diff --git a/ccsrc/include/simulator/vector/detail/cpu_vector_policy.h b/ccsrc/include/simulator/vector/detail/cpu_vector_policy.h
index 2b1a917d1..a824a605f 100644
--- a/ccsrc/include/simulator/vector/detail/cpu_vector_policy.h
+++ b/ccsrc/include/simulator/vector/detail/cpu_vector_policy.h
@@ -95,6 +95,8 @@ struct CPUVectorPolicyBase {
     static py_qs_data_t ExpectationOfCsr(const std::shared_ptr<sparse::CsrHdMatrix<calc_type>>& a,
                                          const std::shared_ptr<sparse::CsrHdMatrix<calc_type>>& b,
                                          const qs_data_p_t& bra, const qs_data_p_t& ket, index_t dim);
+    static VT<calc_type> GetCumulativeProbs(const qs_data_p_t& qs_p, index_t dim);
+    static VT<unsigned> LowerBound(const VT<calc_type>& cum_prob, const VT<calc_type>& sampled_probs);
     // X like operator
     // ========================================================================================================
 
diff --git a/ccsrc/include/simulator/vector/detail/gpu_vector_policy.cuh b/ccsrc/include/simulator/vector/detail/gpu_vector_policy.cuh
index 738706f24..f75bfbfa7 100644
--- a/ccsrc/include/simulator/vector/detail/gpu_vector_policy.cuh
+++ b/ccsrc/include/simulator/vector/detail/gpu_vector_policy.cuh
@@ -86,6 +86,8 @@ struct GPUVectorPolicyBase {
     static py_qs_data_t ExpectationOfCsr(const std::shared_ptr<sparse::CsrHdMatrix<calc_type>>& a,
                                          const std::shared_ptr<sparse::CsrHdMatrix<calc_type>>& b,
                                          const qs_data_p_t& bra, const qs_data_p_t& ket, index_t dim);
+    static VT<calc_type> GetCumulativeProbs(const qs_data_p_t& qs_p, index_t dim);
+    static VT<unsigned> LowerBound(const VT<calc_type>& cum_prob, const VT<calc_type>& sampled_probs);
     // X like operator
     // ========================================================================================================
 
diff --git a/ccsrc/include/simulator/vector/vector_state.tpp b/ccsrc/include/simulator/vector/vector_state.tpp
index 175a5e244..11ca716ef 100644
--- a/ccsrc/include/simulator/vector/vector_state.tpp
+++ b/ccsrc/include/simulator/vector/vector_state.tpp
@@ -1303,8 +1303,8 @@ VT<unsigned> VectorState<qs_policy_t_>::SamplingMeasurementEndingWithoutNoise(co
     auto sim = derived_t(n_qubits, static_cast<unsigned>(rng()), qs);
 
     VT<int> already_measured(this->n_qubits, 0);
-    circuit_t mea_circ;
-
+    VT<int> m_qids;
+    circuit_t other_circ;
     for (auto& g : circ) {
         if (g->GetID() == GateID::M) {
             auto m_qid = g->GetObjQubits()[0];
@@ -1312,12 +1312,28 @@ VT<unsigned> VectorState<qs_policy_t_>::SamplingMeasurementEndingWithoutNoise(co
                 throw std::runtime_error("Quantum circuit is not a measurement ending circuit.");
             }
             already_measured[m_qid] = 1;
-            mea_circ.push_back(g);
+            m_qids.push_back(m_qid);
         } else {
-            sim.ApplyGate(g, pr, false);
+            other_circ.push_back(g);
+        }
+    }
+    sim.ApplyCircuit(other_circ, pr);
+
+    auto cum_prob = qs_policy_t_::GetCumulativeProbs(sim.qs, dim);
+    VT<calc_type> sampled_probs(shots);
+    for (size_t i = 0; i < shots; ++i) {
+        sampled_probs[i] = static_cast<calc_type>(sim.rng_());
+    }
+    std::sort(sampled_probs.begin(), sampled_probs.end());
+    auto res = qs_policy_t_::LowerBound(cum_prob, sampled_probs);
+    VT<unsigned> out;
+    // Measure all qubits and filter actually measurement qubit.
+    for (auto r : res) {
+        for (auto i : m_qids) {
+            out.push_back((r >> i) & 1);
         }
     }
-    return sim.Sampling(mea_circ, pr, shots, key_map, seed);
+    return out;
 }
 }  // namespace mindquantum::sim::vector::detail
 
