"""
********************************************************************************
* Copyright (c) 2026 the Qrisp authors
*
* This program and the accompanying materials are made available under the
* terms of the Eclipse Public License 2.0 which is available at
* http://www.eclipse.org/legal/epl-2.0.
*
* This Source Code may also be made available under the following Secondary
* Licenses when the conditions for such availability set forth in the Eclipse
* Public License, v. 2.0 are satisfied: GNU General Public License, version 2
* with the GNU Classpath Exception which is
* available at https://www.gnu.org/software/classpath/license.html.
*
* SPDX-License-Identifier: EPL-2.0 OR GPL-2.0 WITH Classpath-exception-2.0
********************************************************************************
"""
from functools import lru_cache
import jax
from jax import make_jaxpr
from jax.extend.core import Jaxpr, Literal, ClosedJaxpr
from jax.tree_util import tree_flatten, tree_unflatten
from qrisp.jasp import (
eval_jaxpr,
flatten_environments,
cond_to_cl_control,
extract_invalues,
flatten_environments,
insert_outvalues,
)
from qrisp.jasp.primitives import (
AbstractQuantumState,
QuantumPrimitive,
ParityOperation,
)
from qrisp.jasp.interpreter_tools.interpreters import ProcessedMeasurement
from qrisp.jasp.primitives import AbstractQuantumState, QuantumPrimitive
from qrisp.jasp.jasp_expression import collect_environments, invert_jaspr
[docs]
class Jaspr(ClosedJaxpr):
"""
The ``Jaspr`` class enables an efficient representations of a wide variety
of (hybrid) algorithms. For many applications, the representation is agnostic
to the scale of the problem, implying function calls with 10 or 10000 qubits
can be represented by the same object. The actual unfolding to a circuit-level
description is outsourced to
`established, classical compilation infrastructure <https://mlir.llvm.org/>`_,
implying state-of-the-art compilation speed can be reached.
As a subtype of ``jax.extend.core.ClosedJaxpr``, Jasprs are embedded into the well matured
`Jax ecosystem <https://github.com/n2cholas/awesome-jax>`_,
which facilitates the compilation of classical `real-time computation <https://arxiv.org/abs/2206.12950>`_
using some of the most advanced libraries in the world such as
`CUDA <https://jax.readthedocs.io/en/latest/Custom_Operation_for_GPUs.html>`_.
Especially `machine learning <https://ai.google.dev/gemma/docs/jax_inference>`_
and other scientific computations tasks are particularly well supported.
To get a better understanding of the syntax and semantics of Jaxpr (and with
that also Jaspr) please check `this link <https://jax.readthedocs.io/en/latest/jaxpr.html>`__.
Similar to Jaxpr, Jaspr objects represent (hybrid) quantum
algorithms in the form of a `functional programming language <https://en.wikipedia.org/wiki/Functional_programming>`_
in `SSA-form <https://en.wikipedia.org/wiki/Static_single-assignment_form>`_.
It is possible to compile Jaspr objects into QIR, which is facilitated by the
`Catalyst framework <https://docs.pennylane.ai/projects/catalyst/en/stable/index.html>`__
(check :meth:`qrisp.jasp.jaspr.to_qir` for more details).
Qrisp scripts can be turned into Jaspr objects by
calling the ``make_jaspr`` function, which has similar semantics as
`jax.make_jaxpr <https://jax.readthedocs.io/en/latest/_autosummary/jax.make_jaxpr.html>`_.
::
from qrisp import *
from qrisp.jasp import make_jaspr
def test_fun(i):
qv = QuantumFloat(i, -1)
x(qv[0])
cx(qv[0], qv[i-1])
meas_res = measure(qv)
meas_res += 1
return meas_res
jaspr = make_jaspr(test_fun)(4)
print(jaspr)
This will give you the following output:
.. code-block::
{ lambda ; a:QuantumState b:i32[]. let
c:QuantumState d:QubitArray = create_qubits a b
e:Qubit = get_qubit d 0
f:QuantumState = x c e
g:i32[] = sub b 1
h:Qubit = get_qubit d g
i:QuantumState = cx f e h
j:QuantumState k:i32[] = measure i d
l:f32[] = convert_element_type[new_dtype=float64 weak_type=True] k
m:f32[] = mul l 0.5
n:f32[] = add m 1.0
in (j, n) }
A defining feature of the Jaspr class is that the first input and the
first output are always of QuantumState type. Therefore, Jaspr objects always
represent some (hybrid) quantum operation.
Qrisp comes with a built-in Jaspr interpreter. For that you simply have to
call the object like a function:
>>> print(jaspr(2))
2.5
>>> print(jaspr(4))
5.5
"""
__slots__ = (
"permeability",
"isqfree",
"hashvalue",
"ctrl_jaspr",
"inv_jaspr",
"envs_flattened",
)
def __init__(
self,
*args,
permeability=None,
isqfree=None,
ctrl_jaspr=None,
inv_jaspr=None,
**kwargs,
):
if len(args) == 2:
if not isinstance(args[0], Jaxpr) or not isinstance(args[1], list):
raise Exception(
f"Tried to call the Jaspr constructor with two arguments and signature {type(args[0]), type(args[1])} (allowed is (Jaxpr, list))"
)
kwargs["jaxpr"] = args[0]
kwargs["consts"] = args[1]
elif len(args) == 1:
if not isinstance(args[0], ClosedJaxpr):
raise Exception(
f"Tried to call the Jaspr constructor with one argument and signature {type(args[0])} (allowed is (ClosedJaxpr))"
)
kwargs["jaxpr"] = args[0].jaxpr
kwargs["consts"] = args[0].consts
if "jaxpr" in kwargs:
ClosedJaxpr.__init__(self, kwargs["jaxpr"], kwargs["consts"])
else:
if "consts" in kwargs:
consts = kwargs["consts"]
del kwargs["consts"]
else:
if len(kwargs["constvars"]):
raise Exception(
"Tried to create Jaspr with constvars but no constants"
)
consts = []
ClosedJaxpr.__init__(self, jaxpr=Jaxpr(**kwargs), consts=consts)
self.hashvalue = id(self)
self.permeability = {}
if permeability is None:
permeability = {}
for var in self.constvars + self.invars + self.outvars:
if isinstance(var, Literal):
continue
self.permeability[var] = permeability.get(var, None)
self.isqfree = isqfree
self.ctrl_jaspr = ctrl_jaspr
self.inv_jaspr = inv_jaspr
self.envs_flattened = False
if not isinstance(self.invars[-1].aval, AbstractQuantumState):
raise Exception(
f"Tried to create a Jaspr from data that doesn't have a QuantumState the last argument (got {type(self.invars[-1].aval)} instead)"
)
if not isinstance(self.outvars[-1].aval, AbstractQuantumState):
raise Exception(
f"Tried to create a Jaspr from data that doesn't have a QuantumState the last entry of return type (got {type(self.outvars[-1].aval)} instead)"
)
@property
def constvars(self):
return self.jaxpr.constvars
@property
def eqns(self):
return self.jaxpr.eqns
@property
def invars(self):
return self.jaxpr.invars
@property
def outvars(self):
return self.jaxpr.outvars
@property
def debug_info(self):
return self.jaxpr.debug_info
def __hash__(self):
return self.hashvalue
def __eq__(self, other):
if not isinstance(other, Jaxpr):
return False
return id(self) == id(other)
def copy(self):
if self.ctrl_jaspr is None:
ctrl_jaspr = None
else:
ctrl_jaspr = self.ctrl_jaspr.copy()
res = Jaspr(
permeability=self.permeability,
isqfree=self.isqfree,
ctrl_jaspr=ctrl_jaspr,
constvars=list(self.constvars),
invars=list(self.invars),
outvars=list(self.outvars),
eqns=list(self.eqns),
effects=self.effects,
debug_info=self.debug_info,
)
res.envs_flattened = self.envs_flattened
return res
[docs]
def inverse(self):
"""
Returns the inverse Jaspr (if applicable). For Jaspr that contain realtime
computations or measurements, the inverse does not exist.
Returns
-------
Jaspr
The daggered Jaspr.
Examples
--------
We create a simple script and inspect the daggered version:
::
from qrisp import *
from qrisp.jasp import make_jaspr
def example_function(i):
qv = QuantumVariable(i)
cx(qv[0], qv[1])
t(qv[1])
return qv
jaspr = make_jaspr(example_function)(2)
print(jaspr.inverse())
# Yields
# { lambda ; a:QuantumState b:i32[]. let
# c:QuantumState d:QubitArray = create_qubits a b
# e:Qubit = get_qubit d 0
# f:Qubit = get_qubit d 1
# g:QuantumState = t_dg c f
# h:QuantumState = cx g e f
# in (h, d) }
"""
return invert_jaspr(self)
[docs]
def control(self, num_ctrl, ctrl_state=-1):
"""
Returns the controlled version of the Jaspr. The control qubits are added
to the signature of the Jaspr as the arguments after the QuantumState.
Parameters
----------
num_ctrl : int
The amount of controls to be added.
ctrl_state : int of str, optional
The control state on which to activate. The default is -1.
Returns
-------
Jaspr
The controlled Jaspr.
Examples
--------
We create a simple script and inspect the controlled version:
::
from qrisp import *
from qrisp.jasp import make_jaspr
def example_function(i):
qv = QuantumVariable(i)
cx(qv[0], qv[1])
t(qv[1])
return qv
jaspr = make_jaspr(example_function)(2)
print(jaspr.control(2))
# Yields
# { lambda ; a:QuantumState b:Qubit c:Qubit d:i32[]. let
# e:QuantumState f:QubitArray = create_qubits a 1
# g:Qubit = get_qubit f 0
# h:QuantumState = ccx e b c g
# i:QuantumState j:QubitArray = create_qubits h d
# k:Qubit = get_qubit j 0
# l:Qubit = get_qubit j 1
# m:QuantumState = ccx i g k l
# n:QuantumState = ct m g l
# o:QuantumState = ccx n b c g
# in (o, j) }
We see that the control qubits are part of the function signature
(``a`` and ``b``)
"""
if self.ctrl_jaspr is not None and num_ctrl == 1 and ctrl_state == -1:
return self.ctrl_jaspr
from qrisp.jasp import ControlledJaspr
if isinstance(ctrl_state, int):
if ctrl_state < 0:
ctrl_state += 2**num_ctrl
ctrl_state = bin(ctrl_state)[2:].zfill(num_ctrl)
else:
ctrl_state = str(ctrl_state)
return ControlledJaspr.from_cache(self, ctrl_state)
[docs]
def to_qc(self, *args):
"""
Converts the Jaspr into a :ref:`QuantumCircuit` if applicable. Circuit
conversion of algorithms involving realtime computations is not possible.
Any computations that perform classical postprocessing of measurements
can not be reflected within the QuantumCircuit object itself and will
generate an object of type ``ProcessedMeasurement``. These objects hold
no further information and are simply used as placeholders to emulate
the computation.
Parameters
----------
*args : tuple
The arguments to call the Jaspr with.
Returns
-------
return_values : tuple
The return values of the Jaspr. QuantumVariable return types are
returned as lists of Qubits.
:ref:`QuantumCircuit`
The resulting QuantumCircuit.
Examples
--------
We create a simple script and inspect the QuantumCircuit:
::
from qrisp import *
from qrisp.jasp import make_jaspr
def example_function(i):
qv = QuantumVariable(i)
cx(qv[0], qv[1])
t(qv[1])
return qv
jaspr = make_jaspr(example_function)(2)
qb_list, qc = jaspr.to_qc(2)
print(qc)
# Yields
# qb_0: ──■───────
# ┌─┴─┐┌───┐
# qb_1: ┤ X ├┤ T ├
# └───┘└───┘
To demonstrate the behavior under measurement post-processing, we build
a similar script:
::
from qrisp import ProcessedMeasurement
def example_function(i):
qf = QuantumFloat(i)
cx(qf[0], qf[1])
t(qf[1])
meas_res = measure(qf)
# Perform classical post processing
meas_res *= 2
return meas_res
jaspr = make_jaspr(example_function)(2)
meas_res, qc = jaspr.to_qc(2)
print(isinstance(meas_res, ProcessedMeasurement))
# True
"""
from qrisp.jasp.interpreter_tools.interpreters import jaspr_to_qc
return jaspr_to_qc(self, *args)
def eval(self, *args, eqn_evaluator=lambda x, y: True):
return eval_jaxpr(self, eqn_evaluator=eqn_evaluator)(*args)
def flatten_environments(self):
"""
Flattens all environments by applying the corresponding compilation
routines such that no more ``q_env`` primitives are left.
Returns
-------
Jaspr
The Jaspr with flattened environments.
Examples
--------
We create a Jaspr containing an :ref:`InversionEnvironment` and flatten:
::
def test_function(i):
qv = QuantumVariable(i)
with invert():
t(qv[0])
cx(qv[0], qv[1])
return qv
jaspr = make_jaspr(test_function)(2)
print(jaspr)
::
{ lambda ; a:QuantumState b:i32[]. let
c:QuantumState d:QubitArray = create_qubits a b
e:QuantumState = q_env[
jaspr={ lambda ; f:QuantumState d:QubitArray. let
g:Qubit = get_qubit d 0
h:QuantumState = t f g
i:Qubit = get_qubit d 1
j:QuantumState = cx h g i
in (j,) }
type=InversionEnvironment
] c d
in (e, d) }
You can see how the body of the :ref:`InversionEnvironment` is __collected__
into another Jaspr. This reflects the fact that at their core,
:ref:`QuantumEnvironment <QuantumEnvironment>` describe `higher-order
quantum functions <https://en.wikipedia.org/wiki/Higher-order_function>`_
(ie. functions that operate on functions). In order to apply the
transformations induced by the QuantumEnvironment, we can call
``jaspr.flatten_environments``:
>>> print(jaspr.flatten_environments)
{ lambda ; a:QuantumState b:i32[]. let
c:QuantumState d:QubitArray = create_qubits a b
e:Qubit = get_qubit d 0
f:Qubit = get_qubit d 1
g:QuantumState = cx c e f
h:QuantumState = t_dg g e
in (h, d) }
We see that as expected, the order of the ``cx`` and the ``t`` gate has been switched and the ``t`` gate has been turned into a ``t_dg``.
