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#include <linux/linkage.h>
53
54.text
55
56# Virtual Registers
57Y_0 = %ymm4
58Y_1 = %ymm5
59Y_2 = %ymm6
60Y_3 = %ymm7
61
62YTMP0 = %ymm0
63YTMP1 = %ymm1
64YTMP2 = %ymm2
65YTMP3 = %ymm3
66YTMP4 = %ymm8
67XFER  = YTMP0
68
69BYTE_FLIP_MASK  = %ymm9
70
71# 1st arg is %rdi, which is saved to the stack and accessed later via %r12
72CTX1        = %rdi
73CTX2        = %r12
74# 2nd arg
75INP         = %rsi
76# 3rd arg
77NUM_BLKS    = %rdx
78
79c           = %rcx
80d           = %r8
81e           = %rdx
82y3          = %rsi
83
84TBL   = %rdi # clobbers CTX1
85
86a     = %rax
87b     = %rbx
88
89f     = %r9
90g     = %r10
91h     = %r11
92old_h = %r11
93
94T1    = %r12 # clobbers CTX2
95y0    = %r13
96y1    = %r14
97y2    = %r15
98
99# Local variables (stack frame)
100XFER_SIZE = 4*8
101SRND_SIZE = 1*8
102INP_SIZE = 1*8
103INPEND_SIZE = 1*8
104CTX_SIZE = 1*8
105RSPSAVE_SIZE = 1*8
106GPRSAVE_SIZE = 5*8
107
108frame_XFER = 0
109frame_SRND = frame_XFER + XFER_SIZE
110frame_INP = frame_SRND + SRND_SIZE
111frame_INPEND = frame_INP + INP_SIZE
112frame_CTX = frame_INPEND + INPEND_SIZE
113frame_RSPSAVE = frame_CTX + CTX_SIZE
114frame_GPRSAVE = frame_RSPSAVE + RSPSAVE_SIZE
115frame_size = frame_GPRSAVE + GPRSAVE_SIZE
116
117## assume buffers not aligned
118#define	VMOVDQ vmovdqu
119
120# addm [mem], reg
121# Add reg to mem using reg-mem add and store
122.macro addm p1 p2
123	add	\p1, \p2
124	mov	\p2, \p1
125.endm
126
127
128# COPY_YMM_AND_BSWAP ymm, [mem], byte_flip_mask
129# Load ymm with mem and byte swap each dword
130.macro COPY_YMM_AND_BSWAP p1 p2 p3
131	VMOVDQ \p2, \p1
132	vpshufb \p3, \p1, \p1
133.endm
134# rotate_Ys
135# Rotate values of symbols Y0...Y3
136.macro rotate_Ys
137	Y_ = Y_0
138	Y_0 = Y_1
139	Y_1 = Y_2
140	Y_2 = Y_3
141	Y_3 = Y_
142.endm
143
144# RotateState
145.macro RotateState
146	# Rotate symbols a..h right
147	old_h  = h
148	TMP_   = h
149	h      = g
150	g      = f
151	f      = e
152	e      = d
153	d      = c
154	c      = b
155	b      = a
156	a      = TMP_
157.endm
158
159# macro MY_VPALIGNR	YDST, YSRC1, YSRC2, RVAL
160# YDST = {YSRC1, YSRC2} >> RVAL*8
161.macro MY_VPALIGNR YDST YSRC1 YSRC2 RVAL
162	vperm2f128      $0x3, \YSRC2, \YSRC1, \YDST     # YDST = {YS1_LO, YS2_HI}
163	vpalignr        $\RVAL, \YSRC2, \YDST, \YDST    # YDST = {YDS1, YS2} >> RVAL*8
164.endm
165
166.macro FOUR_ROUNDS_AND_SCHED
167################################### RND N + 0 #########################################
168
169	# Extract w[t-7]
170	MY_VPALIGNR	YTMP0, Y_3, Y_2, 8		# YTMP0 = W[-7]
171	# Calculate w[t-16] + w[t-7]
172	vpaddq		Y_0, YTMP0, YTMP0		# YTMP0 = W[-7] + W[-16]
173	# Extract w[t-15]
174	MY_VPALIGNR	YTMP1, Y_1, Y_0, 8		# YTMP1 = W[-15]
175
176	# Calculate sigma0
177
178	# Calculate w[t-15] ror 1
