xref: /linux/arch/x86/crypto/sha512-avx2-asm.S (revision 1e0731c05c985deb68a97fa44c1adcd3305dda90)
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#include <linux/cfi_types.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
106
107frame_XFER = 0
108frame_SRND = frame_XFER + XFER_SIZE
109frame_INP = frame_SRND + SRND_SIZE
110frame_INPEND = frame_INP + INP_SIZE
111frame_CTX = frame_INPEND + INPEND_SIZE
112frame_size = frame_CTX + CTX_SIZE
113
114## assume buffers not aligned
115#define	VMOVDQ vmovdqu
116
117# addm [mem], reg
118# Add reg to mem using reg-mem add and store
119.macro addm p1 p2
120	add	\p1, \p2
121	mov	\p2, \p1
122.endm
123
124
125# COPY_YMM_AND_BSWAP ymm, [mem], byte_flip_mask
126# Load ymm with mem and byte swap each dword
127.macro COPY_YMM_AND_BSWAP p1 p2 p3
128	VMOVDQ \p2, \p1
129	vpshufb \p3, \p1, \p1
130.endm
131# rotate_Ys
132# Rotate values of symbols Y0...Y3
133.macro rotate_Ys
134	Y_ = Y_0
135	Y_0 = Y_1
136	Y_1 = Y_2
137	Y_2 = Y_3
138	Y_3 = Y_
139.endm
140
141# RotateState
142.macro RotateState
143	# Rotate symbols a..h right
144	old_h  = h
145	TMP_   = h
146	h      = g
147	g      = f
148	f      = e
149	e      = d
150	d      = c
151	c      = b
152	b      = a
153	a      = TMP_
154.endm
155
156# macro MY_VPALIGNR	YDST, YSRC1, YSRC2, RVAL
157# YDST = {YSRC1, YSRC2} >> RVAL*8
158.macro MY_VPALIGNR YDST YSRC1 YSRC2 RVAL
159	vperm2f128      $0x3, \YSRC2, \YSRC1, \YDST     # YDST = {YS1_LO, YS2_HI}
160	vpalignr        $\RVAL, \YSRC2, \YDST, \YDST    # YDST = {YDS1, YS2} >> RVAL*8
161.endm
162
163.macro FOUR_ROUNDS_AND_SCHED
164################################### RND N + 0 #########################################
165
166	# Extract w[t-7]
167	MY_VPALIGNR	YTMP0, Y_3, Y_2, 8		# YTMP0 = W[-7]
168	# Calculate w[t-16] + w[t-7]
169	vpaddq		Y_0, YTMP0, YTMP0		# YTMP0 = W[-7] + W[-16]
170	# Extract w[t-15]
171	MY_VPALIGNR	YTMP1, Y_1, Y_0, 8		# YTMP1 = W[-15]
172
173	# Calculate sigma0
174
175	# Calculate w[t-15] ror 1
176	vpsrlq		$1, YTMP1, YTMP2
177	vpsllq		$(64-1), YTMP1, YTMP3
178	vpor		YTMP2, YTMP3, YTMP3		# YTMP3 = W[-15] ror 1
179	# Calculate w[t-15] shr 7
180	vpsrlq		$7, YTMP1, YTMP4		# YTMP4 = W[-15] >> 7
181
182	mov	a, y3		# y3 = a                                # MAJA
183	rorx	$41, e, y0	# y0 = e >> 41				# S1A
184	rorx	$18, e, y1	# y1 = e >> 18				# S1B
185	add	frame_XFER(%rsp),h		# h = k + w + h         # --
186	or	c, y3		# y3 = a|c                              # MAJA
187	mov	f, y2		# y2 = f                                # CH
188	rorx	$34, a, T1	# T1 = a >> 34				# S0B
189
190	xor	y1, y0		# y0 = (e>>41) ^ (e>>18)		# S1
191	xor	g, y2		# y2 = f^g                              # CH
192	rorx	$14, e, y1	# y1 = (e >> 14)			# S1
193
194	and	e, y2		# y2 = (f^g)&e                          # CH
195	xor	y1, y0		# y0 = (e>>41) ^ (e>>18) ^ (e>>14)	# S1
196	rorx	$39, a, y1	# y1 = a >> 39				# S0A
197	add	h, d		# d = k + w + h + d                     # --
198
199	and	b, y3		# y3 = (a|c)&b                          # MAJA
200	xor	T1, y1		# y1 = (a>>39) ^ (a>>34)		# S0
201	rorx	$28, a, T1	# T1 = (a >> 28)			# S0
