Issue 243132
Summary `[InstCombine]` Fold `exp(subnormal)` and `exp(+inf)` to `x + 1.0`
Labels good first issue, llvm:instcombine, floating-point
Assignees
Reporter ZERICO2005
    https://github.com/llvm/llvm-project/blob/f1b92cc3a1b0d13b01633b1d333a033b2da8006a/llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp#L3194-L3203

Currently, we perform these folds:
- `exp(zero)` --> `1.0`
- `exp(zero | NaN)` --> `x + 1.0`
- `exp(+-inf | NaN)` --> `select (fcmp ueq x, +inf) x, +0.0`

Since `exp(x) = 1.0 + x + ...`, and `1.0 + subnormal == 1.0`, we can also fold `exp(subnormal)` --> `1.0`. Also `exp(+inf) == +inf`.

TODO:
- Fold `exp(zero | subnormal)` --> `1.0` (Invalid for `RoundingMode::Dynamic`, although we don't have to worry about this here).
- Fold `exp(zero | subnormal | +inf | NaN)` --> `x + 1.0`
- Also make sure we do **not** fold `exp(zero | subnormal | +-inf | NaN)` --> `x + 1.0`

These also apply to `exp2` and `exp10`.
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs

Reply via email to