179	vpsrlq		$1, YTMP1, YTMP2
180	vpsllq		$(64-1), YTMP1, YTMP3
181	vpor		YTMP2, YTMP3, YTMP3		# YTMP3 = W[-15] ror 1
182	# Calculate w[t-15] shr 7
183	vpsrlq		$7, YTMP1, YTMP4		# YTMP4 = W[-15] >> 7
184
185	mov	a, y3		# y3 = a                                # MAJA
186	rorx	$41, e, y0	# y0 = e >> 41				# S1A
187	rorx	$18, e, y1	# y1 = e >> 18				# S1B
188	add	frame_XFER(%rsp),h		# h = k + w + h         # --
189	or	c, y3		# y3 = a|c                              # MAJA
190	mov	f, y2		# y2 = f                                # CH
191	rorx	$34, a, T1	# T1 = a >> 34				# S0B
192
193	xor	y1, y0		# y0 = (e>>41) ^ (e>>18)		# S1
194	xor	g, y2		# y2 = f^g                              # CH
195	rorx	$14, e, y1	# y1 = (e >> 14)			# S1
196
197	and	e, y2		# y2 = (f^g)&e                          # CH
198	xor	y1, y0		# y0 = (e>>41) ^ (e>>18) ^ (e>>14)	# S1
199	rorx	$39, a, y1	# y1 = a >> 39				# S0A
200	add	h, d		# d = k + w + h + d                     # --
201
202	and	b, y3		# y3 = (a|c)&b                          # MAJA
203	xor	T1, y1		# y1 = (a>>39) ^ (a>>34)		# S0
204	rorx	$28, a, T1	# T1 = (a >> 28)			# S0
205
206	xor	g, y2		# y2 = CH = ((f^g)&e)^g                 # CH
207	xor	T1, y1		# y1 = (a>>39) ^ (a>>34) ^ (a>>28)	# S0
208	mov	a, T1		# T1 = a                                # MAJB
209	and	c, T1		# T1 = a&c                              # MAJB
210
211	add	y0, y2		# y2 = S1 + CH                          # --
212	or	T1, y3		# y3 = MAJ = (a|c)&b)|(a&c)             # MAJ
213	add	y1, h		# h = k + w + h + S0                    # --
214
215	add	y2, d		# d = k + w + h + d + S1 + CH = d + t1  # --
216
217	add	y2, h		# h = k + w + h + S0 + S1 + CH = t1 + S0# --
218	add	y3, h		# h = t1 + S0 + MAJ                     # --
219
220	RotateState
221
222################################### RND N + 1 #########################################
223
224	# Calculate w[t-15] ror 8
225	vpsrlq		$8, YTMP1, YTMP2
226	vpsllq		$(64-8), YTMP1, YTMP1
227	vpor		YTMP2, YTMP1, YTMP1		# YTMP1 = W[-15] ror 8
228	# XOR the three components
229	vpxor		YTMP4, YTMP3, YTMP3		# YTMP3 = W[-15] ror 1 ^ W[-15] >> 7
230	vpxor		YTMP1, YTMP3, YTMP1		# YTMP1 = s0
231
232
233	# Add three components, w[t-16], w[t-7] and sigma0
234	vpaddq		YTMP1, YTMP0, YTMP0		# YTMP0 = W[-16] + W[-7] + s0
235	# Move to appropriate lanes for calculating w[16] and w[17]
236	vperm2f128	$0x0, YTMP0, YTMP0, Y_0		# Y_0 = W[-16] + W[-7] + s0 {BABA}
237	# Move to appropriate lanes for calculating w[18] and w[19]
238	vpand		MASK_YMM_LO(%rip), YTMP0, YTMP0	# YTMP0 = W[-16] + W[-7] + s0 {DC00}
239
240	# Calculate w[16] and w[17] in both 128 bit lanes
241
242	# Calculate sigma1 for w[16] and w[17] on both 128 bit lanes
243	vperm2f128	$0x11, Y_3, Y_3, YTMP2		# YTMP2 = W[-2] {BABA}
244	vpsrlq		$6, YTMP2, YTMP4		# YTMP4 = W[-2] >> 6 {BABA}