202
203	xor	g, y2		# y2 = CH = ((f^g)&e)^g                 # CH
204	xor	T1, y1		# y1 = (a>>39) ^ (a>>34) ^ (a>>28)	# S0
205	mov	a, T1		# T1 = a                                # MAJB
206	and	c, T1		# T1 = a&c                              # MAJB
207
208	add	y0, y2		# y2 = S1 + CH                          # --
209	or	T1, y3		# y3 = MAJ = (a|c)&b)|(a&c)             # MAJ
210	add	y1, h		# h = k + w + h + S0                    # --
211
212	add	y2, d		# d = k + w + h + d + S1 + CH = d + t1  # --
213
214	add	y2, h		# h = k + w + h + S0 + S1 + CH = t1 + S0# --
215	add	y3, h		# h = t1 + S0 + MAJ                     # --
216
217	RotateState
218
219################################### RND N + 1 #########################################
220
221	# Calculate w[t-15] ror 8
222	vpsrlq		$8, YTMP1, YTMP2
223	vpsllq		$(64-8), YTMP1, YTMP1
224	vpor		YTMP2, YTMP1, YTMP1		# YTMP1 = W[-15] ror 8
225	# XOR the three components
226	vpxor		YTMP4, YTMP3, YTMP3		# YTMP3 = W[-15] ror 1 ^ W[-15] >> 7
227	vpxor		YTMP1, YTMP3, YTMP1		# YTMP1 = s0
228
229
230	# Add three components, w[t-16], w[t-7] and sigma0
231	vpaddq		YTMP1, YTMP0, YTMP0		# YTMP0 = W[-16] + W[-7] + s0
232	# Move to appropriate lanes for calculating w[16] and w[17]
233	vperm2f128	$0x0, YTMP0, YTMP0, Y_0		# Y_0 = W[-16] + W[-7] + s0 {BABA}
234	# Move to appropriate lanes for calculating w[18] and w[19]
235	vpand		MASK_YMM_LO(%rip), YTMP0, YTMP0	# YTMP0 = W[-16] + W[-7] + s0 {DC00}
236
237	# Calculate w[16] and w[17] in both 128 bit lanes
238
239	# Calculate sigma1 for w[16] and w[17] on both 128 bit lanes
240	vperm2f128	$0x11, Y_3, Y_3, YTMP2		# YTMP2 = W[-2] {BABA}
241	vpsrlq		$6, YTMP2, YTMP4		# YTMP4 = W[-2] >> 6 {BABA}
242
243
244	mov	a, y3		# y3 = a                                # MAJA
245	rorx	$41, e, y0	# y0 = e >> 41				# S1A
246	rorx	$18, e, y1	# y1 = e >> 18				# S1B
247	add	1*8+frame_XFER(%rsp), h		# h = k + w + h         # --
248	or	c, y3		# y3 = a|c                              # MAJA
249
250
251	mov	f, y2		# y2 = f                                # CH
252	rorx	$34, a, T1	# T1 = a >> 34				# S0B
253	xor	y1, y0		# y0 = (e>>41) ^ (e>>18)		# S1
254	xor	g, y2		# y2 = f^g                              # CH
255
256
257	rorx	$14, e, y1	# y1 = (e >> 14)			# S1
258	xor	y1, y0		# y0 = (e>>41) ^ (e>>18) ^ (e>>14)	# S1
259	rorx	$39, a, y1	# y1 = a >> 39				# S0A
260	and	e, y2		# y2 = (f^g)&e                          # CH
261	add	h, d		# d = k + w + h + d                     # --
262
263	and	b, y3		# y3 = (a|c)&b                          # MAJA
264	xor	T1, y1		# y1 = (a>>39) ^ (a>>34)		# S0
265
266	rorx	$28, a, T1	# T1 = (a >> 28)			# S0
267	xor	g, y2		# y2 = CH = ((f^g)&e)^g                 # CH
268
269	xor	T1, y1		# y1 = (a>>39) ^ (a>>34) ^ (a>>28)	# S0
270	mov	a, T1		# T1 = a                                # MAJB
271	and	c, T1		# T1 = a&c                              # MAJB
272	add	y0, y2		# y2 = S1 + CH                          # --
273
274	or	T1, y3		# y3 = MAJ = (a|c)&b)|(a&c)             # MAJ
275	add	y1, h		# h = k + w + h + S0                    # --
