From b31ffa86b5a6a501e9cc092b1e1d661687243ba7 Mon Sep 17 00:00:00 2001
From: Hong-Sheng Zheng <mathan0203@gmail.com>
Date: Tue, 30 Sep 2025 09:37:55 -0400
Subject: [PATCH 01/21] Add support of MCM for decompose pass

---
 mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp b/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
index 9dcc4ea1ad..ebaf06b4d5 100644
--- a/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
+++ b/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
@@ -346,6 +346,9 @@ class OpSignatureAnalyzer {
                 }
                 qubit = customOp.getQubitOperands()[0];
             }
+            else if (auto measureOp = dyn_cast_or_null<quantum::MeasureOp>(qubit.getDefiningOp())) {
+                qubit = measureOp.getInQubit();
+            }
         }
 
         return nullptr;
@@ -377,6 +380,9 @@ class OpSignatureAnalyzer {
                     }
                 }
             }
+            else if (auto measureOp = dyn_cast_or_null<quantum::MeasureOp>(qubit.getDefiningOp())) {
+                qubit = measureOp.getInQubit();
+            }
 
             break;
         }

From 407f16fe392cc449164060ec1039e4ea6bff2a38 Mon Sep 17 00:00:00 2001
From: Hong-Sheng Zheng <mathan0203@gmail.com>
Date: Tue, 30 Sep 2025 10:44:12 -0400
Subject: [PATCH 02/21] Add test

---
 mlir/test/Quantum/DecomposeLoweringTest.mlir | 46 ++++++++++++++++++++
 1 file changed, 46 insertions(+)

diff --git a/mlir/test/Quantum/DecomposeLoweringTest.mlir b/mlir/test/Quantum/DecomposeLoweringTest.mlir
index 91bfbe7778..2a5b24eb54 100644
--- a/mlir/test/Quantum/DecomposeLoweringTest.mlir
+++ b/mlir/test/Quantum/DecomposeLoweringTest.mlir
@@ -508,3 +508,49 @@ module @cnot_alternative_decomposition {
     return %out_qubits_2#0, %out_qubits_4 : !quantum.bit, !quantum.bit
   }
 }
+
+// -----
+
+module @mcm_example {
+  func.func public @test_mcm_hadamard() -> tensor<2xf64> {
+    %0 = quantum.alloc( 1) : !quantum.reg
+    %1 = quantum.extract %0[ 0] : !quantum.reg -> !quantum.bit
+    %mres, %out_qubit = quantum.measure %1 : i1, !quantum.bit
+    %2 = quantum.insert %0[ 0], %out_qubit : !quantum.reg, !quantum.bit
+
+    // CHECK: [[RZ_QUBIT:%.+]] = quantum.custom "RZ"([[CST_0:%.+]])
+    // CHECK: [[RY_QUBIT:%.+]] = quantum.custom "RY"([[CST_1:%.+]]) [[RZ_QUBIT]] : !quantum.bit
+    // CHECK: [[REG_1:%.+]] = quantum.insert [[REG:%.+]][[[EXTRACTED:%.+]]], [[RY_QUBIT]] : !quantum.reg, !quantum.bit
+    // CHECK-NOT: quantum.custom "Hadamard"
+    %3 = quantum.extract %2[ 0] : !quantum.reg -> !quantum.bit
+    %out_qubits = quantum.custom "Hadamard"() %3 : !quantum.bit
+    %4 = quantum.insert %2[ 0], %out_qubits : !quantum.reg, !quantum.bit
+
+    %5 = quantum.compbasis qreg %4 : !quantum.obs
+    %6 = quantum.probs %5 : tensor<2xf64>
+    quantum.dealloc %4 : !quantum.reg
+    return %6 : tensor<2xf64>
+  }
+
+  // Decomposition function should be applied and removed from the module
+  // CHECK-NOT: func.func public @rz_ry
+  func.func public @rz_ry(%arg0: !quantum.reg, %arg1: tensor<1xi64>) -> !quantum.reg attributes {llvm.linkage = #llvm.linkage<internal>, num_wires = 1 : i64, target_gate = "Hadamard"} {
+    %cst = arith.constant 3.1415926535897931 : f64
+    %cst_0 = arith.constant 1.5707963267948966 : f64
+    %0 = stablehlo.slice %arg1 [0:1] : (tensor<1xi64>) -> tensor<1xi64>
+    %1 = stablehlo.reshape %0 : (tensor<1xi64>) -> tensor<i64>
+    %extracted = tensor.extract %1[] : tensor<i64>
+    %2 = quantum.extract %arg0[%extracted] : !quantum.reg -> !quantum.bit
+    %out_qubits = quantum.custom "RZ"(%cst_0) %2 : !quantum.bit
+    %3 = stablehlo.slice %arg1 [0:1] : (tensor<1xi64>) -> tensor<1xi64>
+    %4 = stablehlo.reshape %3 : (tensor<1xi64>) -> tensor<i64>
+    %extracted_1 = tensor.extract %1[] : tensor<i64>
+    %5 = quantum.insert %arg0[%extracted_1], %out_qubits : !quantum.reg, !quantum.bit
+    %extracted_2 = tensor.extract %4[] : tensor<i64>
+    %6 = quantum.extract %5[%extracted_2] : !quantum.reg -> !quantum.bit
+    %out_qubits_3 = quantum.custom "RY"(%cst) %6 : !quantum.bit
+    %extracted_4 = tensor.extract %4[] : tensor<i64>
+    %7 = quantum.insert %5[%extracted_4], %out_qubits_3 : !quantum.reg, !quantum.bit
+    return %7 : !quantum.reg
+  }
+}

From ee3b0c2ad05e3bb18de8d00406e615ff983b988a Mon Sep 17 00:00:00 2001
From: Hong-Sheng Zheng <mathan0203@gmail.com>
Date: Tue, 30 Sep 2025 11:32:42 -0400
Subject: [PATCH 03/21] Fix

---
 mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp | 1 +
 1 file changed, 1 insertion(+)

diff --git a/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp b/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
index ebaf06b4d5..2b4bbce295 100644
--- a/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
+++ b/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
@@ -382,6 +382,7 @@ class OpSignatureAnalyzer {
             }
             else if (auto measureOp = dyn_cast_or_null<quantum::MeasureOp>(qubit.getDefiningOp())) {
                 qubit = measureOp.getInQubit();
+                continue;
             }
 
             break;

From d01ef10b237b8e689de87ec478831c16bd5c4064 Mon Sep 17 00:00:00 2001
From: Hong-Sheng Zheng <mathan0203@gmail.com>
Date: Fri, 3 Oct 2025 14:17:02 -0400
Subject: [PATCH 04/21] refactoring

---
 .../Transforms/DecomposeLoweringPatterns.cpp  | 133 +++++++-----------
 1 file changed, 54 insertions(+), 79 deletions(-)

diff --git a/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp b/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
index 2b4bbce295..db13074261 100644
--- a/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
+++ b/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
@@ -12,6 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+#include "Quantum/IR/QuantumInterfaces.h"
 #define DEBUG_TYPE "decompose-lowering"
 
