[bugfix] upcast input tensor to fp64 for cpu reference

[StrongSpoon]

PR Category

OP Test

Type of Change

Bug Fix

Description

upcast input tensor to fp64 for cpu reference. so that weight_norm and upsample_bicubic2d_aa could run fp16 tests in cpu mode.

Issue

Progress

• [ ] Change is properly reviewed (1 reviewer required, 2 recommended).
• [ ] Change is responded to an issue.
• [ ] Change is fully covered by a UT.

Performance

[iclementine]

I wonder why the original upcast value for these tests were False? I thought non-upcasting only applies to some integer related operarions?