b80801b, 0x3de2e2df); 970160814Ssimon &_data_word(0x26ebebcd, 0x6927274e, 0xcdb2b27f, 0x9f7575ea); 971160814Ssimon &_data_word(0x1b090912, 0x9e83831d, 0x742c2c58, 0x2e1a1a34); 972160814Ssimon &_data_word(0x2d1b1b36, 0xb26e6edc, 0xee5a5ab4, 0xfba0a05b); 973160814Ssimon &_data_word(0xf65252a4, 0x4d3b3b76, 0x61d6d6b7, 0xceb3b37d); 974160814Ssimon &_data_word(0x7b292952, 0x3ee3e3dd, 0x712f2f5e, 0x97848413); 975160814Ssimon &_data_word(0xf55353a6, 0x68d1d1b9, 0x00000000, 0x2cededc1); 976160814Ssimon &_data_word(0x60202040, 0x1ffcfce3, 0xc8b1b179, 0xed5b5bb6); 977160814Ssimon &_data_word(0xbe6a6ad4, 0x46cbcb8d, 0xd9bebe67, 0x4b393972); 978160814Ssimon &_data_word(0xde4a4a94, 0xd44c4c98, 0xe85858b0, 0x4acfcf85); 979160814Ssimon &_data_word(0x6bd0d0bb, 0x2aefefc5, 0xe5aaaa4f, 0x16fbfbed); 980160814Ssimon &_data_word(0xc5434386, 0xd74d4d9a, 0x55333366, 0x94858511); 981160814Ssimon &_data_word(0xcf45458a, 0x10f9f9e9, 0x06020204, 0x817f7ffe); 982160814Ssimon &_data_word(0xf05050a0, 0x443c3c78, 0xba9f9f25, 0xe3a8a84b); 983160814Ssimon &_data_word(0xf35151a2, 0xfea3a35d, 0xc0404080, 0x8a8f8f05); 984160814Ssimon &_data_word(0xad92923f, 0xbc9d9d21, 0x48383870, 0x04f5f5f1); 985160814Ssimon &_data_word(0xdfbcbc63, 0xc1b6b677, 0x75dadaaf, 0x63212142); 986160814Ssimon &_data_word(0x30101020, 0x1affffe5, 0x0ef3f3fd, 0x6dd2d2bf); 987160814Ssimon &_data_word(0x4ccdcd81, 0x140c0c18, 0x35131326, 0x2fececc3); 988160814Ssimon &_data_word(0xe15f5fbe, 0xa2979735, 0xcc444488, 0x3917172e); 989160814Ssimon &_data_word(0x57c4c493, 0xf2a7a755, 0x827e7efc, 0x473d3d7a); 990160814Ssimon &_data_word(0xac6464c8, 0xe75d5dba, 0x2b191932, 0x957373e6); 991160814Ssimon &_data_word(0xa06060c0, 0x98818119, 0xd14f4f9e, 0x7fdcdca3); 992160814Ssimon &_data_word(0x66222244, 0x7e2a2a54, 0xab90903b, 0x8388880b); 993160814Ssimon &_data_word(0xca46468c, 0x29eeeec7, 0xd3b8b86b, 0x3c141428); 994160814Ssimon &_data_word(0x79dedea7, 0xe25e5ebc, 0x1d0b0b16, 0x76dbdbad); 995160814Ssimon &_data_word(0x3be0e0db, 0x56323264, 0x4e3a3a74, 0x1e0a0a14); 996160814Ssimon &_data_word(0xdb494992, 0x0a06060c, 0x6c242448, 0xe45c5cb8); 997160814Ssimon &_data_word(0x5dc2c29f, 0x6ed3d3bd, 0xefacac43, 0xa66262c4); 998160814Ssimon &_data_word(0xa8919139, 0xa4959531, 0x37e4e4d3, 0x8b7979f2); 999160814Ssimon &_data_word(0x32e7e7d5, 0x43c8c88b, 0x5937376e, 0xb76d6dda); 1000160814Ssimon &_data_word(0x8c8d8d01, 0x64d5d5b1, 0xd24e4e9c, 0xe0a9a949); 1001160814Ssimon &_data_word(0xb46c6cd8, 0xfa5656ac, 0x07f4f4f3, 0x25eaeacf); 1002160814Ssimon &_data_word(0xaf6565ca, 0x8e7a7af4, 0xe9aeae47, 0x18080810); 1003160814Ssimon &_data_word(0xd5baba6f, 0x887878f0, 0x6f25254a, 0x722e2e5c); 1004160814Ssimon &_data_word(0x241c1c38, 0xf1a6a657, 0xc7b4b473, 0x51c6c697); 1005160814Ssimon &_data_word(0x23e8e8cb, 0x7cdddda1, 0x9c7474e8, 0x211f1f3e); 1006160814Ssimon &_data_word(0xdd4b4b96, 0xdcbdbd61, 0x868b8b0d, 0x858a8a0f); 1007160814Ssimon &_data_word(0x907070e0, 0x423e3e7c, 0xc4b5b571, 0xaa6666cc); 1008160814Ssimon &_data_word(0xd8484890, 0x05030306, 0x01f6f6f7, 0x120e0e1c); 1009160814Ssimon &_data_word(0xa36161c2, 0x5f35356a, 0xf95757ae, 0xd0b9b969); 1010160814Ssimon &_data_word(0x91868617, 0x58c1c199, 0x271d1d3a, 0xb99e9e27); 1011160814Ssimon &_data_word(0x38e1e1d9, 0x13f8f8eb, 0xb398982b, 0x33111122); 1012160814Ssimon &_data_word(0xbb6969d2, 0x70d9d9a9, 0x898e8e07, 0xa7949433); 1013160814Ssimon &_data_word(0xb69b9b2d, 0x221e1e3c, 0x92878715, 0x20e9e9c9); 1014160814Ssimon &_data_word(0x49cece87, 0xff5555aa, 0x78282850, 0x7adfdfa5); 1015160814Ssimon &_data_word(0x8f8c8c03, 0xf8a1a159, 0x80898909, 0x170d0d1a); 1016160814Ssimon &_data_word(0xdabfbf65, 0x31e6e6d7, 0xc6424284, 0xb86868d0); 1017160814Ssimon &_data_word(0xc3414182, 0xb0999929, 0x772d2d5a, 0x110f0f1e); 1018160814Ssimon &_data_word(0xcbb0b07b, 0xfc5454a8, 0xd6bbbb6d, 0x3a16162c); 1019238405Sjkim 1020238405Sjkim#Te4 # four copies of Te4 to choose from to avoid L1 aliasing 1021238405Sjkim &data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5); 1022238405Sjkim &data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76); 1023238405Sjkim &data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0); 1024238405Sjkim &data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0); 1025238405Sjkim &data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc); 1026238405Sjkim &data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15); 1027238405Sjkim &data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a); 1028238405Sjkim &data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75); 1029238405Sjkim &data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0); 1030238405Sjkim &data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84); 1031238405Sjkim &data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b); 1032238405Sjkim &data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf); 1033238405Sjkim &data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85); 1034238405Sjkim &data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8); 1035238405Sjkim &data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5); 1036238405Sjkim &data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2); 1037238405Sjkim &data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17); 1038238405Sjkim &data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73); 1039238405Sjkim &data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88); 1040238405Sjkim &data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb); 1041238405Sjkim &data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c); 1042238405Sjkim &data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79); 1043238405Sjkim &data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9); 1044238405Sjkim &data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08); 1045238405Sjkim &data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6); 1046238405Sjkim &data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a); 1047238405Sjkim &data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e); 1048238405Sjkim &data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e); 1049238405Sjkim &data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94); 1050238405Sjkim &data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf); 1051238405Sjkim &data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68); 1052238405Sjkim &data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16); 1053238405Sjkim 1054238405Sjkim &data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5); 1055238405Sjkim &data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76); 1056238405Sjkim &data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0); 1057238405Sjkim &data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0); 1058238405Sjkim &data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc); 1059238405Sjkim &data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15); 1060238405Sjkim &data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a); 1061238405Sjkim &data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75); 1062238405Sjkim &data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0); 1063238405Sjkim &data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84); 1064238405Sjkim &data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b); 1065238405Sjkim &data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf); 1066238405Sjkim &data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85); 1067238405Sjkim &data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8); 1068238405Sjkim &data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5); 1069238405Sjkim &data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2); 1070238405Sjkim &data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17); 1071238405Sjkim &data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73); 1072238405Sjkim &data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88); 1073238405Sjkim &data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb); 1074238405Sjkim &data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c); 1075238405Sjkim &data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79); 1076238405Sjkim &data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9); 1077238405Sjkim &data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08); 1078238405Sjkim &data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6); 1079238405Sjkim &data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a); 1080238405Sjkim &data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e); 1081238405Sjkim &data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e); 1082238405Sjkim &data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94); 1083238405Sjkim &data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf); 1084238405Sjkim &data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68); 1085238405Sjkim &data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16); 1086238405Sjkim 1087238405Sjkim &data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5); 1088238405Sjkim &data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76); 1089238405Sjkim &data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0); 1090238405Sjkim &data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0); 1091238405Sjkim &data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc); 1092238405Sjkim &data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15); 1093238405Sjkim &data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a); 1094238405Sjkim &data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75); 1095238405Sjkim &data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0); 1096238405Sjkim &data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84); 1097238405Sjkim &data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b); 1098238405Sjkim &data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf); 1099238405Sjkim &data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85); 1100238405Sjkim &data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8); 1101238405Sjkim &data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5); 1102238405Sjkim &data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2); 1103238405Sjkim &data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17); 1104238405Sjkim &data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73); 1105238405Sjkim &data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88); 1106238405Sjkim &data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb); 1107238405Sjkim &data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c); 1108238405Sjkim &data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79); 1109238405Sjkim &data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9); 1110238405Sjkim &data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08); 1111238405Sjkim &data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6); 1112238405Sjkim &data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a); 1113238405Sjkim &data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e); 1114238405Sjkim &data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e); 1115238405Sjkim &data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94); 1116238405Sjkim &data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf); 1117238405Sjkim &data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68); 1118238405Sjkim &data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16); 1119238405Sjkim 1120238405Sjkim &data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5); 1121238405Sjkim &data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76); 1122238405Sjkim &data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0); 1123238405Sjkim &data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0); 1124238405Sjkim &data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc); 1125238405Sjkim &data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15); 1126238405Sjkim &data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a); 1127238405Sjkim &data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75); 1128238405Sjkim &data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0); 1129238405Sjkim &data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84); 1130238405Sjkim &data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b); 1131238405Sjkim &data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf); 1132238405Sjkim &data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85); 1133238405Sjkim &data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8); 1134238405Sjkim &data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5); 1135238405Sjkim &data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2); 1136238405Sjkim &data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17); 1137238405Sjkim &data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73); 1138238405Sjkim &data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88); 1139238405Sjkim &data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb); 1140238405Sjkim &data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c); 1141238405Sjkim &data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79); 1142238405Sjkim &data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9); 1143238405Sjkim &data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08); 1144238405Sjkim &data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6); 1145238405Sjkim &data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a); 1146238405Sjkim &data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e); 1147238405Sjkim &data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e); 1148238405Sjkim &data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94); 1149238405Sjkim &data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf); 1150238405Sjkim &data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68); 1151238405Sjkim &data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16); 1152160814Ssimon#rcon: 1153160814Ssimon &data_word(0x00000001, 0x00000002, 0x00000004, 0x00000008); 1154160814Ssimon &data_word(0x00000010, 0x00000020, 0x00000040, 0x00000080); 1155238405Sjkim &data_word(0x0000001b, 0x00000036, 0x00000000, 0x00000000); 1156238405Sjkim &data_word(0x00000000, 0x00000000, 0x00000000, 0x00000000); 1157160814Ssimon&function_end_B("_x86_AES_encrypt"); 1158160814Ssimon 1159160814Ssimon# void AES_encrypt (const void *inp,void *out,const AES_KEY *key); 1160160814Ssimon&function_begin("AES_encrypt"); 1161160814Ssimon &mov ($acc,&wparam(0)); # load inp 1162160814Ssimon &mov ($key,&wparam(2)); # load key 1163160814Ssimon 1164160814Ssimon &mov ($s0,"esp"); 1165238405Sjkim &sub ("esp",36); 1166238405Sjkim &and ("esp",-64); # align to cache-line 1167160814Ssimon 1168238405Sjkim # place stack frame just "above" the key schedule 1169238405Sjkim &lea ($s1,&DWP(-64-63,$key)); 1170238405Sjkim &sub ($s1,"esp"); 1171238405Sjkim &neg ($s1); 1172238405Sjkim &and ($s1,0x3C0); # modulo 1024, but aligned to cache-line 1173238405Sjkim &sub ("esp",$s1); 1174238405Sjkim &add ("esp",4); # 4 is reserved for caller's return address 1175238405Sjkim &mov ($_esp,$s0); # save stack pointer 1176238405Sjkim 1177160814Ssimon &call (&label("pic_point")); # make it PIC! 1178160814Ssimon &set_label("pic_point"); 1179238405Sjkim &blindpop($tbl); 1180238405Sjkim &picmeup($s0,"OPENSSL_ia32cap_P",$tbl,&label("pic_point")) if (!$x86only); 1181238405Sjkim &lea ($tbl,&DWP(&label("AES_Te")."-".&label("pic_point"),$tbl)); 1182160814Ssimon 1183238405Sjkim # pick Te4 copy which can't "overlap" with stack frame or key schedule 1184238405Sjkim &lea ($s1,&DWP(768-4,"esp")); 1185238405Sjkim &sub ($s1,$tbl); 1186238405Sjkim &and ($s1,0x300); 1187238405Sjkim &lea ($tbl,&DWP(2048+128,$tbl,$s1)); 1188238405Sjkim 1189238405Sjkim if (!$x86only) { 1190238405Sjkim &bt (&DWP(0,$s0),25); # check for SSE bit 1191238405Sjkim &jnc (&label("x86")); 1192238405Sjkim 1193238405Sjkim &movq ("mm0",&QWP(0,$acc)); 1194238405Sjkim &movq ("mm4",&QWP(8,$acc)); 1195238405Sjkim &call ("_sse_AES_encrypt_compact"); 1196238405Sjkim &mov ("esp",$_esp); # restore stack pointer 1197238405Sjkim &mov ($acc,&wparam(1)); # load out 1198238405Sjkim &movq (&QWP(0,$acc),"mm0"); # write output data 1199238405Sjkim &movq (&QWP(8,$acc),"mm4"); 1200238405Sjkim &emms (); 1201238405Sjkim &function_end_A(); 1202238405Sjkim } 1203238405Sjkim &set_label("x86",16); 1204238405Sjkim &mov ($_tbl,$tbl); 1205160814Ssimon &mov ($s0,&DWP(0,$acc)); # load input data 1206160814Ssimon &mov ($s1,&DWP(4,$acc)); 1207160814Ssimon &mov ($s2,&DWP(8,$acc)); 1208160814Ssimon &mov ($s3,&DWP(12,$acc)); 1209238405Sjkim &call ("_x86_AES_encrypt_compact"); 1210238405Sjkim &mov ("esp",$_esp); # restore stack pointer 1211160814Ssimon &mov ($acc,&wparam(1)); # load out 1212160814Ssimon &mov (&DWP(0,$acc),$s0); # write output data 1213160814Ssimon &mov (&DWP(4,$acc),$s1); 1214160814Ssimon &mov (&DWP(8,$acc),$s2); 1215160814Ssimon &mov (&DWP(12,$acc),$s3); 1216160814Ssimon&function_end("AES_encrypt"); 1217160814Ssimon 1218238405Sjkim#--------------------------------------------------------------------# 1219160814Ssimon 1220238405Sjkim###################################################################### 1221238405Sjkim# "Compact" block function 1222238405Sjkim###################################################################### 1223238405Sjkim 1224238405Sjkimsub deccompact() 1225238405Sjkim{ my $Fn = mov; 1226238405Sjkim while ($#_>5) { pop(@_); $Fn=sub{}; } 1227238405Sjkim my ($i,$td,@s)=@_; 1228238405Sjkim my $tmp = $key; 1229238405Sjkim my $out = $i==3?$s[0]:$acc; 1230238405Sjkim 1231238405Sjkim # $Fn is used in first compact round and its purpose is to 1232238405Sjkim # void restoration of some values from stack, so that after 1233238405Sjkim # 4xdeccompact with extra argument $key, $s0 and $s1 values 1234238405Sjkim # are left there... 1235238405Sjkim if($i==3) { &$Fn ($key,$__key); } 1236238405Sjkim else { &mov ($out,$s[0]); } 1237238405Sjkim &and ($out,0xFF); 1238238405Sjkim &movz ($out,&BP(-128,$td,$out,1)); 1239238405Sjkim 1240238405Sjkim if ($i==3) { $tmp=$s[1]; } 1241238405Sjkim &movz ($tmp,&HB($s[1])); 1242238405Sjkim &movz ($tmp,&BP(-128,$td,$tmp,1)); 1243238405Sjkim &shl ($tmp,8); 1244238405Sjkim &xor ($out,$tmp); 1245238405Sjkim 1246238405Sjkim if ($i==3) { $tmp=$s[2]; &mov ($s[1],$acc); } 1247238405Sjkim else { mov ($tmp,$s[2]); } 1248238405Sjkim &shr ($tmp,16); 1249238405Sjkim &and ($tmp,0xFF); 1250238405Sjkim &movz ($tmp,&BP(-128,$td,$tmp,1)); 1251238405Sjkim &shl ($tmp,16); 1252238405Sjkim &xor ($out,$tmp); 1253238405Sjkim 1254238405Sjkim if ($i==3) { $tmp=$s[3]; &$Fn ($s[2],$__s1); } 1255238405Sjkim else { &mov ($tmp,$s[3]); } 1256238405Sjkim &shr ($tmp,24); 1257238405Sjkim &movz ($tmp,&BP(-128,$td,$tmp,1)); 1258238405Sjkim &shl ($tmp,24); 1259238405Sjkim &xor ($out,$tmp); 1260238405Sjkim if ($i<2) { &mov (&DWP(4+4*$i,"esp"),$out); } 1261238405Sjkim if ($i==3) { &$Fn ($s[3],$__s0); } 1262238405Sjkim} 1263238405Sjkim 1264238405Sjkim# must be called with 2,3,0,1 as argument sequence!!! 1265238405Sjkimsub dectransform() 1266238405Sjkim{ my @s = ($s0,$s1,$s2,$s3); 1267238405Sjkim my $i = shift; 1268238405Sjkim my $tmp = $key; 1269238405Sjkim my $tp2 = @s[($i+2)%4]; $tp2 = @s[2] if ($i==1); 1270238405Sjkim my $tp4 = @s[($i+3)%4]; $tp4 = @s[3] if ($i==1); 1271238405Sjkim my $tp8 = $tbl; 1272238405Sjkim 1273238405Sjkim &mov ($acc,$s[$i]); 1274238405Sjkim &and ($acc,0x80808080); 1275238405Sjkim &mov ($tmp,$acc); 1276238405Sjkim &shr ($tmp,7); 1277238405Sjkim &lea ($tp2,&DWP(0,$s[$i],$s[$i])); 1278238405Sjkim &sub ($acc,$tmp); 1279238405Sjkim &and ($tp2,0xfefefefe); 1280238405Sjkim &and ($acc,0x1b1b1b1b); 1281238405Sjkim &xor ($acc,$tp2); 1282238405Sjkim &mov ($tp2,$acc); 1283238405Sjkim 1284238405Sjkim &and ($acc,0x80808080); 1285238405Sjkim &mov ($tmp,$acc); 1286238405Sjkim &shr ($tmp,7); 1287238405Sjkim &lea ($tp4,&DWP(0,$tp2,$tp2)); 1288238405Sjkim &sub ($acc,$tmp); 1289238405Sjkim &and ($tp4,0xfefefefe); 1290238405Sjkim &and ($acc,0x1b1b1b1b); 1291238405Sjkim &xor ($tp2,$s[$i]); # tp2^tp1 1292238405Sjkim &xor ($acc,$tp4); 1293238405Sjkim &mov ($tp4,$acc); 1294238405Sjkim 1295238405Sjkim &and ($acc,0x80808080); 1296238405Sjkim &mov ($tmp,$acc); 1297238405Sjkim &shr ($tmp,7); 1298238405Sjkim &lea ($tp8,&DWP(0,$tp4,$tp4)); 1299238405Sjkim &sub ($acc,$tmp); 1300238405Sjkim &and ($tp8,0xfefefefe); 1301238405Sjkim &and ($acc,0x1b1b1b1b); 1302238405Sjkim &xor ($tp4,$s[$i]); # tp4^tp1 1303238405Sjkim &rotl ($s[$i],8); # = ROTATE(tp1,8) 1304238405Sjkim &xor ($tp8,$acc); 1305238405Sjkim 1306238405Sjkim &xor ($s[$i],$tp2); 1307238405Sjkim &xor ($tp2,$tp8); 1308238405Sjkim &rotl ($tp2,24); 1309238405Sjkim &xor ($s[$i],$tp4); 1310238405Sjkim &xor ($tp4,$tp8); 1311238405Sjkim &rotl ($tp4,16); 1312238405Sjkim &xor ($s[$i],$tp8); # ^= tp8^(tp4^tp1)^(tp2^tp1) 1313238405Sjkim &rotl ($tp8,8); 1314238405Sjkim &xor ($s[$i],$tp2); # ^= ROTATE(tp8^tp2^tp1,24) 1315238405Sjkim &xor ($s[$i],$tp4); # ^= ROTATE(tp8^tp4^tp1,16) 1316238405Sjkim &mov ($s[0],$__s0) if($i==2); #prefetch $s0 1317238405Sjkim &mov ($s[1],$__s1) if($i==3); #prefetch $s1 1318238405Sjkim &mov ($s[2],$__s2) if($i==1); 1319238405Sjkim &xor ($s[$i],$tp8); # ^= ROTATE(tp8,8) 1320238405Sjkim 1321238405Sjkim &mov ($s[3],$__s3) if($i==1); 1322238405Sjkim &mov (&DWP(4+4*$i,"esp"),$s[$i]) if($i>=2); 1323238405Sjkim} 1324238405Sjkim 1325238405Sjkim&function_begin_B("_x86_AES_decrypt_compact"); 1326238405Sjkim # note that caller is expected to allocate stack frame for me! 1327238405Sjkim &mov ($__key,$key); # save key 1328238405Sjkim 1329238405Sjkim &xor ($s0,&DWP(0,$key)); # xor with key 1330238405Sjkim &xor ($s1,&DWP(4,$key)); 1331238405Sjkim &xor ($s2,&DWP(8,$key)); 1332238405Sjkim &xor ($s3,&DWP(12,$key)); 1333238405Sjkim 1334238405Sjkim &mov ($acc,&DWP(240,$key)); # load key->rounds 1335238405Sjkim 1336238405Sjkim &lea ($acc,&DWP(-2,$acc,$acc)); 1337238405Sjkim &lea ($acc,&DWP(0,$key,$acc,8)); 1338238405Sjkim &mov ($__end,$acc); # end of key schedule 1339238405Sjkim 1340238405Sjkim # prefetch Td4 1341238405Sjkim &mov ($key,&DWP(0-128,$tbl)); 1342238405Sjkim &mov ($acc,&DWP(32-128,$tbl)); 1343238405Sjkim &mov ($key,&DWP(64-128,$tbl)); 1344238405Sjkim &mov ($acc,&DWP(96-128,$tbl)); 1345238405Sjkim &mov ($key,&DWP(128-128,$tbl)); 1346238405Sjkim &mov ($acc,&DWP(160-128,$tbl)); 1347238405Sjkim &mov ($key,&DWP(192-128,$tbl)); 1348238405Sjkim &mov ($acc,&DWP(224-128,$tbl)); 1349238405Sjkim 1350238405Sjkim &set_label("loop",16); 1351238405Sjkim 1352238405Sjkim &deccompact(0,$tbl,$s0,$s3,$s2,$s1,1); 1353238405Sjkim &deccompact(1,$tbl,$s1,$s0,$s3,$s2,1); 1354238405Sjkim &deccompact(2,$tbl,$s2,$s1,$s0,$s3,1); 1355238405Sjkim &deccompact(3,$tbl,$s3,$s2,$s1,$s0,1); 1356238405Sjkim &dectransform(2); 1357238405Sjkim &dectransform(3); 1358238405Sjkim &dectransform(0); 1359238405Sjkim &dectransform(1); 1360238405Sjkim &mov ($key,$__key); 1361238405Sjkim &mov ($tbl,$__tbl); 1362238405Sjkim &add ($key,16); # advance rd_key 1363238405Sjkim &xor ($s0,&DWP(0,$key)); 1364238405Sjkim &xor ($s1,&DWP(4,$key)); 1365238405Sjkim &xor ($s2,&DWP(8,$key)); 1366238405Sjkim &xor ($s3,&DWP(12,$key)); 1367238405Sjkim 1368238405Sjkim &cmp ($key,$__end); 1369238405Sjkim &mov ($__key,$key); 1370238405Sjkim &jb (&label("loop")); 1371238405Sjkim 1372238405Sjkim &deccompact(0,$tbl,$s0,$s3,$s2,$s1); 1373238405Sjkim &deccompact(1,$tbl,$s1,$s0,$s3,$s2); 1374238405Sjkim &deccompact(2,$tbl,$s2,$s1,$s0,$s3); 1375238405Sjkim &deccompact(3,$tbl,$s3,$s2,$s1,$s0); 1376238405Sjkim 1377238405Sjkim &xor ($s0,&DWP(16,$key)); 1378238405Sjkim &xor ($s1,&DWP(20,$key)); 1379238405Sjkim &xor ($s2,&DWP(24,$key)); 1380238405Sjkim &xor ($s3,&DWP(28,$key)); 1381238405Sjkim 1382238405Sjkim &ret (); 1383238405Sjkim&function_end_B("_x86_AES_decrypt_compact"); 1384238405Sjkim 1385238405Sjkim###################################################################### 1386238405Sjkim# "Compact" SSE block function. 