Statistics
| Branch: | Revision:

ffmpeg / libavcodec / ppc / float_altivec.c @ 5137235e

History | View | Annotate | Download (9.37 KB)

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 "libavcodec/dsputil.h"
22

    
23
#include "dsputil_altivec.h"
24
#include "util_altivec.h"
25

    
26
static void vector_fmul_altivec(float *dst, const float *src, int len)
27
{
28
    int i;
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);
32
        s = vec_ld(0, src+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);
36
        vec_st(d0, 0, dst+i);
37
        vec_st(d1, 16, dst+i);
38
    }
39
}
40

    
41
static void vector_fmul_reverse_altivec(float *dst, const float *src0,
42
                                        const float *src1, int len)
43
{
44
    int i;
45
    vector float d, s0, s1, h0, l0,
46
                 s2, s3, zero = (vector float)vec_splat_u32(0);
47
    src1 += len-4;
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]
57
                                             // [d,c,b,a]
58
        l0 = vec_mergel(s3, s3);
59
        d = vec_madd(s0, s1, zero);
60
        h0 = vec_mergeh(s3, s3);
61
        vec_st(d, 0, dst+i);
62
        s3 = vec_mergeh(vec_mergel(l0,h0),
63
                        vec_mergeh(l0,h0));
64
        d = vec_madd(s2, s3, zero);
65
        vec_st(d, 16, dst+i);
66
    }
67
}
68

    
69
static void vector_fmul_add_add_altivec(float *dst, const float *src0,
70
                                        const float *src1, const float *src2,
71
                                        int src3, int len, int step)
72
{
73
    int i;
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);
77

    
78
#if 0 //FIXME: there is still something wrong
79
    if (step == 2) {
80
        int y;
81
        vector float d0, d1, s3, t2;
82
        vector unsigned int sel =
83
                vec_mergeh(vec_splat_u32(-1), vec_splat_u32(0));
84
        t1 = vec_ld(16, dst);
85
        for (i=0,y=0; i<len-3; i+=4,y+=8) {
86

87
            s0 = vec_ld(0,src0+i);
88
            s1 = vec_ld(0,src1+i);
89
            s2 = vec_ld(0,src2+i);
90

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]
94

95
            d = vec_madd(s0,s1,s2); // [A B C D]
96

97
                                                 // [A A B B]
98

99
                                                 // [C C D D]
100

101
            d0 = vec_perm(t0, t1, mask); // [a b c d]
102

103
            d0 = vec_sel(vec_mergeh(d, d), d0, sel);   // [A b B d]
104

105
            edges = vec_perm(t1, t0, mask);
106

107
            t0 = vec_perm(edges, d0, align); // [x x x|A]
108

109
            t1 = vec_perm(d0, edges, align); // [b B d|e]
110

111
            vec_stl(t0, 0, dst+y);
112

113
            d1 = vec_perm(t1, t2, mask); // [e f g h]
114

115
            d1 = vec_sel(vec_mergel(d, d), d1, sel); // [C f D h]
116

117
            edges = vec_perm(t2, t1, mask);
118

119
            t1 = vec_perm(edges, d1, align); // [b B d|C]
120

121
            t2 = vec_perm(d1, edges, align); // [f D h|x]
122

123
            vec_stl(t1, 16, dst+y);
124

125
            t0 = t1;
126

127
            vec_stl(t2, 31, dst+y);
128

129
            t1 = t2;
130
        }
131
    } else
132
    #endif
133
    if (step == 1 && src3 == 0)
134
        for (i=0; i<len-3; i+=4) {
135
            t0 = vec_ld(0, dst+i);
136
            t1 = vec_ld(15, dst+i);
137
            s0 = vec_ld(0, src0+i);
138
            s1 = vec_ld(0, src1+i);
139
            s2 = vec_ld(0, src2+i);
140
            edges = vec_perm(t1 ,t0, mask);
141
            d = vec_madd(s0,s1,s2);
142
            t1 = vec_perm(d, edges, align);
143
            t0 = vec_perm(edges, d, align);
144
            vec_st(t1, 15, dst+i);
145
            vec_st(t0, 0, dst+i);
146
        }
147
    else
148
        ff_vector_fmul_add_add_c(dst, src0, src1, src2, src3, len, step);
149
}
150

151
static void vector_fmul_window_altivec(float *dst, const float *src0, const float *src1, const float *win, float add_bias, int len)
152
{
153
    union {
154
        vector float v;
155
        float s[4];
156
    } vadd;
157
    vector float vadd_bias, zero, t0, t1, s0, s1, wi, wj;
158
    const vector unsigned char reverse = vcprm(3,2,1,0);
159
    int i,j;
160

161
    dst += len;
162
    win += len;
163
    src0+= len;
164

165
    vadd.s[0] = add_bias;
166
    vadd_bias = vec_splat(vadd.v, 0);
167
    zero = (vector float)vec_splat_u32(0);
168

169
    for(i=-len*4, j=len*4-16; i<0; i+=16, j-=16) {
170
        s0 = vec_ld(i, src0);
171
        s1 = vec_ld(j, src1);
172
        wi = vec_ld(i, win);
173
        wj = vec_ld(j, win);
174

175
        s1 = vec_perm(s1, s1, reverse);
176
        wj = vec_perm(wj, wj, reverse);
177

178
        t0 = vec_madd(s0, wj, vadd_bias);
179
        t0 = vec_nmsub(s1, wi, t0);
180
        t1 = vec_madd(s0, wi, vadd_bias);
181
        t1 = vec_madd(s1, wj, t1);
182
        t1 = vec_perm(t1, t1, reverse);
183

