xref: /openssl/crypto/aes/asm/aest4-sparcv9.pl (revision 54b40531) 
1#! /usr/bin/env perl
2# Copyright 2012-2021 The OpenSSL Project Authors. All Rights Reserved.
3#
4# Licensed under the Apache License 2.0 (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# Written by David S. Miller and Andy Polyakov.
12# The module is licensed under 2-clause BSD license. October 2012.
13# All rights reserved.
14# ====================================================================
15
16######################################################################
17# AES for SPARC T4.
18#
19# AES round instructions complete in 3 cycles and can be issued every
20# cycle. It means that round calculations should take 4*rounds cycles,
21# because any given round instruction depends on result of *both*
22# previous instructions:
23#
24#	|0 |1 |2 |3 |4
25#	|01|01|01|
26#	   |23|23|23|
27#	            |01|01|...
28#	               |23|...
29#
30# Provided that fxor [with IV] takes 3 cycles to complete, critical
31# path length for CBC encrypt would be 3+4*rounds, or in other words
32# it should process one byte in at least (3+4*rounds)/16 cycles. This
33# estimate doesn't account for "collateral" instructions, such as
34# fetching input from memory, xor-ing it with zero-round key and
35# storing the result. Yet, *measured* performance [for data aligned
36# at 64-bit boundary!] deviates from this equation by less than 0.5%:
37#
38#		128-bit key	192-		256-
39# CBC encrypt	2.70/2.90(*)	3.20/3.40	3.70/3.90
40#			 (*) numbers after slash are for
41#			     misaligned data.
42#
43# Out-of-order execution logic managed to fully overlap "collateral"
44# instructions with those on critical path. Amazing!
45#
46# As with Intel AES-NI, question is if it's possible to improve
47# performance of parallelizable modes by interleaving round
48# instructions. Provided round instruction latency and throughput
49# optimal interleave factor is 2. But can we expect 2x performance
50# improvement? Well, as round instructions can be issued one per
51# cycle, they don't saturate the 2-way issue pipeline and therefore
52# there is room for "collateral" calculations... Yet, 2x speed-up
53# over CBC encrypt remains unattaintable:
54#
55#		128-bit key	192-		256-
56# CBC decrypt	1.64/2.11	1.89/2.37	2.23/2.61
57# CTR		1.64/2.08(*)	1.89/2.33	2.23/2.61
58#			 (*) numbers after slash are for
59#			     misaligned data.
60#
61# Estimates based on amount of instructions under assumption that
62# round instructions are not pairable with any other instruction
63# suggest that latter is the actual case and pipeline runs
64# underutilized. It should be noted that T4 out-of-order execution
65# logic is so capable that performance gain from 2x interleave is
66# not even impressive, ~7-13% over non-interleaved code, largest
67# for 256-bit keys.
68
69# To anchor to something else, software implementation processes
70# one byte in 29 cycles with 128-bit key on same processor. Intel
71# Sandy Bridge encrypts byte in 5.07 cycles in CBC mode and decrypts
72# in 0.93, naturally with AES-NI.
73
74$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
75push(@INC,"${dir}","${dir}../../perlasm");
76require "sparcv9_modes.pl";
77
78$output = pop and open STDOUT,">$output";
79
80$::evp=1;	# if $evp is set to 0, script generates module with
81# AES_[en|de]crypt, AES_set_[en|de]crypt_key and AES_cbc_encrypt entry
82# points. These however are not fully compatible with openssl/aes.h,
83# because they expect AES_KEY to be aligned at 64-bit boundary. When
84# used through EVP, alignment is arranged at EVP layer. Second thing
85# that is arranged by EVP is at least 32-bit alignment of IV.
86
87######################################################################
88# single-round subroutines
89#
90{
91my ($inp,$out,$key,$rounds,$tmp,$mask)=map("%o$_",(0..5));
92
93$code.=<<___;
94#ifndef __ASSEMBLER__
95# define __ASSEMBLER__ 1
96#endif
97#include "crypto/sparc_arch.h"
98
99#ifdef	__arch64__
100.register	%g2,#scratch
101.register	%g3,#scratch
102#endif
103
104.text
105
106.globl	aes_t4_encrypt
107.align	32
108aes_t4_encrypt:
109	andcc		$inp, 7, %g1		! is input aligned?
110	andn		$inp, 7, $inp
111
112	ldx		[$key + 0], %g4
113	ldx		[$key + 8], %g5
114
115	ldx		[$inp + 0], %o4
116	bz,pt		%icc, 1f
117	ldx		[$inp + 8], %o5
118	ldx		[$inp + 16], $inp
119	sll		%g1, 3, %g1
120	sub		%g0, %g1, %o3
121	sllx		%o4, %g1, %o4
122	sllx		%o5, %g1, %g1
123	srlx		%o5, %o3, %o5
124	srlx		$inp, %o3, %o3
125	or		%o5, %o4, %o4
126	or		%o3, %g1, %o5
1271:
128	ld		[$key + 240], $rounds
129	ldd		[$key + 16], %f12
130	ldd		[$key + 24], %f14
131	xor		%g4, %o4, %o4
132	xor		%g5, %o5, %o5
133	movxtod		%o4, %f0
134	movxtod		%o5, %f2
135	srl		$rounds, 1, $rounds
136	ldd		[$key + 32], %f16
137	sub		$rounds, 1, $rounds
138	ldd		[$key + 40], %f18
139	add		$key, 48, $key
140
141.Lenc:
142	aes_eround01	%f12, %f0, %f2, %f4
143	aes_eround23	%f14, %f0, %f2, %f2
144	ldd		[$key + 0], %f12
145	ldd		[$key + 8], %f14
146	sub		$rounds,1,$rounds
147	aes_eround01	%f16, %f4, %f2, %f0
148	aes_eround23	%f18, %f4, %f2, %f2
149	ldd		[$key + 16], %f16
150	ldd		[$key + 24], %f18
151	brnz,pt		$rounds, .Lenc
152	add		$key, 32, $key
153
154	andcc		$out, 7, $tmp		! is output aligned?
155	aes_eround01	%f12, %f0, %f2, %f4
156	aes_eround23	%f14, %f0, %f2, %f2
157	aes_eround01_l	%f16, %f4, %f2, %f0
158	aes_eround23_l	%f18, %f4, %f2, %f2
159
160	bnz,pn		%icc, 2f
161	nop
162
163	std		%f0, [$out + 0]
164	retl
165	std		%f2, [$out + 8]
166
1672:	alignaddrl	$out, %g0, $out
168	mov		0xff, $mask
169	srl		$mask, $tmp, $mask
170
171	faligndata	%f0, %f0, %f4
172	faligndata	%f0, %f2, %f6
173	faligndata	%f2, %f2, %f8
174
175	stda		%f4, [$out + $mask]0xc0	! partial store
176	std		%f6, [$out + 8]
177	add		$out, 16, $out
178	orn		%g0, $mask, $mask
179	retl
180	stda		%f8, [$out + $mask]0xc0	! partial store
181.type	aes_t4_encrypt,#function
182.size	aes_t4_encrypt,.-aes_t4_encrypt
183
184.globl	aes_t4_decrypt
185.align	32
186aes_t4_decrypt:
187	andcc		$inp, 7, %g1		! is input aligned?
188	andn		$inp, 7, $inp
189
190	ldx		[$key + 0], %g4
191	ldx		[$key + 8], %g5
192
193	ldx		[$inp + 0], %o4
194	bz,pt		%icc, 1f
195	ldx		[$inp + 8], %o5
196	ldx		[$inp + 16], $inp
197	sll		%g1, 3, %g1
198	sub		%g0, %g1, %o3
199	sllx		%o4, %g1, %o4
200	sllx		%o5, %g1, %g1
201	srlx		%o5, %o3, %o5
202	srlx		$inp, %o3, %o3
203	or		%o5, %o4, %o4
204	or		%o3, %g1, %o5
2051:
206	ld		[$key + 240], $rounds
207	ldd		[$key + 16], %f12
208	ldd		[$key + 24], %f14
209	xor		%g4, %o4, %o4
210	xor		%g5, %o5, %o5
211	movxtod		%o4, %f0
212	movxtod		%o5, %f2
213	srl		$rounds, 1, $rounds
214	ldd		[$key + 32], %f16
215	sub		$rounds, 1, $rounds
216	ldd		[$key + 40], %f18
217	add		$key, 48, $key
218
219.Ldec:
220	aes_dround01	%f12, %f0, %f2, %f4
221	aes_dround23	%f14, %f0, %f2, %f2
222	ldd		[$key + 0], %f12
223	ldd		[$key + 8], %f14
224	sub		$rounds,1,$rounds
225	aes_dround01	%f16, %f4, %f2, %f0
226	aes_dround23	%f18, %f4, %f2, %f2
227	ldd		[$key + 16], %f16
228	ldd		[$key + 24], %f18
229	brnz,pt		$rounds, .Ldec
230	add		$key, 32, $key
231
232	andcc		$out, 7, $tmp		! is output aligned?
233	aes_dround01	%f12, %f0, %f2, %f4
234	aes_dround23	%f14, %f0, %f2, %f2
235	aes_dround01_l	%f16, %f4, %f2, %f0
236	aes_dround23_l	%f18, %f4, %f2, %f2
237
238	bnz,pn		%icc, 2f
239	nop
240
241	std		%f0, [$out + 0]
242	retl
243	std		%f2, [$out + 8]
244
2452:	alignaddrl	$out, %g0, $out
246	mov		0xff, $mask
247	srl		$mask, $tmp, $mask
248
249	faligndata	%f0, %f0, %f4
250	faligndata	%f0, %f2, %f6
251	faligndata	%f2, %f2, %f8
252
253	stda		%f4, [$out + $mask]0xc0	! partial store
254	std		%f6, [$out + 8]
255	add		$out, 16, $out
256	orn		%g0, $mask, $mask
257	retl
258	stda		%f8, [$out + $mask]0xc0	! partial store
259.type	aes_t4_decrypt,#function
260.size	aes_t4_decrypt,.-aes_t4_decrypt
261___
262}
263
264######################################################################
265# key setup subroutines
266#
267{
268my ($inp,$bits,$out,$tmp)=map("%o$_",(0..5));
269$code.=<<___;
270.globl	aes_t4_set_encrypt_key
271.align	32
272aes_t4_set_encrypt_key:
273.Lset_encrypt_key:
274	and		$inp, 7, $tmp
275	alignaddr	$inp, %g0, $inp
276	cmp		$bits, 192
277	ldd		[$inp + 0], %f0
278	bl,pt		%icc,.L128
279	ldd		[$inp + 8], %f2
280
281	be,pt		%icc,.L192
282	ldd		[$inp + 16], %f4
283	brz,pt		$tmp, .L256aligned
284	ldd		[$inp + 24], %f6
285
286	ldd		[$inp + 32], %f8
287	faligndata	%f0, %f2, %f0
288	faligndata	%f2, %f4, %f2
289	faligndata	%f4, %f6, %f4
290	faligndata	%f6, %f8, %f6
291.L256aligned:
292___
293for ($i=0; $i<6; $i++) {
294    $code.=<<___;
295	std		%f0, [$out + `32*$i+0`]
296	aes_kexpand1	%f0, %f6, $i, %f0
297	std		%f2, [$out + `32*$i+8`]
298	aes_kexpand2	%f2, %f0, %f2
299	std		%f4, [$out + `32*$i+16`]
300	aes_kexpand0	%f4, %f2, %f4
301	std		%f6, [$out + `32*$i+24`]
302	aes_kexpand2	%f6, %f4, %f6
303___
304}
305$code.=<<___;
306	std		%f0, [$out + `32*$i+0`]
307	aes_kexpand1	%f0, %f6, $i, %f0
308	std		%f2, [$out + `32*$i+8`]
309	aes_kexpand2	%f2, %f0, %f2
310	std		%f4, [$out + `32*$i+16`]
311	std		%f6, [$out + `32*$i+24`]
312	std		%f0, [$out + `32*$i+32`]
313	std		%f2, [$out + `32*$i+40`]
314
315	mov		14, $tmp
316	st		$tmp, [$out + 240]
317	retl
318	xor		%o0, %o0, %o0
319
320.align	16
321.L192:
322	brz,pt		$tmp, .L192aligned
323	nop
324
325	ldd		[$inp + 24], %f6
326	faligndata	%f0, %f2, %f0
327	faligndata	%f2, %f4, %f2
328	faligndata	%f4, %f6, %f4
329.L192aligned:
330___
331for ($i=0; $i<7; $i++) {
332    $code.=<<___;
333	std		%f0, [$out + `24*$i+0`]
334	aes_kexpand1	%f0, %f4, $i, %f0
335	std		%f2, [$out + `24*$i+8`]
336	aes_kexpand2	%f2, %f0, %f2
337	std		%f4, [$out + `24*$i+16`]
338	aes_kexpand2	%f4, %f2, %f4
339___
340}
341$code.=<<___;
342	std		%f0, [$out + `24*$i+0`]
343	aes_kexpand1	%f0, %f4, $i, %f0
344	std		%f2, [$out + `24*$i+8`]
345	aes_kexpand2	%f2, %f0, %f2
346	std		%f4, [$out + `24*$i+16`]
347	std		%f0, [$out + `24*$i+24`]
348	std		%f2, [$out + `24*$i+32`]
349
350	mov		12, $tmp
351	st		$tmp, [$out + 240]
352	retl
353	xor		%o0, %o0, %o0
354
355.align	16
356.L128:
357	brz,pt		$tmp, .L128aligned
358	nop
359
360	ldd		[$inp + 16], %f4
361	faligndata	%f0, %f2, %f0
362	faligndata	%f2, %f4, %f2
363.L128aligned:
364___
365for ($i=0; $i<10; $i++) {
366    $code.=<<___;