1387238405Sjkim###################################################################### 1388238405Sjkim 1389238405Sjkimsub sse_deccompact() 1390238405Sjkim{ 1391238405Sjkim &pshufw ("mm1","mm0",0x0c); # 7, 6, 1, 0 1392238405Sjkim &movd ("eax","mm1"); # 7, 6, 1, 0 1393238405Sjkim 1394238405Sjkim &pshufw ("mm5","mm4",0x09); # 13,12,11,10 1395238405Sjkim &movz ($acc,&LB("eax")); # 0 1396238405Sjkim &movz ("ecx",&BP(-128,$tbl,$acc,1)); # 0 1397238405Sjkim &movd ("ebx","mm5"); # 13,12,11,10 1398238405Sjkim &movz ("edx",&HB("eax")); # 1 1399238405Sjkim &movz ("edx",&BP(-128,$tbl,"edx",1)); # 1 1400238405Sjkim &shl ("edx",8); # 1 1401238405Sjkim 1402238405Sjkim &pshufw ("mm2","mm0",0x06); # 3, 2, 5, 4 1403238405Sjkim &movz ($acc,&LB("ebx")); # 10 1404238405Sjkim &movz ($acc,&BP(-128,$tbl,$acc,1)); # 10 1405238405Sjkim &shl ($acc,16); # 10 1406238405Sjkim &or ("ecx",$acc); # 10 1407238405Sjkim &shr ("eax",16); # 7, 6 1408238405Sjkim &movz ($acc,&HB("ebx")); # 11 1409238405Sjkim &movz ($acc,&BP(-128,$tbl,$acc,1)); # 11 1410238405Sjkim &shl ($acc,24); # 11 1411238405Sjkim &or ("edx",$acc); # 11 1412238405Sjkim &shr ("ebx",16); # 13,12 1413238405Sjkim 1414238405Sjkim &pshufw ("mm6","mm4",0x03); # 9, 8,15,14 1415238405Sjkim &movz ($acc,&HB("eax")); # 7 1416238405Sjkim &movz ($acc,&BP(-128,$tbl,$acc,1)); # 7 1417238405Sjkim &shl ($acc,24); # 7 1418238405Sjkim &or ("ecx",$acc); # 7 1419238405Sjkim &movz ($acc,&HB("ebx")); # 13 1420238405Sjkim &movz ($acc,&BP(-128,$tbl,$acc,1)); # 13 1421238405Sjkim &shl ($acc,8); # 13 1422238405Sjkim &or ("ecx",$acc); # 13 1423238405Sjkim &movd ("mm0","ecx"); # t[0] collected 1424238405Sjkim 1425238405Sjkim &movz ($acc,&LB("eax")); # 6 1426238405Sjkim &movd ("eax","mm2"); # 3, 2, 5, 4 1427238405Sjkim &movz ("ecx",&BP(-128,$tbl,$acc,1)); # 6 1428238405Sjkim &shl ("ecx",16); # 6 1429238405Sjkim &movz ($acc,&LB("ebx")); # 12 1430238405Sjkim &movd ("ebx","mm6"); # 9, 8,15,14 1431238405Sjkim &movz ($acc,&BP(-128,$tbl,$acc,1)); # 12 1432238405Sjkim &or ("ecx",$acc); # 12 1433238405Sjkim 1434238405Sjkim &movz ($acc,&LB("eax")); # 4 1435238405Sjkim &movz ($acc,&BP(-128,$tbl,$acc,1)); # 4 1436238405Sjkim &or ("edx",$acc); # 4 1437238405Sjkim &movz ($acc,&LB("ebx")); # 14 1438238405Sjkim &movz ($acc,&BP(-128,$tbl,$acc,1)); # 14 1439238405Sjkim &shl ($acc,16); # 14 1440238405Sjkim &or ("edx",$acc); # 14 1441238405Sjkim &movd ("mm1","edx"); # t[1] collected 1442238405Sjkim 1443238405Sjkim &movz ($acc,&HB("eax")); # 5 1444238405Sjkim &movz ("edx",&BP(-128,$tbl,$acc,1)); # 5 1445238405Sjkim &shl ("edx",8); # 5 1446238405Sjkim &movz ($acc,&HB("ebx")); # 15 1447238405Sjkim &shr ("eax",16); # 3, 2 1448238405Sjkim &movz ($acc,&BP(-128,$tbl,$acc,1)); # 15 1449238405Sjkim &shl ($acc,24); # 15 1450238405Sjkim &or ("edx",$acc); # 15 1451238405Sjkim &shr ("ebx",16); # 9, 8 1452238405Sjkim 1453238405Sjkim &punpckldq ("mm0","mm1"); # t[0,1] collected 1454238405Sjkim 1455238405Sjkim &movz ($acc,&HB("ebx")); # 9 1456238405Sjkim &movz ($acc,&BP(-128,$tbl,$acc,1)); # 9 1457238405Sjkim &shl ($acc,8); # 9 1458238405Sjkim &or ("ecx",$acc); # 9 1459238405Sjkim &and ("ebx",0xff); # 8 1460238405Sjkim &movz ("ebx",&BP(-128,$tbl,"ebx",1)); # 8 1461238405Sjkim &or ("edx","ebx"); # 8 1462238405Sjkim &movz ($acc,&LB("eax")); # 2 1463238405Sjkim &movz ($acc,&BP(-128,$tbl,$acc,1)); # 2 1464238405Sjkim &shl ($acc,16); # 2 1465238405Sjkim &or ("edx",$acc); # 2 1466238405Sjkim &movd ("mm4","edx"); # t[2] collected 1467238405Sjkim &movz ("eax",&HB("eax")); # 3 1468238405Sjkim &movz ("eax",&BP(-128,$tbl,"eax",1)); # 3 1469238405Sjkim &shl ("eax",24); # 3 1470238405Sjkim &or ("ecx","eax"); # 3 1471238405Sjkim &movd ("mm5","ecx"); # t[3] collected 1472238405Sjkim 1473238405Sjkim &punpckldq ("mm4","mm5"); # t[2,3] collected 1474238405Sjkim} 1475238405Sjkim 1476238405Sjkim if (!$x86only) { 1477238405Sjkim&function_begin_B("_sse_AES_decrypt_compact"); 1478238405Sjkim &pxor ("mm0",&QWP(0,$key)); # 7, 6, 5, 4, 3, 2, 1, 0 1479238405Sjkim &pxor ("mm4",&QWP(8,$key)); # 15,14,13,12,11,10, 9, 8 1480238405Sjkim 1481238405Sjkim # note that caller is expected to allocate stack frame for me! 1482238405Sjkim &mov ($acc,&DWP(240,$key)); # load key->rounds 1483238405Sjkim &lea ($acc,&DWP(-2,$acc,$acc)); 1484238405Sjkim &lea ($acc,&DWP(0,$key,$acc,8)); 1485238405Sjkim &mov ($__end,$acc); # end of key schedule 1486238405Sjkim 1487238405Sjkim &mov ($s0,0x1b1b1b1b); # magic constant 1488238405Sjkim &mov (&DWP(8,"esp"),$s0); 1489238405Sjkim &mov (&DWP(12,"esp"),$s0); 1490238405Sjkim 1491238405Sjkim # prefetch Td4 1492238405Sjkim &mov ($s0,&DWP(0-128,$tbl)); 1493238405Sjkim &mov ($s1,&DWP(32-128,$tbl)); 1494238405Sjkim &mov ($s2,&DWP(64-128,$tbl)); 1495238405Sjkim &mov ($s3,&DWP(96-128,$tbl)); 1496238405Sjkim &mov ($s0,&DWP(128-128,$tbl)); 1497238405Sjkim &mov ($s1,&DWP(160-128,$tbl)); 1498238405Sjkim &mov ($s2,&DWP(192-128,$tbl)); 1499238405Sjkim &mov ($s3,&DWP(224-128,$tbl)); 1500238405Sjkim 1501238405Sjkim &set_label("loop",16); 1502238405Sjkim &sse_deccompact(); 1503238405Sjkim &add ($key,16); 1504238405Sjkim &cmp ($key,$__end); 1505238405Sjkim &ja (&label("out")); 1506238405Sjkim 1507238405Sjkim # ROTATE(x^y,N) == ROTATE(x,N)^ROTATE(y,N) 1508238405Sjkim &movq ("mm3","mm0"); &movq ("mm7","mm4"); 1509238405Sjkim &movq ("mm2","mm0",1); &movq ("mm6","mm4",1); 1510238405Sjkim &movq ("mm1","mm0"); &movq ("mm5","mm4"); 1511238405Sjkim &pshufw ("mm0","mm0",0xb1); &pshufw ("mm4","mm4",0xb1);# = ROTATE(tp0,16) 1512238405Sjkim &pslld ("mm2",8); &pslld ("mm6",8); 1513238405Sjkim &psrld ("mm3",8); &psrld ("mm7",8); 1514238405Sjkim &pxor ("mm0","mm2"); &pxor ("mm4","mm6"); # ^= tp0<<8 1515238405Sjkim &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= tp0>>8 1516238405Sjkim &pslld ("mm2",16); &pslld ("mm6",16); 1517238405Sjkim &psrld ("mm3",16); &psrld ("mm7",16); 1518238405Sjkim &pxor ("mm0","mm2"); &pxor ("mm4","mm6"); # ^= tp0<<24 1519238405Sjkim &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= tp0>>24 1520238405Sjkim 1521238405Sjkim &movq ("mm3",&QWP(8,"esp")); 1522238405Sjkim &pxor ("mm2","mm2"); &pxor ("mm6","mm6"); 1523238405Sjkim &pcmpgtb("mm2","mm1"); &pcmpgtb("mm6","mm5"); 1524238405Sjkim &pand ("mm2","mm3"); &pand ("mm6","mm3"); 1525238405Sjkim &paddb ("mm1","mm1"); &paddb ("mm5","mm5"); 1526238405Sjkim &pxor ("mm1","mm2"); &pxor ("mm5","mm6"); # tp2 1527238405Sjkim &movq ("mm3","mm1"); &movq ("mm7","mm5"); 1528238405Sjkim &movq ("mm2","mm1"); &movq ("mm6","mm5"); 1529238405Sjkim &pxor ("mm0","mm1"); &pxor ("mm4","mm5"); # ^= tp2 1530238405Sjkim &pslld ("mm3",24); &pslld ("mm7",24); 1531238405Sjkim &psrld ("mm2",8); &psrld ("mm6",8); 1532238405Sjkim &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= tp2<<24 1533238405Sjkim &pxor ("mm0","mm2"); &pxor ("mm4","mm6"); # ^= tp2>>8 1534238405Sjkim 1535238405Sjkim &movq ("mm2",&QWP(8,"esp")); 1536238405Sjkim &pxor ("mm3","mm3"); &pxor ("mm7","mm7"); 1537238405Sjkim &pcmpgtb("mm3","mm1"); &pcmpgtb("mm7","mm5"); 1538238405Sjkim &pand ("mm3","mm2"); &pand ("mm7","mm2"); 1539238405Sjkim &paddb ("mm1","mm1"); &paddb ("mm5","mm5"); 1540238405Sjkim &pxor ("mm1","mm3"); &pxor ("mm5","mm7"); # tp4 1541238405Sjkim &pshufw ("mm3","mm1",0xb1); &pshufw ("mm7","mm5",0xb1); 1542238405Sjkim &pxor ("mm0","mm1"); &pxor ("mm4","mm5"); # ^= tp4 1543238405Sjkim &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= ROTATE(tp4,16) 1544238405Sjkim 1545238405Sjkim &pxor ("mm3","mm3"); &pxor ("mm7","mm7"); 1546238405Sjkim &pcmpgtb("mm3","mm1"); &pcmpgtb("mm7","mm5"); 1547238405Sjkim &pand ("mm3","mm2"); &pand ("mm7","mm2"); 1548238405Sjkim &paddb ("mm1","mm1"); &paddb ("mm5","mm5"); 1549238405Sjkim &pxor ("mm1","mm3"); &pxor ("mm5","mm7"); # tp8 1550238405Sjkim &pxor ("mm0","mm1"); &pxor ("mm4","mm5"); # ^= tp8 1551238405Sjkim &movq ("mm3","mm1"); &movq ("mm7","mm5"); 1552238405Sjkim &pshufw ("mm2","mm1",0xb1); &pshufw ("mm6","mm5",0xb1); 1553238405Sjkim &pxor ("mm0","mm2"); &pxor ("mm4","mm6"); # ^= ROTATE(tp8,16) 1554238405Sjkim &pslld ("mm1",8); &pslld ("mm5",8); 1555238405Sjkim &psrld ("mm3",8); &psrld ("mm7",8); 1556238405Sjkim &movq ("mm2",&QWP(0,$key)); &movq ("mm6",&QWP(8,$key)); 1557238405Sjkim &pxor ("mm0","mm1"); &pxor ("mm4","mm5"); # ^= tp8<<8 1558238405Sjkim &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= tp8>>8 1559238405Sjkim &mov ($s0,&DWP(0-128,$tbl)); 1560238405Sjkim &pslld ("mm1",16); &pslld ("mm5",16); 1561238405Sjkim &mov ($s1,&DWP(64-128,$tbl)); 1562238405Sjkim &psrld ("mm3",16); &psrld ("mm7",16); 1563238405Sjkim &mov ($s2,&DWP(128-128,$tbl)); 1564238405Sjkim &pxor ("mm0","mm1"); &pxor ("mm4","mm5"); # ^= tp8<<24 1565238405Sjkim &mov ($s3,&DWP(192-128,$tbl)); 1566238405Sjkim &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= tp8>>24 1567238405Sjkim 1568238405Sjkim &pxor ("mm0","mm2"); &pxor ("mm4","mm6"); 1569238405Sjkim &jmp (&label("loop")); 1570238405Sjkim 1571238405Sjkim &set_label("out",16); 1572238405Sjkim &pxor ("mm0",&QWP(0,$key)); 1573238405Sjkim &pxor ("mm4",&QWP(8,$key)); 1574238405Sjkim 1575238405Sjkim &ret (); 1576238405Sjkim&function_end_B("_sse_AES_decrypt_compact"); 1577238405Sjkim } 1578238405Sjkim 1579238405Sjkim###################################################################### 1580238405Sjkim# Vanilla block function. 1581238405Sjkim###################################################################### 1582238405Sjkim 1583160814Ssimonsub decstep() 1584160814Ssimon{ my ($i,$td,@s) = @_; 1585160814Ssimon my $tmp = $key; 1586160814Ssimon my $out = $i==3?$s[0]:$acc; 1587160814Ssimon 1588160814Ssimon # no instructions are reordered, as performance appears 1589160814Ssimon # optimal... or rather that all attempts to reorder didn't 1590160814Ssimon # result in better performance [which by the way is not a 1591160814Ssimon # bit lower than ecryption]. 1592238405Sjkim if($i==3) { &mov ($key,$__key); } 1593160814Ssimon else { &mov ($out,$s[0]); } 1594160814Ssimon &and ($out,0xFF); 1595160814Ssimon &mov ($out,&DWP(0,$td,$out,8)); 1596160814Ssimon 1597160814Ssimon if ($i==3) { $tmp=$s[1]; } 1598160814Ssimon &movz ($tmp,&HB($s[1])); 1599160814Ssimon &xor ($out,&DWP(3,$td,$tmp,8)); 1600160814Ssimon 1601160814Ssimon if ($i==3) { $tmp=$s[2]; &mov ($s[1],$acc); } 1602160814Ssimon else { &mov ($tmp,$s[2]); } 1603160814Ssimon &shr ($tmp,16); 1604160814Ssimon &and ($tmp,0xFF); 1605160814Ssimon &xor ($out,&DWP(2,$td,$tmp,8)); 1606160814Ssimon 1607238405Sjkim if ($i==3) { $tmp=$s[3]; &mov ($s[2],$__s1); } 1608160814Ssimon else { &mov ($tmp,$s[3]); } 1609160814Ssimon &shr ($tmp,24); 1610160814Ssimon &xor ($out,&DWP(1,$td,$tmp,8)); 1611160814Ssimon if ($i<2) { &mov (&DWP(4+4*$i,"esp"),$out); } 1612238405Sjkim if ($i==3) { &mov ($s[3],$__s0); } 1613160814Ssimon &comment(); 1614160814Ssimon} 1615160814Ssimon 1616160814Ssimonsub declast() 1617160814Ssimon{ my ($i,$td,@s)=@_; 1618160814Ssimon my $tmp = $key; 1619160814Ssimon my $out = $i==3?$s[0]:$acc; 1620160814Ssimon 1621238405Sjkim if($i==0) { &lea ($td,&DWP(2048+128,$td)); 1622238405Sjkim &mov ($tmp,&DWP(0-128,$td)); 1623238405Sjkim &mov ($acc,&DWP(32-128,$td)); 1624238405Sjkim &mov ($tmp,&DWP(64-128,$td)); 1625238405Sjkim &mov ($acc,&DWP(96-128,$td)); 1626238405Sjkim &mov ($tmp,&DWP(128-128,$td)); 1627238405Sjkim &mov ($acc,&DWP(160-128,$td)); 1628238405Sjkim &mov ($tmp,&DWP(192-128,$td)); 1629238405Sjkim &mov ($acc,&DWP(224-128,$td)); 1630238405Sjkim &lea ($td,&DWP(-128,$td)); } 1631238405Sjkim if($i==3) { &mov ($key,$__key); } 1632160814Ssimon else { &mov ($out,$s[0]); } 1633160814Ssimon &and ($out,0xFF); 1634238405Sjkim &movz ($out,&BP(0,$td,$out,1)); 1635160814Ssimon 1636160814Ssimon if ($i==3) { $tmp=$s[1]; } 1637160814Ssimon &movz ($tmp,&HB($s[1])); 1638238405Sjkim &movz ($tmp,&BP(0,$td,$tmp,1)); 1639162911Ssimon &shl ($tmp,8); 1640160814Ssimon &xor ($out,$tmp); 1641160814Ssimon 1642160814Ssimon if ($i==3) { $tmp=$s[2]; &mov ($s[1],$acc); } 1643160814Ssimon else { mov ($tmp,$s[2]); } 1644160814Ssimon &shr ($tmp,16); 1645160814Ssimon &and ($tmp,0xFF); 1646238405Sjkim &movz ($tmp,&BP(0,$td,$tmp,1)); 1647162911Ssimon &shl ($tmp,16); 1648160814Ssimon &xor ($out,$tmp); 1649160814Ssimon 1650238405Sjkim if ($i==3) { $tmp=$s[3]; &mov ($s[2],$__s1); } 1651160814Ssimon else { &mov ($tmp,$s[3]); } 1652160814Ssimon &shr ($tmp,24); 1653238405Sjkim &movz ($tmp,&BP(0,$td,$tmp,1)); 1654162911Ssimon &shl ($tmp,24); 1655160814Ssimon &xor ($out,$tmp); 1656160814Ssimon if ($i<2) { &mov (&DWP(4+4*$i,"esp"),$out); } 1657238405Sjkim if ($i==3) { &mov ($s[3],$__s0); 1658238405Sjkim &lea ($td,&DWP(-2048,$td)); } 1659160814Ssimon} 1660160814Ssimon 1661160814Ssimon&function_begin_B("_x86_AES_decrypt"); 1662160814Ssimon # note that caller is expected to allocate stack frame for me! 1663238405Sjkim &mov ($__key,$key); # save key 1664160814Ssimon 1665160814Ssimon &xor ($s0,&DWP(0,$key)); # xor with key 1666160814Ssimon &xor ($s1,&DWP(4,$key)); 1667160814Ssimon &xor ($s2,&DWP(8,$key)); 1668160814Ssimon &xor ($s3,&DWP(12,$key)); 1669160814Ssimon 1670160814Ssimon &mov ($acc,&DWP(240,$key)); # load key->rounds 1671160814Ssimon 1672160814Ssimon if ($small_footprint) { 1673160814Ssimon &lea ($acc,&DWP(-2,$acc,$acc)); 1674160814Ssimon &lea ($acc,&DWP(0,$key,$acc,8)); 1675238405Sjkim &mov ($__end,$acc); # end of key schedule 1676238405Sjkim &set_label("loop",16); 1677238405Sjkim &decstep(0,$tbl,$s0,$s3,$s2,$s1); 1678238405Sjkim &decstep(1,$tbl,$s1,$s0,$s3,$s2); 1679238405Sjkim &decstep(2,$tbl,$s2,$s1,$s0,$s3); 1680238405Sjkim &decstep(3,$tbl,$s3,$s2,$s1,$s0); 1681160814Ssimon &add ($key,16); # advance rd_key 1682160814Ssimon &xor ($s0,&DWP(0,$key)); 1683160814Ssimon &xor ($s1,&DWP(4,$key)); 1684160814Ssimon &xor ($s2,&DWP(8,$key)); 1685160814Ssimon &xor ($s3,&DWP(12,$key)); 1686238405Sjkim &cmp ($key,$__end); 1687238405Sjkim &mov ($__key,$key); 1688160814Ssimon &jb (&label("loop")); 1689160814Ssimon } 1690160814Ssimon else { 1691160814Ssimon &cmp ($acc,10); 1692160814Ssimon &jle (&label("10rounds")); 1693160814Ssimon &cmp ($acc,12); 1694160814Ssimon &jle (&label("12rounds")); 1695160814Ssimon 1696238405Sjkim &set_label("14rounds",4); 1697160814Ssimon for ($i=1;$i<3;$i++) { 1698238405Sjkim &decstep(0,$tbl,$s0,$s3,$s2,$s1); 1699238405Sjkim &decstep(1,$tbl,$s1,$s0,$s3,$s2); 1700238405Sjkim &decstep(2,$tbl,$s2,$s1,$s0,$s3); 1701238405Sjkim &decstep(3,$tbl,$s3,$s2,$s1,$s0); 1702160814Ssimon &xor ($s0,&DWP(16*$i+0,$key)); 1703160814Ssimon &xor ($s1,&DWP(16*$i+4,$key)); 1704160814Ssimon &xor ($s2,&DWP(16*$i+8,$key)); 1705160814Ssimon &xor ($s3,&DWP(16*$i+12,$key)); 1706160814Ssimon } 1707160814Ssimon &add ($key,32); 1708238405Sjkim &mov ($__key,$key); # advance rd_key 1709238405Sjkim &set_label("12rounds",4); 1710160814Ssimon for ($i=1;$i<3;$i++) { 1711238405Sjkim &decstep(0,$tbl,$s0,$s3,$s2,$s1); 1712238405Sjkim &decstep(1,$tbl,$s1,$s0,$s3,$s2); 1713238405Sjkim &decstep(2,$tbl,$s2,$s1,$s0,$s3); 1714238405Sjkim &decstep(3,$tbl,$s3,$s2,$s1,$s0); 1715160814Ssimon &xor ($s0,&DWP(16*$i+0,$key)); 1716160814Ssimon &xor ($s1,&DWP(16*$i+4,$key)); 1717160814Ssimon &xor ($s2,&DWP(16*$i+8,$key)); 1718160814Ssimon &xor ($s3,&DWP(16*$i+12,$key)); 1719160814Ssimon } 1720160814Ssimon &add ($key,32); 1721238405Sjkim &mov ($__key,$key); # advance rd_key 1722238405Sjkim &set_label("10rounds",4); 1723160814Ssimon for ($i=1;$i<10;$i++) { 1724238405Sjkim &decstep(0,$tbl,$s0,$s3,$s2,$s1); 1725238405Sjkim &decstep(1,$tbl,$s1,$s0,$s3,$s2); 1726238405Sjkim &decstep(2,$tbl,$s2,$s1,$s0,$s3); 1727238405Sjkim &decstep(3,$tbl,$s3,$s2,$s1,$s0); 1728160814Ssimon &xor ($s0,&DWP(16*$i+0,$key)); 1729160814Ssimon &xor ($s1,&DWP(16*$i+4,$key)); 1730160814Ssimon &xor ($s2,&DWP(16*$i+8,$key)); 1731160814Ssimon &xor ($s3,&DWP(16*$i+12,$key)); 1732160814Ssimon } 1733160814Ssimon } 1734160814Ssimon 1735238405Sjkim &declast(0,$tbl,$s0,$s3,$s2,$s1); 1736238405Sjkim &declast(1,$tbl,$s1,$s0,$s3,$s2); 1737238405Sjkim &declast(2,$tbl,$s2,$s1,$s0,$s3); 1738238405Sjkim &declast(3,$tbl,$s3,$s2,$s1,$s0); 1739160814Ssimon 1740160814Ssimon &add ($key,$small_footprint?16:160); 1741160814Ssimon &xor ($s0,&DWP(0,$key)); 1742160814Ssimon &xor ($s1,&DWP(4,$key)); 1743160814Ssimon &xor ($s2,&DWP(8,$key)); 1744160814Ssimon &xor ($s3,&DWP(12,$key)); 1745160814Ssimon 1746160814Ssimon &ret (); 1747160814Ssimon 1748160814Ssimon&set_label("AES_Td",64); # Yes! I keep it in the code segment! 1749160814Ssimon &_data_word(0x50a7f451, 0x5365417e, 0xc3a4171a, 0x965e273a); 1750160814Ssimon &_data_word(0xcb6bab3b, 0xf1459d1f, 0xab58faac, 0x9303e34b); 1751160814Ssimon &_data_word(0x55fa3020, 0xf66d76ad, 0x9176cc88, 0x254c02f5); 1752160814Ssimon &_data_word(0xfcd7e54f, 0xd7cb2ac5, 0x80443526, 0x8fa362b5); 1753160814Ssimon &_data_word(0x495ab1de, 0x671bba25, 0x980eea45, 0xe1c0fe5d); 1754160814Ssimon &_data_word(0x02752fc3, 0x12f04c81, 0xa397468d, 0xc6f9d36b); 1755160814Ssimon &_data_word(0xe75f8f03, 0x959c9215, 0xeb7a6dbf, 0xda595295); 1756160814Ssimon &_data_word(0x2d83bed4, 0xd3217458, 0x2969e049, 0x44c8c98e); 1757160814Ssimon &_data_word(0x6a89c275, 0x78798ef4, 0x6b3e5899, 0xdd71b927); 1758160814Ssimon &_data_word(0xb64fe1be, 0x17ad88f0, 0x66ac20c9, 0xb43ace7d); 1759160814Ssimon &_data_word(0x184adf63, 0x82311ae5, 0x60335197, 0x457f5362); 1760160814Ssimon &_data_word(0xe07764b1, 0x84ae6bbb, 0x1ca081fe, 0x942b08f9); 1761160814Ssimon &_data_word(0x58684870, 0x19fd458f, 0x876cde94, 0xb7f87b52); 1762160814Ssimon &_data_word(0x23d373ab, 0xe2024b72, 0x578f1fe3, 0x2aab5566); 1763160814Ssimon &_data_word(0x0728ebb2, 0x03c2b52f, 0x9a7bc586, 0xa50837d3); 1764160814Ssimon &_data_word(0xf2872830, 0xb2a5bf23, 0xba6a0302, 0x5c8216ed); 1765160814Ssimon &_data_word(0x2b1ccf8a, 0x92b479a7, 0xf0f207f3, 0xa1e2694e); 1766160814Ssimon &_data_word(0xcdf4da65, 0xd5be0506, 0x1f6234d1, 0x8afea6c4); 1767160814Ssimon &_data_word(0x9d532e34, 0xa055f3a2, 0x32e18a05, 0x75ebf6a4); 1768160814Ssimon &_data_word(0x39ec830b, 0xaaef6040, 0x069f715e, 0x51106ebd); 1769160814Ssimon &_data_word(0xf98a213e, 0x3d06dd96, 0xae053edd, 0x46bde64d); 1770160814Ssimon &_data_word(0xb58d5491, 0x055dc471, 0x6fd40604, 0xff155060); 1771160814Ssimon &_data_word(0x24fb9819, 0x97e9bdd6, 0xcc434089, 0x779ed967); 1772160814Ssimon &_data_word(0xbd42e8b0, 0x888b8907, 0x385b19e7, 0xdbeec879); 1773160814Ssimon &_data_word(0x470a7ca1, 0xe90f427c, 0xc91e84f8, 0x00000000); 1774160814Ssimon &_data_word(0x83868009, 0x48ed2b32, 0xac70111e, 0x4e725a6c); 1775160814Ssimon &_data_word(0xfbff0efd, 0x5638850f, 0x1ed5ae3d, 0x27392d36); 1776160814Ssimon &_data_word(0x64d90f0a, 0x21a65c68, 0xd1545b9b, 0x3a2e3624); 1777160814Ssimon &_data_word(0xb1670a0c, 0x0fe75793, 0xd296eeb4, 0x9e919b1b); 1778160814Ssimon &_data_word(0x4fc5c080, 0xa220dc61, 0x694b775a, 0x161a121c); 1779160814Ssimon &_data_word(0x0aba93e2, 0xe52aa0c0, 0x43e0223c, 0x1d171b12); 1780160814Ssimon &_data_word(0x0b0d090e, 0xadc78bf2, 0xb9a8b62d, 0xc8a91e14); 1781160814Ssimon &_data_word(0x8519f157, 0x4c0775af, 0xbbdd99ee, 0xfd607fa3); 1782160814Ssimon &_data_word(0x9f2601f7, 0xbcf5725c, 0xc53b6644, 0x347efb5b); 1783160814Ssimon &_data_word(0x7629438b, 0xdcc623cb, 0x68fcedb6, 0x63f1e4b8); 1784160814Ssimon &_data_word(0xcadc31d7, 0x10856342, 0x40229713, 0x2011c684); 1785160814Ssimon &_data_word(0x7d244a85, 0xf83dbbd2, 0x1132f9ae, 0x6da129c7); 1786160814Ssimon &_data_word(0x4b2f9e1d, 0xf330b2dc, 0xec52860d, 0xd0e3c177); 1787160814Ssimon &_data_word(0x6c16b32b, 0x99b970a9, 0xfa489411, 0x2264e947); 1788160814Ssimon &_data_word(0xc48cfca8, 0x1a3ff0a0, 0xd82c7d56, 0xef903322); 1789160814Ssimon &_data_word(0xc74e4987, 0xc1d138d9, 0xfea2ca8c, 0x360bd498); 1790160814Ssimon &_data_word(0xcf81f5a6, 0x28de7aa5, 0x268eb7da, 0xa4bfad3f); 1791160814Ssimon &_data_word(0xe49d3a2c, 0x0d927850, 0x9bcc5f6a, 0x62467e54); 1792160814Ssimon &_data_word(0xc2138df6, 0xe8b8d890, 0x5ef7392e, 0xf5afc382); 1793160814Ssimon &_data_word(0xbe805d9f, 0x7c93d069, 0xa92dd56f, 0xb31225cf); 1794160814Ssimon &_data_word(0x3b99acc8, 0xa77d1810, 0x6e639ce8, 0x7bbb3bdb); 1795160814Ssimon &_data_word(0x097826cd, 0xf418596e, 0x01b79aec, 0xa89a4f83); 1796160814Ssimon &_data_word(0x656e95e6, 0x7ee6ffaa, 0x08cfbc21, 0xe6e815ef); 1797160814Ssimon &_data_word(0xd99be7ba, 0xce366f4a, 0xd4099fea, 0xd67cb029); 1798160814Ssimon &_data_word(0xafb2a431, 0x31233f2a, 0x3094a5c6, 0xc066a235); 1799160814Ssimon &_data_word(0x37bc4e74, 0xa6ca82fc, 0xb0d090e0, 0x15d8a733); 1800160814Ssimon &_data_word(0x4a9804f1, 0xf7daec41, 0x0e50cd7f, 0x2ff69117); 1801160814Ssimon &_data_word(0x8dd64d76, 0x4db0ef43, 0x544daacc, 0xdf0496e4); 1802160814Ssimon &_data_word(0xe3b5d19e, 0x1b886a4c, 0xb81f2cc1, 0x7f516546); 1803160814Ssimon &_data_word(0x04ea5e9d, 0x5d358c01, 0x737487fa, 0x2e410bfb); 1804160814Ssimon &_data_word(0x5a1d67b3, 0x52d2db92, 0x335610e9, 0x1347d66d); 1805160814Ssimon &_data_word(0x8c61d79a, 0x7a0ca137, 0x8e14f859, 0x893c13eb); 1806160814Ssimon &_data_word(0xee27a9ce, 0x35c961b7, 0xede51ce1, 0x3cb1477a); 1807160814Ssimon &_data_word(0x59dfd29c, 0x3f73f255, 0x79ce1418, 0xbf37c773); 1808160814Ssimon &_data_word(0xeacdf753, 0x5baafd5f, 0x146f3ddf, 0x86db4478); 1809160814Ssimon &_data_word(0x81f3afca, 0x3ec468b9, 0x2c342438, 0x5f40a3c2); 1810160814Ssimon &_data_word(0x72c31d16, 0x0c25e2bc, 0x8b493c28, 0x41950dff); 1811160814Ssimon &_data_word(0x7101a839, 0xdeb30c08, 0x9ce4b4d8, 0x90c15664); 1812160814Ssimon &_data_word(0x6184cb7b, 0x70b632d5, 0x745c6c48, 0x4257b8d0); 1813238405Sjkim 1814238405Sjkim#Td4: # four copies of Td4 to choose from to avoid L1 aliasing 1815162911Ssimon &data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38); 1816162911Ssimon &data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb); 1817162911Ssimon &data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87); 1818162911Ssimon &data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb); 1819162911Ssimon &data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d); 1820162911Ssimon &data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e); 1821162911Ssimon &data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2); 1822162911Ssimon &data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25); 1823162911Ssimon &data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16); 1824162911Ssimon &data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92); 1825162911Ssimon &data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda); 1826162911Ssimon &data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84); 1827162911Ssimon &data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a); 1828162911Ssimon &data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06); 1829162911Ssimon &data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02); 1830162911Ssimon &data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b); 1831162911Ssimon &data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea); 1832162911Ssimon &data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73); 1833162911Ssimon &data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85); 1834162911Ssimon &data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e); 1835162911Ssimon &data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89); 1836162911Ssimon &data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b); 1837162911Ssimon &data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20); 1838162911Ssimon &data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4); 1839162911Ssimon &data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31); 1840162911Ssimon &data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f); 1841162911Ssimon &data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d); 1842162911Ssimon &data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef); 1843162911Ssimon &data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0); 1844162911Ssimon &data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61); 1845162911Ssimon &data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26); 1846162911Ssimon &data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d); 1847238405Sjkim 1848238405Sjkim &data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38); 1849238405Sjkim &data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb); 1850238405Sjkim &data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87); 1851238405Sjkim &data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb); 1852238405Sjkim &data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d); 1853238405Sjkim &data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e); 1854238405Sjkim &data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2); 1855238405Sjkim &data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25); 1856238405Sjkim &data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16); 1857238405Sjkim &data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92); 1858238405Sjkim &data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda); 1859238405Sjkim &data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84); 1860238405Sjkim &data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a); 1861238405Sjkim &data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06); 1862238405Sjkim &data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02); 1863238405Sjkim &data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b); 1864238405Sjkim &data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea); 1865238405Sjkim &data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73); 1866238405Sjkim &data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85); 1867238405Sjkim &data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e); 1868238405Sjkim &data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89); 1869238405Sjkim &data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b); 1870238405Sjkim &data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20); 1871238405Sjkim &data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4); 1872238405Sjkim &data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31); 1873238405Sjkim &data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f); 1874238405Sjkim &data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d); 1875238405Sjkim &data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef); 1876238405Sjkim &data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0); 1877238405Sjkim &data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61); 1878238405Sjkim &data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26); 1879238405Sjkim &data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d); 1880238405Sjkim 1881238405Sjkim &data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38); 1882238405Sjkim &data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb); 1883238405Sjkim &data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87); 1884238405Sjkim &data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb); 1885238405Sjkim &data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d); 1886238405Sjkim &data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e); 1887238405Sjkim &data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2); 1888238405Sjkim &data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25); 1889238405Sjkim &data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16); 1890238405Sjkim &data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92); 1891238405Sjkim &data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda); 1892238405Sjkim &data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84); 1893238405Sjkim &data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a); 1894238405Sjkim &data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06); 1895238405Sjkim &data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02); 1896238405Sjkim &data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b); 1897238405Sjkim &data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea); 1898238405Sjkim &data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73); 1899238405Sjkim &data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85); 1900238405Sjkim &data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e); 1901238405Sjkim &data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89); 1902238405Sjkim &data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b); 1903238405Sjkim &data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20); 1904238405Sjkim &data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4); 1905238405Sjkim &data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31); 1906238405Sjkim &data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f); 1907238405Sjkim &data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d); 1908238405Sjkim &data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef); 1909238405Sjkim &data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0); 1910238405Sjkim &data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61); 1911238405Sjkim &data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26); 1912238405Sjkim &data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d); 1913238405Sjkim 1914238405Sjkim &data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38); 1915238405Sjkim &data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb); 1916238405Sjkim &data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87); 1917238405Sjkim &data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb); 1918238405Sjkim &data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d); 1919238405Sjkim &data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e); 1920238405Sjkim &data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2); 1921238405Sjkim &data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25); 1922238405Sjkim &data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16); 1923238405Sjkim &data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92); 1924238405Sjkim &data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda); 1925238405Sjkim &data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84); 1926238405Sjkim &data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a); 1927238405Sjkim &data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06); 1928238405Sjkim &data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02); 1929238405Sjkim &data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b); 1930238405Sjkim &data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea); 1931238405Sjkim &data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73); 1932238405Sjkim &data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85); 1933238405Sjkim &data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e); 1934238405Sjkim &data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89); 1935238405Sjkim &data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b); 1936238405Sjkim &data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20); 1937238405Sjkim &data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4); 1938238405Sjkim &data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31); 1939238405Sjkim &data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f); 1940238405Sjkim &data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d); 1941238405Sjkim &data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef); 1942238405Sjkim &data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0); 1943238405Sjkim &data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61); 1944238405Sjkim &data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26); 1945238405Sjkim &data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d); 1946160814Ssimon&function_end_B("_x86_AES_decrypt"); 1947160814Ssimon 1948160814Ssimon# void AES_decrypt (const void *inp,void *out,const AES_KEY *key); 1949160814Ssimon&function_begin("AES_decrypt"); 1950160814Ssimon &mov ($acc,&wparam(0)); # load inp 1951160814Ssimon &mov ($key,&wparam(2)); # load key 1952160814Ssimon 1953160814Ssimon &mov ($s0,"esp"); 1954238405Sjkim &sub ("esp",36); 1955238405Sjkim &and ("esp",-64); # align to cache-line 1956160814Ssimon 1957238405Sjkim # place stack frame just "above" the key schedule 1958238405Sjkim &lea ($s1,&DWP(-64-63,$key)); 1959238405Sjkim &sub ($s1,"esp"); 1960238405Sjkim &neg ($s1); 1961238405Sjkim &and ($s1,0x3C0); # modulo 1024, but aligned to cache-line 1962238405Sjkim &sub ("esp",$s1); 1963238405Sjkim &add ("esp",4); # 4 is reserved for caller's return address 1964238405Sjkim &mov ($_esp,$s0); # save stack pointer 1965238405Sjkim 1966160814Ssimon &call (&label("pic_point")); # make it PIC! 1967160814Ssimon &set_label("pic_point"); 1968238405Sjkim &blindpop($tbl); 1969238405Sjkim &picmeup($s0,"OPENSSL_ia32cap_P",$tbl,&label("pic_point")) if(!$x86only); 1970238405Sjkim &lea ($tbl,&DWP(&label("AES_Td")."-".&label("pic_point"),$tbl)); 1971160814Ssimon 1972238405Sjkim # pick Td4 copy which can't "overlap" with stack frame or key schedule 1973238405Sjkim &lea ($s1,&DWP(768-4,"esp")); 1974238405Sjkim &sub ($s1,$tbl); 1975238405Sjkim &and ($s1,0x300); 1976238405Sjkim &lea ($tbl,&DWP(2048+128,$tbl,$s1)); 1977162911Ssimon 1978238405Sjkim if (!$x86only) { 1979238405Sjkim &bt (&DWP(0,$s0),25); # check for SSE bit 1980238405Sjkim &jnc (&label("x86")); 1981238405Sjkim 1982238405Sjkim &movq ("mm0",&QWP(0,$acc)); 1983238405Sjkim &movq ("mm4",&QWP(8,$acc)); 1984238405Sjkim &call ("_sse_AES_decrypt_compact"); 1985238405Sjkim &mov ("esp",$_esp); # restore stack pointer 1986238405Sjkim &mov ($acc,&wparam(1)); # load out 1987238405Sjkim &movq (&QWP(0,$acc),"mm0"); # write output data 1988238405Sjkim &movq (&QWP(8,$acc),"mm4"); 1989238405Sjkim &emms (); 1990238405Sjkim &function_end_A(); 1991238405Sjkim } 1992238405Sjkim &set_label("x86",16); 1993238405Sjkim &mov ($_tbl,$tbl); 1994160814Ssimon &mov ($s0,&DWP(0,$acc)); # load input data 1995160814Ssimon &mov ($s1,&DWP(4,$acc)); 1996160814Ssimon &mov ($s2,&DWP(8,$acc)); 1997160814Ssimon &mov ($s3,&DWP(12,$acc)); 1998238405Sjkim &call ("_x86_AES_decrypt_compact"); 1999238405Sjkim &mov ("esp",$_esp); # restore stack pointer 2000160814Ssimon &mov ($acc,&wparam(1)); # load out 2001160814Ssimon &mov (&DWP(0,$acc),$s0); # write output data 2002160814Ssimon &mov (&DWP(4,$acc),$s1); 2003160814Ssimon &mov (&DWP(8,$acc),$s2); 2004160814Ssimon &mov (&DWP(12,$acc),$s3); 2005160814Ssimon&function_end("AES_decrypt"); 2006160814Ssimon 2007160814Ssimon# void AES_cbc_encrypt (const void char *inp, unsigned char *out, 2008160814Ssimon# size_t length, const AES_KEY *key, 2009160814Ssimon# unsigned char *ivp,const int enc); 2010160814Ssimon{ 2011160814Ssimon# stack frame layout 2012238405Sjkim# -4(%esp) # return address 0(%esp) 2013238405Sjkim# 0(%esp) # s0 backing store 4(%esp) 2014238405Sjkim# 4(%esp) # s1 backing store 8(%esp) 2015238405Sjkim# 8(%esp) # s2 backing store 12(%esp) 2016238405Sjkim# 12(%esp) # s3 backing store 16(%esp) 2017238405Sjkim# 16(%esp) # key backup 20(%esp) 2018238405Sjkim# 20(%esp) # end of key schedule 24(%esp) 2019238405Sjkim# 24(%esp) # %ebp backup 28(%esp) 2020238405Sjkim# 28(%esp) # %esp backup 2021238405Sjkimmy $_inp=&DWP(32,"esp"); # copy of wparam(0) 2022238405Sjkimmy $_out=&DWP(36,"esp"); # copy of wparam(1) 2023238405Sjkimmy $_len=&DWP(40,"esp"); # copy of wparam(2) 2024238405Sjkimmy $_key=&DWP(44,"esp"); # copy of wparam(3) 2025238405Sjkimmy $_ivp=&DWP(48,"esp"); # copy of wparam(4) 2026238405Sjkimmy $_tmp=&DWP(52,"esp"); # volatile variable 2027238405Sjkim# 2028238405Sjkimmy $ivec=&DWP(60,"esp"); # ivec[16] 2029238405Sjkimmy $aes_key=&DWP(76,"esp"); # copy of aes_key 2030238405Sjkimmy $mark=&DWP(76+240,"esp"); # copy of aes_key->rounds 2031160814Ssimon 2032160814Ssimon&function_begin("AES_cbc_encrypt"); 2033160814Ssimon &mov ($s2 eq "ecx"? $s2 : "",&wparam(2)); # load len 2034160814Ssimon &cmp ($s2,0); 2035238405Sjkim &je (&label("drop_out")); 2036160814Ssimon 2037160814Ssimon &call (&label("pic_point")); # make it PIC! 2038160814Ssimon &set_label("pic_point"); 2039238405Sjkim &blindpop($tbl); 2040238405Sjkim &picmeup($s0,"OPENSSL_ia32cap_P",$tbl,&label("pic_point")) if(!$x86only); 2041160814Ssimon 2042238405Sjkim &cmp (&wparam(5),0); 2043238405Sjkim &lea ($tbl,&DWP(&label("AES_Te")."-".&label("pic_point"),$tbl)); 2044238405Sjkim &jne (&label("picked_te")); 2045238405Sjkim &lea ($tbl,&DWP(&label("AES_Td")."-".&label("AES_Te"),$tbl)); 2046238405Sjkim &set_label("picked_te"); 2047238405Sjkim 2048238405Sjkim # one can argue if this is required 2049160814Ssimon &pushf (); 2050160814Ssimon &cld (); 2051160814Ssimon 2052238405Sjkim &cmp ($s2,$speed_limit); 2053238405Sjkim &jb (&label("slow_way")); 2054238405Sjkim &test ($s2,15); 2055238405Sjkim &jnz (&label("slow_way")); 2056238405Sjkim if (!$x86only) { 2057238405Sjkim &bt (&DWP(0,$s0),28); # check for hyper-threading bit 2058238405Sjkim &jc (&label("slow_way")); 2059238405Sjkim } 2060238405Sjkim # pre-allocate aligned stack frame... 2061238405Sjkim &lea ($acc,&DWP(-80-244,"esp")); 2062238405Sjkim &and ($acc,-64); 2063160814Ssimon 2064238405Sjkim # ... and make sure it doesn't alias with $tbl modulo 4096 2065238405Sjkim &mov ($s0,$tbl); 2066238405Sjkim &lea ($s1,&DWP(2048+256,$tbl)); 2067238405Sjkim &mov ($s3,$acc); 2068160814Ssimon &and ($s0,0xfff); # s = %ebp&0xfff 2069238405Sjkim &and ($s1,0xfff); # e = (%ebp+2048+256)&0xfff 2070160814Ssimon &and ($s3,0xfff); # p = %esp&0xfff 2071160814Ssimon 2072160814Ssimon &cmp ($s3,$s1); # if (p>=e) %esp =- (p-e); 2073238405Sjkim &jb (&label("tbl_break_out")); 2074160814Ssimon &sub ($s3,$s1); 2075238405Sjkim &sub ($acc,$s3); 2076238405Sjkim &jmp (&label("tbl_ok")); 2077238405Sjkim &set_label("tbl_break_out",4); # else %esp -= (p-s)&0xfff + framesz; 2078160814Ssimon &sub ($s3,$s0); 2079160814Ssimon &and ($s3,0xfff); 2080238405Sjkim &add ($s3,384); 2081238405Sjkim &sub ($acc,$s3); 2082238405Sjkim &set_label("tbl_ok",4); 2083160814Ssimon 2084238405Sjkim &lea ($s3,&wparam(0)); # obtain pointer to parameter block 2085238405Sjkim &exch ("esp",$acc); # allocate stack frame 2086160814Ssimon &add ("esp",4); # reserve for return address! 