2 * Copyright (c) 2006 Luca Barbato <lu_zero@gentoo.org>
4 * This file is part of FFmpeg.
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.
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.
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
21 #include "libavcodec/dsputil.h"
23 #include "dsputil_altivec.h"
24 #include "util_altivec.h"
26 static void vector_fmul_altivec(float *dst, const float *src, int len)
29 vector float d0, d1, s, zero = (vector float)vec_splat_u32(0);
30 for(i=0; i<len-7; i+=8) {
31 d0 = vec_ld(0, dst+i);
33 d1 = vec_ld(16, dst+i);
34 d0 = vec_madd(d0, s, zero);
35 d1 = vec_madd(d1, vec_ld(16,src+i), zero);
37 vec_st(d1, 16, dst+i);
41 static void vector_fmul_reverse_altivec(float *dst, const float *src0,
42 const float *src1, int len)
45 vector float d, s0, s1, h0, l0,
46 s2, s3, zero = (vector float)vec_splat_u32(0);
48 for(i=0; i<len-7; i+=8) {
49 s1 = vec_ld(0, src1-i); // [a,b,c,d]
50 s0 = vec_ld(0, src0+i);
51 l0 = vec_mergel(s1, s1); // [c,c,d,d]
52 s3 = vec_ld(-16, src1-i);
53 h0 = vec_mergeh(s1, s1); // [a,a,b,b]
54 s2 = vec_ld(16, src0+i);
55 s1 = vec_mergeh(vec_mergel(l0,h0), // [d,b,d,b]
56 vec_mergeh(l0,h0)); // [c,a,c,a]
58 l0 = vec_mergel(s3, s3);
59 d = vec_madd(s0, s1, zero);
60 h0 = vec_mergeh(s3, s3);
62 s3 = vec_mergeh(vec_mergel(l0,h0),
64 d = vec_madd(s2, s3, zero);
69 static void vector_fmul_add_altivec(float *dst, const float *src0,
70 const float *src1, const float *src2,
74 vector float d, s0, s1, s2, t0, t1, edges;
75 vector unsigned char align = vec_lvsr(0,dst),
76 mask = vec_lvsl(0, dst);
78 for (i=0; i<len-3; i+=4) {
79 t0 = vec_ld(0, dst+i);
80 t1 = vec_ld(15, dst+i);
81 s0 = vec_ld(0, src0+i);
82 s1 = vec_ld(0, src1+i);
83 s2 = vec_ld(0, src2+i);
84 edges = vec_perm(t1 ,t0, mask);
85 d = vec_madd(s0,s1,s2);
86 t1 = vec_perm(d, edges, align);
87 t0 = vec_perm(edges, d, align);
88 vec_st(t1, 15, dst+i);
93 static void vector_fmul_window_altivec(float *dst, const float *src0, const float *src1, const float *win, float add_bias, int len)
99 vector float vadd_bias, zero, t0, t1, s0, s1, wi, wj;
100 const vector unsigned char reverse = vcprm(3,2,1,0);
107 vadd.s[0] = add_bias;
108 vadd_bias = vec_splat(vadd.v, 0);
109 zero = (vector float)vec_splat_u32(0);
111 for(i=-len*4, j=len*4-16; i<0; i+=16, j-=16) {
112 s0 = vec_ld(i, src0);
113 s1 = vec_ld(j, src1);
117 s1 = vec_perm(s1, s1, reverse);
118 wj = vec_perm(wj, wj, reverse);
120 t0 = vec_madd(s0, wj, vadd_bias);
121 t0 = vec_nmsub(s1, wi, t0);
122 t1 = vec_madd(s0, wi, vadd_bias);
123 t1 = vec_madd(s1, wj, t1);
124 t1 = vec_perm(t1, t1, reverse);
131 static void int32_to_float_fmul_scalar_altivec(float *dst, const int *src, float mul, int len)
138 vector float src1, src2, dst1, dst2, mul_v, zero;
140 zero = (vector float)vec_splat_u32(0);