245
246
247	mov	a, y3		# y3 = a                                # MAJA
248	rorx	$41, e, y0	# y0 = e >> 41				# S1A
249	rorx	$18, e, y1	# y1 = e >> 18				# S1B
250	add	1*8+frame_XFER(%rsp), h		# h = k + w + h         # --
251	or	c, y3		# y3 = a|c                              # MAJA
252
253
254	mov	f, y2		# y2 = f                                # CH
255	rorx	$34, a, T1	# T1 = a >> 34				# S0B
256	xor	y1, y0		# y0 = (e>>41) ^ (e>>18)		# S1
257	xor	g, y2		# y2 = f^g                              # CH
258
259
260	rorx	$14, e, y1	# y1 = (e >> 14)			# S1
261	xor	y1, y0		# y0 = (e>>41) ^ (e>>18) ^ (e>>14)	# S1
262	rorx	$39, a, y1	# y1 = a >> 39				# S0A
263	and	e, y2		# y2 = (f^g)&e                          # CH
264	add	h, d		# d = k + w + h + d                     # --
265
266	and	b, y3		# y3 = (a|c)&b                          # MAJA
267	xor	T1, y1		# y1 = (a>>39) ^ (a>>34)		# S0
268
269	rorx	$28, a, T1	# T1 = (a >> 28)			# S0
270	xor	g, y2		# y2 = CH = ((f^g)&e)^g                 # CH
271
272	xor	T1, y1		# y1 = (a>>39) ^ (a>>34) ^ (a>>28)	# S0
273	mov	a, T1		# T1 = a                                # MAJB
274	and	c, T1		# T1 = a&c                              # MAJB
275	add	y0, y2		# y2 = S1 + CH                          # --
276
277	or	T1, y3		# y3 = MAJ = (a|c)&b)|(a&c)             # MAJ
278	add	y1, h		# h = k + w + h + S0                    # --
279
280	add	y2, d		# d = k + w + h + d + S1 + CH = d + t1  # --
281	add	y2, h		# h = k + w + h + S0 + S1 + CH = t1 + S0# --
282	add	y3, h		# h = t1 + S0 + MAJ                     # --
283
284	RotateState
285
286
287################################### RND N + 2 #########################################
288
289	vpsrlq		$19, YTMP2, YTMP3		# YTMP3 = W[-2] >> 19 {BABA}
290	vpsllq		$(64-19), YTMP2, YTMP1		# YTMP1 = W[-2] << 19 {BABA}
291	vpor		YTMP1, YTMP3, YTMP3		# YTMP3 = W[-2] ror 19 {BABA}
292	vpxor		YTMP3, YTMP4, YTMP4		# YTMP4 = W[-2] ror 19 ^ W[-2] >> 6 {BABA}
293	vpsrlq		$61, YTMP2, YTMP3		# YTMP3 = W[-2] >> 61 {BABA}
294	vpsllq		$(64-61), YTMP2, YTMP1		# YTMP1 = W[-2] << 61 {BABA}
295	vpor		YTMP1, YTMP3, YTMP3		# YTMP3 = W[-2] ror 61 {BABA}
296	vpxor		YTMP3, YTMP4, YTMP4		# YTMP4 = s1 = (W[-2] ror 19) ^
297							#  (W[-2] ror 61) ^ (W[-2] >> 6) {BABA}
298
299	# Add sigma1 to the other compunents to get w[16] and w[17]
300	vpaddq		YTMP4, Y_0, Y_0			# Y_0 = {W[1], W[0], W[1], W[0]}
301
302	# Calculate sigma1 for w[18] and w[19] for upper 128 bit lane
303	vpsrlq		$6, Y_0, YTMP4			# YTMP4 = W[-2] >> 6 {DC--}
304
305	mov	a, y3		# y3 = a                                # MAJA
306	rorx	$41, e, y0	# y0 = e >> 41				# S1A
307	add	2*8+frame_XFER(%rsp), h		# h = k + w + h         # --
308
309	rorx	$18, e, y1	# y1 = e >> 18				# S1B
310	or	c, y3		# y3 = a|c                              # MAJA
