Programs calling libm's pow() exit with SIGILL because it executes the FMA4 instruction vfmaddsd %xmm4,0x8(%rdx),%xmm6,%xmm0 on a system that does not support FMA4. This is apparently due to a regression in commit 107e6a3c2212ba7a3a4ec7cae8d82d73f7c95d0b file sysdeps/x86_64/fpu/multiarch/ifunc-fma4.h where the ifunc checks for "FMA" instead of "FMA4". Originally reported to Arch Linux: https://bugs.archlinux.org/task/67715
Created attachment 12797 [details] patch fixing the regression (correcting typo/backspace-o)
(In reply to Ondřej Hošek from comment #1) > Created attachment 12797 [details] > patch fixing the regression (correcting typo/backspace-o) Can you send it to the glibc mailing list? Thanks.
Patch sent to the mailing list: https://sourceware.org/pipermail/libc-alpha/2020-August/117267.html
Gentoo also encountered it: https://bugs.gentoo.org/740110 Looks like patch was merged into master as: https://sourceware.org/git/?p=glibc.git;a=patch;h=23af890b3f04e80da783ba64e6b6d94822e01d54
Yup, and backported to the 2.32 branch in https://sourceware.org/git/?p=glibc.git;a=commit;h=cebc01cbfd6c257f329e818b68bb1b2eef2afe09 I guess I can change the status to RESOLVED FIXED now...