• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1#! /usr/bin/env perl
2# Copyright 2011-2016 The OpenSSL Project Authors. All Rights Reserved.
3#
4# Licensed under the OpenSSL license (the "License").  You may not use
5# this file except in compliance with the License.  You can obtain a copy
6# in the file LICENSE in the source distribution or at
7# https://www.openssl.org/source/license.html
8
9
10######################################################################
11## Constant-time SSSE3 AES core implementation.
12## version 0.1
13##
14## By Mike Hamburg (Stanford University), 2009
15## Public domain.
16##
17## For details see http://shiftleft.org/papers/vector_aes/ and
18## http://crypto.stanford.edu/vpaes/.
19
20######################################################################
21# September 2011.
22#
23# Interface to OpenSSL as "almost" drop-in replacement for
24# aes-x86_64.pl. "Almost" refers to the fact that AES_cbc_encrypt
25# doesn't handle partial vectors (doesn't have to if called from
26# EVP only). "Drop-in" implies that this module doesn't share key
27# schedule structure with the original nor does it make assumption
28# about its alignment...
29#
30# Performance summary. aes-x86_64.pl column lists large-block CBC
31# encrypt/decrypt/with-hyper-threading-off(*) results in cycles per
32# byte processed with 128-bit key, and vpaes-x86_64.pl column -
33# [also large-block CBC] encrypt/decrypt.
34#
35#		aes-x86_64.pl		vpaes-x86_64.pl
36#
37# Core 2(**)	29.6/41.1/14.3		21.9/25.2(***)
38# Nehalem	29.6/40.3/14.6		10.0/11.8
39# Atom		57.3/74.2/32.1		60.9/77.2(***)
40# Silvermont	52.7/64.0/19.5		48.8/60.8(***)
41# Goldmont	38.9/49.0/17.8		10.6/12.6
42#
43# (*)	"Hyper-threading" in the context refers rather to cache shared
44#	among multiple cores, than to specifically Intel HTT. As vast
45#	majority of contemporary cores share cache, slower code path
46#	is common place. In other words "with-hyper-threading-off"
47#	results are presented mostly for reference purposes.
48#
49# (**)	"Core 2" refers to initial 65nm design, a.k.a. Conroe.
50#
51# (***)	Less impressive improvement on Core 2 and Atom is due to slow
52#	pshufb,	yet it's respectable +36%/62% improvement on Core 2
53#	(as implied, over "hyper-threading-safe" code path).
54#
55#						<appro@openssl.org>
56
57$flavour = shift;
58$output  = shift;
59if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
60
61$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
62
63$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
64( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
65( $xlate="${dir}../../../perlasm/x86_64-xlate.pl" and -f $xlate) or
66die "can't locate x86_64-xlate.pl";
67
68open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\"";
69*STDOUT=*OUT;
70
71$PREFIX="vpaes";
72
73$code.=<<___;
74.text
75
76##
77##  _aes_encrypt_core
78##
79##  AES-encrypt %xmm0.
80##
81##  Inputs:
82##     %xmm0 = input
83##     %xmm9-%xmm15 as in _vpaes_preheat
84##    (%rdx) = scheduled keys
85##
86##  Output in %xmm0
87##  Clobbers  %xmm1-%xmm5, %r9, %r10, %r11, %rax
88##  Preserves %xmm6 - %xmm8 so you get some local vectors
89##
90##
91.type	_vpaes_encrypt_core,\@abi-omnipotent
92.align 16
93_vpaes_encrypt_core:
94.cfi_startproc
95	mov	%rdx,	%r9
96	mov	\$16,	%r11
97	mov	240(%rdx),%eax
98	movdqa	%xmm9,	%xmm1
99	movdqa	.Lk_ipt(%rip), %xmm2	# iptlo
100	pandn	%xmm0,	%xmm1
101	movdqu	(%r9),	%xmm5		# round0 key
102	psrld	\$4,	%xmm1
103	pand	%xmm9,	%xmm0
104	pshufb	%xmm0,	%xmm2
105	movdqa	.Lk_ipt+16(%rip), %xmm0	# ipthi
106	pshufb	%xmm1,	%xmm0
107	pxor	%xmm5,	%xmm2
108	add	\$16,	%r9
109	pxor	%xmm2,	%xmm0
110	lea	.Lk_mc_backward(%rip),%r10
111	jmp	.Lenc_entry
112
113.align 16
114.Lenc_loop:
115	# middle of middle round
116	movdqa  %xmm13,	%xmm4	# 4 : sb1u
117	movdqa  %xmm12,	%xmm0	# 0 : sb1t
118	pshufb  %xmm2,	%xmm4	# 4 = sb1u
119	pshufb  %xmm3,	%xmm0	# 0 = sb1t
120	pxor	%xmm5,	%xmm4	# 4 = sb1u + k
121	movdqa  %xmm15,	%xmm5	# 4 : sb2u
122	pxor	%xmm4,	%xmm0	# 0 = A
123	movdqa	-0x40(%r11,%r10), %xmm1		# .Lk_mc_forward[]
124	pshufb	%xmm2,	%xmm5	# 4 = sb2u
125	movdqa	(%r11,%r10), %xmm4		# .Lk_mc_backward[]
126	movdqa	%xmm14, %xmm2	# 2 : sb2t
127	pshufb	%xmm3,  %xmm2	# 2 = sb2t
128	movdqa	%xmm0,  %xmm3	# 3 = A
129	pxor	%xmm5,	%xmm2	# 2 = 2A
130	pshufb  %xmm1,  %xmm0	# 0 = B
131	add	\$16,	%r9	# next key
132	pxor	%xmm2,  %xmm0	# 0 = 2A+B
133	pshufb	%xmm4,	%xmm3	# 3 = D
134	add	\$16,	%r11	# next mc
135	pxor	%xmm0,	%xmm3	# 3 = 2A+B+D
136	pshufb  %xmm1,	%xmm0	# 0 = 2B+C
137	and	\$0x30,	%r11	# ... mod 4
138	sub	\$1,%rax	# nr--
139	pxor	%xmm3,	%xmm0	# 0 = 2A+3B+C+D
140
141.Lenc_entry:
142	# top of round
143	movdqa  %xmm9, 	%xmm1	# 1 : i
144	movdqa	%xmm11, %xmm5	# 2 : a/k
145	pandn	%xmm0, 	%xmm1	# 1 = i<<4
146	psrld	\$4,   	%xmm1   # 1 = i
147	pand	%xmm9, 	%xmm0   # 0 = k
148	pshufb  %xmm0,  %xmm5	# 2 = a/k
149	movdqa	%xmm10,	%xmm3  	# 3 : 1/i
150	pxor	%xmm1,	%xmm0	# 0 = j
151	pshufb  %xmm1, 	%xmm3  	# 3 = 1/i
152	movdqa	%xmm10,	%xmm4  	# 4 : 1/j
153	pxor	%xmm5, 	%xmm3  	# 3 = iak = 1/i + a/k
154	pshufb	%xmm0, 	%xmm4  	# 4 = 1/j
155	movdqa	%xmm10,	%xmm2  	# 2 : 1/iak
156	pxor	%xmm5, 	%xmm4  	# 4 = jak = 1/j + a/k
157	pshufb  %xmm3,	%xmm2  	# 2 = 1/iak
158	movdqa	%xmm10, %xmm3   # 3 : 1/jak
159	pxor	%xmm0, 	%xmm2  	# 2 = io
160	pshufb  %xmm4,  %xmm3   # 3 = 1/jak
161	movdqu	(%r9),	%xmm5
162	pxor	%xmm1,  %xmm3   # 3 = jo
163	jnz	.Lenc_loop
164
165	# middle of last round
166	movdqa	-0x60(%r10), %xmm4	# 3 : sbou	.Lk_sbo
167	movdqa	-0x50(%r10), %xmm0	# 0 : sbot	.Lk_sbo+16
168	pshufb  %xmm2,  %xmm4	# 4 = sbou
169	pxor	%xmm5,  %xmm4	# 4 = sb1u + k
170	pshufb  %xmm3,	%xmm0	# 0 = sb1t
171	movdqa	0x40(%r11,%r10), %xmm1		# .Lk_sr[]
172	pxor	%xmm4,	%xmm0	# 0 = A
173	pshufb	%xmm1,	%xmm0
174	ret
175.cfi_endproc
176.size	_vpaes_encrypt_core,.-_vpaes_encrypt_core
177
178##
179##  _aes_encrypt_core_2x
180##
181##  AES-encrypt %xmm0 and %xmm6 in parallel.
