[FFmpeg-devel] [PATCH 02/15] libavutil: Implementation of AAC_fixed_decoder (LC-module) [2/5]

[Nedeljko Babic]

>On Fri, Nov 07, 2014 at 04:08:06PM +0100, Nedeljko Babic wrote:
>> From: Djordje Pesut <djordje.pesut at imgtec.com>
>> 
>> Make changes in softfloat.
>> 
>> Functions for sqrt and sincos are added.
>> 
>> Div function is improved.
>> 
>> Some changes are made in order for code in softfloat to be usable in fixed aac
>> decoder code.
>> For example order of fields in structure SoftFloat is changed.
>> This doesn't create any impact on current ffmpeg code since softfloat is
>> currently not in use and this way we don't need to make much changes in
>> implementation of aac fixed point decoder that uses this code.
>> 
>> Signed-off-by: Nedeljko Babic <nedeljko.babic at imgtec.com>
>> ---
>>  libavutil/softfloat.h        | 167 ++++++++++++++++++++----
>>  libavutil/softfloat_tables.h | 294 +++++++++++++++++++++++++++++++++++++++++++
>>  2 files changed, 439 insertions(+), 22 deletions(-)
>>  create mode 100644 libavutil/softfloat_tables.h
>> 
>> diff --git a/libavutil/softfloat.h b/libavutil/softfloat.h
>> index 8647e6a..df91a8b 100644
>> --- a/libavutil/softfloat.h
>> +++ b/libavutil/softfloat.h
>> @@ -23,18 +23,26 @@
>>  
>>  #include <stdint.h>
>>  #include "common.h"
>> -
>>  #include "avassert.h"
>> +#include "softfloat_tables.h"
>>  
>>  #define MIN_EXP -126
>>  #define MAX_EXP  126
>>  #define ONE_BITS 29
>>  
>>  typedef struct SoftFloat{
>> -    int32_t  exp;
>>      int32_t mant;
>> +    int32_t  exp;
>>  }SoftFloat;
>>  
>
>> +static const SoftFloat FLOAT_0          = {         0,   0};
>> +static const SoftFloat FLOAT_05         = { 536870912,   0};
>> +static const SoftFloat FLOAT_1          = { 536870912,   1};
>> +static const SoftFloat FLOAT_EPSILON    = { 703687442, -16};
>> +static const SoftFloat FLOAT_1584893192 = { 850883053,   1};
>> +static const SoftFloat FLOAT_100000     = { 819200000,  17};
>> +static const SoftFloat FLOAT_0999999    = {1073740750,   0};
>
>the mantises should be in some form of base 2 or base 16 in the
>source as this should be more readable
>

I didn't made a change here because for some of the values base 16
representation is not more readable than base 10 (for example 703687442 and
0x29F16B12).
On the other hand I see that it is more readable for most of the values
and I will make a change in the next patch.

>
>> +
>>  static av_const SoftFloat av_normalize_sf(SoftFloat a){
>>      if(a.mant){
>>  #if 1
>> @@ -66,11 +74,12 @@ static inline av_const SoftFloat av_normalize1_sf(SoftFloat a){
>>      av_assert2(a.mant < 0x40000000 && a.mant > -0x40000000);
>>      return a;
>>  #elif 1
>> -    int t= a.mant + 0x40000000 < 0;
>> -    return (SoftFloat){a.exp+t, a.mant>>t};
>> +    int t= a.mant + 0x40000000;
>> +    t = t < 0;
>> +    return (SoftFloat){ a.mant>>t, a.exp+t};
>>  #else
>>      int t= (a.mant + 0x40000000U)>>31;
>
>> -    return (SoftFloat){a.exp+t, a.mant>>t};
>> +    return (SoftFloat){a.mant>>t, a.exp+t};
>
>fliping the order of the argumets, as its cosmetic should be in a
>seperate patch from all functional changes
>this keeps the changes easier to read now for review as well as in
>the future

Ok. It will be moved to the new patch.
Should I resend entire patch set once all the changes are made, or should I just
split this patch in the two new patches and send them in answer to this review?