onnxruntime/winml
Jeff Bloomfield 0180c0429f
Fix DML regression from allocator refactor and enable unrounded weight allocation in ORT API (#17030)
This addresses a DML performance regression from the following PR
resulting in allocations not being rounded and pooled in the DML
execution provider.

https://github.com/microsoft/onnxruntime/pull/15833

This also fixes a pre-existing limitation that allocations during
session initialization (primarily large weights and persistent
resources) only bypassed rounding and pooling while using the Winml API.
The allocator now also respects a caller's rounding mode parameter when
provided.
2023-08-10 17:02:24 -07:00
..
adapter Fix DML regression from allocator refactor and enable unrounded weight allocation in ORT API (#17030) 2023-08-10 17:02:24 -07:00
api Format c++ code under winml/ (#16660) 2023-07-25 21:56:50 -07:00
dll Format c++ code under winml/ (#16660) 2023-07-25 21:56:50 -07:00
lib Fix DML regression from allocator refactor and enable unrounded weight allocation in ORT API (#17030) 2023-08-10 17:02:24 -07:00
test Fix DML regression from allocator refactor and enable unrounded weight allocation in ORT API (#17030) 2023-08-10 17:02:24 -07:00
.clang-format Format c++ code under winml/ (#16660) 2023-07-25 21:56:50 -07:00