182##
183##  Inputs:
184##     %xmm0 and %xmm6 = input
185##     %xmm12-%xmm15 as in _vpaes_preheat
186##    (%rdx) = scheduled keys
187##
188##  Output in %xmm0 and %xmm6
189##  Clobbers  %xmm1-%xmm5, %xmm7-%xmm11, %r9, %r10, %r11, %rax
190##  Preserves %xmm14 and %xmm15
191##
192##  This function stitches two parallel instances of _vpaes_encrypt_core. x86_64
193##  provides 16 XMM registers. _vpaes_encrypt_core computes over six registers
194##  (%xmm0-%xmm5) and additionally uses seven registers with preloaded constants
195##  from _vpaes_preheat (%xmm9-%xmm15). This does not quite fit two instances,
196##  so we spill some of %xmm9 through %xmm15 back to memory. We keep %xmm9 and
197##  %xmm10 in registers as these values are used several times in a row. The
198##  remainder are read once per round and are spilled to memory. This leaves two
199##  registers preserved for the caller.
200##
201##  Thus, of the two _vpaes_encrypt_core instances, the first uses (%xmm0-%xmm5)
202##  as before. The second uses %xmm6-%xmm8,%xmm11-%xmm13. (Add 6 to %xmm2 and
203##  below. Add 8 to %xmm3 and up.) Instructions in the second instance are
204##  indented by one space.
205##
206##
207.type	_vpaes_encrypt_core_2x,\@abi-omnipotent
208.align 16
209_vpaes_encrypt_core_2x:
210.cfi_startproc
211	mov	%rdx,	%r9
212	mov	\$16,	%r11
213	mov	240(%rdx),%eax
214	movdqa	%xmm9,	%xmm1
215	 movdqa	%xmm9,	%xmm7
216	movdqa	.Lk_ipt(%rip), %xmm2	# iptlo
217	 movdqa	%xmm2,	%xmm8
218	pandn	%xmm0,	%xmm1
219	 pandn	%xmm6,	%xmm7
220	movdqu	(%r9),	%xmm5		# round0 key
221	 # Also use %xmm5 in the second instance.
222	psrld	\$4,	%xmm1
223	 psrld	\$4,	%xmm7
224	pand	%xmm9,	%xmm0
225	 pand	%xmm9,	%xmm6
226	pshufb	%xmm0,	%xmm2
227	 pshufb	%xmm6,	%xmm8
228	movdqa	.Lk_ipt+16(%rip), %xmm0	# ipthi
229	 movdqa	%xmm0,	%xmm6
230	pshufb	%xmm1,	%xmm0
231	 pshufb	%xmm7,	%xmm6
232	pxor	%xmm5,	%xmm2
233	 pxor	%xmm5,	%xmm8
234	add	\$16,	%r9
235	pxor	%xmm2,	%xmm0
236	 pxor	%xmm8,	%xmm6
237	lea	.Lk_mc_backward(%rip),%r10
238	jmp	.Lenc2x_entry
239
240.align 16
241.Lenc2x_loop:
242	# middle of middle round
243	movdqa  .Lk_sb1(%rip),	%xmm4		# 4 : sb1u
244	movdqa  .Lk_sb1+16(%rip),%xmm0		# 0 : sb1t
245	 movdqa	%xmm4,	%xmm12
246	 movdqa	%xmm0,	%xmm6
247	pshufb  %xmm2,	%xmm4			# 4 = sb1u
248	 pshufb	%xmm8,	%xmm12
249	pshufb  %xmm3,	%xmm0			# 0 = sb1t
250	 pshufb	%xmm11,	%xmm6
251	pxor	%xmm5,	%xmm4			# 4 = sb1u + k
252	 pxor	%xmm5,	%xmm12
253	movdqa  .Lk_sb2(%rip),	%xmm5		# 4 : sb2u
254	 movdqa	%xmm5,	%xmm13
255	pxor	%xmm4,	%xmm0			# 0 = A
256	 pxor	%xmm12,	%xmm6
257	movdqa	-0x40(%r11,%r10), %xmm1		# .Lk_mc_forward[]
258	 # Also use %xmm1 in the second instance.
259	pshufb	%xmm2,	%xmm5			# 4 = sb2u
260	 pshufb	%xmm8,	%xmm13
261	movdqa	(%r11,%r10), %xmm4		# .Lk_mc_backward[]
262	 # Also use %xmm4 in the second instance.
263	movdqa	.Lk_sb2+16(%rip), %xmm2		# 2 : sb2t
264	 movdqa	%xmm2,	%xmm8
265	pshufb	%xmm3,  %xmm2			# 2 = sb2t
266	 pshufb	%xmm11,	%xmm8
267	movdqa	%xmm0,  %xmm3			# 3 = A
268	 movdqa	%xmm6,	%xmm11
269	pxor	%xmm5,	%xmm2			# 2 = 2A
270	 pxor	%xmm13,	%xmm8
271	pshufb  %xmm1,  %xmm0			# 0 = B
272	 pshufb	%xmm1,	%xmm6
273	add	\$16,	%r9			# next key
274	pxor	%xmm2,  %xmm0			# 0 = 2A+B
275	 pxor	%xmm8,	%xmm6
276	pshufb	%xmm4,	%xmm3			# 3 = D
277	 pshufb	%xmm4,	%xmm11
278	add	\$16,	%r11			# next mc
279	pxor	%xmm0,	%xmm3			# 3 = 2A+B+D
280	 pxor	%xmm6,	%xmm11
281	pshufb  %xmm1,	%xmm0			# 0 = 2B+C
282	 pshufb	%xmm1,	%xmm6
283	and	\$0x30,	%r11			# ... mod 4
284	sub	\$1,%rax			# nr--
285	pxor	%xmm3,	%xmm0			# 0 = 2A+3B+C+D
286	 pxor	%xmm11,	%xmm6
287
288.Lenc2x_entry:
289	# top of round
290	movdqa  %xmm9, 	%xmm1	# 1 : i
291	 movdqa	%xmm9,	%xmm7
292	movdqa	.Lk_inv+16(%rip), %xmm5	# 2 : a/k
293	 movdqa	%xmm5,	%xmm13
294	pandn	%xmm0, 	%xmm1	# 1 = i<<4
295	 pandn	%xmm6,	%xmm7
296	psrld	\$4,   	%xmm1   # 1 = i
297	 psrld	\$4,	%xmm7
298	pand	%xmm9, 	%xmm0   # 0 = k
299	 pand	%xmm9,	%xmm6
300	pshufb  %xmm0,  %xmm5	# 2 = a/k
301	 pshufb	%xmm6,	%xmm13
302	movdqa	%xmm10,	%xmm3  	# 3 : 1/i
303	 movdqa	%xmm10,	%xmm11
304	pxor	%xmm1,	%xmm0	# 0 = j
305	 pxor	%xmm7,	%xmm6
306	pshufb  %xmm1, 	%xmm3  	# 3 = 1/i
307	 pshufb	%xmm7,	%xmm11
308	movdqa	%xmm10,	%xmm4  	# 4 : 1/j
309	 movdqa	%xmm10,	%xmm12
310	pxor	%xmm5, 	%xmm3  	# 3 = iak = 1/i + a/k
311	 pxor	%xmm13,	%xmm11
312	pshufb	%xmm0, 	%xmm4  	# 4 = 1/j
313	 pshufb	%xmm6,	%xmm12
314	movdqa	%xmm10,	%xmm2  	# 2 : 1/iak
315	 movdqa	%xmm10,	%xmm8
316	pxor	%xmm5, 	%xmm4  	# 4 = jak = 1/j + a/k
317	 pxor	%xmm13,	%xmm12
318	pshufb  %xmm3,	%xmm2  	# 2 = 1/iak
319	 pshufb	%xmm11,	%xmm8
320	movdqa	%xmm10, %xmm3   # 3 : 1/jak
321	 movdqa	%xmm10,	%xmm11
322	pxor	%xmm0, 	%xmm2  	# 2 = io
323	 pxor	%xmm6,	%xmm8
324	pshufb  %xmm4,  %xmm3   # 3 = 1/jak
325	 pshufb	%xmm12,	%xmm11
326	movdqu	(%r9),	%xmm5
327	 # Also use %xmm5 in the second instance.
328	pxor	%xmm1,  %xmm3   # 3 = jo
329	 pxor	%xmm7,	%xmm11
330	jnz	.Lenc2x_loop
331
332	# middle of last round
333	movdqa	-0x60(%r10), %xmm4	# 3 : sbou	.Lk_sbo
334	movdqa	-0x50(%r10), %xmm0	# 0 : sbot	.Lk_sbo+16
335	 movdqa	%xmm4,	%xmm12
336	 movdqa	%xmm0,	%xmm6
337	pshufb  %xmm2,  %xmm4	# 4 = sbou
338	 pshufb	%xmm8,	%xmm12
339	pxor	%xmm5,  %xmm4	# 4 = sb1u + k
340	 pxor	%xmm5,	%xmm12
341	pshufb  %xmm3,	%xmm0	# 0 = sb1t
342	 pshufb	%xmm11,	%xmm6
343	movdqa	0x40(%r11,%r10), %xmm1		# .Lk_sr[]
344	 # Also use %xmm1 in the second instance.
345	pxor	%xmm4,	%xmm0	# 0 = A
346	 pxor	%xmm12,	%xmm6
347	pshufb	%xmm1,	%xmm0
348	 pshufb	%xmm1,	%xmm6
349	ret
350.cfi_endproc
351.size	_vpaes_encrypt_core_2x,.-_vpaes_encrypt_core_2x
352
353##
354##  Decryption core
355##
356##  Same API as encryption core.
