Fix for save_expval truncation (#2265)

* Fix for save_expval truncation

* Fixed a bug in pauli truncation and add relevant test

* limit cuQuantum version

* fix setup.py

* add release note

---------

Co-authored-by: U-AzureAD\JUNDOI <[email protected]>

Author: gadial

.github/workflows/build.yml

@@ -77,7 +77,7 @@ jobs:
       - name: Build wheels
         env:
           CIBW_BEFORE_ALL: "sed -i -e 's/^mirrorlist/#mirrorlist/' -e 's/^#baseurl/baseurl/' -e 's/mirror.centos.org/vault.centos.org/' /etc/yum.repos.d/*.repo && yum install -y yum-utils wget && wget -q https://developer.download.nvidia.com/compute/cuda/11.8.0/local_installers/cuda-repo-rhel7-11-8-local-11.8.0_520.61.05-1.x86_64.rpm && rpm -i cuda-repo-rhel7-11-8-local-11.8.0_520.61.05-1.x86_64.rpm && yum clean all && yum -y install cuda && yum -y install openblas-devel && yum-config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel7/x86_64/cuda-rhel7.repo && yum clean all"
-          CIBW_BEFORE_BUILD : "pip cache purge && pip install nvidia-cuda-runtime-cu11 nvidia-cublas-cu11 nvidia-cusolver-cu11 nvidia-cusparse-cu11 cuquantum-cu11"
+          CIBW_BEFORE_BUILD : "pip cache purge && pip install nvidia-cuda-runtime-cu11 nvidia-cublas-cu11 nvidia-cusolver-cu11 nvidia-cusparse-cu11 'cuquantum-cu11<24.11.0'"
           CIBW_SKIP: "*-manylinux_i686 pp* cp36* cp37* cp38* cp39* cp310* cp311* *musllinux*"
           CIBW_ENVIRONMENT: QISKIT_AER_PACKAGE_NAME=qiskit-aer-gpu-cu11 QISKIT_AER_CUDA_MAJOR=11 CMAKE_VERBOSE_MAKEFILE=true AER_THRUST_BACKEND=CUDA CUDACXX=/usr/local/cuda/bin/nvcc AER_CUDA_ARCH="7.0 7.2 7.5 8.0 8.6 8.7" AER_PYTHON_CUDA_ROOT=/opt/_internal AER_CIBUILD=true
           CIBW_REPAIR_WHEEL_COMMAND: 'auditwheel repair --exclude libcudart.so.11.0 --exclude libcustatevec.so.1 --exclude libcutensornet.so.2 --exclude libcutensor.so.1 --exclude libcutensorMg.so.1 --exclude libcusolver.so.11 --exclude libcusolverMg.so.11 --exclude libcusparse.so.11 --exclude libcublas.so.11 --exclude libcublasLt.so.11 -w {dest_dir} {wheel}'
@@ -110,7 +110,7 @@ jobs:
       - name: Build wheels
         env:
           CIBW_BEFORE_ALL: "sed -i -e 's/^mirrorlist/#mirrorlist/' -e 's/^#baseurl/baseurl/' -e 's/mirror.centos.org/vault.centos.org/' /etc/yum.repos.d/*.repo && yum install -y yum-utils wget && wget -q https://developer.download.nvidia.com/compute/cuda/12.5.1/local_installers/cuda-repo-rhel8-12-5-local-12.5.1_555.42.06-1.x86_64.rpm && rpm -i cuda-repo-rhel8-12-5-local-12.5.1_555.42.06-1.x86_64.rpm && yum clean all && yum -y install cuda-toolkit-12-5 && yum -y install openblas-devel && yum clean all"
-          CIBW_BEFORE_BUILD : "pip cache purge && pip install nvidia-cuda-runtime-cu12 nvidia-nvjitlink-cu12 nvidia-cublas-cu12 nvidia-cusolver-cu12 nvidia-cusparse-cu12 cuquantum-cu12"
+          CIBW_BEFORE_BUILD : "pip cache purge && pip install nvidia-cuda-runtime-cu12 nvidia-nvjitlink-cu12 nvidia-cublas-cu12 nvidia-cusolver-cu12 nvidia-cusparse-cu12 'cuquantum-cu12<24.11.0'"
           CIBW_SKIP: "*-manylinux_i686 pp* cp36* cp37* cp38* cp39* cp310* cp311* *musllinux*"
           CIBW_ENVIRONMENT: QISKIT_AER_PACKAGE_NAME=qiskit-aer-gpu QISKIT_AER_CUDA_MAJOR=12 CMAKE_VERBOSE_MAKEFILE=true AER_THRUST_BACKEND=CUDA CUDACXX=/usr/local/cuda/bin/nvcc AER_CUDA_ARCH="7.0 7.2 7.5 8.0 8.6 8.7 8.9 9.0" AER_PYTHON_CUDA_ROOT=/opt/_internal AER_CIBUILD=true
           CIBW_REPAIR_WHEEL_COMMAND: 'auditwheel repair --exclude libcudart.so.12 --exclude libcustatevec.so.1 --exclude libcutensornet.so.2 --exclude libcutensor.so.1 --exclude libcutensorMg.so.1 --exclude libcusolver.so.11 --exclude libcusolverMg.so.11 --exclude libcusolver.so.12 --exclude libcusolverMg.so.12 --exclude libcusparse.so.12 --exclude libcublas.so.12 --exclude libcublasLt.so.12 --exclude libnvJitLink.so.12 -w {dest_dir} {wheel}'

