1########################################################################
2# Implement fast SHA-512 with AVX2 instructions. (x86_64)
3#
4# Copyright (C) 2013 Intel Corporation.
5#
6# Authors:
7#     James Guilford <james.guilford@intel.com>
8#     Kirk Yap <kirk.s.yap@intel.com>
9#     David Cote <david.m.cote@intel.com>
10#     Tim Chen <tim.c.chen@linux.intel.com>
11#
12# This software is available to you under a choice of one of two
13# licenses.  You may choose to be licensed under the terms of the GNU
14# General Public License (GPL) Version 2, available from the file
15# COPYING in the main directory of this source tree, or the
16# OpenIB.org BSD license below:
17#
18#     Redistribution and use in source and binary forms, with or
19#     without modification, are permitted provided that the following
20#     conditions are met:
21#
22#      - Redistributions of source code must retain the above
23#        copyright notice, this list of conditions and the following
24#        disclaimer.
25#
26#      - Redistributions in binary form must reproduce the above
27#        copyright notice, this list of conditions and the following
28#        disclaimer in the documentation and/or other materials
29#        provided with the distribution.
30#
31# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
32# EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
33# MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
34# NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
35# BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
36# ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
37# CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
38# SOFTWARE.
39#
40########################################################################
41#
42# This code is described in an Intel White-Paper:
43# "Fast SHA-512 Implementations on Intel Architecture Processors"
44#
45# To find it, surf to http://www.intel.com/p/en_US/embedded
46# and search for that title.
47#
48########################################################################
49# This code schedules 1 blocks at a time, with 4 lanes per block
50########################################################################
51
52#ifdef CONFIG_AS_AVX2
53#include <linux/linkage.h>
54
55.text
56
57# Virtual Registers
58Y_0 = %ymm4
59Y_1 = %ymm5
60Y_2 = %ymm6
61Y_3 = %ymm7
62
63YTMP0 = %ymm0
64YTMP1 = %ymm1
65YTMP2 = %ymm2
66YTMP3 = %ymm3
67YTMP4 = %ymm8
68XFER  = YTMP0
69
70BYTE_FLIP_MASK  = %ymm9
71
72# 1st arg is %rdi, which is saved to the stack and accessed later via %r12
73CTX1        = %rdi
74CTX2        = %r12
75# 2nd arg
76INP         = %rsi
77# 3rd arg
78NUM_BLKS    = %rdx
79
80c           = %rcx
81d           = %r8
82e           = %rdx
83y3          = %rsi
84
85TBL   = %rdi # clobbers CTX1
86
87a     = %rax
88b     = %rbx
89
90f     = %r9
91g     = %r10
92h     = %r11
93old_h = %r11
94
95T1    = %r12 # clobbers CTX2
96y0    = %r13
97y1    = %r14
98y2    = %r15
99
100# Local variables (stack frame)
101XFER_SIZE = 4*8
102SRND_SIZE = 1*8
103INP_SIZE = 1*8
104INPEND_SIZE = 1*8
105CTX_SIZE = 1*8
106RSPSAVE_SIZE = 1*8
107GPRSAVE_SIZE = 5*8
108
109frame_XFER = 0
110frame_SRND = frame_XFER + XFER_SIZE
111frame_INP = frame_SRND + SRND_SIZE
112frame_INPEND = frame_INP + INP_SIZE
113frame_CTX = frame_INPEND + INPEND_SIZE
114frame_RSPSAVE = frame_CTX + CTX_SIZE
115frame_GPRSAVE = frame_RSPSAVE + RSPSAVE_SIZE
116frame_size = frame_GPRSAVE + GPRSAVE_SIZE
117
118## assume buffers not aligned
119#define	VMOVDQ vmovdqu
120
121# addm [mem], reg