357##
358.type	_vpaes_decrypt_core,\@abi-omnipotent
359.align	16
360_vpaes_decrypt_core:
361.cfi_startproc
362	mov	%rdx,	%r9		# load key
363	mov	240(%rdx),%eax
364	movdqa	%xmm9,	%xmm1
365	movdqa	.Lk_dipt(%rip), %xmm2	# iptlo
366	pandn	%xmm0,	%xmm1
367	mov	%rax,	%r11
368	psrld	\$4,	%xmm1
369	movdqu	(%r9),	%xmm5		# round0 key
370	shl	\$4,	%r11
371	pand	%xmm9,	%xmm0
372	pshufb	%xmm0,	%xmm2
373	movdqa	.Lk_dipt+16(%rip), %xmm0 # ipthi
374	xor	\$0x30,	%r11
375	lea	.Lk_dsbd(%rip),%r10
376	pshufb	%xmm1,	%xmm0
377	and	\$0x30,	%r11
378	pxor	%xmm5,	%xmm2
379	movdqa	.Lk_mc_forward+48(%rip), %xmm5
380	pxor	%xmm2,	%xmm0
381	add	\$16,	%r9
382	add	%r10,	%r11
383	jmp	.Ldec_entry
384
385.align 16
386.Ldec_loop:
387##
388##  Inverse mix columns
389##
390	movdqa  -0x20(%r10),%xmm4	# 4 : sb9u
391	movdqa  -0x10(%r10),%xmm1	# 0 : sb9t
392	pshufb	%xmm2,	%xmm4		# 4 = sb9u
393	pshufb	%xmm3,	%xmm1		# 0 = sb9t
394	pxor	%xmm4,	%xmm0
395	movdqa  0x00(%r10),%xmm4	# 4 : sbdu
396	pxor	%xmm1,	%xmm0		# 0 = ch
397	movdqa  0x10(%r10),%xmm1	# 0 : sbdt
398
399	pshufb	%xmm2,	%xmm4		# 4 = sbdu
400	pshufb	%xmm5,	%xmm0		# MC ch
401	pshufb	%xmm3,	%xmm1		# 0 = sbdt
402	pxor	%xmm4,	%xmm0		# 4 = ch
403	movdqa  0x20(%r10),%xmm4	# 4 : sbbu
404	pxor	%xmm1,	%xmm0		# 0 = ch
405	movdqa  0x30(%r10),%xmm1	# 0 : sbbt
406
407	pshufb	%xmm2,	%xmm4		# 4 = sbbu
408	pshufb	%xmm5,	%xmm0		# MC ch
409	pshufb	%xmm3,	%xmm1		# 0 = sbbt
410	pxor	%xmm4,	%xmm0		# 4 = ch
411	movdqa  0x40(%r10),%xmm4	# 4 : sbeu
412	pxor	%xmm1,	%xmm0		# 0 = ch
413	movdqa  0x50(%r10),%xmm1	# 0 : sbet
414
415	pshufb	%xmm2,	%xmm4		# 4 = sbeu
416	pshufb	%xmm5,	%xmm0		# MC ch
417	pshufb	%xmm3,	%xmm1		# 0 = sbet
418	pxor	%xmm4,	%xmm0		# 4 = ch
419	add	\$16, %r9		# next round key
420	palignr	\$12,	%xmm5,	%xmm5
421	pxor	%xmm1,	%xmm0		# 0 = ch
422	sub	\$1,%rax		# nr--
423
424.Ldec_entry:
425	# top of round
426	movdqa  %xmm9, 	%xmm1	# 1 : i
427	pandn	%xmm0, 	%xmm1	# 1 = i<<4
428	movdqa	%xmm11, %xmm2	# 2 : a/k
429	psrld	\$4,    %xmm1	# 1 = i
430	pand	%xmm9, 	%xmm0	# 0 = k
431	pshufb  %xmm0,  %xmm2	# 2 = a/k
432	movdqa	%xmm10,	%xmm3	# 3 : 1/i
433	pxor	%xmm1,	%xmm0	# 0 = j
434	pshufb  %xmm1, 	%xmm3	# 3 = 1/i
435	movdqa	%xmm10,	%xmm4	# 4 : 1/j
436	pxor	%xmm2, 	%xmm3	# 3 = iak = 1/i + a/k
437	pshufb	%xmm0, 	%xmm4	# 4 = 1/j
438	pxor	%xmm2, 	%xmm4	# 4 = jak = 1/j + a/k
439	movdqa	%xmm10,	%xmm2	# 2 : 1/iak
440	pshufb  %xmm3,	%xmm2	# 2 = 1/iak
441	movdqa	%xmm10, %xmm3	# 3 : 1/jak
442	pxor	%xmm0, 	%xmm2	# 2 = io
443	pshufb  %xmm4,  %xmm3	# 3 = 1/jak
444	movdqu	(%r9),	%xmm0
445	pxor	%xmm1,  %xmm3	# 3 = jo
446	jnz	.Ldec_loop
447
448	# middle of last round
449	movdqa	0x60(%r10), %xmm4	# 3 : sbou
450	pshufb  %xmm2,  %xmm4	# 4 = sbou
451	pxor	%xmm0,  %xmm4	# 4 = sb1u + k
452	movdqa	0x70(%r10), %xmm0	# 0 : sbot
453	movdqa	-0x160(%r11), %xmm2	# .Lk_sr-.Lk_dsbd=-0x160
454	pshufb  %xmm3,	%xmm0	# 0 = sb1t
455	pxor	%xmm4,	%xmm0	# 0 = A
456	pshufb	%xmm2,	%xmm0
457	ret
458.cfi_endproc
459.size	_vpaes_decrypt_core,.-_vpaes_decrypt_core
460
461########################################################
462##                                                    ##
463##                  AES key schedule                  ##
464##                                                    ##
465########################################################
466.type	_vpaes_schedule_core,\@abi-omnipotent
467.align	16
468_vpaes_schedule_core:
469.cfi_startproc
470	# rdi = key
471	# rsi = size in bits
472	# rdx = buffer
473	# rcx = direction.  0=encrypt, 1=decrypt
474
475	call	_vpaes_preheat		# load the tables
476	movdqa	.Lk_rcon(%rip), %xmm8	# load rcon
477	movdqu	(%rdi),	%xmm0		# load key (unaligned)
478
479	# input transform
480	movdqa	%xmm0,	%xmm3
481	lea	.Lk_ipt(%rip), %r11
482	call	_vpaes_schedule_transform
483	movdqa	%xmm0,	%xmm7
484
485	lea	.Lk_sr(%rip),%r10
486	test	%rcx,	%rcx
487	jnz	.Lschedule_am_decrypting
488
489	# encrypting, output zeroth round key after transform
490	movdqu	%xmm0,	(%rdx)
491	jmp	.Lschedule_go
492
493.Lschedule_am_decrypting:
494	# decrypting, output zeroth round key after shiftrows
495	movdqa	(%r8,%r10),%xmm1
496	pshufb  %xmm1,	%xmm3
497	movdqu	%xmm3,	(%rdx)
498	xor	\$0x30, %r8
499
500.Lschedule_go:
501	cmp	\$192,	%esi
502	ja	.Lschedule_256
503	je	.Lschedule_192
504	# 128: fall though
505
506##
507##  .schedule_128
508##
509##  128-bit specific part of key schedule.
510##
511##  This schedule is really simple, because all its parts
512##  are accomplished by the subroutines.
513##
514.Lschedule_128:
515	mov	\$10, %esi
516
517.Loop_schedule_128:
518	call 	_vpaes_schedule_round
519	dec	%rsi
520	jz 	.Lschedule_mangle_last
521	call	_vpaes_schedule_mangle	# write output
522	jmp 	.Loop_schedule_128
523
524##
525##  .aes_schedule_192
526##
527##  192-bit specific part of key schedule.
528##
529##  The main body of this schedule is the same as the 128-bit
530##  schedule, but with more smearing.  The long, high side is
531##  stored in %xmm7 as before, and the short, low side is in
532##  the high bits of %xmm6.
533##
534##  This schedule is somewhat nastier, however, because each
535##  round produces 192 bits of key material, or 1.5 round keys.
536##  Therefore, on each cycle we do 2 rounds and produce 3 round
537##  keys.
538##
539.align	16
540.Lschedule_192:
541	movdqu	8(%rdi),%xmm0		# load key part 2 (very unaligned)
542	call	_vpaes_schedule_transform	# input transform
543	movdqa	%xmm0,	%xmm6		# save short part
544	pxor	%xmm4,	%xmm4		# clear 4
545	movhlps	%xmm4,	%xmm6		# clobber low side with zeros
546	mov	\$4,	%esi
547
548.Loop_schedule_192:
549	call	_vpaes_schedule_round
550	palignr	\$8,%xmm6,%xmm0
551	call	_vpaes_schedule_mangle	# save key n
552	call	_vpaes_schedule_192_smear
553	call	_vpaes_schedule_mangle	# save key n+1
554	call	_vpaes_schedule_round
555	dec	%rsi
556	jz 	.Lschedule_mangle_last
557	call	_vpaes_schedule_mangle	# save key n+2
558	call	_vpaes_schedule_192_smear
559	jmp	.Loop_schedule_192
560
561##
562##  .aes_schedule_256
563##
564##  256-bit specific part of key schedule.