 #include <variant>
@@ -40,6 +41,7 @@ namespace quantum {
 /// - A runtime Value (for dynamic indices computed at runtime)
 /// - An IntegerAttr (for compile-time constant indices)
 /// - Invalid/uninitialized (represented by std::monostate)
+/// And a qreg value to represent the qreg that the index belongs to
 ///
 /// The struct uses std::variant to ensure only one type is active at a time,
 /// preventing invalid states.
@@ -54,47 +56,44 @@ namespace quantum {
 ///       Value idx = dynamicIdx.getValue();   // Get the Value
 ///     }
 ///   }
-struct QubitIndex {
+class QubitIndex {
+  private:
     // use monostate to represent the invalid index
     std::variant<std::monostate, Value, IntegerAttr> index;
+    Value qreg;
 
-    QubitIndex() : index(std::monostate()) {}
-    QubitIndex(Value val) : index(val) {}
-    QubitIndex(IntegerAttr attr) : index(attr) {}
+  public:
+    QubitIndex() : index(std::monostate()), qreg(nullptr) {}
+    QubitIndex(Value val, Value qreg) : index(val), qreg(qreg) {}
+    QubitIndex(IntegerAttr attr, Value qreg) : index(attr), qreg(qreg) {}
 
     bool isValue() const { return std::holds_alternative<Value>(index); }
     bool isAttr() const { return std::holds_alternative<IntegerAttr>(index); }
     operator bool() const { return isValue() || isAttr(); }
+    Value getReg() const { return qreg; }
     Value getValue() const { return isValue() ? std::get<Value>(index) : nullptr; }
     IntegerAttr getAttr() const { return isAttr() ? std::get<IntegerAttr>(index) : nullptr; }
 };
 
-// The goal of this class is to analyze the signature of a custom operation to get the enough
+// The goal of this class is to analyze the signature of a quantum gate operation to get the enough
 // information to prepare the call operands and results for replacing the op to calling the
 // decomposition function.
+// GateType can be QuantumGate or ParametrizedGate
 class OpSignatureAnalyzer {
   public:
     OpSignatureAnalyzer() = delete;
     OpSignatureAnalyzer(CustomOp op, bool enableQregMode)
-        : signature(OpSignature{
-              .params = op.getParams(),
-              .inQubits = op.getInQubits(),
-              .inCtrlQubits = op.getInCtrlQubits(),
-              .inCtrlValues = op.getInCtrlValues(),
-              .outQubits = op.getOutQubits(),
-              .outCtrlQubits = op.getOutCtrlQubits(),
-          })
+        : signature(
+              OpSignature{.params = op.getParams(),
+                          .inQubits = op.getNonCtrlQubitOperands(),
+                          .inCtrlQubits = op.getCtrlQubitOperands(),
+                          .inCtrlValues = op.getCtrlValueOperands(),
+                          .outQubits = op.getNonCtrlQubitResults(),
+                          .outCtrlQubits = op.getCtrlQubitResults()})
     {
         if (!enableQregMode)
             return;
 
-        signature.sourceQreg = getSourceQreg(signature.inQubits.front());
-        if (!signature.sourceQreg) {
-            op.emitError("Cannot get source qreg");
-            isValid = false;
-            return;
-        }
-
         // input wire indices
         for (Value qubit : signature.inQubits) {
             const QubitIndex index = getExtractIndex(qubit);
@@ -117,6 +116,9 @@ class OpSignatureAnalyzer {
             signature.inCtrlWireIndices.emplace_back(index);
         }
 
+        assert((signature.inWireIndices.size() + signature.inCtrlWireIndices.size()) > 0 &&
+               "inWireIndices or inCtrlWireIndices should not be empty");
+
         // Output qubit indices are the same as input qubit indices
         signature.outQubitIndices = signature.inWireIndices;
         signature.outCtrlQubitIndices = signature.inCtrlWireIndices;
@@ -124,6 +126,13 @@ class OpSignatureAnalyzer {
 
     operator bool() const { return isValid; }
 
+    Value getUpdatedQreg(PatternRewriter &rewriter, Location loc)
+    {
+        // FIXME: This will cause an issue when the decomposition function has cross-qreg
+        // inputs and outputs. Now, we just assume has only one qreg input, the global one exists.
+        return signature.inWireIndices[0].getReg();
+    }
+
     // Prepare the operands for calling the decomposition function
     // There are two cases:
     // 1. The first input is a qreg, which means the decomposition function is a qreg mode function
@@ -144,15 +153,8 @@ class OpSignatureAnalyzer {
 
         int operandIdx = 0;
         if (isa<quantum::QuregType>(funcInputs[0])) {
-            Value updatedQreg = signature.sourceQreg;
-            for (auto [i, qubit] : llvm::enumerate(signature.inQubits)) {
-                const QubitIndex &index = signature.inWireIndices[i];
-                updatedQreg =
-                    rewriter.create<quantum::InsertOp>(loc, updatedQreg.getType(), updatedQreg,
-                                                       index.getValue(), index.getAttr(), qubit);
-            }
+            operands[operandIdx++] = getUpdatedQreg(rewriter, loc);
 
-            operands[operandIdx++] = updatedQreg;
             if (!signature.params.empty()) {
                 auto [startIdx, endIdx] =
                     findParamTypeRange(funcInputs, signature.params.size(), operandIdx);
@@ -163,16 +165,11 @@ class OpSignatureAnalyzer {
                 }
             }
 
-            if (!signature.inWireIndices.empty()) {
-                operands[operandIdx] = fromTensorOrAsIs(signature.inWireIndices,
-                                                        funcInputs[operandIdx], rewriter, loc);
-                operandIdx++;
-            }
-
-            if (!signature.inCtrlWireIndices.empty()) {
-                operands[operandIdx] = fromTensorOrAsIs(signature.inCtrlWireIndices,
-                                                        funcInputs[operandIdx], rewriter, loc);
-                operandIdx++;
+            for (const auto &indices : {signature.inWireIndices, signature.inCtrlWireIndices}) {
+                if (!indices.empty()) {
+                    operands[operandIdx++] =
+                        fromTensorOrAsIs(indices, funcInputs[operandIdx], rewriter, loc);
+                }
             }
         }
         else {
@@ -218,18 +215,16 @@ class OpSignatureAnalyzer {
 
         SmallVector<Value> newResults;
         rewriter.setInsertionPointAfter(callOp);
-        for (const QubitIndex &index : signature.outQubitIndices) {
-            auto extractOp = rewriter.create<quantum::ExtractOp>(
-                callOp.getLoc(), rewriter.getType<quantum::QubitType>(), qreg, index.getValue(),
-                index.getAttr());
-            newResults.emplace_back(extractOp.getResult());
-        }
-        for (const QubitIndex &index : signature.outCtrlQubitIndices) {
-            auto extractOp = rewriter.create<quantum::ExtractOp>(
-                callOp.getLoc(), rewriter.getType<quantum::QubitType>(), qreg, index.getValue(),
-                index.getAttr());
-            newResults.emplace_back(extractOp.getResult());
+
+        for (const auto &indices : {signature.outQubitIndices, signature.outCtrlQubitIndices}) {
+            for (const auto &index : indices) {
+                auto extractOp = rewriter.create<quantum::ExtractOp>(
+                    callOp.getLoc(), rewriter.getType<quantum::QubitType>(), qreg, index.getValue(),
+                    index.getAttr());
+                newResults.emplace_back(extractOp.getResult());
+            }
         }
+
         return newResults;
     }
 
