crowd-funded eco-conscious hardware: https://www.crowdsupply.com/eoma68
On Tue, Feb 9, 2021 at 7:43 PM Tulio Magno Quites Machado Filho
Robert Scheck <robert(a)fedoraproject.org> writes:
> Trying to follow the path on how to get Fedora for ppc64le to a
> I had to learn that glibc developers seem to treat IBM POWER9 strictly equal
> to ABI rules rather handling the different features more fine granulated. My
> understanding here might be imprecise or wrong, this is what I understood
> and I'm no expert at this topic.
There are indeed a couple of bad assumptions in glibc code that go beyond
what Florian mentioned about the ELFv2 ABI.
These include fixed cache line size and assuming that VSX is available
POWER ISA 2.06 or later.
Even though we can't change the ELFv2 ABI, I've been (slowly) working to remove
these assumptions and make it easier to use an ELFv2 glibc build on processors
that do not implement VSX.
Tulio this is absolutely fantastic to hear. i have reached out to
bill schmidt and there will be some discussion and ideas thrown back
and forth, here. if you're actively working on fixing the various
#defines this is a huge relief.
If you find any other issues, please let me know. I'll be glad to
> The IBM Power ISA Version 3.1  specifies on page 8 compliancy subsets,
> where the Linux Compliancy Subset (LCS) is explained on page 10 ff.
Notice that SIMD is not a Linux Optional feature in this page. ;-)
ahh you may mean that the other way round? so as not to spam people
with attachments, allow me to post screenshot links instead:
it distinctly says "optional" for the Linux Compliancy Subset at the
top of the page. i initially completely missed this.
now, if you want to have Quad-precision Floating-point (QFP) then SIMD
*is* required (highlighted by the green arrows). likewise if you want
SIMD, you must also have FP.
so this defines the dependency-chain:
* QFP depends on SIMD
* SIMD depends on FP
all these things - including LSM, LS, PCR AMO, they really should all
be in a ppc glibc6 HWCAPS.