diff --git a/ccsrc/lib/simulator/vector/detail/cpu_common/cpu_vector_core_policy.cpp b/ccsrc/lib/simulator/vector/detail/cpu_common/cpu_vector_core_policy.cpp
index 95a0e6399..b9b81e256 100644
--- a/ccsrc/lib/simulator/vector/detail/cpu_common/cpu_vector_core_policy.cpp
+++ b/ccsrc/lib/simulator/vector/detail/cpu_common/cpu_vector_core_policy.cpp
@@ -14,12 +14,16 @@
  * limitations under the License.
  */
 #include <cstddef>
+#include <cstdlib>
+#include <execution>
 #include <limits>
+#include <numeric>
 #include <stdexcept>
 
 #include "config/details/macros.h"
 #include "config/openmp.h"
 #include "config/type_promotion.h"
+#include "core/mq_base_types.h"
 #include "core/utils.h"
 #include "math/pr/parameter_resolver.h"
 #include "simulator/utils.h"
@@ -281,6 +285,56 @@ auto CPUVectorPolicyBase<derived_, calc_type>::GroundStateOfZZs(const std::map<i
     return result;
 }
 
+template <typename T>
+void ShowVec(const std::vector<T>& a) {
+    for (auto i : a) {
+        std::cout << i << ", ";
+    }
+    std::cout << std::endl;
+}
+
+template <typename derived_, typename calc_type>
+VT<calc_type> CPUVectorPolicyBase<derived_, calc_type>::GetCumulativeProbs(const qs_data_p_t& qs_out, index_t dim) {
+    auto qs = qs_out;
+    bool will_free = false;
+    if (qs == nullptr) {
+        qs = derived_::InitState(dim);
+        will_free = true;
+    }
+    VT<calc_type> prob(dim);
+    // Can be optimized by parallel prefix sum algorithm.
+    prob[0] = qs[0].real() * qs[0].real() + qs[0].imag() * qs[0].imag();
+    for (size_t i = 1; i < dim; ++i) {
+        prob[i] = qs[i].real() * qs[i].real() + qs[i].imag() * qs[i].imag() + prob[i - 1];
+    }
+
+    prob[dim - 1] = 1.0;
+    if (will_free) {
+        free(qs);
+    }
+    return prob;
+}
+
+template <typename derived_, typename calc_type>
+VT<unsigned> CPUVectorPolicyBase<derived_, calc_type>::LowerBound(const VT<calc_type>& cum_prob,
+                                                                  const VT<calc_type>& sampled_probs) {
+    size_t samp_size = sampled_probs.size();
+    VT<unsigned> out(samp_size);
+    size_t samp_idx = 0, dist_idx = 0;
+    while (true) {
+        if (samp_idx >= samp_size) {
+            break;
+        }
+        if (sampled_probs[samp_idx] < cum_prob[dist_idx]) {
+            out[samp_idx] = dist_idx;
+            samp_idx += 1;
+        } else {
+            dist_idx += 1;
+        }
+    }
+    return out;
+}
+
 #ifdef __x86_64__
 template struct CPUVectorPolicyBase<CPUVectorPolicyAvxFloat, float>;
 template struct CPUVectorPolicyBase<CPUVectorPolicyAvxDouble, double>;
diff --git a/ccsrc/lib/simulator/vector/detail/gpu/gpu_vector_core_gate_expect.cu b/ccsrc/lib/simulator/vector/detail/gpu/gpu_vector_core_gate_expect.cu
index 3bee4e8c0..36bd0e497 100644
--- a/ccsrc/lib/simulator/vector/detail/gpu/gpu_vector_core_gate_expect.cu
+++ b/ccsrc/lib/simulator/vector/detail/gpu/gpu_vector_core_gate_expect.cu
@@ -931,7 +931,6 @@ auto GPUVectorPolicyBase<derived_, calc_type_>::ExpectDiffGivens(const qs_data_p
     auto obj_rev_high_mask = mask.obj_rev_high_mask;
     auto obj_low_mask = mask.obj_low_mask;
     auto obj_rev_low_mask = mask.obj_rev_low_mask;
-    auto obj_mask = mask.obj_mask;
     auto obj_min_mask = mask.obj_min_mask;
     auto obj_max_mask = mask.obj_max_mask;
     auto ctrl_mask = mask.ctrl_mask;
diff --git a/ccsrc/lib/simulator/vector/detail/gpu/gpu_vector_core_policy.cu b/ccsrc/lib/simulator/vector/detail/gpu/gpu_vector_core_policy.cu
index c19fa29c4..5da11cada 100644
--- a/ccsrc/lib/simulator/vector/detail/gpu/gpu_vector_core_policy.cu
+++ b/ccsrc/lib/simulator/vector/detail/gpu/gpu_vector_core_policy.cu
@@ -16,6 +16,7 @@
 #include <limits>
 #include <stdexcept>
 
