1dnl  SPARC64 mpn_gcd_11.
2
3dnl  Based on the K7 gcd_1.asm, by Kevin Ryde.  Rehacked for SPARC by Torbj��rn
4dnl  Granlund.
5
6dnl  Copyright 2000-2002, 2005, 2009, 2011-2013 Free Software Foundation, Inc.
7
8dnl  This file is part of the GNU MP Library.
9dnl
10dnl  The GNU MP Library is free software; you can redistribute it and/or modify
11dnl  it under the terms of either:
12dnl
13dnl    * the GNU Lesser General Public License as published by the Free
14dnl      Software Foundation; either version 3 of the License, or (at your
15dnl      option) any later version.
16dnl
17dnl  or
18dnl
19dnl    * the GNU General Public License as published by the Free Software
20dnl      Foundation; either version 2 of the License, or (at your option) any
21dnl      later version.
22dnl
23dnl  or both in parallel, as here.
24dnl
25dnl  The GNU MP Library is distributed in the hope that it will be useful, but
26dnl  WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
27dnl  or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
28dnl  for more details.
29dnl
30dnl  You should have received copies of the GNU General Public License and the
31dnl  GNU Lesser General Public License along with the GNU MP Library.  If not,
32dnl  see https://www.gnu.org/licenses/.
33
34include(`../config.m4')
35
36
37C		  cycles/bit (approx)
38C UltraSPARC 1&2:	 5.1
39C UltraSPARC 3:		 5.0
40C UltraSPARC T1:	11.4
41C UltraSPARC T3:	10
42C UltraSPARC T4:	 6
43C Numbers measured with: speed -CD -s32-64 -t32 mpn_gcd_1
44
45C ctz_table[n] is the number of trailing zeros on n, or MAXSHIFT if n==0.
46
47deflit(MAXSHIFT, 7)
48deflit(MASK, eval((m4_lshift(1,MAXSHIFT))-1))
49
50	RODATA
51	TYPE(ctz_table,object)
52ctz_table:
53	.byte	MAXSHIFT
54forloop(i,1,MASK,
55`	.byte	m4_count_trailing_zeros(i)
56')
57	SIZE(ctz_table,.-ctz_table)
58
59define(`u0',    `%o0')
60define(`v0',    `%o1')
61
62ASM_START()
63	REGISTER(%g2,#scratch)
64	REGISTER(%g3,#scratch)
65PROLOGUE(mpn_gcd_11)
66	LEA64(ctz_table, o5, g4)
67	b	L(odd)
68	 mov	u0, %o4
69
70	ALIGN(16)
71L(top):	movcc	%xcc, %o4, v0		C v = min(u,v)
72	movcc	%xcc, %o2, %o0		C u = |v - u]
73L(mid):	ldub	[%o5+%g3], %g5		C
74	brz,a,pn %g3, L(shift_alot)	C
75	 srlx	%o0, MAXSHIFT, %o0
76	srlx	%o0, %g5, %o4		C new u, odd
77L(odd):	subcc	v0, %o4, %o2		C v - u, set flags for branch and movcc
78	sub	%o4, v0, %o0		C u - v
79	bnz,pt	%xcc, L(top)		C
80	 and	%o2, MASK, %g3		C extract low MAXSHIFT bits from (v-u)
81
82	retl
83	 mov	v0, %o0
84
85L(shift_alot):
86	b	L(mid)
87	 and	%o0, MASK, %g3		C
88EPILOGUE()
89