367	std		%f0, [$out + `16*$i+0`]
368	aes_kexpand1	%f0, %f2, $i, %f0
369	std		%f2, [$out + `16*$i+8`]
370	aes_kexpand2	%f2, %f0, %f2
371___
372}
373$code.=<<___;
374	std		%f0, [$out + `16*$i+0`]
375	std		%f2, [$out + `16*$i+8`]
376
377	mov		10, $tmp
378	st		$tmp, [$out + 240]
379	retl
380	xor		%o0, %o0, %o0
381.type	aes_t4_set_encrypt_key,#function
382.size	aes_t4_set_encrypt_key,.-aes_t4_set_encrypt_key
383
384.globl	aes_t4_set_decrypt_key
385.align	32
386aes_t4_set_decrypt_key:
387	mov		%o7, %o5
388	call		.Lset_encrypt_key
389	nop
390
391	mov		%o5, %o7
392	sll		$tmp, 4, $inp		! $tmp is number of rounds
393	add		$tmp, 2, $tmp
394	add		$out, $inp, $inp	! $inp=$out+16*rounds
395	srl		$tmp, 2, $tmp		! $tmp=(rounds+2)/4
396
397.Lkey_flip:
398	ldd		[$out + 0],  %f0
399	ldd		[$out + 8],  %f2
400	ldd		[$out + 16], %f4
401	ldd		[$out + 24], %f6
402	ldd		[$inp + 0],  %f8
403	ldd		[$inp + 8],  %f10
404	ldd		[$inp - 16], %f12
405	ldd		[$inp - 8],  %f14
406	sub		$tmp, 1, $tmp
407	std		%f0, [$inp + 0]
408	std		%f2, [$inp + 8]
409	std		%f4, [$inp - 16]
410	std		%f6, [$inp - 8]
411	std		%f8, [$out + 0]
412	std		%f10, [$out + 8]
413	std		%f12, [$out + 16]
414	std		%f14, [$out + 24]
415	add		$out, 32, $out
416	brnz		$tmp, .Lkey_flip
417	sub		$inp, 32, $inp
418
419	retl
420	xor		%o0, %o0, %o0
421.type	aes_t4_set_decrypt_key,#function
422.size	aes_t4_set_decrypt_key,.-aes_t4_set_decrypt_key
423___
424}
425
426{{{
427my ($inp,$out,$len,$key,$ivec,$enc)=map("%i$_",(0..5));
428my ($ileft,$iright,$ooff,$omask,$ivoff)=map("%l$_",(1..7));
429
430$code.=<<___;
431.align	32
432_aes128_encrypt_1x:
433___
434for ($i=0; $i<4; $i++) {
435    $code.=<<___;
436	aes_eround01	%f`16+8*$i+0`, %f0, %f2, %f4
437	aes_eround23	%f`16+8*$i+2`, %f0, %f2, %f2
438	aes_eround01	%f`16+8*$i+4`, %f4, %f2, %f0
439	aes_eround23	%f`16+8*$i+6`, %f4, %f2, %f2
440___
441}
442$code.=<<___;
443	aes_eround01	%f48, %f0, %f2, %f4
444	aes_eround23	%f50, %f0, %f2, %f2
445	aes_eround01_l	%f52, %f4, %f2, %f0
446	retl
447	aes_eround23_l	%f54, %f4, %f2, %f2
448.type	_aes128_encrypt_1x,#function
449.size	_aes128_encrypt_1x,.-_aes128_encrypt_1x
450
451.align	32
452_aes128_encrypt_2x:
453___
454for ($i=0; $i<4; $i++) {
455    $code.=<<___;
456	aes_eround01	%f`16+8*$i+0`, %f0, %f2, %f8
457	aes_eround23	%f`16+8*$i+2`, %f0, %f2, %f2
458	aes_eround01	%f`16+8*$i+0`, %f4, %f6, %f10
459	aes_eround23	%f`16+8*$i+2`, %f4, %f6, %f6
460	aes_eround01	%f`16+8*$i+4`, %f8, %f2, %f0
461	aes_eround23	%f`16+8*$i+6`, %f8, %f2, %f2
462	aes_eround01	%f`16+8*$i+4`, %f10, %f6, %f4
463	aes_eround23	%f`16+8*$i+6`, %f10, %f6, %f6
464___
465}
466$code.=<<___;
467	aes_eround01	%f48, %f0, %f2, %f8
468	aes_eround23	%f50, %f0, %f2, %f2
469	aes_eround01	%f48, %f4, %f6, %f10
470	aes_eround23	%f50, %f4, %f6, %f6
471	aes_eround01_l	%f52, %f8, %f2, %f0
472	aes_eround23_l	%f54, %f8, %f2, %f2
473	aes_eround01_l	%f52, %f10, %f6, %f4
474	retl
475	aes_eround23_l	%f54, %f10, %f6, %f6
476.type	_aes128_encrypt_2x,#function
477.size	_aes128_encrypt_2x,.-_aes128_encrypt_2x
478
479.align	32
480_aes128_loadkey:
481	ldx		[$key + 0], %g4
482	ldx		[$key + 8], %g5
483___
