1#!/usr/bin/env perl
2
3######################################################################
4## Constant-time SSSE3 AES core implementation.
5## version 0.1
6##
7## By Mike Hamburg (Stanford University), 2009
8## Public domain.
9##
10## For details see http://shiftleft.org/papers/vector_aes/ and
11## http://crypto.stanford.edu/vpaes/.
12
13######################################################################
14# September 2011.
15#
16# Port vpaes-x86_64.pl as 32-bit "almost" drop-in replacement for
17# aes-586.pl. "Almost" refers to the fact that AES_cbc_encrypt
18# doesn't handle partial vectors (doesn't have to if called from
19# EVP only). "Drop-in" implies that this module doesn't share key
20# schedule structure with the original nor does it make assumption
21# about its alignment...
22#
23# Performance summary. aes-586.pl column lists large-block CBC
24# encrypt/decrypt/with-hyper-threading-off(*) results in cycles per
25# byte processed with 128-bit key, and vpaes-x86.pl column - [also
26# large-block CBC] encrypt/decrypt.
27#
28#		aes-586.pl		vpaes-x86.pl
29#
30# Core 2(**)	29.1/42.3/18.3		22.0/25.6(***)
31# Nehalem	27.9/40.4/18.1		10.3/12.0
32# Atom		102./119./60.1		64.5/85.3(***)
33#
34# (*)	"Hyper-threading" in the context refers rather to cache shared
35#	among multiple cores, than to specifically Intel HTT. As vast
36#	majority of contemporary cores share cache, slower code path
37#	is common place. In other words "with-hyper-threading-off"
38#	results are presented mostly for reference purposes.
39#
40# (**)	"Core 2" refers to initial 65nm design, a.k.a. Conroe.
41#
42# (***)	Less impressive improvement on Core 2 and Atom is due to slow
43#	pshufb,	yet it's respectable +32%/65%  improvement on Core 2
44#	and +58%/40% on Atom (as implied, over "hyper-threading-safe"
45#	code path).
46#
47#						<appro@openssl.org>
48
49$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
50push(@INC,"${dir}","${dir}../../perlasm");
51require "x86asm.pl";
52
53&asm_init($ARGV[0],"vpaes-x86.pl",$x86only = $ARGV[$#ARGV] eq "386");
54
55$PREFIX="vpaes";
56
57my  ($round, $base, $magic, $key, $const, $inp, $out)=
58    ("eax",  "ebx", "ecx",  "edx","ebp",  "esi","edi");
59
60	&rodataseg();
61&static_label("_vpaes_consts");
62&static_label("_vpaes_schedule_low_round");
63
64&set_label("_vpaes_consts",64);
65$k_inv=-0x30;		# inv, inva
66	&data_word(0x0D080180,0x0E05060F,0x0A0B0C02,0x04070309);
67	&data_word(0x0F0B0780,0x01040A06,0x02050809,0x030D0E0C);
68
69$k_s0F=-0x10;		# s0F
70	&data_word(0x0F0F0F0F,0x0F0F0F0F,0x0F0F0F0F,0x0F0F0F0F);
71
72$k_ipt=0x00;		# input transform (lo, hi)
73	&data_word(0x5A2A7000,0xC2B2E898,0x52227808,0xCABAE090);
74	&data_word(0x317C4D00,0x4C01307D,0xB0FDCC81,0xCD80B1FC);
75
76$k_sb1=0x20;		# sb1u, sb1t
77	&data_word(0xCB503E00,0xB19BE18F,0x142AF544,0xA5DF7A6E);
78	&data_word(0xFAE22300,0x3618D415,0x0D2ED9EF,0x3BF7CCC1);
79$k_sb2=0x40;		# sb2u, sb2t
80	&data_word(0x0B712400,0xE27A93C6,0xBC982FCD,0x5EB7E955);
81	&data_word(0x0AE12900,0x69EB8840,0xAB82234A,0xC2A163C8);
82$k_sbo=0x60;		# sbou, sbot
83	&data_word(0x6FBDC700,0xD0D26D17,0xC502A878,0x15AABF7A);
84	&data_word(0x5FBB6A00,0xCFE474A5,0x412B35FA,0x8E1E90D1);
85
86$k_mc_forward=0x80;	# mc_forward
87	&data_word(0x00030201,0x04070605,0x080B0A09,0x0C0F0E0D);
88	&data_word(0x04070605,0x080B0A09,0x0C0F0E0D,0x00030201);
