[llvm] [AMDGPU] Add f32 cost model for exp/exp2/exp10 intrinsics (PR #185369)

[Juan Manuel Martinez Caamaño via llvm-commits]

================
@@ -755,6 +755,20 @@ GCNTTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
       return LT.first * NElts * NumOps * get64BitInstrCost(CostKind);
     }
 
+    if (SLT == MVT::f32) {
+      unsigned NElts =
+          LT.second.isVector() ? LT.second.getVectorNumElements() : 1;
----------------
jmmartinez wrote:

The f32 and f64 cases seem to use

```
unsigned NElts =
          LT.second.isVector() ? LT.second.getVectorNumElements() : 1;
```

Can we refactor them together to avoid repetition?

https://github.com/llvm/llvm-project/pull/185369