.github/workflows/deploy.yml

@@ -144,7 +144,7 @@ jobs:
       - name: Build wheels
         env:
           CIBW_BEFORE_ALL: "sed -i -e 's/^mirrorlist/#mirrorlist/' -e 's/^#baseurl/baseurl/' -e 's/mirror.centos.org/vault.centos.org/' /etc/yum.repos.d/*.repo && yum install -y yum-utils wget && wget -q https://developer.download.nvidia.com/compute/cuda/11.8.0/local_installers/cuda-repo-rhel7-11-8-local-11.8.0_520.61.05-1.x86_64.rpm && rpm -i cuda-repo-rhel7-11-8-local-11.8.0_520.61.05-1.x86_64.rpm && yum clean all && yum -y install cuda && yum -y install openblas-devel && yum-config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel7/x86_64/cuda-rhel7.repo && yum clean all"
-          CIBW_BEFORE_BUILD : "pip install nvidia-cuda-runtime-cu11 nvidia-cublas-cu11 nvidia-cusolver-cu11 nvidia-cusparse-cu11 cuquantum-cu11"
+          CIBW_BEFORE_BUILD : "pip install nvidia-cuda-runtime-cu11 nvidia-cublas-cu11 nvidia-cusolver-cu11 nvidia-cusparse-cu11 'cuquantum-cu11<24.11.0'"
           CIBW_SKIP: "*-manylinux_i686 pp* cp36* cp37* *musllinux*"
           CIBW_TEST_SKIP: "*"
           CIBW_ENVIRONMENT: QISKIT_AER_PACKAGE_NAME=qiskit-aer-gpu-cu11 QISKIT_AER_CUDA_MAJOR=11 CMAKE_VERBOSE_MAKEFILE=true AER_THRUST_BACKEND=CUDA CUDACXX=/usr/local/cuda/bin/nvcc AER_CUDA_ARCH="7.0 7.2 7.5 8.0 8.6 8.7" AER_PYTHON_CUDA_ROOT=/opt/_internal AER_CIBUILD=true
@@ -188,7 +188,7 @@ jobs:
       - name: Build wheels
         env:
           CIBW_BEFORE_ALL: "sed -i -e 's/^mirrorlist/#mirrorlist/' -e 's/^#baseurl/baseurl/' -e 's/mirror.centos.org/vault.centos.org/' /etc/yum.repos.d/*.repo && yum install -y yum-utils wget && wget -q https://developer.download.nvidia.com/compute/cuda/12.5.1/local_installers/cuda-repo-rhel8-12-5-local-12.5.1_555.42.06-1.x86_64.rpm && rpm -i cuda-repo-rhel8-12-5-local-12.5.1_555.42.06-1.x86_64.rpm && yum clean all && yum -y install cuda-toolkit-12-5 && yum -y install openblas-devel && yum clean all"
-          CIBW_BEFORE_BUILD : "pip install nvidia-cuda-runtime-cu12 nvidia-nvjitlink-cu12 nvidia-cublas-cu12 nvidia-cusolver-cu12 nvidia-cusparse-cu12 cuquantum-cu12"
+          CIBW_BEFORE_BUILD : "pip install nvidia-cuda-runtime-cu12 nvidia-nvjitlink-cu12 nvidia-cublas-cu12 nvidia-cusolver-cu12 nvidia-cusparse-cu12 'cuquantum-cu12<24.11.0"
           CIBW_SKIP: "*-manylinux_i686 pp* cp36* cp37* *musllinux*"
           CIBW_TEST_SKIP: "*"
           CIBW_ENVIRONMENT: QISKIT_AER_PACKAGE_NAME=qiskit-aer-gpu QISKIT_AER_CUDA_MAJOR=12 CMAKE_VERBOSE_MAKEFILE=true AER_THRUST_BACKEND=CUDA CUDACXX=/usr/local/cuda/bin/nvcc AER_CUDA_ARCH="7.0 7.2 7.5 8.0 8.6 8.7 9.0" AER_PYTHON_CUDA_ROOT=/opt/_internal AER_CIBUILD=true

