Hi, while I haven't given it a thorough look, the comment "Rounding errors make the algorithm behave differently on different platforms" can be spotted very easily from the failed salsa job itself. I am not sure, but perhaps this test wasn't supposed to be run on several architechtures to begin with?
Best, Ananthu On 16 July 2024 9:06:07 pm UTC, "Étienne Mollier" <[email protected]> wrote: >Quick lookup at the test item >shows numpy code, which I believe is sensitive to floating point >unit precision of the underlying hardware.
