0
|
1 /* Copyright (C) 2008, 2009 Free Software Foundation, Inc.
|
|
2
|
|
3 This file is free software; you can redistribute it and/or modify it under
|
|
4 the terms of the GNU General Public License as published by the Free
|
|
5 Software Foundation; either version 3 of the License, or (at your option)
|
|
6 any later version.
|
|
7
|
|
8 This file is distributed in the hope that it will be useful, but WITHOUT
|
|
9 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
|
|
10 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
|
11 for more details.
|
|
12
|
|
13 Under Section 7 of GPL version 3, you are granted additional
|
|
14 permissions described in the GCC Runtime Library Exception, version
|
|
15 3.1, as published by the Free Software Foundation.
|
|
16
|
|
17 You should have received a copy of the GNU General Public License and
|
|
18 a copy of the GCC Runtime Library Exception along with this program;
|
|
19 see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
|
|
20 <http://www.gnu.org/licenses/>. */
|
|
21
|
|
22 #include <spu_intrinsics.h>
|
|
23
|
|
24 typedef unsigned int UTItype __attribute__ ((mode (TI)));
|
|
25 typedef int TItype __attribute__ ((mode (TI)));
|
|
26 TItype __divti3 (TItype u, TItype v);
|
|
27 TItype __modti3 (TItype u, TItype v);
|
|
28 UTItype __udivti3 (UTItype u, UTItype v);
|
|
29 UTItype __umodti3 (UTItype u, UTItype v);
|
|
30 UTItype __udivmodti4 (UTItype u, UTItype v, UTItype *w);
|
|
31
|
|
32 inline static unsigned int
|
|
33 count_leading_zeros (UTItype x)
|
|
34 {
|
|
35 qword c = si_clz (*(qword *) & x);
|
|
36 qword cmp0 = si_cgti (c, 31);
|
|
37 qword cmp1 = si_and (cmp0, si_shlqbyi (cmp0, 4));
|
|
38 qword cmp2 = si_and (cmp1, si_shlqbyi (cmp0, 8));
|
|
39 qword s = si_a (c, si_and (cmp0, si_shlqbyi (c, 4)));
|
|
40 s = si_a (s, si_and (cmp1, si_shlqbyi (c, 8)));
|
|
41 s = si_a (s, si_and (cmp2, si_shlqbyi (c, 12)));
|
|
42 return si_to_uint (s);
|
|
43 }
|
|
44
|
|
45 /* Based on implementation of udivmodsi4, which is essentially
|
|
46 * an optimized version of gcc/config/udivmodsi4.c
|
|
47 clz %7,%2
|
|
48 clz %4,%1
|
|
49 il %5,1
|
|
50 fsmbi %0,0
|
|
51 sf %7,%4,%7
|
|
52 ori %3,%1,0
|
|
53 shl %5,%5,%7
|
|
54 shl %4,%2,%7
|
|
55 1: or %8,%0,%5
|
|
56 rotmi %5,%5,-1
|
|
57 clgt %6,%4,%3
|
|
58 sf %7,%4,%3
|
|
59 rotmi %4,%4,-1
|
|
60 selb %0,%8,%0,%6
|
|
61 selb %3,%7,%3,%6
|
|
62 3: brnz %5,1b
|
|
63 */
|
|
64
|
|
65 UTItype
|
|
66 __udivmodti4 (UTItype num, UTItype den, UTItype * rp)
|
|
67 {
|
|
68 qword shift =
|
|
69 si_from_uint (count_leading_zeros (den) - count_leading_zeros (num));
|
|
70 qword n0 = *(qword *) & num;
|
|
71 qword d0 = *(qword *) & den;
|
|
72 qword bit = si_andi (si_fsmbi (1), 1);
|
|
73 qword r0 = si_il (0);
|
|
74 qword m1 = si_fsmbi (0x000f);
|
|
75 qword mask, r1, n1;
|
|
76
|
|
77 d0 = si_shlqbybi (si_shlqbi (d0, shift), shift);
|
|
78 bit = si_shlqbybi (si_shlqbi (bit, shift), shift);
|
|
79
|
|
80 do
|
|
81 {
|
|
82 r1 = si_or (r0, bit);
|
|
83
|
|
84 // n1 = n0 - d0 in TImode
|
|
85 n1 = si_bg (d0, n0);
|
|
86 n1 = si_shlqbyi (n1, 4);
|
|
87 n1 = si_sf (m1, n1);
|
|
88 n1 = si_bgx (d0, n0, n1);
|
|
89 n1 = si_shlqbyi (n1, 4);
|
|
90 n1 = si_sf (m1, n1);
|
|
91 n1 = si_bgx (d0, n0, n1);
|
|
92 n1 = si_shlqbyi (n1, 4);
|
|
93 n1 = si_sf (m1, n1);
|
|
94 n1 = si_sfx (d0, n0, n1);
|
|
95
|
|
96 mask = si_fsm (si_cgti (n1, -1));
|
|
97 r0 = si_selb (r0, r1, mask);
|
|
98 n0 = si_selb (n0, n1, mask);
|
|
99 bit = si_rotqmbii (bit, -1);
|
|
100 d0 = si_rotqmbii (d0, -1);
|
|
101 }
|
|
102 while (si_to_uint (si_orx (bit)));
|
|
103 if (rp)
|
|
104 *rp = *(UTItype *) & n0;
|
|
105 return *(UTItype *) & r0;
|
|
106 }
|
|
107
|
|
108 UTItype
|
|
109 __udivti3 (UTItype n, UTItype d)
|
|
110 {
|
|
111 return __udivmodti4 (n, d, (UTItype *)0);
|
|
112 }
|
|
113
|
|
114 UTItype
|
|
115 __umodti3 (UTItype n, UTItype d)
|
|
116 {
|
|
117 UTItype w;
|
|
118 __udivmodti4 (n, d, &w);
|
|
119 return w;
|
|
120 }
|
|
121
|
|
122 TItype
|
|
123 __divti3 (TItype n, TItype d)
|
|
124 {
|
|
125 int c = 0;
|
|
126 TItype w;
|
|
127
|
|
128 if (n < 0)
|
|
129 {
|
|
130 c = ~c;
|
|
131 n = -n;
|
|
132 }
|
|
133 if (d < 0)
|
|
134 {
|
|
135 c = ~c;
|
|
136 d = -d;
|
|
137 }
|
|
138
|
|
139 w = __udivmodti4 (n, d, (UTItype *)0);
|
|
140 if (c)
|
|
141 w = -w;
|
|
142 return w;
|
|
143 }
|
|
144
|
|
145 TItype
|
|
146 __modti3 (TItype n, TItype d)
|
|
147 {
|
|
148 int c = 0;
|
|
149 TItype w;
|
|
150
|
|
151 if (n < 0)
|
|
152 {
|
|
153 c = ~c;
|
|
154 n = -n;
|
|
155 }
|
|
156 if (d < 0)
|
|
157 {
|
|
158 c = ~c;
|
|
159 d = -d;
|
|
160 }
|
|
161
|
|
162 __udivmodti4 (n, d, (UTItype *) &w);
|
|
163 if (c)
|
|
164 w = -w;
|
|
165 return w;
|
|
166 }
|