make test_lapack more concise #211
Conversation
tests/test_lapack.cxx
Outdated
| expect_complex_near(h_B(2, 1, 1), 31.0); | ||
| auto expected_B = gt::gtensor<T, 3>( | ||
| {{{1., 2., 3.}, {-3., 7., 31.}}, {{1., 2., 3.}, {-3., 7., 31.}}}); | ||
| EXPECT_NEAR(gt::sum_squares(h_B - expected_B), 0., 1e-10); |
There was a problem hiding this comment.
The expect_complex_near helper can show which elements triggered the failure. This is good for small matrices, and awful for large ones, so my preference would be to keep the helper for this small case.
There was a problem hiding this comment.
Oh but this is not using the array helper but the individual one... this one is good for small arrays: https://github.com/wdmapp/gtensor/blob/main/tests/test_helpers.h#L110
There was a problem hiding this comment.
Well, you had another helper somewhere which flattened the two gtensors and then compared element by element -- that would be a nice compromise, since it's also a concise way of expressing stuff. I guess overall this kinda points out that we don't have proper iterators for gtensors, the best we can do is flatten and do a 1-d loop. Though now that I think about it, I guess we could use gt::launch with EXPECT inside of it, that'd seem neat enough.
germasch
force-pushed
the
pr/test-lapack
branch
from
005e60a
to
a1775ca
Compare
|
Learned two things from the looking at the sycl CI:
|
They are still kinda lengthy compared to what they do -- we should probably have a better way to deal with the host/device mirroring. And generally, there's still more room for improvement, I don't think I made all the best choices here...
germasch
force-pushed
the
pr/test-lapack
branch
from
a1775ca
to
04d4478
Compare
germasch
force-pushed
the
pr/test-lapack
branch
from
04d4478
to
1bf04a3
Compare
We do have a bunch of features that help work with matrices / vectors at a time, so let's use more of them. Still not really
as simple as I'd like things to be, but I think it's an improvement.