89	&data_word(0x080B0A09,0x0C0F0E0D,0x00030201,0x04070605);
90	&data_word(0x0C0F0E0D,0x00030201,0x04070605,0x080B0A09);
91
92$k_mc_backward=0xc0;	# mc_backward
93	&data_word(0x02010003,0x06050407,0x0A09080B,0x0E0D0C0F);
94	&data_word(0x0E0D0C0F,0x02010003,0x06050407,0x0A09080B);
95	&data_word(0x0A09080B,0x0E0D0C0F,0x02010003,0x06050407);
96	&data_word(0x06050407,0x0A09080B,0x0E0D0C0F,0x02010003);
97
98$k_sr=0x100;		# sr
99	&data_word(0x03020100,0x07060504,0x0B0A0908,0x0F0E0D0C);
100	&data_word(0x0F0A0500,0x030E0904,0x07020D08,0x0B06010C);
101	&data_word(0x0B020900,0x0F060D04,0x030A0108,0x070E050C);
102	&data_word(0x070A0D00,0x0B0E0104,0x0F020508,0x0306090C);
103
104$k_rcon=0x140;		# rcon
105	&data_word(0xAF9DEEB6,0x1F8391B9,0x4D7C7D81,0x702A9808);
106
107$k_s63=0x150;		# s63: all equal to 0x63 transformed
108	&data_word(0x5B5B5B5B,0x5B5B5B5B,0x5B5B5B5B,0x5B5B5B5B);
109
110$k_opt=0x160;		# output transform
111	&data_word(0xD6B66000,0xFF9F4929,0xDEBE6808,0xF7974121);
112	&data_word(0x50BCEC00,0x01EDBD51,0xB05C0CE0,0xE10D5DB1);
113
114$k_deskew=0x180;	# deskew tables: inverts the sbox's "skew"
115	&data_word(0x47A4E300,0x07E4A340,0x5DBEF91A,0x1DFEB95A);
116	&data_word(0x83EA6900,0x5F36B5DC,0xF49D1E77,0x2841C2AB);
117##
118##  Decryption stuff
119##  Key schedule constants
120##
121$k_dksd=0x1a0;		# decryption key schedule: invskew x*D
122	&data_word(0xA3E44700,0xFEB91A5D,0x5A1DBEF9,0x0740E3A4);
123	&data_word(0xB5368300,0x41C277F4,0xAB289D1E,0x5FDC69EA);
124$k_dksb=0x1c0;		# decryption key schedule: invskew x*B
125	&data_word(0x8550D500,0x9A4FCA1F,0x1CC94C99,0x03D65386);
126	&data_word(0xB6FC4A00,0x115BEDA7,0x7E3482C8,0xD993256F);
127$k_dkse=0x1e0;		# decryption key schedule: invskew x*E + 0x63
128	&data_word(0x1FC9D600,0xD5031CCA,0x994F5086,0x53859A4C);
129	&data_word(0x4FDC7BE8,0xA2319605,0x20B31487,0xCD5EF96A);
130$k_dks9=0x200;		# decryption key schedule: invskew x*9
131	&data_word(0x7ED9A700,0xB6116FC8,0x82255BFC,0x4AED9334);
132	&data_word(0x27143300,0x45765162,0xE9DAFDCE,0x8BB89FAC);
133
134##
135##  Decryption stuff
136##  Round function constants
137##
138$k_dipt=0x220;		# decryption input transform
139	&data_word(0x0B545F00,0x0F505B04,0x114E451A,0x154A411E);
140	&data_word(0x60056500,0x86E383E6,0xF491F194,0x12771772);
141
142$k_dsb9=0x240;		# decryption sbox output *9*u, *9*t
143	&data_word(0x9A86D600,0x851C0353,0x4F994CC9,0xCAD51F50);
144	&data_word(0xECD74900,0xC03B1789,0xB2FBA565,0x725E2C9E);
145$k_dsbd=0x260;		# decryption sbox output *D*u, *D*t
146	&data_word(0xE6B1A200,0x7D57CCDF,0x882A4439,0xF56E9B13);
147	&data_word(0x24C6CB00,0x3CE2FAF7,0x15DEEFD3,0x2931180D);
148$k_dsbb=0x280;		# decryption sbox output *B*u, *B*t
149	&data_word(0x96B44200,0xD0226492,0xB0F2D404,0x602646F6);
150	&data_word(0xCD596700,0xC19498A6,0x3255AA6B,0xF3FF0C3E);
151$k_dsbe=0x2a0;		# decryption sbox output *E*u, *E*t
152	&data_word(0x26D4D000,0x46F29296,0x64B4F6B0,0x22426004);
153	&data_word(0xFFAAC100,0x0C55A6CD,0x98593E32,0x9467F36B);
154$k_dsbo=0x2c0;		# decryption sbox final output
155	&data_word(0x7EF94000,0x1387EA53,0xD4943E2D,0xC7AA6DB9);
156	&data_word(0x93441D00,0x12D7560F,0xD8C58E9C,0xCA4B8159);
157	&previous();
158
159&function_begin_B("_vpaes_preheat");
160	&movdqa	("xmm7",&QWP($k_inv,$const));
161	&movdqa	("xmm6",&QWP($k_s0F,$const));
162	&ret	();
163&function_end_B("_vpaes_preheat");
164
165##
166##  _aes_encrypt_core
167##
168##  AES-encrypt %xmm0.