@@ -245,7 +240,6 @@ class OpSignatureAnalyzer {
         ValueRange outCtrlQubits;
 
         // Qreg mode specific information
-        Value sourceQreg = nullptr;
         SmallVector<QubitIndex> inWireIndices;
         SmallVector<QubitIndex> inCtrlWireIndices;
         SmallVector<QubitIndex> outQubitIndices;
@@ -333,42 +327,21 @@ class OpSignatureAnalyzer {
         return values.front();
     }
 
-    Value getSourceQreg(Value qubit)
-    {
-        while (qubit) {
-            if (auto extractOp = qubit.getDefiningOp<quantum::ExtractOp>()) {
-                return extractOp.getQreg();
-            }
-
-            if (auto customOp = dyn_cast_or_null<quantum::CustomOp>(qubit.getDefiningOp())) {
-                if (customOp.getQubitOperands().empty()) {
-                    break;
-                }
-                qubit = customOp.getQubitOperands()[0];
-            }
-            else if (auto measureOp = dyn_cast_or_null<quantum::MeasureOp>(qubit.getDefiningOp())) {
-                qubit = measureOp.getInQubit();
-            }
-        }
-
-        return nullptr;
-    }
-
     QubitIndex getExtractIndex(Value qubit)
     {
         while (qubit) {
             if (auto extractOp = qubit.getDefiningOp<quantum::ExtractOp>()) {
                 if (Value idx = extractOp.getIdx()) {
-                    return QubitIndex(idx);
+                    return QubitIndex(idx, extractOp.getQreg());
                 }
                 if (IntegerAttr idxAttr = extractOp.getIdxAttrAttr()) {
-                    return QubitIndex(idxAttr);
+                    return QubitIndex(idxAttr, extractOp.getQreg());
                 }
             }
 
-            if (auto customOp = dyn_cast_or_null<quantum::CustomOp>(qubit.getDefiningOp())) {
-                auto qubitOperands = customOp.getQubitOperands();
-                auto qubitResults = customOp.getQubitResults();
+            if (auto gate = dyn_cast_or_null<quantum::QuantumGate>(qubit.getDefiningOp())) {
+                auto qubitOperands = gate.getQubitOperands();
+                auto qubitResults = gate.getQubitResults();
                 auto it =
                     llvm::find_if(qubitResults, [&](Value result) { return result == qubit; });
 
@@ -401,7 +374,8 @@ struct DecomposeLoweringRewritePattern : public OpRewritePattern<CustomOp> {
     DecomposeLoweringRewritePattern(MLIRContext *context,
                                     const llvm::StringMap<func::FuncOp> &registry,
                                     const llvm::StringSet<llvm::MallocAllocator> &gateSet)
-        : OpRewritePattern(context), decompositionRegistry(registry), targetGateSet(gateSet)
+        : OpRewritePattern<CustomOp>(context), decompositionRegistry(registry),
+          targetGateSet(gateSet)
     {
     }
 
@@ -428,11 +402,12 @@ struct DecomposeLoweringRewritePattern : public OpRewritePattern<CustomOp> {
         assert(decompFunc.getFunctionType().getNumResults() >= 1 &&
                "Decomposition function must have at least one result");
 
+        rewriter.setInsertionPointAfter(op);
+
         auto enableQreg = isa<quantum::QuregType>(decompFunc.getFunctionType().getInput(0));
         auto analyzer = OpSignatureAnalyzer(op, enableQreg);
         assert(analyzer && "Analyzer should be valid");
 
-        rewriter.setInsertionPointAfter(op);
         auto callOperands = analyzer.prepareCallOperands(decompFunc, rewriter, op.getLoc());
         auto callOp =
             rewriter.create<func::CallOp>(op.getLoc(), decompFunc.getFunctionType().getResults(),
@@ -460,4 +435,4 @@ void populateDecomposeLoweringPatterns(RewritePatternSet &patterns,
 }
 
 } // namespace quantum
-} // namespace catalyst
+} // namespace catalyst
\ No newline at end of file

From 29debbb94c349107ebad429cd4f6c8837179bd4b Mon Sep 17 00:00:00 2001
From: Hong-Sheng Zheng <mathan0203@gmail.com>
Date: Fri, 3 Oct 2025 14:17:13 -0400
Subject: [PATCH 05/21] formatting

---
 .../Transforms/DecomposeLoweringPatterns.cpp        | 13 ++++++-------
 1 file changed, 6 insertions(+), 7 deletions(-)

diff --git a/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp b/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
index db13074261..e21b3cf4fb 100644
--- a/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
+++ b/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
@@ -83,13 +83,12 @@ class OpSignatureAnalyzer {
   public:
     OpSignatureAnalyzer() = delete;
     OpSignatureAnalyzer(CustomOp op, bool enableQregMode)
-        : signature(
-              OpSignature{.params = op.getParams(),
-                          .inQubits = op.getNonCtrlQubitOperands(),
-                          .inCtrlQubits = op.getCtrlQubitOperands(),
-                          .inCtrlValues = op.getCtrlValueOperands(),
-                          .outQubits = op.getNonCtrlQubitResults(),
-                          .outCtrlQubits = op.getCtrlQubitResults()})
+        : signature(OpSignature{.params = op.getParams(),
+                                .inQubits = op.getNonCtrlQubitOperands(),
+                                .inCtrlQubits = op.getCtrlQubitOperands(),
+                                .inCtrlValues = op.getCtrlValueOperands(),
+                                .outQubits = op.getNonCtrlQubitResults(),
+                                .outCtrlQubits = op.getCtrlQubitResults()})
     {
         if (!enableQregMode)
             return;

From 211135c40ccdc9c5262eff0d8a5f014362ba797a Mon Sep 17 00:00:00 2001
From: Hong-Sheng Zheng <mathan0203@gmail.com>
Date: Fri, 3 Oct 2025 14:24:18 -0400
Subject: [PATCH 06/21] fix

---
 mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp b/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
index e21b3cf4fb..e5ae71b824 100644
--- a/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
+++ b/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
@@ -166,8 +166,9 @@ class OpSignatureAnalyzer {
 
             for (const auto &indices : {signature.inWireIndices, signature.inCtrlWireIndices}) {
                 if (!indices.empty()) {
-                    operands[operandIdx++] =
+                    operands[operandIdx] =
                         fromTensorOrAsIs(indices, funcInputs[operandIdx], rewriter, loc);
+                    operandIdx++;
                 }
             }
         }