276
277	add	y2, d		# d = k + w + h + d + S1 + CH = d + t1  # --
278	add	y2, h		# h = k + w + h + S0 + S1 + CH = t1 + S0# --
279	add	y3, h		# h = t1 + S0 + MAJ                     # --
280
281	RotateState
282
283
284################################### RND N + 2 #########################################
285
286	vpsrlq		$19, YTMP2, YTMP3		# YTMP3 = W[-2] >> 19 {BABA}
287	vpsllq		$(64-19), YTMP2, YTMP1		# YTMP1 = W[-2] << 19 {BABA}
288	vpor		YTMP1, YTMP3, YTMP3		# YTMP3 = W[-2] ror 19 {BABA}
289	vpxor		YTMP3, YTMP4, YTMP4		# YTMP4 = W[-2] ror 19 ^ W[-2] >> 6 {BABA}
290	vpsrlq		$61, YTMP2, YTMP3		# YTMP3 = W[-2] >> 61 {BABA}
291	vpsllq		$(64-61), YTMP2, YTMP1		# YTMP1 = W[-2] << 61 {BABA}
292	vpor		YTMP1, YTMP3, YTMP3		# YTMP3 = W[-2] ror 61 {BABA}
293	vpxor		YTMP3, YTMP4, YTMP4		# YTMP4 = s1 = (W[-2] ror 19) ^
294							#  (W[-2] ror 61) ^ (W[-2] >> 6) {BABA}
295
296	# Add sigma1 to the other compunents to get w[16] and w[17]
297	vpaddq		YTMP4, Y_0, Y_0			# Y_0 = {W[1], W[0], W[1], W[0]}
298
299	# Calculate sigma1 for w[18] and w[19] for upper 128 bit lane
300	vpsrlq		$6, Y_0, YTMP4			# YTMP4 = W[-2] >> 6 {DC--}
301
302	mov	a, y3		# y3 = a                                # MAJA
303	rorx	$41, e, y0	# y0 = e >> 41				# S1A
304	add	2*8+frame_XFER(%rsp), h		# h = k + w + h         # --
305
306	rorx	$18, e, y1	# y1 = e >> 18				# S1B
307	or	c, y3		# y3 = a|c                              # MAJA
308	mov	f, y2		# y2 = f                                # CH
309	xor	g, y2		# y2 = f^g                              # CH
310
311	rorx	$34, a, T1	# T1 = a >> 34				# S0B
312	xor	y1, y0		# y0 = (e>>41) ^ (e>>18)		# S1
313	and	e, y2		# y2 = (f^g)&e                          # CH
314
315	rorx	$14, e, y1	# y1 = (e >> 14)			# S1
316	add	h, d		# d = k + w + h + d                     # --
317	and	b, y3		# y3 = (a|c)&b                          # MAJA
318
319	xor	y1, y0		# y0 = (e>>41) ^ (e>>18) ^ (e>>14)	# S1
320	rorx	$39, a, y1	# y1 = a >> 39				# S0A
321	xor	g, y2		# y2 = CH = ((f^g)&e)^g                 # CH
322
323	xor	T1, y1		# y1 = (a>>39) ^ (a>>34)		# S0
324	rorx	$28, a, T1	# T1 = (a >> 28)			# S0
325
326	xor	T1, y1		# y1 = (a>>39) ^ (a>>34) ^ (a>>28)	# S0
327	mov	a, T1		# T1 = a                                # MAJB
328	and	c, T1		# T1 = a&c                              # MAJB
329	add	y0, y2		# y2 = S1 + CH                          # --
330
331	or	T1, y3		# y3 = MAJ = (a|c)&b)|(a&c)             # MAJ
332	add	y1, h		# h = k + w + h + S0                    # --
333	add	y2, d		# d = k + w + h + d + S1 + CH = d + t1  # --
334	add	y2, h		# h = k + w + h + S0 + S1 + CH = t1 + S0# --
335
336	add	y3, h		# h = t1 + S0 + MAJ                     # --
337
338	RotateState
339
340################################### RND N + 3 #########################################
341
342	vpsrlq		$19, Y_0, YTMP3			# YTMP3 = W[-2] >> 19 {DC--}
343	vpsllq		$(64-19), Y_0, YTMP1		# YTMP1 = W[-2] << 19 {DC--}
344	vpor		YTMP1, YTMP3, YTMP3		# YTMP3 = W[-2] ror 19 {DC--}
345	vpxor		YTMP3, YTMP4, YTMP4		# YTMP4 = W[-2] ror 19 ^ W[-2] >> 6 {DC--}