142 mul_v = vec_splat(mul_u.v, 0);
144 for(i=0; i<len; i+=8) {
145 src1 = vec_ctf(vec_ld(0, src+i), 0);
146 src2 = vec_ctf(vec_ld(16, src+i), 0);
147 dst1 = vec_madd(src1, mul_v, zero);
148 dst2 = vec_madd(src2, mul_v, zero);
149 vec_st(dst1, 0, dst+i);
150 vec_st(dst2, 16, dst+i);
155 static vector signed short
156 float_to_int16_one_altivec(const float *src)
158 vector float s0 = vec_ld(0, src);
159 vector float s1 = vec_ld(16, src);
160 vector signed int t0 = vec_cts(s0, 0);
161 vector signed int t1 = vec_cts(s1, 0);
162 return vec_packs(t0,t1);
165 static void float_to_int16_altivec(int16_t *dst, const float *src, long len)
168 vector signed short d0, d1, d;
169 vector unsigned char align;
170 if(((long)dst)&15) //FIXME
171 for(i=0; i<len-7; i+=8) {
172 d0 = vec_ld(0, dst+i);
173 d = float_to_int16_one_altivec(src+i);
174 d1 = vec_ld(15, dst+i);
175 d1 = vec_perm(d1, d0, vec_lvsl(0,dst+i));
176 align = vec_lvsr(0, dst+i);
177 d0 = vec_perm(d1, d, align);
178 d1 = vec_perm(d, d1, align);
179 vec_st(d0, 0, dst+i);
180 vec_st(d1,15, dst+i);
183 for(i=0; i<len-7; i+=8) {
184 d = float_to_int16_one_altivec(src+i);
190 float_to_int16_interleave_altivec(int16_t *dst, const float **src,
191 long len, int channels)
194 vector signed short d0, d1, d2, c0, c1, t0, t1;
195 vector unsigned char align;
197 float_to_int16_altivec(dst, src[0], len);
201 for(i=0; i<len-7; i+=8) {
202 d0 = vec_ld(0, dst + i);
203 t0 = float_to_int16_one_altivec(src[0] + i);
204 d1 = vec_ld(31, dst + i);
205 t1 = float_to_int16_one_altivec(src[1] + i);
206 c0 = vec_mergeh(t0, t1);
207 c1 = vec_mergel(t0, t1);
208 d2 = vec_perm(d1, d0, vec_lvsl(0, dst + i));
209 align = vec_lvsr(0, dst + i);
210 d0 = vec_perm(d2, c0, align);
211 d1 = vec_perm(c0, c1, align);
212 vec_st(d0, 0, dst + i);
213 d0 = vec_perm(c1, d2, align);
214 vec_st(d1, 15, dst + i);
215 vec_st(d0, 31, dst + i);
219 for(i=0; i<len-7; i+=8) {
220 t0 = float_to_int16_one_altivec(src[0] + i);
221 t1 = float_to_int16_one_altivec(src[1] + i);
222 d0 = vec_mergeh(t0, t1);
223 d1 = vec_mergel(t0, t1);
224 vec_st(d0, 0, dst + i);
225 vec_st(d1, 16, dst + i);
229 DECLARE_ALIGNED(16, int16_t, tmp)[len];
231 for (c = 0; c < channels; c++) {
232 float_to_int16_altivec(tmp, src[c], len);
233 for (i = 0, j = c; i < len; i++, j+=channels) {
240 void float_init_altivec(DSPContext* c, AVCodecContext *avctx)
242 c->vector_fmul = vector_fmul_altivec;
243 c->vector_fmul_reverse = vector_fmul_reverse_altivec;
244 c->vector_fmul_add = vector_fmul_add_altivec;
245 c->int32_to_float_fmul_scalar = int32_to_float_fmul_scalar_altivec;
246 if(!(avctx->flags & CODEC_FLAG_BITEXACT)) {
247 c->vector_fmul_window = vector_fmul_window_altivec;
248 c->float_to_int16 = float_to_int16_altivec;
249 c->float_to_int16_interleave = float_to_int16_interleave_altivec;