You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Reductions currently have several issues in graphs, though they work in some limited cases (where the device supports USM and it happens through record and replay) but there are issues particularly in explicit mode (reduction code expects a device in the handler). These should be addressed for both modes and with buffers and USM.
Accessor support was added in #151 which should help with this issue.
Note: Buffers may be hard to test if we can't force reductions to use them.
The text was updated successfully, but these errors were encountered:
Reduction support is currently
[incomplete](#159) and we
workaround this in the tests with an ifdef. However, this isn't
documented and hides the reduction implementation status.
Instead, move reductions out into their own tests. These current
pass in Recording mode when a device supports USM, but not in
Explicit.
As part of #174 some reduction tests have been added, the explicit ones have been marked as XFAIL. As part of actioning this ticket, all these tests should be reliable passes.
Reductions currently have several issues in graphs, though they work in some limited cases (where the device supports USM and it happens through record and replay) but there are issues particularly in explicit mode (reduction code expects a device in the handler). These should be addressed for both modes and with buffers and USM.
Accessor support was added in #151 which should help with this issue.
Note: Buffers may be hard to test if we can't force reductions to use them.
The text was updated successfully, but these errors were encountered: