2013 SDK miscompiles reduction kernel with structs

[Kloeckner__Andreas]

Hi there,

the attached PyOpenCL example demonstrates an issue with the below version of the Intel CPU CL runtime, observed on Intel Sandy Bridge and AMD Piledriver. Specifically, the assertion at the end of the code ("the bounding box is found correctly") is violated when running the example with the 2013 runtime. Everything is fine when using the 2012 runtime or the AMD runtime. I'm aware that this is not very efficient code for a CPU device, but I claim that it is correct.

Thanks,

Andreas

SDK Version: intel_sdk_for_ocl_applications_2013_xe_beta_sdk_3.0.56860_x64

[Yuri_K_Intel]

Hi Andreas, Thank you for reporting the issue. Indeed, it is reproducible for me on the beta version (56860). But it runs OK on the latest development version. So the fix for it will be available in the next release later this year. Thanks, Yuri

[Kloeckner__Andreas]

Hi Yuri,

Thanks for the update. I'd of course love an ETA for that release, but I imagine you won't be able to help me with that... :)

Andreas