lib.libc.gen.floatunditf_test.floatunditf: https://ci.freebsd.org/job/FreeBSD-head-i386-test/4868/testReport/lib.libc.gen/floatunditf_test/floatunditf/ lib.libc.stdio.printfloat_test.hexadecimal_rounding: https://ci.freebsd.org/job/FreeBSD-head-i386-test/4868/testReport/lib.libc.stdio/printfloat_test/hexadecimal_rounding/ lib.msun.ctrig_test.test_small_inputs: https://ci.freebsd.org/job/FreeBSD-head-i386-test/4868/testReport/lib.msun/ctrig_test/test_small_inputs/ lib.msun.precision_test.t_precision: https://ci.freebsd.org/job/FreeBSD-head-i386-test/4868/testReport/lib.msun/precision_test/t_precision/
What was the method you used to point out this specific commit ? I tried the same i386 test binary on stable/12 _amd64_ kernel (floatunditf) and it failed the same way, giving unexpected success. That said, I do not see any difference between i386 FPU state on HEAD/i386 vs. stable/12 amd64 compat32. So this makes me wonder if the change is somewhere else, and I am asking why do you blame r345562.
These 4 tests start failing since: https://ci.freebsd.org/job/FreeBSD-head-i386-test/4868/ (r345562) But not in: https://ci.freebsd.org/job/FreeBSD-head-i386-test/4867/ (r345560) (r345561 is a MFC to stable/11) Then I checked the i386 image in bhyve: https://artifact.ci.freebsd.org/snapshot/head/r345562/i386/i386/disk-test.img.xz https://artifact.ci.freebsd.org/snapshot/head/r345560/i386/i386/disk-test.img.xz Running these commands: root@:/usr/tests/lib/libc/gen # kyua test floatunditf_test:floatunditf root@:/usr/tests/lib/libc/stdio # kyua test printfloat_test:hexadecimal_rounding root@:/usr/tests/lib/msun # kyua test ctrig_test:test_small_inputs root@:/usr/tests/lib/msun # kyua test precision_test:t_precision All failed in r345562 but not in r345560. BTW I am sorry that I didn't emphasize the floatunditf_test:floatunditf and precision_test:t_precision case are marked as expected_failure, but now no failure occurs so we need check if r345562 fixes these case and rmeove that expected failure mark from test cases. While other 2 do report some checks failed.
(In reply to Li-Wen Hsu from comment #2) I cannot care less about failing/non-failing marking for the tests, I only care about changed behavior. And, if I take binaries from i386 HEAD and run them on stable/12 amd64 kernel, the tests behave same as on HEAD, while the commit you mentioned r345560 is not in stable/12 kernel. Can you run stable/12 i386 test binaries on i386 HEAD kernel ?
https://reviews.freebsd.org/rS346294