From ef06aa46636566ab1429995cf5f1ddc467f56820 Mon Sep 17 00:00:00 2001
From: Hong-Sheng Zheng <mathan0203@gmail.com>
Date: Fri, 3 Oct 2025 15:02:39 -0400
Subject: [PATCH 07/21] reorder include

---
 mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp b/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
index e5ae71b824..3fb388017b 100644
--- a/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
+++ b/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
@@ -12,7 +12,6 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include "Quantum/IR/QuantumInterfaces.h"
 #define DEBUG_TYPE "decompose-lowering"
 
 #include <variant>
@@ -26,6 +25,7 @@
 #include "mlir/IR/PatternMatch.h"
 #include "mlir/IR/ValueRange.h"
 
+#include "Quantum/IR/QuantumInterfaces.h"
 #include "Quantum/IR/QuantumOps.h"
 #include "Quantum/Transforms/Patterns.h"
 

From 9aa4697ab2a96924d3e72733d5efbae789df7fd9 Mon Sep 17 00:00:00 2001
From: Hong-Sheng Zheng <mathan0203@gmail.com>
Date: Fri, 3 Oct 2025 15:16:26 -0400
Subject: [PATCH 08/21] update comment

---
 mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp b/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
index 3fb388017b..7e0c9728e7 100644
--- a/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
+++ b/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
@@ -75,10 +75,9 @@ class QubitIndex {
     IntegerAttr getAttr() const { return isAttr() ? std::get<IntegerAttr>(index) : nullptr; }
 };
 
-// The goal of this class is to analyze the signature of a quantum gate operation to get the enough
+// The goal of this class is to analyze the signature of a custom operation to get the enough
 // information to prepare the call operands and results for replacing the op to calling the
 // decomposition function.
-// GateType can be QuantumGate or ParametrizedGate
 class OpSignatureAnalyzer {
   public:
     OpSignatureAnalyzer() = delete;

From 790446621ef568237378cf8c1f472f8ccd5e262f Mon Sep 17 00:00:00 2001
From: Paul <79805239+paul0403@users.noreply.github.com>
Date: Mon, 6 Oct 2025 17:07:59 -0400
Subject: [PATCH 09/21] Prepare for next development cycle (v14) (#2097)

Context: Setting up the repository for the next development cycle after
v13 release.

Description of the Change:
- Add new changelog-dev.md file for v0.14.0
- Add changelog-dev reference to release_notes.rst
- Update version in _version.py to 0.14.0-dev0
- Update rc_sync.yml for release manager name
- Update build-nightly-release-candidate.yaml's schedule for this
release

---------

Co-authored-by: Mehrdad Malek <39844030+mehrdad2m@users.noreply.github.com>
---
 .../build-nightly-release-candidate.yaml      |  8 ++++----
 .github/workflows/rc_sync.yaml                |  2 +-
 doc/dev/release_notes.rst                     |  2 ++
 doc/releases/changelog-dev.md                 | 19 +++++++++++++++++++
 frontend/catalyst/_version.py                 |  2 +-
 5 files changed, 27 insertions(+), 6 deletions(-)
 create mode 100644 doc/releases/changelog-dev.md

diff --git a/.github/workflows/build-nightly-release-candidate.yaml b/.github/workflows/build-nightly-release-candidate.yaml
index 34e5518697..c20c5216ad 100644
--- a/.github/workflows/build-nightly-release-candidate.yaml
+++ b/.github/workflows/build-nightly-release-candidate.yaml
@@ -2,14 +2,14 @@ name: Build Release Branch Nightly for TestPyPI
 
 on:
   schedule:
-    # Run from July 8th to July 13th at 02:00 UTC (10:00 PM EDT)
-    - cron: "0 2 8-13 7 *"
+    # Run from October 7th to October 12th at 02:00 UTC (10:00 PM EDT)
+    - cron: "0 2 7-12 10 *"
   workflow_dispatch:
     inputs:
       branch:
         description: 'Branch to build from'
         required: true
-        default: 'v0.12.0-rc'
+        default: 'v0.13.0-rc'
 
 jobs:
   setup:
@@ -20,7 +20,7 @@ jobs:
     steps:
     - name: Set branch
       id: set_branch
-      run: echo "branch=${{ github.event.inputs.branch || 'v0.12.0-rc' }}" >> $GITHUB_OUTPUT
+      run: echo "branch=${{ github.event.inputs.branch || 'v0.13.0-rc' }}" >> $GITHUB_OUTPUT
 
     - name: Checkout Catalyst repo release branch
       uses: actions/checkout@v4
diff --git a/.github/workflows/rc_sync.yaml b/.github/workflows/rc_sync.yaml
index 9f8c8868c5..7d3eadd1b8 100644
--- a/.github/workflows/rc_sync.yaml
+++ b/.github/workflows/rc_sync.yaml
@@ -83,4 +83,4 @@ jobs:
           pr_body: "Automatic sync from the release candidate to main during a feature freeze."
           pr_allow_empty: false
           pr_draft: false
-          pr_reviewer: "dime10,mehrdad2m"
+          pr_reviewer: "dime10,paul0403"
diff --git a/doc/dev/release_notes.rst b/doc/dev/release_notes.rst
index 201a4da558..7b6aba4e56 100644
--- a/doc/dev/release_notes.rst
+++ b/doc/dev/release_notes.rst
@@ -3,6 +3,8 @@ Release notes
 
 This page contains the release notes for Catalyst.
 
+.. mdinclude:: ../releases/changelog-dev.md
+
 .. mdinclude:: ../releases/changelog-0.13.0.md
 
 .. mdinclude:: ../releases/changelog-0.12.0.md
diff --git a/doc/releases/changelog-dev.md b/doc/releases/changelog-dev.md
new file mode 100644
index 0000000000..4bdf910e6b
--- /dev/null
+++ b/doc/releases/changelog-dev.md
@@ -0,0 +1,19 @@
+# Release 0.14.0 (development release)
+
+<h3>New features since last release</h3>
+
+<h3>Improvements 🛠</h3>
+
+<h3>Breaking changes 💔</h3>
+
+<h3>Deprecations 👋</h3>
+
+<h3>Bug fixes 🐛</h3>
+
+<h3>Internal changes ⚙️</h3>
+
+<h3>Documentation 📝</h3>
+
+<h3>Contributors ✍️</h3>
+
+This release contains contributions from (in alphabetical order):
diff --git a/frontend/catalyst/_version.py b/frontend/catalyst/_version.py
index 0a02dda8fb..e03910ca65 100644
--- a/frontend/catalyst/_version.py
+++ b/frontend/catalyst/_version.py
@@ -16,4 +16,4 @@
 Version number (major.minor.patch[-label])
 """
 
-__version__ = "0.13.0"
+__version__ = "0.14.0-dev0"

From bed0d69311a646ee3bd350e6b7bd11477352c3f2 Mon Sep 17 00:00:00 2001
From: ringo-but-quantum <github-ringo-but-quantum@xanadu.ai>
Date: Tue, 7 Oct 2025 03:45:34 +0000
Subject: [PATCH 10/21] [no ci] bump nightly version

---
 frontend/catalyst/_version.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/frontend/catalyst/_version.py b/frontend/catalyst/_version.py
index e03910ca65..9c7bfbd24c 100644
--- a/frontend/catalyst/_version.py
+++ b/frontend/catalyst/_version.py
@@ -16,4 +16,4 @@
 Version number (major.minor.patch[-label])
 """
 
-__version__ = "0.14.0-dev0"
+__version__ = "0.14.0-dev1"

From 4d065d1a33dcb52789bf28cc8b17da058918fb6f Mon Sep 17 00:00:00 2001
From: "github-actions[bot]"
 <41898282+github-actions[bot]@users.noreply.github.com>
Date: Tue, 7 Oct 2025 10:10:57 -0400
Subject: [PATCH 11/21] Daily rc sync to main (#2098)

Automatic sync from the release candidate to main during a feature
freeze.

