Compilation on a Cell blade

[Erwin Coumans]

Result is segmentation fault. Presumably unaligned memory access.

The crash is due to use of vector instructions for copying members, with alignment limitations/assumptions in the gcc modifications in the IBM Cell SDK. The nasty bit is that the compiler/linker doesn't produce warnings or errors, but it leads to unexpected and buggy behaviour that is hard to track down. As a workaround, the alignment for btQuadWord is disabled when compiling using the IBM Cell SDK (#ifndef USE_LIBSPE2) in the latest Bullet trunk.

Do you want to integrate the benchmarks from Demos/BenchmarkDemo by your own? Do you need help on that?
How can I help you to move forward with this topic?

Can you help copying the benchmarks from Demos/BenchmarkDemo into the CellSpuDemo? Note that the current Demos/CellSpuDemo is purely command-line, without graphics. It would be good to keep an option to compile/link the CellSpuDemo without OpenGL.

Here is an issue to track progress:
http://code.google.com/p/bullet/issues/detail?id=112

Thanks for offering help, you patch/contribution is welcome,
Erwin

[Joczhen]

Can you help copying the benchmarks from Demos/BenchmarkDemo into the CellSpuDemo? Note that the current Demos/CellSpuDemo is purely command-line, without graphics. It would be good to keep an option to compile/link the CellSpuDemo without OpenGL.

Alright I'll take care of that and send you a first patch within the next days.

@Daniel: I'll also try to solve this gcc compiler issue.