122# Add reg to mem using reg-mem add and store
123.macro addm p1 p2
124	add	\p1, \p2
125	mov	\p2, \p1
126.endm
127
128
129# COPY_YMM_AND_BSWAP ymm, [mem], byte_flip_mask
130# Load ymm with mem and byte swap each dword
131.macro COPY_YMM_AND_BSWAP p1 p2 p3
132	VMOVDQ \p2, \p1
133	vpshufb \p3, \p1, \p1
134.endm
135# rotate_Ys
136# Rotate values of symbols Y0...Y3
137.macro rotate_Ys
138	Y_ = Y_0
139	Y_0 = Y_1
140	Y_1 = Y_2
141	Y_2 = Y_3
142	Y_3 = Y_
143.endm
144
145# RotateState
146.macro RotateState
147	# Rotate symbols a..h right
148	old_h  = h
149	TMP_   = h
150	h      = g
151	g      = f
152	f      = e
153	e      = d
154	d      = c
155	c      = b
156	b      = a
157	a      = TMP_
158.endm
159
160# macro MY_VPALIGNR	YDST, YSRC1, YSRC2, RVAL
161# YDST = {YSRC1, YSRC2} >> RVAL*8
162.macro MY_VPALIGNR YDST YSRC1 YSRC2 RVAL
163	vperm2f128      $0x3, \YSRC2, \YSRC1, \YDST     # YDST = {YS1_LO, YS2_HI}
164	vpalignr        $\RVAL, \YSRC2, \YDST, \YDST    # YDST = {YDS1, YS2} >> RVAL*8
165.endm
166
167.macro FOUR_ROUNDS_AND_SCHED
168################################### RND N + 0 #########################################
169
170	# Extract w[t-7]
171	MY_VPALIGNR	YTMP0, Y_3, Y_2, 8		# YTMP0 = W[-7]
172	# Calculate w[t-16] + w[t-7]
173	vpaddq		Y_0, YTMP0, YTMP0		# YTMP0 = W[-7] + W[-16]
174	# Extract w[t-15]
175	MY_VPALIGNR	YTMP1, Y_1, Y_0, 8		# YTMP1 = W[-15]
176
177	# Calculate sigma0
178
179	# Calculate w[t-15] ror 1
180	vpsrlq		$1, YTMP1, YTMP2
181	vpsllq		$(64-1), YTMP1, YTMP3
182	vpor		YTMP2, YTMP3, YTMP3		# YTMP3 = W[-15] ror 1
183	# Calculate w[t-15] shr 7
184	vpsrlq		$7, YTMP1, YTMP4		# YTMP4 = W[-15] >> 7
185
186	mov	a, y3		# y3 = a                                # MAJA
187	rorx	$41, e, y0	# y0 = e >> 41				# S1A
188	rorx	$18, e, y1	# y1 = e >> 18				# S1B
189	add	frame_XFER(%rsp),h		# h = k + w + h         # --
190	or	c, y3		# y3 = a|c                              # MAJA
191	mov	f, y2		# y2 = f                                # CH
192	rorx	$34, a, T1	# T1 = a >> 34				# S0B
193
194	xor	y1, y0		# y0 = (e>>41) ^ (e>>18)		# S1
195	xor	g, y2		# y2 = f^g                              # CH
196	rorx	$14, e, y1	# y1 = (e >> 14)			# S1
197
198	and	e, y2		# y2 = (f^g)&e                          # CH
199	xor	y1, y0		# y0 = (e>>41) ^ (e>>18) ^ (e>>14)	# S1
200	rorx	$39, a, y1	# y1 = a >> 39				# S0A
201	add	h, d		# d = k + w + h + d                     # --
202
203	and	b, y3		# y3 = (a|c)&b                          # MAJA
204	xor	T1, y1		# y1 = (a>>39) ^ (a>>34)		# S0
205	rorx	$28, a, T1	# T1 = (a >> 28)			# S0
206
207	xor	g, y2		# y2 = CH = ((f^g)&e)^g                 # CH
208	xor	T1, y1		# y1 = (a>>39) ^ (a>>34) ^ (a>>28)	# S0
209	mov	a, T1		# T1 = a                                # MAJB
210	and	c, T1		# T1 = a&c                              # MAJB
211
212	add	y0, y2		# y2 = S1 + CH                          # --
213	or	T1, y3		# y3 = MAJ = (a|c)&b)|(a&c)             # MAJ
214	add	y1, h		# h = k + w + h + S0                    # --
215
216	add	y2, d		# d = k + w + h + d + S1 + CH = d + t1  # --
217
218	add	y2, h		# h = k + w + h + S0 + S1 + CH = t1 + S0# --
219	add	y3, h		# h = t1 + S0 + MAJ                     # --
220
221	RotateState
222
223################################### RND N + 1 #########################################
224
225	# Calculate w[t-15] ror 8
226	vpsrlq		$8, YTMP1, YTMP2
227	vpsllq		$(64-8), YTMP1, YTMP1
228	vpor		YTMP2, YTMP1, YTMP1		# YTMP1 = W[-15] ror 8
229	# XOR the three components
230	vpxor		YTMP4, YTMP3, YTMP3		# YTMP3 = W[-15] ror 1 ^ W[-15] >> 7
