Improve cmd options

[KelvinChung2000]

There is no documentation on the -i switch so I have added a bit of doc about it.

The -i option is a bit clunky, so I have improved the pass option to allow a semi-colon-separated list to specify a pass pipeline.

Finally, the help message listing all the passes in one go is difficult to read. So, I have added a new variable to keep track of the original pass, which can better preserve the help message provided.

[rachitnigam]

The -p option supports this behavior already because you can repeat it (-p p1 -p p2) will run passes p1 and p2 in a sequence.

[KelvinChung2000]

After implementing the feature, I noticed this was possible. The documentation only implies that it runs a pass and does not mention ordering. However, the order in which each -p is applied is unclear since command-line input does not always indicate order.

[rachitnigam]

However, the order in which each -p is applied is unclear since command-line input does not always indicate order.

I'm not sure I follow. Command-line order is exactly the pass order.

[KelvinChung2000]

Not all programs necessarily take argument order, parse, and store it in input order.

And I also think that -p "p1;p2" is a lot more concise and clean compared to -p p1 -p p2

[rachitnigam]

the problem is that now the two styles can be combined and you can write: -p "p1;p2" -p "p3;p4". i would rather there be one uniform way of doing things.

most compiler drivers (llvm-opt, mlir-opt) do actually take passes like this in command-line order.

[KelvinChung2000]

I would also suggest renaming the insert to reorder and going from -i to -r since it does not insert a pass into the pipeline, as a:b requires both a and b to be in the pipeline in the first place.

[rachitnigam]

What does this change help with?

[KelvinChung2000]

It allows printing the alias pass name instead of the fully expanded pass. So the help message at "all" will be "validate,pre-opt..." instead of the long chain of passes.