311	mov	f, y2		# y2 = f                                # CH
312	xor	g, y2		# y2 = f^g                              # CH
313
314	rorx	$34, a, T1	# T1 = a >> 34				# S0B
315	xor	y1, y0		# y0 = (e>>41) ^ (e>>18)		# S1
316	and	e, y2		# y2 = (f^g)&e                          # CH
317
318	rorx	$14, e, y1	# y1 = (e >> 14)			# S1
319	add	h, d		# d = k + w + h + d                     # --
320	and	b, y3		# y3 = (a|c)&b                          # MAJA
321
322	xor	y1, y0		# y0 = (e>>41) ^ (e>>18) ^ (e>>14)	# S1
323	rorx	$39, a, y1	# y1 = a >> 39				# S0A
324	xor	g, y2		# y2 = CH = ((f^g)&e)^g                 # CH
325
326	xor	T1, y1		# y1 = (a>>39) ^ (a>>34)		# S0
327	rorx	$28, a, T1	# T1 = (a >> 28)			# S0
328
329	xor	T1, y1		# y1 = (a>>39) ^ (a>>34) ^ (a>>28)	# S0
330	mov	a, T1		# T1 = a                                # MAJB
331	and	c, T1		# T1 = a&c                              # MAJB
332	add	y0, y2		# y2 = S1 + CH                          # --
333
334	or	T1, y3		# y3 = MAJ = (a|c)&b)|(a&c)             # MAJ
335	add	y1, h		# h = k + w + h + S0                    # --
336	add	y2, d		# d = k + w + h + d + S1 + CH = d + t1  # --
337	add	y2, h		# h = k + w + h + S0 + S1 + CH = t1 + S0# --
338
339	add	y3, h		# h = t1 + S0 + MAJ                     # --
340
341	RotateState
342
343################################### RND N + 3 #########################################
344
345	vpsrlq		$19, Y_0, YTMP3			# YTMP3 = W[-2] >> 19 {DC--}
346	vpsllq		$(64-19), Y_0, YTMP1		# YTMP1 = W[-2] << 19 {DC--}
347	vpor		YTMP1, YTMP3, YTMP3		# YTMP3 = W[-2] ror 19 {DC--}
348	vpxor		YTMP3, YTMP4, YTMP4		# YTMP4 = W[-2] ror 19 ^ W[-2] >> 6 {DC--}
349	vpsrlq		$61, Y_0, YTMP3			# YTMP3 = W[-2] >> 61 {DC--}
350	vpsllq		$(64-61), Y_0, YTMP1		# YTMP1 = W[-2] << 61 {DC--}
351	vpor		YTMP1, YTMP3, YTMP3		# YTMP3 = W[-2] ror 61 {DC--}
352	vpxor		YTMP3, YTMP4, YTMP4		# YTMP4 = s1 = (W[-2] ror 19) ^
353							#  (W[-2] ror 61) ^ (W[-2] >> 6) {DC--}
354
355	# Add the sigma0 + w[t-7] + w[t-16] for w[18] and w[19]
356	# to newly calculated sigma1 to get w[18] and w[19]
357	vpaddq		YTMP4, YTMP0, YTMP2		# YTMP2 = {W[3], W[2], --, --}
358
359	# Form w[19, w[18], w17], w[16]
360	vpblendd		$0xF0, YTMP2, Y_0, Y_0		# Y_0 = {W[3], W[2], W[1], W[0]}
361
362	mov	a, y3		# y3 = a                                # MAJA
363	rorx	$41, e, y0	# y0 = e >> 41				# S1A
364	rorx	$18, e, y1	# y1 = e >> 18				# S1B
365	add	3*8+frame_XFER(%rsp), h		# h = k + w + h         # --
366	or	c, y3		# y3 = a|c                              # MAJA
367
368
369	mov	f, y2		# y2 = f                                # CH
370	rorx	$34, a, T1	# T1 = a >> 34				# S0B
371	xor	y1, y0		# y0 = (e>>41) ^ (e>>18)		# S1
372	xor	g, y2		# y2 = f^g                              # CH
373
374
375	rorx	$14, e, y1	# y1 = (e >> 14)			# S1
