[FFmpeg-devel] The state of FATE

[Måns Rullgård]

Here's a quick summary of the current errors FATE is reporting.

AVR32
-----

Linux, gcc 4.2.4 + AVR32 patches:
- Regression tests fail on WMA.  This is caused by floating-point
  rounding anomalies, possibly a bug in gcc's softfloat implementation
  for AVR32.  My sources say a new AVR32 patchset for gcc is to be
  released Real Soon Now (tm).

PowerPC
-------

Linux, gcc svn 150925: 185/311 tests pass
- All(?) h264 tests fail.  Any clues as to why?
- Regression tests fail on h263p and snowll.  Don't know why.

Sparc
-----

No updates on Linux/Sparc for 10 days.  Something needs a kick.

IA64
----

Linux, gcc 4.1.3: 310/311 tests pass
- Regression tests fail on dnxhd.  Are we certain this is a gcc bug and
  not an obscure code error with undefined behaviour?

Linux, gcc snapshot 4.5.0 20090923: build fail
- ICE in ra144.c

x86_32
------

Linux, gcc svn 150925: build fail
- FATE misconfigured

Linux, icc 10.1.022: 310/311 tests pass
- Regression tests fail on mp2.

Linux, llvm svn 79390: all tests pass
- Using --arch=c, why?

NetBSD, gcc-4.1.3 20080704 prerelease: all tests pass
- Using --extra-cflags=-D_NETBSD_SOURCE, why?

Cygwin, gcc 4.2.4 and 3.4.6: 310/311 tests pass
- Regression tests fail on wma, probably float rounding problems.

FreeBSD, gcc 4.2.5 20090325: all pass
- Using --extra-cflags=-D__BSD_VISIBLE, why?

x86_64
------

FreeBSD, gcc 4.2.5 20090325 and clang 1.1: all pass
- Using --extra-cflags=-D__BSD_VISIBLE, why?

Linux, gcc svn 150925: 310/311 tests pass
- Regression tests fail on snowll.

Linux, suncc 5.10: all pass
- Using various cflags hacks, patch pending.

Linux, icc 11.0.083: xx/127 tests pass
- Only 127 tests are run, why?
- All h264 tests fail, various compiler bugs confirmed.
- Regression tests fail on snow, snow53, svq1.  Any known cause?

Linux, llvm svn 79390: all pass
- Using --arch=c, why?

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