Adding cudf::get_current_device_resource_ref in place of rmm calls
#2398
Conversation
There was a problem hiding this comment.
I would actually not rush this PR in. We are discovering issues with the Cython/C++ interface that are slowing the transition down. We have stepped back to do more codesign with the CCCL team to get this right. Things may change.
It's fine to merge this, just saying you may have to change it again later whether you merge this now or not. It will work with or without this change.
| @@ -256,7 +256,7 @@ TEST_F(TimeZoneTest, ConvertFromUTCMicroseconds) | |||
| 568036800000000L, | |||
| }; | |||
| auto const actual = spark_rapids_jni::convert_utc_timestamp_to_timezone( | |||
| ts_col, *transitions, 1, cudf::get_default_stream(), rmm::mr::get_current_device_resource()); | |||
| ts_col, *transitions, 1, cudf::get_default_stream(), cudf::get_current_device_resource_ref()); | |||
|
|
|||
| CUDF_TEST_EXPECT_COLUMNS_EQUAL(expected, *actual); | |||
| } | |||
There was a problem hiding this comment.
This file needs a newline at the end (clang-format should be catching that).
Do you have an issue or something to monitor for these issues and changes? |
|
|
Actually, the rest of libcudf has been updated to use |
|
@hyperbolic2346 can you please retarget to branch-24.12? |
…_current_device_resource_ref
Signed-off-by: Mike Wilson <[email protected]>
|
build |
|
Please hold off. The design discussion on the CCCL side continues. I will let you know when it settles down. |
This really closes #2358. It was closed earlier, but @harrism pointed out that we should wait for rapidsai/cudf#16679 and use the new cudf function. This is that update.