"""
res = flatten_environments(self)
if self.ctrl_jaspr is not None:
res.ctrl_jaspr = self.ctrl_jaspr.flatten_environments()
if self.inv_jaspr is not None:
if not self.inv_jaspr.envs_flattened:
res.inv_jaspr = self.inv_jaspr.flatten_environments()
else:
res.inv_jaspr = self.inv_jaspr
res.inv_jaspr.inv_jaspr = res
return res
def __call__(self, *args):
from qrisp.jasp.evaluation_tools.jaspification import simulate_jaspr
return simulate_jaspr(self, *args)
if len(self.outvars) == 1:
return None
from qrisp.simulator import BufferedQuantumState
args = [BufferedQuantumState()] + list(tree_flatten(args)[0])
from qrisp.jasp import eval_jaxpr, extract_invalues, insert_outvalues
flattened_jaspr = self
def eqn_evaluator(eqn, context_dic):
if eqn.primitive.name == "jit":
if eqn.params["name"] == "expectation_value_eval_function":
from qrisp.jasp.program_control import sampling_evaluator
sampling_evaluator("ev")(
eqn, context_dic, eqn_evaluator=eqn_evaluator
)
return
if eqn.params["name"] == "sampling_eval_function":
from qrisp.jasp.program_control import sampling_evaluator
sampling_evaluator("array")(
eqn, context_dic, eqn_evaluator=eqn_evaluator
)
return
invalues = extract_invalues(eqn, context_dic)
outvalues = eval_jaxpr(
eqn.params["jaxpr"], eqn_evaluator=eqn_evaluator
)(*invalues)
if not isinstance(outvalues, (list, tuple)):
outvalues = [outvalues]
insert_outvalues(eqn, context_dic, outvalues)
elif eqn.primitive.name == "jasp.create_quantum_kernel":
insert_outvalues(eqn, context_dic, BufferedQuantumState())
else:
return True
res = eval_jaxpr(flattened_jaspr, eqn_evaluator=eqn_evaluator)(
*(args + self.consts)
)
if len(self.outvars) == 2:
return res[1]
else:
return res[1:]
def inline(self, *args):
from qrisp.jasp import TracingQuantumSession
qs = TracingQuantumSession.get_instance()
abs_qst = qs.abs_qst
ammended_args = list(args) + [abs_qst]
res = eval_jaxpr(self)(*ammended_args)
if isinstance(res, tuple):
new_abs_qst = res[-1]
res = res[:-1]
else:
new_abs_qst = res
res = None
qs.abs_qst = new_abs_qst
return res
def count_ops(self, *args, meas_behavior):
from qrisp.jasp.evaluation_tools import profile_jaspr
return profile_jaspr(self, "count_ops", meas_behavior)(*args)
def depth(self, *args, meas_behavior, max_qubits=1024):
from qrisp.jasp.evaluation_tools import profile_jaspr
return profile_jaspr(self, "depth", meas_behavior, max_qubits=max_qubits)(*args)
def num_qubits(self, *args, meas_behavior, max_allocations=1000):
from qrisp.jasp.evaluation_tools import profile_jaspr
return profile_jaspr(
self, "num_qubits", meas_behavior, max_allocations=max_allocations
)(*args)
def embedd(self, *args, name=None, inline=False):
from qrisp.jasp import TracingQuantumSession, get_last_equation
qs = TracingQuantumSession.get_instance()
abs_qst = qs.abs_qst
ammended_args = list(args) + [abs_qst]
if not inline:
res = jax.jit(eval_jaxpr(self))(*ammended_args)
eqn = get_last_equation()
eqn.params["jaxpr"] = self
if name is not None:
eqn.params["name"] = name
else:
res = eval_jaxpr(self)(*ammended_args)
if isinstance(res, tuple):
new_abs_qst = res[-1]
res = res[:-1]
else:
new_abs_qst = res
res = None
qs.abs_qst = new_abs_qst
return res
[docs]
def qjit(self, *args, function_name="jaspr_function", device=None):
"""
Leverages the Catalyst pipeline to compile a QIR representation of
this function and executes that function using the Catalyst QIR runtime.
Parameters
----------
*args : iterable
The arguments to call the function with.
device : object
The `PennyLane device <https://docs.pennylane.ai/projects/catalyst/en/stable/dev/devices.html>`_ to execute the function.
The default device is `"lightning.qubit" <https://docs.pennylane.ai/projects/lightning/en/stable/lightning_qubit/device.html>`_,
a fast state-vector qubit simulator.
Returns
-------
The values returned by the compiled, executed function.
"""
flattened_jaspr = self
from qrisp.jasp.evaluation_tools.catalyst_interface import (
jaspr_to_catalyst_qjit,
)
qjit_obj = jaspr_to_catalyst_qjit(
flattened_jaspr, function_name=function_name, device=device
)
res = qjit_obj.compiled_function(*args)
if not isinstance(res, (tuple, list)):
return res
elif len(res) == 1:
return res[0]
else:
return res
@classmethod
@lru_cache(maxsize=int(1e5))
def from_cache(cls, closed_jaxpr):
res = Jaspr(jaxpr=closed_jaxpr.jaxpr, consts=closed_jaxpr.consts)
remove_redundant_allocations(res)
return res
def update_eqns(self, eqns):
return Jaspr(
constvars=list(self.constvars),
invars=list(self.invars),
outvars=list(self.outvars),
eqns=list(eqns),
consts=list(self.consts),
debug_info=self.debug_info,
)
[docs]
def to_qir(self):
"""
Compiles the Jaspr to QIR using the `Catalyst framework <https://docs.pennylane.ai/projects/catalyst/en/stable/index.html>`__.
Parameters
----------
None
Returns
-------
str
The QIR string.
Examples
--------
We create a simple script and inspect the QIR string:
::
from qrisp import *
from qrisp.jasp import make_jaspr
def example_function(i):
qv = QuantumFloat(i)
cx(qv[0], qv[1])
t(qv[1])
meas_res = measure(qv)
meas_res += 1
return meas_res
jaspr = make_jaspr(example_function)(2)
print(jaspr.to_qir())
Yields:
.. code-block:: none
; ModuleID = 'LLVMDialectModule'
source_filename = "LLVMDialectModule"
@"{'shots': 0, 'mcmc': False, 'num_burnin': 0, 'kernel_name': None}" = internal constant [66 x i8] c"{'shots': 0, 'mcmc': False, 'num_burnin': 0, 'kernel_name': None}\00"
@lightning.qubit = internal constant [16 x i8] c"lightning.qubit\00"
@"/home/positr0nium/miniconda3/envs/qrisp/lib/python3.10/site-packages/catalyst/utils/../lib/librtd_lightning.so" = internal constant [111 x i8] c"/home/positr0nium/miniconda3/envs/qrisp/lib/python3.10/site-packages/catalyst/utils/../lib/librtd_lightning.so\00"
declare void @__catalyst__rt__finalize() local_unnamed_addr
