xref: /isa-l_crypto/sha512_mb/sha512_mb_mgr_submit_avx512.asm (revision d28f1034f736e3eb791c3cf6bff3e2fa81fb5331)
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 "sha512_job.asm"
31%include "sha512_mb_mgr_datastruct.asm"
32%include "reg_sizes.asm"
33
34extern sha512_mb_x8_avx512
35
36%ifidn __OUTPUT_FORMAT__, elf64
37; LINUX register definitions
38%define arg1    rdi ; rcx
39%define arg2    rsi ; rdx
40
41; idx needs to be other than arg1, arg2, rbx, r12
42%define idx             rdx ; rsi
43%define last_len        rdx ; rsi
44
45%define size_offset     rcx ; rdi
46%define tmp2            rcx ; rdi
47
48%else
49; WINDOWS register definitions
50%define arg1    rcx
51%define arg2    rdx
52
53; idx needs to be other than arg1, arg2, rbx, r12
54%define last_len        rsi
55%define idx             rsi
56
57%define size_offset     rdi
58%define tmp2            rdi
59
60%endif
61
62; Common definitions
63%define state   arg1
64%define job     arg2
65%define len2    arg2
66%define p2      arg2
67
68%define p               r11
69%define start_offset    r11
70
71%define unused_lanes    rbx
72
73%define job_rax         rax
74%define len             rax
75
76%define lane            rbp
77%define tmp3            rbp
78%define lens3           rbp
79
80%define extra_blocks    r8
81%define lens0           r8
82
83%define num_lanes_inuse r9
84%define tmp             r9
85%define lens1           r9
86
87%define lane_data       r10
88%define lens2           r10
89
90struc stack_frame
91	.xmm: resb 16*10
92	.gpr: resb 8*8
93	.rsp: resb 8
94endstruc
95
96; STACK_SPACE needs to be an odd multiple of 8
97%define _XMM_SAVE       stack_frame.gpr
98%define _GPR_SAVE       stack_frame.rsp
99%define STACK_SPACE     stack_frame_size
100
101; ISAL_SHA512_JOB* _sha512_mb_mgr_submit_avx512(ISAL_SHA512_MB_JOB_MGR *state, ISAL_SHA512_JOB *job)
102; arg 1 : rcx : state
103; arg 2 : rdx : job
104mk_global _sha512_mb_mgr_submit_avx512, function, internal
105_sha512_mb_mgr_submit_avx512:
106	endbranch
107
108	mov     rax, rsp
109
110	sub     rsp, STACK_SPACE
111
112	mov     [rsp + stack_frame.rsp], rax
113
114	mov     [rsp + _XMM_SAVE + 8*0], rbx
115	mov     [rsp + _XMM_SAVE + 8*1], rbp
116	mov     [rsp + _XMM_SAVE + 8*2], r12
117	mov     [rsp + _XMM_SAVE + 8*5], r13
118	mov     [rsp + _XMM_SAVE + 8*6], r14
119	mov     [rsp + _XMM_SAVE + 8*7], r15
120%ifidn __OUTPUT_FORMAT__, win64
121	mov     [rsp + _XMM_SAVE + 8*3], rsi
122	mov     [rsp + _XMM_SAVE + 8*4], rdi
123	vmovdqu  [rsp + 16*0], xmm6
124	vmovdqu  [rsp + 16*1], xmm7
125	vmovdqu  [rsp + 16*2], xmm8
126	vmovdqu  [rsp + 16*3], xmm9
127	vmovdqu  [rsp + 16*4], xmm10
128	vmovdqu  [rsp + 16*5], xmm11
129	vmovdqu  [rsp + 16*6], xmm12
130	vmovdqu  [rsp + 16*7], xmm13
131	vmovdqu  [rsp + 16*8], xmm14
132	vmovdqu  [rsp + 16*9], xmm15
133%endif
134
135	mov     unused_lanes, [state + _unused_lanes]
136	movzx   lane, BYTE(unused_lanes)
137	shr     unused_lanes, 8
138	imul    lane_data, lane, _LANE_DATA_size
139	mov     dword [job + _status], ISAL_STS_BEING_PROCESSED
140	lea     lane_data, [state + _ldata + lane_data]
141	mov     [state + _unused_lanes], unused_lanes
142	mov     DWORD(len), [job + _len]
143
144	mov     [lane_data + _job_in_lane], job
145	mov     [state + _lens + 4 + 8*lane], DWORD(len)
146
147
148	; Load digest words from result_digest
149	vmovdqa  xmm0, [job + _result_digest + 0*16]
150	vmovdqa  xmm1, [job + _result_digest + 1*16]
151	vmovdqa  xmm2, [job + _result_digest + 2*16]
152	vmovdqa  xmm3, [job + _result_digest + 3*16]
153	vmovq    [state + _args_digest + 8*lane + 0*64], xmm0
154	vpextrq  [state + _args_digest + 8*lane + 1*64], xmm0, 1
155	vmovq    [state + _args_digest + 8*lane + 2*64], xmm1
156	vpextrq  [state + _args_digest + 8*lane + 3*64], xmm1, 1
157	vmovq    [state + _args_digest + 8*lane + 4*64], xmm2
158	vpextrq  [state + _args_digest + 8*lane + 5*64], xmm2, 1
159	vmovq    [state + _args_digest + 8*lane + 6*64], xmm3
