[ceph.git] / ceph / src / spdk / intel-ipsec-mb / avx / mb_mgr_hmac_md5_flush_avx.asm

;;
;; Copyright (c) 2012-2018, Intel Corporation
;;
;; Redistribution and use in source and binary forms, with or without
;; modification, are permitted provided that the following conditions are met:
;;
;;     * Redistributions of source code must retain the above copyright notice,
;;       this list of conditions and the following disclaimer.
;;     * Redistributions in binary form must reproduce the above copyright
;;       notice, this list of conditions and the following disclaimer in the
;;       documentation and/or other materials provided with the distribution.
;;     * Neither the name of Intel Corporation nor the names of its contributors
;;       may be used to endorse or promote products derived from this software
;;       without specific prior written permission.
;;
;; THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
;; AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
;; IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
;; DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE
;; FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
;; DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
;; SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
;; CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
;; OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
;; OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
;;

%include "os.asm"
%include "job_aes_hmac.asm"
%include "mb_mgr_datastruct.asm"
%include "reg_sizes.asm"

extern md5_x4x2_avx

section .data
default rel
align 16
dupw:	;ddq 0x01000100010001000100010001000100
	dq 0x0100010001000100, 0x0100010001000100
x80:    ;ddq 0x00000000000000000000000000000080
        dq 0x0000000000000080, 0x0000000000000000
x00:    ;ddq 0x00000000000000000000000000000000
        dq 0x0000000000000000, 0x0000000000000000
len_masks:
	;ddq 0x0000000000000000000000000000FFFF
	dq 0x000000000000FFFF, 0x0000000000000000
	;ddq 0x000000000000000000000000FFFF0000
	dq 0x00000000FFFF0000, 0x0000000000000000
	;ddq 0x00000000000000000000FFFF00000000
	dq 0x0000FFFF00000000, 0x0000000000000000
	;ddq 0x0000000000000000FFFF000000000000
	dq 0xFFFF000000000000, 0x0000000000000000
	;ddq 0x000000000000FFFF0000000000000000
	dq 0x0000000000000000, 0x000000000000FFFF
	;ddq 0x00000000FFFF00000000000000000000
	dq 0x0000000000000000, 0x00000000FFFF0000
	;ddq 0x0000FFFF000000000000000000000000
	dq 0x0000000000000000, 0x0000FFFF00000000
	;ddq 0xFFFF0000000000000000000000000000
	dq 0x0000000000000000, 0xFFFF000000000000
one:	dq  1
two:	dq  2
three:	dq  3
four:	dq  4
five:	dq  5
six:	dq  6
seven:	dq  7

section .text

%if 1
%ifdef LINUX
%define arg1	rdi
%define arg2	rsi
%else
%define arg1	rcx
%define arg2	rdx
%endif

%define state	arg1
%define job	arg2
%define len2	arg2


; idx needs to be in rbp
%define idx             rbp

; unused_lanes must be in rax-rdx
%define unused_lanes    rbx
%define lane_data       rbx
%define tmp2		rbx

%define job_rax         rax
%define	tmp1		rax
%define size_offset     rax
%define tmp             rax
%define start_offset    rax

%define tmp3		arg1

%define extra_blocks    arg2
%define p               arg2

%define tmp4		r8
%define tmp5		r9

%endif

; This routine and/or the called routine clobbers all GPRs
struc STACK
_gpr_save:	resq	8
_rsp_save:	resq	1
endstruc

%define APPEND(a,b) a %+ b

; JOB* flush_job_hmac_md5_avx(MB_MGR_HMAC_MD5_OOO *state)
; arg 1 : rcx : state
MKGLOBAL(flush_job_hmac_md5_avx,function,internal)
flush_job_hmac_md5_avx:

        mov	rax, rsp
        sub	rsp, STACK_size
        and	rsp, -16

	mov	[rsp + _gpr_save + 8*0], rbx
	mov	[rsp + _gpr_save + 8*1], rbp
	mov	[rsp + _gpr_save + 8*2], r12
	mov	[rsp + _gpr_save + 8*3], r13
	mov	[rsp + _gpr_save + 8*4], r14
	mov	[rsp + _gpr_save + 8*5], r15
%ifndef LINUX
	mov	[rsp + _gpr_save + 8*6], rsi
	mov	[rsp + _gpr_save + 8*7], rdi
%endif
	mov	[rsp + _rsp_save], rax	; original SP

	mov	unused_lanes, [state + _unused_lanes_md5]
	bt	unused_lanes, 32+3
	jc	return_null

