Conversation
ndgrigorian
requested review from
antonwolfy,
vlad-perevezentsev and
vtavana
as code owners
|
View rendered docs @ https://intelpython.github.io/dpctl/pulls/2144/index.html |
|
Array API standard conformance tests for dpctl=0.21.0dev0=py310h93fe807_143 ran successfully. |
|
Array API standard conformance tests for dpctl=0.21.0dev0=py310h93fe807_144 ran successfully. |
| . $CONDA/etc/profile.d/conda.sh | ||
| conda activate ${{ env.TEST_ENV_NAME }} | ||
| python -c "import dpctl; dpctl.lsplatform(verbosity=2)" | ||
| - name: Install gdb |
There was a problem hiding this comment.
It might be helpful in the future, when we will need to debug some test.
Would it be better to comment instead?
There was a problem hiding this comment.
I figured we'd just write it back in later if it becomes necessary again, we could just comment though
ndgrigorian
force-pushed
the
reduce-bloat-in-dpctl-test-suite
branch
from
8f68787 to
86965dd
Compare
|
Array API standard conformance tests for dpctl=0.21.0dev0=py310h93fe807_155 ran successfully. |
|
Array API standard conformance tests for dpctl=0.21.0dev0=py310h93fe807_155 ran successfully. |
|
Array API standard conformance tests for dpctl=0.21.0dev0=py310h93fe807_160 ran successfully. |
|
Array API standard conformance tests for dpctl=0.21.0dev0=py310h93fe807_164 ran successfully. |
also fixes a typo
no longer needed since crash was resolved
also consolidates two add tests
this test keeps test coverage of some branches related to type resolution
ndgrigorian
force-pushed
the
reduce-bloat-in-dpctl-test-suite
branch
from
d2d4e5a to
d97f840
Compare
|
Array API standard conformance tests for dpctl=0.21.0dev0=py310h93fe807_191 ran successfully. |
|
Changes in this PR reduce time of execution of test suite on my CPU from 6 minutes and 39 seconds to 5 minutes and 43 seconds. Coverage drop of -0.02% is minimal and is likely a fluke. |
| x = dpt.asarray( | ||
| [1, 2, 3, 4], dtype=dtype, usm_type=usm_type, sycl_queue=q | ||
| ) | ||
| return getattr(dpt, fn)(x) |
There was a problem hiding this comment.
Should an assertion that the result vector has usm_type be added here? Costs very little.
I would also make strided x inputs, such as x[::2] or dpt.flip(x).
There was a problem hiding this comment.
I opted to add two individual usm_type coercion tests, just to minimize redundancy, checking in each test would be fine too
I can also add for strided inputs.
4 participants
This PR proposes reducing the scope of the dpctl test suite by redacting tests which are redundant
For example, the dpctl test suite for element-wise functions repeatedly tests for things like
usm_typeof the output array. But the output arrayusm_typeis determined in the function which dispatches to the element-wise kernel, and therefore, only needs to be tested once