markalle

Update: okay, testing shows not quite ready still. I think the detection is still a slight weak spot

Repushed based on ppc testing and my comment above. That path worked fine in the default mode (where __float128 is available) but I also tested artificially turning off HAVE__FLOAT128 just...

So the systems I've tested on so far are * x86_64 mac (which doesn't have __float128, uses f80_to_f128) * ppc9 linux (uses __float128 conversion path) * ppc9 linux (with __float128...

For tracing through the changes, I think main pieces are: * long double detection: based on LDMANTDIG and LDEXPSIZE in arch * added from_arch and to_arch arguments to pFunction *...

I'm trying to refresh my memory if there were any concerns left in this PR. One test I ran that I didn't mention in my comments above is for big-endian,...

I found an ubuntu image for s390x and ran it under QEMU anyway, and it worked there. The OMPI code looks at LDBL_MANT_DIG and LDBL_MAX_EXP and sets OPAL_ARCH_LDMANTDIGIS113 and OPAL_ARCH_LDEXPSIZEIS15...

Are these jobs running at the same time? If they're not running at the same time then I don't think there's any overlap, they both look like 2-host jobs where...

@gpaulsen asked me to post how many symbols there are that need changed, based on my run just now it's ~250. Admittedly if we're only being pragmatic and non-strict we...

Ah, yes I did make an actual test that shows how redefining one of those global symbols in an MPI user applications breaks OMPI: *** demonstrating why pollution is a...

Thanks for the review, if treematch isn't to be modified, then I think PR8132 has got to leave it out of libmpi.so, since it's by far the largest offender in...