[llvm] Reland [llvm] Add support for llvm IR atomicrmw fminimum/fmaximum instructions (PR #137701)

[Kazu Hirata via llvm-commits]

kazutakahirata wrote:

@jthackray Thank you for taking a look.

I'm using:

```
    cmake -G Ninja -B "$1" \
      -DLLVM_ENABLE_PROJECTS=flang \
      -DCMAKE_BUILD_TYPE=Release \
      -DCMAKE_C_COMPILER=clang \
      -DCMAKE_CXX_COMPILER=clang++ \
      -DLLVM_USE_LINKER=lld \
      -DLLVM_PARALLEL_LINK_JOBS=4 \
      -DCMAKE_EXPORT_COMPILE_COMMANDS=ON \
      -DCMAKE_C_COMPILER_LAUNCHER=ccache \
      -DCMAKE_CXX_COMPILER_LAUNCHER=ccache \
      -DLLVM_ENABLE_WERROR=On \
      -DLLVM_ENABLE_ASSERTIONS=On \
      $LLVM_PROJECT/llvm

```

and:

```
ninja bin/flang
```

with the host compiler being clang-16.

With this setup, I get:

```
tools/mlir/include/mlir/Dialect/LLVMIR/LLVMConversionEnumsFromLLVM.inc:11:11: error: enumeration values 'FMaximum' and 'FMinimum' not handled in switch [-Werror,-Wswitch]
   11 |   switch (value) {
      |           ^~~~~
1 error generated.
```

I think one relevant difference between your cmake invocation and mine is that I have:

```
-DLLVM_ENABLE_WERROR=On
```

instead of:

```
-DFLANG_ENABLE_WERROR=ON
```


https://github.com/llvm/llvm-project/pull/137701