https://bugs.kde.org/show_bug.cgi?id=435665
--- Comment #7 from Julian Seward <jsew...@acm.org> --- I looked at the PowerISA 3.0b documentation you linked to. Given that this is a copy from "normal" memory to an accelerator, I think you don't have the option to implement it precisely. I'd say the least worst is to implement it so that, for the copy, Memcheck will flag an error if any of the 128 bytes are undefined, and for the paste we make it look as if the 128 bytes are completely defined. That way, at least you'll know if you're sending undefined values to the accelerator. Exactly how to achieve this I am not sure. It will take some fiddling with the annotations on the dirty helper calls. I'll contemplate it more. -- You are receiving this mail because: You are watching all bug changes.
