[ceph.git] / ceph / src / erasure-code / isa / isa-l / erasure_code / gf_6vect_mad_sse.asm.s

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;  Copyright(c) 2011-2015 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.
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;

;;;
;;; gf_6vect_mad_sse(len, vec, vec_i, mul_array, src, dest);
;;;

%include "reg_sizes.asm"

%define PS 8

%ifidn __OUTPUT_FORMAT__, win64
 %define arg0  rcx
 %define arg0.w ecx
 %define arg1  rdx
 %define arg2  r8
 %define arg3  r9
 %define arg4  r12
 %define arg5  r15
 %define tmp   r11
 %define tmp.w r11d
 %define tmp2   r10
 %define tmp3   r13
 %define tmp4   r14
 %define tmp5   rdi
 %define return rax
 %define return.w eax
 %define stack_size 16*10 + 5*8
 %define arg(x)      [rsp + stack_size + PS + PS*x]
 %define func(x) proc_frame x

%macro FUNC_SAVE 0
	sub	rsp, stack_size
	movdqa	[rsp+16*0],xmm6
	movdqa	[rsp+16*1],xmm7
	movdqa	[rsp+16*2],xmm8
	movdqa	[rsp+16*3],xmm9
	movdqa	[rsp+16*4],xmm10
	movdqa	[rsp+16*5],xmm11
	movdqa	[rsp+16*6],xmm12
	movdqa	[rsp+16*7],xmm13
	movdqa	[rsp+16*8],xmm14
	movdqa	[rsp+16*9],xmm15
	save_reg	r12,  10*16 + 0*8
	save_reg	r13,  10*16 + 1*8
	save_reg	r14,  10*16 + 2*8
	save_reg	r15,  10*16 + 3*8
	save_reg	rdi,  10*16 + 4*8
	end_prolog
	mov	arg4, arg(4)
	mov	arg5, arg(5)
%endmacro

%macro FUNC_RESTORE 0
	movdqa	xmm6, [rsp+16*0]
	movdqa	xmm7, [rsp+16*1]
	movdqa	xmm8, [rsp+16*2]
	movdqa	xmm9, [rsp+16*3]
	movdqa	xmm10, [rsp+16*4]
	movdqa	xmm11, [rsp+16*5]
	movdqa	xmm12, [rsp+16*6]
	movdqa	xmm13, [rsp+16*7]
	movdqa	xmm14, [rsp+16*8]
	movdqa	xmm15, [rsp+16*9]
	mov	r12,  [rsp + 10*16 + 0*8]
	mov	r13,  [rsp + 10*16 + 1*8]
	mov	r14,  [rsp + 10*16 + 2*8]
	mov	r15,  [rsp + 10*16 + 3*8]
	mov	rdi,  [rsp + 10*16 + 4*8]
	add	rsp, stack_size
%endmacro

%elifidn __OUTPUT_FORMAT__, elf64
 %define arg0  rdi
 %define arg0.w edi
 %define arg1  rsi
 %define arg2  rdx
 %define arg3  rcx
 %define arg4  r8
 %define arg5  r9
 %define tmp   r11
 %define tmp.w r11d
 %define tmp2   r10
 %define tmp3   r12
 %define tmp4   r13
 %define tmp5   r14
 %define return rax
 %define return.w eax

 %define func(x) x:
 %macro FUNC_SAVE 0
	push	r12
	push	r13
	push	r14
 %endmacro
 %macro FUNC_RESTORE 0
	pop	r14
	pop	r13
	pop	r12
 %endmacro
%endif

;;; gf_6vect_mad_sse(len, vec, vec_i, mul_array, src, dest)
%define len   arg0
%define len.w arg0.w
%define vec    arg1
%define vec_i    arg2
%define mul_array arg3
%define	src   arg4
%define dest1  arg5
%define pos   return
%define pos.w return.w

%define dest2 mul_array
%define dest3 tmp2
%define dest4 tmp4
%define dest5 tmp5
%define dest6 vec_i