releasenotes/notes/fix_gpu_build-acc09ca02f2d54ea.yaml

@@ -0,0 +1,6 @@
+---
+fixes:
+  - |
+    This fix limit the version of cuQuantum temporary to solve link issue
+    of GLIBC version containing in the latest cuQuantum.
+    Make sure to unlimit the version when cuQuantum will be fixed

releasenotes/notes/fix_save_expval_truncation-2ea6b9224b22956e.yaml

@@ -0,0 +1,4 @@
+---
+fixes:
+  - |
+    Fixes a small bug preventing the expval truncation mechanism from activating correctly.

setup.py

@@ -56,7 +56,7 @@
             "nvidia-cublas-cu11>=11.11.3.6",
             "nvidia-cusolver-cu11>=11.4.1.48",
             "nvidia-cusparse-cu11>=11.7.5.86",
-            "cuquantum-cu11>=23.3.0",
+            "cuquantum-cu11>=23.3.0,<24.11.0",
         ]
         classifiers_cuda = [
             "Environment :: GPU :: NVIDIA CUDA :: 11",
@@ -68,7 +68,7 @@
             "nvidia-cublas-cu12>=12.1.3.1",
             "nvidia-cusolver-cu12>=11.4.5.107",
             "nvidia-cusparse-cu12>=12.1.0.106",
-            "cuquantum-cu12>=23.3.0",
+            "cuquantum-cu12>=23.3.0,<24.11.0",
         ]
         classifiers_cuda = [
             "Environment :: GPU :: NVIDIA CUDA :: 12",