346	vpsrlq		$61, Y_0, YTMP3			# YTMP3 = W[-2] >> 61 {DC--}
347	vpsllq		$(64-61), Y_0, YTMP1		# YTMP1 = W[-2] << 61 {DC--}
348	vpor		YTMP1, YTMP3, YTMP3		# YTMP3 = W[-2] ror 61 {DC--}
349	vpxor		YTMP3, YTMP4, YTMP4		# YTMP4 = s1 = (W[-2] ror 19) ^
350							#  (W[-2] ror 61) ^ (W[-2] >> 6) {DC--}
351
352	# Add the sigma0 + w[t-7] + w[t-16] for w[18] and w[19]
353	# to newly calculated sigma1 to get w[18] and w[19]
354	vpaddq		YTMP4, YTMP0, YTMP2		# YTMP2 = {W[3], W[2], --, --}
355
356	# Form w[19, w[18], w17], w[16]
357	vpblendd		$0xF0, YTMP2, Y_0, Y_0		# Y_0 = {W[3], W[2], W[1], W[0]}
358
359	mov	a, y3		# y3 = a                                # MAJA
360	rorx	$41, e, y0	# y0 = e >> 41				# S1A
361	rorx	$18, e, y1	# y1 = e >> 18				# S1B
362	add	3*8+frame_XFER(%rsp), h		# h = k + w + h         # --
363	or	c, y3		# y3 = a|c                              # MAJA
364
365
366	mov	f, y2		# y2 = f                                # CH
367	rorx	$34, a, T1	# T1 = a >> 34				# S0B
368	xor	y1, y0		# y0 = (e>>41) ^ (e>>18)		# S1
369	xor	g, y2		# y2 = f^g                              # CH
370
371
372	rorx	$14, e, y1	# y1 = (e >> 14)			# S1
373	and	e, y2		# y2 = (f^g)&e                          # CH
374	add	h, d		# d = k + w + h + d                     # --
375	and	b, y3		# y3 = (a|c)&b                          # MAJA
376
377	xor	y1, y0		# y0 = (e>>41) ^ (e>>18) ^ (e>>14)	# S1
378	xor	g, y2		# y2 = CH = ((f^g)&e)^g                 # CH
379
380	rorx	$39, a, y1	# y1 = a >> 39				# S0A
381	add	y0, y2		# y2 = S1 + CH                          # --
382
383	xor	T1, y1		# y1 = (a>>39) ^ (a>>34)		# S0
384	add	y2, d		# d = k + w + h + d + S1 + CH = d + t1  # --
385
386	rorx	$28, a, T1	# T1 = (a >> 28)			# S0
387
388	xor	T1, y1		# y1 = (a>>39) ^ (a>>34) ^ (a>>28)	# S0
389	mov	a, T1		# T1 = a                                # MAJB
390	and	c, T1		# T1 = a&c                              # MAJB
391	or	T1, y3		# y3 = MAJ = (a|c)&b)|(a&c)             # MAJ
392
393	add	y1, h		# h = k + w + h + S0                    # --
394	add	y2, h		# h = k + w + h + S0 + S1 + CH = t1 + S0# --
395	add	y3, h		# h = t1 + S0 + MAJ                     # --
396
397	RotateState
398
399	rotate_Ys
400.endm
401
402.macro DO_4ROUNDS
403
404################################### RND N + 0 #########################################
405
406	mov	f, y2		# y2 = f                                # CH
407	rorx	$41, e, y0	# y0 = e >> 41				# S1A
408	rorx	$18, e, y1	# y1 = e >> 18				# S1B
409	xor	g, y2		# y2 = f^g                              # CH
410
411	xor	y1, y0		# y0 = (e>>41) ^ (e>>18)		# S1
412	rorx	$14, e, y1	# y1 = (e >> 14)			# S1
413	and	e, y2		# y2 = (f^g)&e                          # CH
414
415	xor	y1, y0		# y0 = (e>>41) ^ (e>>18) ^ (e>>14)	# S1
416	rorx	$34, a, T1	# T1 = a >> 34				# S0B
417	xor	g, y2		# y2 = CH = ((f^g)&e)^g                 # CH
418	rorx	$39, a, y1	# y1 = a >> 39				# S0A
419	mov	a, y3		# y3 = a                                # MAJA
420
421	xor	T1, y1		# y1 = (a>>39) ^ (a>>34)		# S0
422	rorx	$28, a, T1	# T1 = (a >> 28)			# S0
423	add	frame_XFER(%rsp), h		# h = k + w + h         # --