376	and	e, y2		# y2 = (f^g)&e                          # CH
377	add	h, d		# d = k + w + h + d                     # --
378	and	b, y3		# y3 = (a|c)&b                          # MAJA
379
380	xor	y1, y0		# y0 = (e>>41) ^ (e>>18) ^ (e>>14)	# S1
381	xor	g, y2		# y2 = CH = ((f^g)&e)^g                 # CH
382
383	rorx	$39, a, y1	# y1 = a >> 39				# S0A
384	add	y0, y2		# y2 = S1 + CH                          # --
385
386	xor	T1, y1		# y1 = (a>>39) ^ (a>>34)		# S0
387	add	y2, d		# d = k + w + h + d + S1 + CH = d + t1  # --
388
389	rorx	$28, a, T1	# T1 = (a >> 28)			# S0
390
391	xor	T1, y1		# y1 = (a>>39) ^ (a>>34) ^ (a>>28)	# S0
392	mov	a, T1		# T1 = a                                # MAJB
393	and	c, T1		# T1 = a&c                              # MAJB
394	or	T1, y3		# y3 = MAJ = (a|c)&b)|(a&c)             # MAJ
395
396	add	y1, h		# h = k + w + h + S0                    # --
397	add	y2, h		# h = k + w + h + S0 + S1 + CH = t1 + S0# --
398	add	y3, h		# h = t1 + S0 + MAJ                     # --
399
400	RotateState
401
402	rotate_Ys
403.endm
404
405.macro DO_4ROUNDS
406
407################################### RND N + 0 #########################################
408
409	mov	f, y2		# y2 = f                                # CH
410	rorx	$41, e, y0	# y0 = e >> 41				# S1A
411	rorx	$18, e, y1	# y1 = e >> 18				# S1B
412	xor	g, y2		# y2 = f^g                              # CH
413
414	xor	y1, y0		# y0 = (e>>41) ^ (e>>18)		# S1
415	rorx	$14, e, y1	# y1 = (e >> 14)			# S1
416	and	e, y2		# y2 = (f^g)&e                          # CH
417
418	xor	y1, y0		# y0 = (e>>41) ^ (e>>18) ^ (e>>14)	# S1
419	rorx	$34, a, T1	# T1 = a >> 34				# S0B
420	xor	g, y2		# y2 = CH = ((f^g)&e)^g                 # CH
421	rorx	$39, a, y1	# y1 = a >> 39				# S0A
422	mov	a, y3		# y3 = a                                # MAJA
423
424	xor	T1, y1		# y1 = (a>>39) ^ (a>>34)		# S0
425	rorx	$28, a, T1	# T1 = (a >> 28)			# S0
426	add	frame_XFER(%rsp), h		# h = k + w + h         # --
427	or	c, y3		# y3 = a|c                              # MAJA
428
429	xor	T1, y1		# y1 = (a>>39) ^ (a>>34) ^ (a>>28)	# S0
430	mov	a, T1		# T1 = a                                # MAJB
431	and	b, y3		# y3 = (a|c)&b                          # MAJA
432	and	c, T1		# T1 = a&c                              # MAJB
433	add	y0, y2		# y2 = S1 + CH                          # --
434
435	add	h, d		# d = k + w + h + d                     # --
436	or	T1, y3		# y3 = MAJ = (a|c)&b)|(a&c)             # MAJ
437	add	y1, h		# h = k + w + h + S0                    # --
438
439	add	y2, d		# d = k + w + h + d + S1 + CH = d + t1  # --
440
441	RotateState
442
443################################### RND N + 1 #########################################
444
445	add	y2, old_h	# h = k + w + h + S0 + S1 + CH = t1 + S0# --