%ifndef EC_ALIGNED_ADDR
;;; Use Un-aligned load/store
 %define XLDR movdqu
 %define XSTR movdqu
%else
;;; Use Non-temporal load/stor
 %ifdef NO_NT_LDST
  %define XLDR movdqa
  %define XSTR movdqa
 %else
  %define XLDR movntdqa
  %define XSTR movntdq
 %endif
%endif

default rel

[bits 64]
section .text

%define xmask0f  xmm15
%define xgft4_lo  xmm14
%define xgft4_hi  xmm13
%define xgft5_lo  xmm12
%define xgft5_hi  xmm11
%define xgft6_lo  xmm10
%define xgft6_hi  xmm9

%define x0         xmm0
%define xtmpa      xmm1
%define xtmph1     xmm2
%define xtmpl1     xmm3
%define xtmph2     xmm4
%define xtmpl2     xmm5
%define xtmph3     xmm6
%define xtmpl3     xmm7
%define xd1        xmm8
%define xd2        xtmpl1
%define xd3        xtmph1


align 16
global gf_6vect_mad_sse:function
func(gf_6vect_mad_sse)
	FUNC_SAVE
	sub	len, 16
	jl	.return_fail

	xor	pos, pos
	movdqa	xmask0f, [mask0f]	;Load mask of lower nibble in each byte

	mov	tmp, vec
	sal	vec_i, 5		;Multiply by 32
	lea	tmp3, [mul_array + vec_i]
	sal	tmp, 6			;Multiply by 64

	sal	vec, 5			;Multiply by 32
	lea	vec_i, [tmp + vec]	;vec_i = 96
	lea	mul_array, [tmp + vec_i]	;mul_array = 160

	movdqu	xgft5_lo, [tmp3+2*tmp]		;Load array Ex{00}, Ex{01}, ..., Ex{0f}
	movdqu	xgft5_hi, [tmp3+2*tmp+16]	;     "     Ex{00}, Ex{10}, ..., Ex{f0}
	movdqu	xgft4_lo, [tmp3+vec_i]		;Load array Dx{00}, Dx{01}, Dx{02}, ...
	movdqu	xgft4_hi, [tmp3+vec_i+16]	; " Dx{00}, Dx{10}, Dx{20}, ... , Dx{f0}
	movdqu	xgft6_lo, [tmp3+mul_array]	;Load array Fx{00}, Fx{01}, ..., Fx{0f}
	movdqu	xgft6_hi, [tmp3+mul_array+16]	;     "     Fx{00}, Fx{10}, ..., Fx{f0}

	mov	dest2, [dest1+PS]
	mov	dest3, [dest1+2*PS]
	mov	dest4, [dest1+3*PS]  ; reuse mul_array
	mov	dest5, [dest1+4*PS]
	mov	dest6, [dest1+5*PS]  ; reuse vec_i
	mov	dest1, [dest1]

.loop16:
	XLDR	x0, [src+pos]		;Get next source vector

	movdqu	xtmpl1, [tmp3]	;Load array Ax{00}, Ax{01}, Ax{02}, ...
	movdqu	xtmph1, [tmp3+16]	; " Ax{00}, Ax{10}, Ax{20}, ... , Ax{f0}
	movdqu	xtmpl2, [tmp3+vec]	;Load array Bx{00}, Bx{01}, Bx{02}, ...
	movdqu	xtmph2, [tmp3+vec+16]	; " Bx{00}, Bx{10}, Bx{20}, ... , Bx{f0}
	movdqu	xtmpl3, [tmp3+2*vec]	;Load array Cx{00}, Cx{01}, Cx{02}, ...
	movdqu	xtmph3, [tmp3+2*vec+16]	; " Cx{00}, Cx{10}, Cx{20}, ... , Cx{f0}
	XLDR	xd1, [dest1+pos]	;Get next dest vector

	movdqa	xtmpa, x0		;Keep unshifted copy of src
	psraw	x0, 4			;Shift to put high nibble into bits 4-0
	pand	x0, xmask0f		;Mask high src nibble in bits 4-0
	pand	xtmpa, xmask0f		;Mask low src nibble in bits 4-0

	;dest1
	pshufb	xtmph1, x0		;Lookup mul table of high nibble
	pshufb	xtmpl1, xtmpa		;Lookup mul table of low nibble
	pxor	xtmph1, xtmpl1		;GF add high and low partials
	pxor	xd1, xtmph1

