FFmpeg
mpegvideoencdsp.c
Go to the documentation of this file.
1 /*
2  * This file is part of FFmpeg.
3  *
4  * FFmpeg 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 2.1 of the License, or (at your option) any later version.
8  *
9  * FFmpeg 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.
13  *
14  * You should have received a copy of the GNU Lesser General Public
15  * License along with FFmpeg; if not, write to the Free Software
16  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17  */
18 
19 #include "config.h"
20 
21 #include <stdint.h>
22 
23 #include "libavutil/attributes.h"
24 #include "libavutil/cpu.h"
25 #include "libavutil/ppc/cpu.h"
27 
29 
30 #if HAVE_ALTIVEC
31 
32 #if HAVE_VSX
33 static int pix_norm1_altivec(const uint8_t *pix, int line_size)
34 {
35  int i, s = 0;
36  const vector unsigned int zero =
37  (const vector unsigned int) vec_splat_u32(0);
38  vector unsigned int sv = (vector unsigned int) vec_splat_u32(0);
39  vector signed int sum;
40 
41  for (i = 0; i < 16; i++) {
42  /* Read the potentially unaligned pixels. */
43  //vector unsigned char pixl = vec_ld(0, pix);
44  //vector unsigned char pixr = vec_ld(15, pix);
45  //vector unsigned char pixv = vec_perm(pixl, pixr, perm);
46  vector unsigned char pixv = vec_vsx_ld(0, pix);
47 
48  /* Square the values, and add them to our sum. */
49  sv = vec_msum(pixv, pixv, sv);
50 
51  pix += line_size;
52  }
53  /* Sum up the four partial sums, and put the result into s. */
54  sum = vec_sums((vector signed int) sv, (vector signed int) zero);
55  sum = vec_splat(sum, 3);
56  vec_ste(sum, 0, &s);
57  return s;
58 }
59 #else
60 static int pix_norm1_altivec(const uint8_t *pix, int line_size)
61 {
62  int i, s = 0;
63  const vector unsigned int zero =
64  (const vector unsigned int) vec_splat_u32(0);
65  vector unsigned char perm = vec_lvsl(0, pix);
66  vector unsigned int sv = (vector unsigned int) vec_splat_u32(0);
67  vector signed int sum;
68 
69  for (i = 0; i < 16; i++) {
70  /* Read the potentially unaligned pixels. */
71  vector unsigned char pixl = vec_ld(0, pix);
72  vector unsigned char pixr = vec_ld(15, pix);
73  vector unsigned char pixv = vec_perm(pixl, pixr, perm);
74 
75  /* Square the values, and add them to our sum. */
76  sv = vec_msum(pixv, pixv, sv);
77 
78  pix += line_size;
79  }
80  /* Sum up the four partial sums, and put the result into s. */
81  sum = vec_sums((vector signed int) sv, (vector signed int) zero);
82  sum = vec_splat(sum, 3);
83  vec_ste(sum, 0, &s);
84 
85  return s;
86 }
87 #endif /* HAVE_VSX */
88 
89 #if HAVE_VSX
90 static int pix_sum_altivec(const uint8_t *pix, int line_size)
91 {
92  int i, s;
93  const vector unsigned int zero =
94  (const vector unsigned int) vec_splat_u32(0);
95  vector unsigned int sad = (vector unsigned int) vec_splat_u32(0);
96  vector signed int sumdiffs;
97 
98  for (i = 0; i < 16; i++) {
99  /* Read the potentially unaligned 16 pixels into t1. */
100  //vector unsigned char pixl = vec_ld(0, pix);
101  //vector unsigned char pixr = vec_ld(15, pix);
102  //vector unsigned char t1 = vec_perm(pixl, pixr, perm);
103  vector unsigned char t1 = vec_vsx_ld(0, pix);
104 
105  /* Add each 4 pixel group together and put 4 results into sad. */
106  sad = vec_sum4s(t1, sad);
107 
108  pix += line_size;
109  }
110 
111  /* Sum up the four partial sums, and put the result into s. */
112  sumdiffs = vec_sums((vector signed int) sad, (vector signed int) zero);
113  sumdiffs = vec_splat(sumdiffs, 3);
114  vec_ste(sumdiffs, 0, &s);
115  return s;
116 }
117 #else
118 static int pix_sum_altivec(const uint8_t *pix, int line_size)
119 {
120  int i, s;
121  const vector unsigned int zero =
122  (const vector unsigned int) vec_splat_u32(0);
123  vector unsigned char perm = vec_lvsl(0, pix);
124  vector unsigned int sad = (vector unsigned int) vec_splat_u32(0);
125  vector signed int sumdiffs;
126 
127  for (i = 0; i < 16; i++) {
128  /* Read the potentially unaligned 16 pixels into t1. */
129  vector unsigned char pixl = vec_ld(0, pix);
130  vector unsigned char pixr = vec_ld(15, pix);
131  vector unsigned char t1 = vec_perm(pixl, pixr, perm);
132 
133  /* Add each 4 pixel group together and put 4 results into sad. */
134  sad = vec_sum4s(t1, sad);
135 
136  pix += line_size;
137  }
138 
139  /* Sum up the four partial sums, and put the result into s. */
140  sumdiffs = vec_sums((vector signed int) sad, (vector signed int) zero);
141  sumdiffs = vec_splat(sumdiffs, 3);
142  vec_ste(sumdiffs, 0, &s);
143 
144  return s;
145 }
146 
147 #endif /* HAVE_VSX */
148 
149 #endif /* HAVE_ALTIVEC */
150 
152  AVCodecContext *avctx)
153 {
154 #if HAVE_ALTIVEC
156  return;
157 
158  c->pix_norm1 = pix_norm1_altivec;
159  c->pix_sum = pix_sum_altivec;
160 #endif /* HAVE_ALTIVEC */
161 }
t1
#define t1
Definition: regdef.h:29
ff_mpegvideoencdsp_init_ppc
av_cold void ff_mpegvideoencdsp_init_ppc(MpegvideoEncDSPContext *c, AVCodecContext *avctx)
Definition: mpegvideoencdsp.c:151
av_get_cpu_flags
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition: cpu.c:103
perm
perm
Definition: f_perms.c:75
av_cold
#define av_cold
Definition: attributes.h:90
s
#define s(width, name)
Definition: cbs_vp9.c:198
c
Undefined Behavior In the C some operations are like signed integer dereferencing freed accessing outside allocated Undefined Behavior must not occur in a C it is not safe even if the output of undefined operations is unused The unsafety may seem nit picking but Optimizing compilers have in fact optimized code on the assumption that no undefined Behavior occurs Optimizing code based on wrong assumptions can and has in some cases lead to effects beyond the output of computations The signed integer overflow problem in speed critical code Code which is highly optimized and works with signed integers sometimes has the problem that often the output of the computation does not c
Definition: undefined.txt:32
PPC_ALTIVEC
#define PPC_ALTIVEC(flags)
Definition: cpu.h:25
cpu.h
attributes.h
MpegvideoEncDSPContext
Definition: mpegvideoencdsp.h:32
i
#define i(width, name, range_min, range_max)
Definition: cbs_h2645.c:255
AVCodecContext
main external API structure.
Definition: avcodec.h:445
mpegvideoencdsp.h
zero
#define zero
Definition: regdef.h:64
util_altivec.h
cpu.h
int
int
Definition: ffmpeg_filter.c:409