Serialize output arg

[pfultz2]

No description provided.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 92.18%. Comparing base (2f97579) to head (4f168d1).
Report is 132 commits behind head on develop.

Additional details and impacted files

@@           Coverage Diff            @@
##           develop    #3613   +/-   ##
========================================
  Coverage    92.18%   92.18%           
========================================
  Files          513      513           
  Lines        21548    21548           
========================================
  Hits         19863    19863           
  Misses        1685     1685           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[CharlieL7]

What's the use case for the output_arg?

[turneram]

I'm seeing correct results from mxr with this change.

[pfultz2]

What's the use case for the output_arg?

Originally it was needed for MLIR as the last parameter wasnt the output parameter. Instead it would take the output buffer and then the sizes(ie lens, strides) to the buffer, that is no longer the case.

In this case, the concat_past_present_kernel outputs the first parameter because it modifies the input inplace rather than taking an output buffer to copy the ouput to. So we set it to 0 instead of the default -1.