---------

Co-authored-by: Jeffrey Kam <jeffrey.kam@xanadu.ai>
Co-authored-by: Isaac De Vlugt <34751083+isaacdevlugt@users.noreply.github.com>
Co-authored-by: Ali Asadi <10773383+maliasadi@users.noreply.github.com>
Co-authored-by: ringo-but-quantum <github-ringo-but-quantum@xanadu.ai>
Co-authored-by: Paul <79805239+paul0403@users.noreply.github.com>
Co-authored-by: GitHub Actions Bot <>
Co-authored-by: paul0403 <paulhaochen.wang@gmail.com>
---
 doc/releases/changelog-0.13.0.md                     | 1 +
 frontend/catalyst/qfunc.py                           | 4 ++++
 frontend/test/lit/test_mid_circuit_measurement.py    | 6 +++---
 frontend/test/pytest/test_mid_circuit_measurement.py | 7 ++++---
 4 files changed, 12 insertions(+), 6 deletions(-)

diff --git a/doc/releases/changelog-0.13.0.md b/doc/releases/changelog-0.13.0.md
index 39e60f1082..828aedc2b8 100644
--- a/doc/releases/changelog-0.13.0.md
+++ b/doc/releases/changelog-0.13.0.md
@@ -134,6 +134,7 @@
 
 * Catalyst now supports returning classical and MCM values with the dynamic one-shot MCM method.
   [(#2004)](https://github.com/PennyLaneAI/catalyst/pull/2004)
+  [(#2090)](https://github.com/PennyLaneAI/catalyst/pull/2090)
 
   For example, the code below will generate 10 values, with an equal probability of 42 and 43
   appearing.
diff --git a/frontend/catalyst/qfunc.py b/frontend/catalyst/qfunc.py
index b25453107e..38173c5eda 100644
--- a/frontend/catalyst/qfunc.py
+++ b/frontend/catalyst/qfunc.py
@@ -431,6 +431,10 @@ def _process_terminal_measurements(mcm_method, cpy_tape, out, snapshots, shot_ve
 
         result = jnp.squeeze(out[idx])
         max_ndim = min(len(out[idx].shape), 2)
+        if out[idx].shape[0] == 1:
+            # Adding the first axis back when the first axis in the original
+            # array is 1, since it corresponds to the shot's dimension.
+            result = jnp.expand_dims(result, axis=0)
         if result.ndim == 1 and max_ndim == 2:
             result = jnp.expand_dims(result, axis=1)
 
diff --git a/frontend/test/lit/test_mid_circuit_measurement.py b/frontend/test/lit/test_mid_circuit_measurement.py
index e0544e3322..efea172a15 100644
--- a/frontend/test/lit/test_mid_circuit_measurement.py
+++ b/frontend/test/lit/test_mid_circuit_measurement.py
@@ -41,9 +41,9 @@ def test_one_shot_with_static_argnums(N):
 
     # CHECK: func.func private @one_shot_wrapper() -> tensor<1024xf64>
     # CHECK-DAG: [[one:%.+]] = arith.constant 1 : index
-    # CHECK-DAG: [[eleven:%.+]] = arith.constant 11 : index
+    # CHECK-DAG: [[ten:%.+]] = arith.constant 10 : index
     # CHECK-DAG: [[zero:%.+]] = arith.constant 0 : index
-    # CHECK: scf.for %arg0 = [[zero]] to [[eleven]] step [[one]]
+    # CHECK: scf.for %arg0 = [[zero]] to [[ten]] step [[one]]
     # CHECK: {{%.+}} = catalyst.launch_kernel @module_circ::@circ() : () -> tensor<1024xf64>
 
     # CHECK: func.func public @circ() -> tensor<1024xf64>
@@ -53,7 +53,7 @@ def test_one_shot_with_static_argnums(N):
 
     dev = qml.device("lightning.qubit", wires=N)
 
-    @qml.set_shots(N + 1)
+    @qml.set_shots(N)
     @qml.qnode(dev, mcm_method="one-shot")
     def circ():
         return qml.probs()
diff --git a/frontend/test/pytest/test_mid_circuit_measurement.py b/frontend/test/pytest/test_mid_circuit_measurement.py
index 5fe4fd9a4a..4e46569a75 100644
--- a/frontend/test/pytest/test_mid_circuit_measurement.py
+++ b/frontend/test/pytest/test_mid_circuit_measurement.py
@@ -424,7 +424,8 @@ def circuit(x):
 class TestDynamicOneShotIntegration:
     """Integration tests for QNodes using mcm_method="one-shot"/dynamic_one_shot."""
 
-    def test_dynamic_one_shot_static_argnums(self, backend):
+    @pytest.mark.parametrize("shots", [1, 2])
+    def test_dynamic_one_shot_static_argnums(self, backend, shots):
         """
         Test static argnums is passed correctly to the one shot qnodes.
         """
@@ -433,14 +434,14 @@ def test_dynamic_one_shot_static_argnums(self, backend):
         def workflow(N):
             dev = qml.device(backend, wires=N)
 
-            @qml.set_shots(N + 1)
+            @qml.set_shots(N)
             @qml.qnode(dev, mcm_method="one-shot")
             def circ():
                 return qml.probs()
 
             return circ()
 
-        assert np.allclose(workflow(1), [1, 0])
+        assert np.allclose(workflow(shots), [1 if i == 0 else 0 for i in range(2**shots)])
 
     # pylint: disable=too-many-arguments
     @pytest.mark.parametrize(

From f93aab781da4c4bf8d2889502df2b1823b20529d Mon Sep 17 00:00:00 2001
From: Hong-Sheng Zheng <mathan0203@gmail.com>
Date: Tue, 7 Oct 2025 10:43:05 -0400
Subject: [PATCH 12/21] Add test

---
 .../from_plxpr/test_capture_integration.py    | 32 +++++++++++++++++++
 1 file changed, 32 insertions(+)

diff --git a/frontend/test/pytest/from_plxpr/test_capture_integration.py b/frontend/test/pytest/from_plxpr/test_capture_integration.py
index 3f67d2b94b..68d0ad1c71 100644
--- a/frontend/test/pytest/from_plxpr/test_capture_integration.py
+++ b/frontend/test/pytest/from_plxpr/test_capture_integration.py
@@ -1313,6 +1313,38 @@ def circuit(x: float, y: float, z: float):
 
         assert jnp.allclose(circuit(1.5, 2.5, 3.5), capture_result)
 