484for ($i=2; $i<22;$i++) {			# load key schedule
485    $code.=<<___;
486	ldd		[$key + `8*$i`], %f`12+2*$i`
487___
488}
489$code.=<<___;
490	retl
491	nop
492.type	_aes128_loadkey,#function
493.size	_aes128_loadkey,.-_aes128_loadkey
494_aes128_load_enckey=_aes128_loadkey
495_aes128_load_deckey=_aes128_loadkey
496
497___
498
499&alg_cbc_encrypt_implement("aes",128);
500if ($::evp) {
501    &alg_ctr32_implement("aes",128);
502    &alg_xts_implement("aes",128,"en");
503    &alg_xts_implement("aes",128,"de");
504}
505&alg_cbc_decrypt_implement("aes",128);
506
507$code.=<<___;
508.align	32
509_aes128_decrypt_1x:
510___
511for ($i=0; $i<4; $i++) {
512    $code.=<<___;
513	aes_dround01	%f`16+8*$i+0`, %f0, %f2, %f4
514	aes_dround23	%f`16+8*$i+2`, %f0, %f2, %f2
515	aes_dround01	%f`16+8*$i+4`, %f4, %f2, %f0
516	aes_dround23	%f`16+8*$i+6`, %f4, %f2, %f2
517___
518}
519$code.=<<___;
520	aes_dround01	%f48, %f0, %f2, %f4
521	aes_dround23	%f50, %f0, %f2, %f2
522	aes_dround01_l	%f52, %f4, %f2, %f0
523	retl
524	aes_dround23_l	%f54, %f4, %f2, %f2
525.type	_aes128_decrypt_1x,#function
526.size	_aes128_decrypt_1x,.-_aes128_decrypt_1x
527
528.align	32
529_aes128_decrypt_2x:
530___
531for ($i=0; $i<4; $i++) {
532    $code.=<<___;
533	aes_dround01	%f`16+8*$i+0`, %f0, %f2, %f8
534	aes_dround23	%f`16+8*$i+2`, %f0, %f2, %f2
535	aes_dround01	%f`16+8*$i+0`, %f4, %f6, %f10
536	aes_dround23	%f`16+8*$i+2`, %f4, %f6, %f6
537	aes_dround01	%f`16+8*$i+4`, %f8, %f2, %f0
538	aes_dround23	%f`16+8*$i+6`, %f8, %f2, %f2
539	aes_dround01	%f`16+8*$i+4`, %f10, %f6, %f4
540	aes_dround23	%f`16+8*$i+6`, %f10, %f6, %f6
541___
542}
543$code.=<<___;
544	aes_dround01	%f48, %f0, %f2, %f8
545	aes_dround23	%f50, %f0, %f2, %f2
546	aes_dround01	%f48, %f4, %f6, %f10
547	aes_dround23	%f50, %f4, %f6, %f6
548	aes_dround01_l	%f52, %f8, %f2, %f0
549	aes_dround23_l	%f54, %f8, %f2, %f2
550	aes_dround01_l	%f52, %f10, %f6, %f4
551	retl
552	aes_dround23_l	%f54, %f10, %f6, %f6
553.type	_aes128_decrypt_2x,#function
554.size	_aes128_decrypt_2x,.-_aes128_decrypt_2x
555___
556
557$code.=<<___;
558.align	32
559_aes192_encrypt_1x:
560___
561for ($i=0; $i<5; $i++) {
562    $code.=<<___;
563	aes_eround01	%f`16+8*$i+0`, %f0, %f2, %f4
564	aes_eround23	%f`16+8*$i+2`, %f0, %f2, %f2
565	aes_eround01	%f`16+8*$i+4`, %f4, %f2, %f0
566	aes_eround23	%f`16+8*$i+6`, %f4, %f2, %f2
567___
568}
569$code.=<<___;
570	aes_eround01	%f56, %f0, %f2, %f4
571	aes_eround23	%f58, %f0, %f2, %f2
572	aes_eround01_l	%f60, %f4, %f2, %f0
573	retl
574	aes_eround23_l	%f62, %f4, %f2, %f2
575.type	_aes192_encrypt_1x,#function
576.size	_aes192_encrypt_1x,.-_aes192_encrypt_1x
577
578.align	32
579_aes192_encrypt_2x:
580___
581for ($i=0; $i<5; $i++) {
582    $code.=<<___;
583	aes_eround01	%f`16+8*$i+0`, %f0, %f2, %f8
584	aes_eround23	%f`16+8*$i+2`, %f0, %f2, %f2
585	aes_eround01	%f`16+8*$i+0`, %f4, %f6, %f10
586	aes_eround23	%f`16+8*$i+2`, %f4, %f6, %f6
587	aes_eround01	%f`16+8*$i+4`, %f8, %f2, %f0
588	aes_eround23	%f`16+8*$i+6`, %f8, %f2, %f2
589	aes_eround01	%f`16+8*$i+4`, %f10, %f6, %f4
590	aes_eround23	%f`16+8*$i+6`, %f10, %f6, %f6
591___
592}
593$code.=<<___;
594	aes_eround01	%f56, %f0, %f2, %f8