160	vpextrq  [state + _args_digest + 8*lane + 7*64], xmm3, 1
161
162	mov     p, [job + _buffer]
163	mov     [state + _args_data_ptr + 8*lane], p
164
165	mov	DWORD(num_lanes_inuse), [state + _num_lanes_inuse]
166        add     num_lanes_inuse, 1
167	mov	[state + _num_lanes_inuse], DWORD(num_lanes_inuse)
168        cmp     num_lanes_inuse, 8
169	jne     return_null
170
171start_loop:
172	; Find min length, len in sha512_mgr is 64bit, high 32bit is block num, low 8bit is idx
173	vmovdqu ymm0, [state + _lens + 0*32]	; ymm0 has {D,d,C,c,B,b,A,a}
174	vmovdqu ymm1, [state + _lens + 1*32]
175
176	vpminuq ymm2, ymm0, ymm1	; ymm2 has {D,i,C,i,B,i,A,i}
177	vpalignr ymm3, ymm3, ymm2, 8	; ymm3 has {x,i,D,i,x,i,B,i}
178	vpminuq ymm2, ymm2, ymm3	; ymm2 has {x,i,F,i,x,i,E,i}
179	vperm2i128 ymm3, ymm2, ymm2, 1	; ymm3 has {x,i,x,i,x,i,F,i}
180	vpminuq ymm2, ymm2, ymm3	; ymm2 has min value in high dword
181
182	vmovq   idx, xmm2
183	mov     len2, idx
184	and     idx, 0xF
185	shr     len2, 32
186	jz      len_is_0
187
188
189	vperm2i128 ymm2, ymm2, ymm2, 0	; ymm2 has {x,x,E,i,x,x,E,i}
190	vpand   ymm2, ymm2, [rel clear_low_nibble]	; ymm2 has {0,0,E,0,0,0,E,0}
191	vpshufd ymm2, ymm2, 0x44	; ymm2 has {E,0,E,0,E,0,E,0}
192
193	vpsubd  ymm0, ymm0, ymm2
194	vpsubd  ymm1, ymm1, ymm2
195
196	vmovdqu [state + _lens + 0*32], ymm0
197	vmovdqu [state + _lens + 1*32], ymm1
198
199	; "state" and "args" are the same address, arg1
200	; len is arg2
201	call    sha512_mb_x8_avx512
202	; state and idx are intact
203
204len_is_0:
205
206	; process completed job "idx"
207	imul    lane_data, idx, _LANE_DATA_size
208	lea     lane_data, [state + _ldata + lane_data]
209
210	mov     job_rax, [lane_data + _job_in_lane]
211
212
213	mov     unused_lanes, [state + _unused_lanes]
214	mov     qword [lane_data + _job_in_lane], 0
215	mov     dword [job_rax + _status], ISAL_STS_COMPLETED
216	shl     unused_lanes, 8
217	or      unused_lanes, idx
218	mov     [state + _unused_lanes], unused_lanes
219
220        mov     DWORD(num_lanes_inuse), [state + _num_lanes_inuse]
221        sub     num_lanes_inuse, 1
222        mov     [state + _num_lanes_inuse], DWORD(num_lanes_inuse)
223	vmovq    xmm0, [state + _args_digest + 8*idx + 0*64]
224	vpinsrq  xmm0, [state + _args_digest + 8*idx + 1*64], 1
225	vmovq    xmm1, [state + _args_digest + 8*idx + 2*64]
226	vpinsrq  xmm1, [state + _args_digest + 8*idx + 3*64], 1
227	vmovq    xmm2, [state + _args_digest + 8*idx + 4*64]
228	vpinsrq  xmm2, [state + _args_digest + 8*idx + 5*64], 1
229	vmovq    xmm3, [state + _args_digest + 8*idx + 6*64]
230	vpinsrq  xmm3, [state + _args_digest + 8*idx + 7*64], 1
231	vmovdqa  [job_rax + _result_digest + 0*16], xmm0
232	vmovdqa  [job_rax + _result_digest + 1*16], xmm1
233	vmovdqa  [job_rax + _result_digest + 2*16], xmm2
234	vmovdqa  [job_rax + _result_digest + 3*16], xmm3
235
236return:
237
238%ifidn __OUTPUT_FORMAT__, win64
239	vmovdqu  xmm6,  [rsp + 16*0]
240	vmovdqu  xmm7,  [rsp + 16*1]
241	vmovdqu  xmm8,  [rsp + 16*2]
242	vmovdqu  xmm9,  [rsp + 16*3]
243	vmovdqu  xmm10, [rsp + 16*4]
244	vmovdqu  xmm11, [rsp + 16*5]
245	vmovdqu  xmm12, [rsp + 16*6]
246	vmovdqu  xmm13, [rsp + 16*7]
247	vmovdqu  xmm14, [rsp + 16*8]
248	vmovdqu  xmm15, [rsp + 16*9]
249	mov     rsi, [rsp + _XMM_SAVE + 8*3]
250	mov     rdi, [rsp + _XMM_SAVE + 8*4]
251%endif
252	mov     rbx, [rsp + _XMM_SAVE + 8*0]
253	mov     rbp, [rsp + _XMM_SAVE + 8*1]
254	mov     r12, [rsp + _XMM_SAVE + 8*2]
255	mov     r13, [rsp + _XMM_SAVE + 8*5]
256	mov     r14, [rsp + _XMM_SAVE + 8*6]
257	mov     r15, [rsp + _XMM_SAVE + 8*7]
258
259	mov	rsp, [rsp + stack_frame.rsp]
260
261	ret
262
263return_null:
264	xor     job_rax, job_rax
265	jmp     return
266
267section .data align=32
268
269align 32
270clear_low_nibble:	; mgr len element 0xnnnnnnnn 0000000m, nnnnnnnn is blocknum, m is index
271	dq 0xFFFFFFFF00000000, 0x0000000000000000
272	dq 0xFFFFFFFF00000000, 0x0000000000000000
273