declare void @__catalyst__rt__initialize() local_unnamed_addr
declare ptr @__catalyst__qis__Measure(ptr, i32) local_unnamed_addr
declare void @__catalyst__qis__T(ptr, ptr) local_unnamed_addr
declare void @__catalyst__qis__CNOT(ptr, ptr, ptr) local_unnamed_addr
declare ptr @__catalyst__rt__array_get_element_ptr_1d(ptr, i64) local_unnamed_addr
declare ptr @__catalyst__rt__qubit_allocate_array(i64) local_unnamed_addr
declare void @__catalyst__rt__device_init(ptr, ptr, ptr) local_unnamed_addr
declare void @_mlir_memref_to_llvm_free(ptr) local_unnamed_addr
declare ptr @_mlir_memref_to_llvm_alloc(i64) local_unnamed_addr
define { ptr, ptr, i64 } @jit_jaspr_function(ptr nocapture readnone %0, ptr nocapture readonly %1, i64 %2) local_unnamed_addr {
tail call void @__catalyst__rt__device_init(ptr nonnull @"/home/positr0nium/miniconda3/envs/qrisp/lib/python3.10/site-packages/catalyst/utils/../lib/librtd_lightning.so", ptr nonnull @lightning.qubit, ptr nonnull @"{'shots': 0, 'mcmc': False, 'num_burnin': 0, 'kernel_name': None}")
%4 = tail call ptr @__catalyst__rt__qubit_allocate_array(i64 20)
%5 = tail call ptr @__catalyst__rt__array_get_element_ptr_1d(ptr %4, i64 0)
%6 = load ptr, ptr %5, align 8
%7 = tail call ptr @__catalyst__rt__array_get_element_ptr_1d(ptr %4, i64 1)
%8 = load ptr, ptr %7, align 8
tail call void @__catalyst__qis__CNOT(ptr %6, ptr %8, ptr null)
%9 = tail call ptr @__catalyst__rt__array_get_element_ptr_1d(ptr %4, i64 1)
%10 = load ptr, ptr %9, align 8
tail call void @__catalyst__qis__T(ptr %10, ptr null)
%11 = tail call ptr @_mlir_memref_to_llvm_alloc(i64 65)
%12 = ptrtoint ptr %11 to i64
%13 = add i64 %12, 63
%14 = and i64 %13, -64
%15 = inttoptr i64 %14 to ptr
%16 = tail call ptr @_mlir_memref_to_llvm_alloc(i64 65)
%17 = ptrtoint ptr %16 to i64
%18 = add i64 %17, 63
%19 = and i64 %18, -64
%20 = inttoptr i64 %19 to ptr
%21 = tail call ptr @_mlir_memref_to_llvm_alloc(i64 72)
%22 = ptrtoint ptr %21 to i64
%23 = add i64 %22, 63
%24 = and i64 %23, -64
%25 = inttoptr i64 %24 to ptr
%26 = tail call ptr @_mlir_memref_to_llvm_alloc(i64 72)
%27 = ptrtoint ptr %26 to i64
%28 = add i64 %27, 63
%29 = and i64 %28, -64
%30 = inttoptr i64 %29 to ptr
%31 = tail call ptr @_mlir_memref_to_llvm_alloc(i64 72)
%32 = ptrtoint ptr %31 to i64
%33 = add i64 %32, 63
%34 = and i64 %33, -64
%35 = inttoptr i64 %34 to ptr
%36 = tail call ptr @_mlir_memref_to_llvm_alloc(i64 72)
%37 = ptrtoint ptr %36 to i64
%38 = add i64 %37, 63
%39 = and i64 %38, -64
%40 = inttoptr i64 %39 to ptr
%41 = tail call ptr @_mlir_memref_to_llvm_alloc(i64 8)
store i64 0, ptr %41, align 1
%42 = tail call ptr @_mlir_memref_to_llvm_alloc(i64 8)
store i64 0, ptr %42, align 1
%43 = load i64, ptr %1, align 4
%44 = icmp slt i64 %43, 1
store i1 %44, ptr %15, align 64
%45 = tail call ptr @_mlir_memref_to_llvm_alloc(i64 8)
%46 = load i64, ptr %42, align 1
store i64 %46, ptr %45, align 1
tail call void @_mlir_memref_to_llvm_free(ptr nonnull %42)
%47 = tail call ptr @_mlir_memref_to_llvm_alloc(i64 8)
%48 = load i64, ptr %41, align 1
store i64 %48, ptr %47, align 1
tail call void @_mlir_memref_to_llvm_free(ptr nonnull %41)
br i1 %44, label %.lr.ph, label %._crit_edge
.lr.ph: ; preds = %3, %.lr.ph
%49 = phi ptr [ %87, %.lr.ph ], [ %47, %3 ]
%50 = phi ptr [ %85, %.lr.ph ], [ %45, %3 ]
%51 = load i64, ptr %50, align 4
%52 = tail call ptr @__catalyst__rt__array_get_element_ptr_1d(ptr %4, i64 %51)
%53 = load ptr, ptr %52, align 8
%54 = tail call ptr @__catalyst__qis__Measure(ptr %53, i32 -1)
%55 = load i1, ptr %54, align 1
store i1 %55, ptr %20, align 64
%56 = load i64, ptr %50, align 4
store i64 %56, ptr %25, align 64
%57 = shl i64 2, %56
%58 = icmp ult i64 %56, 64
%59 = select i1 %58, i64 %57, i64 0
store i64 %59, ptr %30, align 64
%60 = load i1, ptr %20, align 64
%61 = zext i1 %60 to i64
store i64 %61, ptr %35, align 64
%62 = load i64, ptr %30, align 64
%63 = select i1 %60, i64 %62, i64 0
store i64 %63, ptr %40, align 64
%64 = tail call ptr @_mlir_memref_to_llvm_alloc(i64 72)
%65 = ptrtoint ptr %64 to i64
%66 = add i64 %65, 63
%67 = and i64 %66, -64
%68 = inttoptr i64 %67 to ptr
%69 = load i64, ptr %49, align 4
%70 = load i64, ptr %40, align 64
%71 = add i64 %70, %69
store i64 %71, ptr %68, align 64
tail call void @_mlir_memref_to_llvm_free(ptr nonnull %49)
%72 = tail call ptr @_mlir_memref_to_llvm_alloc(i64 72)
%73 = ptrtoint ptr %72 to i64
%74 = add i64 %73, 63
%75 = and i64 %74, -64
%76 = inttoptr i64 %75 to ptr
%77 = load i64, ptr %50, align 4
%78 = add i64 %77, 1
store i64 %78, ptr %76, align 64
tail call void @_mlir_memref_to_llvm_free(ptr nonnull %50)
%79 = tail call ptr @_mlir_memref_to_llvm_alloc(i64 8)
%80 = load i64, ptr %68, align 64
store i64 %80, ptr %79, align 1
tail call void @_mlir_memref_to_llvm_free(ptr %64)
%81 = tail call ptr @_mlir_memref_to_llvm_alloc(i64 8)
%82 = load i64, ptr %76, align 64
store i64 %82, ptr %81, align 1
tail call void @_mlir_memref_to_llvm_free(ptr %72)
%.pre = load i64, ptr %81, align 4
%83 = load i64, ptr %1, align 4
%84 = icmp sge i64 %.pre, %83
store i1 %84, ptr %15, align 64
%85 = tail call ptr @_mlir_memref_to_llvm_alloc(i64 8)
%86 = load i64, ptr %81, align 1
store i64 %86, ptr %85, align 1
tail call void @_mlir_memref_to_llvm_free(ptr nonnull %81)
%87 = tail call ptr @_mlir_memref_to_llvm_alloc(i64 8)
%88 = load i64, ptr %79, align 1
store i64 %88, ptr %87, align 1
tail call void @_mlir_memref_to_llvm_free(ptr nonnull %79)
br i1 %84, label %.lr.ph, label %._crit_edge
._crit_edge: ; preds = %.lr.ph, %3
%.lcssa20 = phi ptr [ %45, %3 ], [ %85, %.lr.ph ]
%.lcssa = phi ptr [ %47, %3 ], [ %87, %.lr.ph ]
tail call void @_mlir_memref_to_llvm_free(ptr nonnull %.lcssa20)
tail call void @_mlir_memref_to_llvm_free(ptr %36)
