[mlir][llvm] Add extra attributes to the atomic ops.

[gysit]

Summary

The revision adds a number of extra arguments to the
atomic read modify write and compare and exchange
operations. The extra arguments include the volatile,
weak, syncscope, and alignment attributes.

The implementation also adapts the fence operation to use
a assembly format and generalizes the helper used
to obtain the syncscope name.

[Dinistro]

mlir/lib/Target/LLVMIR/ModuleImport.cpp

107

Might it make sense to change the return type to std::optional<StringRef> to make it clear that some instructions do not have a syncscope?

[gysit]

mlir/lib/Target/LLVMIR/ModuleImport.cpp

107

I originally implemented a variant of the method that returns an optional StringRef. However, LLVM uses the empty string to model system level synchronization. I thus decided to consistently use the empty string as "no syncscope" value which is also inline with the already existing fence operation implementation.

I can add a comment that clarifies that the default value of the syncscope name is the empty string and that we do not want to print this default value.

[gysit]

Rebase and add comment addressing the review concern.

[Dinistro]

Thanks for clarifying the syncscope handling. LGTM!