harrisonGPU wrote: @overmighty @lntue I used a template to implement log_eval, which supports both float and double. I think this is a better way to implement a smaller float version of log_eval. https://github.com/llvm/llvm-project/pull/132612