https://bugs.kde.org/show_bug.cgi?id=385412
--- Comment #14 from Julian Seward <jsew...@acm.org> --- I looked at the patches. They all seem fine to me, except for one thing I wasn't happy about: the use of the three dirty helpers to implement PPNO. In particular I don't think s390x_dirtyhelper_PPNO_sha512_load_param_block will do what you want, because it merely says memory is loaded. But it doesn't show, at the IR level, any connection betweeen the loaded values and the rest of the computation -- they are simply ignored. I assume that they are really inputs somehow. Because of this, any undefinedness in the loaded memory will be ignored by Memcheck. I understand that PPNO is difficult to implement because it has so many inputs and outputs. We have had similar problems implementing crypto instructions on Intel and ARM before now. Can you tell me briefly what PPNO does? Specifically, what inputs and what outputs (registers and memory) does it have? Then I can maybe suggest an alternative implementation. -- You are receiving this mail because: You are watching all bug changes.