	XLDR	xd2, [dest2+pos]	;reuse xtmpl1. Get next dest vector
	XLDR	xd3, [dest3+pos]	;reuse xtmph1. Get next dest3 vector

	;dest2
	pshufb	xtmph2, x0		;Lookup mul table of high nibble
	pshufb	xtmpl2, xtmpa		;Lookup mul table of low nibble
	pxor	xtmph2, xtmpl2		;GF add high and low partials
	pxor	xd2, xtmph2

	;dest3
	pshufb	xtmph3, x0		;Lookup mul table of high nibble
	pshufb	xtmpl3, xtmpa		;Lookup mul table of low nibble
	pxor	xtmph3, xtmpl3		;GF add high and low partials
	pxor	xd3, xtmph3

	XSTR	[dest1+pos], xd1	;Store result into dest1
	XSTR	[dest2+pos], xd2	;Store result into dest2
	XSTR	[dest3+pos], xd3	;Store result into dest3

	movdqa	xtmph1, xgft4_hi	;Reload const array registers
	movdqa	xtmpl1, xgft4_lo	;Reload const array registers
	movdqa	xtmph2, xgft5_hi	;Reload const array registers
	movdqa	xtmpl2, xgft5_lo	;Reload const array registers
	movdqa	xtmph3, xgft6_hi	;Reload const array registers
	movdqa	xtmpl3, xgft6_lo	;Reload const array registers

	;dest4
	XLDR	xd1, [dest4+pos]	;Get next dest vector
	pshufb	xtmph1, x0		;Lookup mul table of high nibble
	pshufb	xtmpl1, xtmpa		;Lookup mul table of low nibble
	pxor	xtmph1, xtmpl1		;GF add high and low partials
	pxor	xd1, xtmph1

	XLDR	xd2, [dest5+pos]	;reuse xtmpl1. Get next dest vector
	XLDR	xd3, [dest6+pos]	;reuse xtmph1. Get next dest vector

	;dest5
	pshufb	xtmph2, x0		;Lookup mul table of high nibble
	pshufb	xtmpl2, xtmpa		;Lookup mul table of low nibble
	pxor	xtmph2, xtmpl2		;GF add high and low partials
	pxor	xd2, xtmph2

	;dest6
	pshufb	xtmph3, x0		;Lookup mul table of high nibble
	pshufb	xtmpl3, xtmpa		;Lookup mul table of low nibble
	pxor	xtmph3, xtmpl3		;GF add high and low partials
	pxor	xd3, xtmph3

	XSTR	[dest4+pos], xd1	;Store result into dest4
	XSTR	[dest5+pos], xd2	;Store result into dest5
	XSTR	[dest6+pos], xd3	;Store result into dest6

	add	pos, 16			;Loop on 16 bytes at a time
	cmp	pos, len
	jle	.loop16

	lea	tmp, [len + 16]
	cmp	pos, tmp
	je	.return_pass

.lessthan16:
	;; Tail len
	;; Do one more overlap pass
	;; Overlapped offset length-16
	mov	tmp, len		;Backup len as len=rdi

	XLDR	x0, [src+tmp]		;Get next source vector
	XLDR	xd1, [dest4+tmp]	;Get next dest vector
	XLDR	xd2, [dest5+tmp]	;reuse xtmpl1. Get next dest vector
	XLDR	xd3, [dest6+tmp]	;reuse xtmph1. Get next dest vector

	sub	len, pos

	movdqa	xtmph3, [constip16]	;Load const of i + 16
	pinsrb	xtmpl3, len.w, 15
	pshufb	xtmpl3, xmask0f		;Broadcast len to all bytes
	pcmpgtb	xtmpl3, xtmph3

	movdqa	xtmpa, x0		;Keep unshifted copy of src
	psraw	x0, 4			;Shift to put high nibble into bits 4-0
	pand	x0, xmask0f		;Mask high src nibble in bits 4-0
	pand	xtmpa, xmask0f		;Mask low src nibble in bits 4-0

	;dest4
	pshufb	xgft4_hi, x0		;Lookup mul table of high nibble
	pshufb	xgft4_lo, xtmpa		;Lookup mul table of low nibble
	pxor	xgft4_hi, xgft4_lo	;GF add high and low partials
	pand	xgft4_hi, xtmpl3
	pxor	xd1, xgft4_hi