231	vpxor		YTMP1, YTMP3, YTMP1		# YTMP1 = s0
232
233
234	# Add three components, w[t-16], w[t-7] and sigma0
235	vpaddq		YTMP1, YTMP0, YTMP0		# YTMP0 = W[-16] + W[-7] + s0
236	# Move to appropriate lanes for calculating w[16] and w[17]
237	vperm2f128	$0x0, YTMP0, YTMP0, Y_0		# Y_0 = W[-16] + W[-7] + s0 {BABA}
238	# Move to appropriate lanes for calculating w[18] and w[19]
239	vpand		MASK_YMM_LO(%rip), YTMP0, YTMP0	# YTMP0 = W[-16] + W[-7] + s0 {DC00}
240
241	# Calculate w[16] and w[17] in both 128 bit lanes
242
243	# Calculate sigma1 for w[16] and w[17] on both 128 bit lanes
244	vperm2f128	$0x11, Y_3, Y_3, YTMP2		# YTMP2 = W[-2] {BABA}
245	vpsrlq		$6, YTMP2, YTMP4		# YTMP4 = W[-2] >> 6 {BABA}
246
247
248	mov	a, y3		# y3 = a                                # MAJA
249	rorx	$41, e, y0	# y0 = e >> 41				# S1A
250	rorx	$18, e, y1	# y1 = e >> 18				# S1B
251	add	1*8+frame_XFER(%rsp), h		# h = k + w + h         # --
252	or	c, y3		# y3 = a|c                              # MAJA
253
254
255	mov	f, y2		# y2 = f                                # CH
256	rorx	$34, a, T1	# T1 = a >> 34				# S0B
257	xor	y1, y0		# y0 = (e>>41) ^ (e>>18)		# S1
258	xor	g, y2		# y2 = f^g                              # CH
259
260
261	rorx	$14, e, y1	# y1 = (e >> 14)			# S1
262	xor	y1, y0		# y0 = (e>>41) ^ (e>>18) ^ (e>>14)	# S1
263	rorx	$39, a, y1	# y1 = a >> 39				# S0A
264	and	e, y2		# y2 = (f^g)&e                          # CH
265	add	h, d		# d = k + w + h + d                     # --
266
267	and	b, y3		# y3 = (a|c)&b                          # MAJA
268	xor	T1, y1		# y1 = (a>>39) ^ (a>>34)		# S0
269
270	rorx	$28, a, T1	# T1 = (a >> 28)			# S0
271	xor	g, y2		# y2 = CH = ((f^g)&e)^g                 # CH
272
273	xor	T1, y1		# y1 = (a>>39) ^ (a>>34) ^ (a>>28)	# S0
274	mov	a, T1		# T1 = a                                # MAJB
275	and	c, T1		# T1 = a&c                              # MAJB
276	add	y0, y2		# y2 = S1 + CH                          # --
277
278	or	T1, y3		# y3 = MAJ = (a|c)&b)|(a&c)             # MAJ
279	add	y1, h		# h = k + w + h + S0                    # --
280
281	add	y2, d		# d = k + w + h + d + S1 + CH = d + t1  # --
282	add	y2, h		# h = k + w + h + S0 + S1 + CH = t1 + S0# --
283	add	y3, h		# h = t1 + S0 + MAJ                     # --
284
285	RotateState
286
287
288################################### RND N + 2 #########################################
289
290	vpsrlq		$19, YTMP2, YTMP3		# YTMP3 = W[-2] >> 19 {BABA}
291	vpsllq		$(64-19), YTMP2, YTMP1		# YTMP1 = W[-2] << 19 {BABA}
292	vpor		YTMP1, YTMP3, YTMP3		# YTMP3 = W[-2] ror 19 {BABA}
293	vpxor		YTMP3, YTMP4, YTMP4		# YTMP4 = W[-2] ror 19 ^ W[-2] >> 6 {BABA}
294	vpsrlq		$61, YTMP2, YTMP3		# YTMP3 = W[-2] >> 61 {BABA}
295	vpsllq		$(64-61), YTMP2, YTMP1		# YTMP1 = W[-2] << 61 {BABA}
296	vpor		YTMP1, YTMP3, YTMP3		# YTMP3 = W[-2] ror 61 {BABA}
297	vpxor		YTMP3, YTMP4, YTMP4		# YTMP4 = s1 = (W[-2] ror 19) ^
298							#  (W[-2] ror 61) ^ (W[-2] >> 6) {BABA}
299
300	# Add sigma1 to the other compunents to get w[16] and w[17]
301	vpaddq		YTMP4, Y_0, Y_0			# Y_0 = {W[1], W[0], W[1], W[0]}
302
303	# Calculate sigma1 for w[18] and w[19] for upper 128 bit lane
304	vpsrlq		$6, Y_0, YTMP4			# YTMP4 = W[-2] >> 6 {DC--}
305
306	mov	a, y3		# y3 = a                                # MAJA
307	rorx	$41, e, y0	# y0 = e >> 41				# S1A
308	add	2*8+frame_XFER(%rsp), h		# h = k + w + h         # --
309
310	rorx	$18, e, y1	# y1 = e >> 18				# S1B
311	or	c, y3		# y3 = a|c                              # MAJA
312	mov	f, y2		# y2 = f                                # CH