565##
566##  The structure here is very similar to the 128-bit
567##  schedule, but with an additional "low side" in
568##  %xmm6.  The low side's rounds are the same as the
569##  high side's, except no rcon and no rotation.
570##
571.align	16
572.Lschedule_256:
573	movdqu	16(%rdi),%xmm0		# load key part 2 (unaligned)
574	call	_vpaes_schedule_transform	# input transform
575	mov	\$7, %esi
576
577.Loop_schedule_256:
578	call	_vpaes_schedule_mangle	# output low result
579	movdqa	%xmm0,	%xmm6		# save cur_lo in xmm6
580
581	# high round
582	call	_vpaes_schedule_round
583	dec	%rsi
584	jz 	.Lschedule_mangle_last
585	call	_vpaes_schedule_mangle
586
587	# low round. swap xmm7 and xmm6
588	pshufd	\$0xFF,	%xmm0,	%xmm0
589	movdqa	%xmm7,	%xmm5
590	movdqa	%xmm6,	%xmm7
591	call	_vpaes_schedule_low_round
592	movdqa	%xmm5,	%xmm7
593
594	jmp	.Loop_schedule_256
595
596
597##
598##  .aes_schedule_mangle_last
599##
600##  Mangler for last round of key schedule
601##  Mangles %xmm0
602##    when encrypting, outputs out(%xmm0) ^ 63
603##    when decrypting, outputs unskew(%xmm0)
604##
605##  Always called right before return... jumps to cleanup and exits
606##
607.align	16
608.Lschedule_mangle_last:
609	# schedule last round key from xmm0
610	lea	.Lk_deskew(%rip),%r11	# prepare to deskew
611	test	%rcx, 	%rcx
612	jnz	.Lschedule_mangle_last_dec
613
614	# encrypting
615	movdqa	(%r8,%r10),%xmm1
616	pshufb	%xmm1,	%xmm0		# output permute
617	lea	.Lk_opt(%rip),	%r11	# prepare to output transform
618	add	\$32,	%rdx
619
620.Lschedule_mangle_last_dec:
621	add	\$-16,	%rdx
622	pxor	.Lk_s63(%rip),	%xmm0
623	call	_vpaes_schedule_transform # output transform
624	movdqu	%xmm0,	(%rdx)		# save last key
625
626	# cleanup
627	pxor	%xmm0,  %xmm0
628	pxor	%xmm1,  %xmm1
629	pxor	%xmm2,  %xmm2
630	pxor	%xmm3,  %xmm3
631	pxor	%xmm4,  %xmm4
632	pxor	%xmm5,  %xmm5
633	pxor	%xmm6,  %xmm6
634	pxor	%xmm7,  %xmm7
635	ret
636.cfi_endproc
637.size	_vpaes_schedule_core,.-_vpaes_schedule_core
638
639##
640##  .aes_schedule_192_smear
641##
642##  Smear the short, low side in the 192-bit key schedule.
643##
644##  Inputs:
645##    %xmm7: high side, b  a  x  y
646##    %xmm6:  low side, d  c  0  0
647##    %xmm13: 0
648##
649##  Outputs:
650##    %xmm6: b+c+d  b+c  0  0
651##    %xmm0: b+c+d  b+c  b  a
652##
653.type	_vpaes_schedule_192_smear,\@abi-omnipotent
654.align	16
655_vpaes_schedule_192_smear:
656.cfi_startproc
657	pshufd	\$0x80,	%xmm6,	%xmm1	# d c 0 0 -> c 0 0 0
658	pshufd	\$0xFE,	%xmm7,	%xmm0	# b a _ _ -> b b b a
659	pxor	%xmm1,	%xmm6		# -> c+d c 0 0
660	pxor	%xmm1,	%xmm1
661	pxor	%xmm0,	%xmm6		# -> b+c+d b+c b a
662	movdqa	%xmm6,	%xmm0
663	movhlps	%xmm1,	%xmm6		# clobber low side with zeros
664	ret
665.cfi_endproc
666.size	_vpaes_schedule_192_smear,.-_vpaes_schedule_192_smear
667
668##
669##  .aes_schedule_round
670##
671##  Runs one main round of the key schedule on %xmm0, %xmm7
672##
673##  Specifically, runs subbytes on the high dword of %xmm0
674##  then rotates it by one byte and xors into the low dword of
675##  %xmm7.
676##
677##  Adds rcon from low byte of %xmm8, then rotates %xmm8 for
678##  next rcon.
679##
680##  Smears the dwords of %xmm7 by xoring the low into the
681##  second low, result into third, result into highest.
682##
683##  Returns results in %xmm7 = %xmm0.
684##  Clobbers %xmm1-%xmm4, %r11.
685##
686.type	_vpaes_schedule_round,\@abi-omnipotent
687.align	16
688_vpaes_schedule_round:
689.cfi_startproc
690	# extract rcon from xmm8
691	pxor	%xmm1,	%xmm1
692	palignr	\$15,	%xmm8,	%xmm1
693	palignr	\$15,	%xmm8,	%xmm8
694	pxor	%xmm1,	%xmm7
695
696	# rotate
697	pshufd	\$0xFF,	%xmm0,	%xmm0
698	palignr	\$1,	%xmm0,	%xmm0
699
700	# fall through...
701
702	# low round: same as high round, but no rotation and no rcon.
703_vpaes_schedule_low_round:
704	# smear xmm7
705	movdqa	%xmm7,	%xmm1
706	pslldq	\$4,	%xmm7
707	pxor	%xmm1,	%xmm7
708	movdqa	%xmm7,	%xmm1
709	pslldq	\$8,	%xmm7
710	pxor	%xmm1,	%xmm7
711	pxor	.Lk_s63(%rip), %xmm7
712
713	# subbytes
714	movdqa  %xmm9, 	%xmm1
715	pandn	%xmm0, 	%xmm1
716	psrld	\$4,    %xmm1		# 1 = i
717	pand	%xmm9, 	%xmm0		# 0 = k
718	movdqa	%xmm11, %xmm2		# 2 : a/k
719	pshufb  %xmm0,  %xmm2		# 2 = a/k
720	pxor	%xmm1,	%xmm0		# 0 = j
721	movdqa	%xmm10,	%xmm3		# 3 : 1/i
722	pshufb  %xmm1, 	%xmm3		# 3 = 1/i
723	pxor	%xmm2, 	%xmm3		# 3 = iak = 1/i + a/k
724	movdqa	%xmm10,	%xmm4		# 4 : 1/j
725	pshufb	%xmm0, 	%xmm4		# 4 = 1/j
726	pxor	%xmm2, 	%xmm4		# 4 = jak = 1/j + a/k
727	movdqa	%xmm10,	%xmm2		# 2 : 1/iak
728	pshufb  %xmm3,	%xmm2		# 2 = 1/iak
729	pxor	%xmm0, 	%xmm2		# 2 = io
730	movdqa	%xmm10, %xmm3		# 3 : 1/jak
731	pshufb  %xmm4,  %xmm3		# 3 = 1/jak
732	pxor	%xmm1,  %xmm3		# 3 = jo
733	movdqa	%xmm13, %xmm4		# 4 : sbou
734	pshufb  %xmm2,  %xmm4		# 4 = sbou
735	movdqa	%xmm12, %xmm0		# 0 : sbot
736	pshufb  %xmm3,	%xmm0		# 0 = sb1t
737	pxor	%xmm4, 	%xmm0		# 0 = sbox output
738
739	# add in smeared stuff
740	pxor	%xmm7,	%xmm0
741	movdqa	%xmm0,	%xmm7
742	ret
743.cfi_endproc
744.size	_vpaes_schedule_round,.-_vpaes_schedule_round
745
746##
747##  .aes_schedule_transform
748##
749##  Linear-transform %xmm0 according to tables at (%r11)
750##
751##  Requires that %xmm9 = 0x0F0F... as in preheat
752##  Output in %xmm0
753##  Clobbers %xmm1, %xmm2
754##
755.type	_vpaes_schedule_transform,\@abi-omnipotent
756.align	16
757_vpaes_schedule_transform:
758.cfi_startproc
759	movdqa	%xmm9,	%xmm1
760	pandn	%xmm0,	%xmm1
761	psrld	\$4,	%xmm1
762	pand	%xmm9,	%xmm0
763	movdqa	(%r11), %xmm2 	# lo
764	pshufb	%xmm0,	%xmm2
765	movdqa	16(%r11), %xmm0 # hi
766	pshufb	%xmm1,	%xmm0
767	pxor	%xmm2,	%xmm0
768	ret
769.cfi_endproc
770.size	_vpaes_schedule_transform,.-_vpaes_schedule_transform
771
772##
773##  .aes_schedule_mangle
774##
775##  Mangle xmm0 from (basis-transformed) standard version
776##  to our version.