	; find a lane with a non-null job
	xor	idx, idx
	cmp	qword [state + _ldata_md5 + 1 * _HMAC_SHA1_LANE_DATA_size + _job_in_lane],0
	cmovne	idx, [rel one]
	cmp	qword [state + _ldata_md5 + 2 * _HMAC_SHA1_LANE_DATA_size + _job_in_lane],0
	cmovne	idx, [rel two]
	cmp	qword [state + _ldata_md5 + 3 * _HMAC_SHA1_LANE_DATA_size + _job_in_lane],0
	cmovne	idx, [rel three]
	cmp	qword [state + _ldata_md5 + 4 * _HMAC_SHA1_LANE_DATA_size + _job_in_lane],0
	cmovne	idx, [rel four]
	cmp	qword [state + _ldata_md5 + 5 * _HMAC_SHA1_LANE_DATA_size + _job_in_lane],0
	cmovne	idx, [rel five]
	cmp	qword [state + _ldata_md5 + 6 * _HMAC_SHA1_LANE_DATA_size + _job_in_lane],0
	cmovne	idx, [rel six]
	cmp	qword [state + _ldata_md5 + 7 * _HMAC_SHA1_LANE_DATA_size + _job_in_lane],0
	cmovne	idx, [rel seven]

copy_lane_data:
	; copy good lane (idx) to empty lanes
	vmovdqa	xmm0, [state + _lens_md5]
	mov	tmp, [state + _args_data_ptr_md5 + PTR_SZ*idx]

%assign I 0
%rep 8
	cmp	qword [state + _ldata_md5 + I * _HMAC_SHA1_LANE_DATA_size + _job_in_lane], 0
	jne	APPEND(skip_,I)
	mov	[state + _args_data_ptr_md5 + PTR_SZ*I], tmp
	vpor	xmm0, xmm0, [rel len_masks + 16*I]
APPEND(skip_,I):
%assign I (I+1)
%endrep

	vmovdqa	[state + _lens_md5], xmm0

	vphminposuw	xmm1, xmm0
	vpextrw	DWORD(len2), xmm1, 0	; min value
	vpextrw	DWORD(idx), xmm1, 1	; min index (0...3)
	cmp	len2, 0
	je	len_is_0

	vpshufb	xmm1, [rel dupw]	; duplicate words across all lanes
	vpsubw	xmm0, xmm0, xmm1
	vmovdqa	[state + _lens_md5], xmm0

	; "state" and "args" are the same address, arg1
	; len is arg2
	call	md5_x4x2_avx
	; state and idx are intact

len_is_0:
	; process completed job "idx"
	imul	lane_data, idx, _HMAC_SHA1_LANE_DATA_size
	lea	lane_data, [state + _ldata_md5 + lane_data]
	mov	DWORD(extra_blocks), [lane_data + _extra_blocks]
	cmp	extra_blocks, 0
	jne	proc_extra_blocks
	cmp	dword [lane_data + _outer_done], 0
	jne	end_loop

proc_outer:
	mov	dword [lane_data + _outer_done], 1
	mov	DWORD(size_offset), [lane_data + _size_offset]
	mov	qword [lane_data + _extra_block + size_offset], 0
	mov	word [state + _lens_md5 + 2*idx], 1
	lea	tmp, [lane_data + _outer_block]
	mov	job, [lane_data + _job_in_lane]
	mov	[state + _args_data_ptr_md5 + PTR_SZ*idx], tmp

	vmovd	xmm0, [state + _args_digest_md5 + MD5_DIGEST_WORD_SIZE*idx + 0*MD5_DIGEST_ROW_SIZE]
	vpinsrd	xmm0, [state + _args_digest_md5 + MD5_DIGEST_WORD_SIZE*idx + 1*MD5_DIGEST_ROW_SIZE], 1
	vpinsrd	xmm0, [state + _args_digest_md5 + MD5_DIGEST_WORD_SIZE*idx + 2*MD5_DIGEST_ROW_SIZE], 2
	vpinsrd	xmm0, [state + _args_digest_md5 + MD5_DIGEST_WORD_SIZE*idx + 3*MD5_DIGEST_ROW_SIZE], 3
	vmovdqa	[lane_data + _outer_block], xmm0