169##
170##  Inputs:
171##     %xmm0 = input
172##     %xmm6-%xmm7 as in _vpaes_preheat
173##    (%edx) = scheduled keys
174##
175##  Output in %xmm0
176##  Clobbers  %xmm1-%xmm5, %eax, %ebx, %ecx, %edx
177##
178##
179&function_begin_B("_vpaes_encrypt_core");
180	&mov	($magic,16);
181	&mov	($round,&DWP(240,$key));
182	&movdqa	("xmm1","xmm6")
183	&movdqa	("xmm2",&QWP($k_ipt,$const));
184	&pandn	("xmm1","xmm0");
185	&movdqu	("xmm5",&QWP(0,$key));
186	&psrld	("xmm1",4);
187	&pand	("xmm0","xmm6");
188	&pshufb	("xmm2","xmm0");
189	&movdqa	("xmm0",&QWP($k_ipt+16,$const));
190	&pshufb	("xmm0","xmm1");
191	&pxor	("xmm2","xmm5");
192	&pxor	("xmm0","xmm2");
193	&add	($key,16);
194	&lea	($base,&DWP($k_mc_backward,$const));
195	&jmp	(&label("enc_entry"));
196
197
198&set_label("enc_loop",16);
199	# middle of middle round
200	&movdqa	("xmm4",&QWP($k_sb1,$const));	# 4 : sb1u
201	&pshufb	("xmm4","xmm2");		# 4 = sb1u
202	&pxor	("xmm4","xmm5");		# 4 = sb1u + k
203	&movdqa	("xmm0",&QWP($k_sb1+16,$const));# 0 : sb1t
204	&pshufb	("xmm0","xmm3");		# 0 = sb1t
205	&pxor	("xmm0","xmm4");		# 0 = A
206	&movdqa	("xmm5",&QWP($k_sb2,$const));	# 4 : sb2u
207	&pshufb	("xmm5","xmm2");		# 4 = sb2u
208	&movdqa	("xmm1",&QWP(-0x40,$base,$magic));# .Lk_mc_forward[]
209	&movdqa	("xmm2",&QWP($k_sb2+16,$const));# 2 : sb2t
210	&pshufb	("xmm2","xmm3");		# 2 = sb2t
211	&pxor	("xmm2","xmm5");		# 2 = 2A
212	&movdqa	("xmm4",&QWP(0,$base,$magic));	# .Lk_mc_backward[]
213	&movdqa	("xmm3","xmm0");		# 3 = A
214	&pshufb	("xmm0","xmm1");		# 0 = B
215	&add	($key,16);			# next key
216	&pxor	("xmm0","xmm2");		# 0 = 2A+B
217	&pshufb	("xmm3","xmm4");		# 3 = D
218	&add	($magic,16);			# next mc
219	&pxor	("xmm3","xmm0");		# 3 = 2A+B+D
220	&pshufb	("xmm0","xmm1");		# 0 = 2B+C
221	&and	($magic,0x30);			# ... mod 4
222	&pxor	("xmm0","xmm3");		# 0 = 2A+3B+C+D
223	&sub	($round,1);			# nr--
224
225&set_label("enc_entry");
226	# top of round
227	&movdqa	("xmm1","xmm6");		# 1 : i
228	&pandn	("xmm1","xmm0");		# 1 = i<<4
229	&psrld	("xmm1",4);			# 1 = i
230	&pand	("xmm0","xmm6");		# 0 = k
231	&movdqa	("xmm5",&QWP($k_inv+16,$const));# 2 : a/k
232	&pshufb	("xmm5","xmm0");		# 2 = a/k
233	&pxor	("xmm0","xmm1");		# 0 = j
234	&movdqa	("xmm3","xmm7");		# 3 : 1/i
235	&pshufb	("xmm3","xmm1");		# 3 = 1/i
236	&pxor	("xmm3","xmm5");		# 3 = iak = 1/i + a/k
237	&movdqa	("xmm4","xmm7");		# 4 : 1/j
238	&pshufb	("xmm4","xmm0");		# 4 = 1/j
239	&pxor	("xmm4","xmm5");		# 4 = jak = 1/j + a/k
240	&movdqa	("xmm2","xmm7");		# 2 : 1/iak
241	&pshufb	("xmm2","xmm3");		# 2 = 1/iak
242	&pxor	("xmm2","xmm0");		# 2 = io
243	&movdqa	("xmm3","xmm7");		# 3 : 1/jak
244	&movdqu	("xmm5",&QWP(0,$key));
245	&pshufb	("xmm3","xmm4");		# 3 = 1/jak
246	&pxor	("xmm3","xmm1");		# 3 = jo
247	&jnz	(&label("enc_loop"));
248
249	# middle of last round
250	&movdqa	("xmm4",&QWP($k_sbo,$const));	# 3 : sbou      .Lk_sbo
251	&movdqa	("xmm0",&QWP($k_sbo+16,$const));# 3 : sbot      .Lk_sbo+16
252	&pshufb	("xmm4","xmm2");		# 4 = sbou
253	&pxor	("xmm4","xmm5");		# 4 = sb1u + k
254	&pshufb	("xmm0","xmm3");		# 0 = sb1t
255	&movdqa	("xmm1",&QWP(0x40,$base,$magic));# .Lk_sr[]
256	&pxor	("xmm0","xmm4");		# 0 = A
257	&pshufb	("xmm0","xmm1");
258	&ret	();
259&function_end_B("_vpaes_encrypt_core");
260
261##
262##  Decryption core
263##
264##  Same API as encryption core.
