|Re: [eigen] Accidental push/merge for s390x SIMD :)|
[ Thread Index |
| More lists.tuxfamily.org/eigen Archives
Στις 13-04-2016, ημέρα Τετ, και ώρα 14:25 +0200, ο/η Gael Guennebaud έγραψε:
> sorry for late reply,
Hi Gael, no problem.
> What are those 10 tests? Do you consider this support mature enough to be shipped with the 3.3
> release, or shall it be delayed for 3.4?
99% tests passed, 10 tests failed out of 716
Label Time Summary:
Official = 9214.82 sec
Unsupported = 581.59 sec
Total Test time (real) = 9871.08 sec
The following tests FAILED:
7 - nomalloc_3 (OTHER_FAULT)
27 - mixingtypes_5 (OTHER_FAULT)
33 - packetmath_5 (OTHER_FAULT)
113 - product_small_5 (OTHER_FAULT)
470 - bdcsvd_9 (OTHER_FAULT)
487 - geo_quaternion_2 (OTHER_FAULT)
639 - fastmath (OTHER_FAULT)
641 - NonLinearOptimization (OTHER_FAULT)
682 - matrix_square_root_3 (OTHER_FAULT)
715 - levenberg_marquardt (OTHER_FAULT)
Most are alignment issues (some are for float, which I'm going to ignore for now, but there are a
couple for double), I'm looking at the packetmath_5 right now n particular:
Test packetmath_scatter_gather<std::complex<double> >() failed in
isApproxAbs(buffer[i], data1[i/stride], refvalue) && "pscatter"
Apart from that, I think it's in a good state, and when I'm done with the pscatter issue, it should
worth including in 3.3 already.
I'm in the process of setting up proper nightly tests for s390x and VSX platforms, so it would help
in monitoring the ports' status (I've left the Altivec/VSX one unattended for a while), plan to get
it up to par with the rest, maybe implement some missing functions here and there). I'd like to be
done with all ports (except for arm64, no machine yet, but maybe the compile farm that Marc
suggested might be a good choice, so I may give it a shot).
> Since ZVector does not address single precision, I guess that users of such systems do not mind
> about float at all, so I would not waste time for that.
Yeah, it's on a very low priority for me as well.