xref: /isa-l_crypto/sha512_mb/sha512_sse4.asm (revision d57a697cfda8b34985b3e1eba01f8a39e8486ff3)
1;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
2;  Copyright(c) 2011-2016 Intel Corporation All rights reserved.
3;
4;  Redistribution and use in source and binary forms, with or without
5;  modification, are permitted provided that the following conditions
6;  are met:
7;    * Redistributions of source code must retain the above copyright
8;      notice, this list of conditions and the following disclaimer.
9;    * Redistributions in binary form must reproduce the above copyright
10;      notice, this list of conditions and the following disclaimer in
11;      the documentation and/or other materials provided with the
12;      distribution.
13;    * Neither the name of Intel Corporation nor the names of its
14;      contributors may be used to endorse or promote products derived
15;      from this software without specific prior written permission.
16;
17;  THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
18;  "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
19;  LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
20;  A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
21;  OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
22;  SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
23;  LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
24;  DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
25;  THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
26;  (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
27;  OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
29
30%include "reg_sizes.asm"
31
32[bits 64]
33default rel
34section .text
35
36; Virtual Registers
37%ifidn __OUTPUT_FORMAT__, win64
38	%define msg     rcx ; ARG1
39	%define digest  rdx ; ARG2
40	%define msglen  r8  ; ARG3
41	%define T1      rsi
42	%define T2      rdi
43%else
44	%define msg     rdi ; ARG1
45	%define digest  rsi ; ARG2
46	%define msglen  rdx ; ARG3
47	%define T1      rcx
48	%define T2      r8
49%endif
50%define a_64    r9
51%define b_64    r10
52%define c_64    r11
53%define d_64    r12
54%define e_64    r13
55%define f_64    r14
56%define g_64    r15
57%define h_64    rbx
58%define tmp0    rax
59
60; Local variables (stack frame)
61; Note: frame_size must be an odd multiple of 8 bytes to XMM align RSP
62struc frame
63	.W:       resq 80 ; Message Schedule
64	.WK:      resq  2 ; W[t] + K[t] | W[t+1] + K[t+1]
65
66%ifidn __OUTPUT_FORMAT__, win64
67	.GPRSAVE: resq 7
68%else
69	.GPRSAVE: resq 5
70%endif
71endstruc
72
73; Useful QWORD "arrays" for simpler memory references
74%define MSG(i)    msg    + 8*(i)               ; Input message (arg1)
75%define DIGEST(i) digest + 8*(i)               ; Output Digest (arg2)
76%define K_t(i)    K512   + 8*(i)               ; SHA Constants (static mem)
77%define W_t(i)    rsp + frame.W  + 8*(i)       ; Message Schedule (stack frame)
78%define WK_2(i)   rsp + frame.WK + 8*((i) % 2) ; W[t]+K[t] (stack frame)
79; MSG, DIGEST, K_t, W_t are arrays