595	aes_eround23	%f58, %f0, %f2, %f2
596	aes_eround01	%f56, %f4, %f6, %f10
597	aes_eround23	%f58, %f4, %f6, %f6
598	aes_eround01_l	%f60, %f8, %f2, %f0
599	aes_eround23_l	%f62, %f8, %f2, %f2
600	aes_eround01_l	%f60, %f10, %f6, %f4
601	retl
602	aes_eround23_l	%f62, %f10, %f6, %f6
603.type	_aes192_encrypt_2x,#function
604.size	_aes192_encrypt_2x,.-_aes192_encrypt_2x
605
606.align	32
607_aes256_encrypt_1x:
608	aes_eround01	%f16, %f0, %f2, %f4
609	aes_eround23	%f18, %f0, %f2, %f2
610	ldd		[$key + 208], %f16
611	ldd		[$key + 216], %f18
612	aes_eround01	%f20, %f4, %f2, %f0
613	aes_eround23	%f22, %f4, %f2, %f2
614	ldd		[$key + 224], %f20
615	ldd		[$key + 232], %f22
616___
617for ($i=1; $i<6; $i++) {
618    $code.=<<___;
619	aes_eround01	%f`16+8*$i+0`, %f0, %f2, %f4
620	aes_eround23	%f`16+8*$i+2`, %f0, %f2, %f2
621	aes_eround01	%f`16+8*$i+4`, %f4, %f2, %f0
622	aes_eround23	%f`16+8*$i+6`, %f4, %f2, %f2
623___
624}
625$code.=<<___;
626	aes_eround01	%f16, %f0, %f2, %f4
627	aes_eround23	%f18, %f0, %f2, %f2
628	ldd		[$key + 16], %f16
629	ldd		[$key + 24], %f18
630	aes_eround01_l	%f20, %f4, %f2, %f0
631	aes_eround23_l	%f22, %f4, %f2, %f2
632	ldd		[$key + 32], %f20
633	retl
634	ldd		[$key + 40], %f22
635.type	_aes256_encrypt_1x,#function
636.size	_aes256_encrypt_1x,.-_aes256_encrypt_1x
637
638.align	32
639_aes256_encrypt_2x:
640	aes_eround01	%f16, %f0, %f2, %f8
641	aes_eround23	%f18, %f0, %f2, %f2
642	aes_eround01	%f16, %f4, %f6, %f10
643	aes_eround23	%f18, %f4, %f6, %f6
644	ldd		[$key + 208], %f16
645	ldd		[$key + 216], %f18
646	aes_eround01	%f20, %f8, %f2, %f0
647	aes_eround23	%f22, %f8, %f2, %f2
648	aes_eround01	%f20, %f10, %f6, %f4
649	aes_eround23	%f22, %f10, %f6, %f6
650	ldd		[$key + 224], %f20
651	ldd		[$key + 232], %f22
652___
653for ($i=1; $i<6; $i++) {
654    $code.=<<___;
655	aes_eround01	%f`16+8*$i+0`, %f0, %f2, %f8
656	aes_eround23	%f`16+8*$i+2`, %f0, %f2, %f2
657	aes_eround01	%f`16+8*$i+0`, %f4, %f6, %f10
658	aes_eround23	%f`16+8*$i+2`, %f4, %f6, %f6
659	aes_eround01	%f`16+8*$i+4`, %f8, %f2, %f0
660	aes_eround23	%f`16+8*$i+6`, %f8, %f2, %f2
661	aes_eround01	%f`16+8*$i+4`, %f10, %f6, %f4
662	aes_eround23	%f`16+8*$i+6`, %f10, %f6, %f6
663___
664}
665$code.=<<___;
666	aes_eround01	%f16, %f0, %f2, %f8
667	aes_eround23	%f18, %f0, %f2, %f2
668	aes_eround01	%f16, %f4, %f6, %f10
669	aes_eround23	%f18, %f4, %f6, %f6
670	ldd		[$key + 16], %f16
671	ldd		[$key + 24], %f18
672	aes_eround01_l	%f20, %f8, %f2, %f0
673	aes_eround23_l	%f22, %f8, %f2, %f2
674	aes_eround01_l	%f20, %f10, %f6, %f4
675	aes_eround23_l	%f22, %f10, %f6, %f6
676	ldd		[$key + 32], %f20
677	retl
678	ldd		[$key + 40], %f22
679.type	_aes256_encrypt_2x,#function
680.size	_aes256_encrypt_2x,.-_aes256_encrypt_2x
681
682.align	32
683_aes192_loadkey:
684	ldx		[$key + 0], %g4
685	ldx		[$key + 8], %g5
686___
687for ($i=2; $i<26;$i++) {			# load key schedule
688    $code.=<<___;
689	ldd		[$key + `8*$i`], %f`12+2*$i`
690___
691}
692$code.=<<___;
693	retl
694	nop
695.type	_aes192_loadkey,#function
696.size	_aes192_loadkey,.-_aes192_loadkey
697_aes256_loadkey=_aes192_loadkey
698_aes192_load_enckey=_aes192_loadkey
