> Summary: it seems that libdeflate explicitly uses some gcc-intrinsic > functions on gcc 11+, and it is expected that the assembler can handle > the result. So this doesn't depend on a compiler option and it looks > like we cannot disable it either.
This seems vaguely similar to the build failure I've observed for libvpx on aarch64 9.0, with the error being Error: selected processor does not support `udot ... I even tried specifying -march=armv8.2-a+dotprod to no avail. I also tried using binutils from pkgsrc, also to no avail. Which processor variant is our gcc configured for on amd64? Does that target include the required avx variant? The gcc man page under "x86 Options" lists a plethora of options to select specific variants. The "as" man page has a similar list. In my case it turned out that building this on 10.0 works without any modifications, while it doesn't on 9.0, despite binutils in pkgsrc being a newer variant than what's in-tree in 10.0. Is it perhaps so that the particular source file needs to be built with the appropriate -m option? Regards, - Håvard
