RepoMirrors
/
ffmpeg
mirror of https://git.ffmpeg.org/ffmpeg.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

avcodec/mpegaudio_tablegen: speed up dynamic table creation 

This does some miscellaneous stuff mainly avoiding the usage of pow to
achieve significant speedups. This is not speed critical, but is
unnecessary latency and cycles wasted for a user.

All tables tested and are identical to the old ones
(bit-exact even in floating point case).

Sample benchmark (x86-64, Haswell, GNU/Linux):
old:
102329530 decicycles in mpegaudio_tableinit,       1 runs,      0 skips

new:
34111900 decicycles in mpegaudio_tableinit,       1 runs,      0 skips

Reviewed-by: Ronald S. Bultje <rsbultje@gmail.com>
Signed-off-by: Ganesh Ajjanagadde <gajjanagadde@gmail.com>
This commit is contained in:
Ganesh Ajjanagadde 2015-11-25 16:26:08 -05:00
parent 538e8ab512
commit 16d4c28cbb
2 changed files with 14 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
libavcodec/mpegaudio_tablegen.c
View File

@ -23,6 +23,7 @@
#include <stdlib.h> #include <stdlib.h>
#define CONFIG_HARDCODED_TABLES 0 #define CONFIG_HARDCODED_TABLES 0
#include "mpegaudio_tablegen.h" #include "mpegaudio_tablegen.h"
#include "libavutil/tablegen.h"
#include "tableprint.h" #include "tableprint.h"
int main(void) int main(void)

19
libavcodec/mpegaudio_tablegen.h
View File

@ -45,12 +45,21 @@ static float expval_table_float[512][16];
static av_cold void mpegaudio_tableinit(void) static av_cold void mpegaudio_tableinit(void)
{ {
int i, value, exponent; int i, value, exponent;
double exp2_lut[4] = {
1.00000000000000000000, /* 2 ^ (0 * 0.25) */
1.18920711500272106672, /* 2 ^ (1 * 0.25) */
M_SQRT2 , /* 2 ^ (2 * 0.25) */
1.68179283050742908606, /* 2 ^ (3 * 0.25) */
};
double cbrt_lut[16];
for (i = 0; i < 16; ++i)
cbrt_lut[i] = cbrt(i);
for (i = 1; i < TABLE_4_3_SIZE; i++) { for (i = 1; i < TABLE_4_3_SIZE; i++) {
double value = i / 4; double value = i / 4;
double f, fm; double f, fm;
int e, m; int e, m;
/* cbrtf() isn't available on all systems, so we use powf(). */ f = value / IMDCT_SCALAR * cbrt(value) * exp2_lut[i & 3];
f = value / IMDCT_SCALAR * pow(value, 1.0 / 3.0) * pow(2, (i & 3) * 0.25);
fm = frexp(f, &e); fm = frexp(f, &e);
m = (uint32_t)(fm * (1LL << 31) + 0.5); m = (uint32_t)(fm * (1LL << 31) + 0.5);
e += FRAC_BITS - 31 + 5 - 100; e += FRAC_BITS - 31 + 5 - 100;
@ -61,10 +70,8 @@ static av_cold void mpegaudio_tableinit(void)
} }
for (exponent = 0; exponent < 512; exponent++) { for (exponent = 0; exponent < 512; exponent++) {
for (value = 0; value < 16; value++) { for (value = 0; value < 16; value++) {
/* cbrtf() isn't available on all systems, so we use powf(). */ double f = value * cbrt_lut[value] * pow(2, (exponent - 400) * 0.25 + FRAC_BITS + 5) / IMDCT_SCALAR;
double f = (double)value * pow(value, 1.0 / 3.0) * pow(2, (exponent - 400) * 0.25 + FRAC_BITS + 5) / IMDCT_SCALAR; expval_table_fixed[exponent][value] = (f < 0xFFFFFFFF ? llrint(f) : 0xFFFFFFFF);
/* llrint() isn't always available, so round and cast manually. */
expval_table_fixed[exponent][value] = (long long int) (f < 0xFFFFFFFF ? floor(f + 0.5) : 0xFFFFFFFF);
expval_table_float[exponent][value] = f; expval_table_float[exponent][value] = f;
} }
exp_table_fixed[exponent] = expval_table_fixed[exponent][1]; exp_table_fixed[exponent] = expval_table_fixed[exponent][1];