184
        vec_st(t0, i, dst);
185
        vec_st(t1, j, dst);
186
    }
187
}
188

189
static void int32_to_float_fmul_scalar_altivec(float *dst, const int *src, float mul, int len)
190
{
191
    union {
192
        vector float v;
193
        float s[4];
194
    } mul_u;
195
    int i;
196
    vector float src1, src2, dst1, dst2, mul_v, zero;
197

198
    zero = (vector float)vec_splat_u32(0);
199
    mul_u.s[0] = mul;
200
    mul_v = vec_splat(mul_u.v, 0);
201

202
    for(i=0; i<len; i+=8) {
203
        src1 = vec_ctf(vec_ld(0,  src+i), 0);
204
        src2 = vec_ctf(vec_ld(16, src+i), 0);
205
        dst1 = vec_madd(src1, mul_v, zero);
206
        dst2 = vec_madd(src2, mul_v, zero);
207
        vec_st(dst1,  0, dst+i);
208
        vec_st(dst2, 16, dst+i);
209
    }
210
}
211

212

213
static vector signed short
214
float_to_int16_one_altivec(const float *src)
215
{
216
    vector float s0 = vec_ld(0, src);
217
    vector float s1 = vec_ld(16, src);
218
    vector signed int t0 = vec_cts(s0, 0);
219
    vector signed int t1 = vec_cts(s1, 0);
220
    return vec_packs(t0,t1);
221
}
222

223
static void float_to_int16_altivec(int16_t *dst, const float *src, long len)
224
{
225
    int i;
226
    vector signed short d0, d1, d;
227
    vector unsigned char align;
228
    if(((long)dst)&15) //FIXME
229
    for(i=0; i<len-7; i+=8) {
230
        d0 = vec_ld(0, dst+i);
231
        d = float_to_int16_one_altivec(src+i);
232
        d1 = vec_ld(15, dst+i);
233
        d1 = vec_perm(d1, d0, vec_lvsl(0,dst+i));
234
        align = vec_lvsr(0, dst+i);
235
        d0 = vec_perm(d1, d, align);
236
        d1 = vec_perm(d, d1, align);
237
        vec_st(d0, 0, dst+i);
238
        vec_st(d1,15, dst+i);
239
    }
240
    else
241
    for(i=0; i<len-7; i+=8) {
242
        d = float_to_int16_one_altivec(src+i);
243
        vec_st(d, 0, dst+i);
244
    }
245
}
246

247
static void
248
float_to_int16_interleave_altivec(int16_t *dst, const float **src,
249
                                  long len, int channels)
250
{
251
    int i;
252
    vector signed short d0, d1, d2, c0, c1, t0, t1;
253
    vector unsigned char align;
254
    if(channels == 1)
255
        float_to_int16_altivec(dst, src[0], len);
256
    else
257
        if (channels == 2) {
258
        if(((long)dst)&15)
259
        for(i=0; i<len-7; i+=8) {
260
            d0 = vec_ld(0, dst + i);
261
            t0 = float_to_int16_one_altivec(src[0] + i);
262
            d1 = vec_ld(31, dst + i);
263
            t1 = float_to_int16_one_altivec(src[1] + i);
264
            c0 = vec_mergeh(t0, t1);
265
            c1 = vec_mergel(t0, t1);
266
            d2 = vec_perm(d1, d0, vec_lvsl(0, dst + i));
267
            align = vec_lvsr(0, dst + i);
268
            d0 = vec_perm(d2, c0, align);
269
            d1 = vec_perm(c0, c1, align);
270
            vec_st(d0,  0, dst + i);
271
            d0 = vec_perm(c1, d2, align);
272
            vec_st(d1, 15, dst + i);
273
            vec_st(d0, 31, dst + i);
274
            dst+=8;
275
        }
276
        else
277
        for(i=0; i<len-7; i+=8) {
278
            t0 = float_to_int16_one_altivec(src[0] + i);
279
            t1 = float_to_int16_one_altivec(src[1] + i);
280
            d0 = vec_mergeh(t0, t1);
281
            d1 = vec_mergel(t0, t1);
282
            vec_st(d0,  0, dst + i);
283
            vec_st(d1, 16, dst + i);
284
            dst+=8;
285
        }
286
    } else {
287
        DECLARE_ALIGNED(16, int16_t, tmp[len]);
288
        int c, j;
289
        for (c = 0; c < channels; c++) {
290
            float_to_int16_altivec(tmp, src[c], len);
291
            for (i = 0, j = c; i < len; i++, j+=channels) {
292
                dst[j] = tmp[i];
293
            }
294
        }
295
   }
296
}
297

298
void float_init_altivec(DSPContext* c, AVCodecContext *avctx)
299
{
300
    c->vector_fmul = vector_fmul_altivec;
301
    c->vector_fmul_reverse = vector_fmul_reverse_altivec;
302
    c->vector_fmul_add_add = vector_fmul_add_add_altivec;
303
    c->int32_to_float_fmul_scalar = int32_to_float_fmul_scalar_altivec;
304
    if(!(avctx->flags & CODEC_FLAG_BITEXACT)) {
305
        c->vector_fmul_window = vector_fmul_window_altivec;
306
        c->float_to_int16 = float_to_int16_altivec;
307
        c->float_to_int16_interleave = float_to_int16_interleave_altivec;
308
    }
309
}
310