446	mov	f, y2		# y2 = f                                # CH
447	rorx	$41, e, y0	# y0 = e >> 41				# S1A
448	rorx	$18, e, y1	# y1 = e >> 18				# S1B
449	xor	g, y2		# y2 = f^g                              # CH
450
451	xor	y1, y0		# y0 = (e>>41) ^ (e>>18)		# S1
452	rorx	$14, e, y1	# y1 = (e >> 14)			# S1
453	and	e, y2		# y2 = (f^g)&e                          # CH
454	add	y3, old_h	# h = t1 + S0 + MAJ                     # --
455
456	xor	y1, y0		# y0 = (e>>41) ^ (e>>18) ^ (e>>14)	# S1
457	rorx	$34, a, T1	# T1 = a >> 34				# S0B
458	xor	g, y2		# y2 = CH = ((f^g)&e)^g                 # CH
459	rorx	$39, a, y1	# y1 = a >> 39				# S0A
460	mov	a, y3		# y3 = a                                # MAJA
461
462	xor	T1, y1		# y1 = (a>>39) ^ (a>>34)		# S0
463	rorx	$28, a, T1	# T1 = (a >> 28)			# S0
464	add	8*1+frame_XFER(%rsp), h		# h = k + w + h         # --
465	or	c, y3		# y3 = a|c                              # MAJA
466
467	xor	T1, y1		# y1 = (a>>39) ^ (a>>34) ^ (a>>28)	# S0
468	mov	a, T1		# T1 = a                                # MAJB
469	and	b, y3		# y3 = (a|c)&b                          # MAJA
470	and	c, T1		# T1 = a&c                              # MAJB
471	add	y0, y2		# y2 = S1 + CH                          # --
472
473	add	h, d		# d = k + w + h + d                     # --
474	or	T1, y3		# y3 = MAJ = (a|c)&b)|(a&c)             # MAJ
475	add	y1, h		# h = k + w + h + S0                    # --
476
477	add	y2, d		# d = k + w + h + d + S1 + CH = d + t1  # --
478
479	RotateState
480
481################################### RND N + 2 #########################################
482
483	add	y2, old_h	# h = k + w + h + S0 + S1 + CH = t1 + S0# --
484	mov	f, y2		# y2 = f                                # CH
485	rorx	$41, e, y0	# y0 = e >> 41				# S1A
486	rorx	$18, e, y1	# y1 = e >> 18				# S1B
487	xor	g, y2		# y2 = f^g                              # CH
488
489	xor	y1, y0		# y0 = (e>>41) ^ (e>>18)		# S1
490	rorx	$14, e, y1	# y1 = (e >> 14)			# S1
491	and	e, y2		# y2 = (f^g)&e                          # CH
492	add	y3, old_h	# h = t1 + S0 + MAJ                     # --
493
494	xor	y1, y0		# y0 = (e>>41) ^ (e>>18) ^ (e>>14)	# S1
495	rorx	$34, a, T1	# T1 = a >> 34				# S0B
496	xor	g, y2		# y2 = CH = ((f^g)&e)^g                 # CH
497	rorx	$39, a, y1	# y1 = a >> 39				# S0A
498	mov	a, y3		# y3 = a                                # MAJA
499
500	xor	T1, y1		# y1 = (a>>39) ^ (a>>34)		# S0
501	rorx	$28, a, T1	# T1 = (a >> 28)			# S0
502	add	8*2+frame_XFER(%rsp), h		# h = k + w + h         # --
503	or	c, y3		# y3 = a|c                              # MAJA
504
505	xor	T1, y1		# y1 = (a>>39) ^ (a>>34) ^ (a>>28)	# S0
506	mov	a, T1		# T1 = a                                # MAJB
507	and	b, y3		# y3 = (a|c)&b                          # MAJA
508	and	c, T1		# T1 = a&c                              # MAJB