	mov	tmp, [job + _auth_key_xor_opad]
	vmovdqu	xmm0, [tmp]
	vmovd	[state + _args_digest_md5 + MD5_DIGEST_WORD_SIZE*idx + 0*MD5_DIGEST_ROW_SIZE], xmm0
	vpextrd	[state + _args_digest_md5 + MD5_DIGEST_WORD_SIZE*idx + 1*MD5_DIGEST_ROW_SIZE], xmm0, 1
	vpextrd	[state + _args_digest_md5 + MD5_DIGEST_WORD_SIZE*idx + 2*MD5_DIGEST_ROW_SIZE], xmm0, 2
	vpextrd	[state + _args_digest_md5 + MD5_DIGEST_WORD_SIZE*idx + 3*MD5_DIGEST_ROW_SIZE], xmm0, 3
	jmp	copy_lane_data

	align	16
proc_extra_blocks:
	mov	DWORD(start_offset), [lane_data + _start_offset]
	mov	[state + _lens_md5 + 2*idx], WORD(extra_blocks)
	lea	tmp, [lane_data + _extra_block + start_offset]
	mov	[state + _args_data_ptr_md5 + PTR_SZ*idx], tmp
	mov	dword [lane_data + _extra_blocks], 0
	jmp	copy_lane_data

return_null:
	xor	job_rax, job_rax
	jmp	return

	align	16
end_loop:
	mov	job_rax, [lane_data + _job_in_lane]
	mov	qword [lane_data + _job_in_lane], 0
	or	dword [job_rax + _status], STS_COMPLETED_HMAC
	mov	unused_lanes, [state + _unused_lanes_md5]
	shl	unused_lanes, 4
	or	unused_lanes, idx
	mov	[state + _unused_lanes_md5], unused_lanes

	mov	p, [job_rax + _auth_tag_output]

	; copy 12 bytes
	mov	DWORD(tmp2), [state + _args_digest_md5 + MD5_DIGEST_WORD_SIZE*idx + 0*MD5_DIGEST_ROW_SIZE]
	mov	DWORD(tmp4), [state + _args_digest_md5 + MD5_DIGEST_WORD_SIZE*idx + 1*MD5_DIGEST_ROW_SIZE]
	mov	DWORD(tmp5), [state + _args_digest_md5 + MD5_DIGEST_WORD_SIZE*idx + 2*MD5_DIGEST_ROW_SIZE]
;	bswap	DWORD(tmp2)
;	bswap	DWORD(tmp4)
;	bswap	DWORD(tmp3)
	mov	[p + 0*4], DWORD(tmp2)
	mov	[p + 1*4], DWORD(tmp4)
	mov	[p + 2*4], DWORD(tmp5)

        cmp     DWORD [job_rax + _auth_tag_output_len_in_bytes], 12
        je      return

        ; copy 16 bytes
        mov	DWORD(tmp5), [state + _args_digest_md5 + MD5_DIGEST_WORD_SIZE*idx + 3*MD5_DIGEST_ROW_SIZE]
        mov	[p + 3*4], DWORD(tmp5)

return:

	mov	rbx, [rsp + _gpr_save + 8*0]
	mov	rbp, [rsp + _gpr_save + 8*1]
	mov	r12, [rsp + _gpr_save + 8*2]
	mov	r13, [rsp + _gpr_save + 8*3]
	mov	r14, [rsp + _gpr_save + 8*4]
	mov	r15, [rsp + _gpr_save + 8*5]
%ifndef LINUX
	mov	rsi, [rsp + _gpr_save + 8*6]
	mov	rdi, [rsp + _gpr_save + 8*7]
%endif
	mov	rsp, [rsp + _rsp_save]	; original SP

	ret

%ifdef LINUX
section .note.GNU-stack noalloc noexec nowrite progbits
%endif
Commit	Line	Data
11fdf7f2 TL	1	;;
	2	;; Copyright (c) 2012-2018, Intel Corporation
	3	;;
	4	;; Redistribution and use in source and binary forms, with or without
	5	;; modification, are permitted provided that the following conditions are met:
	6	;;
	7	;; * Redistributions of source code must retain the above copyright notice,
	8	;; 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 the
	11	;; documentation and/or other materials provided with the distribution.
	12	;; * Neither the name of Intel Corporation nor the names of its contributors
	13	;; may be used to endorse or promote products derived from this software
	14	;; without specific prior written permission.
	15	;;
	16	;; THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
	17	;; AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
	18	;; IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
	19	;; DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE
	20	;; FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
	21	;; DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
	22	;; SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
	23	;; CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
	24	;; OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
	25	;; OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
	26	;;
	27
	28	%include "os.asm"
	29	%include "job_aes_hmac.asm"
	30	%include "mb_mgr_datastruct.asm"
	31	%include "reg_sizes.asm"
	32
	33	extern md5_x4x2_avx
	34
	35	section .data
	36	default rel
	37	align 16
	38	dupw: ;ddq 0x01000100010001000100010001000100
	39	dq 0x0100010001000100, 0x0100010001000100
	40	x80: ;ddq 0x00000000000000000000000000000080
	41	dq 0x0000000000000080, 0x0000000000000000
	42	x00: ;ddq 0x00000000000000000000000000000000
	43	dq 0x0000000000000000, 0x0000000000000000
	44	len_masks:
	45	;ddq 0x0000000000000000000000000000FFFF
	46	dq 0x000000000000FFFF, 0x0000000000000000
	47	;ddq 0x000000000000000000000000FFFF0000
	48	dq 0x00000000FFFF0000, 0x0000000000000000
	49	;ddq 0x00000000000000000000FFFF00000000
	50	dq 0x0000FFFF00000000, 0x0000000000000000
	51	;ddq 0x0000000000000000FFFF000000000000
	52	dq 0xFFFF000000000000, 0x0000000000000000
	53	;ddq 0x000000000000FFFF0000000000000000
	54	dq 0x0000000000000000, 0x000000000000FFFF
	55	;ddq 0x00000000FFFF00000000000000000000
	56	dq 0x0000000000000000, 0x00000000FFFF0000
	57	;ddq 0x0000FFFF000000000000000000000000
	58	dq 0x0000000000000000, 0x0000FFFF00000000
	59	;ddq 0xFFFF0000000000000000000000000000
	60	dq 0x0000000000000000, 0xFFFF000000000000
	61	one: dq 1
	62	two: dq 2
	63	three: dq 3
	64	four: dq 4
65	five: dq 5
66	six: dq 6
67	seven: dq 7
68
69	section .text
70
71	%if 1
72	%ifdef LINUX
73	%define arg1 rdi
74	%define arg2 rsi
75	%else
76	%define arg1 rcx
77	%define arg2 rdx
78	%endif
79
80	%define state arg1
81	%define job arg2
82	%define len2 arg2
83
84
85	; idx needs to be in rbp
86	%define idx rbp
87
88	; unused_lanes must be in rax-rdx
89	%define unused_lanes rbx
90	%define lane_data rbx
91	%define tmp2 rbx
92
93	%define job_rax rax
94	%define tmp1 rax
95	%define size_offset rax
96	%define tmp rax
97	%define start_offset rax
98
