2 * Copyright (c) 2006 Luca Barbato <lu_zero@gentoo.org>
4 * This library is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU Lesser General Public
6 * License as published by the Free Software Foundation; either
7 * version 2 of the License, or (at your option) any later version.
9 * This library is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * Lesser General Public License for more details.
14 * You should have received a copy of the GNU Lesser General Public
15 * License along with this library; if not, write to the Free Software
16 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 #include "../dsputil.h"
21 #include "gcc_fixes.h"
23 #include "dsputil_altivec.h"
25 static void vector_fmul_altivec(float *dst, const float *src, int len)
28 vector float d0, d1, s, zero = (vector float)vec_splat_u32(0);
29 for(i=0; i<len-7; i+=8) {
30 d0 = vec_ld(0, dst+i);
32 d1 = vec_ld(16, dst+i);
33 d0 = vec_madd(d0, s, zero);
34 d1 = vec_madd(d1, vec_ld(16,src+i), zero);
36 vec_st(d1, 16, dst+i);
40 static void vector_fmul_reverse_altivec(float *dst, const float *src0,
41 const float *src1, int len)
44 vector float d, s0, s1, h0, l0,
45 s2, s3, zero = (vector float)vec_splat_u32(0);
47 for(i=0; i<len-7; i+=8) {
48 s1 = vec_ld(0, src1-i); // [a,b,c,d]
49 s0 = vec_ld(0, src0+i);
50 l0 = vec_mergel(s1, s1); // [c,c,d,d]
51 s3 = vec_ld(-16, src1-i);
52 h0 = vec_mergeh(s1, s1); // [a,a,b,b]
53 s2 = vec_ld(16, src0+i);
54 s1 = vec_mergeh(vec_mergel(l0,h0), // [d,b,d,b]
55 vec_mergeh(l0,h0)); // [c,a,c,a]
57 l0 = vec_mergel(s3, s3);
58 d = vec_madd(s0, s1, zero);
59 h0 = vec_mergeh(s3, s3);
61 s3 = vec_mergeh(vec_mergel(l0,h0),
63 d = vec_madd(s2, s3, zero);
68 static void vector_fmul_add_add_altivec(float *dst, const float *src0,
69 const float *src1, const float *src2,
70 int src3, int len, int step)
73 vector float d, s0, s1, s2, t0, t1, edges;
74 vector unsigned char align = vec_lvsr(0,dst),
75 mask = vec_lvsl(0, dst);
78 #if 0 //FIXME: there is still something wrong
81 vector float d0, d1, s3, t2;
82 vector unsigned int sel =
83 vec_mergeh(vec_splat_u32(-1), vec_splat_u32(0));
85 for (i=0,y=0; i<len-3; i+=4,y+=8) {
87 s0 = vec_ld(0,src0+i);
88 s1 = vec_ld(0,src1+i);
89 s2 = vec_ld(0,src2+i);
91 // t0 = vec_ld(0, dst+y); //[x x x|a]
92 // t1 = vec_ld(16, dst+y); //[b c d|e]
93 t2 = vec_ld(31, dst+y); //[f g h|x]
95 d = vec_madd(s0,s1,s2); // [A B C D]
101 d0 = vec_perm(t0, t1, mask); // [a b c d]
103 d0 = vec_sel(vec_mergeh(d, d), d0, sel); // [A b B d]
105 edges = vec_perm(t1, t0, mask);
107 t0 = vec_perm(edges, d0, align); // [x x x|A]
109 t1 = vec_perm(d0, edges, align); // [b B d|e]
111 vec_stl(t0, 0, dst+y);
113 d1 = vec_perm(t1, t2, mask); // [e f g h]
115 d1 = vec_sel(vec_mergel(d, d), d1, sel); // [C f D h]
117 edges = vec_perm(t2, t1, mask);
119 t1 = vec_perm(edges, d1, align); // [b B d|C]
121 t2 = vec_perm(d1, edges, align); // [f D h|x]
123 vec_stl(t1, 16, dst+y);
127 vec_stl(t2, 31, dst+y);
133 if (step == 1 && src3 == 0)
134 for (i=0; i<len-3; i+=4) {
135 t1 = vec_ld(15, dst+i);
136 s0 = vec_ld(0, src0+i);
137 s1 = vec_ld(0, src1+i);
138 s2 = vec_ld(0, src2+i);
139 edges = vec_perm(t1 ,t0, mask);
140 d = vec_madd(s0,s1,s2);
141 t1 = vec_perm(d, edges, align);
142 t0 = vec_perm(edges, d, align);
143 vec_st(t1, 15, dst+i);
144 vec_st(t0, 0, dst+i);
148 ff_vector_fmul_add_add_c(dst, src0, src1, src2, src3, len, step);
151 void float_to_int16_altivec(int16_t *dst, const float *src, int len)
155 vector signed int t0, t1;
156 vector signed short d0, d1, d;
157 vector unsigned char align;
158 if(((long)dst)&15) //FIXME
159 for(i=0; i<len-7; i+=8) {
160 s0 = vec_ld(0, src+i);
161 s1 = vec_ld(16, src+i);
163 d0 = vec_ld(0, dst+i);
165 d1 = vec_ld(15, dst+i);
166 d = vec_packs(t0,t1);
167 d1 = vec_perm(d1, d0, vec_lvsl(0,dst+i));
168 align = vec_lvsr(0, dst+i);
169 d0 = vec_perm(d1, d, align);
170 d1 = vec_perm(d, d1, align);
171 vec_st(d0, 0, dst+i);
172 vec_st(d1,15, dst+i);
175 for(i=0; i<len-7; i+=8) {
176 s0 = vec_ld(0, src+i);
177 s1 = vec_ld(16, src+i);
180 d = vec_packs(t0,t1);
185 void float_init_altivec(DSPContext* c, AVCodecContext *avctx)
187 c->vector_fmul = vector_fmul_altivec;
188 c->vector_fmul_reverse = vector_fmul_reverse_altivec;
189 c->vector_fmul_add_add = vector_fmul_add_add_altivec;
190 if(!(avctx->flags & CODEC_FLAG_BITEXACT))
191 c->float_to_int16 = float_to_int16_altivec;