424	or	c, y3		# y3 = a|c                              # MAJA
425
426	xor	T1, y1		# y1 = (a>>39) ^ (a>>34) ^ (a>>28)	# S0
427	mov	a, T1		# T1 = a                                # MAJB
428	and	b, y3		# y3 = (a|c)&b                          # MAJA
429	and	c, T1		# T1 = a&c                              # MAJB
430	add	y0, y2		# y2 = S1 + CH                          # --
431
432	add	h, d		# d = k + w + h + d                     # --
433	or	T1, y3		# y3 = MAJ = (a|c)&b)|(a&c)             # MAJ
434	add	y1, h		# h = k + w + h + S0                    # --
435
436	add	y2, d		# d = k + w + h + d + S1 + CH = d + t1  # --
437
438	RotateState
439
440################################### RND N + 1 #########################################
441
442	add	y2, old_h	# h = k + w + h + S0 + S1 + CH = t1 + S0# --
443	mov	f, y2		# y2 = f                                # CH
444	rorx	$41, e, y0	# y0 = e >> 41				# S1A
445	rorx	$18, e, y1	# y1 = e >> 18				# S1B
446	xor	g, y2		# y2 = f^g                              # CH
447
448	xor	y1, y0		# y0 = (e>>41) ^ (e>>18)		# S1
449	rorx	$14, e, y1	# y1 = (e >> 14)			# S1
450	and	e, y2		# y2 = (f^g)&e                          # CH
451	add	y3, old_h	# h = t1 + S0 + MAJ                     # --
452
453	xor	y1, y0		# y0 = (e>>41) ^ (e>>18) ^ (e>>14)	# S1
454	rorx	$34, a, T1	# T1 = a >> 34				# S0B
455	xor	g, y2		# y2 = CH = ((f^g)&e)^g                 # CH
456	rorx	$39, a, y1	# y1 = a >> 39				# S0A
457	mov	a, y3		# y3 = a                                # MAJA
458
459	xor	T1, y1		# y1 = (a>>39) ^ (a>>34)		# S0
460	rorx	$28, a, T1	# T1 = (a >> 28)			# S0
461	add	8*1+frame_XFER(%rsp), h		# h = k + w + h         # --
462	or	c, y3		# y3 = a|c                              # MAJA
463
464	xor	T1, y1		# y1 = (a>>39) ^ (a>>34) ^ (a>>28)	# S0
465	mov	a, T1		# T1 = a                                # MAJB
466	and	b, y3		# y3 = (a|c)&b                          # MAJA
467	and	c, T1		# T1 = a&c                              # MAJB
468	add	y0, y2		# y2 = S1 + CH                          # --
469
470	add	h, d		# d = k + w + h + d                     # --
471	or	T1, y3		# y3 = MAJ = (a|c)&b)|(a&c)             # MAJ
472	add	y1, h		# h = k + w + h + S0                    # --
473
474	add	y2, d		# d = k + w + h + d + S1 + CH = d + t1  # --
475
476	RotateState
477
478################################### RND N + 2 #########################################
479
480	add	y2, old_h	# h = k + w + h + S0 + S1 + CH = t1 + S0# --
481	mov	f, y2		# y2 = f                                # CH
482	rorx	$41, e, y0	# y0 = e >> 41				# S1A
483	rorx	$18, e, y1	# y1 = e >> 18				# S1B
484	xor	g, y2		# y2 = f^g                              # CH
485
486	xor	y1, y0		# y0 = (e>>41) ^ (e>>18)		# S1
487	rorx	$14, e, y1	# y1 = (e >> 14)			# S1
488	and	e, y2		# y2 = (f^g)&e                          # CH
489	add	y3, old_h	# h = t1 + S0 + MAJ                     # --
490
491	xor	y1, y0		# y0 = (e>>41) ^ (e>>18) ^ (e>>14)	# S1
492	rorx	$34, a, T1	# T1 = a >> 34				# S0B
493	xor	g, y2		# y2 = CH = ((f^g)&e)^g                 # CH