777##
778##  On encrypt,
779##    xor with 0x63
780##    multiply by circulant 0,1,1,1
781##    apply shiftrows transform
782##
783##  On decrypt,
784##    xor with 0x63
785##    multiply by "inverse mixcolumns" circulant E,B,D,9
786##    deskew
787##    apply shiftrows transform
788##
789##
790##  Writes out to (%rdx), and increments or decrements it
791##  Keeps track of round number mod 4 in %r8
792##  Preserves xmm0
793##  Clobbers xmm1-xmm5
794##
795.type	_vpaes_schedule_mangle,\@abi-omnipotent
796.align	16
797_vpaes_schedule_mangle:
798.cfi_startproc
799	movdqa	%xmm0,	%xmm4	# save xmm0 for later
800	movdqa	.Lk_mc_forward(%rip),%xmm5
801	test	%rcx, 	%rcx
802	jnz	.Lschedule_mangle_dec
803
804	# encrypting
805	add	\$16,	%rdx
806	pxor	.Lk_s63(%rip),%xmm4
807	pshufb	%xmm5,	%xmm4
808	movdqa	%xmm4,	%xmm3
809	pshufb	%xmm5,	%xmm4
810	pxor	%xmm4,	%xmm3
811	pshufb	%xmm5,	%xmm4
812	pxor	%xmm4,	%xmm3
813
814	jmp	.Lschedule_mangle_both
815.align	16
816.Lschedule_mangle_dec:
817	# inverse mix columns
818	lea	.Lk_dksd(%rip),%r11
819	movdqa	%xmm9,	%xmm1
820	pandn	%xmm4,	%xmm1
821	psrld	\$4,	%xmm1	# 1 = hi
822	pand	%xmm9,	%xmm4	# 4 = lo
823
824	movdqa	0x00(%r11), %xmm2
825	pshufb	%xmm4,	%xmm2
826	movdqa	0x10(%r11), %xmm3
827	pshufb	%xmm1,	%xmm3
828	pxor	%xmm2,	%xmm3
829	pshufb	%xmm5,	%xmm3
830
831	movdqa	0x20(%r11), %xmm2
832	pshufb	%xmm4,	%xmm2
833	pxor	%xmm3,	%xmm2
834	movdqa	0x30(%r11), %xmm3
835	pshufb	%xmm1,	%xmm3
836	pxor	%xmm2,	%xmm3
837	pshufb	%xmm5,	%xmm3
838
839	movdqa	0x40(%r11), %xmm2
840	pshufb	%xmm4,	%xmm2
841	pxor	%xmm3,	%xmm2
842	movdqa	0x50(%r11), %xmm3
843	pshufb	%xmm1,	%xmm3
844	pxor	%xmm2,	%xmm3
845	pshufb	%xmm5,	%xmm3
846
847	movdqa	0x60(%r11), %xmm2
848	pshufb	%xmm4,	%xmm2
849	pxor	%xmm3,	%xmm2
850	movdqa	0x70(%r11), %xmm3
851	pshufb	%xmm1,	%xmm3
852	pxor	%xmm2,	%xmm3
853
854	add	\$-16,	%rdx
855
856.Lschedule_mangle_both:
857	movdqa	(%r8,%r10),%xmm1
858	pshufb	%xmm1,%xmm3
859	add	\$-16,	%r8
860	and	\$0x30,	%r8
861	movdqu	%xmm3,	(%rdx)
862	ret
863.cfi_endproc
864.size	_vpaes_schedule_mangle,.-_vpaes_schedule_mangle
865
866#
867# Interface to OpenSSL
868#
869.globl	${PREFIX}_set_encrypt_key
870.type	${PREFIX}_set_encrypt_key,\@function,3
871.align	16
872${PREFIX}_set_encrypt_key:
873.cfi_startproc
874#ifndef NDEBUG
875#ifndef BORINGSSL_FIPS
876.extern        BORINGSSL_function_hit
877       movb \$1, BORINGSSL_function_hit+5(%rip)
878#endif
879#endif
880
881___
882$code.=<<___ if ($win64);
883	lea	-0xb8(%rsp),%rsp
884	movaps	%xmm6,0x10(%rsp)
885	movaps	%xmm7,0x20(%rsp)
886	movaps	%xmm8,0x30(%rsp)
887	movaps	%xmm9,0x40(%rsp)
888	movaps	%xmm10,0x50(%rsp)
889	movaps	%xmm11,0x60(%rsp)
890	movaps	%xmm12,0x70(%rsp)
891	movaps	%xmm13,0x80(%rsp)
892	movaps	%xmm14,0x90(%rsp)
893	movaps	%xmm15,0xa0(%rsp)
894.Lenc_key_body:
895___
896$code.=<<___;
897	mov	%esi,%eax
898	shr	\$5,%eax
899	add	\$5,%eax
900	mov	%eax,240(%rdx)	# AES_KEY->rounds = nbits/32+5;
901
902	mov	\$0,%ecx
903	mov	\$0x30,%r8d
904	call	_vpaes_schedule_core
905___
906$code.=<<___ if ($win64);
907	movaps	0x10(%rsp),%xmm6
908	movaps	0x20(%rsp),%xmm7
909	movaps	0x30(%rsp),%xmm8
910	movaps	0x40(%rsp),%xmm9
911	movaps	0x50(%rsp),%xmm10
912	movaps	0x60(%rsp),%xmm11
913	movaps	0x70(%rsp),%xmm12
914	movaps	0x80(%rsp),%xmm13
915	movaps	0x90(%rsp),%xmm14
916	movaps	0xa0(%rsp),%xmm15
917	lea	0xb8(%rsp),%rsp
918.Lenc_key_epilogue:
919___
920$code.=<<___;
921	xor	%eax,%eax
922	ret
923.cfi_endproc
924.size	${PREFIX}_set_encrypt_key,.-${PREFIX}_set_encrypt_key
925
926.globl	${PREFIX}_set_decrypt_key
927.type	${PREFIX}_set_decrypt_key,\@function,3
928.align	16
929${PREFIX}_set_decrypt_key:
930.cfi_startproc
931___
932$code.=<<___ if ($win64);
933	lea	-0xb8(%rsp),%rsp
934	movaps	%xmm6,0x10(%rsp)
935	movaps	%xmm7,0x20(%rsp)
936	movaps	%xmm8,0x30(%rsp)
937	movaps	%xmm9,0x40(%rsp)
938	movaps	%xmm10,0x50(%rsp)
939	movaps	%xmm11,0x60(%rsp)
940	movaps	%xmm12,0x70(%rsp)
941	movaps	%xmm13,0x80(%rsp)
942	movaps	%xmm14,0x90(%rsp)
943	movaps	%xmm15,0xa0(%rsp)
944.Ldec_key_body:
945___
946$code.=<<___;
947	mov	%esi,%eax
948	shr	\$5,%eax
949	add	\$5,%eax
950	mov	%eax,240(%rdx)	# AES_KEY->rounds = nbits/32+5;
951	shl	\$4,%eax
952	lea	16(%rdx,%rax),%rdx
953
954	mov	\$1,%ecx
955	mov	%esi,%r8d
956	shr	\$1,%r8d
957	and	\$32,%r8d
958	xor	\$32,%r8d	# nbits==192?0:32
959	call	_vpaes_schedule_core
960___
961$code.=<<___ if ($win64);
962	movaps	0x10(%rsp),%xmm6
963	movaps	0x20(%rsp),%xmm7
964	movaps	0x30(%rsp),%xmm8
965	movaps	0x40(%rsp),%xmm9
966	movaps	0x50(%rsp),%xmm10
967	movaps	0x60(%rsp),%xmm11
968	movaps	0x70(%rsp),%xmm12
969	movaps	0x80(%rsp),%xmm13
970	movaps	0x90(%rsp),%xmm14
971	movaps	0xa0(%rsp),%xmm15
972	lea	0xb8(%rsp),%rsp
973.Ldec_key_epilogue:
974___
975$code.=<<___;
976	xor	%eax,%eax
977	ret
978.cfi_endproc
979.size	${PREFIX}_set_decrypt_key,.-${PREFIX}_set_decrypt_key
980
981.globl	${PREFIX}_encrypt
982.type	${PREFIX}_encrypt,\@function,3
983.align	16
984${PREFIX}_encrypt:
985.cfi_startproc
986#ifndef NDEBUG
987#ifndef BORINGSSL_FIPS
988.extern        BORINGSSL_function_hit
989       movb \$1, BORINGSSL_function_hit+4(%rip)
990#endif
991#endif
992___
993$code.=<<___ if ($win64);
994	lea	-0xb8(%rsp),%rsp
995	movaps	%xmm6,0x10(%rsp)
996	movaps	%xmm7,0x20(%rsp)
997	movaps	%xmm8,0x30(%rsp)
998	movaps	%xmm9,0x40(%rsp)
