https://bugs.kde.org/show_bug.cgi?id=385409
--- Comment #32 from Julian Seward <jsew...@acm.org> --- (In reply to Florian Weimer from comment #30) > Created attachment 114305 [details] > Implement early exit in s390_vr_loadWithLength > > Updated patch. This seems to fix the VLL issue. Limited testing only. It would be better to implement guarded loads and use that scheme, so that we don't have to end the superblock every time we see VLL. > However, these VLL instructions are typically part of the GCC inline > expansion of strlen, and the result is used with VFENEZBS. Since VLL is > used to over-load bytes until the next boundary, the vector result is > partially undefined. VFENEZBS is implemented with a host assist only, > though, so we get tons of warnings from inline copies of strlen: What does VFENEZBS do? I imagine this will need to be fixed by making a translation of it that is more Memcheck-friendly, and/or adjusting Memcheck to deal with the underlying IROps better. -- You are receiving this mail because: You are watching all bug changes.