494	rorx	$39, a, y1	# y1 = a >> 39				# S0A
495	mov	a, y3		# y3 = a                                # MAJA
496
497	xor	T1, y1		# y1 = (a>>39) ^ (a>>34)		# S0
498	rorx	$28, a, T1	# T1 = (a >> 28)			# S0
499	add	8*2+frame_XFER(%rsp), h		# h = k + w + h         # --
500	or	c, y3		# y3 = a|c                              # MAJA
501
502	xor	T1, y1		# y1 = (a>>39) ^ (a>>34) ^ (a>>28)	# S0
503	mov	a, T1		# T1 = a                                # MAJB
504	and	b, y3		# y3 = (a|c)&b                          # MAJA
505	and	c, T1		# T1 = a&c                              # MAJB
506	add	y0, y2		# y2 = S1 + CH                          # --
507
508	add	h, d		# d = k + w + h + d                     # --
509	or	T1, y3		# y3 = MAJ = (a|c)&b)|(a&c)             # MAJ
510	add	y1, h		# h = k + w + h + S0                    # --
511
512	add	y2, d		# d = k + w + h + d + S1 + CH = d + t1  # --
513
514	RotateState
515
516################################### RND N + 3 #########################################
517
518	add	y2, old_h	# h = k + w + h + S0 + S1 + CH = t1 + S0# --
519	mov	f, y2		# y2 = f                                # CH
520	rorx	$41, e, y0	# y0 = e >> 41				# S1A
521	rorx	$18, e, y1	# y1 = e >> 18				# S1B
522	xor	g, y2		# y2 = f^g                              # CH
523
524	xor	y1, y0		# y0 = (e>>41) ^ (e>>18)		# S1
525	rorx	$14, e, y1	# y1 = (e >> 14)			# S1
526	and	e, y2		# y2 = (f^g)&e                          # CH
527	add	y3, old_h	# h = t1 + S0 + MAJ                     # --
528
529	xor	y1, y0		# y0 = (e>>41) ^ (e>>18) ^ (e>>14)	# S1
530	rorx	$34, a, T1	# T1 = a >> 34				# S0B
531	xor	g, y2		# y2 = CH = ((f^g)&e)^g                 # CH
532	rorx	$39, a, y1	# y1 = a >> 39				# S0A
533	mov	a, y3		# y3 = a                                # MAJA
534
535	xor	T1, y1		# y1 = (a>>39) ^ (a>>34)		# S0
536	rorx	$28, a, T1	# T1 = (a >> 28)			# S0
537	add	8*3+frame_XFER(%rsp), h		# h = k + w + h         # --
538	or	c, y3		# y3 = a|c                              # MAJA
539
540	xor	T1, y1		# y1 = (a>>39) ^ (a>>34) ^ (a>>28)	# S0
541	mov	a, T1		# T1 = a                                # MAJB
542	and	b, y3		# y3 = (a|c)&b                          # MAJA
543	and	c, T1		# T1 = a&c                              # MAJB
544	add	y0, y2		# y2 = S1 + CH                          # --
545
546
547	add	h, d		# d = k + w + h + d                     # --
548	or	T1, y3		# y3 = MAJ = (a|c)&b)|(a&c)             # MAJ
549	add	y1, h		# h = k + w + h + S0                    # --
550
551	add	y2, d		# d = k + w + h + d + S1 + CH = d + t1  # --
552
553	add	y2, h		# h = k + w + h + S0 + S1 + CH = t1 + S0# --
554
555	add	y3, h		# h = t1 + S0 + MAJ                     # --
556
557	RotateState
558
559.endm
560
561########################################################################
562# void sha512_transform_rorx(sha512_state *state, const u8 *data, int blocks)
563# Purpose: Updates the SHA512 digest stored at "state" with the message
564# stored in "data".
565# The size of the message pointed to by "data" must be an integer multiple
566# of SHA512 message blocks.