+    def test_transform_graph_decompose_workflow(self, backend):
+        """Test the integration for a circuit with a 'decompose' graph transform."""
+
+        # Capture enabled
+
+        qml.capture.enable()
+        qml.decomposition.enable_graph()
+
+        @qjit(target="mlir")
+        @partial(qml.transforms.decompose, gate_set=[qml.RX, qml.RY, qml.RZ])
+        @qml.qnode(qml.device(backend, wires=2))
+        def captured_circuit(x: float, y: float, z: float):
+            qml.measure(0)
+            qml.Rot(x, y, z, 0)
+            return qml.expval(qml.PauliZ(0))
+
+        capture_result = captured_circuit(1.5, 2.5, 3.5)
+
+        qml.decomposition.disable_graph()
+        qml.capture.disable()
+
+        # Capture disabled
+        @qjit
+        @partial(qml.transforms.decompose, gate_set=[qml.RX, qml.RY, qml.RZ])
+        @qml.qnode(qml.device(backend, wires=2))
+        def circuit(x: float, y: float, z: float):
+            catalyst.measure(0)
+            qml.Rot(x, y, z, 0)
+            return qml.expval(qml.PauliZ(0))
+
+        assert jnp.allclose(circuit(1.5, 2.5, 3.5), capture_result)
+
     def test_transform_map_wires_workflow(self, backend):
         """Test the integration for a circuit with a 'map_wires' transform."""
 

From d07e9f05b130cf2be0384d257cba68c60aa71160 Mon Sep 17 00:00:00 2001
From: Hong-Sheng Zheng <mathan0203@gmail.com>
Date: Tue, 7 Oct 2025 11:38:49 -0400
Subject: [PATCH 13/21] revert changes from latest main

---
 .../build-nightly-release-candidate.yaml      |  8 ++++----
 .github/workflows/rc_sync.yaml                |  4 ++--
 doc/dev/release_notes.rst                     |  2 --
 doc/releases/changelog-dev.md                 | 19 -------------------
 4 files changed, 6 insertions(+), 27 deletions(-)
 delete mode 100644 doc/releases/changelog-dev.md

diff --git a/.github/workflows/build-nightly-release-candidate.yaml b/.github/workflows/build-nightly-release-candidate.yaml
index c20c5216ad..34e5518697 100644
--- a/.github/workflows/build-nightly-release-candidate.yaml
+++ b/.github/workflows/build-nightly-release-candidate.yaml
@@ -2,14 +2,14 @@ name: Build Release Branch Nightly for TestPyPI
 
 on:
   schedule:
-    # Run from October 7th to October 12th at 02:00 UTC (10:00 PM EDT)
-    - cron: "0 2 7-12 10 *"
+    # Run from July 8th to July 13th at 02:00 UTC (10:00 PM EDT)
+    - cron: "0 2 8-13 7 *"
   workflow_dispatch:
     inputs:
       branch:
         description: 'Branch to build from'
         required: true
-        default: 'v0.13.0-rc'
+        default: 'v0.12.0-rc'
 
 jobs:
   setup:
@@ -20,7 +20,7 @@ jobs:
     steps:
     - name: Set branch
       id: set_branch
-      run: echo "branch=${{ github.event.inputs.branch || 'v0.13.0-rc' }}" >> $GITHUB_OUTPUT
+      run: echo "branch=${{ github.event.inputs.branch || 'v0.12.0-rc' }}" >> $GITHUB_OUTPUT
 
     - name: Checkout Catalyst repo release branch
       uses: actions/checkout@v4
diff --git a/.github/workflows/rc_sync.yaml b/.github/workflows/rc_sync.yaml
index 7d3eadd1b8..860bee3d5d 100644
--- a/.github/workflows/rc_sync.yaml
+++ b/.github/workflows/rc_sync.yaml
@@ -28,7 +28,7 @@ jobs:
       - uses: actions/checkout@v4
         with:
           fetch-depth: 0
-      
+
       # Import Numpy/Pybind11 because it is imported in setup.py
       - name: Install Numpy and Pybind11
         run: |
@@ -83,4 +83,4 @@ jobs:
           pr_body: "Automatic sync from the release candidate to main during a feature freeze."
           pr_allow_empty: false
           pr_draft: false
-          pr_reviewer: "dime10,paul0403"
+          pr_reviewer: "dime10,mehrdad2m"
diff --git a/doc/dev/release_notes.rst b/doc/dev/release_notes.rst
index 7b6aba4e56..201a4da558 100644
--- a/doc/dev/release_notes.rst
+++ b/doc/dev/release_notes.rst
@@ -3,8 +3,6 @@ Release notes
 
 This page contains the release notes for Catalyst.
 
-.. mdinclude:: ../releases/changelog-dev.md
-
 .. mdinclude:: ../releases/changelog-0.13.0.md
 
 .. mdinclude:: ../releases/changelog-0.12.0.md
diff --git a/doc/releases/changelog-dev.md b/doc/releases/changelog-dev.md
deleted file mode 100644
index 4bdf910e6b..0000000000
--- a/doc/releases/changelog-dev.md
+++ /dev/null
@@ -1,19 +0,0 @@
-# Release 0.14.0 (development release)
-
-<h3>New features since last release</h3>
-
-<h3>Improvements 🛠</h3>
-
-<h3>Breaking changes 💔</h3>
-
-<h3>Deprecations 👋</h3>
-
-<h3>Bug fixes 🐛</h3>
-
-<h3>Internal changes ⚙️</h3>
-
-<h3>Documentation 📝</h3>
-
-<h3>Contributors ✍️</h3>
-
-This release contains contributions from (in alphabetical order):

From 0d1fe0f4ecc6683aeb6970f5df196fc3fd7b6f4c Mon Sep 17 00:00:00 2001
From: Hong-Sheng Zheng <mathan0203@gmail.com>
Date: Tue, 7 Oct 2025 11:40:33 -0400
Subject: [PATCH 14/21] revert changes from main

---
 doc/releases/changelog-0.13.0.md | 19 +++++++++----------
 1 file changed, 9 insertions(+), 10 deletions(-)

diff --git a/doc/releases/changelog-0.13.0.md b/doc/releases/changelog-0.13.0.md
index ccc7d65588..e497540570 100644
--- a/doc/releases/changelog-0.13.0.md
+++ b/doc/releases/changelog-0.13.0.md
@@ -19,7 +19,7 @@
   [(#2002)](https://github.com/PennyLaneAI/catalyst/pull/2002)
 
   Two new functions, ``qml.allocate()`` and ``qml.deallocate()``, [have been added to
-  PennyLane](https://docs.pennylane.ai/en/stable/development/release_notes.html#release-0-43-0) 
+  PennyLane](https://docs.pennylane.ai/en/stable/development/release_notes.html#release-0-43-0)
   to support dynamic wire allocation. With Catalyst, these features can be accessed on
    ``lightning.qubit``, ``lightning.kokkos``, and ``lightning.gpu``.
 
