10a7169d1b4933943545cf18c16e134bb98c292b
[deb_ffmpeg.git] / fmtconvert_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 "config.h"
22 #include "libavutil/attributes.h"
23 #include "libavutil/cpu.h"
24 #include "libavutil/mem.h"
25 #include "libavutil/ppc/cpu.h"
26 #include "libavutil/ppc/util_altivec.h"
27 #include "libavcodec/fmtconvert.h"
28
29 #if HAVE_ALTIVEC
30
31 static void int32_to_float_fmul_scalar_altivec(float *dst, const int32_t *src,
32 float mul, int len)
33 {
34 union {
35 vector float v;
36 float s[4];
37 } mul_u;
38 int i;
39 vector float src1, src2, dst1, dst2, mul_v, zero;
40
41 zero = (vector float)vec_splat_u32(0);
42 mul_u.s[0] = mul;
43 mul_v = vec_splat(mul_u.v, 0);
44
45 for (i = 0; i < len; i += 8) {
46 src1 = vec_ctf(vec_ld(0, src+i), 0);
47 src2 = vec_ctf(vec_ld(16, src+i), 0);
48 dst1 = vec_madd(src1, mul_v, zero);
49 dst2 = vec_madd(src2, mul_v, zero);
50 vec_st(dst1, 0, dst+i);
51 vec_st(dst2, 16, dst+i);
52 }
53 }
54
55
56 static vector signed short float_to_int16_one_altivec(const float *src)
57 {
58 vector float s0 = vec_ld(0, src);
59 vector float s1 = vec_ld(16, src);
60 vector signed int t0 = vec_cts(s0, 0);
61 vector signed int t1 = vec_cts(s1, 0);
62 return vec_packs(t0,t1);
63 }
64
65 static void float_to_int16_altivec(int16_t *dst, const float *src, long len)
66 {
67 int i;
68 vector signed short d0, d1, d;
69 vector unsigned char align;
70 if (((long)dst) & 15) { //FIXME
71 for (i = 0; i < len - 7; i += 8) {
72 d0 = vec_ld(0, dst+i);
73 d = float_to_int16_one_altivec(src + i);
74 d1 = vec_ld(15, dst+i);
75 d1 = vec_perm(d1, d0, vec_lvsl(0, dst + i));
76 align = vec_lvsr(0, dst + i);
77 d0 = vec_perm(d1, d, align);
78 d1 = vec_perm(d, d1, align);
79 vec_st(d0, 0, dst + i);
80 vec_st(d1, 15, dst + i);
81 }
82 } else {
83 for (i = 0; i < len - 7; i += 8) {
84 d = float_to_int16_one_altivec(src + i);
85 vec_st(d, 0, dst + i);
86 }
87 }
88 }
89
90 #define VSTE_INC(dst, v, elem, inc) do { \
91 vector signed short s = vec_splat(v, elem); \
92 vec_ste(s, 0, dst); \
93 dst += inc; \
94 } while (0)
95
96 static void float_to_int16_stride_altivec(int16_t *dst, const float *src,
97 long len, int stride)
98 {
99 int i;
100 vector signed short d;
101
102 for (i = 0; i < len - 7; i += 8) {
103 d = float_to_int16_one_altivec(src + i);
104 VSTE_INC(dst, d, 0, stride);
105 VSTE_INC(dst, d, 1, stride);
106 VSTE_INC(dst, d, 2, stride);
107 VSTE_INC(dst, d, 3, stride);
108 VSTE_INC(dst, d, 4, stride);
109 VSTE_INC(dst, d, 5, stride);
110 VSTE_INC(dst, d, 6, stride);
111 VSTE_INC(dst, d, 7, stride);
112 }
113 }
114
115 static void float_to_int16_interleave_altivec(int16_t *dst, const float **src,
116 long len, int channels)
117 {
118 int i;
119 vector signed short d0, d1, d2, c0, c1, t0, t1;
120 vector unsigned char align;
121
122 if (channels == 1)
123 float_to_int16_altivec(dst, src[0], len);
124 else {
125 if (channels == 2) {
126 if (((long)dst) & 15) {
127 for (i = 0; i < len - 7; i += 8) {
128 d0 = vec_ld(0, dst + i);
129 t0 = float_to_int16_one_altivec(src[0] + i);
130 d1 = vec_ld(31, dst + i);
131 t1 = float_to_int16_one_altivec(src[1] + i);
132 c0 = vec_mergeh(t0, t1);
133 c1 = vec_mergel(t0, t1);
134 d2 = vec_perm(d1, d0, vec_lvsl(0, dst + i));
135 align = vec_lvsr(0, dst + i);
136 d0 = vec_perm(d2, c0, align);
137 d1 = vec_perm(c0, c1, align);
138 vec_st(d0, 0, dst + i);
139 d0 = vec_perm(c1, d2, align);
140 vec_st(d1, 15, dst + i);
141 vec_st(d0, 31, dst + i);
142 dst += 8;
143 }
144 } else {
145 for (i = 0; i < len - 7; i += 8) {
146 t0 = float_to_int16_one_altivec(src[0] + i);
147 t1 = float_to_int16_one_altivec(src[1] + i);
148 d0 = vec_mergeh(t0, t1);
149 d1 = vec_mergel(t0, t1);
150 vec_st(d0, 0, dst + i);
151 vec_st(d1, 16, dst + i);
152 dst += 8;
153 }
154 }
155 } else {
156 for (i = 0; i < channels; i++)
157 float_to_int16_stride_altivec(dst + i, src[i], len, channels);
158 }
159 }
160 }
161
162 #endif /* HAVE_ALTIVEC */
163
164 av_cold void ff_fmt_convert_init_ppc(FmtConvertContext *c,
165 AVCodecContext *avctx)
166 {
167 #if HAVE_ALTIVEC
168 if (!PPC_ALTIVEC(av_get_cpu_flags()))
169 return;
170
171 c->int32_to_float_fmul_scalar = int32_to_float_fmul_scalar_altivec;
172 if (!(avctx->flags & CODEC_FLAG_BITEXACT)) {
173 c->float_to_int16 = float_to_int16_altivec;
174 c->float_to_int16_interleave = float_to_int16_interleave_altivec;
175 }
176 #endif /* HAVE_ALTIVEC */
177 }