Fix trim UnusedYield

[ehua7365]

Fix the problem with unused yield removing the wrong arguments.
Fixes Issue #452.

The reason root cause is the following.
For example, consider the example in the Issue.

@structural
def main(n: int):
    x = 0
    for i in range(n):
        x += i
    return x

which becomes the following IR

func.func @main(n : !py.int) -> !Any {
  ^0(%main_self, %n):
  │         %x = py.constant.constant 0 : !py.int
  │         %0 = py.constant.constant 0 : !py.int
  │         %1 = py.constant.constant 1 : !py.int
  │         %2 = py.ilist.range(start=%0, stop=%n : !py.int, step=%1) : !py.IList[!py.int, !Any]
  │ %x_1, %x_2 = scf.for %i in %2 -> ~T, ~T
  │              │ iter_args(%x_3 : !py.int = %x, %x_4 : ~T = %x) {
  │              │ %x_5 = py.binop.add(%x_3 : !py.int, %i) : ~T
  │              │        scf.yield %x_5, %x_5
  │              } -> purity=False
  │              func.return %x_2
} // func.func main

In this case, each iteration of the for loop yields the the same value twice.
The old behaviour of scf.trim.UnusedYield rewrites the for loop to this:

func.func @main(n : !py.int) -> !Any {
  ^0(%main_self, %n):
  │   %x = py.constant.constant 0 : !py.int
  │   %0 = py.constant.constant 0 : !py.int
  │   %1 = py.constant.constant 1 : !py.int
  │   %2 = py.ilist.range(start=%0, stop=%n : !py.int, step=%1) : !py.IList[!py.int, !Any]
  │ %x_1 = scf.for %i in %2 -> ~T
  │        │ iter_args(%x_2 : ~T = %x) {
  │        │ %x_3 = py.binop.add(%x, %i) : ~T
  │        │        scf.yield %x_3
  │        } -> purity=False
  │        func.return %x_1
} // func.func main

It seems to be deleting the first argument it sees everywhere, without regard for whether or not it is used later on when there are other equal values which should be deleted.
This is incorrect, since the iter_arg value %x_2 is clearly not used in the arguments of the add statement for %x_3 which is what is returned. Instead %x is used, which would be equivalent to returning 0 + n-1, not using anything from the previous iterations.

Instead this should be the correct behaviour:

func.func @main(n : !py.int) -> !Any {
  ^0(%main_self, %n):
  │   %x = py.constant.constant 0 : !py.int
  │   %0 = py.constant.constant 0 : !py.int
  │   %1 = py.constant.constant 1 : !py.int
  │   %2 = py.ilist.range(start=%0, stop=%n : !py.int, step=%1) : !py.IList[!py.int, !Any]
  │ %x_1 = scf.for %i in %2 -> ~T
  │        │ iter_args(%x_2 : !py.int = %x) {
  │        │ %x_3 = py.binop.add(%x_2 : !py.int, %i) : ~T
  │        │        scf.yield %x_3
  │        } -> purity=False
  │        func.return %x_1
} // func.func main

In the above the %x_2 argument is actually used in the add statement.

This PR corrects this incorrect behaviour to correctly delete the right argument when they are both the same.

[github-actions]

PR Preview Action v1.6.2
Preview removed because the pull request was closed.
2025-07-31 15:12 UTC

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[github-actions]

☂️ Python Coverage

current status: ✅

Overall Coverage

Lines	Covered	Coverage	Threshold	Status
9995	8906	89%	0%	🟢

New Files

No new covered files...

Modified Files

File	Coverage	Status
src/kirin/dialects/scf/trim.py	97%	🟢
TOTAL	97%	🟢

updated for commit: 2f66828 by action🐍

[cduck]

The test cases look good but I'm not familiar enough with kirin internals to understand the fix. Can @weinbe58 look this over as well?

[weinbe58]

@cduck I don't think this is the correct solution here I've opened up another PR that should more cleanly fix the issue: #456