999	movaps	%xmm10,0x50(%rsp)
1000	movaps	%xmm11,0x60(%rsp)
1001	movaps	%xmm12,0x70(%rsp)
1002	movaps	%xmm13,0x80(%rsp)
1003	movaps	%xmm14,0x90(%rsp)
1004	movaps	%xmm15,0xa0(%rsp)
1005.Lenc_body:
1006___
1007$code.=<<___;
1008	movdqu	(%rdi),%xmm0
1009	call	_vpaes_preheat
1010	call	_vpaes_encrypt_core
1011	movdqu	%xmm0,(%rsi)
1012___
1013$code.=<<___ if ($win64);
1014	movaps	0x10(%rsp),%xmm6
1015	movaps	0x20(%rsp),%xmm7
1016	movaps	0x30(%rsp),%xmm8
1017	movaps	0x40(%rsp),%xmm9
1018	movaps	0x50(%rsp),%xmm10
1019	movaps	0x60(%rsp),%xmm11
1020	movaps	0x70(%rsp),%xmm12
1021	movaps	0x80(%rsp),%xmm13
1022	movaps	0x90(%rsp),%xmm14
1023	movaps	0xa0(%rsp),%xmm15
1024	lea	0xb8(%rsp),%rsp
1025.Lenc_epilogue:
1026___
1027$code.=<<___;
1028	ret
1029.cfi_endproc
1030.size	${PREFIX}_encrypt,.-${PREFIX}_encrypt
1031
1032.globl	${PREFIX}_decrypt
1033.type	${PREFIX}_decrypt,\@function,3
1034.align	16
1035${PREFIX}_decrypt:
1036.cfi_startproc
1037___
1038$code.=<<___ if ($win64);
1039	lea	-0xb8(%rsp),%rsp
1040	movaps	%xmm6,0x10(%rsp)
1041	movaps	%xmm7,0x20(%rsp)
1042	movaps	%xmm8,0x30(%rsp)
1043	movaps	%xmm9,0x40(%rsp)
1044	movaps	%xmm10,0x50(%rsp)
1045	movaps	%xmm11,0x60(%rsp)
1046	movaps	%xmm12,0x70(%rsp)
1047	movaps	%xmm13,0x80(%rsp)
1048	movaps	%xmm14,0x90(%rsp)
1049	movaps	%xmm15,0xa0(%rsp)
1050.Ldec_body:
1051___
1052$code.=<<___;
1053	movdqu	(%rdi),%xmm0
1054	call	_vpaes_preheat
1055	call	_vpaes_decrypt_core
1056	movdqu	%xmm0,(%rsi)
1057___
1058$code.=<<___ if ($win64);
1059	movaps	0x10(%rsp),%xmm6
1060	movaps	0x20(%rsp),%xmm7
1061	movaps	0x30(%rsp),%xmm8
1062	movaps	0x40(%rsp),%xmm9
1063	movaps	0x50(%rsp),%xmm10
1064	movaps	0x60(%rsp),%xmm11
1065	movaps	0x70(%rsp),%xmm12
1066	movaps	0x80(%rsp),%xmm13
1067	movaps	0x90(%rsp),%xmm14
1068	movaps	0xa0(%rsp),%xmm15
1069	lea	0xb8(%rsp),%rsp
1070.Ldec_epilogue:
1071___
1072$code.=<<___;
1073	ret
1074.cfi_endproc
1075.size	${PREFIX}_decrypt,.-${PREFIX}_decrypt
1076___
1077{
1078my ($inp,$out,$len,$key,$ivp,$enc)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9");
1079# void AES_cbc_encrypt (const void char *inp, unsigned char *out,
1080#                       size_t length, const AES_KEY *key,
1081#                       unsigned char *ivp,const int enc);
1082$code.=<<___;
1083.globl	${PREFIX}_cbc_encrypt
1084.type	${PREFIX}_cbc_encrypt,\@function,6
1085.align	16
1086${PREFIX}_cbc_encrypt:
1087.cfi_startproc
1088	xchg	$key,$len
1089___
1090($len,$key)=($key,$len);
1091$code.=<<___;
1092	sub	\$16,$len
1093	jc	.Lcbc_abort
1094___
1095$code.=<<___ if ($win64);
1096	lea	-0xb8(%rsp),%rsp
1097	movaps	%xmm6,0x10(%rsp)
1098	movaps	%xmm7,0x20(%rsp)
1099	movaps	%xmm8,0x30(%rsp)
1100	movaps	%xmm9,0x40(%rsp)
1101	movaps	%xmm10,0x50(%rsp)
1102	movaps	%xmm11,0x60(%rsp)
1103	movaps	%xmm12,0x70(%rsp)
1104	movaps	%xmm13,0x80(%rsp)
1105	movaps	%xmm14,0x90(%rsp)
1106	movaps	%xmm15,0xa0(%rsp)
1107.Lcbc_body:
1108___
1109$code.=<<___;
1110	movdqu	($ivp),%xmm6		# load IV
1111	sub	$inp,$out
1112	call	_vpaes_preheat
1113	cmp	\$0,${enc}d
1114	je	.Lcbc_dec_loop
1115	jmp	.Lcbc_enc_loop
1116.align	16
1117.Lcbc_enc_loop:
1118	movdqu	($inp),%xmm0
1119	pxor	%xmm6,%xmm0
1120	call	_vpaes_encrypt_core
1121	movdqa	%xmm0,%xmm6
1122	movdqu	%xmm0,($out,$inp)
1123	lea	16($inp),$inp
1124	sub	\$16,$len
1125	jnc	.Lcbc_enc_loop
1126	jmp	.Lcbc_done
1127.align	16
1128.Lcbc_dec_loop:
1129	movdqu	($inp),%xmm0
1130	movdqa	%xmm0,%xmm7
1131	call	_vpaes_decrypt_core
1132	pxor	%xmm6,%xmm0
1133	movdqa	%xmm7,%xmm6
1134	movdqu	%xmm0,($out,$inp)
1135	lea	16($inp),$inp
1136	sub	\$16,$len
1137	jnc	.Lcbc_dec_loop
1138.Lcbc_done:
1139	movdqu	%xmm6,($ivp)		# save IV
1140___
1141$code.=<<___ if ($win64);
1142	movaps	0x10(%rsp),%xmm6
1143	movaps	0x20(%rsp),%xmm7
1144	movaps	0x30(%rsp),%xmm8
1145	movaps	0x40(%rsp),%xmm9
1146	movaps	0x50(%rsp),%xmm10
1147	movaps	0x60(%rsp),%xmm11
1148	movaps	0x70(%rsp),%xmm12
1149	movaps	0x80(%rsp),%xmm13
1150	movaps	0x90(%rsp),%xmm14
1151	movaps	0xa0(%rsp),%xmm15
1152	lea	0xb8(%rsp),%rsp
1153.Lcbc_epilogue:
1154___
1155$code.=<<___;
1156.Lcbc_abort:
1157	ret
1158.cfi_endproc
1159.size	${PREFIX}_cbc_encrypt,.-${PREFIX}_cbc_encrypt
1160___
1161}
1162{
1163my ($inp,$out,$blocks,$key,$ivp)=("%rdi","%rsi","%rdx","%rcx","%r8");
1164# void vpaes_ctr32_encrypt_blocks(const uint8_t *inp, uint8_t *out,
1165#                                 size_t blocks, const AES_KEY *key,
1166#                                 const uint8_t ivp[16]);
1167$code.=<<___;
1168.globl	${PREFIX}_ctr32_encrypt_blocks
1169.type	${PREFIX}_ctr32_encrypt_blocks,\@function,5
1170.align	16
1171${PREFIX}_ctr32_encrypt_blocks:
1172.cfi_startproc
1173	# _vpaes_encrypt_core and _vpaes_encrypt_core_2x expect the key in %rdx.
1174	xchg	$key, $blocks
1175___
1176($blocks,$key)=($key,$blocks);
1177$code.=<<___;
1178	test	$blocks, $blocks
1179	jz	.Lctr32_abort
1180___
1181$code.=<<___ if ($win64);
1182	lea	-0xb8(%rsp),%rsp
1183	movaps	%xmm6,0x10(%rsp)
1184	movaps	%xmm7,0x20(%rsp)
1185	movaps	%xmm8,0x30(%rsp)
1186	movaps	%xmm9,0x40(%rsp)
1187	movaps	%xmm10,0x50(%rsp)
1188	movaps	%xmm11,0x60(%rsp)
1189	movaps	%xmm12,0x70(%rsp)
1190	movaps	%xmm13,0x80(%rsp)
1191	movaps	%xmm14,0x90(%rsp)
1192	movaps	%xmm15,0xa0(%rsp)
1193.Lctr32_body:
1194___
1195$code.=<<___;
1196	movdqu	($ivp), %xmm0		# Load IV.
