[llvm] [AMDGPU] Support double type in atomic optimizer. (PR #84307)

[Pravin Jagtap via llvm-commits]

pravinjagtap wrote:

> It looks like this patch only affects "noret" cases with uniform input values, so it does not compute any reductions or scans, so it does not exercise either the Iterative or the DPP algorithm.

It affects both `ret` & `no-ret` cases but only for uniform values (in this patch). Will add support for Iterative and DPP strategies (for divergent values) of scan in next patches.

https://github.com/llvm/llvm-project/pull/84307