exp2 suffices here. Some trivial speedup is done in addition here by
reusing results.
This retains accuracy, and in particular results in identical values
with GNU libm + gcc/clang.
sample benchmark (Haswell, GNU/Linux):
proposed : 424160 decicycles in pow_table, 512 runs, 0 skips
exp2 only:
1262093 decicycles in pow_table, 512 runs, 0 skips
old :
2849085 decicycles in pow_table, 512 runs, 0 skips
Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: Ganesh Ajjanagadde <gajjanagadde@gmail.com>
/* Generate overlap window */
ff_init_ff_sine_windows(7);
+ /* faster way of doing
for (i = 0; i < POW_TABLE_SIZE; i++)
- pow_table[i] = pow(2, -i / 2048.0 - 3.0 + POW_TABLE_OFFSET);
+ pow_table[i] = 2^(-i / 2048.0 - 3.0 + POW_TABLE_OFFSET); */
+ pow_table[0] = 1;
+ pow_table[1024] = M_SQRT1_2;
+ for (i = 1; i < 513; i++) {
+ double tmp = exp2(-i / 2048.0);
+ pow_table[i] = tmp;
+ pow_table[1024-i] = M_SQRT1_2 / tmp;
+ pow_table[1024+i] = tmp * M_SQRT1_2;
+ pow_table[2048-i] = 0.5 / tmp;
+ }
if (s->avctx->trellis) {
s->opt = av_malloc(NELLY_BANDS * OPT_SIZE * sizeof(float ));