	;dest5
	pshufb	xgft5_hi, x0		;Lookup mul table of high nibble
	pshufb	xgft5_lo, xtmpa		;Lookup mul table of low nibble
	pxor	xgft5_hi, xgft5_lo	;GF add high and low partials
	pand	xgft5_hi, xtmpl3
	pxor	xd2, xgft5_hi

	;dest6
	pshufb	xgft6_hi, x0		;Lookup mul table of high nibble
	pshufb	xgft6_lo, xtmpa		;Lookup mul table of low nibble
	pxor	xgft6_hi, xgft6_lo	;GF add high and low partials
	pand	xgft6_hi, xtmpl3
	pxor	xd3, xgft6_hi

	XSTR	[dest4+tmp], xd1	;Store result into dest4
	XSTR	[dest5+tmp], xd2	;Store result into dest5
	XSTR	[dest6+tmp], xd3	;Store result into dest6

	movdqu	xgft4_lo, [tmp3]	;Load array Ax{00}, Ax{01}, Ax{02}, ...
	movdqu	xgft4_hi, [tmp3+16]	; " Ax{00}, Ax{10}, Ax{20}, ... , Ax{f0}
	movdqu	xgft5_lo, [tmp3+vec]	;Load array Bx{00}, Bx{01}, Bx{02}, ...
	movdqu	xgft5_hi, [tmp3+vec+16]	; " Bx{00}, Bx{10}, Bx{20}, ... , Bx{f0}
	movdqu	xgft6_lo, [tmp3+2*vec]	;Load array Cx{00}, Cx{01}, Cx{02}, ...
	movdqu	xgft6_hi, [tmp3+2*vec+16]	; " Cx{00}, Cx{10}, Cx{20}, ... , Cx{f0}
	XLDR	xd1, [dest1+tmp]	;Get next dest vector
	XLDR	xd2, [dest2+tmp]	;reuse xtmpl1. Get next dest vector
	XLDR	xd3, [dest3+tmp]	;reuse xtmph1. Get next dest3 vector

	;dest1
	pshufb	xgft4_hi, x0		;Lookup mul table of high nibble
	pshufb	xgft4_lo, xtmpa		;Lookup mul table of low nibble
	pxor	xgft4_hi, xgft4_lo		;GF add high and low partials
	pand	xgft4_hi, xtmpl3
	pxor	xd1, xgft4_hi

	;dest2
	pshufb	xgft5_hi, x0		;Lookup mul table of high nibble
	pshufb	xgft5_lo, xtmpa		;Lookup mul table of low nibble
	pxor	xgft5_hi, xgft5_lo	;GF add high and low partials
	pand	xgft5_hi, xtmpl3
	pxor	xd2, xgft5_hi

	;dest3
	pshufb	xgft6_hi, x0		;Lookup mul table of high nibble
	pshufb	xgft6_lo, xtmpa		;Lookup mul table of low nibble
	pxor	xgft6_hi, xgft6_lo	;GF add high and low partials
	pand	xgft6_hi, xtmpl3
	pxor	xd3, xgft6_hi

	XSTR	[dest1+tmp], xd1	;Store result into dest1
	XSTR	[dest2+tmp], xd2	;Store result into dest2
	XSTR	[dest3+tmp], xd3	;Store result into dest3

.return_pass:
	FUNC_RESTORE
	mov	return, 0
	ret

.return_fail:
	FUNC_RESTORE
	mov	return, 1
	ret

endproc_frame

section .data

align 16

mask0f: ddq 0x0f0f0f0f0f0f0f0f0f0f0f0f0f0f0f0f
constip16:
	ddq 0xf0f1f2f3f4f5f6f7f8f9fafbfcfdfeff