509	add	y0, y2		# y2 = S1 + CH                          # --
510
511	add	h, d		# d = k + w + h + d                     # --
512	or	T1, y3		# y3 = MAJ = (a|c)&b)|(a&c)             # MAJ
513	add	y1, h		# h = k + w + h + S0                    # --
514
515	add	y2, d		# d = k + w + h + d + S1 + CH = d + t1  # --
516
517	RotateState
518
519################################### RND N + 3 #########################################
520
521	add	y2, old_h	# h = k + w + h + S0 + S1 + CH = t1 + S0# --
522	mov	f, y2		# y2 = f                                # CH
523	rorx	$41, e, y0	# y0 = e >> 41				# S1A
524	rorx	$18, e, y1	# y1 = e >> 18				# S1B
525	xor	g, y2		# y2 = f^g                              # CH
526
527	xor	y1, y0		# y0 = (e>>41) ^ (e>>18)		# S1
528	rorx	$14, e, y1	# y1 = (e >> 14)			# S1
529	and	e, y2		# y2 = (f^g)&e                          # CH
530	add	y3, old_h	# h = t1 + S0 + MAJ                     # --
531
532	xor	y1, y0		# y0 = (e>>41) ^ (e>>18) ^ (e>>14)	# S1
533	rorx	$34, a, T1	# T1 = a >> 34				# S0B
534	xor	g, y2		# y2 = CH = ((f^g)&e)^g                 # CH
535	rorx	$39, a, y1	# y1 = a >> 39				# S0A
536	mov	a, y3		# y3 = a                                # MAJA
537
538	xor	T1, y1		# y1 = (a>>39) ^ (a>>34)		# S0
539	rorx	$28, a, T1	# T1 = (a >> 28)			# S0
540	add	8*3+frame_XFER(%rsp), h		# h = k + w + h         # --
541	or	c, y3		# y3 = a|c                              # MAJA
542
543	xor	T1, y1		# y1 = (a>>39) ^ (a>>34) ^ (a>>28)	# S0
544	mov	a, T1		# T1 = a                                # MAJB
545	and	b, y3		# y3 = (a|c)&b                          # MAJA
546	and	c, T1		# T1 = a&c                              # MAJB
547	add	y0, y2		# y2 = S1 + CH                          # --
548
549
550	add	h, d		# d = k + w + h + d                     # --
551	or	T1, y3		# y3 = MAJ = (a|c)&b)|(a&c)             # MAJ
552	add	y1, h		# h = k + w + h + S0                    # --
553
554	add	y2, d		# d = k + w + h + d + S1 + CH = d + t1  # --
555
556	add	y2, h		# h = k + w + h + S0 + S1 + CH = t1 + S0# --
557
558	add	y3, h		# h = t1 + S0 + MAJ                     # --
559
560	RotateState
561
562.endm
563
564########################################################################
565# void sha512_transform_rorx(sha512_state *state, const u8 *data, int blocks)
566# Purpose: Updates the SHA512 digest stored at "state" with the message
567# stored in "data".
568# The size of the message pointed to by "data" must be an integer multiple
569# of SHA512 message blocks.
570# "blocks" is the message length in SHA512 blocks
571########################################################################
572SYM_FUNC_START(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
685SYM_FUNC_END(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