src/framework/circuit.hpp

@@ -420,13 +420,14 @@ void Circuit::reset_metadata() {
 void Circuit::add_op_metadata(const Op &op) {
   has_conditional |= op.conditional;
   opset_.insert(op);
-  if (!qubitset_.empty() &&
-      (op.type == OpType::save_expval || op.type == OpType::save_expval_var)) {
+  if (op.type == OpType::save_expval || op.type == OpType::save_expval_var) {
     for (int_t j = 0; j < op.expval_params.size(); j++) {
       const std::string &pauli = std::get<0>(op.expval_params[j]);
       for (int_t i = 0; i < op.qubits.size(); i++) {
         // add qubit with non-I operator
-        if (pauli[pauli.size() - 1 - i] != 'I') {
+        // we also add one qubit if the qubitset is empty to prevent edge cases
+        // with empty circuits
+        if (qubitset_.empty() || pauli[pauli.size() - 1 - i] != 'I') {
           qubitset_.insert(op.qubits[i]);
         }
       }
@@ -682,17 +683,23 @@ void Circuit::set_params(bool truncation) {
 void Circuit::remap_qubits(Op &op) const {
   // truncate save_expval
   if (op.type == OpType::save_expval || op.type == OpType::save_expval_var) {
+    // map each qubit to its location in the pauli strings
+    std::unordered_map<uint_t, uint_t> ops_pauli_qubit_map;
+    for (size_t i = 0; i < op.qubits.size(); ++i) {
+      ops_pauli_qubit_map[op.qubits[i]] = op.qubits.size() - 1 - i;
+    }
     int_t nparams = op.expval_params.size();
     for (int_t i = 0; i < nparams; i++) {
       std::string &pauli = std::get<0>(op.expval_params[i]);
       std::string new_pauli;
       new_pauli.resize(qubitmap_.size());
       for (auto q = qubitmap_.cbegin(); q != qubitmap_.cend(); q++) {
         new_pauli[qubitmap_.size() - 1 - q->second] =
-            pauli[pauli.size() - 1 - q->first];
+            pauli[ops_pauli_qubit_map[q->first]];
       }
       pauli = new_pauli;
     }
+    op.qubits.resize(qubitmap_.size());
     for (int_t i = 0; i < qubitmap_.size(); i++) {
       op.qubits[i] = i;
     }

test/terra/backends/aer_simulator/test_save_expval.py

@@ -271,3 +271,21 @@ def test_save_expval_var_stabilizer_pauli_cache_blocking(self, method, device, p
             blocking_qubits=2,
             max_parallel_threads=1,
         )
+
+    @supported_methods(
+        [
+            "automatic",
+            "stabilizer",
+            "statevector",
+            "density_matrix",
+            "matrix_product_state",
+            "tensor_network",
+        ],
+    )
+    def test_save_expval_truncation(self, method, device):
+        """Test save expval when qubits are not in use and are truncated"""
+        circ = QuantumCircuit(3)
+        circ.x(1)
+        oper = qi.Pauli("ZZ")
+        qubits = [1, 2]  # qubit 0 will be truncated
+        self._test_save_expval(circ, oper, qubits, False, method=method, device=device)

test/terra/primitives/test_estimator_v2.py

@@ -445,6 +445,33 @@ def test_truncate(self):
             result_5[0].data["evs"][0], result_2[0].data["evs"][0], delta=self._rtol
         )
 
+    def test_truncate_large_backends(self):
+        """Test truncation allows us to run few-qubit circuits on many-qubit backends"""
+        N = 12
+        qc = QuantumCircuit(N)
+
+        qc.x(range(N))
+        qc.h(range(N))
+
+        for kk in range(N // 2, 0, -1):
+            qc.ch(kk, kk - 1)
+        for kk in range(N // 2, N - 1):
+            qc.ch(kk, kk + 1)
+
+        op = SparsePauliOp("Z" * N)
+        backend_127 = GenericBackendV2(num_qubits=127)
+        pm = generate_preset_pass_manager(backend=backend_127, optimization_level=1)
+        isa_circuit = pm.run(qc)
+        mapped_observable = op.apply_layout(isa_circuit.layout)
+
+        ref_est = StatevectorEstimator()
+        ref_result = ref_est.run(pubs=[(qc, op)]).result()
+
+        est = EstimatorV2()
+        res = est.run(pubs=[(isa_circuit, mapped_observable)]).result()
+
+        self.assertAlmostEqual(ref_result[0].data.evs, res[0].data.evs)
+
 
 if __name__ == "__main__":
     unittest.main()