313	xor	g, y2		# y2 = f^g                              # CH
314
315	rorx	$34, a, T1	# T1 = a >> 34				# S0B
316	xor	y1, y0		# y0 = (e>>41) ^ (e>>18)		# S1
317	and	e, y2		# y2 = (f^g)&e                          # CH
318
319	rorx	$14, e, y1	# y1 = (e >> 14)			# S1
320	add	h, d		# d = k + w + h + d                     # --
321	and	b, y3		# y3 = (a|c)&b                          # MAJA
322
323	xor	y1, y0		# y0 = (e>>41) ^ (e>>18) ^ (e>>14)	# S1
324	rorx	$39, a, y1	# y1 = a >> 39				# S0A
325	xor	g, y2		# y2 = CH = ((f^g)&e)^g                 # CH
326
327	xor	T1, y1		# y1 = (a>>39) ^ (a>>34)		# S0
328	rorx	$28, a, T1	# T1 = (a >> 28)			# S0
329
330	xor	T1, y1		# y1 = (a>>39) ^ (a>>34) ^ (a>>28)	# S0
331	mov	a, T1		# T1 = a                                # MAJB
332	and	c, T1		# T1 = a&c                              # MAJB
333	add	y0, y2		# y2 = S1 + CH                          # --
334
335	or	T1, y3		# y3 = MAJ = (a|c)&b)|(a&c)             # MAJ
336	add	y1, h		# h = k + w + h + S0                    # --
337	add	y2, d		# d = k + w + h + d + S1 + CH = d + t1  # --
338	add	y2, h		# h = k + w + h + S0 + S1 + CH = t1 + S0# --
339
340	add	y3, h		# h = t1 + S0 + MAJ                     # --
341
342	RotateState
343
344################################### RND N + 3 #########################################
345
346	vpsrlq		$19, Y_0, YTMP3			# YTMP3 = W[-2] >> 19 {DC--}
347	vpsllq		$(64-19), Y_0, YTMP1		# YTMP1 = W[-2] << 19 {DC--}
348	vpor		YTMP1, YTMP3, YTMP3		# YTMP3 = W[-2] ror 19 {DC--}
349	vpxor		YTMP3, YTMP4, YTMP4		# YTMP4 = W[-2] ror 19 ^ W[-2] >> 6 {DC--}
350	vpsrlq		$61, Y_0, YTMP3			# YTMP3 = W[-2] >> 61 {DC--}
351	vpsllq		$(64-61), Y_0, YTMP1		# YTMP1 = W[-2] << 61 {DC--}
352	vpor		YTMP1, YTMP3, YTMP3		# YTMP3 = W[-2] ror 61 {DC--}
353	vpxor		YTMP3, YTMP4, YTMP4		# YTMP4 = s1 = (W[-2] ror 19) ^
354							#  (W[-2] ror 61) ^ (W[-2] >> 6) {DC--}
355
356	# Add the sigma0 + w[t-7] + w[t-16] for w[18] and w[19]
357	# to newly calculated sigma1 to get w[18] and w[19]
358	vpaddq		YTMP4, YTMP0, YTMP2		# YTMP2 = {W[3], W[2], --, --}
359
360	# Form w[19, w[18], w17], w[16]
361	vpblendd		$0xF0, YTMP2, Y_0, Y_0		# Y_0 = {W[3], W[2], W[1], W[0]}
362
363	mov	a, y3		# y3 = a                                # MAJA
364	rorx	$41, e, y0	# y0 = e >> 41				# S1A
365	rorx	$18, e, y1	# y1 = e >> 18				# S1B
366	add	3*8+frame_XFER(%rsp), h		# h = k + w + h         # --
367	or	c, y3		# y3 = a|c                              # MAJA
368
369
370	mov	f, y2		# y2 = f                                # CH
371	rorx	$34, a, T1	# T1 = a >> 34				# S0B
372	xor	y1, y0		# y0 = (e>>41) ^ (e>>18)		# S1
373	xor	g, y2		# y2 = f^g                              # CH
374
375
376	rorx	$14, e, y1	# y1 = (e >> 14)			# S1
377	and	e, y2		# y2 = (f^g)&e                          # CH
378	add	h, d		# d = k + w + h + d                     # --
379	and	b, y3		# y3 = (a|c)&b                          # MAJA
380
381	xor	y1, y0		# y0 = (e>>41) ^ (e>>18) ^ (e>>14)	# S1
382	xor	g, y2		# y2 = CH = ((f^g)&e)^g                 # CH
383
384	rorx	$39, a, y1	# y1 = a >> 39				# S0A
385	add	y0, y2		# y2 = S1 + CH                          # --
386
387	xor	T1, y1		# y1 = (a>>39) ^ (a>>34)		# S0
388	add	y2, d		# d = k + w + h + d + S1 + CH = d + t1  # --
389
390	rorx	$28, a, T1	# T1 = (a >> 28)			# S0
391
392	xor	T1, y1		# y1 = (a>>39) ^ (a>>34) ^ (a>>28)	# S0
393	mov	a, T1		# T1 = a                                # MAJB
394	and	c, T1		# T1 = a&c                              # MAJB
395	or	T1, y3		# y3 = MAJ = (a|c)&b)|(a&c)             # MAJ
396