tail call void @_mlir_memref_to_llvm_free(ptr %31)
tail call void @_mlir_memref_to_llvm_free(ptr %26)
tail call void @_mlir_memref_to_llvm_free(ptr %21)
tail call void @_mlir_memref_to_llvm_free(ptr %16)
tail call void @_mlir_memref_to_llvm_free(ptr %11)
%89 = tail call ptr @_mlir_memref_to_llvm_alloc(i64 68)
%90 = ptrtoint ptr %89 to i64
%91 = add i64 %90, 63
%92 = and i64 %91, -64
%93 = inttoptr i64 %92 to ptr
%94 = load i64, ptr %.lcssa, align 4
%95 = trunc i64 %94 to i32
store i32 %95, ptr %93, align 64
tail call void @_mlir_memref_to_llvm_free(ptr nonnull %.lcssa)
%96 = tail call ptr @_mlir_memref_to_llvm_alloc(i64 68)
%97 = ptrtoint ptr %96 to i64
%98 = add i64 %97, 63
%99 = and i64 %98, -64
%100 = inttoptr i64 %99 to ptr
%101 = load i32, ptr %93, align 64
%102 = add i32 %101, 1
store i32 %102, ptr %100, align 64
tail call void @_mlir_memref_to_llvm_free(ptr %89)
%103 = icmp eq ptr %96, inttoptr (i64 3735928559 to ptr)
br i1 %103, label %104, label %107
104: ; preds = %._crit_edge
%105 = tail call ptr @_mlir_memref_to_llvm_alloc(i64 4)
%106 = load i32, ptr %100, align 64
store i32 %106, ptr %105, align 1
br label %107
107: ; preds = %104, %._crit_edge
%.pn16 = phi ptr [ %105, %104 ], [ %96, %._crit_edge ]
%.pn14 = phi ptr [ %105, %104 ], [ %100, %._crit_edge ]
%.pn13 = insertvalue { ptr, ptr, i64 } undef, ptr %.pn16, 0
%.pn = insertvalue { ptr, ptr, i64 } %.pn13, ptr %.pn14, 1
%108 = insertvalue { ptr, ptr, i64 } %.pn, i64 0, 2
ret { ptr, ptr, i64 } %108
}
define void @_catalyst_pyface_jit_jaspr_function(ptr nocapture writeonly %0, ptr nocapture readonly %1) local_unnamed_addr {
%.unpack = load ptr, ptr %1, align 8
%.elt1.i = getelementptr inbounds { ptr, ptr, i64 }, ptr %.unpack, i64 0, i32 1
%.unpack2.i = load ptr, ptr %.elt1.i, align 8
%3 = tail call { ptr, ptr, i64 } @jit_jaspr_function(ptr poison, ptr %.unpack2.i, i64 poison)
%.elt.i = extractvalue { ptr, ptr, i64 } %3, 0
store ptr %.elt.i, ptr %0, align 8
%.repack5.i = getelementptr inbounds { ptr, ptr, i64 }, ptr %0, i64 0, i32 1
%.elt6.i = extractvalue { ptr, ptr, i64 } %3, 1
store ptr %.elt6.i, ptr %.repack5.i, align 8
%.repack7.i = getelementptr inbounds { ptr, ptr, i64 }, ptr %0, i64 0, i32 2
%.elt8.i = extractvalue { ptr, ptr, i64 } %3, 2
store i64 %.elt8.i, ptr %.repack7.i, align 8
ret void
}
define void @_catalyst_ciface_jit_jaspr_function(ptr nocapture writeonly %0, ptr nocapture readonly %1) local_unnamed_addr {
%.elt1 = getelementptr inbounds { ptr, ptr, i64 }, ptr %1, i64 0, i32 1
%.unpack2 = load ptr, ptr %.elt1, align 8
%3 = tail call { ptr, ptr, i64 } @jit_jaspr_function(ptr poison, ptr %.unpack2, i64 poison)
%.elt = extractvalue { ptr, ptr, i64 } %3, 0
store ptr %.elt, ptr %0, align 8
%.repack5 = getelementptr inbounds { ptr, ptr, i64 }, ptr %0, i64 0, i32 1
%.elt6 = extractvalue { ptr, ptr, i64 } %3, 1
store ptr %.elt6, ptr %.repack5, align 8
%.repack7 = getelementptr inbounds { ptr, ptr, i64 }, ptr %0, i64 0, i32 2
%.elt8 = extractvalue { ptr, ptr, i64 } %3, 2
store i64 %.elt8, ptr %.repack7, align 8
ret void
}
define void @setup() local_unnamed_addr {
tail call void @__catalyst__rt__initialize()
ret void
}
define void @teardown() local_unnamed_addr {
tail call void @__catalyst__rt__finalize()
ret void
}
!llvm.module.flags = !{!0}
!0 = !{i32 2, !"Debug Info Version", i32 3}
"""
from qrisp.jasp.evaluation_tools.catalyst_interface import jaspr_to_qir
return jaspr_to_qir(self.flatten_environments())
[docs]
def to_mlir(self):
"""
Compiles the Jaspr to an xDSL module using the Jasp Dialect.
Requires the xDSL package to be installed (``pip install xdsl``).
.. note::
An xDSL module can be visualized via:
::
print(xdsl_module)
and serialized to a string using:
::
from xdsl.printer import Printer
Printer().print_op(xdsl_module)
Returns
-------
xdsl.dialects.builtin.ModuleOp
An xDSL module representing the quantum computation.
Examples
--------
We create a simple script and inspect the MLIR string:
::
from qrisp import *
from qrisp.jasp import make_jaspr
def example_function(i):
qv = QuantumVariable(i)
cx(qv[0], qv[1])
t(qv[1])
meas_res = measure(qv)
meas_res += 1
return meas_res
jaspr = make_jaspr(example_function)(2)
print(jaspr.to_mlir())
.. code-block:: none
builtin.module @jasp_module {
func.func public @main(%arg0 : tensor<i64>, %arg1 : !jasp.QuantumState) -> (tensor<i64>, !jasp.QuantumState) {
%0, %1 = "jasp.create_qubits"(%arg0, %arg1) : (tensor<i64>, !jasp.QuantumState) -> (!jasp.QubitArray, !jasp.QuantumState)
%2 = "stablehlo.constant"() <{value = dense<0> : tensor<i64>}> : () -> tensor<i64>
%3 = "jasp.get_qubit"(%0, %2) : (!jasp.QubitArray, tensor<i64>) -> !jasp.Qubit
%4 = "stablehlo.constant"() <{value = dense<1> : tensor<i64>}> : () -> tensor<i64>
%5 = "jasp.get_qubit"(%0, %4) : (!jasp.QubitArray, tensor<i64>) -> !jasp.Qubit
%6 = "jasp.quantum_gate"(%3, %5, %1) {gate_type = "cx"} : (!jasp.Qubit, !jasp.Qubit, !jasp.QuantumState) -> !jasp.QuantumState
%7 = "jasp.quantum_gate"(%5, %6) {gate_type = "t"} : (!jasp.Qubit, !jasp.QuantumState) -> !jasp.QuantumState
%8, %9 = "jasp.measure"(%0, %7) : (!jasp.QubitArray, !jasp.QuantumState) -> (tensor<i64>, !jasp.QuantumState)
%10 = "stablehlo.add"(%8, %4) : (tensor<i64>, tensor<i64>) -> tensor<i64>
%11 = "jasp.reset"(%0, %9) : (!jasp.QubitArray, !jasp.QuantumState) -> !jasp.QuantumState
%12 = "jasp.delete_qubits"(%0, %11) : (!jasp.QubitArray, !jasp.QuantumState) -> !jasp.QuantumState
func.return %10, %12 : tensor<i64>, !jasp.QuantumState
}
}
"""
from qrisp.jasp.mlir import jaspr_to_mlir
return jaspr_to_mlir(self)
[docs]
def to_catalyst_mlir(self):
"""
Compiles the Jaspr to MLIR using the `Catalyst dialect <https://docs.pennylane.ai/projects/catalyst/en/stable/index.html>`__.
Parameters
----------
None
Returns
-------
str
The MLIR string.
