[llvm] [AMDGPU] Use s_fmamk_f32 when the folded addend is inlinable (PR #218955)

[Manuel Carrasco via llvm-commits]

mgcarrasco wrote:

Hi! This comment was generated using [fuzz-fill](https://github.com/ROCm/fuzz-fill), a new tool we're building to help improve LLVM test coverage by spotting lines introduced in a PR that don't yet appear to be exercised by the full test suite, including any tests added in this PR.

Looking at PR head fb96fe53981e, we found 2 lines added or modified here that we couldn't find a test for:

1. https://github.com/llvm/llvm-project/blob/fb96fe53981ea71196a714c4d85fc07c4d714cf2/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp#L927
2. https://github.com/llvm/llvm-project/blob/fb96fe53981ea71196a714c4d85fc07c4d714cf2/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp#L934

If it makes sense, adding a regression test that validates these lines would be great. And if any of them are already tested and we missed it, we'd love to hear so we can improve the tool. Feel free to give us some feedback on whether this review was useful from your perspective. Thanks!


https://github.com/llvm/llvm-project/pull/218955