397	add	y1, h		# h = k + w + h + S0                    # --
398	add	y2, h		# h = k + w + h + S0 + S1 + CH = t1 + S0# --
399	add	y3, h		# h = t1 + S0 + MAJ                     # --
400
401	RotateState
402
403	rotate_Ys
404.endm
405
406.macro DO_4ROUNDS
407
408################################### RND N + 0 #########################################
409
410	mov	f, y2		# y2 = f                                # CH
411	rorx	$41, e, y0	# y0 = e >> 41				# S1A
412	rorx	$18, e, y1	# y1 = e >> 18				# S1B
413	xor	g, y2		# y2 = f^g                              # CH
414
415	xor	y1, y0		# y0 = (e>>41) ^ (e>>18)		# S1
416	rorx	$14, e, y1	# y1 = (e >> 14)			# S1
417	and	e, y2		# y2 = (f^g)&e                          # CH
418
419	xor	y1, y0		# y0 = (e>>41) ^ (e>>18) ^ (e>>14)	# S1
420	rorx	$34, a, T1	# T1 = a >> 34				# S0B
421	xor	g, y2		# y2 = CH = ((f^g)&e)^g                 # CH
422	rorx	$39, a, y1	# y1 = a >> 39				# S0A
423	mov	a, y3		# y3 = a                                # MAJA
424
425	xor	T1, y1		# y1 = (a>>39) ^ (a>>34)		# S0
426	rorx	$28, a, T1	# T1 = (a >> 28)			# S0
427	add	frame_XFER(%rsp), h		# h = k + w + h         # --
428	or	c, y3		# y3 = a|c                              # MAJA
429
430	xor	T1, y1		# y1 = (a>>39) ^ (a>>34) ^ (a>>28)	# S0
431	mov	a, T1		# T1 = a                                # MAJB
432	and	b, y3		# y3 = (a|c)&b                          # MAJA
433	and	c, T1		# T1 = a&c                              # MAJB
434	add	y0, y2		# y2 = S1 + CH                          # --
435
436	add	h, d		# d = k + w + h + d                     # --
437	or	T1, y3		# y3 = MAJ = (a|c)&b)|(a&c)             # MAJ
438	add	y1, h		# h = k + w + h + S0                    # --
439
440	add	y2, d		# d = k + w + h + d + S1 + CH = d + t1  # --
441
442	RotateState
443
444################################### RND N + 1 #########################################
445
446	add	y2, old_h	# h = k + w + h + S0 + S1 + CH = t1 + S0# --
447	mov	f, y2		# y2 = f                                # CH
448	rorx	$41, e, y0	# y0 = e >> 41				# S1A
449	rorx	$18, e, y1	# y1 = e >> 18				# S1B
450	xor	g, y2		# y2 = f^g                              # CH
451
452	xor	y1, y0		# y0 = (e>>41) ^ (e>>18)		# S1
453	rorx	$14, e, y1	# y1 = (e >> 14)			# S1
454	and	e, y2		# y2 = (f^g)&e                          # CH
455	add	y3, old_h	# h = t1 + S0 + MAJ                     # --
456
457	xor	y1, y0		# y0 = (e>>41) ^ (e>>18) ^ (e>>14)	# S1
458	rorx	$34, a, T1	# T1 = a >> 34				# S0B
459	xor	g, y2		# y2 = CH = ((f^g)&e)^g                 # CH
460	rorx	$39, a, y1	# y1 = a >> 39				# S0A
461	mov	a, y3		# y3 = a                                # MAJA
462
463	xor	T1, y1		# y1 = (a>>39) ^ (a>>34)		# S0
464	rorx	$28, a, T1	# T1 = (a >> 28)			# S0
465	add	8*1+frame_XFER(%rsp), h		# h = k + w + h         # --
466	or	c, y3		# y3 = a|c                              # MAJA
467
468	xor	T1, y1		# y1 = (a>>39) ^ (a>>34) ^ (a>>28)	# S0
469	mov	a, T1		# T1 = a                                # MAJB
470	and	b, y3		# y3 = (a|c)&b                          # MAJA
471	and	c, T1		# T1 = a&c                              # MAJB
472	add	y0, y2		# y2 = S1 + CH                          # --
473
474	add	h, d		# d = k + w + h + d                     # --
475	or	T1, y3		# y3 = MAJ = (a|c)&b)|(a&c)             # MAJ
476	add	y1, h		# h = k + w + h + S0                    # --
477
478	add	y2, d		# d = k + w + h + d + S1 + CH = d + t1  # --
479
480	RotateState
481
482################################### RND N + 2 #########################################
483
484	add	y2, old_h	# h = k + w + h + S0 + S1 + CH = t1 + S0# --
485	mov	f, y2		# y2 = f                                # CH
486	rorx	$41, e, y0	# y0 = e >> 41				# S1A
487	rorx	$18, e, y1	# y1 = e >> 18				# S1B
488	xor	g, y2		# y2 = f^g                              # CH
489
490	xor	y1, y0		# y0 = (e>>41) ^ (e>>18)		# S1
491	rorx	$14, e, y1	# y1 = (e >> 14)			# S1
492	and	e, y2		# y2 = (f^g)&e                          # CH
493	add	y3, old_h	# h = t1 + S0 + MAJ                     # --
494
495	xor	y1, y0		# y0 = (e>>41) ^ (e>>18) ^ (e>>14)	# S1
496	rorx	$34, a, T1	# T1 = a >> 34				# S0B
497	xor	g, y2		# y2 = CH = ((f^g)&e)^g                 # CH
498	rorx	$39, a, y1	# y1 = a >> 39				# S0A
499	mov	a, y3		# y3 = a                                # MAJA
500
501	xor	T1, y1		# y1 = (a>>39) ^ (a>>34)		# S0
502	rorx	$28, a, T1	# T1 = (a >> 28)			# S0
503	add	8*2+frame_XFER(%rsp), h		# h = k + w + h         # --
504	or	c, y3		# y3 = a|c                              # MAJA
505
506	xor	T1, y1		# y1 = (a>>39) ^ (a>>34) ^ (a>>28)	# S0
507	mov	a, T1		# T1 = a                                # MAJB
508	and	b, y3		# y3 = (a|c)&b                          # MAJA
509	and	c, T1		# T1 = a&c                              # MAJB
510	add	y0, y2		# y2 = S1 + CH                          # --
511
512	add	h, d		# d = k + w + h + d                     # --
513	or	T1, y3		# y3 = MAJ = (a|c)&b)|(a&c)             # MAJ
514	add	y1, h		# h = k + w + h + S0                    # --
515
516	add	y2, d		# d = k + w + h + d + S1 + CH = d + t1  # --
517
518	RotateState
519
520################################### RND N + 3 #########################################
521
522	add	y2, old_h	# h = k + w + h + S0 + S1 + CH = t1 + S0# --
523	mov	f, y2		# y2 = f                                # CH
524	rorx	$41, e, y0	# y0 = e >> 41				# S1A
525	rorx	$18, e, y1	# y1 = e >> 18				# S1B
526	xor	g, y2		# y2 = f^g                              # CH
527
528	xor	y1, y0		# y0 = (e>>41) ^ (e>>18)		# S1
529	rorx	$14, e, y1	# y1 = (e >> 14)			# S1
530	and	e, y2		# y2 = (f^g)&e                          # CH
531	add	y3, old_h	# h = t1 + S0 + MAJ                     # --
532
533	xor	y1, y0		# y0 = (e>>41) ^ (e>>18) ^ (e>>14)	# S1
534	rorx	$34, a, T1	# T1 = a >> 34				# S0B
535	xor	g, y2		# y2 = CH = ((f^g)&e)^g                 # CH
536	rorx	$39, a, y1	# y1 = a >> 39				# S0A
537	mov	a, y3		# y3 = a                                # MAJA
538
539	xor	T1, y1		# y1 = (a>>39) ^ (a>>34)		# S0
540	rorx	$28, a, T1	# T1 = (a >> 28)			# S0
541	add	8*3+frame_XFER(%rsp), h		# h = k + w + h         # --
542	or	c, y3		# y3 = a|c                              # MAJA
543
544	xor	T1, y1		# y1 = (a>>39) ^ (a>>34) ^ (a>>28)	# S0
545	mov	a, T1		# T1 = a                                # MAJB
546	and	b, y3		# y3 = (a|c)&b                          # MAJA
547	and	c, T1		# T1 = a&c                              # MAJB
548	add	y0, y2		# y2 = S1 + CH                          # --
549
550
551	add	h, d		# d = k + w + h + d                     # --
552	or	T1, y3		# y3 = MAJ = (a|c)&b)|(a&c)             # MAJ
553	add	y1, h		# h = k + w + h + S0                    # --
554
555	add	y2, d		# d = k + w + h + d + S1 + CH = d + t1  # --
556
557	add	y2, h		# h = k + w + h + S0 + S1 + CH = t1 + S0# --
558
559	add	y3, h		# h = t1 + S0 + MAJ                     # --
560
561	RotateState
562
563.endm
564
565########################################################################
566# void sha512_transform_rorx(void* D, const void* M, uint64_t L)#
567# Purpose: Updates the SHA512 digest stored at D with the message stored in M.
