paleolimbot commented on code in PR #527:
URL: https://github.com/apache/arrow-nanoarrow/pull/527#discussion_r1639998943
##########
src/nanoarrow/nanoarrow_device_metal_test.cc:
##########
@@ -237,15 +246,25 @@ TEST_P(StringTypeParameterizedTestFixture,
ArrowDeviceMetalArrayViewString) {
EXPECT_EQ(device_array_view.array_view.buffer_views[2].size_bytes, 7);
- // We should be able to move arrays to Metal (this may depend on MacOS
version
- // ...wrapping an arbitrary sequence of bytes as an MTL::Buffer failed under
- // previous versions of MacOS but now seems to work)
+ // In some MacOS environments, ArrowDeviceArrayMoveToDevice() with buffers
allocated
+ // using ArrowMalloc() will work (i.e., apparently MTL::Buffer can wrap
+ // arbitrary bytes, but only sometimes)
Review Comment:
I actually have no clue! I don't think it's MacOS version (becuase both me
and the GitHub runner are on 14.5), but it's also not hardware (because I wrote
the initial version on the same computer where the test is now failing).
Metal is sort of a mystery to most people, including me...I did this
initially to ensure that the framework I'd come up with was going to work for
something other than CUDA. Eventually I'll have to check with
https://github.com/ml-explore/mlx to get a better handle on what's possible
here.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