699_aes192_load_deckey=_aes192_loadkey
700_aes256_load_enckey=_aes192_loadkey
701_aes256_load_deckey=_aes192_loadkey
702___
703
704&alg_cbc_encrypt_implement("aes",256);
705&alg_cbc_encrypt_implement("aes",192);
706if ($::evp) {
707    &alg_ctr32_implement("aes",256);
708    &alg_xts_implement("aes",256,"en");
709    &alg_xts_implement("aes",256,"de");
710    &alg_ctr32_implement("aes",192);
711}
712&alg_cbc_decrypt_implement("aes",192);
713&alg_cbc_decrypt_implement("aes",256);
714
715$code.=<<___;
716.align	32
717_aes256_decrypt_1x:
718	aes_dround01	%f16, %f0, %f2, %f4
719	aes_dround23	%f18, %f0, %f2, %f2
720	ldd		[$key + 208], %f16
721	ldd		[$key + 216], %f18
722	aes_dround01	%f20, %f4, %f2, %f0
723	aes_dround23	%f22, %f4, %f2, %f2
724	ldd		[$key + 224], %f20
725	ldd		[$key + 232], %f22
726___
727for ($i=1; $i<6; $i++) {
728    $code.=<<___;
729	aes_dround01	%f`16+8*$i+0`, %f0, %f2, %f4
730	aes_dround23	%f`16+8*$i+2`, %f0, %f2, %f2
731	aes_dround01	%f`16+8*$i+4`, %f4, %f2, %f0
732	aes_dround23	%f`16+8*$i+6`, %f4, %f2, %f2
733___
734}
735$code.=<<___;
736	aes_dround01	%f16, %f0, %f2, %f4
737	aes_dround23	%f18, %f0, %f2, %f2
738	ldd		[$key + 16], %f16
739	ldd		[$key + 24], %f18
740	aes_dround01_l	%f20, %f4, %f2, %f0
741	aes_dround23_l	%f22, %f4, %f2, %f2
742	ldd		[$key + 32], %f20
743	retl
744	ldd		[$key + 40], %f22
745.type	_aes256_decrypt_1x,#function
746.size	_aes256_decrypt_1x,.-_aes256_decrypt_1x
747
748.align	32
749_aes256_decrypt_2x:
750	aes_dround01	%f16, %f0, %f2, %f8
751	aes_dround23	%f18, %f0, %f2, %f2
752	aes_dround01	%f16, %f4, %f6, %f10
753	aes_dround23	%f18, %f4, %f6, %f6
754	ldd		[$key + 208], %f16
755	ldd		[$key + 216], %f18
756	aes_dround01	%f20, %f8, %f2, %f0
757	aes_dround23	%f22, %f8, %f2, %f2
758	aes_dround01	%f20, %f10, %f6, %f4
759	aes_dround23	%f22, %f10, %f6, %f6
760	ldd		[$key + 224], %f20
761	ldd		[$key + 232], %f22
762___
763for ($i=1; $i<6; $i++) {
764    $code.=<<___;
765	aes_dround01	%f`16+8*$i+0`, %f0, %f2, %f8
766	aes_dround23	%f`16+8*$i+2`, %f0, %f2, %f2
767	aes_dround01	%f`16+8*$i+0`, %f4, %f6, %f10
768	aes_dround23	%f`16+8*$i+2`, %f4, %f6, %f6
769	aes_dround01	%f`16+8*$i+4`, %f8, %f2, %f0
770	aes_dround23	%f`16+8*$i+6`, %f8, %f2, %f2
771	aes_dround01	%f`16+8*$i+4`, %f10, %f6, %f4
772	aes_dround23	%f`16+8*$i+6`, %f10, %f6, %f6
773___
774}
775$code.=<<___;
776	aes_dround01	%f16, %f0, %f2, %f8
777	aes_dround23	%f18, %f0, %f2, %f2
778	aes_dround01	%f16, %f4, %f6, %f10
779	aes_dround23	%f18, %f4, %f6, %f6
780	ldd		[$key + 16], %f16
781	ldd		[$key + 24], %f18
782	aes_dround01_l	%f20, %f8, %f2, %f0
783	aes_dround23_l	%f22, %f8, %f2, %f2
784	aes_dround01_l	%f20, %f10, %f6, %f4
785	aes_dround23_l	%f22, %f10, %f6, %f6
786	ldd		[$key + 32], %f20
787	retl
788	ldd		[$key + 40], %f22
789.type	_aes256_decrypt_2x,#function
790.size	_aes256_decrypt_2x,.-_aes256_decrypt_2x
791
792.align	32
793_aes192_decrypt_1x:
794___
795for ($i=0; $i<5; $i++) {
796    $code.=<<___;
797	aes_dround01	%f`16+8*$i+0`, %f0, %f2, %f4
798	aes_dround23	%f`16+8*$i+2`, %f0, %f2, %f2
799	aes_dround01	%f`16+8*$i+4`, %f4, %f2, %f0
800	aes_dround23	%f`16+8*$i+6`, %f4, %f2, %f2
