On Sat, Mar 19, 2016 at 05:42:32PM -0400, Vittorio Giovara wrote: > --- > tests/fate/microsoft.mak | 3 + > tests/ref/fate/wmv8-intrax8 | 475 > ++++++++++++++++++++++++++++++++++++++++++++ > 2 files changed, 478 insertions(+) > create mode 100644 tests/ref/fate/wmv8-intrax8 > > This test breaks on an oracle box with gcc-4.4. I carried out a minimum > investigation and managed to isolate the issue to optimization level: when > compiled with -O2 gcc-4.4 produces the expected results, while -O3 generates > differing hashes. Note that with -O2 and the other 6-7 additional > optimizations > that -O3 enables, the test still works fine, it's something that is triggered > when -O3 is explicitly passed in. > > I don't have sample code to demonstrate this is a compiler bug, but I would > rather have this section of the code covered by this test and update > the gcc version on the oracle box, than trying to isolate the issue more. > However, I'd be more than happy to help anybody willing to work on this.
This instance I presume? https://oracle.libav.org/x86_64-gcc-4.4-c99conv/20160301213050 Another option is for that instance to add --extra-cflags="-O2" to its set of flags. That should work around the problem (wherever it may be) and also broaden the test coverage of FATE/ORACLE, IIRC we have no -O2 instance so far. > Since I don't want to create controversy, if there is no consensus I'll simply > drop this patch, and leave this portion of code untested. The changes brought > in by this set do not affect this test at all. Nah :) The test file seems a tad long; maybe you could cut it down to size a bit? Diego _______________________________________________ libav-devel mailing list libav-devel@libav.org https://lists.libav.org/mailman/listinfo/libav-devel
