[PATCH] D152766: AMDGPU: Use correct lowering for llvm.log2.f32

Thu Jun 22 12:58:16 PDT 2023

rampitec added a comment.

In D152766#4442408 <https://reviews.llvm.org/D152766#4442408>, @JonChesterfield wrote:

> In D152766#4442379 <https://reviews.llvm.org/D152766#4442379>, @rampitec wrote:
>
>> OK, got the explanation from @b-sumner. This looks very unusual, but it is correct.
>
> Please could you summarise it here? Also curious what's going on here

Quote from Brian:

In general, ULP(x) is defined for any real value of x and is roughly the distance from x to the nearest floating point number divided by the distance between the two nearest floating point numbers.

So, statements about a maximum relative error being 0.51 or 3.14159 ULP are quite meaningful.

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D152766/new/

https://reviews.llvm.org/D152766