568# The size of the message pointed to by M must be an integer multiple of SHA512
569#   message blocks.
570# L is the message length in SHA512 blocks
571########################################################################
572ENTRY(sha512_transform_rorx)
573	# Allocate Stack Space
574	mov	%rsp, %rax
575	sub	$frame_size, %rsp
576	and	$~(0x20 - 1), %rsp
577	mov	%rax, frame_RSPSAVE(%rsp)
578
579	# Save GPRs
580	mov	%rbx, 8*0+frame_GPRSAVE(%rsp)
581	mov	%r12, 8*1+frame_GPRSAVE(%rsp)
582	mov	%r13, 8*2+frame_GPRSAVE(%rsp)
583	mov	%r14, 8*3+frame_GPRSAVE(%rsp)
584	mov	%r15, 8*4+frame_GPRSAVE(%rsp)
585
586	shl	$7, NUM_BLKS	# convert to bytes
587	jz	done_hash
588	add	INP, NUM_BLKS	# pointer to end of data
589	mov	NUM_BLKS, frame_INPEND(%rsp)
590
591	## load initial digest
592	mov	8*0(CTX1), a
593	mov	8*1(CTX1), b
594	mov	8*2(CTX1), c
595	mov	8*3(CTX1), d
596	mov	8*4(CTX1), e
597	mov	8*5(CTX1), f
598	mov	8*6(CTX1), g
599	mov	8*7(CTX1), h
600
601	# save %rdi (CTX) before it gets clobbered
602	mov	%rdi, frame_CTX(%rsp)
603
604	vmovdqa	PSHUFFLE_BYTE_FLIP_MASK(%rip), BYTE_FLIP_MASK
605
606loop0:
607	lea	K512(%rip), TBL
608
609	## byte swap first 16 dwords
610	COPY_YMM_AND_BSWAP	Y_0, (INP), BYTE_FLIP_MASK
611	COPY_YMM_AND_BSWAP	Y_1, 1*32(INP), BYTE_FLIP_MASK
612	COPY_YMM_AND_BSWAP	Y_2, 2*32(INP), BYTE_FLIP_MASK
613	COPY_YMM_AND_BSWAP	Y_3, 3*32(INP), BYTE_FLIP_MASK
614
615	mov	INP, frame_INP(%rsp)
616
617	## schedule 64 input dwords, by doing 12 rounds of 4 each
618	movq	$4, frame_SRND(%rsp)
619
620.align 16
621loop1:
622	vpaddq	(TBL), Y_0, XFER
623	vmovdqa XFER, frame_XFER(%rsp)
624	FOUR_ROUNDS_AND_SCHED
625
626	vpaddq	1*32(TBL), Y_0, XFER
627	vmovdqa XFER, frame_XFER(%rsp)
628	FOUR_ROUNDS_AND_SCHED
629
630	vpaddq	2*32(TBL), Y_0, XFER
631	vmovdqa XFER, frame_XFER(%rsp)
632	FOUR_ROUNDS_AND_SCHED
633
634	vpaddq	3*32(TBL), Y_0, XFER
635	vmovdqa XFER, frame_XFER(%rsp)
636	add	$(4*32), TBL
637	FOUR_ROUNDS_AND_SCHED
638
639	subq	$1, frame_SRND(%rsp)
640	jne	loop1
641
642	movq	$2, frame_SRND(%rsp)
643loop2:
644	vpaddq	(TBL), Y_0, XFER
645	vmovdqa XFER, frame_XFER(%rsp)
646	DO_4ROUNDS
647	vpaddq	1*32(TBL), Y_1, XFER
648	vmovdqa XFER, frame_XFER(%rsp)
649	add	$(2*32), TBL
650	DO_4ROUNDS
651
652	vmovdqa	Y_2, Y_0
653	vmovdqa	Y_3, Y_1
654
655	subq	$1, frame_SRND(%rsp)
656	jne	loop2
657
658	mov	frame_CTX(%rsp), CTX2
659	addm	8*0(CTX2), a
660	addm	8*1(CTX2), b
661	addm	8*2(CTX2), c
662	addm	8*3(CTX2), d
663	addm	8*4(CTX2), e
664	addm	8*5(CTX2), f
665	addm	8*6(CTX2), g
666	addm	8*7(CTX2), h
667
668	mov	frame_INP(%rsp), INP
669	add	$128, INP
670	cmp	frame_INPEND(%rsp), INP
671	jne	loop0
672
673done_hash:
674
675# Restore GPRs
676	mov	8*0+frame_GPRSAVE(%rsp), %rbx
677	mov	8*1+frame_GPRSAVE(%rsp), %r12
678	mov	8*2+frame_GPRSAVE(%rsp), %r13
679	mov	8*3+frame_GPRSAVE(%rsp), %r14
680	mov	8*4+frame_GPRSAVE(%rsp), %r15
681
682	# Restore Stack Pointer
683	mov	frame_RSPSAVE(%rsp), %rsp
684	ret
685ENDPROC(sha512_transform_rorx)
686
687########################################################################
688### Binary Data
689
690
691# Mergeable 640-byte rodata section. This allows linker to merge the table
692# with other, exactly the same 640-byte fragment of another rodata section
693# (if such section exists).