567# "blocks" is the message length in SHA512 blocks
568########################################################################
569SYM_TYPED_FUNC_START(sha512_transform_rorx)
570	# Save GPRs
571	push	%rbx
572	push	%r12
573	push	%r13
574	push	%r14
575	push	%r15
576
577	# Allocate Stack Space
578	push	%rbp
579	mov	%rsp, %rbp
580	sub	$frame_size, %rsp
581	and	$~(0x20 - 1), %rsp
582
583	shl	$7, NUM_BLKS	# convert to bytes
584	jz	.Ldone_hash
585	add	INP, NUM_BLKS	# pointer to end of data
586	mov	NUM_BLKS, frame_INPEND(%rsp)
587
588	## load initial digest
589	mov	8*0(CTX1), a
590	mov	8*1(CTX1), b
591	mov	8*2(CTX1), c
592	mov	8*3(CTX1), d
593	mov	8*4(CTX1), e
594	mov	8*5(CTX1), f
595	mov	8*6(CTX1), g
596	mov	8*7(CTX1), h
597
598	# save %rdi (CTX) before it gets clobbered
599	mov	%rdi, frame_CTX(%rsp)
600
601	vmovdqa	PSHUFFLE_BYTE_FLIP_MASK(%rip), BYTE_FLIP_MASK
602
603.Lloop0:
604	lea	K512(%rip), TBL
605
606	## byte swap first 16 dwords
607	COPY_YMM_AND_BSWAP	Y_0, (INP), BYTE_FLIP_MASK
608	COPY_YMM_AND_BSWAP	Y_1, 1*32(INP), BYTE_FLIP_MASK
609	COPY_YMM_AND_BSWAP	Y_2, 2*32(INP), BYTE_FLIP_MASK
610	COPY_YMM_AND_BSWAP	Y_3, 3*32(INP), BYTE_FLIP_MASK
611
612	mov	INP, frame_INP(%rsp)
613
614	## schedule 64 input dwords, by doing 12 rounds of 4 each
615	movq	$4, frame_SRND(%rsp)
616
617.align 16
618.Lloop1:
619	vpaddq	(TBL), Y_0, XFER
620	vmovdqa XFER, frame_XFER(%rsp)
621	FOUR_ROUNDS_AND_SCHED
622
623	vpaddq	1*32(TBL), Y_0, XFER
624	vmovdqa XFER, frame_XFER(%rsp)
625	FOUR_ROUNDS_AND_SCHED
626
627	vpaddq	2*32(TBL), Y_0, XFER
628	vmovdqa XFER, frame_XFER(%rsp)
629	FOUR_ROUNDS_AND_SCHED
630
631	vpaddq	3*32(TBL), Y_0, XFER
632	vmovdqa XFER, frame_XFER(%rsp)
633	add	$(4*32), TBL
634	FOUR_ROUNDS_AND_SCHED
635
636	subq	$1, frame_SRND(%rsp)
637	jne	.Lloop1
638
639	movq	$2, frame_SRND(%rsp)
640.Lloop2:
641	vpaddq	(TBL), Y_0, XFER
642	vmovdqa XFER, frame_XFER(%rsp)
643	DO_4ROUNDS
644	vpaddq	1*32(TBL), Y_1, XFER
645	vmovdqa XFER, frame_XFER(%rsp)
646	add	$(2*32), TBL
647	DO_4ROUNDS
648
649	vmovdqa	Y_2, Y_0
650	vmovdqa	Y_3, Y_1
651
652	subq	$1, frame_SRND(%rsp)
653	jne	.Lloop2
654
655	mov	frame_CTX(%rsp), CTX2
656	addm	8*0(CTX2), a
657	addm	8*1(CTX2), b
658	addm	8*2(CTX2), c
659	addm	8*3(CTX2), d
660	addm	8*4(CTX2), e
661	addm	8*5(CTX2), f
662	addm	8*6(CTX2), g
663	addm	8*7(CTX2), h
664
665	mov	frame_INP(%rsp), INP
666	add	$128, INP
667	cmp	frame_INPEND(%rsp), INP
668	jne	.Lloop0
669
670.Ldone_hash:
671
672	# Restore Stack Pointer
673	mov	%rbp, %rsp
674	pop	%rbp
675
676	# Restore GPRs
677	pop	%r15
678	pop	%r14
679	pop	%r13
680	pop	%r12
681	pop	%rbx
682
683	RET
684SYM_FUNC_END(sha512_transform_rorx)
685
686########################################################################
687### Binary Data
688
689
690# Mergeable 640-byte rodata section. This allows linker to merge the table
691# with other, exactly the same 640-byte fragment of another rodata section
692# (if such section exists).