265##
266&function_begin_B("_vpaes_decrypt_core");
267	&mov	($round,&DWP(240,$key));
268	&lea	($base,&DWP($k_dsbd,$const));
269	&movdqa	("xmm1","xmm6");
270	&movdqa	("xmm2",&QWP($k_dipt-$k_dsbd,$base));
271	&pandn	("xmm1","xmm0");
272	&mov	($magic,$round);
273	&psrld	("xmm1",4)
274	&movdqu	("xmm5",&QWP(0,$key));
275	&shl	($magic,4);
276	&pand	("xmm0","xmm6");
277	&pshufb	("xmm2","xmm0");
278	&movdqa	("xmm0",&QWP($k_dipt-$k_dsbd+16,$base));
279	&xor	($magic,0x30);
280	&pshufb	("xmm0","xmm1");
281	&and	($magic,0x30);
282	&pxor	("xmm2","xmm5");
283	&movdqa	("xmm5",&QWP($k_mc_forward+48,$const));
284	&pxor	("xmm0","xmm2");
285	&add	($key,16);
286	&lea	($magic,&DWP($k_sr-$k_dsbd,$base,$magic));
287	&jmp	(&label("dec_entry"));
288
289&set_label("dec_loop",16);
290##
291##  Inverse mix columns
292##
293	&movdqa	("xmm4",&QWP(-0x20,$base));	# 4 : sb9u
294	&pshufb	("xmm4","xmm2");		# 4 = sb9u
295	&pxor	("xmm4","xmm0");
296	&movdqa	("xmm0",&QWP(-0x10,$base));	# 0 : sb9t
297	&pshufb	("xmm0","xmm3");		# 0 = sb9t
298	&pxor	("xmm0","xmm4");		# 0 = ch
299	&add	($key,16);			# next round key
300
301	&pshufb	("xmm0","xmm5");		# MC ch
302	&movdqa	("xmm4",&QWP(0,$base));		# 4 : sbdu
303	&pshufb	("xmm4","xmm2");		# 4 = sbdu
304	&pxor	("xmm4","xmm0");		# 4 = ch
305	&movdqa	("xmm0",&QWP(0x10,$base));	# 0 : sbdt
306	&pshufb	("xmm0","xmm3");		# 0 = sbdt
307	&pxor	("xmm0","xmm4");		# 0 = ch
308	&sub	($round,1);			# nr--
309
310	&pshufb	("xmm0","xmm5");		# MC ch
311	&movdqa	("xmm4",&QWP(0x20,$base));	# 4 : sbbu
312	&pshufb	("xmm4","xmm2");		# 4 = sbbu
313	&pxor	("xmm4","xmm0");		# 4 = ch
314	&movdqa	("xmm0",&QWP(0x30,$base));	# 0 : sbbt
315	&pshufb	("xmm0","xmm3");		# 0 = sbbt
316	&pxor	("xmm0","xmm4");		# 0 = ch
317
318	&pshufb	("xmm0","xmm5");		# MC ch
319	&movdqa	("xmm4",&QWP(0x40,$base));	# 4 : sbeu
320	&pshufb	("xmm4","xmm2");		# 4 = sbeu
321	&pxor	("xmm4","xmm0");		# 4 = ch
322	&movdqa	("xmm0",&QWP(0x50,$base));	# 0 : sbet
323	&pshufb	("xmm0","xmm3");		# 0 = sbet
324	&pxor	("xmm0","xmm4");		# 0 = ch
325
326	&palignr("xmm5","xmm5",12);
327
328&set_label("dec_entry");
329	# top of round
330	&movdqa	("xmm1","xmm6");		# 1 : i
331	&pandn	("xmm1","xmm0");		# 1 = i<<4
332	&psrld	("xmm1",4);			# 1 = i
333	&pand	("xmm0","xmm6");		# 0 = k
334	&movdqa	("xmm2",&QWP($k_inv+16,$const));# 2 : a/k
335	&pshufb	("xmm2","xmm0");		# 2 = a/k
336	&pxor	("xmm0","xmm1");		# 0 = j
337	&movdqa	("xmm3","xmm7");		# 3 : 1/i
338	&pshufb	("xmm3","xmm1");		# 3 = 1/i
339	&pxor	("xmm3","xmm2");		# 3 = iak = 1/i + a/k
340	&movdqa	("xmm4","xmm7");		# 4 : 1/j
341	&pshufb	("xmm4","xmm0");		# 4 = 1/j
342	&pxor	("xmm4","xmm2");		# 4 = jak = 1/j + a/k
343	&movdqa	("xmm2","xmm7");		# 2 : 1/iak
344	&pshufb	("xmm2","xmm3");		# 2 = 1/iak
345	&pxor	("xmm2","xmm0");		# 2 = io
346	&movdqa	("xmm3","xmm7");		# 3 : 1/jak
347	&pshufb	("xmm3","xmm4");		# 3 = 1/jak
348	&pxor	("xmm3","xmm1");		# 3 = jo
349	&movdqu	("xmm0",&QWP(0,$key));
350	&jnz	(&label("dec_loop"));
351
352	# middle of last round
353	&movdqa	("xmm4",&QWP(0x60,$base));	# 3 : sbou
354	&pshufb	("xmm4","xmm2");		# 4 = sbou
355	&pxor	("xmm4","xmm0");		# 4 = sb1u + k
356	&movdqa	("xmm0",&QWP(0x70,$base));	# 0 : sbot
357	&movdqa	("xmm2",&QWP(0,$magic));
358	&pshufb	("xmm0","xmm3");		# 0 = sb1t
359	&pxor	("xmm0","xmm4");		# 0 = A
360	&pshufb	("xmm0","xmm2");
361	&ret	();
362&function_end_B("_vpaes_decrypt_core");
363
364########################################################
365##                                                    ##
366##                  AES key schedule                  ##
367##                                                    ##
368########################################################
369&function_begin_B("_vpaes_schedule_core");
370	&movdqu	("xmm0",&QWP(0,$inp));		# load key (unaligned)
371	&movdqa	("xmm2",&QWP($k_rcon,$const));	# load rcon
372
373	# input transform
374	&movdqa	("xmm3","xmm0");
375	&lea	($base,&DWP($k_ipt,$const));
376	&movdqa	(&QWP(4,"esp"),"xmm2");		# xmm8
377	&call	("_vpaes_schedule_transform");
378	&movdqa	("xmm7","xmm0");
379
380	&test	($out,$out);
381	&jnz	(&label("schedule_am_decrypting"));
382
383	# encrypting, output zeroth round key after transform
384	&movdqu	(&QWP(0,$key),"xmm0");
385	&jmp	(&label("schedule_go"));
386
387&set_label("schedule_am_decrypting");
388	# decrypting, output zeroth round key after shiftrows
389	&movdqa	("xmm1",&QWP($k_sr,$const,$magic));
390	&pshufb	("xmm3","xmm1");
391	&movdqu	(&QWP(0,$key),"xmm3");
392	&xor	($magic,0x30);
393
394&set_label("schedule_go");
395	&cmp	($round,192);
396	&ja	(&label("schedule_256"));
397	&je	(&label("schedule_192"));
398	# 128: fall though
399
400##
401##  .schedule_128
402##
403##  128-bit specific part of key schedule.
