;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;  Copyright(c) 2011-2016 Intel Corporation All rights reserved.
;
;  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 "md5_job.asm"
%include "md5_mb_mgr_datastruct.asm"
%include "reg_sizes.asm"

%ifdef HAVE_AS_KNOWS_AVX512
extern md5_mb_x16x2_avx512

[bits 64]
default rel
section .text

%if 1
%ifidn __OUTPUT_FORMAT__, win64
; WINDOWS register definitions
%define arg1    rcx
%define arg2    rdx

%define lane	rsi

%else
; UN*X register definitions
%define arg1    rdi
%define arg2    rsi

%define lane	rdx

%endif

; Common definitions
%define state   arg1
%define job     arg2
%define len2    arg2

; idx needs to be in a register not clobberred by md5_mb_x16_avx512
%define idx	rbp

%define p	r11

%define unused_lanes	ymm7

%define job_rax	rax
%define len	rax

%define num_lanes_inuse r9

%define lane_data	r10

%endif ; if 1

; STACK_SPACE needs to be an odd multiple of 8
%define STACK_SPACE	8*8 + 16*10 + 8

;; Byte shift in MEM addr, read a extra byte [addr+16]
%macro MEM_VPSRLDDQ 2
%define %%addr	  %1
%define %%TMP_YMM %2
	vmovdqu	%%TMP_YMM, [%%addr + 1]
	vmovdqu [%%addr], %%TMP_YMM
	mov	[%%addr + 31], byte 0
%endmacro

;; Byte shift in MEM addr, read a extra byte [addr-1]
%macro MEM_VPSLLDDQ 2
%define %%addr	  %1
%define %%TMP_YMM %2
	vmovdqu	%%TMP_YMM, [%%addr-1]
	vmovdqu [%%addr], %%TMP_YMM
	mov	[%%addr], byte 0
%endmacro

align 64

; JOB* submit_job(MB_MGR *state, JOB_MD5 *job)
; arg 1 : rcx : state
; arg 2 : rdx : job
mk_global md5_mb_mgr_submit_avx512, function
md5_mb_mgr_submit_avx512:
	endbranch

	sub	rsp, STACK_SPACE
	; we need to save/restore all GPRs because lower layer clobbers them
	mov	[rsp + 8*0], rbx
	mov	[rsp + 8*1], rbp
	mov	[rsp + 8*2], r12
	mov	[rsp + 8*3], r13
	mov	[rsp + 8*4], r14
	mov	[rsp + 8*5], r15
%ifidn __OUTPUT_FORMAT__, win64
	mov	[rsp + 8*6], rsi
	mov	[rsp + 8*7], rdi
	vmovdqa  [rsp + 8*8 + 16*0], xmm6
	vmovdqa  [rsp + 8*8 + 16*1], xmm7
	vmovdqa  [rsp + 8*8 + 16*2], xmm8
	vmovdqa  [rsp + 8*8 + 16*3], xmm9
	vmovdqa  [rsp + 8*8 + 16*4], xmm10
	vmovdqa  [rsp + 8*8 + 16*5], xmm11
	vmovdqa  [rsp + 8*8 + 16*6], xmm12
	vmovdqa  [rsp + 8*8 + 16*7], xmm13
	vmovdqa  [rsp + 8*8 + 16*8], xmm14
	vmovdqa  [rsp + 8*8 + 16*9], xmm15
%endif

	mov	lane, [state + _unused_lanes]
	and	lane, 0x3F
	MEM_VPSRLDDQ (state + _unused_lanes), unused_lanes
	imul	lane_data, lane, _LANE_DATA_size
	mov	dword [job + _status], STS_BEING_PROCESSED
	lea	lane_data, [state + _ldata + lane_data]
	mov	DWORD(len), [job + _len]

	shl	len, 6	; low 5 bits store idx
	or	len, lane

	mov	[lane_data + _job_in_lane], job
	mov	[state + _lens + 4*lane], DWORD(len)

	; Load digest words from result_digest
	vmovdqu	xmm0, [job + _result_digest + 0*16]
	vmovd	[state + _args_digest + 4*lane + 0*4*16*2], xmm0
	vpextrd [state + _args_digest + 4*lane + 1*4*16*2], xmm0, 1
	vpextrd [state + _args_digest + 4*lane + 2*4*16*2], xmm0, 2
	vpextrd [state + _args_digest + 4*lane + 3*4*16*2], xmm0, 3

	mov	p, [job + _buffer]
	mov	[state + _args_data_ptr + 8*lane], p

	mov	DWORD(num_lanes_inuse), [state + _num_lanes_inuse]
	add	num_lanes_inuse, 1
	mov	[state + _num_lanes_inuse], DWORD(num_lanes_inuse)
	cmp	num_lanes_inuse, 32
	jne	return_null

start_loop:
	; Find min length
	vmovdqu ymm0, [state + _lens + 0*32]
	vmovdqu ymm1, [state + _lens + 1*32]

	vpminud ymm2, ymm0, ymm1	; ymm2 has {D,C,B,A}
	vpalignr ymm3, ymm3, ymm2, 8    ; ymm3 has {x,x,D,C}
	vpminud ymm2, ymm2, ymm3	; ymm2 has {x,x,E,F}
	vpalignr ymm3, ymm3, ymm2, 4    ; ymm3 has {x,x,x,E}
	vpminud ymm2, ymm2, ymm3	; ymm2 has min value in low dword
	vperm2i128 ymm3, ymm2, ymm2, 1	; ymm3 has halves of ymm2 reversed
	vpminud ymm2, ymm2, ymm3	; ymm2 has min value in low dword