80; WK_2(t) points to 1 of 2 qwords at frame.WK depending on t being odd/even
81
82%macro RotateState 0
83	; Rotate symbles a..h right
84	%xdefine %%TMP h_64
85	%xdefine h_64  g_64
86	%xdefine g_64  f_64
87	%xdefine f_64  e_64
88	%xdefine e_64  d_64
89	%xdefine d_64  c_64
90	%xdefine c_64  b_64
91	%xdefine b_64  a_64
92	%xdefine a_64  %%TMP
93%endmacro
94
95%macro SHA512_Round 1
96%assign %%t   (%1)
97
98	; Compute Round %%t
99	mov     T1,   f_64        ; T1 = f
100	mov     tmp0, e_64        ; tmp = e
101	xor     T1,   g_64        ; T1 = f ^ g
102	ror     tmp0, 23 ; 41     ; tmp = e ror 23
103	and     T1,   e_64        ; T1 = (f ^ g) & e
104	xor     tmp0, e_64        ; tmp = (e ror 23) ^ e
105	xor     T1,   g_64        ; T1 = ((f ^ g) & e) ^ g = CH(e,f,g)
106	add     T1,   [WK_2(%%t)] ; W[t] + K[t] from message scheduler
107	ror     tmp0, 4 ; 18      ; tmp = ((e ror 23) ^ e) ror 4
108	xor     tmp0, e_64        ; tmp = (((e ror 23) ^ e) ror 4) ^ e
109	mov     T2,   a_64        ; T2 = a
110	add     T1,   h_64        ; T1 = CH(e,f,g) + W[t] + K[t] + h
111	ror     tmp0, 14 ; 14     ; tmp = ((((e ror23)^e)ror4)^e)ror14 = S1(e)
112	add     T1,   tmp0        ; T1 = CH(e,f,g) + W[t] + K[t] + S1(e)
113	mov     tmp0, a_64        ; tmp = a
114	xor     T2,   c_64        ; T2 = a ^ c
115	and     tmp0, c_64        ; tmp = a & c
116	and     T2,   b_64        ; T2 = (a ^ c) & b
117	xor     T2,   tmp0        ; T2 = ((a ^ c) & b) ^ (a & c) = Maj(a,b,c)
118	mov     tmp0, a_64        ; tmp = a
119	ror     tmp0, 5 ; 39      ; tmp = a ror 5
120	xor     tmp0, a_64        ; tmp = (a ror 5) ^ a
121	add     d_64, T1          ; e(next_state) = d + T1
122	ror     tmp0, 6 ; 34      ; tmp = ((a ror 5) ^ a) ror 6
123	xor     tmp0, a_64        ; tmp = (((a ror 5) ^ a) ror 6) ^ a
124	lea     h_64, [T1 + T2]   ; a(next_state) = T1 + Maj(a,b,c)
125	ror     tmp0, 28 ; 28     ; tmp = ((((a ror5)^a)ror6)^a)ror28 = S0(a)
126	add     h_64, tmp0        ; a(next_state) = T1 + Maj(a,b,c) S0(a)
127	RotateState
128%endmacro
129
130%macro SHA512_2Sched_2Round_sse 1
131%assign %%t (%1)
132
133	; Compute rounds %%t-2 and %%t-1
134	; Compute message schedule QWORDS %%t and %%t+1
135
136	;   Two rounds are computed based on the values for K[t-2]+W[t-2] and
137	; K[t-1]+W[t-1] which were previously stored at WK_2 by the message
138	; scheduler.
139	;   The two new schedule QWORDS are stored at [W_t(%%t)] and [W_t(%%t+1)].
140	; They are then added to their respective SHA512 constants at
141	; [K_t(%%t)] and [K_t(%%t+1)] and stored at dqword [WK_2(%%t)]
142	;   For brievity, the comments following vectored instructions only refer to
143	; the first of a pair of QWORDS.
144	; Eg. XMM2=W[t-2] really means XMM2={W[t-2]|W[t-1]}
145	;   The computation of the message schedule and the rounds are tightly
146	; stitched to take advantage of instruction-level parallelism.
147	; For clarity, integer instructions (for the rounds calculation) are indented
148	; by one tab. Vectored instructions (for the message scheduler) are indented
149	; by two tabs.
