Wow, great find!  Sounds like a definite step forward.  Thanks very
much for all your work on this!

On Thu, Aug 22, 2013 at 8:50 PM, Michael Gogins
 wrote:
> I have discovered part of the reason for problems I described earlier with
> the Windows build of Csound 6.
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48659 for some time now causes
> memory misalignment in stack frames and thus segfaults when OpenMP and SSE
> are used together, as we often do in Csound.
>
> There is a workaround. All code that uses both SSE and OpenMP should be
> compiled with the -mstackrealign option. For our purposes, that means all
> code that is part of Csound and all code that is used by Csound. Reports are
> that there is no performance penalty.
>
> After rebuilding Csound and CsoundQt with MinGW 4.8.1 32 bits dwarf2
> exception handling Posix threads, just as I have been doing, but with
> -mstackrealign, the CsoundPerformanceThread unit test passes, and CsoundQt
> runs and performs Csound pieces.
>
> I also ran CppCheck on the Csound source code and fixed a number of errors
> and warnings.
>
> The memory allocation error I described earlier still occurs.
>
> The problem with CsoundQt not running audio properly a second time still
> occurs.
>
> The problem with the Java examples not running still occurs.
>
> Nevertheless, I am replacing the Csound installer on SourceForge with this
> new build.
>
> In the future, if debugging does not soon reveal the cause of a bug, you can
> be sure I am heading straight for the compiler bug trackers.
>
> The next minor release of Csound will include all third party code compiled
> with the same options.
>
> Regards,
> Mike
>
>
> ===========================
> Michael Gogins
> Irreducible Productions
> http://michaelgogins.tumblr.com
> Michael dot Gogins at gmail dot com