Log2p1 was implemented in commit id(79c52daf4740b00bef43e4f5a4281b3480459b38)and maximum allowed error was updated for powerpc in the commit (f83e461f1014598a5cb4c89407ce303b9f0bd8ac). We are seeing the tests was failing with "--disable-multi-arch" in power8 LE. The issue with those test cases is that the expected ULP from "sysdeps/powerpc/fpu/libm-test-ulps" does not match the result. The expected ULP for float128 is 2, but we are getting 3 for Power8. However, there are no issues on Power9 and Power10 since their output has error of 2 ulp. Can we change the ulps for float128 to 3? Is this a reasonable change.
It's generally reasonable to update ulps like that (small differences can often result from e.g. compiler choices about whether to use fused operations).
This issue has been resolved by changing the ULP value for float128. Issue resolved by the following commit. commit 71dafdf5f19dd2b0729e4774149944911a405bc6 Author: Florian Weimer <fweimer@redhat.com> AuthorDate: Thu Jun 20 11:07:17 2024 +0200 Commit: Florian Weimer <fweimer@redhat.com> CommitDate: Thu Jun 20 12:15:31 2024 +0200 powerpc: Update ulps Results based on POWER8 and POWER9 machines running powerpc64-linux-gnu, with and without --disable-multi-arch.