;;;       func             core, ver, snum
slversion gf_6vect_mad_sse, 00,  01,  020f
Commit	Line	Data
7c673cae FG	1	;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
	2	; Copyright(c) 2011-2015 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	;;;
	31	;;; gf_6vect_mad_sse(len, vec, vec_i, mul_array, src, dest);
	32	;;;
	33
	34	%include "reg_sizes.asm"
	35
	36	%define PS 8
	37
	38	%ifidn __OUTPUT_FORMAT__, win64
	39	%define arg0 rcx
	40	%define arg0.w ecx
	41	%define arg1 rdx
	42	%define arg2 r8
	43	%define arg3 r9
	44	%define arg4 r12
	45	%define arg5 r15
	46	%define tmp r11
	47	%define tmp.w r11d
	48	%define tmp2 r10
	49	%define tmp3 r13
	50	%define tmp4 r14
	51	%define tmp5 rdi
	52	%define return rax
	53	%define return.w eax
	54	%define stack_size 1610 + 58
	55	%define arg(x) [rsp + stack_size + PS + PS*x]
	56	%define func(x) proc_frame x
	57
	58	%macro FUNC_SAVE 0
	59	sub rsp, stack_size
	60	movdqa [rsp+16*0],xmm6
	61	movdqa [rsp+16*1],xmm7
	62	movdqa [rsp+16*2],xmm8
	63	movdqa [rsp+16*3],xmm9
	64	movdqa [rsp+16*4],xmm10
65	movdqa [rsp+16*5],xmm11
66	movdqa [rsp+16*6],xmm12
67	movdqa [rsp+16*7],xmm13
68	movdqa [rsp+16*8],xmm14
69	movdqa [rsp+16*9],xmm15
70	save_reg r12, 1016 + 08
71	save_reg r13, 1016 + 18
72	save_reg r14, 1016 + 28
73	save_reg r15, 1016 + 38
74	save_reg rdi, 1016 + 48
75	end_prolog
76	mov arg4, arg(4)
77	mov arg5, arg(5)
78	%endmacro
79
80	%macro FUNC_RESTORE 0
81	movdqa xmm6, [rsp+16*0]
82	movdqa xmm7, [rsp+16*1]
83	movdqa xmm8, [rsp+16*2]
84	movdqa xmm9, [rsp+16*3]
85	movdqa xmm10, [rsp+16*4]
86	movdqa xmm11, [rsp+16*5]
87	movdqa xmm12, [rsp+16*6]
88	movdqa xmm13, [rsp+16*7]
89	movdqa xmm14, [rsp+16*8]
90	movdqa xmm15, [rsp+16*9]
91	mov r12, [rsp + 1016 + 08]
92	mov r13, [rsp + 1016 + 18]
93	mov r14, [rsp + 1016 + 28]
94	mov r15, [rsp + 1016 + 38]
95	mov rdi, [rsp + 1016 + 48]
96	add rsp, stack_size
97	%endmacro
98
99	%elifidn __OUTPUT_FORMAT__, elf64
100	%define arg0 rdi
101	%define arg0.w edi
102	%define arg1 rsi
103	%define arg2 rdx
104	%define arg3 rcx
105	%define arg4 r8
106	%define arg5 r9
107	%define tmp r11
108	%define tmp.w r11d
109	%define tmp2 r10
110	%define tmp3 r12
111	%define tmp4 r13
112	%define tmp5 r14
113	%define return rax
114	%define return.w eax
115
116	%define func(x) x:
117	%macro FUNC_SAVE 0
118	push r12
119	push r13
120	push r14
121	%endmacro
122	%macro FUNC_RESTORE 0
123	pop r14
124	pop r13
125	pop r12
126	%endmacro
127	%endif
128
129	;;; gf_6vect_mad_sse(len, vec, vec_i, mul_array, src, dest)
130	%define len arg0
131	%define len.w arg0.w
132	%define vec arg1
133	%define vec_i arg2
134	%define mul_array arg3
135	%define src arg4
136	%define dest1 arg5
137	%define pos return
138	%define pos.w return.w
139
140	%define dest2 mul_array
141	%define dest3 tmp2
142	%define dest4 tmp4
143	%define dest5 tmp5
144	%define dest6 vec_i
145
