dialect: (riscv) add folding patterns for shift ops

[osmanyasar05]

This PR adds constant folding patterns to shift operations Slli and Srli.

[osmanyasar05]

I think there are multiple ways that we can add this function. We can either make py_operation a static method and pass both operands as arguments, or make it an instance method and only pass the register operand, not the immediate. I followed #5497 here. What do you think @superlopuh ?

[codecov]

Codecov Report

❌ Patch coverage is 94.44444% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 86.23%. Comparing base (22285d6) to head (ee82909).
⚠️ Report is 5 commits behind head on main.

Files with missing lines	Patch %	Lines
xdsl/transforms/canonicalization_patterns/riscv.py	87.50%	0 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #5656      +/-   ##
==========================================
- Coverage   86.23%   86.23%   -0.01%     
==========================================
  Files         403      403              
  Lines       57216    57209       -7     
  Branches     6647     6646       -1     
==========================================
- Hits        49342    49335       -7     
  Misses       6319     6319              
  Partials     1555     1555              

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[osmanyasar05]

added new test cases to ensure correctness of srai and slli with signed numbers. they work as intended.

(srai -1, 1) = -1
(srli -1,1) = 2147483647

[superlopuh]

This is looking great! Can you please add tests for the new ops, and probably also new tests for the ops that already had the pattern to show the difference in behaviour between the shifts that respect the sign bits and not?

[osmanyasar05]

This is looking great! Can you please add tests for the new ops, and probably also new tests for the ops that already had the pattern to show the difference in behaviour between the shifts that respect the sign bits and not?

 %srli_neg = riscv.srli %c_neg1, 1 : (!riscv.reg) -> !riscv.reg<a0>
  "test.op"(%srli_neg) : (!riscv.reg<a0>) -> ()

  %srai_neg = riscv.srai %c_neg1, 1 : (!riscv.reg) -> !riscv.reg<a0>
  "test.op"(%srai_neg) : (!riscv.reg<a0>) -> ()

  %slli_neg = riscv.slli %c_neg1, 1 : (!riscv.reg) -> !riscv.reg<a0>
  "test.op"(%slli_neg) : (!riscv.reg<a0>) -> ()

I have added these tests to canonicalize.mlir. I believe these should cover what you requested? Logical and arithmetic shifting of -1.

[superlopuh]

let's normalize the tests here? I mean it would be good to have tests for all the shift ops with the same positive and negative inputs to show what the results should be

[superlopuh]

One more idea, can you please use py_operation in the interpreter implementation? For now, we would need to construct the IntegerAttr from the int passed in, but I'd like to migrate the interpreter to use attributes instead of arbitrary Python objects so it shouldn't have overhead in the future