westonpace opened a new pull request, #35565:
URL: https://github.com/apache/arrow/pull/35565

   ### Rationale for this change
   
   Various compute kernels and Acero internals rely on type punning.  This is 
only safe when the buffer has appropriate alignment (e.g. casting uint8_t* to 
uint32_t* is only safe if the buffer has 4-byte alignment).  To avoid errors we 
enforced 64-byte alignment in Acero.  However, this is too strict.  While 
Arrow's allocators will always generate 64-byte aligned buffers this is not the 
case for numpy's allocators (and presumably many others).  This PR relaxes the 
constraint so that we only require value-aligned buffers.
   
   ### What changes are included in this PR?
   
   The main complexity here is determining which buffers need aligned and how 
much.  A special flag kMallocAlignment is added which can be specified when 
calling CheckAlignment or EnforceAlignment to only require value-alignment and 
not a particular number.
   
   ### Are these changes tested?
   
   TODO
   
   ### Are there any user-facing changes?
   
   No


-- 
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]

Reply via email to