801___
802}
803$code.=<<___;
804	aes_dround01	%f56, %f0, %f2, %f4
805	aes_dround23	%f58, %f0, %f2, %f2
806	aes_dround01_l	%f60, %f4, %f2, %f0
807	retl
808	aes_dround23_l	%f62, %f4, %f2, %f2
809.type	_aes192_decrypt_1x,#function
810.size	_aes192_decrypt_1x,.-_aes192_decrypt_1x
811
812.align	32
813_aes192_decrypt_2x:
814___
815for ($i=0; $i<5; $i++) {
816    $code.=<<___;
817	aes_dround01	%f`16+8*$i+0`, %f0, %f2, %f8
818	aes_dround23	%f`16+8*$i+2`, %f0, %f2, %f2
819	aes_dround01	%f`16+8*$i+0`, %f4, %f6, %f10
820	aes_dround23	%f`16+8*$i+2`, %f4, %f6, %f6
821	aes_dround01	%f`16+8*$i+4`, %f8, %f2, %f0
822	aes_dround23	%f`16+8*$i+6`, %f8, %f2, %f2
823	aes_dround01	%f`16+8*$i+4`, %f10, %f6, %f4
824	aes_dround23	%f`16+8*$i+6`, %f10, %f6, %f6
825___
826}
827$code.=<<___;
828	aes_dround01	%f56, %f0, %f2, %f8
829	aes_dround23	%f58, %f0, %f2, %f2
830	aes_dround01	%f56, %f4, %f6, %f10
831	aes_dround23	%f58, %f4, %f6, %f6
832	aes_dround01_l	%f60, %f8, %f2, %f0
833	aes_dround23_l	%f62, %f8, %f2, %f2
834	aes_dround01_l	%f60, %f10, %f6, %f4
835	retl
836	aes_dround23_l	%f62, %f10, %f6, %f6
837.type	_aes192_decrypt_2x,#function
838.size	_aes192_decrypt_2x,.-_aes192_decrypt_2x
839___
840}}}
841
842if (!$::evp) {
843$code.=<<___;
844.global	AES_encrypt
845AES_encrypt=aes_t4_encrypt
846.global	AES_decrypt
847AES_decrypt=aes_t4_decrypt
848.global	AES_set_encrypt_key
849.align	32
850AES_set_encrypt_key:
851	andcc		%o2, 7, %g0		! check alignment
852	bnz,a,pn	%icc, 1f
853	mov		-1, %o0
854	brz,a,pn	%o0, 1f
855	mov		-1, %o0
856	brz,a,pn	%o2, 1f
857	mov		-1, %o0
858	andncc		%o1, 0x1c0, %g0
859	bnz,a,pn	%icc, 1f
860	mov		-2, %o0
861	cmp		%o1, 128
862	bl,a,pn		%icc, 1f
863	mov		-2, %o0
864	b		aes_t4_set_encrypt_key
865	nop
8661:	retl
867	nop
868.type	AES_set_encrypt_key,#function
869.size	AES_set_encrypt_key,.-AES_set_encrypt_key
870
871.global	AES_set_decrypt_key
872.align	32
873AES_set_decrypt_key:
874	andcc		%o2, 7, %g0		! check alignment
875	bnz,a,pn	%icc, 1f
876	mov		-1, %o0
877	brz,a,pn	%o0, 1f
878	mov		-1, %o0
879	brz,a,pn	%o2, 1f
880	mov		-1, %o0
881	andncc		%o1, 0x1c0, %g0
882	bnz,a,pn	%icc, 1f
883	mov		-2, %o0
884	cmp		%o1, 128
885	bl,a,pn		%icc, 1f
886	mov		-2, %o0
887	b		aes_t4_set_decrypt_key
888	nop
8891:	retl
890	nop
891.type	AES_set_decrypt_key,#function
892.size	AES_set_decrypt_key,.-AES_set_decrypt_key
893___
894
895my ($inp,$out,$len,$key,$ivec,$enc)=map("%o$_",(0..5));
896
897$code.=<<___;
898.globl	AES_cbc_encrypt
899.align	32
900AES_cbc_encrypt:
901	ld		[$key + 240], %g1
902	nop
903	brz		$enc, .Lcbc_decrypt
904	cmp		%g1, 12
905
906	bl,pt		%icc, aes128_t4_cbc_encrypt
907	nop
908	be,pn		%icc, aes192_t4_cbc_encrypt
909	nop
910	ba		aes256_t4_cbc_encrypt
911	nop
912
913.Lcbc_decrypt:
914	bl,pt		%icc, aes128_t4_cbc_decrypt
915	nop
916	be,pn		%icc, aes192_t4_cbc_decrypt
917	nop
918	ba		aes256_t4_cbc_decrypt
919	nop
920.type	AES_cbc_encrypt,#function
921.size	AES_cbc_encrypt,.-AES_cbc_encrypt
922___
923}
924$code.=<<___;
925.asciz	"AES for SPARC T4, David S. Miller, Andy Polyakov"
926.align	4
927___
928
929&emit_assembler();
930
931close STDOUT or die "error closing STDOUT: $!";
932