@@ -46,13 +46,13 @@
   ```
 
   In the above program, 2 qubits are allocated during device initialization, and 1
-  additional qubit is allocated inside the circuit with ``qml.allocate(1)``. 
+  additional qubit is allocated inside the circuit with ``qml.allocate(1)``.
 
   For more information on what ``qml.allocate`` and ``qml.deallocate`` do, please consult the
   [PennyLane v0.43 release notes](https://docs.pennylane.ai/en/stable/development/release_notes.html#release-0-43-0).
 
   However, there are some notable differences between the behaviour of these features
-  with ``qjit`` versus without. For details, please see the relevant sections in the 
+  with ``qjit`` versus without. For details, please see the relevant sections in the
   [Catalyst sharp bits page](https://docs.pennylane.ai/projects/catalyst/en/stable/dev/sharp_bits.html#functionality-differences-from-pennylane).
 
 * A new quantum compilation pass that reduces the depth and count of non-Clifford Pauli product
@@ -106,7 +106,7 @@
               qml.H(wires=i)
               qml.T(wires=i)
 
-          return [measure(wires=i) for i in range(n)] 
+          return [measure(wires=i) for i in range(n)]
 
       print(ppm_specs(circuit))
   ```
@@ -116,7 +116,7 @@
   {'circuit_0': {'depth_pi8_ppr': 4, 'depth_ppm': 1, 'logical_qubits': 3, 'max_weight_pi8': 3, 'num_of_ppm': 3, 'pi8_ppr': 6}}
   ```
 
-  After performing the :func:`~.passes.to_ppr`, :func:`~.passes.commute_ppr`, and :func:`~.passes.merge_ppr_ppm`, 
+  After performing the :func:`~.passes.to_ppr`, :func:`~.passes.commute_ppr`, and :func:`~.passes.merge_ppr_ppm`,
   passes, the circuit contains a depth of four of non-Clifford PPRs (`depth_pi8_ppr`). Subsequently applying the
   :func:`~.passes.t_layer_reduction` pass will move PPRs around via commutation, resulting in a circuit with a
   smaller PPR depth of three.
@@ -141,8 +141,7 @@
   ```
 
 * Catalyst now supports returning classical and MCM values with the dynamic one-shot MCM method.
-  [(#2004)](https://github.com/PennyLaneAI/catalyst/pull/2004)
-  [(#2090)](https://github.com/PennyLaneAI/catalyst/pull/2090)
+  [(#2004)](https://github.com/PennyLaneAI/catalyst/pull/2004) [(#2090)](https://github.com/PennyLaneAI/catalyst/pull/2090)
 
   For example, the code below will generate 10 values, with an equal probability of 42 and 43
   appearing.
@@ -462,7 +461,7 @@ for example the one-shot mid circuit measurement transform.
   [(#2057)](https://github.com/PennyLaneAI/catalyst/pull/2057)
 
   This pass is part of a bottom-of-stack MBQC execution pathway, with a thin shim between the
-  PPR/PPM layer and MBQC to enable end-to-end compilation on a mocked backend.  Also, in an MBQC gate 
+  PPR/PPM layer and MBQC to enable end-to-end compilation on a mocked backend.  Also, in an MBQC gate
   set, one of the gate `RotXZX` cannot yet be executed on available backends.
 