99	%define tmp3 arg1
100
101	%define extra_blocks arg2
102	%define p arg2
103
104	%define tmp4 r8
9f95a23c	105	%define tmp5 r9
11fdf7f2 TL	106
	107	%endif
	108
	109	; This routine and/or the called routine clobbers all GPRs
	110	struc STACK
	111	_gpr_save: resq 8
	112	_rsp_save: resq 1
	113	endstruc
	114
	115	%define APPEND(a,b) a %+ b
	116
	117	; JOB* flush_job_hmac_md5_avx(MB_MGR_HMAC_MD5_OOO *state)
	118	; arg 1 : rcx : state
	119	MKGLOBAL(flush_job_hmac_md5_avx,function,internal)
	120	flush_job_hmac_md5_avx:
	121
	122	mov rax, rsp
	123	sub rsp, STACK_size
	124	and rsp, -16
	125
	126	mov [rsp + _gpr_save + 8*0], rbx
	127	mov [rsp + _gpr_save + 8*1], rbp
	128	mov [rsp + _gpr_save + 8*2], r12
	129	mov [rsp + _gpr_save + 8*3], r13
	130	mov [rsp + _gpr_save + 8*4], r14
	131	mov [rsp + _gpr_save + 8*5], r15
	132	%ifndef LINUX
	133	mov [rsp + _gpr_save + 8*6], rsi
	134	mov [rsp + _gpr_save + 8*7], rdi
	135	%endif
	136	mov [rsp + _rsp_save], rax ; original SP
	137
	138	mov unused_lanes, [state + _unused_lanes_md5]
	139	bt unused_lanes, 32+3
	140	jc return_null
	141
	142	; find a lane with a non-null job
	143	xor idx, idx
	144	cmp qword [state + _ldata_md5 + 1 * _HMAC_SHA1_LANE_DATA_size + _job_in_lane],0
	145	cmovne idx, [rel one]
	146	cmp qword [state + _ldata_md5 + 2 * _HMAC_SHA1_LANE_DATA_size + _job_in_lane],0
	147	cmovne idx, [rel two]
	148	cmp qword [state + _ldata_md5 + 3 * _HMAC_SHA1_LANE_DATA_size + _job_in_lane],0
	149	cmovne idx, [rel three]
	150	cmp qword [state + _ldata_md5 + 4 * _HMAC_SHA1_LANE_DATA_size + _job_in_lane],0
	151	cmovne idx, [rel four]
	152	cmp qword [state + _ldata_md5 + 5 * _HMAC_SHA1_LANE_DATA_size + _job_in_lane],0
	153	cmovne idx, [rel five]
	154	cmp qword [state + _ldata_md5 + 6 * _HMAC_SHA1_LANE_DATA_size + _job_in_lane],0
	155	cmovne idx, [rel six]
	156	cmp qword [state + _ldata_md5 + 7 * _HMAC_SHA1_LANE_DATA_size + _job_in_lane],0
	157	cmovne idx, [rel seven]
	158
	159	copy_lane_data:
	160	; copy good lane (idx) to empty lanes
	161	vmovdqa xmm0, [state + _lens_md5]
	162	mov tmp, [state + _args_data_ptr_md5 + PTR_SZ*idx]
	163
	164	%assign I 0
	165	%rep 8
	166	cmp qword [state + _ldata_md5 + I * _HMAC_SHA1_LANE_DATA_size + _job_in_lane], 0
	167	jne APPEND(skip_,I)
	168	mov [state + _args_data_ptr_md5 + PTR_SZ*I], tmp
	169	vpor xmm0, xmm0, [rel len_masks + 16*I]
170	APPEND(skip_,I):
171	%assign I (I+1)
172	%endrep
173
174	vmovdqa [state + _lens_md5], xmm0
175
176	vphminposuw xmm1, xmm0
177	vpextrw DWORD(len2), xmm1, 0 ; min value
178	vpextrw DWORD(idx), xmm1, 1 ; min index (0...3)
179	cmp len2, 0
180	je len_is_0
181
182	vpshufb xmm1, [rel dupw] ; duplicate words across all lanes
183	vpsubw xmm0, xmm0, xmm1
184	vmovdqa [state + _lens_md5], xmm0
185
186	; "state" and "args" are the same address, arg1
187	; len is arg2
188	call md5_x4x2_avx
189	; state and idx are intact
190
191	len_is_0:
192	; process completed job "idx"
193	imul lane_data, idx, _HMAC_SHA1_LANE_DATA_size
194	lea lane_data, [state + _ldata_md5 + lane_data]
195	mov DWORD(extra_blocks), [lane_data + _extra_blocks]
196	cmp extra_blocks, 0
197	jne proc_extra_blocks
198	cmp dword [lane_data + _outer_done], 0
199	jne end_loop
200
201	proc_outer:
202	mov dword [lane_data + _outer_done], 1