150
151	mov     T1, f_64
152	movdqa  xmm2, [W_t(%%t-2)]  ; XMM2 = W[t-2]
153	xor     T1,   g_64
154	and     T1,   e_64
155	movdqa  xmm0, xmm2          ; XMM0 = W[t-2]
156	xor     T1,   g_64
157	add     T1,   [WK_2(%%t)]
158	movdqu  xmm5, [W_t(%%t-15)] ; XMM5 = W[t-15]
159	mov     tmp0, e_64
160	ror     tmp0, 23 ; 41
161	movdqa  xmm3, xmm5          ; XMM3 = W[t-15]
162	xor     tmp0, e_64
163	ror     tmp0, 4 ; 18
164	psrlq   xmm0, 61 - 19       ; XMM0 = W[t-2] >> 42
165	xor     tmp0, e_64
166	ror     tmp0, 14 ; 14
167	psrlq   xmm3, (8 - 7)       ; XMM3 = W[t-15] >> 1
168	add     T1,   tmp0
169	add     T1,   h_64
170	pxor    xmm0, xmm2          ; XMM0 = (W[t-2] >> 42) ^ W[t-2]
171	mov     T2,   a_64
172	xor     T2,   c_64
173	pxor    xmm3, xmm5          ; XMM3 = (W[t-15] >> 1) ^ W[t-15]
174	and     T2,   b_64
175	mov     tmp0, a_64
176	psrlq   xmm0, 19 - 6        ; XMM0 = ((W[t-2]>>42)^W[t-2])>>13
177	and     tmp0, c_64
178	xor     T2,   tmp0
179	psrlq   xmm3, (7 - 1)       ; XMM3 = ((W[t-15]>>1)^W[t-15])>>6
180	mov     tmp0, a_64
181	ror     tmp0, 5 ; 39
182	pxor    xmm0, xmm2          ; XMM0 = (((W[t-2]>>42)^W[t-2])>>13)^W[t-2]
183	xor     tmp0, a_64
184	ror     tmp0, 6 ; 34
185	pxor    xmm3, xmm5          ; XMM3 = (((W[t-15]>>1)^W[t-15])>>6)^W[t-15]
186	xor     tmp0, a_64
187	ror     tmp0, 28 ; 28
188	psrlq   xmm0, 6             ; XMM0 = ((((W[t-2]>>42)^W[t-2])>>13)^W[t-2])>>6
189	add     T2,   tmp0
190	add     d_64, T1
191	psrlq   xmm3, 1             ; XMM3 = (((W[t-15]>>1)^W[t-15])>>6)^W[t-15]>>1
192	lea     h_64, [T1 + T2]
193	RotateState
194	movdqa  xmm1, xmm2          ; XMM1 = W[t-2]
195	mov     T1, f_64
196	xor     T1,   g_64
197	movdqa  xmm4, xmm5          ; XMM4 = W[t-15]
198	and     T1,   e_64
199	xor     T1,   g_64
200	psllq   xmm1, (64 - 19) - (64 - 61) ; XMM1 = W[t-2] << 42
201	add     T1,   [WK_2(%%t+1)]
202	mov     tmp0, e_64
203	psllq   xmm4, (64 - 1) - (64 - 8) ; XMM4 = W[t-15] << 7
204	ror     tmp0, 23 ; 41
205	xor     tmp0, e_64
206	pxor    xmm1, xmm2          ; XMM1 = (W[t-2] << 42)^W[t-2]
207	ror     tmp0, 4 ; 18
208	xor     tmp0, e_64
209	pxor    xmm4, xmm5          ; XMM4 = (W[t-15]<<7)^W[t-15]
210	ror     tmp0, 14 ; 14
211	add     T1,   tmp0
212	psllq   xmm1, (64 - 61)     ; XMM1 = ((W[t-2] << 42)^W[t-2])<<3
213	add     T1,   h_64
214	mov     T2,   a_64
215	psllq   xmm4, (64 - 8)      ; XMM4 = ((W[t-15]<<7)^W[t-15])<<56
216	xor     T2,   c_64
217	and     T2,   b_64
218	pxor    xmm0, xmm1          ; XMM0 = s1(W[t-2])
219	mov     tmp0, a_64
220	and     tmp0, c_64
221	movdqu  xmm1, [W_t(%%t- 7)] ; XMM1 = W[t-7]
222	xor     T2,   tmp0
223	pxor    xmm3, xmm4          ; XMM3 = s0(W[t-15])
224	mov     tmp0, a_64
225	paddq   xmm0, xmm3          ; XMM0 = s1(W[t-2]) + s0(W[t-15])
226	ror     tmp0, 5 ; 39
227	paddq   xmm0, [W_t(%%t-16)] ; XMM0 = s1(W[t-2]) + s0(W[t-15]) + W[t-16]
228	xor     tmp0, a_64
229	paddq   xmm0, xmm1          ; XMM0 = s1(W[t-2]) + W[t-7] + s0(W[t-15]) + W[t-16]
230	ror     tmp0, 6 ; 34
231	movdqa  [W_t(%%t)], xmm0    ; Store scheduled qwords
232	xor     tmp0, a_64
233	paddq   xmm0, [K_t(t)]      ; Compute W[t]+K[t]
234	ror     tmp0, 28 ; 28
235	movdqa  [WK_2(t)], xmm0     ; Store W[t]+K[t] for next rounds
236	add     T2,   tmp0
237	add     d_64, T1
238	lea     h_64, [T1 + T2]
239	RotateState
240%endmacro
241
242;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
243; void _sha512_sse4(const void* M, void* D, uint64_t L);
244; Purpose: Updates the SHA512 digest stored at D with the message stored in M.