Examples
--------
We create a simple script and inspect the MLIR string:
::
from qrisp import *
from qrisp.jasp import make_jaspr
def example_function(i):
qv = QuantumFloat(i)
cx(qv[0], qv[1])
t(qv[1])
meas_res = measure(qv)
meas_res += 1
return meas_res
jaspr = make_jaspr(example_function)(2)
print(jaspr.to_catalyst_mlir())
.. code-block:: none
module @jaspr_function {
func.func public @jit_jaspr_function(%arg0: tensor<i64>) -> tensor<i32> attributes {llvm.emit_c_interface} {
%0 = stablehlo.constant dense<1> : tensor<i32>
%1 = stablehlo.constant dense<2> : tensor<i64>
%2 = stablehlo.constant dense<1> : tensor<i64>
%3 = stablehlo.constant dense<0> : tensor<i64>
quantum.device["/home/positr0nium/miniconda3/envs/qrisp/lib/python3.10/site-packages/catalyst/utils/../lib/librtd_lightning.so", "lightning.qubit", "{'shots': 0, 'mcmc': False, 'num_burnin': 0, 'kernel_name': None}"]
%4 = quantum.alloc( 20) : !quantum.reg
%5 = quantum.extract %4[ 0] : !quantum.reg -> !quantum.bit
%6 = quantum.extract %4[ 1] : !quantum.reg -> !quantum.bit
%out_qubits:2 = quantum.custom "CNOT"() %5, %6 : !quantum.bit, !quantum.bit
%7 = quantum.insert %4[ 0], %out_qubits#0 : !quantum.reg, !quantum.bit
%8 = quantum.insert %7[ 1], %out_qubits#1 : !quantum.reg, !quantum.bit
%9 = quantum.extract %8[ 1] : !quantum.reg -> !quantum.bit
%out_qubits_0 = quantum.custom "T"() %9 : !quantum.bit
%10 = quantum.insert %8[ 1], %out_qubits_0 : !quantum.reg, !quantum.bit
%11 = stablehlo.add %3, %arg0 : tensor<i64>
%12:3 = scf.while (%arg1 = %3, %arg2 = %3, %arg3 = %10) : (tensor<i64>, tensor<i64>, !quantum.reg) -> (tensor<i64>, tensor<i64>, !quantum.reg) {
%16 = stablehlo.compare GE, %arg1, %11, SIGNED : (tensor<i64>, tensor<i64>) -> tensor<i1>
%extracted = tensor.extract %16[] : tensor<i1>
scf.condition(%extracted) %arg1, %arg2, %arg3 : tensor<i64>, tensor<i64>, !quantum.reg
} do {
^bb0(%arg1: tensor<i64>, %arg2: tensor<i64>, %arg3: !quantum.reg):
%extracted = tensor.extract %arg1[] : tensor<i64>
%16 = quantum.extract %arg3[%extracted] : !quantum.reg -> !quantum.bit
%mres, %out_qubit = quantum.measure %16 : i1, !quantum.bit
%from_elements = tensor.from_elements %mres : tensor<i1>
%extracted_1 = tensor.extract %arg1[] : tensor<i64>
%17 = quantum.insert %arg3[%extracted_1], %out_qubit : !quantum.reg, !quantum.bit
%18 = stablehlo.subtract %arg1, %3 : tensor<i64>
%19 = stablehlo.shift_left %1, %18 : tensor<i64>
%20 = stablehlo.convert %from_elements : (tensor<i1>) -> tensor<i64>
%21 = stablehlo.multiply %19, %20 : tensor<i64>
%22 = stablehlo.add %arg2, %21 : tensor<i64>
%23 = stablehlo.add %arg1, %2 : tensor<i64>
scf.yield %23, %22, %17 : tensor<i64>, tensor<i64>, !quantum.reg
}
%13 = stablehlo.convert %12#1 : (tensor<i64>) -> tensor<i32>
%14 = stablehlo.multiply %13, %0 : tensor<i32>
%15 = stablehlo.add %14, %0 : tensor<i32>
return %15 : tensor<i32>
}
func.func @setup() {
quantum.init
return
}
func.func @teardown() {
quantum.finalize
return
}
}
"""
from qrisp.jasp.evaluation_tools.catalyst_interface import jaspr_to_mlir
return jaspr_to_mlir(self.flatten_environments())
[docs]
def to_qasm(self, *args):
"""
Compiles the Jaspr into an OpenQASM 2 string. Real-time control is possible
as long as no computations on the measurement results are performed.
Parameters
----------
*args : list
The arguments to call the :ref:`QuantumCircuit` evaluation with.
Returns
-------
str
The OpenQASM 2 string.
Examples
--------
We create a simple script and inspect the QASM 2 string:
::
from qrisp import *
from qrisp.jasp import make_jaspr
def main(i):
qv = QuantumVariable(i)
cx(qv[0], qv[1])
t(qv[1])
return qv
jaspr = make_jaspr(main)(2)
qasm_str = jaspr.to_qasm(2)
print(qasm_str)
# Yields
# OPENQASM 2.0;
# include "qelib1.inc";
# qreg qb_59[1];
# qreg qb_60[1];
# cx qb_59[0],qb_60[0];
# t qb_60[0];
It is also possible to compile simple real-time control features:
::
def main(phi):
qf = QuantumFloat(5)
h(qf)
bl = measure(qf[0])
with control(bl):
rz(phi, qf[1])
x(qf[1])
return
jaspr = make_jaspr(main)(0.5)
print(jaspr.to_qasm(0.5))
This gives:
::
OPENQASM 2.0;
include "qelib1.inc";
qreg qb_59[1];
qreg qb_60[1];
qreg qb_61[1];
qreg qb_62[1];
qreg qb_63[1];
creg cb_0[1];
h qb_59[0];
h qb_60[0];
h qb_61[0];
reset qb_61[0];
h qb_62[0];
reset qb_62[0];
h qb_63[0];
reset qb_63[0];
measure qb_59[0] -> cb_0[0];
reset qb_59[0];
if(cb_0==1) rz(0.5) qb_60[0];
if(cb_0==1) x qb_60[0];
reset qb_60[0];
"""
res = self.to_qc(*args)
if len(self.outvars) == 1:
res = [res]
qrisp_qc = res[-1]
return qrisp_qc.qasm()
[docs]
def to_catalyst_jaxpr(self):
"""
Compiles the jaspr to the corresponding `Catalyst jaxpr <https://docs.pennylane.ai/projects/catalyst/en/stable/index.html>`__.
Parameters
----------
*args : tuple
The arguments to call the jaspr with.
Returns
-------
jax.core.Jaxpr
The Jaxpr using Catalyst primitives.
