[FFmpeg-devel] [PATCH] avcodec/xbmenc: Avoid snprintf() for data->hex conversion

[Andreas Rheinhardt]

Marton Balint:
> 
> 
> On Fri, 8 Mar 2024, Andreas Rheinhardt wrote:
> 
>> Marton Balint:
>>>
>>>
>>> On Fri, 8 Mar 2024, Andreas Rheinhardt wrote:
>>>
>>>> Andreas Rheinhardt:
>>>>> Use a small LUT instead. Improves performance.
>>>>>
>>>>> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt at outlook.com>
>>>>> ---
>>>>>  libavcodec/xbmenc.c | 21 +++++++++++++++------
>>>>>  1 file changed, 15 insertions(+), 6 deletions(-)
>>>>>
>>>>> diff --git a/libavcodec/xbmenc.c b/libavcodec/xbmenc.c
>>>>> index cd8b73afa3..5231d4691d 100644
>>>>> --- a/libavcodec/xbmenc.c
>>>>> +++ b/libavcodec/xbmenc.c
>>>>> @@ -20,11 +20,9 @@
>>>>>   * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
>>>>> 02110-1301 USA
>>>>>   */
>>>>>
>>>>> -#include "libavutil/reverse.h"
>>>>>  #include "avcodec.h"
>>>>>  #include "codec_internal.h"
>>>>>  #include "encode.h"
>>>>> -#include "mathops.h"
>>>>>
>>>>>  #define ANSI_MIN_READLINE 509
>>>>>
>>>>> @@ -57,14 +55,25 @@ static int xbm_encode_frame(AVCodecContext
>>>>> *avctx, AVPacket *pkt,
>>>>>      buf += snprintf(buf, 39, "static unsigned char image_bits[] =
>>>>> {\n");
>>>>>      for (i = 0, l = lineout; i < avctx->height; i++) {
>>>>>          for (j = 0; j < linesize; j++) {
>>>>> -            buf += snprintf(buf, 6, " 0x%02X", ff_reverse[*ptr++]);
>>>>> +            // 0..15 bitreversed as chars
>>>>> +            static const char lut[] = {
>>>>> +                '0', '8', '4', 'C', '2', 'A', '6', 'E',
>>>>> +                '1', '9', '5', 'D', '3', 'B', '7', 'F'
>>>>> +            };
>>>>> +            buf[0] = ' ';
>>>>> +            buf[1] = '0';
>>>>> +            buf[2] = 'x';
>>>>> +            buf[3] = lut[*ptr & 0xF];
>>>>> +            buf[4] = lut[*ptr >> 4];
>>>
>>> Maybe you could use *buf++ = ... here as well, to avoid the next line.
>>> But fine either way I guess.
>>>
>>
>> You mean *ptr++ to avoid the line after the next line? That would make
>> the two lut accesses unsymmetric. And actually I prefer that both
>> pointers are incremented side-by-side.
>>
> 
> I meant this:
> 
> diff --git a/libavcodec/xbmenc.c b/libavcodec/xbmenc.c
> index 5231d4691d..20f8951f93 100644
> --- a/libavcodec/xbmenc.c
> +++ b/libavcodec/xbmenc.c
> @@ -60,12 +60,11 @@ static int xbm_encode_frame(AVCodecContext *avctx,
> AVPacket *pkt,
>                  '0', '8', '4', 'C', '2', 'A', '6', 'E',
>                  '1', '9', '5', 'D', '3', 'B', '7', 'F'
>              };
> -            buf[0] = ' ';
> -            buf[1] = '0';
> -            buf[2] = 'x';
> -            buf[3] = lut[*ptr & 0xF];
> -            buf[4] = lut[*ptr >> 4];
> -            buf += 5;
> +            *buf++ = ' ';
> +            *buf++ = '0';
> +            *buf++ = 'x';
> +            *buf++ = lut[*ptr & 0xF];
> +            *buf++ = lut[*ptr >> 4];
>              ptr++;
>              if (--commas <= 0) {
>                  *buf++ = '\n';
> 
> 

Ok, I misunderstood. But I still prefer my version.

- Andreas