10l, there is a corner case afterall...
[libav.git] / libavcodec / ppc / float_altivec.c
1 /*
2 * Copyright (c) 2006 Luca Barbato <lu_zero@gentoo.org>
3 *
4 * This file is part of FFmpeg.
5 *
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
10 *
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20
21 #include "../dsputil.h"
22
23 #include "gcc_fixes.h"
24
25 #include "dsputil_altivec.h"
26
27 static void vector_fmul_altivec(float *dst, const float *src, int len)
28 {
29 int i;
30 vector float d0, d1, s, zero = (vector float)vec_splat_u32(0);
31 for(i=0; i<len-7; i+=8) {
32 d0 = vec_ld(0, dst+i);
33 s = vec_ld(0, src+i);
34 d1 = vec_ld(16, dst+i);
35 d0 = vec_madd(d0, s, zero);
36 d1 = vec_madd(d1, vec_ld(16,src+i), zero);
37 vec_st(d0, 0, dst+i);
38 vec_st(d1, 16, dst+i);
39 }
40 }
41
42 static void vector_fmul_reverse_altivec(float *dst, const float *src0,
43 const float *src1, int len)
44 {
45 int i;
46 vector float d, s0, s1, h0, l0,
47 s2, s3, zero = (vector float)vec_splat_u32(0);
48 src1 += len-4;
49 for(i=0; i<len-7; i+=8) {
50 s1 = vec_ld(0, src1-i); // [a,b,c,d]
51 s0 = vec_ld(0, src0+i);
52 l0 = vec_mergel(s1, s1); // [c,c,d,d]
53 s3 = vec_ld(-16, src1-i);
54 h0 = vec_mergeh(s1, s1); // [a,a,b,b]
55 s2 = vec_ld(16, src0+i);
56 s1 = vec_mergeh(vec_mergel(l0,h0), // [d,b,d,b]
57 vec_mergeh(l0,h0)); // [c,a,c,a]
58 // [d,c,b,a]
59 l0 = vec_mergel(s3, s3);
60 d = vec_madd(s0, s1, zero);
61 h0 = vec_mergeh(s3, s3);
62 vec_st(d, 0, dst+i);
63 s3 = vec_mergeh(vec_mergel(l0,h0),
64 vec_mergeh(l0,h0));
65 d = vec_madd(s2, s3, zero);
66 vec_st(d, 16, dst+i);
67 }
68 }
69
70 static void vector_fmul_add_add_altivec(float *dst, const float *src0,
71 const float *src1, const float *src2,
72 int src3, int len, int step)
73 {
74 int i;
75 vector float d, s0, s1, s2, t0, t1, edges;
76 vector unsigned char align = vec_lvsr(0,dst),
77 mask = vec_lvsl(0, dst);
78
79 #if 0 //FIXME: there is still something wrong
80 if (step == 2) {
81 int y;
82 vector float d0, d1, s3, t2;
83 vector unsigned int sel =
84 vec_mergeh(vec_splat_u32(-1), vec_splat_u32(0));
85 t1 = vec_ld(16, dst);
86 for (i=0,y=0; i<len-3; i+=4,y+=8) {
87
88 s0 = vec_ld(0,src0+i);
89 s1 = vec_ld(0,src1+i);
90 s2 = vec_ld(0,src2+i);
91
92 // t0 = vec_ld(0, dst+y); //[x x x|a]
93 // t1 = vec_ld(16, dst+y); //[b c d|e]
94 t2 = vec_ld(31, dst+y); //[f g h|x]
95
96 d = vec_madd(s0,s1,s2); // [A B C D]
97
98 // [A A B B]
99
100 // [C C D D]
101
102 d0 = vec_perm(t0, t1, mask); // [a b c d]
103
104 d0 = vec_sel(vec_mergeh(d, d), d0, sel); // [A b B d]
105
106 edges = vec_perm(t1, t0, mask);
107
108 t0 = vec_perm(edges, d0, align); // [x x x|A]
109
110 t1 = vec_perm(d0, edges, align); // [b B d|e]
111
112 vec_stl(t0, 0, dst+y);
113
114 d1 = vec_perm(t1, t2, mask); // [e f g h]
115
116 d1 = vec_sel(vec_mergel(d, d), d1, sel); // [C f D h]
117
118 edges = vec_perm(t2, t1, mask);
119
120 t1 = vec_perm(edges, d1, align); // [b B d|C]
121
122 t2 = vec_perm(d1, edges, align); // [f D h|x]
123
124 vec_stl(t1, 16, dst+y);
125
126 t0 = t1;
127
128 vec_stl(t2, 31, dst+y);
129
130 t1 = t2;
131 }
132 } else
133 #endif
134 if (step == 1 && src3 == 0)
135 for (i=0; i<len-3; i+=4) {
136 t0 = vec_ld(0, dst+i);
137 t1 = vec_ld(15, dst+i);
138 s0 = vec_ld(0, src0+i);
139 s1 = vec_ld(0, src1+i);
140 s2 = vec_ld(0, src2+i);
141 edges = vec_perm(t1 ,t0, mask);
142 d = vec_madd(s0,s1,s2);
143 t1 = vec_perm(d, edges, align);
144 t0 = vec_perm(edges, d, align);
145 vec_st(t1, 15, dst+i);
146 vec_st(t0, 0, dst+i);
147 }
148 else
149 ff_vector_fmul_add_add_c(dst, src0, src1, src2, src3, len, step);
150 }
151
152 void float_to_int16_altivec(int16_t *dst, const float *src, int len)
153 {
154 int i;
155 vector float s0, s1;
156 vector signed int t0, t1;
157 vector signed short d0, d1, d;
158 vector unsigned char align;
159 if(((long)dst)&15) //FIXME
160 for(i=0; i<len-7; i+=8) {
161 s0 = vec_ld(0, src+i);
162 s1 = vec_ld(16, src+i);
163 t0 = vec_cts(s0, 0);
164 d0 = vec_ld(0, dst+i);
165 t1 = vec_cts(s1, 0);
166 d1 = vec_ld(15, dst+i);
167 d = vec_packs(t0,t1);
168 d1 = vec_perm(d1, d0, vec_lvsl(0,dst+i));
169 align = vec_lvsr(0, dst+i);
170 d0 = vec_perm(d1, d, align);
171 d1 = vec_perm(d, d1, align);
172 vec_st(d0, 0, dst+i);
173 vec_st(d1,15, dst+i);
174 }
175 else
176 for(i=0; i<len-7; i+=8) {
177 s0 = vec_ld(0, src+i);
178 s1 = vec_ld(16, src+i);
179 t0 = vec_cts(s0, 0);
180 t1 = vec_cts(s1, 0);
181 d = vec_packs(t0,t1);
182 vec_st(d, 0, dst+i);
183 }
184 }
185
186 void float_init_altivec(DSPContext* c, AVCodecContext *avctx)
187 {
188 c->vector_fmul = vector_fmul_altivec;
189 c->vector_fmul_reverse = vector_fmul_reverse_altivec;
190 c->vector_fmul_add_add = vector_fmul_add_add_altivec;
191 if(!(avctx->flags & CODEC_FLAG_BITEXACT))
192 c->float_to_int16 = float_to_int16_altivec;
193 }