this looks perfect: https://refspecs.linuxfoundation.org/ELF/ppc64/PPC-elf64abi.html no VSX. https://openpowerfoundation.org/?resource_lib=64-bit-elf-v2-abi-specification-power-architecture this is where it goes awry. whilst the Compliancy Levels make VSX entirely optional, even at the Embedded FP level, it's the V2 ABI which forces VSX.
http://lists.mailinglist.openpowerfoundation.org/pipermail/syssw-elfv2abi/
http://lists.mailinglist.openpowerfoundation.org/pipermail/syssw-elfv2abi/2018-December/000014.html found contact details for bill schmidtt
https://lists.fedoraproject.org/archives/list/ppc@lists.fedoraproject.org/thread/F6XHJEPXFLUKXQP3SAMBXVUYC4GPTPTA/
if we can, we should try to use the v2 abi without altivec since it also has many other improvements. *vsx is actually a misnomer when applied to powerpc's SIMD support in general*, it refers specifically to the new vector-scalar interop instructions added in POWER9 (or 8? icr), not to SIMD in general, we should use either altivec or SIMD as terms that people will recognize otherwise we may get software support for SIMD without vsx -- not the intended endpoint.
(In reply to Jacob Lifshay from comment #4) > if we can, we should try to use the v2 abi without altivec since it also has > many other improvements. > > *vsx is actually a misnomer when applied to powerpc's SIMD support in > general*, it refers specifically to the new vector-scalar interop > instructions added in POWER9 (or 8? icr), not to SIMD in general, we should > use either altivec or SIMD as terms that people will recognize otherwise we > may get software support for SIMD without vsx -- not the intended endpoint. vmx is also an acceptable term for altivec: https://en.wikipedia.org/wiki/AltiVec
https://libre-soc.org/irclog/%23libre-soc.2021-02-12.log.html#t2021-02-12T16:33:42
https://gcc.gnu.org/legacy-ml/gcc/2015-11/msg00068.html
good example https://patches-gcc.linaro.org/patch/4834/
this turns out not to be true (many thanks to steve, author of pveclib) there *is* an ABI, it is currently v2.0 soon to be v2.1, and VSX *is* optional, as are many other things that suit SFFS and SFS perfectly fine. steve kindly explained to me that going back to v1.4 for example would be a serious mistake, as there were a *lot* of issues fixed. it also turns out to be straightforward to pass the options to today's gcc to get it to disable VSX and 128-bit FP: https://bugs.libre-soc.org/show_bug.cgi?id=1131#c1