Repository navigation
[Feature]: FP64 alpha/beta/gamma for elementwise binary and trinary ops in hipTensor #14403
Description
Suggestion Description
Follow-up to #13081 (review comment by @evedovelli). Binary and trinary ops accept R_64F, but alpha/beta/gamma are converted to float before reaching the kernel, so fp64 runs use float-precision scalars. Permute had the same issue, fixed in #13081 with ScaleF64. I’d like to implement this the same way.
Operating System
No response
GPU
No response
ROCm Component
hipTensor
Activity