146	%ifndef EC_ALIGNED_ADDR
147	;;; Use Un-aligned load/store
148	%define XLDR movdqu
149	%define XSTR movdqu
150	%else
151	;;; Use Non-temporal load/stor
152	%ifdef NO_NT_LDST
153	%define XLDR movdqa
154	%define XSTR movdqa
155	%else
156	%define XLDR movntdqa
157	%define XSTR movntdq
158	%endif
159	%endif
160
161	default rel
162
163	[bits 64]
164	section .text
165
166	%define xmask0f xmm15
167	%define xgft4_lo xmm14
168	%define xgft4_hi xmm13
169	%define xgft5_lo xmm12
170	%define xgft5_hi xmm11
171	%define xgft6_lo xmm10
172	%define xgft6_hi xmm9
173
174	%define x0 xmm0
175	%define xtmpa xmm1
176	%define xtmph1 xmm2
177	%define xtmpl1 xmm3
178	%define xtmph2 xmm4
179	%define xtmpl2 xmm5
180	%define xtmph3 xmm6
181	%define xtmpl3 xmm7
182	%define xd1 xmm8
183	%define xd2 xtmpl1
184	%define xd3 xtmph1
185
186
187	align 16
188	global gf_6vect_mad_sse:function
189	func(gf_6vect_mad_sse)
190	FUNC_SAVE
191	sub len, 16
192	jl .return_fail
193
194	xor pos, pos
195	movdqa xmask0f, [mask0f] ;Load mask of lower nibble in each byte
196
197	mov tmp, vec
198	sal vec_i, 5 ;Multiply by 32
199	lea tmp3, [mul_array + vec_i]
200	sal tmp, 6 ;Multiply by 64
201
202	sal vec, 5 ;Multiply by 32
203	lea vec_i, [tmp + vec] ;vec_i = 96
204	lea mul_array, [tmp + vec_i] ;mul_array = 160
205
206	movdqu xgft5_lo, [tmp3+2*tmp] ;Load array Ex{00}, Ex{01}, ..., Ex{0f}
207	movdqu xgft5_hi, [tmp3+2*tmp+16] ; " Ex{00}, Ex{10}, ..., Ex{f0}
208	movdqu xgft4_lo, [tmp3+vec_i] ;Load array Dx{00}, Dx{01}, Dx{02}, ...
209	movdqu xgft4_hi, [tmp3+vec_i+16] ; " Dx{00}, Dx{10}, Dx{20}, ... , Dx{f0}
210	movdqu xgft6_lo, [tmp3+mul_array] ;Load array Fx{00}, Fx{01}, ..., Fx{0f}
211	movdqu xgft6_hi, [tmp3+mul_array+16] ; " Fx{00}, Fx{10}, ..., Fx{f0}
212
213	mov dest2, [dest1+PS]
214	mov dest3, [dest1+2*PS]
215	mov dest4, [dest1+3*PS] ; reuse mul_array
216	mov dest5, [dest1+4*PS]
217	mov dest6, [dest1+5*PS] ; reuse vec_i
218	mov dest1, [dest1]
219
220	.loop16:
221	XLDR x0, [src+pos] ;Get next source vector
222
223	movdqu xtmpl1, [tmp3] ;Load array Ax{00}, Ax{01}, Ax{02}, ...
224	movdqu xtmph1, [tmp3+16] ; " Ax{00}, Ax{10}, Ax{20}, ... , Ax{f0}
225	movdqu xtmpl2, [tmp3+vec] ;Load array Bx{00}, Bx{01}, Bx{02}, ...
226	movdqu xtmph2, [tmp3+vec+16] ; " Bx{00}, Bx{10}, Bx{20}, ... , Bx{f0}
227	movdqu xtmpl3, [tmp3+2*vec] ;Load array Cx{00}, Cx{01}, Cx{02}, ...
228	movdqu xtmph3, [tmp3+2*vec+16] ; " Cx{00}, Cx{10}, Cx{20}, ... , Cx{f0}
229	XLDR xd1, [dest1+pos] ;Get next dest vector
230
231	movdqa xtmpa, x0 ;Keep unshifted copy of src
232	psraw x0, 4 ;Shift to put high nibble into bits 4-0
233	pand x0, xmask0f ;Mask high src nibble in bits 4-0
234	pand xtmpa, xmask0f ;Mask low src nibble in bits 4-0
235
236	;dest1
237	pshufb xtmph1, x0 ;Lookup mul table of high nibble
238	pshufb xtmpl1, xtmpa ;Lookup mul table of low nibble
239	pxor xtmph1, xtmpl1 ;GF add high and low partials
