Open derekbruening opened 3 months ago
This passed on a re-run. It passes every time on the non-sve machine (don't have simple direct access to the sve machine).
This issue happens only on sve it seems, but it is similar to a recent non-sve issue that was observed on signalNNN1 tests; where the result of some trig math floating-point ops changed and became non-deterministic after PR #6725.
Since #6725 adjusted some sve and non-sve logic too, maybe both of these are indeed related.
PR #6758 may help with this issue. I haven't used tmate to verify the failure count post that PR though.
The tool.record_filter_bycore_multi test failed on a64 sve at https://github.com/DynamoRIO/dynamorio/actions/runs/8546804195/job/23417839931?pr=6754
The output mismatch looks like it's b/c it estimated pi at 3.94... instead of 3.14...