+#include <thrust/host_vector.h>
 #include <thrust/transform_reduce.h>
 
 #include "config/openmp.h"
@@ -334,6 +335,54 @@ auto GPUVectorPolicyBase<derived_, calc_type_>::Copy(const qs_data_p_t& qs, inde
     return out;
 };
 
+template <typename derived_, typename calc_type>
+VT<calc_type> GPUVectorPolicyBase<derived_, calc_type>::GetCumulativeProbs(const qs_data_p_t& qs_out, index_t dim) {
+    auto qs = qs_out;
+    bool will_free = false;
+    if (qs == nullptr) {
+        qs = derived_::InitState(dim);
+        will_free = true;
+    }
+    calc_type* prob;
+    auto state = cudaMalloc((void**) &prob, sizeof(calc_type) * dim);  // NOLINT
+    if (state != cudaSuccess) {
+        throw std::runtime_error("Malloc GPU memory failed: " + std::string(cudaGetErrorName(state)) + ", "
+                                 + cudaGetErrorString(state));
+    }
+    thrust::counting_iterator<index_t> i(0);
+    thrust::for_each(
+        i, i + dim, [=] __device__(index_t i) { prob[i] = qs[i].real() * qs[i].real() + qs[i].imag() * qs[i].imag(); });
+    thrust::device_ptr<calc_type> prob_ptr(prob);
+    thrust::inclusive_scan(thrust::device, prob, prob + dim, prob);
+    if (will_free) {
+        derived::FreeState(&qs);
+    }
+    VT<calc_type> out(dim);
+    thrust::copy(prob_ptr, prob_ptr + dim, out.begin());
+    cudaFree(prob);
+    return out;
+}
+
+template <typename derived_, typename calc_type>
+VT<unsigned> GPUVectorPolicyBase<derived_, calc_type>::LowerBound(const VT<calc_type>& cum_prob,
+                                                                  const VT<calc_type>& sampled_probs) {
+    size_t samp_size = sampled_probs.size();
+    VT<unsigned> out(samp_size);
+    size_t samp_idx = 0, dist_idx = 0;
+    while (true) {
+        if (samp_idx >= samp_size) {
+            break;
+        }
+        if (sampled_probs[samp_idx] < cum_prob[dist_idx]) {
+            out[samp_idx] = dist_idx;
+            samp_idx += 1;
+        } else {
+            dist_idx += 1;
+        }
+    }
+    return out;
+}
+
 template struct GPUVectorPolicyBase<GPUVectorPolicyFloat, float>;
 template struct GPUVectorPolicyBase<GPUVectorPolicyDouble, double>;
 
diff --git a/mindquantum/core/circuit/circuit.py b/mindquantum/core/circuit/circuit.py
index d49fbddba..759e91321 100644
--- a/mindquantum/core/circuit/circuit.py
+++ b/mindquantum/core/circuit/circuit.py
@@ -835,7 +835,7 @@ def is_measure_end(self):
                 high[idx] += 1
             if isinstance(gate, mq_gates.Measure):
                 m_idx = gate.obj_qubits[0]
-                if high[m_idx] != self.all_qubits.map[m_idx]:
+                if high[m_idx] != circ.all_qubits.map[m_idx]:
                     return False
         return True
 
diff --git a/tests/st/test_io/test_svg.py b/tests/st/test_io/test_svg.py
index 3c345f45b..7316eb70a 100644
--- a/tests/st/test_io/test_svg.py
+++ b/tests/st/test_io/test_svg.py
@@ -29,7 +29,7 @@ def test_measure_svg():
     res = sim.sampling(circ, shots=100, seed=42)
     text = res.svg()._repr_svg_().split('bar')  # pylint: disable=protected-access
     text = "bar".join([text[0]] + ['"'.join(i.split('"')[1:]) for i in text[1:]])
-    len_text_exp = 9257
+    len_text_exp = 9258
     assert len(text) == len_text_exp