Adding circuit executor classes and shot-branching

qiskit_aer/backends/aer_compiler.py

@@ -465,6 +465,8 @@ def compile_circuit(circuits, basis_gates=None, optypes=None):
     "chunk_swap_buffer_qubits": (int, np.integer),
     "batched_shots_gpu": (bool, np.bool_),
     "batched_shots_gpu_max_qubits": (int, np.integer),
+    "shot_branching_enable": (bool, np.bool_),
+    "shot_branching_sampling_enable": (bool, np.bool_),
     "num_threads_per_device": (int, np.integer),
     "statevector_parallel_threshold": (int, np.integer),
     "statevector_sample_measure_opt": (int, np.integer),
@@ -488,6 +490,7 @@ def compile_circuit(circuits, basis_gates=None, optypes=None):
     "use_cuTensorNet_autotuning": (bool, np.bool_),
     "parameterizations": (list),
     "fusion_parallelization_threshold": (int, np.integer),
+    "target_gpus": (list),
 }
 
 

qiskit_aer/backends/aer_simulator.py

@@ -170,6 +170,10 @@ class AerSimulator(AerBackend):
     If AerSimulator is built with cuStateVec support, cuStateVec APIs are enabled
     by setting ``cuStateVec_enable=True``.
 
+    * ``target_gpus`` (list): List of GPU's IDs starting from 0 sets
+      the target GPUs used for the simulation.
+      If this option is not specified, all the available GPUs are used for
+      chunks/shots distribution.
 
     **Additional Backend Options**
 
@@ -287,6 +291,30 @@ class AerSimulator(AerBackend):
       threads per GPU. This parameter is used to optimize Pauli noise
       simulation with multiple-GPUs (Default: 1).
 
+    * ``shot_branching_enable`` (bool): This option enables/disables
+      applying shot-branching technique to speed up multi-shots of dynamic
+      circutis simulations or circuits simulations with noise models.
+      (Default: False).
+      Starting from single state shared with multiple shots and
+      state will be branched dynamically at runtime.
+      This option can decrease runs of shots if there will be less branches
+      than number of total shots.
+      This option is available for ``"statevector"``, ``"density_matrix"``
+      and ``"tensor_network"``.
+
+    * ``shot_branching_sampling_enable`` (bool): This option enables/disables
+      applying sampling measure if the input circuit has all the measure
+      operations at the end of the circuit. (Default: False).
+      Because measure operation branches state into 2 states, it is not
+      efficient to apply branching for measure.
+      Sampling measure improves speed to get counts for multiple-shots
+      sharing the same state.
+      Note that the counts obtained by sampling measure may not be as same as
+      the counts calculated by multiple measure operations,
+      becuase sampling measure takes only one randome number per shot.
+      This option is available for ``"statevector"``, ``"density_matrix"``
+      and ``"tensor_network"``.
+
     * ``accept_distributed_results`` (bool): This option enables storing
       results independently in each process (Default: None).
 
@@ -709,6 +737,9 @@ def _default_options(cls):
             batched_shots_gpu=False,
             batched_shots_gpu_max_qubits=16,
             num_threads_per_device=1,
+            # multi-shot branching
+            shot_branching_enable=False,
+            shot_branching_sampling_enable=False,
             # statevector options
             statevector_parallel_threshold=14,
             statevector_sample_measure_opt=10,

qiskit_aer/backends/wrappers/aer_controller_binding.hpp

@@ -182,6 +182,11 @@ void bind_aer_controller(MODULE m) {
       [](Config &config, uint_t val) {
         config.num_threads_per_device.value(val);
       });
+  // # multi-shot branching
+  aer_config.def_readwrite("shot_branching_enable",
+                           &Config::shot_branching_enable);
+  aer_config.def_readwrite("shot_branching_sampling_enable",
+                           &Config::shot_branching_sampling_enable);
   // # statevector options
   aer_config.def_readwrite("statevector_parallel_threshold",
                            &Config::statevector_parallel_threshold);
@@ -403,6 +408,10 @@ void bind_aer_controller(MODULE m) {
       [](Config &config, uint_t val) {
         config.extended_stabilizer_norm_estimation_default_samples.value(val);
       });
+  aer_config.def_property(
+      "target_gpus",
+      [](const Config &config) { return config.target_gpus.val; },
+      [](Config &config, reg_t val) { config.target_gpus.value(val); });
 
   aer_config.def(py::pickle(
       [](const AER::Config &config) {
@@ -488,12 +497,14 @@ void bind_aer_controller(MODULE m) {
             write_value(77, config.unitary_parallel_threshold),
             write_value(78, config.memory_blocking_bits),
             write_value(
-                79,
-                config.extended_stabilizer_norm_estimation_default_samples));
+                79, config.extended_stabilizer_norm_estimation_default_samples),
+            write_value(80, config.shot_branching_enable),
+            write_value(81, config.shot_branching_sampling_enable),
+            write_value(82, config.target_gpus));
       },
       [](py::tuple t) {
         AER::Config config;
-        if (t.size() != 79)
+        if (t.size() != 82)
           throw std::runtime_error("Invalid serialization format.");
 
         read_value(t, 0, config.shots);
@@ -580,6 +591,9 @@ void bind_aer_controller(MODULE m) {
         read_value(t, 78, config.memory_blocking_bits);
         read_value(t, 79,
                    config.extended_stabilizer_norm_estimation_default_samples);
+        read_value(t, 80, config.shot_branching_enable);
+        read_value(t, 81, config.shot_branching_sampling_enable);
+        read_value(t, 82, config.target_gpus);
         return config;
       }));
 }

releasenotes/notes/add_executor-a03f2d23cf6f4ca9.yaml

@@ -0,0 +1,30 @@
+---
+features:
+  - |
+    This release restructures ``State`` classes.
+    Adding circuit executor classes that runs a circuit and manages multiple
+    states for multi-shots simulations or multi-chunk simulations for large
+    number of qubits.
+    Previously ``StateChunk`` class manages multiple chunks for multi-shots or
+    multi-chunk simulations but now ``State`` class only has one state
+    and all the parallelization codes are moved to ``Executor`` classes.
+    Now all ``State`` classes are independent from parallelization.
+    Also some of the functions in ``Aer::Controller`` class are moved to
+    ``CircuitExecutor::Executor`` class.
+  - |
+    Shot-branching technique that accelerates dynamic circuits simulations
+    is implemented with restructured ``Executor`` classes.
+    Shot-branching is currently applicable to statevector density_matrix
+    and tensor_network methods.
+    Shot-branching provides dynamic distribution of multi-shots
+    by branching states when applying dynamic operations
+    (measure, reset, initialize, noises)
+    By default ``shot_branching_enable`` is disabled.
+    And by setting ``shot_branching_sampling_enable``, final measures will be
+    done by sampling measure that will speed up to get counts for multiple shots
+    sharing the same state.
+  - |
+    A new option ``target_gpus`` is added to select GPUs used for the
+    simulation. A list of target GPU's ID is passed for example
+    ``target_gpus=[0, 2]`` select 2 GPUs to be used.
+    Without this option, all the available GPUs are used.