2087238405Sjkim &mov ($_tbl,$tbl); # save %ebp 2088238405Sjkim &mov ($_esp,$acc); # save %esp 2089160814Ssimon 2090238405Sjkim &mov ($s0,&DWP(0,$s3)); # load inp 2091238405Sjkim &mov ($s1,&DWP(4,$s3)); # load out 2092238405Sjkim #&mov ($s2,&DWP(8,$s3)); # load len 2093238405Sjkim &mov ($key,&DWP(12,$s3)); # load key 2094238405Sjkim &mov ($acc,&DWP(16,$s3)); # load ivp 2095238405Sjkim &mov ($s3,&DWP(20,$s3)); # load enc flag 2096238405Sjkim 2097160814Ssimon &mov ($_inp,$s0); # save copy of inp 2098160814Ssimon &mov ($_out,$s1); # save copy of out 2099160814Ssimon &mov ($_len,$s2); # save copy of len 2100238405Sjkim &mov ($_key,$key); # save copy of key 2101160814Ssimon &mov ($_ivp,$acc); # save copy of ivp 2102160814Ssimon 2103162911Ssimon &mov ($mark,0); # copy of aes_key->rounds = 0; 2104162911Ssimon # do we copy key schedule to stack? 2105238405Sjkim &mov ($s1 eq "ebx" ? $s1 : "",$key); 2106162911Ssimon &mov ($s2 eq "ecx" ? $s2 : "",244/4); 2107238405Sjkim &sub ($s1,$tbl); 2108238405Sjkim &mov ("esi",$key); 2109162911Ssimon &and ($s1,0xfff); 2110160814Ssimon &lea ("edi",$aes_key); 2111238405Sjkim &cmp ($s1,2048+256); 2112238405Sjkim &jb (&label("do_copy")); 2113162911Ssimon &cmp ($s1,4096-244); 2114238405Sjkim &jb (&label("skip_copy")); 2115238405Sjkim &set_label("do_copy",4); 2116162911Ssimon &mov ($_key,"edi"); 2117162911Ssimon &data_word(0xA5F3F689); # rep movsd 2118238405Sjkim &set_label("skip_copy"); 2119160814Ssimon 2120160814Ssimon &mov ($key,16); 2121238405Sjkim &set_label("prefetch_tbl",4); 2122238405Sjkim &mov ($s0,&DWP(0,$tbl)); 2123238405Sjkim &mov ($s1,&DWP(32,$tbl)); 2124238405Sjkim &mov ($s2,&DWP(64,$tbl)); 2125238405Sjkim &mov ($acc,&DWP(96,$tbl)); 2126238405Sjkim &lea ($tbl,&DWP(128,$tbl)); 2127238405Sjkim &sub ($key,1); 2128238405Sjkim &jnz (&label("prefetch_tbl")); 2129238405Sjkim &sub ($tbl,2048); 2130160814Ssimon 2131238405Sjkim &mov ($acc,$_inp); 2132160814Ssimon &mov ($key,$_ivp); 2133160814Ssimon 2134238405Sjkim &cmp ($s3,0); 2135238405Sjkim &je (&label("fast_decrypt")); 2136238405Sjkim 2137238405Sjkim#----------------------------- ENCRYPT -----------------------------# 2138160814Ssimon &mov ($s0,&DWP(0,$key)); # load iv 2139160814Ssimon &mov ($s1,&DWP(4,$key)); 2140160814Ssimon 2141238405Sjkim &set_label("fast_enc_loop",16); 2142160814Ssimon &mov ($s2,&DWP(8,$key)); 2143160814Ssimon &mov ($s3,&DWP(12,$key)); 2144160814Ssimon 2145160814Ssimon &xor ($s0,&DWP(0,$acc)); # xor input data 2146160814Ssimon &xor ($s1,&DWP(4,$acc)); 2147160814Ssimon &xor ($s2,&DWP(8,$acc)); 2148160814Ssimon &xor ($s3,&DWP(12,$acc)); 2149160814Ssimon 2150160814Ssimon &mov ($key,$_key); # load key 2151160814Ssimon &call ("_x86_AES_encrypt"); 2152160814Ssimon 2153160814Ssimon &mov ($acc,$_inp); # load inp 2154160814Ssimon &mov ($key,$_out); # load out 2155160814Ssimon 2156160814Ssimon &mov (&DWP(0,$key),$s0); # save output data 2157160814Ssimon &mov (&DWP(4,$key),$s1); 2158160814Ssimon &mov (&DWP(8,$key),$s2); 2159160814Ssimon &mov (&DWP(12,$key),$s3); 2160160814Ssimon 2161238405Sjkim &lea ($acc,&DWP(16,$acc)); # advance inp 2162160814Ssimon &mov ($s2,$_len); # load len 2163160814Ssimon &mov ($_inp,$acc); # save inp 2164238405Sjkim &lea ($s3,&DWP(16,$key)); # advance out 2165160814Ssimon &mov ($_out,$s3); # save out 2166238405Sjkim &sub ($s2,16); # decrease len 2167160814Ssimon &mov ($_len,$s2); # save len 2168238405Sjkim &jnz (&label("fast_enc_loop")); 2169160814Ssimon &mov ($acc,$_ivp); # load ivp 2170238405Sjkim &mov ($s2,&DWP(8,$key)); # restore last 2 dwords 2171160814Ssimon &mov ($s3,&DWP(12,$key)); 2172160814Ssimon &mov (&DWP(0,$acc),$s0); # save ivec 2173160814Ssimon &mov (&DWP(4,$acc),$s1); 2174160814Ssimon &mov (&DWP(8,$acc),$s2); 2175160814Ssimon &mov (&DWP(12,$acc),$s3); 2176160814Ssimon 2177162911Ssimon &cmp ($mark,0); # was the key schedule copied? 2178160814Ssimon &mov ("edi",$_key); 2179162911Ssimon &je (&label("skip_ezero")); 2180160814Ssimon # zero copy of key schedule 2181160814Ssimon &mov ("ecx",240/4); 2182160814Ssimon &xor ("eax","eax"); 2183160814Ssimon &align (4); 2184162911Ssimon &data_word(0xABF3F689); # rep stosd 2185162911Ssimon &set_label("skip_ezero") 2186194206Ssimon &mov ("esp",$_esp); 2187160814Ssimon &popf (); 2188238405Sjkim &set_label("drop_out"); 2189160814Ssimon &function_end_A(); 2190160814Ssimon &pushf (); # kludge, never executed 2191160814Ssimon 2192160814Ssimon#----------------------------- DECRYPT -----------------------------# 2193238405Sjkim&set_label("fast_decrypt",16); 2194160814Ssimon 2195160814Ssimon &cmp ($acc,$_out); 2196238405Sjkim &je (&label("fast_dec_in_place")); # in-place processing... 2197160814Ssimon 2198160814Ssimon &mov ($_tmp,$key); 2199160814Ssimon 2200160814Ssimon &align (4); 2201238405Sjkim &set_label("fast_dec_loop",16); 2202160814Ssimon &mov ($s0,&DWP(0,$acc)); # read input 2203160814Ssimon &mov ($s1,&DWP(4,$acc)); 2204160814Ssimon &mov ($s2,&DWP(8,$acc)); 2205160814Ssimon &mov ($s3,&DWP(12,$acc)); 2206160814Ssimon 2207160814Ssimon &mov ($key,$_key); # load key 2208160814Ssimon &call ("_x86_AES_decrypt"); 2209160814Ssimon 2210160814Ssimon &mov ($key,$_tmp); # load ivp 2211160814Ssimon &mov ($acc,$_len); # load len 2212160814Ssimon &xor ($s0,&DWP(0,$key)); # xor iv 2213160814Ssimon &xor ($s1,&DWP(4,$key)); 2214160814Ssimon &xor ($s2,&DWP(8,$key)); 2215160814Ssimon &xor ($s3,&DWP(12,$key)); 2216160814Ssimon 2217238405Sjkim &mov ($key,$_out); # load out 2218160814Ssimon &mov ($acc,$_inp); # load inp 2219160814Ssimon 2220160814Ssimon &mov (&DWP(0,$key),$s0); # write output 2221160814Ssimon &mov (&DWP(4,$key),$s1); 2222160814Ssimon &mov (&DWP(8,$key),$s2); 2223160814Ssimon &mov (&DWP(12,$key),$s3); 2224160814Ssimon 2225238405Sjkim &mov ($s2,$_len); # load len 2226160814Ssimon &mov ($_tmp,$acc); # save ivp 2227238405Sjkim &lea ($acc,&DWP(16,$acc)); # advance inp 2228160814Ssimon &mov ($_inp,$acc); # save inp 2229238405Sjkim &lea ($key,&DWP(16,$key)); # advance out 2230160814Ssimon &mov ($_out,$key); # save out 2231238405Sjkim &sub ($s2,16); # decrease len 2232238405Sjkim &mov ($_len,$s2); # save len 2233238405Sjkim &jnz (&label("fast_dec_loop")); 2234160814Ssimon &mov ($key,$_tmp); # load temp ivp 2235160814Ssimon &mov ($acc,$_ivp); # load user ivp 2236160814Ssimon &mov ($s0,&DWP(0,$key)); # load iv 2237160814Ssimon &mov ($s1,&DWP(4,$key)); 2238160814Ssimon &mov ($s2,&DWP(8,$key)); 2239160814Ssimon &mov ($s3,&DWP(12,$key)); 2240160814Ssimon &mov (&DWP(0,$acc),$s0); # copy back to user 2241160814Ssimon &mov (&DWP(4,$acc),$s1); 2242160814Ssimon &mov (&DWP(8,$acc),$s2); 2243160814Ssimon &mov (&DWP(12,$acc),$s3); 2244238405Sjkim &jmp (&label("fast_dec_out")); 2245160814Ssimon 2246238405Sjkim &set_label("fast_dec_in_place",16); 2247238405Sjkim &set_label("fast_dec_in_place_loop"); 2248160814Ssimon &mov ($s0,&DWP(0,$acc)); # read input 2249160814Ssimon &mov ($s1,&DWP(4,$acc)); 2250160814Ssimon &mov ($s2,&DWP(8,$acc)); 2251160814Ssimon &mov ($s3,&DWP(12,$acc)); 2252160814Ssimon 2253238405Sjkim &lea ($key,$ivec); 2254160814Ssimon &mov (&DWP(0,$key),$s0); # copy to temp 2255160814Ssimon &mov (&DWP(4,$key),$s1); 2256160814Ssimon &mov (&DWP(8,$key),$s2); 2257160814Ssimon &mov (&DWP(12,$key),$s3); 2258160814Ssimon 2259160814Ssimon &mov ($key,$_key); # load key 2260160814Ssimon &call ("_x86_AES_decrypt"); 2261160814Ssimon 2262160814Ssimon &mov ($key,$_ivp); # load ivp 2263160814Ssimon &mov ($acc,$_out); # load out 2264160814Ssimon &xor ($s0,&DWP(0,$key)); # xor iv 2265160814Ssimon &xor ($s1,&DWP(4,$key)); 2266160814Ssimon &xor ($s2,&DWP(8,$key)); 2267160814Ssimon &xor ($s3,&DWP(12,$key)); 2268160814Ssimon 2269160814Ssimon &mov (&DWP(0,$acc),$s0); # write output 2270160814Ssimon &mov (&DWP(4,$acc),$s1); 2271160814Ssimon &mov (&DWP(8,$acc),$s2); 2272160814Ssimon &mov (&DWP(12,$acc),$s3); 2273160814Ssimon 2274238405Sjkim &lea ($acc,&DWP(16,$acc)); # advance out 2275160814Ssimon &mov ($_out,$acc); # save out 2276160814Ssimon 2277160814Ssimon &lea ($acc,$ivec); 2278160814Ssimon &mov ($s0,&DWP(0,$acc)); # read temp 2279160814Ssimon &mov ($s1,&DWP(4,$acc)); 2280160814Ssimon &mov ($s2,&DWP(8,$acc)); 2281160814Ssimon &mov ($s3,&DWP(12,$acc)); 2282160814Ssimon 2283160814Ssimon &mov (&DWP(0,$key),$s0); # copy iv 2284160814Ssimon &mov (&DWP(4,$key),$s1); 2285160814Ssimon &mov (&DWP(8,$key),$s2); 2286160814Ssimon &mov (&DWP(12,$key),$s3); 2287160814Ssimon 2288160814Ssimon &mov ($acc,$_inp); # load inp 2289238405Sjkim &mov ($s2,$_len); # load len 2290238405Sjkim &lea ($acc,&DWP(16,$acc)); # advance inp 2291238405Sjkim &mov ($_inp,$acc); # save inp 2292238405Sjkim &sub ($s2,16); # decrease len 2293238405Sjkim &mov ($_len,$s2); # save len 2294238405Sjkim &jnz (&label("fast_dec_in_place_loop")); 2295160814Ssimon 2296238405Sjkim &set_label("fast_dec_out",4); 2297238405Sjkim &cmp ($mark,0); # was the key schedule copied? 2298238405Sjkim &mov ("edi",$_key); 2299238405Sjkim &je (&label("skip_dzero")); 2300238405Sjkim # zero copy of key schedule 2301238405Sjkim &mov ("ecx",240/4); 2302238405Sjkim &xor ("eax","eax"); 2303238405Sjkim &align (4); 2304238405Sjkim &data_word(0xABF3F689); # rep stosd 2305238405Sjkim &set_label("skip_dzero") 2306238405Sjkim &mov ("esp",$_esp); 2307238405Sjkim &popf (); 2308238405Sjkim &function_end_A(); 2309238405Sjkim &pushf (); # kludge, never executed 2310238405Sjkim 2311238405Sjkim#--------------------------- SLOW ROUTINE ---------------------------# 2312238405Sjkim&set_label("slow_way",16); 2313238405Sjkim 2314238405Sjkim &mov ($s0,&DWP(0,$s0)) if (!$x86only);# load OPENSSL_ia32cap 2315238405Sjkim &mov ($key,&wparam(3)); # load key 2316238405Sjkim 2317238405Sjkim # pre-allocate aligned stack frame... 2318238405Sjkim &lea ($acc,&DWP(-80,"esp")); 2319238405Sjkim &and ($acc,-64); 2320238405Sjkim 2321238405Sjkim # ... and make sure it doesn't alias with $key modulo 1024 2322238405Sjkim &lea ($s1,&DWP(-80-63,$key)); 2323238405Sjkim &sub ($s1,$acc); 2324238405Sjkim &neg ($s1); 2325238405Sjkim &and ($s1,0x3C0); # modulo 1024, but aligned to cache-line 2326238405Sjkim &sub ($acc,$s1); 2327238405Sjkim 2328238405Sjkim # pick S-box copy which can't overlap with stack frame or $key 2329238405Sjkim &lea ($s1,&DWP(768,$acc)); 2330238405Sjkim &sub ($s1,$tbl); 2331238405Sjkim &and ($s1,0x300); 2332238405Sjkim &lea ($tbl,&DWP(2048+128,$tbl,$s1)); 2333238405Sjkim 2334238405Sjkim &lea ($s3,&wparam(0)); # pointer to parameter block 2335238405Sjkim 2336238405Sjkim &exch ("esp",$acc); 2337238405Sjkim &add ("esp",4); # reserve for return address! 2338238405Sjkim &mov ($_tbl,$tbl); # save %ebp 2339238405Sjkim &mov ($_esp,$acc); # save %esp 2340238405Sjkim &mov ($_tmp,$s0); # save OPENSSL_ia32cap 2341238405Sjkim 2342238405Sjkim &mov ($s0,&DWP(0,$s3)); # load inp 2343238405Sjkim &mov ($s1,&DWP(4,$s3)); # load out 2344238405Sjkim #&mov ($s2,&DWP(8,$s3)); # load len 2345238405Sjkim #&mov ($key,&DWP(12,$s3)); # load key 2346238405Sjkim &mov ($acc,&DWP(16,$s3)); # load ivp 2347238405Sjkim &mov ($s3,&DWP(20,$s3)); # load enc flag 2348238405Sjkim 2349238405Sjkim &mov ($_inp,$s0); # save copy of inp 2350238405Sjkim &mov ($_out,$s1); # save copy of out 2351238405Sjkim &mov ($_len,$s2); # save copy of len 2352238405Sjkim &mov ($_key,$key); # save copy of key 2353238405Sjkim &mov ($_ivp,$acc); # save copy of ivp 2354238405Sjkim 2355238405Sjkim &mov ($key,$acc); 2356238405Sjkim &mov ($acc,$s0); 2357238405Sjkim 2358238405Sjkim &cmp ($s3,0); 2359238405Sjkim &je (&label("slow_decrypt")); 2360238405Sjkim 2361238405Sjkim#--------------------------- SLOW ENCRYPT ---------------------------# 2362238405Sjkim &cmp ($s2,16); 2363238405Sjkim &mov ($s3,$s1); 2364238405Sjkim &jb (&label("slow_enc_tail")); 2365238405Sjkim 2366238405Sjkim if (!