203	mov DWORD(size_offset), [lane_data + _size_offset]
204	mov qword [lane_data + _extra_block + size_offset], 0
205	mov word [state + _lens_md5 + 2*idx], 1
206	lea tmp, [lane_data + _outer_block]
207	mov job, [lane_data + _job_in_lane]
208	mov [state + _args_data_ptr_md5 + PTR_SZ*idx], tmp
209
210	vmovd xmm0, [state + _args_digest_md5 + MD5_DIGEST_WORD_SIZEidx + 0MD5_DIGEST_ROW_SIZE]
211	vpinsrd xmm0, [state + _args_digest_md5 + MD5_DIGEST_WORD_SIZEidx + 1MD5_DIGEST_ROW_SIZE], 1
212	vpinsrd xmm0, [state + _args_digest_md5 + MD5_DIGEST_WORD_SIZEidx + 2MD5_DIGEST_ROW_SIZE], 2
213	vpinsrd xmm0, [state + _args_digest_md5 + MD5_DIGEST_WORD_SIZEidx + 3MD5_DIGEST_ROW_SIZE], 3
214	vmovdqa [lane_data + _outer_block], xmm0
215
216	mov tmp, [job + _auth_key_xor_opad]
217	vmovdqu xmm0, [tmp]
218	vmovd [state + _args_digest_md5 + MD5_DIGEST_WORD_SIZEidx + 0MD5_DIGEST_ROW_SIZE], xmm0
219	vpextrd [state + _args_digest_md5 + MD5_DIGEST_WORD_SIZEidx + 1MD5_DIGEST_ROW_SIZE], xmm0, 1
220	vpextrd [state + _args_digest_md5 + MD5_DIGEST_WORD_SIZEidx + 2MD5_DIGEST_ROW_SIZE], xmm0, 2
221	vpextrd [state + _args_digest_md5 + MD5_DIGEST_WORD_SIZEidx + 3MD5_DIGEST_ROW_SIZE], xmm0, 3
222	jmp copy_lane_data
223
224	align 16
225	proc_extra_blocks:
226	mov DWORD(start_offset), [lane_data + _start_offset]
227	mov [state + _lens_md5 + 2*idx], WORD(extra_blocks)
228	lea tmp, [lane_data + _extra_block + start_offset]
229	mov [state + _args_data_ptr_md5 + PTR_SZ*idx], tmp
230	mov dword [lane_data + _extra_blocks], 0
231	jmp copy_lane_data
232
233	return_null:
234	xor job_rax, job_rax
235	jmp return
236
237	align 16
238	end_loop:
239	mov job_rax, [lane_data + _job_in_lane]
240	mov qword [lane_data + _job_in_lane], 0
241	or dword [job_rax + _status], STS_COMPLETED_HMAC
242	mov unused_lanes, [state + _unused_lanes_md5]
243	shl unused_lanes, 4
244	or unused_lanes, idx
245	mov [state + _unused_lanes_md5], unused_lanes
246
247	mov p, [job_rax + _auth_tag_output]
248
249	; copy 12 bytes
250	mov DWORD(tmp2), [state + _args_digest_md5 + MD5_DIGEST_WORD_SIZEidx + 0MD5_DIGEST_ROW_SIZE]
251	mov DWORD(tmp4), [state + _args_digest_md5 + MD5_DIGEST_WORD_SIZEidx + 1MD5_DIGEST_ROW_SIZE]
9f95a23c	252	mov DWORD(tmp5), [state + _args_digest_md5 + MD5_DIGEST_WORD_SIZEidx + 2MD5_DIGEST_ROW_SIZE]
11fdf7f2 TL	253	; bswap DWORD(tmp2)
	254	; bswap DWORD(tmp4)
	255	; bswap DWORD(tmp3)
	256	mov [p + 0*4], DWORD(tmp2)
	257	mov [p + 1*4], DWORD(tmp4)
9f95a23c TL	258	mov [p + 2*4], DWORD(tmp5)
	259
	260	cmp DWORD [job_rax + _auth_tag_output_len_in_bytes], 12
	261	je return
	262
	263	; copy 16 bytes
	264	mov DWORD(tmp5), [state + _args_digest_md5 + MD5_DIGEST_WORD_SIZEidx + 3MD5_DIGEST_ROW_SIZE]
	265	mov [p + 3*4], DWORD(tmp5)
11fdf7f2 TL	266
	267	return:
	268
	269	mov rbx, [rsp + _gpr_save + 8*0]
	270	mov rbp, [rsp + _gpr_save + 8*1]
	271	mov r12, [rsp + _gpr_save + 8*2]
	272	mov r13, [rsp + _gpr_save + 8*3]
	273	mov r14, [rsp + _gpr_save + 8*4]
	274	mov r15, [rsp + _gpr_save + 8*5]
	275	%ifndef LINUX
	276	mov rsi, [rsp + _gpr_save + 8*6]
	277	mov rdi, [rsp + _gpr_save + 8*7]
	278	%endif
	279	mov rsp, [rsp + _rsp_save] ; original SP
	280
	281	ret
	282
	283	%ifdef LINUX
	284	section .note.GNU-stack noalloc noexec nowrite progbits
	285	%endif