	; Find min length
	vmovdqu ymm5, [state + _lens + 2*32]
	vmovdqu ymm6, [state + _lens + 3*32]

	vpminud ymm4, ymm5, ymm6	; ymm4 has {D,C,B,A}
	vpalignr ymm3, ymm3, ymm4, 8    ; ymm3 has {x,x,D,C}
	vpminud ymm4, ymm4, ymm3	; ymm4 has {x,x,E,F}
	vpalignr ymm3, ymm3, ymm4, 4    ; ymm3 has {x,x,x,E}
	vpminud ymm4, ymm4, ymm3	; ymm4 has min value in low dword
	vperm2i128 ymm3, ymm4, ymm4, 1	; ymm3 has halves of ymm4 reversed
	vpminud ymm4, ymm4, ymm3	; ymm4 has min value in low dword

	vpminud ymm2, ymm2, ymm4	; ymm2 has min value in low dword
	vmovd	DWORD(idx), xmm2
	mov	len2, idx
	and	idx, 0x3F
	shr	len2, 6
	jz	len_is_0

	vpand	ymm2, ymm2, [rel clear_low_6bits]
	vpshufd ymm2, ymm2, 0

	vpsubd  ymm0, ymm0, ymm2
	vpsubd  ymm1, ymm1, ymm2
	vpsubd  ymm5, ymm5, ymm2
	vpsubd  ymm6, ymm6, ymm2

	vmovdqu [state + _lens + 0*32], ymm0
	vmovdqu [state + _lens + 1*32], ymm1
	vmovdqu [state + _lens + 2*32], ymm5
	vmovdqu [state + _lens + 3*32], ymm6

	; "state" and "args" are the same address, arg1
	; len is arg2
	call	md5_mb_x16x2_avx512
	; state and idx are intact

len_is_0:
	; process completed job "idx"
	imul	lane_data, idx, _LANE_DATA_size
	lea	lane_data, [state + _ldata + lane_data]

	mov	job_rax, [lane_data + _job_in_lane]
	mov	lane, [state + _unused_lanes]
	mov	qword [lane_data + _job_in_lane], 0
	mov	dword [job_rax + _status], STS_COMPLETED

	shl	lane, 8
	or	 lane, idx
	MEM_VPSLLDDQ	(state + _unused_lanes), unused_lanes
	mov	[state + _unused_lanes], lane

	mov	DWORD(num_lanes_inuse), [state + _num_lanes_inuse]
	sub	num_lanes_inuse, 1
	mov	[state + _num_lanes_inuse], DWORD(num_lanes_inuse)

	mov	dword [state + _lens + 4*idx], 0xFFFFFFFF

	vmovd	 xmm0, [state + _args_digest + 4*idx + 0*4*16*2]
	vpinsrd  xmm0, [state + _args_digest + 4*idx + 1*4*16*2], 1
	vpinsrd  xmm0, [state + _args_digest + 4*idx + 2*4*16*2], 2
	vpinsrd  xmm0, [state + _args_digest + 4*idx + 3*4*16*2], 3

	vmovdqa  [job_rax + _result_digest + 0*16], xmm0

return:
%ifidn __OUTPUT_FORMAT__, win64
	vmovdqa  xmm6, [rsp + 8*8 + 16*0]
	vmovdqa  xmm7, [rsp + 8*8 + 16*1]
	vmovdqa  xmm8, [rsp + 8*8 + 16*2]
	vmovdqa  xmm9, [rsp + 8*8 + 16*3]
	vmovdqa  xmm10, [rsp + 8*8 + 16*4]
	vmovdqa  xmm11, [rsp + 8*8 + 16*5]
	vmovdqa  xmm12, [rsp + 8*8 + 16*6]
	vmovdqa  xmm13, [rsp + 8*8 + 16*7]
	vmovdqa  xmm14, [rsp + 8*8 + 16*8]
	vmovdqa  xmm15, [rsp + 8*8 + 16*9]
	mov	rsi, [rsp + 8*6]
	mov	rdi, [rsp + 8*7]
%endif
	mov	rbx, [rsp + 8*0]
	mov	rbp, [rsp + 8*1]
	mov	r12, [rsp + 8*2]
	mov	r13, [rsp + 8*3]
	mov	r14, [rsp + 8*4]
	mov	r15, [rsp + 8*5]

	add	rsp, STACK_SPACE

	ret

return_null:
	xor	job_rax, job_rax
	jmp	return


section .data align=32

align 32
clear_low_6bits:
	dq 0x00000000FFFFFFC0, 0x0000000000000000
	dq 0x00000000FFFFFFC0, 0x0000000000000000

%else
%ifidn __OUTPUT_FORMAT__, win64
global no_md5_mb_mgr_submit_avx512
no_md5_mb_mgr_submit_avx512:
%endif
%endif ; HAVE_AS_KNOWS_AVX512