1197	movdqa	.Lctr_add_one(%rip), %xmm8
1198	sub	$inp, $out		# This allows only incrementing $inp.
1199	call	_vpaes_preheat
1200	movdqa	%xmm0, %xmm6
1201	pshufb	.Lrev_ctr(%rip), %xmm6
1202
1203	test	\$1, $blocks
1204	jz	.Lctr32_prep_loop
1205
1206	# Handle one block so the remaining block count is even for
1207	# _vpaes_encrypt_core_2x.
1208	movdqu	($inp), %xmm7		# Load input.
1209	call	_vpaes_encrypt_core
1210	pxor	%xmm7, %xmm0
1211	paddd	%xmm8, %xmm6
1212	movdqu	%xmm0, ($out,$inp)
1213	sub	\$1, $blocks
1214	lea	16($inp), $inp
1215	jz	.Lctr32_done
1216
1217.Lctr32_prep_loop:
1218	# _vpaes_encrypt_core_2x leaves only %xmm14 and %xmm15 as spare
1219	# registers. We maintain two byte-swapped counters in them.
1220	movdqa	%xmm6, %xmm14
1221	movdqa	%xmm6, %xmm15
1222	paddd	%xmm8, %xmm15
1223
1224.Lctr32_loop:
1225	movdqa	.Lrev_ctr(%rip), %xmm1	# Set up counters.
1226	movdqa	%xmm14, %xmm0
1227	movdqa	%xmm15, %xmm6
1228	pshufb	%xmm1, %xmm0
1229	pshufb	%xmm1, %xmm6
1230	call	_vpaes_encrypt_core_2x
1231	movdqu	($inp), %xmm1		# Load input.
1232	movdqu	16($inp), %xmm2
1233	movdqa	.Lctr_add_two(%rip), %xmm3
1234	pxor	%xmm1, %xmm0		# XOR input.
1235	pxor	%xmm2, %xmm6
1236	paddd	%xmm3, %xmm14		# Increment counters.
1237	paddd	%xmm3, %xmm15
1238	movdqu	%xmm0, ($out,$inp)	# Write output.
1239	movdqu	%xmm6, 16($out,$inp)
1240	sub	\$2, $blocks		# Advance loop.
1241	lea	32($inp), $inp
1242	jnz	.Lctr32_loop
1243
1244.Lctr32_done:
1245___
1246$code.=<<___ if ($win64);
1247	movaps	0x10(%rsp),%xmm6
1248	movaps	0x20(%rsp),%xmm7
1249	movaps	0x30(%rsp),%xmm8
1250	movaps	0x40(%rsp),%xmm9
1251	movaps	0x50(%rsp),%xmm10
1252	movaps	0x60(%rsp),%xmm11
1253	movaps	0x70(%rsp),%xmm12
1254	movaps	0x80(%rsp),%xmm13
1255	movaps	0x90(%rsp),%xmm14
1256	movaps	0xa0(%rsp),%xmm15
1257	lea	0xb8(%rsp),%rsp
1258.Lctr32_epilogue:
1259___
1260$code.=<<___;
1261.Lctr32_abort:
1262	ret
1263.cfi_endproc
1264.size	${PREFIX}_ctr32_encrypt_blocks,.-${PREFIX}_ctr32_encrypt_blocks
1265___
1266}
1267$code.=<<___;
1268##
1269##  _aes_preheat
1270##
1271##  Fills register %r10 -> .aes_consts (so you can -fPIC)
1272##  and %xmm9-%xmm15 as specified below.
1273##
1274.type	_vpaes_preheat,\@abi-omnipotent
1275.align	16
1276_vpaes_preheat:
1277.cfi_startproc
1278	lea	.Lk_s0F(%rip), %r10
1279	movdqa	-0x20(%r10), %xmm10	# .Lk_inv
1280	movdqa	-0x10(%r10), %xmm11	# .Lk_inv+16
1281	movdqa	0x00(%r10), %xmm9	# .Lk_s0F
1282	movdqa	0x30(%r10), %xmm13	# .Lk_sb1
1283	movdqa	0x40(%r10), %xmm12	# .Lk_sb1+16
1284	movdqa	0x50(%r10), %xmm15	# .Lk_sb2
1285	movdqa	0x60(%r10), %xmm14	# .Lk_sb2+16
1286	ret
1287.cfi_endproc
1288.size	_vpaes_preheat,.-_vpaes_preheat
1289########################################################
1290##                                                    ##
1291##                     Constants                      ##
1292##                                                    ##
1293########################################################
1294.type	_vpaes_consts,\@object
1295.align	64
1296_vpaes_consts:
1297.Lk_inv:	# inv, inva
1298	.quad	0x0E05060F0D080180, 0x040703090A0B0C02
1299	.quad	0x01040A060F0B0780, 0x030D0E0C02050809
1300
1301.Lk_s0F:	# s0F
1302	.quad	0x0F0F0F0F0F0F0F0F, 0x0F0F0F0F0F0F0F0F
1303
1304.Lk_ipt:	# input transform (lo, hi)
1305	.quad	0xC2B2E8985A2A7000, 0xCABAE09052227808
1306	.quad	0x4C01307D317C4D00, 0xCD80B1FCB0FDCC81
1307
1308.Lk_sb1:	# sb1u, sb1t
1309	.quad	0xB19BE18FCB503E00, 0xA5DF7A6E142AF544
1310	.quad	0x3618D415FAE22300, 0x3BF7CCC10D2ED9EF
1311.Lk_sb2:	# sb2u, sb2t
1312	.quad	0xE27A93C60B712400, 0x5EB7E955BC982FCD
1313	.quad	0x69EB88400AE12900, 0xC2A163C8AB82234A
1314.Lk_sbo:	# sbou, sbot
1315	.quad	0xD0D26D176FBDC700, 0x15AABF7AC502A878
1316	.quad	0xCFE474A55FBB6A00, 0x8E1E90D1412B35FA
1317
1318.Lk_mc_forward:	# mc_forward
1319	.quad	0x0407060500030201, 0x0C0F0E0D080B0A09
1320	.quad	0x080B0A0904070605, 0x000302010C0F0E0D
1321	.quad	0x0C0F0E0D080B0A09, 0x0407060500030201
1322	.quad	0x000302010C0F0E0D, 0x080B0A0904070605
1323
1324.Lk_mc_backward:# mc_backward
1325	.quad	0x0605040702010003, 0x0E0D0C0F0A09080B
1326	.quad	0x020100030E0D0C0F, 0x0A09080B06050407
1327	.quad	0x0E0D0C0F0A09080B, 0x0605040702010003
1328	.quad	0x0A09080B06050407, 0x020100030E0D0C0F
1329
1330.Lk_sr:		# sr
1331	.quad	0x0706050403020100, 0x0F0E0D0C0B0A0908
1332	.quad	0x030E09040F0A0500, 0x0B06010C07020D08
1333	.quad	0x0F060D040B020900, 0x070E050C030A0108
1334	.quad	0x0B0E0104070A0D00, 0x0306090C0F020508
1335
1336.Lk_rcon:	# rcon
1337	.quad	0x1F8391B9AF9DEEB6, 0x702A98084D7C7D81
1338
1339.Lk_s63:	# s63: all equal to 0x63 transformed
1340	.quad	0x5B5B5B5B5B5B5B5B, 0x5B5B5B5B5B5B5B5B
1341
1342.Lk_opt:	# output transform
1343	.quad	0xFF9F4929D6B66000, 0xF7974121DEBE6808
1344	.quad	0x01EDBD5150BCEC00, 0xE10D5DB1B05C0CE0
1345
1346.Lk_deskew:	# deskew tables: inverts the sbox's "skew"
1347	.quad	0x07E4A34047A4E300, 0x1DFEB95A5DBEF91A
1348	.quad	0x5F36B5DC83EA6900, 0x2841C2ABF49D1E77
1349
1350##
1351##  Decryption stuff
1352##  Key schedule constants
1353##
1354.Lk_dksd:	# decryption key schedule: invskew x*D
1355	.quad	0xFEB91A5DA3E44700, 0x0740E3A45A1DBEF9
1356	.quad	0x41C277F4B5368300, 0x5FDC69EAAB289D1E
1357.Lk_dksb:	# decryption key schedule: invskew x*B
1358	.quad	0x9A4FCA1F8550D500, 0x03D653861CC94C99
1359	.quad	0x115BEDA7B6FC4A00, 0xD993256F7E3482C8
1360.Lk_dkse:	# decryption key schedule: invskew x*E + 0x63
1361	.quad	0xD5031CCA1FC9D600, 0x53859A4C994F5086
1362	.quad	0xA23196054FDC7BE8, 0xCD5EF96A20B31487
1363.Lk_dks9:	# decryption key schedule: invskew x*9
1364	.quad	0xB6116FC87ED9A700, 0x4AED933482255BFC
1365	.quad	0x4576516227143300, 0x8BB89FACE9DAFDCE
1366
1367##
1368##  Decryption stuff
1369##  Round function constants
1370##
1371.Lk_dipt:	# decryption input transform
1372	.quad	0x0F505B040B545F00, 0x154A411E114E451A
1373	.quad	0x86E383E660056500, 0x12771772F491F194
1374
1375.Lk_dsb9:	# decryption sbox output *9*u, *9*t
1376	.quad	0x851C03539A86D600, 0xCAD51F504F994CC9
1377	.quad	0xC03B1789ECD74900, 0x725E2C9EB2FBA565
1378.Lk_dsbd:	# decryption sbox output *D*u, *D*t
1379	.quad	0x7D57CCDFE6B1A200, 0xF56E9B13882A4439
1380	.quad	0x3CE2FAF724C6CB00, 0x2931180D15DEEFD3
1381.Lk_dsbb:	# decryption sbox output *B*u, *B*t
1382	.quad	0xD022649296B44200, 0x602646F6B0F2D404
1383	.quad	0xC19498A6CD596700, 0xF3FF0C3E3255AA6B
1384.Lk_dsbe:	# decryption sbox output *E*u, *E*t
1385	.quad	0x46F2929626D4D000, 0x2242600464B4F6B0
1386	.quad	0x0C55A6CDFFAAC100, 0x9467F36B98593E32
1387.Lk_dsbo:	# decryption sbox final output
1388	.quad	0x1387EA537EF94000, 0xC7AA6DB9D4943E2D
1389	.quad	0x12D7560F93441D00, 0xCA4B8159D8C58E9C
1390
1391# .Lrev_ctr is a permutation which byte-swaps the counter portion of the IV.