$x86only) { 2367238405Sjkim &bt ($_tmp,25); # check for SSE bit 2368238405Sjkim &jnc (&label("slow_enc_x86")); 2369238405Sjkim 2370238405Sjkim &movq ("mm0",&QWP(0,$key)); # load iv 2371238405Sjkim &movq ("mm4",&QWP(8,$key)); 2372238405Sjkim 2373238405Sjkim &set_label("slow_enc_loop_sse",16); 2374238405Sjkim &pxor ("mm0",&QWP(0,$acc)); # xor input data 2375238405Sjkim &pxor ("mm4",&QWP(8,$acc)); 2376238405Sjkim 2377238405Sjkim &mov ($key,$_key); 2378238405Sjkim &call ("_sse_AES_encrypt_compact"); 2379238405Sjkim 2380238405Sjkim &mov ($acc,$_inp); # load inp 2381238405Sjkim &mov ($key,$_out); # load out 2382238405Sjkim &mov ($s2,$_len); # load len 2383238405Sjkim 2384238405Sjkim &movq (&QWP(0,$key),"mm0"); # save output data 2385238405Sjkim &movq (&QWP(8,$key),"mm4"); 2386238405Sjkim 2387238405Sjkim &lea ($acc,&DWP(16,$acc)); # advance inp 2388160814Ssimon &mov ($_inp,$acc); # save inp 2389238405Sjkim &lea ($s3,&DWP(16,$key)); # advance out 2390238405Sjkim &mov ($_out,$s3); # save out 2391238405Sjkim &sub ($s2,16); # decrease len 2392238405Sjkim &cmp ($s2,16); 2393238405Sjkim &mov ($_len,$s2); # save len 2394238405Sjkim &jae (&label("slow_enc_loop_sse")); 2395238405Sjkim &test ($s2,15); 2396238405Sjkim &jnz (&label("slow_enc_tail")); 2397238405Sjkim &mov ($acc,$_ivp); # load ivp 2398238405Sjkim &movq (&QWP(0,$acc),"mm0"); # save ivec 2399238405Sjkim &movq (&QWP(8,$acc),"mm4"); 2400238405Sjkim &emms (); 2401238405Sjkim &mov ("esp",$_esp); 2402238405Sjkim &popf (); 2403238405Sjkim &function_end_A(); 2404238405Sjkim &pushf (); # kludge, never executed 2405238405Sjkim } 2406238405Sjkim &set_label("slow_enc_x86",16); 2407238405Sjkim &mov ($s0,&DWP(0,$key)); # load iv 2408238405Sjkim &mov ($s1,&DWP(4,$key)); 2409160814Ssimon 2410238405Sjkim &set_label("slow_enc_loop_x86",4); 2411238405Sjkim &mov ($s2,&DWP(8,$key)); 2412238405Sjkim &mov ($s3,&DWP(12,$key)); 2413238405Sjkim 2414238405Sjkim &xor ($s0,&DWP(0,$acc)); # xor input data 2415238405Sjkim &xor ($s1,&DWP(4,$acc)); 2416238405Sjkim &xor ($s2,&DWP(8,$acc)); 2417238405Sjkim &xor ($s3,&DWP(12,$acc)); 2418238405Sjkim 2419238405Sjkim &mov ($key,$_key); # load key 2420238405Sjkim &call ("_x86_AES_encrypt_compact"); 2421238405Sjkim 2422238405Sjkim &mov ($acc,$_inp); # load inp 2423238405Sjkim &mov ($key,$_out); # load out 2424238405Sjkim 2425238405Sjkim &mov (&DWP(0,$key),$s0); # save output data 2426238405Sjkim &mov (&DWP(4,$key),$s1); 2427238405Sjkim &mov (&DWP(8,$key),$s2); 2428238405Sjkim &mov (&DWP(12,$key),$s3); 2429238405Sjkim 2430160814Ssimon &mov ($s2,$_len); # load len 2431238405Sjkim &lea ($acc,&DWP(16,$acc)); # advance inp 2432238405Sjkim &mov ($_inp,$acc); # save inp 2433238405Sjkim &lea ($s3,&DWP(16,$key)); # advance out 2434238405Sjkim &mov ($_out,$s3); # save out 2435238405Sjkim &sub ($s2,16); # decrease len 2436238405Sjkim &cmp ($s2,16); 2437160814Ssimon &mov ($_len,$s2); # save len 2438238405Sjkim &jae (&label("slow_enc_loop_x86")); 2439238405Sjkim &test ($s2,15); 2440238405Sjkim &jnz (&label("slow_enc_tail")); 2441238405Sjkim &mov ($acc,$_ivp); # load ivp 2442238405Sjkim &mov ($s2,&DWP(8,$key)); # restore last dwords 2443238405Sjkim &mov ($s3,&DWP(12,$key)); 2444238405Sjkim &mov (&DWP(0,$acc),$s0); # save ivec 2445238405Sjkim &mov (&DWP(4,$acc),$s1); 2446238405Sjkim &mov (&DWP(8,$acc),$s2); 2447238405Sjkim &mov (&DWP(12,$acc),$s3); 2448160814Ssimon 2449238405Sjkim &mov ("esp",$_esp); 2450238405Sjkim &popf (); 2451238405Sjkim &function_end_A(); 2452238405Sjkim &pushf (); # kludge, never executed 2453238405Sjkim 2454238405Sjkim &set_label("slow_enc_tail",16); 2455238405Sjkim &emms () if (!$x86only); 2456238405Sjkim &mov ($key eq "edi"? $key:"",$s3); # load out to edi 2457238405Sjkim &mov ($s1,16); 2458238405Sjkim &sub ($s1,$s2); 2459238405Sjkim &cmp ($key,$acc eq "esi"? $acc:""); # compare with inp 2460238405Sjkim &je (&label("enc_in_place")); 2461238405Sjkim &align (4); 2462238405Sjkim &data_word(0xA4F3F689); # rep movsb # copy input 2463238405Sjkim &jmp (&label("enc_skip_in_place")); 2464238405Sjkim &set_label("enc_in_place"); 2465160814Ssimon &lea ($key,&DWP(0,$key,$s2)); 2466238405Sjkim &set_label("enc_skip_in_place"); 2467238405Sjkim &mov ($s2,$s1); 2468238405Sjkim &xor ($s0,$s0); 2469238405Sjkim &align (4); 2470238405Sjkim &data_word(0xAAF3F689); # rep stosb # zero tail 2471160814Ssimon 2472238405Sjkim &mov ($key,$_ivp); # restore ivp 2473238405Sjkim &mov ($acc,$s3); # output as input 2474238405Sjkim &mov ($s0,&DWP(0,$key)); 2475238405Sjkim &mov ($s1,&DWP(4,$key)); 2476238405Sjkim &mov ($_len,16); # len=16 2477238405Sjkim &jmp (&label("slow_enc_loop_x86")); # one more spin... 2478238405Sjkim 2479238405Sjkim#--------------------------- SLOW DECRYPT ---------------------------# 2480238405Sjkim&set_label("slow_decrypt",16); 2481238405Sjkim if (!$x86only) { 2482238405Sjkim &bt ($_tmp,25); # check for SSE bit 2483238405Sjkim &jnc (&label("slow_dec_loop_x86")); 2484238405Sjkim 2485238405Sjkim &set_label("slow_dec_loop_sse",4); 2486238405Sjkim &movq ("mm0",&QWP(0,$acc)); # read input 2487238405Sjkim &movq ("mm4",&QWP(8,$acc)); 2488238405Sjkim 2489238405Sjkim &mov ($key,$_key); 2490238405Sjkim &call ("_sse_AES_decrypt_compact"); 2491238405Sjkim 2492238405Sjkim &mov ($acc,$_inp); # load inp 2493238405Sjkim &lea ($s0,$ivec); 2494238405Sjkim &mov ($s1,$_out); # load out 2495238405Sjkim &mov ($s2,$_len); # load len 2496238405Sjkim &mov ($key,$_ivp); # load ivp 2497238405Sjkim 2498238405Sjkim &movq ("mm1",&QWP(0,$acc)); # re-read input 2499238405Sjkim &movq ("mm5",&QWP(8,$acc)); 2500238405Sjkim 2501238405Sjkim &pxor ("mm0",&QWP(0,$key)); # xor iv 2502238405Sjkim &pxor ("mm4",&QWP(8,$key)); 2503238405Sjkim 2504238405Sjkim &movq (&QWP(0,$key),"mm1"); # copy input to iv 2505238405Sjkim &movq (&QWP(8,$key),"mm5"); 2506238405Sjkim 2507238405Sjkim &sub ($s2,16); # decrease len 2508238405Sjkim &jc (&label("slow_dec_partial_sse")); 2509238405Sjkim 2510238405Sjkim &movq (&QWP(0,$s1),"mm0"); # write output 2511238405Sjkim &movq (&QWP(8,$s1),"mm4"); 2512238405Sjkim 2513238405Sjkim &lea ($s1,&DWP(16,$s1)); # advance out 2514238405Sjkim &mov ($_out,$s1); # save out 2515238405Sjkim &lea ($acc,&DWP(16,$acc)); # advance inp 2516238405Sjkim &mov ($_inp,$acc); # save inp 2517238405Sjkim &mov ($_len,$s2); # save len 2518238405Sjkim &jnz (&label("slow_dec_loop_sse")); 2519238405Sjkim &emms (); 2520238405Sjkim &mov ("esp",$_esp); 2521238405Sjkim &popf (); 2522238405Sjkim &function_end_A(); 2523238405Sjkim &pushf (); # kludge, never executed 2524238405Sjkim 2525238405Sjkim &set_label("slow_dec_partial_sse",16); 2526238405Sjkim &movq (&QWP(0,$s0),"mm0"); # save output to temp 2527238405Sjkim &movq (&QWP(8,$s0),"mm4"); 2528238405Sjkim &emms (); 2529238405Sjkim 2530238405Sjkim &add ($s2 eq "ecx" ? "ecx":"",16); 2531238405Sjkim &mov ("edi",$s1); # out 2532238405Sjkim &mov ("esi",$s0); # temp 2533238405Sjkim &align (4); 2534238405Sjkim &data_word(0xA4F3F689); # rep movsb # copy partial output 2535238405Sjkim 2536238405Sjkim &mov ("esp",$_esp); 2537238405Sjkim &popf (); 2538238405Sjkim &function_end_A(); 2539238405Sjkim &pushf (); # kludge, never executed 2540238405Sjkim } 2541238405Sjkim &set_label("slow_dec_loop_x86",16); 2542238405Sjkim &mov ($s0,&DWP(0,$acc)); # read input 2543238405Sjkim &mov ($s1,&DWP(4,$acc)); 2544238405Sjkim &mov ($s2,&DWP(8,$acc)); 2545238405Sjkim &mov ($s3,&DWP(12,$acc)); 2546238405Sjkim 2547238405Sjkim &lea ($key,$ivec); 2548238405Sjkim &mov (&DWP(0,$key),$s0); # copy to temp 2549238405Sjkim &mov (&DWP(4,$key),$s1); 2550238405Sjkim &mov (&DWP(8,$key),$s2); 2551238405Sjkim &mov (&DWP(12,$key),$s3); 2552238405Sjkim 2553238405Sjkim &mov ($key,$_key); # load key 2554238405Sjkim &call ("_x86_AES_decrypt_compact"); 2555238405Sjkim 2556238405Sjkim &mov ($key,$_ivp); # load ivp 2557238405Sjkim &mov ($acc,$_len); # load len 2558238405Sjkim &xor ($s0,&DWP(0,$key)); # xor iv 2559238405Sjkim &xor ($s1,&DWP(4,$key)); 2560238405Sjkim &xor ($s2,&DWP(8,$key)); 2561238405Sjkim &xor ($s3,&DWP(12,$key)); 2562238405Sjkim 2563238405Sjkim &sub ($acc,16); 2564238405Sjkim &jc (&label("slow_dec_partial_x86")); 2565238405Sjkim 2566238405Sjkim &mov ($_len,$acc); # save len 2567238405Sjkim &mov ($acc,$_out); # load out 2568238405Sjkim 2569238405Sjkim &mov (&DWP(0,$acc),$s0); # write output 2570238405Sjkim &mov (&DWP(4,$acc),$s1); 2571238405Sjkim &mov (&DWP(8,$acc),$s2); 2572238405Sjkim &mov (&DWP(12,$acc),$s3); 2573238405Sjkim 2574238405Sjkim &lea ($acc,&DWP(16,$acc)); # advance out 2575238405Sjkim &mov ($_out,$acc); # save out 2576238405Sjkim 2577238405Sjkim &lea ($acc,$ivec); 2578238405Sjkim &mov ($s0,&DWP(0,$acc)); # read temp 2579238405Sjkim &mov ($s1,&DWP(4,$acc)); 2580238405Sjkim &mov ($s2,&DWP(8,$acc)); 2581238405Sjkim &mov ($s3,&DWP(12,$acc)); 2582238405Sjkim 2583238405Sjkim &mov (&DWP(0,$key),$s0); # copy it to iv 2584238405Sjkim &mov (&DWP(4,$key),$s1); 2585238405Sjkim &mov (&DWP(8,$key),$s2); 2586238405Sjkim &mov (&DWP(12,$key),$s3); 2587238405Sjkim 2588238405Sjkim &mov ($acc,$_inp); # load inp 2589238405Sjkim &lea ($acc,&DWP(16,$acc)); # advance inp 2590238405Sjkim &mov ($_inp,$acc); # save inp 2591238405Sjkim &jnz (&label("slow_dec_loop_x86")); 2592238405Sjkim &mov ("esp",$_esp); 2593238405Sjkim &popf (); 2594238405Sjkim &function_end_A(); 2595238405Sjkim &pushf (); # kludge, never executed 2596238405Sjkim 2597238405Sjkim &set_label("slow_dec_partial_x86",16); 2598238405Sjkim &lea ($acc,$ivec); 2599238405Sjkim &mov (&DWP(0,$acc),$s0); # save output to temp 2600238405Sjkim &mov (&DWP(4,$acc),$s1); 2601238405Sjkim &mov (&DWP(8,$acc),$s2); 2602238405Sjkim &mov (&DWP(12,$acc),$s3); 2603238405Sjkim 2604238405Sjkim &mov ($acc,$_inp); 2605238405Sjkim &mov ($s0,&DWP(0,$acc)); # re-read input 2606238405Sjkim &mov ($s1,&DWP(4,$acc)); 2607238405Sjkim &mov ($s2,&DWP(8,$acc)); 2608238405Sjkim &mov ($s3,&DWP(12,$acc)); 2609238405Sjkim 2610238405Sjkim &mov (&DWP(0,$key),$s0); # copy it to iv 2611238405Sjkim &mov (&DWP(4,$key),$s1); 2612238405Sjkim &mov (&DWP(8,$key),$s2); 2613238405Sjkim &mov (&DWP(12,$key),$s3); 2614238405Sjkim 2615238405Sjkim &mov ("ecx",$_len); 2616238405Sjkim &mov ("edi",$_out); 2617238405Sjkim &lea ("esi",$ivec); 2618238405Sjkim &align (4); 2619238405Sjkim &data_word(0xA4F3F689); # rep movsb # copy partial output 2620238405Sjkim 2621238405Sjkim &mov ("esp",$_esp); 2622238405Sjkim &popf (); 2623160814Ssimon&function_end("AES_cbc_encrypt"); 2624160814Ssimon} 2625160814Ssimon 2626160814Ssimon#------------------------------------------------------------------# 2627160814Ssimon 2628160814Ssimonsub enckey() 2629160814Ssimon{ 2630160814Ssimon &movz ("esi",&LB("edx")); # rk[i]>>0 2631238405Sjkim &movz ("ebx",&BP(-128,$tbl,"esi",1)); 2632160814Ssimon &movz ("esi",&HB("edx")); # rk[i]>>8 2633238405Sjkim &shl ("ebx",24); 2634160814Ssimon &xor ("eax","ebx"); 2635160814Ssimon 2636238405Sjkim &movz ("ebx",&BP(-128,$tbl,"esi",1)); 2637160814Ssimon &shr ("edx",16); 2638160814Ssimon &movz ("esi",&LB("edx")); # rk[i]>>16 2639160814Ssimon &xor ("eax","ebx"); 2640160814Ssimon 2641238405Sjkim &movz ("ebx",&BP(-128,$tbl,"esi",1)); 2642160814Ssimon &movz ("esi",&HB("edx")); # rk[i]>>24 2643238405Sjkim &shl ("ebx",8); 2644160814Ssimon &xor ("eax","ebx"); 2645160814Ssimon 2646238405Sjkim &movz ("ebx",&BP(-128,$tbl,"esi",1)); 2647238405Sjkim &shl ("ebx",16); 2648160814Ssimon &xor ("eax","ebx"); 2649160814Ssimon 2650238405Sjkim &xor ("eax",&DWP(1024-128,$tbl,"ecx",4)); # rcon 2651160814Ssimon} 2652160814Ssimon 2653238405Sjkim&function_begin("_x86_AES_set_encrypt_key"); 2654238405Sjkim &mov ("esi",&wparam(1)); # user supplied key 2655238405Sjkim &mov ("edi",&wparam(3)); # private key schedule 2656160814Ssimon 2657160814Ssimon &test ("esi",-1); 2658160814Ssimon &jz (&label("badpointer")); 2659160814Ssimon &test ("edi",-1); 2660160814Ssimon &jz (&label("badpointer")); 2661160814Ssimon 2662160814Ssimon &call (&label("pic_point")); 2663160814Ssimon &set_label("pic_point"); 2664238405Sjkim &blindpop($tbl); 2665238405Sjkim &lea ($tbl,&DWP(&label("AES_Te")."-".