Split non commuting pass is added to Catalyst in MLIR

[rniczh]

Context:

Description of the Change:

Adds the split-non-commuting pass, which splits quantum functions that measure non-commuting observables into multiple execution groups (In this inital implemenation, we treat each expval as different group). Each group runs in a separate call from the main function.

Steps:

1. We first run the split-to-single-terms pass to decompose Hamiltonian expval values into single-term observables before this pass runs.
2. Then we do the transformation to construct groups for each expval.

Before:

func.func @circuit() -> (f64, f64, f64) attributes {qnode} {
  %c100 = arith.constant 100 : i64
  quantum.device shots(%c100) ["", "", ""]
  %0 = quantum.alloc(3) : !quantum.reg
  ...
  %ev0 = quantum.expval %obs_X {group = 0} : f64
  %ev1 = quantum.expval %obs_Y {group = 1} : f64
  %ev2 = quantum.expval %obs_Z {group = 2} : f64
  return %ev0, %ev1, %ev2 : f64, f64, f64
}

After:

func.func @circuit() -> (f64, f64, f64) {
  %r0 = call @circuit.group.0() : () -> f64
  %r1 = call @circuit.group.1() : () -> f64
  %r2 = call @circuit.group.2() : () -> f64
  return %r0, %r1, %r2 : f64, f64, f64
}

func.func private @circuit.group.0() -> f64 attributes {qnode} {
  %c100 = arith.constant 100 : i64
  %c3 = arith.constant 3 : i64
  %divided = arith.divsi %c100, %c3 : i64
  quantum.device shots(%divided) ["", "", ""]
  ...
}

// ... group 1 ...
// ... group 2 ...

Shots are distributed among groups using integer division (rounded down); e.g., 100 shots with 3 groups yields 33 shots per group

Benefits:

Possible Drawbacks:

Related GitHub Issues:
[sc-93667]

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 96.87%. Comparing base (06b405a) to head (91dcf84).
⚠️ Report is 2 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2437      +/-   ##
==========================================
+ Coverage   96.76%   96.87%   +0.10%     
==========================================
  Files         156      155       -1     
  Lines       17030    16838     -192     
  Branches     1664     1623      -41     
==========================================
- Hits        16479    16311     -168     
+ Misses        403      392      -11     
+ Partials      148      135      -13     

☔ 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.

[sengthai]

Well done, well written! I have a few comments below.

[dime10]

Technically this is for the other pass, but split-to-single-terms seems to rename the qnode to <original name>.quantum, which is maybe not the most descriptive? Maybe something like <original name>.single_terms would describe the action better, and is less likely to conflict with other passes.
This pass's <original name>.group.<n> is pretty clear I think :)

[rniczh]

Maybe something like .single_terms would describe the action better

Thanks @dime10 ! That makes sense. I do this in this PR, because the goal of this PR is to combine both split-to-single-terms and split-non-commuting and unblock the XAS execution for FTQC team. 93df371

[dime10]

Thanks @rniczh 🚀

[lillian542]

Thanks @rniczh! Looks good, everything seems to behave as expected :) I've made a couple of minor suggestions to the docs and tests, but after that no further comments from me!

[lillian542]

Thanks @rniczh, this looks great 🚀