404##
405##  This schedule is really simple, because all its parts
406##  are accomplished by the subroutines.
407##
408&set_label("schedule_128");
409	&mov	($round,10);
410
411&set_label("loop_schedule_128");
412	&call	("_vpaes_schedule_round");
413	&dec	($round);
414	&jz	(&label("schedule_mangle_last"));
415	&call	("_vpaes_schedule_mangle");	# write output
416	&jmp	(&label("loop_schedule_128"));
417
418##
419##  .aes_schedule_192
420##
421##  192-bit specific part of key schedule.
422##
423##  The main body of this schedule is the same as the 128-bit
424##  schedule, but with more smearing.  The long, high side is
425##  stored in %xmm7 as before, and the short, low side is in
426##  the high bits of %xmm6.
427##
428##  This schedule is somewhat nastier, however, because each
429##  round produces 192 bits of key material, or 1.5 round keys.
430##  Therefore, on each cycle we do 2 rounds and produce 3 round
431##  keys.
432##
433&set_label("schedule_192",16);
434	&movdqu	("xmm0",&QWP(8,$inp));		# load key part 2 (very unaligned)
435	&call	("_vpaes_schedule_transform");	# input transform
436	&movdqa	("xmm6","xmm0");		# save short part
437	&pxor	("xmm4","xmm4");		# clear 4
438	&movhlps("xmm6","xmm4");		# clobber low side with zeros
439	&mov	($round,4);
440
441&set_label("loop_schedule_192");
442	&call	("_vpaes_schedule_round");
443	&palignr("xmm0","xmm6",8);
444	&call	("_vpaes_schedule_mangle");	# save key n
445	&call	("_vpaes_schedule_192_smear");
446	&call	("_vpaes_schedule_mangle");	# save key n+1
447	&call	("_vpaes_schedule_round");
448	&dec	($round);
449	&jz	(&label("schedule_mangle_last"));
450	&call	("_vpaes_schedule_mangle");	# save key n+2
451	&call	("_vpaes_schedule_192_smear");
452	&jmp	(&label("loop_schedule_192"));
453
454##
455##  .aes_schedule_256
456##
457##  256-bit specific part of key schedule.
458##
459##  The structure here is very similar to the 128-bit
460##  schedule, but with an additional "low side" in
461##  %xmm6.  The low side's rounds are the same as the
462##  high side's, except no rcon and no rotation.
463##
464&set_label("schedule_256",16);
465	&movdqu	("xmm0",&QWP(16,$inp));		# load key part 2 (unaligned)
466	&call	("_vpaes_schedule_transform");	# input transform
467	&mov	($round,7);
468
469&set_label("loop_schedule_256");
470	&call	("_vpaes_schedule_mangle");	# output low result
471	&movdqa	("xmm6","xmm0");		# save cur_lo in xmm6
472
473	# high round
474	&call	("_vpaes_schedule_round");
475	&dec	($round);
476	&jz	(&label("schedule_mangle_last"));
477	&call	("_vpaes_schedule_mangle");
478
479	# low round. swap xmm7 and xmm6
480	&pshufd	("xmm0","xmm0",0xFF);
481	&movdqa	(&QWP(20,"esp"),"xmm7");
482	&movdqa	("xmm7","xmm6");
483	&call	("_vpaes_schedule_low_round");
484	&movdqa	("xmm7",&QWP(20,"esp"));
485
486	&jmp	(&label("loop_schedule_256"));
487
488##
489##  .aes_schedule_mangle_last
490##
491##  Mangler for last round of key schedule
492##  Mangles %xmm0
493##    when encrypting, outputs out(%xmm0) ^ 63
494##    when decrypting, outputs unskew(%xmm0)
495##
496##  Always called right before return... jumps to cleanup and exits
497##
498&set_label("schedule_mangle_last",16);
499	# schedule last round key from xmm0
500	&lea	($base,&DWP($k_deskew,$const));
501	&test	($out,$out);
502	&jnz	(&label("schedule_mangle_last_dec"));
503
504	# encrypting
505	&movdqa	("xmm1",&QWP($k_sr,$const,$magic));
506	&pshufb	("xmm0","xmm1");		# output permute
507	&lea	($base,&DWP($k_opt,$const));	# prepare to output transform
508	&add	($key,32);
509
510&set_label("schedule_mangle_last_dec");
511	&add	($key,-16);
512	&pxor	("xmm0",&QWP($k_s63,$const));
513	&call	("_vpaes_schedule_transform");	# output transform
514	&movdqu	(&QWP(0,$key),"xmm0");		# save last key
515
516	# cleanup
517	&pxor	("xmm0","xmm0");
518	&pxor	("xmm1","xmm1");
519	&pxor	("xmm2","xmm2");
520	&pxor	("xmm3","xmm3");
521	&pxor	("xmm4","xmm4");
522	&pxor	("xmm5","xmm5");
523	&pxor	("xmm6","xmm6");
524	&pxor	("xmm7","xmm7");
525	&ret	();
526&function_end_B("_vpaes_schedule_core");
527
528##
529##  .aes_schedule_192_smear
530##
531##  Smear the short, low side in the 192-bit key schedule.