&label("pic_point"),$tbl)); 2666238405Sjkim &lea ($tbl,&DWP(2048+128,$tbl)); 2667160814Ssimon 2668238405Sjkim # prefetch Te4 2669238405Sjkim &mov ("eax",&DWP(0-128,$tbl)); 2670238405Sjkim &mov ("ebx",&DWP(32-128,$tbl)); 2671238405Sjkim &mov ("ecx",&DWP(64-128,$tbl)); 2672238405Sjkim &mov ("edx",&DWP(96-128,$tbl)); 2673238405Sjkim &mov ("eax",&DWP(128-128,$tbl)); 2674238405Sjkim &mov ("ebx",&DWP(160-128,$tbl)); 2675238405Sjkim &mov ("ecx",&DWP(192-128,$tbl)); 2676238405Sjkim &mov ("edx",&DWP(224-128,$tbl)); 2677238405Sjkim 2678238405Sjkim &mov ("ecx",&wparam(2)); # number of bits in key 2679160814Ssimon &cmp ("ecx",128); 2680160814Ssimon &je (&label("10rounds")); 2681160814Ssimon &cmp ("ecx",192); 2682160814Ssimon &je (&label("12rounds")); 2683160814Ssimon &cmp ("ecx",256); 2684160814Ssimon &je (&label("14rounds")); 2685160814Ssimon &mov ("eax",-2); # invalid number of bits 2686160814Ssimon &jmp (&label("exit")); 2687160814Ssimon 2688160814Ssimon &set_label("10rounds"); 2689160814Ssimon &mov ("eax",&DWP(0,"esi")); # copy first 4 dwords 2690160814Ssimon &mov ("ebx",&DWP(4,"esi")); 2691160814Ssimon &mov ("ecx",&DWP(8,"esi")); 2692160814Ssimon &mov ("edx",&DWP(12,"esi")); 2693160814Ssimon &mov (&DWP(0,"edi"),"eax"); 2694160814Ssimon &mov (&DWP(4,"edi"),"ebx"); 2695160814Ssimon &mov (&DWP(8,"edi"),"ecx"); 2696160814Ssimon &mov (&DWP(12,"edi"),"edx"); 2697160814Ssimon 2698160814Ssimon &xor ("ecx","ecx"); 2699160814Ssimon &jmp (&label("10shortcut")); 2700160814Ssimon 2701160814Ssimon &align (4); 2702160814Ssimon &set_label("10loop"); 2703160814Ssimon &mov ("eax",&DWP(0,"edi")); # rk[0] 2704160814Ssimon &mov ("edx",&DWP(12,"edi")); # rk[3] 2705160814Ssimon &set_label("10shortcut"); 2706160814Ssimon &enckey (); 2707160814Ssimon 2708160814Ssimon &mov (&DWP(16,"edi"),"eax"); # rk[4] 2709160814Ssimon &xor ("eax",&DWP(4,"edi")); 2710160814Ssimon &mov (&DWP(20,"edi"),"eax"); # rk[5] 2711160814Ssimon &xor ("eax",&DWP(8,"edi")); 2712160814Ssimon &mov (&DWP(24,"edi"),"eax"); # rk[6] 2713160814Ssimon &xor ("eax",&DWP(12,"edi")); 2714160814Ssimon &mov (&DWP(28,"edi"),"eax"); # rk[7] 2715160814Ssimon &inc ("ecx"); 2716160814Ssimon &add ("edi",16); 2717160814Ssimon &cmp ("ecx",10); 2718160814Ssimon &jl (&label("10loop")); 2719160814Ssimon 2720160814Ssimon &mov (&DWP(80,"edi"),10); # setup number of rounds 2721160814Ssimon &xor ("eax","eax"); 2722160814Ssimon &jmp (&label("exit")); 2723160814Ssimon 2724160814Ssimon &set_label("12rounds"); 2725160814Ssimon &mov ("eax",&DWP(0,"esi")); # copy first 6 dwords 2726160814Ssimon &mov ("ebx",&DWP(4,"esi")); 2727160814Ssimon &mov ("ecx",&DWP(8,"esi")); 2728160814Ssimon &mov ("edx",&DWP(12,"esi")); 2729160814Ssimon &mov (&DWP(0,"edi"),"eax"); 2730160814Ssimon &mov (&DWP(4,"edi"),"ebx"); 2731160814Ssimon &mov (&DWP(8,"edi"),"ecx"); 2732160814Ssimon &mov (&DWP(12,"edi"),"edx"); 2733160814Ssimon &mov ("ecx",&DWP(16,"esi")); 2734160814Ssimon &mov ("edx",&DWP(20,"esi")); 2735160814Ssimon &mov (&DWP(16,"edi"),"ecx"); 2736160814Ssimon &mov (&DWP(20,"edi"),"edx"); 2737160814Ssimon 2738160814Ssimon &xor ("ecx","ecx"); 2739160814Ssimon &jmp (&label("12shortcut")); 2740160814Ssimon 2741160814Ssimon &align (4); 2742160814Ssimon &set_label("12loop"); 2743160814Ssimon &mov ("eax",&DWP(0,"edi")); # rk[0] 2744160814Ssimon &mov ("edx",&DWP(20,"edi")); # rk[5] 2745160814Ssimon &set_label("12shortcut"); 2746160814Ssimon &enckey (); 2747160814Ssimon 2748160814Ssimon &mov (&DWP(24,"edi"),"eax"); # rk[6] 2749160814Ssimon &xor ("eax",&DWP(4,"edi")); 2750160814Ssimon &mov (&DWP(28,"edi"),"eax"); # rk[7] 2751160814Ssimon &xor ("eax",&DWP(8,"edi")); 2752160814Ssimon &mov (&DWP(32,"edi"),"eax"); # rk[8] 2753160814Ssimon &xor ("eax",&DWP(12,"edi")); 2754160814Ssimon &mov (&DWP(36,"edi"),"eax"); # rk[9] 2755160814Ssimon 2756160814Ssimon &cmp ("ecx",7); 2757160814Ssimon &je (&label("12break")); 2758160814Ssimon &inc ("ecx"); 2759160814Ssimon 2760160814Ssimon &xor ("eax",&DWP(16,"edi")); 2761160814Ssimon &mov (&DWP(40,"edi"),"eax"); # rk[10] 2762160814Ssimon &xor ("eax",&DWP(20,"edi")); 2763160814Ssimon &mov (&DWP(44,"edi"),"eax"); # rk[11] 2764160814Ssimon 2765160814Ssimon &add ("edi",24); 2766160814Ssimon &jmp (&label("12loop")); 2767160814Ssimon 2768160814Ssimon &set_label("12break"); 2769160814Ssimon &mov (&DWP(72,"edi"),12); # setup number of rounds 2770160814Ssimon &xor ("eax","eax"); 2771160814Ssimon &jmp (&label("exit")); 2772160814Ssimon 2773160814Ssimon &set_label("14rounds"); 2774160814Ssimon &mov ("eax",&DWP(0,"esi")); # copy first 8 dwords 2775160814Ssimon &mov ("ebx",&DWP(4,"esi")); 2776160814Ssimon &mov ("ecx",&DWP(8,"esi")); 2777160814Ssimon &mov ("edx",&DWP(12,"esi")); 2778160814Ssimon &mov (&DWP(0,"edi"),"eax"); 2779160814Ssimon &mov (&DWP(4,"edi"),"ebx"); 2780160814Ssimon &mov (&DWP(8,"edi"),"ecx"); 2781160814Ssimon &mov (&DWP(12,"edi"),"edx"); 2782160814Ssimon &mov ("eax",&DWP(16,"esi")); 2783160814Ssimon &mov ("ebx",&DWP(20,"esi")); 2784160814Ssimon &mov ("ecx",&DWP(24,"esi")); 2785160814Ssimon &mov ("edx",&DWP(28,"esi")); 2786160814Ssimon &mov (&DWP(16,"edi"),"eax"); 2787160814Ssimon &mov (&DWP(20,"edi"),"ebx"); 2788160814Ssimon &mov (&DWP(24,"edi"),"ecx"); 2789160814Ssimon &mov (&DWP(28,"edi"),"edx"); 2790160814Ssimon 2791160814Ssimon &xor ("ecx","ecx"); 2792160814Ssimon &jmp (&label("14shortcut")); 2793160814Ssimon 2794160814Ssimon &align (4); 2795160814Ssimon &set_label("14loop"); 2796160814Ssimon &mov ("edx",&DWP(28,"edi")); # rk[7] 2797160814Ssimon &set_label("14shortcut"); 2798160814Ssimon &mov ("eax",&DWP(0,"edi")); # rk[0] 2799160814Ssimon 2800160814Ssimon &enckey (); 2801160814Ssimon 2802160814Ssimon &mov (&DWP(32,"edi"),"eax"); # rk[8] 2803160814Ssimon &xor ("eax",&DWP(4,"edi")); 2804160814Ssimon &mov (&DWP(36,"edi"),"eax"); # rk[9] 2805160814Ssimon &xor ("eax",&DWP(8,"edi")); 2806160814Ssimon &mov (&DWP(40,"edi"),"eax"); # rk[10] 2807160814Ssimon &xor ("eax",&DWP(12,"edi")); 2808160814Ssimon &mov (&DWP(44,"edi"),"eax"); # rk[11] 2809160814Ssimon 2810160814Ssimon &cmp ("ecx",6); 2811160814Ssimon &je (&label("14break")); 2812160814Ssimon &inc ("ecx"); 2813160814Ssimon 2814160814Ssimon &mov ("edx","eax"); 2815160814Ssimon &mov ("eax",&DWP(16,"edi")); # rk[4] 2816160814Ssimon &movz ("esi",&LB("edx")); # rk[11]>>0 2817238405Sjkim &movz ("ebx",&BP(-128,$tbl,"esi",1)); 2818160814Ssimon &movz ("esi",&HB("edx")); # rk[11]>>8 2819160814Ssimon &xor ("eax","ebx"); 2820160814Ssimon 2821238405Sjkim &movz ("ebx",&BP(-128,$tbl,"esi",1)); 2822160814Ssimon &shr ("edx",16); 2823238405Sjkim &shl ("ebx",8); 2824160814Ssimon &movz ("esi",&LB("edx")); # rk[11]>>16 2825160814Ssimon &xor ("eax","ebx"); 2826160814Ssimon 2827238405Sjkim &movz ("ebx",&BP(-128,$tbl,"esi",1)); 2828160814Ssimon &movz ("esi",&HB("edx")); # rk[11]>>24 2829238405Sjkim &shl ("ebx",16); 2830160814Ssimon &xor ("eax","ebx"); 2831160814Ssimon 2832238405Sjkim &movz ("ebx",&BP(-128,$tbl,"esi",1)); 2833238405Sjkim &shl ("ebx",24); 2834160814Ssimon &xor ("eax","ebx"); 2835160814Ssimon 2836160814Ssimon &mov (&DWP(48,"edi"),"eax"); # rk[12] 2837160814Ssimon &xor ("eax",&DWP(20,"edi")); 2838160814Ssimon &mov (&DWP(52,"edi"),"eax"); # rk[13] 2839160814Ssimon &xor ("eax",&DWP(24,"edi")); 2840160814Ssimon &mov (&DWP(56,"edi"),"eax"); # rk[14] 2841160814Ssimon &xor ("eax",&DWP(28,"edi")); 2842160814Ssimon &mov (&DWP(60,"edi"),"eax"); # rk[15] 2843160814Ssimon 2844160814Ssimon &add ("edi",32); 2845160814Ssimon &jmp (&label("14loop")); 2846160814Ssimon 2847160814Ssimon &set_label("14break"); 2848160814Ssimon &mov (&DWP(48,"edi"),14); # setup number of rounds 2849160814Ssimon &xor ("eax","eax"); 2850160814Ssimon &jmp (&label("exit")); 2851160814Ssimon 2852160814Ssimon &set_label("badpointer"); 2853160814Ssimon &mov ("eax",-1); 2854160814Ssimon &set_label("exit"); 2855238405Sjkim&function_end("_x86_AES_set_encrypt_key"); 2856160814Ssimon 2857238405Sjkim# int private_AES_set_encrypt_key(const unsigned char *userKey, const int bits, 2858238405Sjkim# AES_KEY *key) 2859238405Sjkim&function_begin_B("private_AES_set_encrypt_key"); 2860238405Sjkim &call ("_x86_AES_set_encrypt_key"); 2861238405Sjkim &ret (); 2862238405Sjkim&function_end_B("private_AES_set_encrypt_key"); 2863238405Sjkim 2864160814Ssimonsub deckey() 2865238405Sjkim{ my ($i,$key,$tp1,$tp2,$tp4,$tp8) = @_; 2866238405Sjkim my $tmp = $tbl; 2867160814Ssimon 2868238405Sjkim &mov ($acc,$tp1); 2869238405Sjkim &and ($acc,0x80808080); 2870238405Sjkim &mov ($tmp,$acc); 2871238405Sjkim &shr ($tmp,7); 2872238405Sjkim &lea ($tp2,&DWP(0,$tp1,$tp1)); 2873238405Sjkim &sub ($acc,$tmp); 2874238405Sjkim &and ($tp2,0xfefefefe); 2875238405Sjkim &and ($acc,0x1b1b1b1b); 2876238405Sjkim &xor ($acc,$tp2); 2877238405Sjkim &mov ($tp2,$acc); 2878238405Sjkim 2879238405Sjkim &and ($acc,0x80808080); 2880238405Sjkim &mov ($tmp,$acc); 2881238405Sjkim &shr ($tmp,7); 2882238405Sjkim &lea ($tp4,&DWP(0,$tp2,$tp2)); 2883238405Sjkim &sub ($acc,$tmp); 2884238405Sjkim &and ($tp4,0xfefefefe); 2885238405Sjkim &and ($acc,0x1b1b1b1b); 2886238405Sjkim &xor ($tp2,$tp1); # tp2^tp1 2887238405Sjkim &xor ($acc,$tp4); 2888238405Sjkim &mov ($tp4,$acc); 2889238405Sjkim 2890238405Sjkim &and ($acc,0x80808080); 2891238405Sjkim &mov ($tmp,$acc); 2892238405Sjkim &shr ($tmp,7); 2893238405Sjkim &lea ($tp8,&DWP(0,$tp4,$tp4)); 2894238405Sjkim &xor ($tp4,$tp1); # tp4^tp1 2895238405Sjkim &sub ($acc,$tmp); 2896238405Sjkim &and ($tp8,0xfefefefe); 2897238405Sjkim &and ($acc,0x1b1b1b1b); 2898238405Sjkim &rotl ($tp1,8); # = ROTATE(tp1,8) 2899238405Sjkim &xor ($tp8,$acc); 2900238405Sjkim 2901238405Sjkim &mov ($tmp,&DWP(4*($i+1),$key)); # modulo-scheduled load 2902238405Sjkim 2903238405Sjkim &xor ($tp1,$tp2); 2904238405Sjkim &xor ($tp2,$tp8); 2905238405Sjkim &xor ($tp1,$tp4); 2906238405Sjkim &rotl ($tp2,24); 2907238405Sjkim &xor ($tp4,$tp8); 2908238405Sjkim &xor ($tp1,$tp8); # ^= tp8^(tp4^tp1)^(tp2^tp1) 2909238405Sjkim &rotl ($tp4,16); 2910238405Sjkim &xor ($tp1,$tp2); # ^= ROTATE(tp8^tp2^tp1,24) 2911238405Sjkim &rotl ($tp8,8); 2912238405Sjkim &xor ($tp1,$tp4); # ^= ROTATE(tp8^tp4^tp1,16) 2913238405Sjkim &mov ($tp2,$tmp); 2914238405Sjkim &xor ($tp1,$tp8); # ^= ROTATE(tp8,8) 2915238405Sjkim 2916238405Sjkim &mov (&DWP(4*$i,$key),$tp1); 2917160814Ssimon} 2918160814Ssimon 2919238405Sjkim# int private_AES_set_decrypt_key(const unsigned char *userKey, const int bits, 2920160814Ssimon# AES_KEY *key) 2921238405Sjkim&function_begin_B("private_AES_set_decrypt_key"); 2922238405Sjkim &call ("_x86_AES_set_encrypt_key"); 2923160814Ssimon &cmp ("eax",0); 2924160814Ssimon &je (&label("proceed")); 2925160814Ssimon &ret (); 2926160814Ssimon 2927160814Ssimon &set_label("proceed"); 2928160814Ssimon &push ("ebp"); 2929160814Ssimon &push ("ebx"); 2930160814Ssimon &push ("esi"); 2931160814Ssimon &push ("edi"); 2932160814Ssimon 2933160814Ssimon &mov ("esi",&wparam(2)); 2934160814Ssimon &mov ("ecx",&DWP(240,"esi")); # pull number of rounds 2935160814Ssimon &lea ("ecx",&DWP(0,"","ecx",4)); 2936160814Ssimon &lea ("edi",&DWP(0,"esi","ecx",4)); # pointer to last chunk 2937160814Ssimon 2938238405Sjkim &set_label("invert",4); # invert order of chunks 2939160814Ssimon &mov ("eax",&DWP(0,"esi")); 2940160814Ssimon &mov ("ebx",&DWP(4,"esi")); 2941160814Ssimon &mov ("ecx",&DWP(0,"edi")); 2942160814Ssimon &mov ("edx",&DWP(4,"edi")); 2943160814Ssimon &mov (&DWP(0,"edi"),"eax"); 2944160814Ssimon &mov (&DWP(4,"edi"),"ebx"); 2945160814Ssimon &mov (&DWP(0,"esi"),"ecx"); 2946160814Ssimon &mov (&DWP(4,"esi"),"edx"); 2947160814Ssimon &mov ("eax",&DWP(8,"esi")); 2948160814Ssimon &mov ("ebx",&DWP(12,"esi")); 2949160814Ssimon &mov ("ecx",&DWP(8,"edi")); 2950160814Ssimon &mov ("edx",&DWP(12,"edi")); 2951160814Ssimon &mov (&DWP(8,"edi"),"eax"); 2952160814Ssimon &mov (&DWP(12,"edi"),"ebx"); 2953160814Ssimon &mov (&DWP(8,"esi"),"ecx"); 2954160814Ssimon &mov (&DWP(12,"esi"),"edx"); 2955160814Ssimon &add ("esi",16); 2956160814Ssimon &sub ("edi",16); 2957160814Ssimon &cmp ("esi","edi"); 2958160814Ssimon &jne (&label("invert")); 2959160814Ssimon 2960238405Sjkim &mov ($key,&wparam(2)); 2961238405Sjkim &mov ($acc,&DWP(240,$key)); # pull number of rounds 2962238405Sjkim &lea ($acc,&DWP(-2,$acc,$acc)); 2963238405Sjkim &lea ($acc,&DWP(0,$key,$acc,8)); 2964238405Sjkim &mov (&wparam(2),$acc); 2965160814Ssimon 2966238405Sjkim &mov ($s0,&DWP(16,$key)); # modulo-scheduled load 2967238405Sjkim &set_label("permute",4); # permute the key schedule 2968238405Sjkim &add ($key,16); 2969238405Sjkim &deckey (0,$key,$s0,$s1,$s2,$s3); 2970238405Sjkim &deckey (1,$key,$s1,$s2,$s3,$s0); 2971238405Sjkim &deckey (2,$key,$s2,$s3,$s0,$s1); 2972238405Sjkim &deckey (3,$key,$s3,$s0,$s1,$s2); 2973238405Sjkim &cmp ($key,&wparam(2)); 2974238405Sjkim &jb (&label("permute")); 2975160814Ssimon 2976160814Ssimon &xor ("eax","eax"); # return success 2977238405Sjkim&function_end("private_AES_set_decrypt_key"); 2978238405Sjkim&asciz("AES for x86, CRYPTOGAMS by <appro\@openssl.org>"); 2979160814Ssimon 2980160814Ssimon&asm_finish(); 2981