Bug in ATI Stream SDK Beta3 for linux x64

[sabby]

Thought I would post here since maybe there's more ATI linux folks here than in their forum... I was playing around with the opencl sdk for linux and came across an SSE4 instruction that seems to be compiled in the .so file even though the spec state SSE3 cpu and later. To reproduce compile and run the binary search sample on non SSE4 cpu and you should get an illegal instruction, the illegal instruction is

pinsrd $0x0,%edx,%xmm0

the stack

#0 0x00007f41828fd6d8 in __OpenCL_binarySearch_stub () from /tmp/OCLJDiw5W.so
#1 0x00007f4183ddff2f in cpu::WorkGroupOperation::execute ()
from /home/dev/ati-stream-sdk-v2.0-beta3-lnx64/lib/x86_64/libOpenCL.so

[sabby]

FYI I still get this error with the newly released skd beta 4. I didn't try yet to download beta driver to run it on the gpu so don't know if it works or not but I still get the same error when running --device cpu.

[bridgman]

The stream folks are pretty active on the AMD forums - the OpenCL forum on developer.amd.com might be your best bet.

[sabby]

The stream folks are pretty active on the AMD forums - the OpenCL forum on developer.amd.com might be your best bet.

Thanks posted there see if they reply. Again looks like a great SDK, lots of potential to integrate in existing apps. Have to see still what kind of speed up we can get. I'm interested in the AES encryption and integrating that in the kernel to provide transparent on the fly encryption of the hdd. Lots of other potential area that are interesting...