[FFmpeg-trac] #2713(avutil:new): av_d2q regression on sparc64 and ia64

[FFmpeg]

#2713: av_d2q regression on sparc64 and ia64
-------------------------------------+-------------------------------------
             Reporter:  cehoyos      |                    Owner:
                 Type:  defect       |                   Status:  new
             Priority:  important    |                Component:  avutil
              Version:  git-master   |               Resolution:
             Keywords:  regression   |               Blocked By:
  ia64 sparc64                       |  Reproduced by developer:  0
             Blocking:               |
Analyzed by developer:  0            |
-------------------------------------+-------------------------------------
Description changed by cehoyos:

Old description:

> The parseutils fate test fails on sparc64 Linux and ia64 Linux since
> 887d74c. The problem is not limited to parseutils, many fate tests fail
> because of an incorrect timebase.
> The reason is that for some values of d and den - local variables in
> av_d2q() used when calling av_reduce() - {{{(uint64_t)rint(d * den)}}}
> and {{{llrint(d * den)}}} provide different results than {{{(int64_t)(d *
> den + 0.5)}}} which was used before.
> {{{
> $ make fate-parseutils
> TEST    parseutils
> --- ./tests/ref/fate/parseutils 2013-06-25 21:37:23.758635630 +0200
> +++ tests/data/fate/parseutils  2013-06-25 21:40:36.629074659 +0200
> @@ -6,7 +6,7 @@
>  '-123 / 0' -> -1/0 ERROR
>  '' -> 0/0 ERROR
>  '/' -> 0/0 ERROR
> -' 123  /  321' -> 41/107 OK
> +' 123  /  321' -> 213623/557504 OK
>  'foo/foo' -> 0/0 ERROR
>  'foo/1' -> 0/0 ERROR
>  '1/foo' -> 1/0 ERROR
> @@ -17,13 +17,13 @@
>  '0' -> 0/1 ERROR
>  '-123/123' -> -1/1 ERROR
>  '-foo' -> 0/0 ERROR
> -'123.23' -> 12323/100 OK
> -'.23' -> 23/100 OK
> -'-.23' -> -23/100 ERROR
> -'-0.234' -> -117/500 ERROR
> +'123.23' -> 697112/5657 OK
> +'.23' -> 209687/911683 OK
> +'-.23' -> -209687/911683 ERROR
> +'-0.234' -> -177357/757936 ERROR
>  '-0.0000001' -> -1/10000000 ERROR
> -'  21332.2324   ' -> 917286/43 OK
> -' -21332.2324   ' -> -917286/43 ERROR
> +'  21332.2324   ' -> 682631/32 OK
> +' -21332.2324   ' -> -682631/32 ERROR
>
>  Testing av_parse_color()
>  bikeshed -> R(80) G(64) B(140) A(59)
> Test parseutils failed. Look at tests/data/fate/parseutils.err for
> details.
> make: *** [fate-parseutils] Error 1
> }}}
>
> Not reproducible on ppc64.

New description:

 The parseutils fate test fails on sparc64 Linux and ia64 Linux since
 64eacb8. The problem is not limited to parseutils, many fate tests fail
 because of an incorrect timebase.
 The reason is that for some values of d and den - local variables in
 av_d2q() used when calling av_reduce() - {{{(uint64_t)rint(d * den)}}} and
 {{{llrint(d * den)}}} provide different results than {{{(int64_t)(d * den
 + 0.5)}}} which was used before.
 {{{
 $ make fate-parseutils
 TEST    parseutils
 --- ./tests/ref/fate/parseutils 2013-06-25 21:37:23.758635630 +0200
 +++ tests/data/fate/parseutils  2013-06-25 21:40:36.629074659 +0200
 @@ -6,7 +6,7 @@
  '-123 / 0' -> -1/0 ERROR
  '' -> 0/0 ERROR
  '/' -> 0/0 ERROR
 -' 123  /  321' -> 41/107 OK
 +' 123  /  321' -> 213623/557504 OK
  'foo/foo' -> 0/0 ERROR
  'foo/1' -> 0/0 ERROR
  '1/foo' -> 1/0 ERROR
 @@ -17,13 +17,13 @@
  '0' -> 0/1 ERROR
  '-123/123' -> -1/1 ERROR
  '-foo' -> 0/0 ERROR
 -'123.23' -> 12323/100 OK
 -'.23' -> 23/100 OK
 -'-.23' -> -23/100 ERROR
 -'-0.234' -> -117/500 ERROR
 +'123.23' -> 697112/5657 OK
 +'.23' -> 209687/911683 OK
 +'-.23' -> -209687/911683 ERROR
 +'-0.234' -> -177357/757936 ERROR
  '-0.0000001' -> -1/10000000 ERROR
 -'  21332.2324   ' -> 917286/43 OK
 -' -21332.2324   ' -> -917286/43 ERROR
 +'  21332.2324   ' -> 682631/32 OK
 +' -21332.2324   ' -> -682631/32 ERROR

  Testing av_parse_color()
  bikeshed -> R(80) G(64) B(140) A(59)
 Test parseutils failed. Look at tests/data/fate/parseutils.err for
 details.
 make: *** [fate-parseutils] Error 1
 }}}

 Not reproducible on ppc64.

--

-- 
Ticket URL: <https://ffmpeg.org/trac/ffmpeg/ticket/2713#comment:5>
FFmpeg <http://ffmpeg.org>
FFmpeg issue tracker