FFmpeg
af_afir.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 modify
5  * it under the terms of the GNU General Public License as published by
6  * the Free Software Foundation; either version 2 of the License, or
7  * (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
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License along
15  * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
16  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
17  */
18 
19 #include "config.h"
20 
21 #include <float.h>
22 #include <stdint.h>
23 
24 #include "libavfilter/af_afirdsp.h"
25 #include "libavutil/internal.h"
26 #include "libavutil/mem_internal.h"
27 #include "checkasm.h"
28 
29 #define LEN 256
30 
31 static void test_fcmul_add(AudioFIRDSPContext *fir)
32 {
33 #define BUF_SIZE LEN*2+8
34  LOCAL_ALIGNED_32(float, src0, [BUF_SIZE]);
35  LOCAL_ALIGNED_32(float, src1, [BUF_SIZE]);
36  LOCAL_ALIGNED_32(float, src2, [BUF_SIZE]);
37 
38  randomize_stddev(src0, BUF_SIZE, 10.0);
39  randomize_stddev(src1, BUF_SIZE, 10.0);
40  randomize_stddev(src2, BUF_SIZE, 10.0);
41 
42  if (check_func(fir->fcmul_add, "fcmul_add")) {
43  LOCAL_ALIGNED_32(float, cdst, [BUF_SIZE]);
44  LOCAL_ALIGNED_32(float, odst, [BUF_SIZE]);
45  int i;
46 
47  declare_func(void, float *sum, const float *t, const float *c,
48  ptrdiff_t len);
49 
50  memcpy(cdst, src0, (BUF_SIZE) * sizeof(float));
51  memcpy(odst, src0, (BUF_SIZE) * sizeof(float));
52  call_ref(cdst, src1, src2, LEN);
53  call_new(odst, src1, src2, LEN);
54  for (i = 0; i <= LEN*2; i++) {
55  int idx = i & ~1;
56  float cre = src2[idx];
57  float cim = src2[idx + 1];
58  float tre = src1[idx];
59  float tim = src1[idx + 1];
60  double t = fabs(src0[i]) +
61  fabs(tre) + fabs(tim) + fabs(cre) + fabs(cim) +
62  fabs(tre * cre) + fabs(tim * cim) +
63  fabs(tre * cim) + fabs(tim * cre) +
64  fabs(tre * cre - tim * cim) +
65  fabs(tre * cim + tim * cre) +
66  fabs(cdst[i]) + 1.0;
67  if (!float_near_abs_eps(cdst[i], odst[i], t * 2 * FLT_EPSILON)) {
68  fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n",
69  i, cdst[i], odst[i], cdst[i] - odst[i]);
70  fail();
71  break;
72  }
73  }
74  memcpy(odst, src0, (BUF_SIZE) * sizeof(float));
75  bench_new(odst, src1, src2, LEN);
76  }
77 
78  report("fcmul_add");
79 }
80 
81 static void test_dcmul_add(AudioFIRDSPContext *fir)
82 {
83 #define BUF_SIZE LEN*2+8
84  LOCAL_ALIGNED_32(double, src0, [BUF_SIZE]);
85  LOCAL_ALIGNED_32(double, src1, [BUF_SIZE]);
86  LOCAL_ALIGNED_32(double, src2, [BUF_SIZE]);
87 
88  randomize_stddev_dbl(src0, BUF_SIZE, 10.0);
89  randomize_stddev_dbl(src1, BUF_SIZE, 10.0);
90  randomize_stddev_dbl(src2, BUF_SIZE, 10.0);
91 
92  if (check_func(fir->dcmul_add, "dcmul_add")) {
93  LOCAL_ALIGNED_32(double, cdst, [BUF_SIZE]);
94  LOCAL_ALIGNED_32(double, odst, [BUF_SIZE]);
95  int i;
96 
97  declare_func(void, double *sum, const double *t, const double *c,
98  ptrdiff_t len);
99 
100  memcpy(cdst, src0, (BUF_SIZE) * sizeof(double));
101  memcpy(odst, src0, (BUF_SIZE) * sizeof(double));
102  call_ref(cdst, src1, src2, LEN);