1392.Lrev_ctr:
1393	.quad	0x0706050403020100, 0x0c0d0e0f0b0a0908
1394# .Lctr_add_* may be added to a byte-swapped xmm register to increment the
1395# counter. The register must be byte-swapped again to form the actual input.
1396.Lctr_add_one:
1397	.quad	0x0000000000000000, 0x0000000100000000
1398.Lctr_add_two:
1399	.quad	0x0000000000000000, 0x0000000200000000
1400
1401.asciz	"Vector Permutation AES for x86_64/SSSE3, Mike Hamburg (Stanford University)"
1402.align	64
1403.size	_vpaes_consts,.-_vpaes_consts
1404___
1405
1406if ($win64) {
1407# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
1408#		CONTEXT *context,DISPATCHER_CONTEXT *disp)
1409$rec="%rcx";
1410$frame="%rdx";
1411$context="%r8";
1412$disp="%r9";
1413
1414$code.=<<___;
1415.extern	__imp_RtlVirtualUnwind
1416.type	se_handler,\@abi-omnipotent
1417.align	16
1418se_handler:
1419	push	%rsi
1420	push	%rdi
1421	push	%rbx
1422	push	%rbp
1423	push	%r12
1424	push	%r13
1425	push	%r14
1426	push	%r15
1427	pushfq
1428	sub	\$64,%rsp
1429
1430	mov	120($context),%rax	# pull context->Rax
1431	mov	248($context),%rbx	# pull context->Rip
1432
1433	mov	8($disp),%rsi		# disp->ImageBase
1434	mov	56($disp),%r11		# disp->HandlerData
1435
1436	mov	0(%r11),%r10d		# HandlerData[0]
1437	lea	(%rsi,%r10),%r10	# prologue label
1438	cmp	%r10,%rbx		# context->Rip<prologue label
1439	jb	.Lin_prologue
1440
1441	mov	152($context),%rax	# pull context->Rsp
1442
1443	mov	4(%r11),%r10d		# HandlerData[1]
1444	lea	(%rsi,%r10),%r10	# epilogue label
1445	cmp	%r10,%rbx		# context->Rip>=epilogue label
1446	jae	.Lin_prologue
1447
1448	lea	16(%rax),%rsi		# %xmm save area
1449	lea	512($context),%rdi	# &context.Xmm6
1450	mov	\$20,%ecx		# 10*sizeof(%xmm0)/sizeof(%rax)
1451	.long	0xa548f3fc		# cld; rep movsq
1452	lea	0xb8(%rax),%rax		# adjust stack pointer
1453
1454.Lin_prologue:
1455	mov	8(%rax),%rdi
1456	mov	16(%rax),%rsi
1457	mov	%rax,152($context)	# restore context->Rsp
1458	mov	%rsi,168($context)	# restore context->Rsi
1459	mov	%rdi,176($context)	# restore context->Rdi
1460
1461	mov	40($disp),%rdi		# disp->ContextRecord
1462	mov	$context,%rsi		# context
1463	mov	\$`1232/8`,%ecx		# sizeof(CONTEXT)
1464	.long	0xa548f3fc		# cld; rep movsq
1465
1466	mov	$disp,%rsi
1467	xor	%rcx,%rcx		# arg1, UNW_FLAG_NHANDLER
1468	mov	8(%rsi),%rdx		# arg2, disp->ImageBase
1469	mov	0(%rsi),%r8		# arg3, disp->ControlPc
1470	mov	16(%rsi),%r9		# arg4, disp->FunctionEntry
1471	mov	40(%rsi),%r10		# disp->ContextRecord
1472	lea	56(%rsi),%r11		# &disp->HandlerData
1473	lea	24(%rsi),%r12		# &disp->EstablisherFrame
1474	mov	%r10,32(%rsp)		# arg5
1475	mov	%r11,40(%rsp)		# arg6
1476	mov	%r12,48(%rsp)		# arg7
1477	mov	%rcx,56(%rsp)		# arg8, (NULL)
1478	call	*__imp_RtlVirtualUnwind(%rip)
1479
1480	mov	\$1,%eax		# ExceptionContinueSearch
1481	add	\$64,%rsp
1482	popfq
1483	pop	%r15
1484	pop	%r14
1485	pop	%r13
1486	pop	%r12
1487	pop	%rbp
1488	pop	%rbx
1489	pop	%rdi
1490	pop	%rsi
1491	ret
1492.size	se_handler,.-se_handler
1493
1494.section	.pdata
1495.align	4
1496	.rva	.LSEH_begin_${PREFIX}_set_encrypt_key
1497	.rva	.LSEH_end_${PREFIX}_set_encrypt_key
1498	.rva	.LSEH_info_${PREFIX}_set_encrypt_key
1499
1500	.rva	.LSEH_begin_${PREFIX}_set_decrypt_key
1501	.rva	.LSEH_end_${PREFIX}_set_decrypt_key
1502	.rva	.LSEH_info_${PREFIX}_set_decrypt_key
1503
1504	.rva	.LSEH_begin_${PREFIX}_encrypt
1505	.rva	.LSEH_end_${PREFIX}_encrypt
1506	.rva	.LSEH_info_${PREFIX}_encrypt
1507
1508	.rva	.LSEH_begin_${PREFIX}_decrypt
1509	.rva	.LSEH_end_${PREFIX}_decrypt
1510	.rva	.LSEH_info_${PREFIX}_decrypt
1511
1512	.rva	.LSEH_begin_${PREFIX}_cbc_encrypt
1513	.rva	.LSEH_end_${PREFIX}_cbc_encrypt
1514	.rva	.LSEH_info_${PREFIX}_cbc_encrypt
1515
1516	.rva	.LSEH_begin_${PREFIX}_ctr32_encrypt_blocks
1517	.rva	.LSEH_end_${PREFIX}_ctr32_encrypt_blocks
1518	.rva	.LSEH_info_${PREFIX}_ctr32_encrypt_blocks
1519
1520.section	.xdata
1521.align	8
1522.LSEH_info_${PREFIX}_set_encrypt_key:
1523	.byte	9,0,0,0
1524	.rva	se_handler
1525	.rva	.Lenc_key_body,.Lenc_key_epilogue	# HandlerData[]
1526.LSEH_info_${PREFIX}_set_decrypt_key:
1527	.byte	9,0,0,0
1528	.rva	se_handler
1529	.rva	.Ldec_key_body,.Ldec_key_epilogue	# HandlerData[]
1530.LSEH_info_${PREFIX}_encrypt:
1531	.byte	9,0,0,0
1532	.rva	se_handler
1533	.rva	.Lenc_body,.Lenc_epilogue		# HandlerData[]
1534.LSEH_info_${PREFIX}_decrypt:
1535	.byte	9,0,0,0
1536	.rva	se_handler
1537	.rva	.Ldec_body,.Ldec_epilogue		# HandlerData[]
1538.LSEH_info_${PREFIX}_cbc_encrypt:
1539	.byte	9,0,0,0
1540	.rva	se_handler
1541	.rva	.Lcbc_body,.Lcbc_epilogue		# HandlerData[]
1542.LSEH_info_${PREFIX}_ctr32_encrypt_blocks:
1543	.byte	9,0,0,0
1544	.rva	se_handler
1545	.rva	.Lctr32_body,.Lctr32_epilogue		# HandlerData[]
1546___
1547}
1548
1549$code =~ s/\`([^\`]*)\`/eval($1)/gem;
1550
1551print $code;
1552
1553close STDOUT;
1554