532##
533##  Inputs:
534##    %xmm7: high side, b  a  x  y
535##    %xmm6:  low side, d  c  0  0
536##    %xmm13: 0
537##
538##  Outputs:
539##    %xmm6: b+c+d  b+c  0  0
540##    %xmm0: b+c+d  b+c  b  a
541##
542&function_begin_B("_vpaes_schedule_192_smear");
543	&pshufd	("xmm0","xmm6",0x80);		# d c 0 0 -> c 0 0 0
544	&pxor	("xmm6","xmm0");		# -> c+d c 0 0
545	&pshufd	("xmm0","xmm7",0xFE);		# b a _ _ -> b b b a
546	&pxor	("xmm6","xmm0");		# -> b+c+d b+c b a
547	&movdqa	("xmm0","xmm6");
548	&pxor	("xmm1","xmm1");
549	&movhlps("xmm6","xmm1");		# clobber low side with zeros
550	&ret	();
551&function_end_B("_vpaes_schedule_192_smear");
552
553##
554##  .aes_schedule_round
555##
556##  Runs one main round of the key schedule on %xmm0, %xmm7
557##
558##  Specifically, runs subbytes on the high dword of %xmm0
559##  then rotates it by one byte and xors into the low dword of
560##  %xmm7.
561##
562##  Adds rcon from low byte of %xmm8, then rotates %xmm8 for
563##  next rcon.
564##
565##  Smears the dwords of %xmm7 by xoring the low into the
566##  second low, result into third, result into highest.
567##
568##  Returns results in %xmm7 = %xmm0.
569##  Clobbers %xmm1-%xmm5.
570##
571&function_begin_B("_vpaes_schedule_round");
572	# extract rcon from xmm8
573	&movdqa	("xmm2",&QWP(8,"esp"));		# xmm8
574	&pxor	("xmm1","xmm1");
575	&palignr("xmm1","xmm2",15);
576	&palignr("xmm2","xmm2",15);
577	&pxor	("xmm7","xmm1");
578
579	# rotate
580	&pshufd	("xmm0","xmm0",0xFF);
581	&palignr("xmm0","xmm0",1);
582
583	# fall through...
584	&movdqa	(&QWP(8,"esp"),"xmm2");		# xmm8
585
586	# low round: same as high round, but no rotation and no rcon.
587&set_label("_vpaes_schedule_low_round");
588	# smear xmm7
589	&movdqa	("xmm1","xmm7");
590	&pslldq	("xmm7",4);
591	&pxor	("xmm7","xmm1");
592	&movdqa	("xmm1","xmm7");
593	&pslldq	("xmm7",8);
594	&pxor	("xmm7","xmm1");
595	&pxor	("xmm7",&QWP($k_s63,$const));
596
597	# subbyte
598	&movdqa	("xmm4",&QWP($k_s0F,$const));
599	&movdqa	("xmm5",&QWP($k_inv,$const));	# 4 : 1/j
600	&movdqa	("xmm1","xmm4");
601	&pandn	("xmm1","xmm0");
602	&psrld	("xmm1",4);			# 1 = i
603	&pand	("xmm0","xmm4");		# 0 = k
604	&movdqa	("xmm2",&QWP($k_inv+16,$const));# 2 : a/k
605	&pshufb	("xmm2","xmm0");		# 2 = a/k
606	&pxor	("xmm0","xmm1");		# 0 = j
607	&movdqa	("xmm3","xmm5");		# 3 : 1/i
608	&pshufb	("xmm3","xmm1");		# 3 = 1/i
609	&pxor	("xmm3","xmm2");		# 3 = iak = 1/i + a/k
610	&movdqa	("xmm4","xmm5");		# 4 : 1/j
611	&pshufb	("xmm4","xmm0");		# 4 = 1/j
612	&pxor	("xmm4","xmm2");		# 4 = jak = 1/j + a/k
613	&movdqa	("xmm2","xmm5");		# 2 : 1/iak
614	&pshufb	("xmm2","xmm3");		# 2 = 1/iak
615	&pxor	("xmm2","xmm0");		# 2 = io
616	&movdqa	("xmm3","xmm5");		# 3 : 1/jak
617	&pshufb	("xmm3","xmm4");		# 3 = 1/jak
618	&pxor	("xmm3","xmm1");		# 3 = jo
619	&movdqa	("xmm4",&QWP($k_sb1,$const));	# 4 : sbou
620	&pshufb	("xmm4","xmm2");		# 4 = sbou
621	&movdqa	("xmm0",&QWP($k_sb1+16,$const));# 0 : sbot
622	&pshufb	("xmm0","xmm3");		# 0 = sb1t
623	&pxor	("xmm0","xmm4");		# 0 = sbox output
624
625	# add in smeared stuff
626	&pxor	("xmm0","xmm7");
627	&movdqa	("xmm7","xmm0");
628	&ret	();
629&function_end_B("_vpaes_schedule_round");
630
631##
632##  .aes_schedule_transform
633##
634##  Linear-transform %xmm0 according to tables at (%ebx)
635##
636##  Output in %xmm0
637##  Clobbers %xmm1, %xmm2
638##
639&function_begin_B("_vpaes_schedule_transform");
640	&movdqa	("xmm2",&QWP($k_s0F,$const));
641	&movdqa	("xmm1","xmm2");
642	&pandn	("xmm1","xmm0");
643	&psrld	("xmm1",4);
644	&pand	("xmm0","xmm2");
645	&movdqa	("xmm2",&QWP(0,$base));
646	&pshufb	("xmm2","xmm0");
647	&movdqa	("xmm0",&QWP(16,$base));
648	&pshufb	("xmm0","xmm1");
649	&pxor	("xmm0","xmm2");
650	&ret	();
651&function_end_B("_vpaes_schedule_transform");
652
653##
654##  .aes_schedule_mangle
655##
656##  Mangle xmm0 from (basis-transformed) standard version
657##  to our version.
