1dnl  AMD64 mpn_sec_tabselect.
2
3dnl  Contributed to the GNU project by Torbj��rn Granlund.
4
5dnl  Copyright 2011-2013 Free Software Foundation, Inc.
6
7dnl  This file is part of the GNU MP Library.
8dnl
9dnl  The GNU MP Library is free software; you can redistribute it and/or modify
10dnl  it under the terms of either:
11dnl
12dnl    * the GNU Lesser General Public License as published by the Free
13dnl      Software Foundation; either version 3 of the License, or (at your
14dnl      option) any later version.
15dnl
16dnl  or
17dnl
18dnl    * the GNU General Public License as published by the Free Software
19dnl      Foundation; either version 2 of the License, or (at your option) any
20dnl      later version.
21dnl
22dnl  or both in parallel, as here.
23dnl
24dnl  The GNU MP Library is distributed in the hope that it will be useful, but
25dnl  WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
26dnl  or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
27dnl  for more details.
28dnl
29dnl  You should have received copies of the GNU General Public License and the
30dnl  GNU Lesser General Public License along with the GNU MP Library.  If not,
31dnl  see https://www.gnu.org/licenses/.
32
33include(`../config.m4')
34
35
36C	     cycles/limb          good for cpu
37C AMD K8,K9	 1.5			Y
38C AMD K10	 1.4
39C AMD bd1	 2.64
40C AMD bobcat	 2.15			Y
41C Intel P4	 4
42C Intel core2	 1.38
43C Intel NHM	 1.75
44C Intel SBR	 1.25
45C Intel atom	 2.5			Y
46C VIA nano	 1.75			Y
47
48C NOTES
49C  * This has not been tuned for any specific processor.  Its speed should not
50C    be too bad, though.
51C  * Using SSE2/AVX2 could result in many-fold speedup.
52C  * WORKS FOR n mod 4 = 0 ONLY!
53
54C mpn_sec_tabselect (mp_limb_t *rp, mp_limb_t *tp, mp_size_t n, mp_size_t nents, mp_size_t which)
55define(`rp',     `%rdi')
56define(`tp',     `%rsi')
57define(`n',      `%rdx')
58define(`nents',  `%rcx')
59define(`which',  `%r8')
60
61define(`i',      `%rbp')
62define(`j',      `%r9')
63
64C rax  rbx  rcx  rdx  rdi  rsi  rbp   r8   r9  r10  r11  r12  r13  r14  r15
65C          nents  n   rp   tab   i   which j    *    *    *    *    *    *
66
67ABI_SUPPORT(DOS64)
68ABI_SUPPORT(STD64)
69
70ASM_START()
71	TEXT
72	ALIGN(16)
73PROLOGUE(mpn_sec_tabselect)
74	FUNC_ENTRY(4)
75IFDOS(`	mov	56(%rsp), %r8d	')
76
77	push	%rbx
78	push	%rbp
79	push	%r12
80	push	%r13
81	push	%r14
82	push	%r15
83
84	mov	n, j
85	add	$-4, j
86	js	L(outer_end)
87
88L(outer_top):
89	mov	nents, i
90	push	tp
91	xor	R32(%r12), R32(%r12)
92	xor	R32(%r13), R32(%r13)
93	xor	R32(%r14), R32(%r14)
94	xor	R32(%r15), R32(%r15)
95	mov	which, %rbx
96
97	ALIGN(16)
98L(top):	sub	$1, %rbx
99	sbb	%rax, %rax
100	mov	0(tp), %r10
101	mov	8(tp), %r11
102	and	%rax, %r10
103	and	%rax, %r11
104	or	%r10, %r12
105	or	%r11, %r13
106	mov	16(tp), %r10
107	mov	24(tp), %r11
108	and	%rax, %r10
109	and	%rax, %r11
110	or	%r10, %r14
111	or	%r11, %r15
112	lea	(tp,n,8), tp
113	add	$-1, i
114	jne	L(top)
115
116	mov	%r12, 0(rp)
117	mov	%r13, 8(rp)
118	mov	%r14, 16(rp)
119	mov	%r15, 24(rp)
120	pop	tp
121	lea	32(tp), tp
122	lea	32(rp), rp
123	add	$-4, j
124	jns	L(outer_top)
125L(outer_end):
126
127	test	$2, R8(n)
128	jz	L(b0x)
129L(b1x):	mov	nents, i
130	push	tp
131	xor	R32(%r12), R32(%r12)
132	xor	R32(%r13), R32(%r13)
133	mov	which, %rbx
134	ALIGN(16)
135L(tp2):	sub	$1, %rbx
136	sbb	%rax, %rax
137	mov	0(tp), %r10
138	mov	8(tp), %r11
139	and	%rax, %r10
140	and	%rax, %r11
141	or	%r10, %r12
142	or	%r11, %r13
143	lea	(tp,n,8), tp
144	add	$-1, i
145	jne	L(tp2)
146	mov	%r12, 0(rp)
147	mov	%r13, 8(rp)
148	pop	tp
149	lea	16(tp), tp
150	lea	16(rp), rp
151
152L(b0x):	test	$1, R8(n)
153	jz	L(b00)
154L(b01):	mov	nents, i
155	xor	R32(%r12), R32(%r12)
156	mov	which, %rbx
157	ALIGN(16)
158L(tp1):	sub	$1, %rbx
159	sbb	%rax, %rax
160	mov	0(tp), %r10
161	and	%rax, %r10
162	or	%r10, %r12
163	lea	(tp,n,8), tp
164	add	$-1, i
165	jne	L(tp1)
166	mov	%r12, 0(rp)
167
168L(b00):	pop	%r15
169	pop	%r14
170	pop	%r13
171	pop	%r12
172	pop	%rbp
173	pop	%rbx
174	FUNC_EXIT()
175	ret
176EPILOGUE()
177