694.section	.rodata.cst640.K512, "aM", @progbits, 640
695.align 64
696# K[t] used in SHA512 hashing
697K512:
698	.quad	0x428a2f98d728ae22,0x7137449123ef65cd
699	.quad	0xb5c0fbcfec4d3b2f,0xe9b5dba58189dbbc
700	.quad	0x3956c25bf348b538,0x59f111f1b605d019
701	.quad	0x923f82a4af194f9b,0xab1c5ed5da6d8118
702	.quad	0xd807aa98a3030242,0x12835b0145706fbe
703	.quad	0x243185be4ee4b28c,0x550c7dc3d5ffb4e2
704	.quad	0x72be5d74f27b896f,0x80deb1fe3b1696b1
705	.quad	0x9bdc06a725c71235,0xc19bf174cf692694
706	.quad	0xe49b69c19ef14ad2,0xefbe4786384f25e3
707	.quad	0x0fc19dc68b8cd5b5,0x240ca1cc77ac9c65
708	.quad	0x2de92c6f592b0275,0x4a7484aa6ea6e483
709	.quad	0x5cb0a9dcbd41fbd4,0x76f988da831153b5
710	.quad	0x983e5152ee66dfab,0xa831c66d2db43210
711	.quad	0xb00327c898fb213f,0xbf597fc7beef0ee4
712	.quad	0xc6e00bf33da88fc2,0xd5a79147930aa725
713	.quad	0x06ca6351e003826f,0x142929670a0e6e70
714	.quad	0x27b70a8546d22ffc,0x2e1b21385c26c926
715	.quad	0x4d2c6dfc5ac42aed,0x53380d139d95b3df
716	.quad	0x650a73548baf63de,0x766a0abb3c77b2a8
717	.quad	0x81c2c92e47edaee6,0x92722c851482353b
718	.quad	0xa2bfe8a14cf10364,0xa81a664bbc423001
719	.quad	0xc24b8b70d0f89791,0xc76c51a30654be30
720	.quad	0xd192e819d6ef5218,0xd69906245565a910
721	.quad	0xf40e35855771202a,0x106aa07032bbd1b8
722	.quad	0x19a4c116b8d2d0c8,0x1e376c085141ab53
723	.quad	0x2748774cdf8eeb99,0x34b0bcb5e19b48a8
724	.quad	0x391c0cb3c5c95a63,0x4ed8aa4ae3418acb
725	.quad	0x5b9cca4f7763e373,0x682e6ff3d6b2b8a3
726	.quad	0x748f82ee5defb2fc,0x78a5636f43172f60
727	.quad	0x84c87814a1f0ab72,0x8cc702081a6439ec
728	.quad	0x90befffa23631e28,0xa4506cebde82bde9
729	.quad	0xbef9a3f7b2c67915,0xc67178f2e372532b
730	.quad	0xca273eceea26619c,0xd186b8c721c0c207
731	.quad	0xeada7dd6cde0eb1e,0xf57d4f7fee6ed178
732	.quad	0x06f067aa72176fba,0x0a637dc5a2c898a6
733	.quad	0x113f9804bef90dae,0x1b710b35131c471b
734	.quad	0x28db77f523047d84,0x32caab7b40c72493
735	.quad	0x3c9ebe0a15c9bebc,0x431d67c49c100d4c
736	.quad	0x4cc5d4becb3e42b6,0x597f299cfc657e2a
737	.quad	0x5fcb6fab3ad6faec,0x6c44198c4a475817
738
739.section	.rodata.cst32.PSHUFFLE_BYTE_FLIP_MASK, "aM", @progbits, 32
740.align 32
741# Mask for byte-swapping a couple of qwords in an XMM register using (v)pshufb.
742PSHUFFLE_BYTE_FLIP_MASK:
743	.octa 0x08090a0b0c0d0e0f0001020304050607
744	.octa 0x18191a1b1c1d1e1f1011121314151617
745
746.section	.rodata.cst32.MASK_YMM_LO, "aM", @progbits, 32
747.align 32
748MASK_YMM_LO:
749	.octa 0x00000000000000000000000000000000
750	.octa 0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF
751
752#endif
753