658##
659##  On encrypt,
660##    xor with 0x63
661##    multiply by circulant 0,1,1,1
662##    apply shiftrows transform
663##
664##  On decrypt,
665##    xor with 0x63
666##    multiply by "inverse mixcolumns" circulant E,B,D,9
667##    deskew
668##    apply shiftrows transform
669##
670##
671##  Writes out to (%edx), and increments or decrements it
672##  Keeps track of round number mod 4 in %ecx
673##  Preserves xmm0
674##  Clobbers xmm1-xmm5
675##
676&function_begin_B("_vpaes_schedule_mangle");
677	&movdqa	("xmm4","xmm0");	# save xmm0 for later
678	&movdqa	("xmm5",&QWP($k_mc_forward,$const));
679	&test	($out,$out);
680	&jnz	(&label("schedule_mangle_dec"));
681
682	# encrypting
683	&add	($key,16);
684	&pxor	("xmm4",&QWP($k_s63,$const));
685	&pshufb	("xmm4","xmm5");
686	&movdqa	("xmm3","xmm4");
687	&pshufb	("xmm4","xmm5");
688	&pxor	("xmm3","xmm4");
689	&pshufb	("xmm4","xmm5");
690	&pxor	("xmm3","xmm4");
691
692	&jmp	(&label("schedule_mangle_both"));
693
694&set_label("schedule_mangle_dec",16);
695	# inverse mix columns
696	&movdqa	("xmm2",&QWP($k_s0F,$const));
697	&lea	($inp,&DWP($k_dksd,$const));
698	&movdqa	("xmm1","xmm2");
699	&pandn	("xmm1","xmm4");
700	&psrld	("xmm1",4);			# 1 = hi
701	&pand	("xmm4","xmm2");		# 4 = lo
702
703	&movdqa	("xmm2",&QWP(0,$inp));
704	&pshufb	("xmm2","xmm4");
705	&movdqa	("xmm3",&QWP(0x10,$inp));
706	&pshufb	("xmm3","xmm1");
707	&pxor	("xmm3","xmm2");
708	&pshufb	("xmm3","xmm5");
709
710	&movdqa	("xmm2",&QWP(0x20,$inp));
711	&pshufb	("xmm2","xmm4");
712	&pxor	("xmm2","xmm3");
713	&movdqa	("xmm3",&QWP(0x30,$inp));
714	&pshufb	("xmm3","xmm1");
715	&pxor	("xmm3","xmm2");
716	&pshufb	("xmm3","xmm5");
717
718	&movdqa	("xmm2",&QWP(0x40,$inp));
719	&pshufb	("xmm2","xmm4");
720	&pxor	("xmm2","xmm3");
721	&movdqa	("xmm3",&QWP(0x50,$inp));
722	&pshufb	("xmm3","xmm1");
723	&pxor	("xmm3","xmm2");
724	&pshufb	("xmm3","xmm5");
725
726	&movdqa	("xmm2",&QWP(0x60,$inp));
727	&pshufb	("xmm2","xmm4");
728	&pxor	("xmm2","xmm3");
729	&movdqa	("xmm3",&QWP(0x70,$inp));
730	&pshufb	("xmm3","xmm1");
731	&pxor	("xmm3","xmm2");
732
733	&add	($key,-16);
734
735&set_label("schedule_mangle_both");
736	&movdqa	("xmm1",&QWP($k_sr,$const,$magic));
737	&pshufb	("xmm3","xmm1");
738	&add	($magic,-16);
739	&and	($magic,0x30);
740	&movdqu	(&QWP(0,$key),"xmm3");
741	&ret	();
742&function_end_B("_vpaes_schedule_mangle");
743
744#
745# Interface to OpenSSL
746#
747&function_begin("${PREFIX}_set_encrypt_key");
748	&mov	($inp,&wparam(0));		# inp
749	&lea	($base,&DWP(-56,"esp"));
750	&mov	($round,&wparam(1));		# bits
751	&and	($base,-16);
752	&mov	($key,&wparam(2));		# key
753	&xchg	($base,"esp");			# alloca
754	&mov	(&DWP(48,"esp"),$base);
755
756	&mov	($base,$round);
757	&shr	($base,5);
758	&add	($base,5);
759	&mov	(&DWP(240,$key),$base);		# AES_KEY->rounds = nbits/32+5;
760	&mov	($magic,0x30);
761	&mov	($out,0);
762
763	&picsetup($const);
764	&picsymbol($const, &label("_vpaes_consts"), $const);
765	&lea	($const,&DWP(0x30,$const))
766
767	&call	("_vpaes_schedule_core");
768
769	&mov	("esp",&DWP(48,"esp"));
770	&xor	("eax","eax");
771&function_end("${PREFIX}_set_encrypt_key");
772
773&function_begin("${PREFIX}_set_decrypt_key");
774	&mov	($inp,&wparam(0));		# inp
775	&lea	($base,&DWP(-56,"esp"));
776	&mov	($round,&wparam(1));		# bits
777	&and	($base,-16);
778	&mov	($key,&wparam(2));		# key
779	&xchg	($base,"esp");			# alloca
780	&mov	(&DWP(48,"esp"),$base);
781
782	&mov	($base,$round);
783	&shr	($base,5);
784	&add	($base,5);
785	&mov	(&DWP(240,$key),$base);	# AES_KEY->rounds = nbits/32+5;
786	&shl	($base,4);
787	&lea	($key,&DWP(16,$key,$base));
788
789	&mov	($out,1);
790	&mov	($magic,$round);
791	&shr	($magic,1);
792	&and	($magic,32);