245; The size of the message pointed to by M must be an integer multiple of SHA512
246;   message blocks.
247; L is the message length in SHA512 blocks.
248mk_global _sha512_sse4, function, internal
249_sha512_sse4:
250	endbranch
251	cmp msglen, 0
252	je .nowork
253
254	; Allocate Stack Space
255	sub     rsp, frame_size
256
257	; Save GPRs
258	mov     [rsp + frame.GPRSAVE + 8 * 0], rbx
259	mov     [rsp + frame.GPRSAVE + 8 * 1], r12
260	mov     [rsp + frame.GPRSAVE + 8 * 2], r13
261	mov     [rsp + frame.GPRSAVE + 8 * 3], r14
262	mov     [rsp + frame.GPRSAVE + 8 * 4], r15
263%ifidn __OUTPUT_FORMAT__, win64
264	mov     [rsp + frame.GPRSAVE + 8 * 5], rsi
265	mov     [rsp + frame.GPRSAVE + 8 * 6], rdi
266%endif
267
268.updateblock:
269
270	; Load state variables
271	mov     a_64, [DIGEST(0)]
272	mov     b_64, [DIGEST(1)]
273	mov     c_64, [DIGEST(2)]
274	mov     d_64, [DIGEST(3)]
275	mov     e_64, [DIGEST(4)]
276	mov     f_64, [DIGEST(5)]
277	mov     g_64, [DIGEST(6)]
278	mov     h_64, [DIGEST(7)]
279
280	%assign t 0
281	%rep 80/2 + 1
282	; (80 rounds) / (2 rounds/iteration) + (1 iteration)
283	; +1 iteration because the scheduler leads hashing by 1 iteration
284		%if t < 2
285			; BSWAP 2 QWORDS
286			movdqa  xmm1, [XMM_QWORD_BSWAP]
287			movdqu  xmm0, [MSG(t)]
288			pshufb  xmm0, xmm1      ; BSWAP
289			movdqa  [W_t(t)], xmm0  ; Store Scheduled Pair
290			paddq   xmm0, [K_t(t)]  ; Compute W[t]+K[t]
291			movdqa  [WK_2(t)], xmm0 ; Store into WK for rounds
292		%elif t < 16
293			; BSWAP 2 QWORDS; Compute 2 Rounds
294			movdqu  xmm0, [MSG(t)]
295			pshufb  xmm0, xmm1      ; BSWAP
296			SHA512_Round t - 2      ; Round t-2
297			movdqa  [W_t(t)], xmm0  ; Store Scheduled Pair
298			paddq   xmm0, [K_t(t)]  ; Compute W[t]+K[t]
299			SHA512_Round t - 1      ; Round t-1
300			movdqa  [WK_2(t)], xmm0 ; Store W[t]+K[t] into WK
301		%elif t < 79
302			; Schedule 2 QWORDS; Compute 2 Rounds
303			SHA512_2Sched_2Round_sse t
304		%else
305			; Compute 2 Rounds
306			SHA512_Round t - 2
307			SHA512_Round t - 1
308		%endif
309	%assign t t+2
310	%endrep
311
312	; Update digest
313	add     [DIGEST(0)], a_64
314	add     [DIGEST(1)], b_64
315	add     [DIGEST(2)], c_64
316	add     [DIGEST(3)], d_64
317	add     [DIGEST(4)], e_64
318	add     [DIGEST(5)], f_64
319	add     [DIGEST(6)], g_64
320	add     [DIGEST(7)], h_64
321
322	; Advance to next message block
323	add     msg, 16*8
324	dec     msglen
325	jnz     .updateblock
326
327	; Restore GPRs
328	mov     rbx, [rsp + frame.GPRSAVE + 8 * 0]
329	mov     r12, [rsp + frame.GPRSAVE + 8 * 1]
330	mov     r13, [rsp + frame.GPRSAVE + 8 * 2]
331	mov     r14, [rsp + frame.GPRSAVE + 8 * 3]
332	mov     r15, [rsp + frame.GPRSAVE + 8 * 4]
333%ifidn __OUTPUT_FORMAT__, win64
334	mov     rsi, [rsp + frame.GPRSAVE + 8 * 5]
335	mov     rdi, [rsp + frame.GPRSAVE + 8 * 6]
336%endif
337	; Restore Stack Pointer
338	add     rsp, frame_size
339
340.nowork:
341	ret
342
343;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
344;;; Binary Data
345
346section .data
347
348ALIGN 16
349
350; Mask for byte-swapping a couple of qwords in an XMM register using (v)pshufb.