693.section	.rodata.cst640.K512, "aM", @progbits, 640
694.align 64
695# K[t] used in SHA512 hashing
696K512:
697	.quad	0x428a2f98d728ae22,0x7137449123ef65cd
698	.quad	0xb5c0fbcfec4d3b2f,0xe9b5dba58189dbbc
699	.quad	0x3956c25bf348b538,0x59f111f1b605d019
700	.quad	0x923f82a4af194f9b,0xab1c5ed5da6d8118
701	.quad	0xd807aa98a3030242,0x12835b0145706fbe
702	.quad	0x243185be4ee4b28c,0x550c7dc3d5ffb4e2
703	.quad	0x72be5d74f27b896f,0x80deb1fe3b1696b1
704	.quad	0x9bdc06a725c71235,0xc19bf174cf692694
705	.quad	0xe49b69c19ef14ad2,0xefbe4786384f25e3
706	.quad	0x0fc19dc68b8cd5b5,0x240ca1cc77ac9c65
707	.quad	0x2de92c6f592b0275,0x4a7484aa6ea6e483
708	.quad	0x5cb0a9dcbd41fbd4,0x76f988da831153b5
709	.quad	0x983e5152ee66dfab,0xa831c66d2db43210
710	.quad	0xb00327c898fb213f,0xbf597fc7beef0ee4
711	.quad	0xc6e00bf33da88fc2,0xd5a79147930aa725
712	.quad	0x06ca6351e003826f,0x142929670a0e6e70
713	.quad	0x27b70a8546d22ffc,0x2e1b21385c26c926
714	.quad	0x4d2c6dfc5ac42aed,0x53380d139d95b3df
715	.quad	0x650a73548baf63de,0x766a0abb3c77b2a8
716	.quad	0x81c2c92e47edaee6,0x92722c851482353b
717	.quad	0xa2bfe8a14cf10364,0xa81a664bbc423001
718	.quad	0xc24b8b70d0f89791,0xc76c51a30654be30
719	.quad	0xd192e819d6ef5218,0xd69906245565a910
720	.quad	0xf40e35855771202a,0x106aa07032bbd1b8
721	.quad	0x19a4c116b8d2d0c8,0x1e376c085141ab53
722	.quad	0x2748774cdf8eeb99,0x34b0bcb5e19b48a8
723	.quad	0x391c0cb3c5c95a63,0x4ed8aa4ae3418acb
724	.quad	0x5b9cca4f7763e373,0x682e6ff3d6b2b8a3
725	.quad	0x748f82ee5defb2fc,0x78a5636f43172f60
726	.quad	0x84c87814a1f0ab72,0x8cc702081a6439ec
727	.quad	0x90befffa23631e28,0xa4506cebde82bde9
728	.quad	0xbef9a3f7b2c67915,0xc67178f2e372532b
729	.quad	0xca273eceea26619c,0xd186b8c721c0c207
730	.quad	0xeada7dd6cde0eb1e,0xf57d4f7fee6ed178
731	.quad	0x06f067aa72176fba,0x0a637dc5a2c898a6
732	.quad	0x113f9804bef90dae,0x1b710b35131c471b
733	.quad	0x28db77f523047d84,0x32caab7b40c72493
734	.quad	0x3c9ebe0a15c9bebc,0x431d67c49c100d4c
735	.quad	0x4cc5d4becb3e42b6,0x597f299cfc657e2a
736	.quad	0x5fcb6fab3ad6faec,0x6c44198c4a475817
737
738.section	.rodata.cst32.PSHUFFLE_BYTE_FLIP_MASK, "aM", @progbits, 32
739.align 32
740# Mask for byte-swapping a couple of qwords in an XMM register using (v)pshufb.
741PSHUFFLE_BYTE_FLIP_MASK:
742	.octa 0x08090a0b0c0d0e0f0001020304050607
743	.octa 0x18191a1b1c1d1e1f1011121314151617
744
745.section	.rodata.cst32.MASK_YMM_LO, "aM", @progbits, 32
746.align 32
747MASK_YMM_LO:
748	.octa 0x00000000000000000000000000000000
749	.octa 0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF
750