240	pxor xd1, xtmph1
241
242	XLDR xd2, [dest2+pos] ;reuse xtmpl1. Get next dest vector
243	XLDR xd3, [dest3+pos] ;reuse xtmph1. Get next dest3 vector
244
245	;dest2
246	pshufb xtmph2, x0 ;Lookup mul table of high nibble
247	pshufb xtmpl2, xtmpa ;Lookup mul table of low nibble
248	pxor xtmph2, xtmpl2 ;GF add high and low partials
249	pxor xd2, xtmph2
250
251	;dest3
252	pshufb xtmph3, x0 ;Lookup mul table of high nibble
253	pshufb xtmpl3, xtmpa ;Lookup mul table of low nibble
254	pxor xtmph3, xtmpl3 ;GF add high and low partials
255	pxor xd3, xtmph3
256
257	XSTR [dest1+pos], xd1 ;Store result into dest1
258	XSTR [dest2+pos], xd2 ;Store result into dest2
259	XSTR [dest3+pos], xd3 ;Store result into dest3
260
261	movdqa xtmph1, xgft4_hi ;Reload const array registers
262	movdqa xtmpl1, xgft4_lo ;Reload const array registers
263	movdqa xtmph2, xgft5_hi ;Reload const array registers
264	movdqa xtmpl2, xgft5_lo ;Reload const array registers
265	movdqa xtmph3, xgft6_hi ;Reload const array registers
266	movdqa xtmpl3, xgft6_lo ;Reload const array registers
267
268	;dest4
269	XLDR xd1, [dest4+pos] ;Get next dest vector
270	pshufb xtmph1, x0 ;Lookup mul table of high nibble
271	pshufb xtmpl1, xtmpa ;Lookup mul table of low nibble
272	pxor xtmph1, xtmpl1 ;GF add high and low partials
273	pxor xd1, xtmph1
274
275	XLDR xd2, [dest5+pos] ;reuse xtmpl1. Get next dest vector
276	XLDR xd3, [dest6+pos] ;reuse xtmph1. Get next dest vector
277
278	;dest5
279	pshufb xtmph2, x0 ;Lookup mul table of high nibble
280	pshufb xtmpl2, xtmpa ;Lookup mul table of low nibble
281	pxor xtmph2, xtmpl2 ;GF add high and low partials
282	pxor xd2, xtmph2
283
284	;dest6
285	pshufb xtmph3, x0 ;Lookup mul table of high nibble
286	pshufb xtmpl3, xtmpa ;Lookup mul table of low nibble
287	pxor xtmph3, xtmpl3 ;GF add high and low partials
288	pxor xd3, xtmph3
289
290	XSTR [dest4+pos], xd1 ;Store result into dest4
291	XSTR [dest5+pos], xd2 ;Store result into dest5
292	XSTR [dest6+pos], xd3 ;Store result into dest6
293
294	add pos, 16 ;Loop on 16 bytes at a time
295	cmp pos, len
296	jle .loop16
297
298	lea tmp, [len + 16]
299	cmp pos, tmp
300	je .return_pass
301
302	.lessthan16:
303	;; Tail len
304	;; Do one more overlap pass
305	;; Overlapped offset length-16
306	mov tmp, len ;Backup len as len=rdi
307
308	XLDR x0, [src+tmp] ;Get next source vector
309	XLDR xd1, [dest4+tmp] ;Get next dest vector
310	XLDR xd2, [dest5+tmp] ;reuse xtmpl1. Get next dest vector
311	XLDR xd3, [dest6+tmp] ;reuse xtmph1. Get next dest vector
312
313	sub len, pos
314
315	movdqa xtmph3, [constip16] ;Load const of i + 16
316	pinsrb xtmpl3, len.w, 15
317	pshufb xtmpl3, xmask0f ;Broadcast len to all bytes
318	pcmpgtb xtmpl3, xtmph3
319
320	movdqa xtmpa, x0 ;Keep unshifted copy of src
321	psraw x0, 4 ;Shift to put high nibble into bits 4-0
