[FFmpeg-devel] [PATCH 3/3] lavc/wmadec: replace pow by faster functions

Paul B Mahol onemda at gmail.com
Fri Jan 22 12:15:26 CET 2016


On 1/22/16, Ganesh Ajjanagadde <gajjanagadde at gmail.com> wrote:
> On Sat, Jan 16, 2016 at 12:31 AM, Ganesh Ajjanagadde
> <gajjanagadde at gmail.com> wrote:
>> Further speedups possible by getting rid of exp2f...
>>
>> Signed-off-by: Ganesh Ajjanagadde <gajjanagadde at gmail.com>
>> ---
>>  libavcodec/wmadec.c | 5 +++--
>>  1 file changed, 3 insertions(+), 2 deletions(-)
>>
>> diff --git a/libavcodec/wmadec.c b/libavcodec/wmadec.c
>> index da54182..1a84323 100644
>> --- a/libavcodec/wmadec.c
>> +++ b/libavcodec/wmadec.c
>> @@ -35,6 +35,7 @@
>>
>>  #include "libavutil/attributes.h"
>>  #include "libavutil/internal.h"
>> +#include "libavutil/libm.h"
>>
>>  #include "avcodec.h"
>>  #include "internal.h"
>> @@ -164,7 +165,7 @@ static av_cold void
>> wma_lsp_to_curve_init(WMACodecContext *s, int frame_len)
>>      /* tables for x^-0.25 computation */
>>      for (i = 0; i < 256; i++) {
>>          e                     = i - 126;
>> -        s->lsp_pow_e_table[i] = pow(2.0, e * -0.25);
>> +        s->lsp_pow_e_table[i] = exp2f(e * -0.25);
>>      }
>>
>>      /* NOTE: these two tables are needed to avoid two operations in
>> @@ -173,7 +174,7 @@ static av_cold void
>> wma_lsp_to_curve_init(WMACodecContext *s, int frame_len)
>>      for (i = (1 << LSP_POW_BITS) - 1; i >= 0; i--) {
>>          m                      = (1 << LSP_POW_BITS) + i;
>>          a                      = (float) m * (0.5 / (1 << LSP_POW_BITS));
>> -        a                      = pow(a, -0.25);
>> +        a                      = 1/sqrt(sqrt(a));
>>          s->lsp_pow_m_table1[i] = 2 * a - b;
>>          s->lsp_pow_m_table2[i] = b - a;
>>          b                      = a;
>> --
>> 2.7.0
>>
>
> ping, note that second can be made sqrtf(sqrtf(x)) at some error which
> I deliberately avoided here.
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>

ok, if fate passes.


More information about the ffmpeg-devel mailing list