103  call_new(odst, src1, src2, LEN);
104  for (i = 0; i <= LEN*2; i++) {
105  int idx = i & ~1;
106  double cre = src2[idx];
107  double cim = src2[idx + 1];
108  double tre = src1[idx];
109  double tim = src1[idx + 1];
110  double t = fabs(src0[i]) +
111  fabs(tre) + fabs(tim) + fabs(cre) + fabs(cim) +
112  fabs(tre * cre) + fabs(tim * cim) +
113  fabs(tre * cim) + fabs(tim * cre) +
114  fabs(tre * cre - tim * cim) +
115  fabs(tre * cim + tim * cre) +
116  fabs(cdst[i]) + 1.0;
117  if (!double_near_abs_eps(cdst[i], odst[i], t * 2 * FLT_EPSILON)) {
118  fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n",
119  i, cdst[i], odst[i], cdst[i] - odst[i]);
120  fail();
121  break;
122  }
123  }
124  memcpy(odst, src0, (BUF_SIZE) * sizeof(double));
125  bench_new(odst, src1, src2, LEN);
126  }
127 
128  report("dcmul_add");
129 }
130 
131 
132 void checkasm_check_afir(void)
133 {
134  AudioFIRDSPContext fir = { 0 };
135 
136  ff_afir_init(&fir);
137  test_fcmul_add(&fir);
138  test_dcmul_add(&fir);
139 }
mem_internal.h
src1
const pixel * src1
Definition: h264pred_template.c:420
checkasm_check_afir
void checkasm_check_afir(void)
Definition: af_afir.c:132
AudioFIRDSPContext::fcmul_add
void(* fcmul_add)(float *sum, const float *t, const float *c, ptrdiff_t len)
Definition: af_afirdsp.h:30
check_func
#define check_func
Definition: test.h:480
float.h
bench_new
#define bench_new
Definition: test.h:486
test_dcmul_add
static void test_dcmul_add(AudioFIRDSPContext *fir)
Definition: af_afir.c:81
call_ref
#define call_ref
Definition: test.h:484
af_afirdsp.h
checkasm.h
ff_afir_init
static av_unused void ff_afir_init(AudioFIRDSPContext *dsp)
Definition: af_afirdsp.h:73
declare_func
#define declare_func
Definition: test.h:488
test_fcmul_add
static void test_fcmul_add(AudioFIRDSPContext *fir)
Definition: af_afir.c:31
fail
#define fail
Definition: test.h:478
float_near_abs_eps
#define float_near_abs_eps
Definition: utils.h:371
fabs
static __device__ float fabs(float a)
Definition: cuda_runtime.h:182
LOCAL_ALIGNED_32
#define LOCAL_ALIGNED_32(t, v,...)
Definition: mem_internal.h:132
randomize_stddev_dbl
#define randomize_stddev_dbl(buf, size, stddev)
Definition: checkasm.h:146
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
i
#define i(width, name, range_min, range_max)
Definition: cbs_h264.c:63
AudioFIRDSPContext
Definition: af_afirdsp.h:29
LEN
#define LEN
Definition: af_afir.c:29
internal.h
src2
const pixel * src2
Definition: h264pred_template.c:421
AudioFIRDSPContext::dcmul_add
void(* dcmul_add)(double *sum, const double *t, const double *c, ptrdiff_t len)
Definition: af_afirdsp.h:32
len
int len
Definition: vorbis_enc_data.h:426
call_new
#define call_new
Definition: test.h:485
double_near_abs_eps
#define double_near_abs_eps
Definition: utils.h:376
BUF_SIZE
#define BUF_SIZE
src0
const pixel *const src0
Definition: h264pred_template.c:419
report
#define report
Definition: test.h:479
randomize_stddev
#define randomize_stddev(buf, size, stddev)
Definition: checkasm.h:144
Generated on Tue Jul 7 2026 19:23:15 for FFmpeg by   doxygen 1.8.17