322	pand x0, xmask0f ;Mask high src nibble in bits 4-0
323	pand xtmpa, xmask0f ;Mask low src nibble in bits 4-0
324
325	;dest4
326	pshufb xgft4_hi, x0 ;Lookup mul table of high nibble
327	pshufb xgft4_lo, xtmpa ;Lookup mul table of low nibble
328	pxor xgft4_hi, xgft4_lo ;GF add high and low partials
329	pand xgft4_hi, xtmpl3
330	pxor xd1, xgft4_hi
331
332	;dest5
333	pshufb xgft5_hi, x0 ;Lookup mul table of high nibble
334	pshufb xgft5_lo, xtmpa ;Lookup mul table of low nibble
335	pxor xgft5_hi, xgft5_lo ;GF add high and low partials
336	pand xgft5_hi, xtmpl3
337	pxor xd2, xgft5_hi
338
339	;dest6
340	pshufb xgft6_hi, x0 ;Lookup mul table of high nibble
341	pshufb xgft6_lo, xtmpa ;Lookup mul table of low nibble
342	pxor xgft6_hi, xgft6_lo ;GF add high and low partials
343	pand xgft6_hi, xtmpl3
344	pxor xd3, xgft6_hi
345
346	XSTR [dest4+tmp], xd1 ;Store result into dest4
347	XSTR [dest5+tmp], xd2 ;Store result into dest5
348	XSTR [dest6+tmp], xd3 ;Store result into dest6
349
350	movdqu xgft4_lo, [tmp3] ;Load array Ax{00}, Ax{01}, Ax{02}, ...
351	movdqu xgft4_hi, [tmp3+16] ; " Ax{00}, Ax{10}, Ax{20}, ... , Ax{f0}
352	movdqu xgft5_lo, [tmp3+vec] ;Load array Bx{00}, Bx{01}, Bx{02}, ...
353	movdqu xgft5_hi, [tmp3+vec+16] ; " Bx{00}, Bx{10}, Bx{20}, ... , Bx{f0}
354	movdqu xgft6_lo, [tmp3+2*vec] ;Load array Cx{00}, Cx{01}, Cx{02}, ...
355	movdqu xgft6_hi, [tmp3+2*vec+16] ; " Cx{00}, Cx{10}, Cx{20}, ... , Cx{f0}
356	XLDR xd1, [dest1+tmp] ;Get next dest vector
357	XLDR xd2, [dest2+tmp] ;reuse xtmpl1. Get next dest vector
358	XLDR xd3, [dest3+tmp] ;reuse xtmph1. Get next dest3 vector
359
360	;dest1
361	pshufb xgft4_hi, x0 ;Lookup mul table of high nibble
362	pshufb xgft4_lo, xtmpa ;Lookup mul table of low nibble
363	pxor xgft4_hi, xgft4_lo ;GF add high and low partials
364	pand xgft4_hi, xtmpl3
365	pxor xd1, xgft4_hi
366
367	;dest2
368	pshufb xgft5_hi, x0 ;Lookup mul table of high nibble
369	pshufb xgft5_lo, xtmpa ;Lookup mul table of low nibble
370	pxor xgft5_hi, xgft5_lo ;GF add high and low partials
371	pand xgft5_hi, xtmpl3
372	pxor xd2, xgft5_hi
373
374	;dest3
375	pshufb xgft6_hi, x0 ;Lookup mul table of high nibble
376	pshufb xgft6_lo, xtmpa ;Lookup mul table of low nibble
377	pxor xgft6_hi, xgft6_lo ;GF add high and low partials
378	pand xgft6_hi, xtmpl3
379	pxor xd3, xgft6_hi
380
381	XSTR [dest1+tmp], xd1 ;Store result into dest1
382	XSTR [dest2+tmp], xd2 ;Store result into dest2
383	XSTR [dest3+tmp], xd3 ;Store result into dest3
384
385	.return_pass:
386	FUNC_RESTORE
387	mov return, 0
388	ret
389
390	.return_fail:
391	FUNC_RESTORE
392	mov return, 1
393	ret
394
395	endproc_frame
396
397	section .data
398
399	align 16
400
401	mask0f: ddq 0x0f0f0f0f0f0f0f0f0f0f0f0f0f0f0f0f
402	constip16:
403	ddq 0xf0f1f2f3f4f5f6f7f8f9fafbfcfdfeff
404
405	;;; func core, ver, snum
406	slversion gf_6vect_mad_sse, 00, 01, 020f