793	&xor	($magic,32);			# nbist==192?0:32;
794
795	&picsetup($const);
796	&picsymbol($const, &label("_vpaes_consts"), $const);
797	&lea	($const,&DWP(0x30,$const))
798
799	&call	("_vpaes_schedule_core");
800
801	&mov	("esp",&DWP(48,"esp"));
802	&xor	("eax","eax");
803&function_end("${PREFIX}_set_decrypt_key");
804
805&function_begin("${PREFIX}_encrypt");
806	&picsetup($const);
807	&picsymbol($const, &label("_vpaes_consts"), $const);
808	&lea	($const,&DWP(0x30,$const))
809
810	&call	("_vpaes_preheat");
811	&mov	($inp,&wparam(0));		# inp
812	&lea	($base,&DWP(-56,"esp"));
813	&mov	($out,&wparam(1));		# out
814	&and	($base,-16);
815	&mov	($key,&wparam(2));		# key
816	&xchg	($base,"esp");			# alloca
817	&mov	(&DWP(48,"esp"),$base);
818
819	&movdqu	("xmm0",&QWP(0,$inp));
820	&call	("_vpaes_encrypt_core");
821	&movdqu	(&QWP(0,$out),"xmm0");
822
823	&mov	("esp",&DWP(48,"esp"));
824&function_end("${PREFIX}_encrypt");
825
826&function_begin("${PREFIX}_decrypt");
827	&picsetup($const);
828	&picsymbol($const, &label("_vpaes_consts"), $const);
829	&lea	($const,&DWP(0x30,$const))
830
831	&call	("_vpaes_preheat");
832	&mov	($inp,&wparam(0));		# inp
833	&lea	($base,&DWP(-56,"esp"));
834	&mov	($out,&wparam(1));		# out
835	&and	($base,-16);
836	&mov	($key,&wparam(2));		# key
837	&xchg	($base,"esp");			# alloca
838	&mov	(&DWP(48,"esp"),$base);
839
840	&movdqu	("xmm0",&QWP(0,$inp));
841	&call	("_vpaes_decrypt_core");
842	&movdqu	(&QWP(0,$out),"xmm0");
843
844	&mov	("esp",&DWP(48,"esp"));
845&function_end("${PREFIX}_decrypt");
846
847&function_begin("${PREFIX}_cbc_encrypt");
848	&mov	($inp,&wparam(0));		# inp
849	&mov	($out,&wparam(1));		# out
850	&mov	($round,&wparam(2));		# len
851	&mov	($key,&wparam(3));		# key
852	&sub	($round,16);
853	&jc	(&label("cbc_abort"));
854	&lea	($base,&DWP(-56,"esp"));
855	&mov	($const,&wparam(4));		# ivp
856	&and	($base,-16);
857	&mov	($magic,&wparam(5));		# enc
858	&xchg	($base,"esp");			# alloca
859	&movdqu	("xmm1",&QWP(0,$const));	# load IV
860	&sub	($out,$inp);
861	&mov	(&DWP(48,"esp"),$base);
862
863	&mov	(&DWP(0,"esp"),$out);		# save out
864	&mov	(&DWP(4,"esp"),$key)		# save key
865	&mov	(&DWP(8,"esp"),$const);		# save ivp
866	&mov	($out,$round);			# $out works as $len
867
868	&picsetup($const);
869	&picsymbol($const, &label("_vpaes_consts"), $const);
870	&lea	($const,&DWP(0x30,$const))
871
872	&call	("_vpaes_preheat");
873	&cmp	($magic,0);
874	&je	(&label("cbc_dec_loop"));
875	&jmp	(&label("cbc_enc_loop"));
876
877&set_label("cbc_enc_loop",16);
878	&movdqu	("xmm0",&QWP(0,$inp));		# load input
879	&pxor	("xmm0","xmm1");		# inp^=iv
880	&call	("_vpaes_encrypt_core");
881	&mov	($base,&DWP(0,"esp"));		# restore out
882	&mov	($key,&DWP(4,"esp"));		# restore key
883	&movdqa	("xmm1","xmm0");
884	&movdqu	(&QWP(0,$base,$inp),"xmm0");	# write output
885	&lea	($inp,&DWP(16,$inp));
886	&sub	($out,16);
887	&jnc	(&label("cbc_enc_loop"));
888	&jmp	(&label("cbc_done"));
889
890&set_label("cbc_dec_loop",16);
891	&movdqu	("xmm0",&QWP(0,$inp));		# load input
892	&movdqa	(&QWP(16,"esp"),"xmm1");	# save IV
893	&movdqa	(&QWP(32,"esp"),"xmm0");	# save future IV
894	&call	("_vpaes_decrypt_core");
895	&mov	($base,&DWP(0,"esp"));		# restore out
896	&mov	($key,&DWP(4,"esp"));		# restore key
897	&pxor	("xmm0",&QWP(16,"esp"));	# out^=iv
898	&movdqa	("xmm1",&QWP(32,"esp"));	# load next IV
899	&movdqu	(&QWP(0,$base,$inp),"xmm0");	# write output
900	&lea	($inp,&DWP(16,$inp));
901	&sub	($out,16);
902	&jnc	(&label("cbc_dec_loop"));
903
904&set_label("cbc_done");
905	&mov	($base,&DWP(8,"esp"));		# restore ivp
906	&mov	("esp",&DWP(48,"esp"));
907	&movdqu	(&QWP(0,$base),"xmm1");		# write IV
908&set_label("cbc_abort");
909&function_end("${PREFIX}_cbc_encrypt");
910
911&asm_finish();
912