core: (traits) Add SameOperandsAndResultType trait

[compor]

This PR:

• Adds the SameOperandsAndResultType trait
• Tests of the above

A few examples of this being tested in upstream MLIR.

[math-fehr]

Quick question. Looking at it, it seems that this is more complex than just the SameOperandsAndResultType in MLIR, right?
MLIR only checks that the operands and results are the same, but here you seem to check across containers? What does the check do exactly? I would probably rename the trait to not confuse it with the MLIR one. (if I'm correct to what MLIR does)

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 91.26%. Comparing base (dee76d4) to head (c181348).
Report is 13 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3751      +/-   ##
==========================================
+ Coverage   91.24%   91.26%   +0.02%     
==========================================
  Files         459      461       +2     
  Lines       57298    57454     +156     
  Branches     5532     5545      +13     
==========================================
+ Hits        52279    52437     +158     
+ Misses       3593     3591       -2     
  Partials     1426     1426              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[compor]

Quick question. Looking at it, it seems that this is more complex than just the SameOperandsAndResultType in MLIR, right? MLIR only checks that the operands and results are the same, but here you seem to check across containers? What does the check do exactly? I would probably rename the trait to not confuse it with the MLIR one. (if I'm correct to what MLIR does)

Is it? I thought it also checks within the container type?
https://github.com/llvm/llvm-project/blob/8d9dcd111e5ced8135387917859dd64d67886be0/mlir/lib/IR/Operation.cpp#L1100

[math-fehr]

Okay my bad then, I indeed misunderstood what MLIR was doing!

[math-fehr]

Thanks for the pointer!

[alexarice]

Does this achieve anything that you can't do with range constraints? Is it just so you can write the same thing as you can in mlir?

[math-fehr]

My feeling is that have_compatible_shape is either impossible, or very very annoying to write in xDSL. So here, range constraints are not enough AFAIK

[alexarice]

I think i've misunderstood what this does then, I'll have a proper look tomorrow

[alexarice]

Ah, I see what this does now, I feel the name is confusing but I guess that's what mlir uses

[alexarice]

My feeling is that have_compatible_shape is either impossible, or very very annoying to write in xDSL. So here, range constraints are not enough AFAIK

We should figure out how to do shapes properly with constraints/irdl at some point