Possible bugs in OpenCL drivers

[Chris_C_1]

Hi,

I have discovered several cases where programs compiled and executed on HD Graphics Haswell GT2 Desktop produce different output to on other OpenCL implementations. Please see attached bug reports, which provide additional details and steps to reproduce.

• https://pastebin.com/Ue5LzWsr
• https://pastebin.com/88s0dpWY
• https://pastebin.com/y9jhcfmE

On Xeon-Phi, this kernel causes a segmentation fault during compilation:

• http://paste.ubuntu.com/25125643/

Cheers,
Chris

[Chris_C_1]

I also have program which trigger assertion errors from within LLVM:

• https://pastebin.com/kCq0pea1
• http://paste.ubuntu.com/25027521/
• http://paste.ubuntu.com/25027522/
• http://paste.ubuntu.com/25027523/
• http://paste.ubuntu.com/25027524/
• http://paste.ubuntu.com/25027525/
• http://paste.ubuntu.com/25027526/
• http://paste.ubuntu.com/25027527/

Cheers,
Chris

[Chris_C_1]

Here are standalone C files containing kernels which segfault the Vectorizer (pass 'Post-Dominance Frontier Construction') on E5-2620 and i5-4570:

http://paste.ubuntu.com/25037714/
http://paste.ubuntu.com/25037715/
http://paste.ubuntu.com/25037716/
http://paste.ubuntu.com/25037717/
http://paste.ubuntu.com/25037718/
http://paste.ubuntu.com/25037719/
http://paste.ubuntu.com/25037720/
http://paste.ubuntu.com/25037721/
http://paste.ubuntu.com/25037722/

Cheers,
Chris

[Chris_C_1]

Another segfault during compilation: http://paste.ubuntu.com/25051842/

[Jeffrey_M_Intel1]

Sorry for the delayed reply.  Thanks for this report.  I am investigating these issues and will get back to you soon.

[Chris_C_1]

No problem Jeffrey. In the cases where there's no error messages, I can't weed out obvious duplicates, so apologies if any of those reports are dupes.

Cheers,
Chris

[Chris_C_1]

Compilation of this 4 line kernel hangs the compiler on GT2: http://paste.ubuntu.com/25180351/

This 1 line kernel segfaults every Intel compiler we're tested, except an i5-4570: http://paste.ubuntu.com/25180582/