Examples
--------
We create a simple script and inspect the Catalyst Jaxpr:
::
from qrisp import *
from qrisp.jasp import make_jaspr
def example_function(i):
qv = QuantumFloat(i)
cx(qv[0], qv[1])
t(qv[1])
meas_res = measure(qv)
meas_res += 1
return meas_res
jaspr = make_jaspr(example_function)(2)
print(jaspr.to_catalyst_jaxpr())
# Yields
# { lambda ; a:AbstractQreg() b:i64[] c:i32[]. let
# d:i64[] = convert_element_type[new_dtype=int64 weak_type=True] c
# e:i64[] = add b d
# f:i64[] = add b 0
# g:i64[] = add b 1
# h:AbstractQbit() = qextract a f
# i:AbstractQbit() = qextract a g
# j:AbstractQbit() k:AbstractQbit() = qinst[op=CNOT qubits_len=2] h i
# l:AbstractQreg() = qinsert a f j
# m:AbstractQreg() = qinsert l g k
# n:AbstractQbit() = qextract m g
# o:AbstractQbit() = qinst[op=T qubits_len=1] n
# p:AbstractQreg() = qinsert m g o
# q:i64[] = convert_element_type[new_dtype=int64 weak_type=True] c
# r:i64[] = add b q
# _:i64[] s:i64[] t:AbstractQreg() _:i64[] _:i64[] = while_loop[
# body_jaxpr={ lambda ; u:i64[] v:i64[] w:AbstractQreg() x:i64[] y:i64[]. let
# z:AbstractQbit() = qextract w u
# ba:bool[] bb:AbstractQbit() = qmeasure z
# bc:AbstractQreg() = qinsert w u bb
# bd:i64[] = sub u x
# be:i64[] = shift_left 2 bd
# bf:i64[] = convert_element_type[new_dtype=int64 weak_type=True] ba
# bg:i64[] = mul be bf
# bh:i64[] = add v bg
# bi:i64[] = add u 1
# in (bi, bh, bc, x, y) }
# body_nconsts=0
# cond_jaxpr={ lambda ; bj:i64[] bk:i64[] bl:AbstractQreg() bm:i64[] bn:i64[]. let
# bo:bool[] = ge bj bn
# in (bo,) }
# cond_nconsts=0
# nimplicit=0
# preserve_dimensions=True
# ] b 0 p b r
# bp:i32[] = convert_element_type[new_dtype=int64 weak_type=False] s
# bq:i32[] = mul bp 1
# br:i32[] = add bq 1
# in (t, e, br) }
"""
from qrisp.jasp.evaluation_tools.catalyst_interface import (
jaspr_to_catalyst_jaxpr,
)
return jaspr_to_catalyst_jaxpr(self.flatten_environments())
def make_jaxpr_mod(fun, static_argnums=(), return_shape=False, abstracted_axes=None):
"""
Creates a function that produces the jaxpr of a traced function.
This is a modified version of JAX's ``make_jaxpr`` that supports
``return_shape=True`` even when the function returns custom abstract
types that don't have ``shape``/``dtype`` attributes (such as
``AbstractQuantumState``).
The interface is identical to ``jax.make_jaxpr``.
Parameters
----------
fun : Callable
The function whose jaxpr is to be computed.
static_argnums : int or Sequence[int], optional
Indices of arguments that should be treated as static (not traced).
Default is ``()``.
return_shape : bool, optional
If True, the returned function produces a tuple ``(jaxpr, out_tree)``
where ``out_tree`` is a PyTreeDef representing the structure of the
output. This can be used to reconstruct PyTree objects from flat
output lists using ``jax.tree_util.tree_unflatten``.
Default is False.
abstracted_axes : optional
Specification for which axes to abstract over. Default is None.
Returns
-------
Callable
A function that, when called with example arguments, returns either:
- A ClosedJaxpr representation of ``fun`` (if ``return_shape=False``)
- A tuple ``(ClosedJaxpr, out_tree)`` (if ``return_shape=True``)
Notes
-----
JAX's native ``make_jaxpr(return_shape=True)`` fails on custom abstract
types because it tries to create ``ShapeDtypeStruct`` objects from the
outputs, which requires ``shape`` and ``dtype`` attributes. This function
avoids that by using ``jit(...).trace()`` to directly access the output
tree structure.
Examples
--------
>>> def f(x):
... return {"a": x + 1, "b": x * 2}
>>> jaxpr, out_tree = make_jaxpr_mod(f, return_shape=True)(1.0)
>>> # out_tree can be used with tree_unflatten to reconstruct the dict
"""
from jax._src.interpreters import partial_eval as pe
from jax._src.util import split_list
def jaxpr_creator(*args, **kwargs):
if return_shape:
# Use jit(...).trace() directly to get access to _out_tree
# This avoids JAX's make_jaxpr return_shape logic which fails on
# custom abstract types that don't have shape/dtype attributes.
traced = jax.jit(
fun, static_argnums=static_argnums, abstracted_axes=abstracted_axes
).trace(*args, **kwargs)
# Extract the jaxpr, handling constants if needed
# (same logic as JAX's make_jaxpr)
if traced._num_consts:
consts, _ = split_list(traced._args_flat, [traced._num_consts])
jaxpr_ = pe.convert_invars_to_constvars(
traced.jaxpr.jaxpr, traced._num_consts
)
closed_jaxpr = ClosedJaxpr(jaxpr_, consts)
else:
closed_jaxpr = traced.jaxpr
return closed_jaxpr, traced._out_tree
else:
return make_jaxpr(
fun, static_argnums=static_argnums, abstracted_axes=abstracted_axes
)(*args, **kwargs)
return jaxpr_creator
[docs]
def make_jaspr(fun, flatten_envs=True, return_shape=False, **jax_kwargs):
"""
Creates a function that returns the Jaspr representation of a quantum function.
This function is analogous to JAX's ``make_jaxpr``, but produces a Jaspr
(a Jaxpr enhanced with quantum primitives) from a Qrisp quantum function.
Parameters
----------
fun : Callable
The quantum function whose Jaspr is to be computed.
flatten_envs : bool, optional
If True (default), flatten quantum environments in the resulting Jaspr.
return_shape : bool, optional
If True, the returned function produces a tuple ``(jaspr, out_tree)``
where ``out_tree`` is a PyTreeDef representing the structure of the
output of ``fun``. This can be used to reconstruct PyTree objects
from flat output lists using ``jax.tree_util.tree_unflatten``.
Default is False.
**jax_kwargs
Additional keyword arguments passed to ``jax.make_jaxpr``, such as
``static_argnums``.
Returns
-------
Callable
A function that, when called with example arguments, returns either:
- A Jaspr representation of ``fun`` (if ``return_shape=False``)
- A tuple ``(Jaspr, out_tree)`` (if ``return_shape=True``) where ``out_tree`` is a PyTreeDef that can be used with ``tree_unflatten``
Examples
--------
**Basic quantum circuit with measurement**
Create a Jaspr for a simple Bell state circuit:
::
from qrisp import QuantumVariable, h, cx, measure
from qrisp.jasp import make_jaspr
def simple_circuit():
qv = QuantumVariable(2)
h(qv[0])
cx(qv[0], qv[1])
return measure(qv)
jaspr = make_jaspr(simple_circuit)()
result = jaspr() # Returns 0 or 3 with equal probability
**Parameterized quantum circuit**
Create a Jaspr with parameterized gates that can be executed with different parameters:
::
from qrisp import QuantumVariable, h, p, measure
from qrisp.jasp import make_jaspr
def rotation_circuit(angle):
qv = QuantumVariable(1)
h(qv)
p(angle, qv)
return measure(qv)
jaspr = make_jaspr(rotation_circuit)(0.5)
result1 = jaspr(0.5) # Execute with angle=0.5
result2 = jaspr(1.0) # Execute with angle=1.0
**Using return_shape for PyTree reconstruction**
Retrieve the output tree structure alongside the Jaspr for reconstructing
complex return values:
::
from qrisp import QuantumVariable, h, cx, x, measure
from qrisp.jasp import make_jaspr
from jax.tree_util import tree_unflatten, tree_flatten
def multi_output_circuit():
qa = QuantumVariable(2)
qb = QuantumVariable(2)
h(qa[0])
cx(qa[0], qa[1])
x(qb)
return measure(qa), measure(qb)
jaspr, out_tree = make_jaspr(multi_output_circuit, return_shape=True)()
result_a, result_b = jaspr()
# Use out_tree to reconstruct the output structure
flat_results, _ = tree_flatten((result_a, result_b))
reconstructed = tree_unflatten(out_tree, flat_results)
"""
from qrisp import recursive_qv_search
from qrisp.jasp import (
AbstractQuantumState,
TracingQuantumSession,
check_for_tracing_mode,
)
# Handle static_argnums adjustment for the ammended function
adjusted_jax_kwargs = dict(jax_kwargs)
if "static_argnums" in adjusted_jax_kwargs:
if isinstance(adjusted_jax_kwargs["static_argnums"], list):
adjusted_jax_kwargs["static_argnums"] = list(
adjusted_jax_kwargs["static_argnums"]
)
for i in range(len(adjusted_jax_kwargs["static_argnums"])):
adjusted_jax_kwargs["static_argnums"][i] += 1
else:
adjusted_jax_kwargs["static_argnums"] += 1
def jaspr_creator(*args, **kwargs):
qs = TracingQuantumSession.get_instance()
