[FFmpeg-devel] [PATCH 1/2] Remove the add bias hack for the C version of DSPContext.float_to_int16_*().

[Måns Rullgård]

"Ronald S. Bultje" <rsbultje at gmail.com> writes:

> Hi,
>
> On Wed, Jan 26, 2011 at 1:46 PM, Justin Ruggles
> <justin.ruggles at gmail.com> wrote:
>> ?libavcodec/aac.h ? ? ? ? ? | ? ?1 -
>> ?libavcodec/aacdec.c ? ? ? ?| ? 42 +++++++++++++++------------------------
>> ?libavcodec/aacsbr.c ? ? ? ?| ? 13 ++---------
>> ?libavcodec/ac3dec.c ? ? ? ?| ? 13 +----------
>> ?libavcodec/ac3dec.h ? ? ? ?| ? ?1 -
>> ?libavcodec/binkaudio.c ? ? | ? ?5 ----
>> ?libavcodec/dca.c ? ? ? ? ? | ? 47 +++++++++++++++++--------------------------
>> ?libavcodec/dsputil.c ? ? ? | ? ?9 +-------
>> ?libavcodec/dsputil.h ? ? ? | ? ?3 +-
>> ?libavcodec/nellymoserdec.c | ? ?9 +-------
>> ?libavcodec/wmadec.c ? ? ? ?| ? 17 +---------------
>> ?11 files changed, 44 insertions(+), 116 deletions(-)
> [..]
>> @@ -637,13 +628,13 @@ static inline void do_imdct(AC3DecodeContext *s, int channels)
>>              for(i=0; i<128; i++)
>>                  x[i] = s->transform_coeffs[ch][2*i];
>>              ff_imdct_half(&s->imdct_256, s->tmp_output, x);
>> -            s->dsp.vector_fmul_window(s->output[ch-1], s->delay[ch-1], s->tmp_output, s->window, add_bias, 128);
>> +            s->dsp.vector_fmul_window(s->output[ch-1], s->delay[ch-1], s->tmp_output, s->window, 0, 128);
>>              for(i=0; i<128; i++)
>>                  x[i] = s->transform_coeffs[ch][2*i+1];
>>              ff_imdct_half(&s->imdct_256, s->delay[ch-1], x);
>>          } else {
>>              ff_imdct_half(&s->imdct_512, s->tmp_output, s->transform_coeffs[ch]);
>> -            s->dsp.vector_fmul_window(s->output[ch-1], s->delay[ch-1], s->tmp_output, s->window, add_bias, 128);
>> +            s->dsp.vector_fmul_window(s->output[ch-1], s->delay[ch-1], s->tmp_output, s->window, 0, 128);
>>              memcpy(s->delay[ch-1], s->tmp_output+128, 128*sizeof(float));
>>          }
>>      }
>
> I see you removed this add_bias in various other places, so where you
> didn't: we can probably simplify such functions now also and remove
> the add_bias argument completely? Or is this used for proper purposes
> elsewhere?

The argument has to stay as a placeholder until we've updated all the asm.
It's easier to do this in two steps.

-- 
M?ns Rullg?rd
mans at mansr.com