xref: /netbsd-src/external/lgpl3/gmp/dist/mpn/x86_64/invert_limb.asm (revision e6c7e151de239c49d2e38720a061ed9d1fa99309)
1dnl  AMD64 mpn_invert_limb -- Invert a normalized limb.
2
3dnl  Contributed to the GNU project by Torbjorn Granlund and Niels Möller.
4
5dnl  Copyright 2004, 2007-2009, 2011, 2012 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 (approx)	div
37C AMD K8,K9	 48			 71
38C AMD K10	 48			 77
39C Intel P4	135			161
40C Intel core2	 69			116
41C Intel corei	 55			 89
42C Intel atom	129			191
43C VIA nano	 79			157
44
45C rax rcx rdx rdi rsi r8
46
47ABI_SUPPORT(DOS64)
48ABI_SUPPORT(STD64)
49
50PROTECT(`mpn_invert_limb_table')
51
52ASM_START()
53	TEXT
54	ALIGN(16)
55PROLOGUE(mpn_invert_limb)		C			Kn	C2	Ci
56	FUNC_ENTRY(1)
57	mov	%rdi, %rax		C			 0	 0	 0
58	shr	$55, %rax		C			 1	 1	 1
59ifdef(`PIC',`
60ifdef(`DARWIN',`
61	mov	mpn_invert_limb_table@GOTPCREL(%rip), %r8
62	add	$-512, %r8
63',`
64	lea	-512+mpn_invert_limb_table(%rip), %r8
65')',`
66	movabs	$-512+mpn_invert_limb_table, %r8
67')
68	movzwl	(%r8,%rax,2), R32(%rcx)	C	%rcx = v0
69
70	C v1 = (v0 << 11) - (v0*v0*d40 >> 40) - 1
71	mov	%rdi, %rsi		C			 0	 0	 0
72	mov	R32(%rcx), R32(%rax)	C			 4	 5	 5
73	imul	R32(%rcx), R32(%rcx)	C			 4	 5	 5
74	shr	$24, %rsi		C			 1	 1	 1
75	inc	%rsi			C	%rsi = d40
76	imul	%rsi, %rcx		C			 8	10	 8
77	shr	$40, %rcx		C			12	15	11
78	sal	$11, R32(%rax)		C			 5	 6	 6
79	dec	R32(%rax)
80	sub	R32(%rcx), R32(%rax)	C	%rax = v1
81
82	C v2 = (v1 << 13) + (v1 * (2^60 - v1*d40) >> 47)
83	mov	$0x1000000000000000, %rcx
84	imul	%rax, %rsi		C			14	17	13
85	sub	%rsi, %rcx
86	imul	%rax, %rcx
87	sal	$13, %rax
88	shr	$47, %rcx
89	add	%rax, %rcx		C	%rcx = v2
90
91	C v3 = (v2 << 31) + (v2 * (2^96 - v2 * d63 + ((v2 >> 1) & mask)) >> 65
92	mov	%rdi, %rsi		C			 0	 0	 0
93	shr	%rsi			C d/2
94	sbb	%rax, %rax		C -d0 = -(d mod 2)
95	sub	%rax, %rsi		C d63 = ceil(d/2)
96	imul	%rcx, %rsi		C v2 * d63
97	and	%rcx, %rax		C v2 * d0
98	shr	%rax			C (v2>>1) * d0
99	sub	%rsi, %rax		C (v2>>1) * d0 - v2 * d63
100	mul	%rcx
101	sal	$31, %rcx
102	shr	%rdx
103	add	%rdx, %rcx		C	%rcx = v3
104
105	mov	%rdi, %rax
106	mul	%rcx
107	add	%rdi, %rax
108	mov	%rcx, %rax
109	adc	%rdi, %rdx
110	sub	%rdx, %rax
111
112	FUNC_EXIT()
113	ret
114EPILOGUE()
115ASM_END()
116