# Close any tracing quantum sessions that might have not been
# properly closed due to whatever reason.
if not check_for_tracing_mode():
while qs.abs_qst is not None:
qs.conclude_tracing()
# This function will be traced by Jax.
# Note that we add the abs_qst keyword as the tracing quantum circuit
def ammended_function(*args, **kwargs):
abs_qst = kwargs[10 * "~"]
del kwargs[10 * "~"]
qs.start_tracing(abs_qst)
# If the signature contains QuantumVariables, these QuantumVariables went
# through a flattening/unflattening procedure. The unflattening creates
# a copy of the QuantumVariable object, which is however not yet registered in any
# QuantumSession. We register these QuantumVariables in the current QuantumSession.
arg_qvs = recursive_qv_search(args)
for qv in arg_qvs:
qs.register_qv(qv, None)
try:
res = fun(*args, **kwargs)
except Exception as e:
qs.conclude_tracing()
raise e
res_qc = qs.conclude_tracing()
return res, res_qc
ammended_kwargs = dict(kwargs)
ammended_kwargs[10 * "~"] = AbstractQuantumState()
# Use make_jaxpr_mod to trace the function
# Pass return_shape through to get the output tree if requested
static_argnums = adjusted_jax_kwargs.get("static_argnums", ())
abstracted_axes = adjusted_jax_kwargs.get("abstracted_axes", None)
result = make_jaxpr_mod(
ammended_function,
static_argnums=static_argnums,
return_shape=return_shape,
abstracted_axes=abstracted_axes,
)(*args, **ammended_kwargs)
if return_shape:
closed_jaxpr, full_out_tree = result
# full_out_tree is a PyTreeDef for (res, res_qc)
# We use .children() to get the subtrees, and take the first one (res)
user_out_tree = full_out_tree.children()[0]
else:
closed_jaxpr = result
# Collect the environments
# This means that the quantum environments no longer appear as
# enter/exit primitives but as primitive that "call" a certain Jaspr.
res = Jaspr.from_cache(collect_environments(closed_jaxpr))
if flatten_envs:
res = res.flatten_environments()
if return_shape:
return res, user_out_tree
else:
return res
return jaspr_creator
def check_aval_equivalence(invars_1, invars_2):
avals_1 = [invar.aval for invar in invars_1]
avals_2 = [invar.aval for invar in invars_2]
return all([type(avals_1[i]) == type(avals_2[i]) for i in range(len(avals_1))])
def remove_redundant_allocations(closed_jaxpr):
"""
Optimizes the Jaspr by removing redundant qubit allocations.
Strategy:
1. Map usages of all variables to identify how QubitArrays are consumed.
2. Identify `jasp.create_qubits` operations that are redundant. An allocation is redundant if:
- The resulting QubitArray is not returned by the function.
- The QubitArray is either unused (DropVar) or ONLY used by `free` primitives
(like `get_size`, `delete_qubits`) that don't actually require the physical qubits
if we know the allocation parameters.
3. Plan removals and replacements:
- Mark redundant `create_qubits` equations for removal.
- Mark dependent `get_size`, `reset`, `delete_qubits` equations for removal.
- Map the output circuit of removed equations to their input circuit (bypassing the operation).
- Map the output of `get_size` to the size input of the `create_qubits` equation.
4. Rewrite the Jaxpr:
- Filter out removed equations.
- Update variable references in remaining equations and output variables using the replacement map.
"""
from jax.core import DropVar
try:
from jax.extend.core import Literal
except ImportError:
from jax.core import Literal
jaxpr = closed_jaxpr.jaxpr
eqns = jaxpr.eqns
usages = {}
# 1. Build usage map
# We iterate over all equations and record which equation uses which input variable.
for eqn in eqns:
for var in eqn.invars:
if isinstance(var, DropVar):
continue
if isinstance(var, Literal):
continue
if var not in usages:
usages[var] = []
usages[var].append(eqn)
# Identify variables that are returned by the function, as these cannot be optimized away.
returned_vars = set()
for var in jaxpr.outvars:
if not isinstance(var, DropVar) and not isinstance(var, Literal):
returned_vars.add(var)
replacements = {}
eqns_to_remove = set()
# Primitives that are "safe" to exist on a redundant qubit array.
# These operations can be resolved without the actual qubits if the allocation is known.
allowed_primitives = {"jasp.get_size", "jasp.delete_qubits"}
# 2. Identify and plan removals
for eqn in eqns:
if eqn.primitive.name == "jasp.create_qubits":
out_qa = eqn.outvars[0]
out_qc = eqn.outvars[1]
# If the QubitArray is part of the output, we must keep it.
if out_qa in returned_vars:
continue
# Check if out_qa is only used by allowed primitives or not at all
is_redundant = False
if isinstance(out_qa, DropVar) or out_qa not in usages:
is_redundant = True
else:
only_allowed_usages = True
for u_eqn in usages[out_qa]:
if u_eqn.primitive.name not in allowed_primitives:
# Found a usage (e.g., gate) that requires real qubits.
only_allowed_usages = False
break
if only_allowed_usages:
is_redundant = True
if is_redundant:
eqns_to_remove.add(id(eqn))
# Rewire circuit: Bypass the allocation by mapping output circuit to input circuit.
replacements[out_qc] = eqn.invars[1]
# Handle dependent equations (safe primitives)
if not isinstance(out_qa, DropVar) and out_qa in usages:
for u_eqn in usages[out_qa]:
eqns_to_remove.add(id(u_eqn))
if u_eqn.primitive.name == "jasp.get_size":
# Optimize get_size: Use the input size from create_qubits directly.
replacements[u_eqn.outvars[0]] = eqn.invars[0]
elif u_eqn.primitive.name == "jasp.delete_qubits":
# Bypass cleanup: Map the cleanup's output circuit to its input circuit.
replacements[u_eqn.outvars[0]] = u_eqn.invars[1]
if not eqns_to_remove:
return
# 3. Apply replacements
def resolve(var):
# Literals are not hashable and cannot be in replacements
if isinstance(var, Literal):
return var
try:
# Resolving chains of replacements
while var in replacements:
var = replacements[var]
except TypeError:
# Handle other potentially unhashable types
pass
return var
new_eqns = []
for eqn in eqns:
if id(eqn) in eqns_to_remove:
continue
# Update inputs of remaining equations using the replacement map
eqn.invars[:] = [resolve(var) for var in eqn.invars]
new_eqns.append(eqn)
# Update the equations list in-place
jaxpr.eqns[:] = new_eqns
# Update the output variables of the jaxpr
jaxpr.outvars[:] = [resolve(var) for var in jaxpr.outvars]
