Custom implementation of isfinite #1383
Conversation
|
View rendered docs @ https://intelpython.github.io/dpctl/pulls/1383/index.html |
|
Array API standard conformance tests for dpctl=0.14.6dev4=py310ha25a700_37 ran successfully. |
|
Array API standard conformance tests for dpctl=0.14.6dev4=py310ha25a700_38 ran successfully. |
|
@ndgrigorian This PR has fixed SYCL-OS workflow. |
| @@ -45,6 +46,7 @@ namespace isnan | |||
|
|
|||
| namespace py = pybind11; | |||
| namespace td_ns = dpctl::tensor::type_dispatch; | |||
| namespace mu_ns = dpctl::tensor::math_utils; | |||
There was a problem hiding this comment.
In isnan.hpp, math_utils is imported and the namespace is set, but never used to call mu_ns::isnan
There was a problem hiding this comment.
Pushed the fix. Thanks for spotting that out @ndgrigorian . I was after std::isnan instances and missed sycl::isnan.
|
Array API standard conformance tests for dpctl=0.14.6dev4=py310ha25a700_39 ran successfully. |
|
Since the issue has been fixed in SYCLOS, perhaps this PR can wait. |
oleksandr-pavlyk
force-pushed
the
custom-is-finite
branch
from
2999014
to
065b2d2
Compare
|
Array API standard conformance tests for dpctl=0.14.6dev4=py310ha25a700_54 ran successfully. |
1 similar comment
|
Array API standard conformance tests for dpctl=0.14.6dev4=py310ha25a700_54 ran successfully. |
This works-around a bug in std::isinfite recently introduced in SYCLOS where std::isfinite(-0.0) returns false for floating point types, including sycl::half, float, double.
oleksandr-pavlyk
force-pushed
the
custom-is-finite
branch
from
065b2d2
to
4ff0543
Compare
|
Array API standard conformance tests for dpctl=0.14.6dev5=py310ha25a700_4 ran successfully. |
|
Closing as no longer relevant |
This PR should address issues test failures from gh-1378
It works around very annoying SYCLOS bug introduced with pull down of clang 18.0
It adds a custom implementation of
math_utils::isfiniteto work around this failure.