351XMM_QWORD_BSWAP:
352	dq 0x0001020304050607, 0x08090a0b0c0d0e0f
353
354; K[t] used in SHA512 hashing
355K512:
356	dq 0x428a2f98d728ae22,0x7137449123ef65cd
357	dq 0xb5c0fbcfec4d3b2f,0xe9b5dba58189dbbc
358	dq 0x3956c25bf348b538,0x59f111f1b605d019
359	dq 0x923f82a4af194f9b,0xab1c5ed5da6d8118
360	dq 0xd807aa98a3030242,0x12835b0145706fbe
361	dq 0x243185be4ee4b28c,0x550c7dc3d5ffb4e2
362	dq 0x72be5d74f27b896f,0x80deb1fe3b1696b1
363	dq 0x9bdc06a725c71235,0xc19bf174cf692694
364	dq 0xe49b69c19ef14ad2,0xefbe4786384f25e3
365	dq 0x0fc19dc68b8cd5b5,0x240ca1cc77ac9c65
366	dq 0x2de92c6f592b0275,0x4a7484aa6ea6e483
367	dq 0x5cb0a9dcbd41fbd4,0x76f988da831153b5
368	dq 0x983e5152ee66dfab,0xa831c66d2db43210
369	dq 0xb00327c898fb213f,0xbf597fc7beef0ee4
370	dq 0xc6e00bf33da88fc2,0xd5a79147930aa725
371	dq 0x06ca6351e003826f,0x142929670a0e6e70
372	dq 0x27b70a8546d22ffc,0x2e1b21385c26c926
373	dq 0x4d2c6dfc5ac42aed,0x53380d139d95b3df
374	dq 0x650a73548baf63de,0x766a0abb3c77b2a8
375	dq 0x81c2c92e47edaee6,0x92722c851482353b
376	dq 0xa2bfe8a14cf10364,0xa81a664bbc423001
377	dq 0xc24b8b70d0f89791,0xc76c51a30654be30
378	dq 0xd192e819d6ef5218,0xd69906245565a910
379	dq 0xf40e35855771202a,0x106aa07032bbd1b8
380	dq 0x19a4c116b8d2d0c8,0x1e376c085141ab53
381	dq 0x2748774cdf8eeb99,0x34b0bcb5e19b48a8
382	dq 0x391c0cb3c5c95a63,0x4ed8aa4ae3418acb
383	dq 0x5b9cca4f7763e373,0x682e6ff3d6b2b8a3
384	dq 0x748f82ee5defb2fc,0x78a5636f43172f60
385	dq 0x84c87814a1f0ab72,0x8cc702081a6439ec
386	dq 0x90befffa23631e28,0xa4506cebde82bde9
387	dq 0xbef9a3f7b2c67915,0xc67178f2e372532b
388	dq 0xca273eceea26619c,0xd186b8c721c0c207
389	dq 0xeada7dd6cde0eb1e,0xf57d4f7fee6ed178
390	dq 0x06f067aa72176fba,0x0a637dc5a2c898a6
391	dq 0x113f9804bef90dae,0x1b710b35131c471b
392	dq 0x28db77f523047d84,0x32caab7b40c72493
393	dq 0x3c9ebe0a15c9bebc,0x431d67c49c100d4c
394	dq 0x4cc5d4becb3e42b6,0x597f299cfc657e2a
395	dq 0x5fcb6fab3ad6faec,0x6c44198c4a475817
396
397