   ```python
@@ -486,8 +485,8 @@ for example the one-shot mid circuit measurement transform.
 
 <h3>Documentation 📝</h3>
 
-* Typos were fixed and supplemental information was added to the 
-  docstrings for ``ppm_compilaion``, ``to_ppr``, ``commute_ppr``, 
+* Typos were fixed and supplemental information was added to the
+  docstrings for ``ppm_compilaion``, ``to_ppr``, ``commute_ppr``,
   ``ppr_to_ppm``, ``merge_ppr_ppm``, and ``ppm_specs``.
   [(#2050)](https://github.com/PennyLaneAI/catalyst/pull/2050)
 

From 65cdd45bd97d33b6c4acb378724790923b1fe001 Mon Sep 17 00:00:00 2001
From: Hong-Sheng Zheng <mathan0203@gmail.com>
Date: Tue, 7 Oct 2025 13:38:21 -0400
Subject: [PATCH 15/21] update test

---
 .../from_plxpr/test_capture_integration.py     | 18 ++++++++++++++----
 1 file changed, 14 insertions(+), 4 deletions(-)

diff --git a/frontend/test/pytest/from_plxpr/test_capture_integration.py b/frontend/test/pytest/from_plxpr/test_capture_integration.py
index 68d0ad1c71..7d1224b9e9 100644
--- a/frontend/test/pytest/from_plxpr/test_capture_integration.py
+++ b/frontend/test/pytest/from_plxpr/test_capture_integration.py
@@ -1325,8 +1325,13 @@ def test_transform_graph_decompose_workflow(self, backend):
         @partial(qml.transforms.decompose, gate_set=[qml.RX, qml.RY, qml.RZ])
         @qml.qnode(qml.device(backend, wires=2))
         def captured_circuit(x: float, y: float, z: float):
-            qml.measure(0)
-            qml.Rot(x, y, z, 0)
+            m = qml.measure(0)
+
+            @qml.cond(m)
+            def cond_fn():
+                qml.Rot(x, y, z, 0)
+
+            cond_fn()
             return qml.expval(qml.PauliZ(0))
 
         capture_result = captured_circuit(1.5, 2.5, 3.5)
@@ -1339,8 +1344,13 @@ def captured_circuit(x: float, y: float, z: float):
         @partial(qml.transforms.decompose, gate_set=[qml.RX, qml.RY, qml.RZ])
         @qml.qnode(qml.device(backend, wires=2))
         def circuit(x: float, y: float, z: float):
-            catalyst.measure(0)
-            qml.Rot(x, y, z, 0)
+            m = catalyst.measure(0)
+
+            @catalyst.cond(m)
+            def cond_fn():
+                qml.Rot(x, y, z, 0)
+
+            cond_fn()
             return qml.expval(qml.PauliZ(0))
 
         assert jnp.allclose(circuit(1.5, 2.5, 3.5), capture_result)

From ee232d3d5ae9810866beb733d4e840cf39e14854 Mon Sep 17 00:00:00 2001
From: Hong-Sheng Zheng <mathan0203@gmail.com>
Date: Tue, 7 Oct 2025 13:38:39 -0400
Subject: [PATCH 16/21] update changelog

---
 doc/releases/changelog-0.13.0.md | 1 +
 1 file changed, 1 insertion(+)

diff --git a/doc/releases/changelog-0.13.0.md b/doc/releases/changelog-0.13.0.md
index e497540570..695fd0be14 100644
--- a/doc/releases/changelog-0.13.0.md
+++ b/doc/releases/changelog-0.13.0.md
@@ -13,6 +13,7 @@
   [(#2091)](https://github.com/PennyLaneAI/catalyst/pull/2091)
   [(#2029)](https://github.com/PennyLaneAI/catalyst/pull/2029)
   [(#2001)](https://github.com/PennyLaneAI/catalyst/pull/2001)
+  [(#2068)](https://github.com/PennyLaneAI/catalyst/pull/2068)
 
 * Catalyst now supports dynamic wire allocation with ``qml.allocate()`` and
   ``qml.deallocate()`` when program capture is enabled.

From d2aa495d200bce9d32e7a52629a8d1152a468fcc Mon Sep 17 00:00:00 2001
From: Hong-Sheng Zheng <mathan0203@gmail.com>
Date: Tue, 7 Oct 2025 13:39:35 -0400
Subject: [PATCH 17/21] Update
 mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp

Co-authored-by: Ali Asadi <10773383+maliasadi@users.noreply.github.com>
---
 mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp b/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
index 7e0c9728e7..5ef19272c7 100644
--- a/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
+++ b/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
@@ -434,4 +434,4 @@ void populateDecomposeLoweringPatterns(RewritePatternSet &patterns,
 }
 
 } // namespace quantum
-} // namespace catalyst
\ No newline at end of file
+} // namespace catalyst

From 5c654229d2cb6162ab72cce0057dd823880d6396 Mon Sep 17 00:00:00 2001
From: Hong-Sheng Zheng <mathan0203@gmail.com>
Date: Wed, 8 Oct 2025 10:52:47 -0400
Subject: [PATCH 18/21] Update doc/releases/changelog-0.13.0.md

Co-authored-by: Ali Asadi <10773383+maliasadi@users.noreply.github.com>
---
 doc/releases/changelog-0.13.0.md | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/doc/releases/changelog-0.13.0.md b/doc/releases/changelog-0.13.0.md
index 695fd0be14..4324a7c5bd 100644
--- a/doc/releases/changelog-0.13.0.md
+++ b/doc/releases/changelog-0.13.0.md
@@ -11,9 +11,9 @@
   and a ``UserWarning`` is raised.
   [(#2099)](https://github.com/PennyLaneAI/catalyst/pull/2099)
   [(#2091)](https://github.com/PennyLaneAI/catalyst/pull/2091)
+  [(#2068)](https://github.com/PennyLaneAI/catalyst/pull/2068)
   [(#2029)](https://github.com/PennyLaneAI/catalyst/pull/2029)
   [(#2001)](https://github.com/PennyLaneAI/catalyst/pull/2001)
-  [(#2068)](https://github.com/PennyLaneAI/catalyst/pull/2068)
 
 * Catalyst now supports dynamic wire allocation with ``qml.allocate()`` and
   ``qml.deallocate()`` when program capture is enabled.

From dafca0e87ad1b9da66b0688ddd77eb106c727112 Mon Sep 17 00:00:00 2001
From: Hong-Sheng Zheng <mathan0203@gmail.com>
Date: Thu, 9 Oct 2025 13:23:38 -0400
Subject: [PATCH 19/21] Assert the decomp when reaching different qreg from an
 op

---
 .../Quantum/Transforms/DecomposeLoweringPatterns.cpp   | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)

diff --git a/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp b/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
index 5ef19272c7..cbc586d51d 100644
--- a/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
+++ b/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
@@ -128,7 +128,15 @@ class OpSignatureAnalyzer {
     {
         // FIXME: This will cause an issue when the decomposition function has cross-qreg
         // inputs and outputs. Now, we just assume has only one qreg input, the global one exists.
-        return signature.inWireIndices[0].getReg();
+        // raise an error if the qreg is not the same
+        Value qreg = signature.inWireIndices[0].getReg();
+
+        bool sameQreg = llvm::any_of(
+            llvm::concat<QubitIndex>(signature.inWireIndices, signature.inCtrlWireIndices),
+            [qreg](const auto &index) { return index.getReg() != qreg; });
+
+        assert(sameQreg && "The qreg of the input wires should be the same");
+        return qreg;
     }
 
     // Prepare the operands for calling the decomposition function

From 1ad5c8b94ac9115a64d645f4e504da6f22151702 Mon Sep 17 00:00:00 2001
From: Hong-Sheng Zheng <mathan0203@gmail.com>
Date: Thu, 9 Oct 2025 13:29:55 -0400
Subject: [PATCH 20/21] reorder the PR in changelog

---
 doc/releases/changelog-0.13.0.md | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

diff --git a/doc/releases/changelog-0.13.0.md b/doc/releases/changelog-0.13.0.md
index eec37eb429..66dd138ac2 100644
--- a/doc/releases/changelog-0.13.0.md
+++ b/doc/releases/changelog-0.13.0.md
@@ -9,11 +9,12 @@
   Similar to PennyLane's behaviour, this experimental feature will fall back to the old system
   whenever the graph cannot find decomposition rules for all unsupported operators in the program,
   and a ``UserWarning`` is raised.
-  [(#2099)](https://github.com/PennyLaneAI/catalyst/pull/2099)
-  [(#2091)](https://github.com/PennyLaneAI/catalyst/pull/2091)
-  [(#2068)](https://github.com/PennyLaneAI/catalyst/pull/2068)
-  [(#2029)](https://github.com/PennyLaneAI/catalyst/pull/2029)
   [(#2001)](https://github.com/PennyLaneAI/catalyst/pull/2001)
+  [(#2029)](https://github.com/PennyLaneAI/catalyst/pull/2029)
+  [(#2068)](https://github.com/PennyLaneAI/catalyst/pull/2068)
+  [(#2091)](https://github.com/PennyLaneAI/catalyst/pull/2091)
+  [(#2099)](https://github.com/PennyLaneAI/catalyst/pull/2099)
+
 
 * Catalyst now supports dynamic wire allocation with ``qml.allocate()`` and
   ``qml.deallocate()`` when program capture is enabled.

From e09be4149efde3a36a34c1423fdfc39dfa9f1ec4 Mon Sep 17 00:00:00 2001
From: Hong-Sheng Zheng <mathan0203@gmail.com>
Date: Thu, 9 Oct 2025 13:49:19 -0400
Subject: [PATCH 21/21] fix

---
 .../Quantum/Transforms/DecomposeLoweringPatterns.cpp   | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp b/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
index cbc586d51d..7d39c1d9f1 100644
--- a/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
+++ b/mlir/lib/Quantum/Transforms/DecomposeLoweringPatterns.cpp
@@ -131,9 +131,13 @@ class OpSignatureAnalyzer {
         // raise an error if the qreg is not the same
         Value qreg = signature.inWireIndices[0].getReg();
 
-        bool sameQreg = llvm::any_of(
-            llvm::concat<QubitIndex>(signature.inWireIndices, signature.inCtrlWireIndices),
-            [qreg](const auto &index) { return index.getReg() != qreg; });
+        bool sameQreg = true;
+        for (const auto &index : signature.inWireIndices) {
+            sameQreg &= index.getReg() == qreg;
+        }
+        for (const auto &index : signature.inCtrlWireIndices) {
+            sameQreg &= index.getReg() == qreg;
+        }
 
         assert(sameQreg && "The qreg of the input wires should be the same");
         return qreg;