Qasm exporter fails with ECRGate QasmError: "Duplicate declaration for gate 'rzx'

[MattePalte]

Environment

• Qiskit Terra version: 0.19.1
• Python version: 3.8
• Operating system: Ubuntu 18.04.6 LTS

What is happening?

The qasm exporter on a circuit with an ECRGate outputs an invalid QASM code which cannot be parsed back.

How can we reproduce the issue?

Run this circuit:

import qiskit
from qiskit import QuantumRegister
from qiskit import Aer, transpile, execute
qc = QuantumCircuit(2, 2, name='qc')
qc.ecr(0, 1)
qc.measure_all()
qc.qasm(formatted=True)

Output

OPENQASM 2.0;
include "qelib1.inc";
gate rzx(param0) q0,q1 { h q1; cx q0,q1; rz(-pi/4) q1; cx q0,q1; h q1; }
gate rzx(param0) q0,q1 { h q1; cx q0,q1; rz(pi/4) q1; cx q0,q1; h q1; }
gate ecr q0,q1 { rzx(pi/4) q0,q1; x q0; rzx(-pi/4) q0,q1; }
qreg q[2];
creg c[2];
creg meas[2];
ecr q[0],q[1];
barrier q[0],q[1];
measure q[0] -> meas[0];
measure q[1] -> meas[1];

Read the QASM back:

qc = QuantumCircuit.from_qasm_str(qc.qasm())

Output

QasmError                                 Traceback (most recent call last)
/tmp/ipykernel_35760/1770840892.py in <module>
----> 1 qc = QuantumCircuit.from_qasm_str(qc.qasm())
      2 counts = execute(qc, backend=Aer.get_backend('qasm_simulator'), shots=1024).result().get_counts(qc)
      3 counts

/qiskit/circuit/quantumcircuit.py in from_qasm_str(qasm_str)
   2362         """
   2363         qasm = Qasm(data=qasm_str)
-> 2364         return _circuit_from_qasm(qasm)
   2365 
   2366     @property

/qiskit/circuit/quantumcircuit.py in _circuit_from_qasm(qasm)
   4695     from qiskit.converters import dag_to_circuit
   4696 
-> 4697     ast = qasm.parse()
   4698     dag = ast_to_dag(ast)
   4699     return dag_to_circuit(dag)

/qiskit/qasm/qasm.py in parse(self)
     51         with QasmParser(self._filename) as qasm_p:
     52             qasm_p.parse_debug(False)
---> 53             return qasm_p.parse(self._data)

/qiskit/qasm/qasmparser.py in parse(self, data)
   1138     def parse(self, data):
   1139         """Parse some data."""
-> 1140         self.parser.parse(data, lexer=self.lexer, debug=self.parse_deb)
   1141         if self.qasm is None:
   1142             raise QasmError("Uncaught exception in parser; " + "see previous messages for details.")

/ply/yacc.py in parse(self, input, lexer, debug, tracking, tokenfunc)
    331             return self.parseopt(input, lexer, debug, tracking, tokenfunc)
    332         else:
--> 333             return self.parseopt_notrack(input, lexer, debug, tracking, tokenfunc)
    334 
    335 

/ply/yacc.py in parseopt_notrack(self, input, lexer, debug, tracking, tokenfunc)
   1118                             del symstack[-plen:]
   1119                             self.state = state
-> 1120                             p.callable(pslice)
   1121                             del statestack[-plen:]
   1122                             symstack.append(sym)

/qiskit/qasm/qasmparser.py in p_gate_decl_2(self, program)
    613             )
    614         self.pop_scope()
--> 615         self.update_symtab(program[0])
    616 
    617     def p_gate_scope(self, _):

/qiskit/qasm/qasmparser.py in update_symtab(self, obj)
     70         if obj.name in self.current_symtab:
     71             prev = self.current_symtab[obj.name]
---> 72             raise QasmError(
     73                 "Duplicate declaration for",
     74                 obj.type + " '" + obj.name + "' at line",

QasmError: "Duplicate declaration for gate 'rzx' at line 4, file .\nPrevious occurrence at line 3, file "

What should happen?

I would have expected the final qasm to define a "general" rzx gate and reuse it:

gate rzx(param0) q0,q1 { h q1; cx q0,q1; rz(param0) q1; cx q0,q1; h q1; }
gate ecr q0,q1 { rzx(pi/4) q0,q1; x q0; rzx(-pi/4) q0,q1; }

Any suggestions?

The problem is in the qasm() function of the QuantumCircuit, in particular it fails to convert ecr which has two identical rzx instructions. I noticed that by having two RZXGate directly in the circuit, gives a correct result (even if it seems a bit odd, given that the parametrization is not used):

qc = QuantumCircuit(2, 2, name='qc')
qc.rzx(0.78, 0, 1)
qc.rzx(-0.78, 0, 1)

OPENQASM 2.0;
include "qelib1.inc";
gate rzx(param0) q0,q1 { h q1; cx q0,q1; rz(0.78) q1; cx q0,q1; h q1; }
gate rzx_140348495844112(param0) q0,q1 { h q1; cx q0,q1; rz(-0.78) q1; cx q0,q1; h q1; }
qreg q[2];
creg c[2];
creg meas[2];
rzx(0.78) q[0],q[1];
rzx_140348495844112(-0.78) q[0],q[1];
barrier q[0],q[1];
measure q[0] -> meas[0];
measure q[1] -> meas[1];

Maybe, as a quick workaround, we should add this part below within the function _add_sub_instruction_to_existing_composite_circuits so that the two rzx gates get two identifiers.
https://github.com/Qiskit/qiskit-terra/blob/ee0d76052411230848ab2830c5741c14c2450439/qiskit/circuit/quantumcircuit.py#L1689-L1693

Insert it above this line:
https://github.com/Qiskit/qiskit-terra/blob/ee0d76052411230848ab2830c5741c14c2450439/qiskit/circuit/quantumcircuit.py#L4738

Although, the most elegant solution would be to actually get a parametrizable rzx

gate rzx(param0) q0,q1 { h q1; cx q0,q1; rz(param0) q1; cx q0,q1; h q1; }

Looking forward to hearing your feedback on this odd situation, thanks in advance

[jakelishman]

Thanks for the report! Right now there are systemic failures in the QASM 2 and QASM 3 exporters with slightly exotic parameterised gates, which stem from design choices in how the Instruction class stores its definition. If a gate that the exporter doesn't have special handling for takes a parameter, but the first time it is encountered the parameter is already bound, there's no way for the exporter to retrive the general definition from the instruction instance.

The QASM 3 exporter should at least be "smart" enough to munge the gate names to avoid a collision (see #7335 and its workaround in #7336), but the QASM 2 exporter has had this problem for a long time, so we didn't prioritise folding in a change during the 0.19 rush when the QASM 3 exporter was written.

The new approach to parameters that we're moving towards (see #7624) will let us solve this problem. I'll update #7335 to mention the QASM 2 exporter as well, but then I'll close this as duplicate of that, just to make it a little easier for us to track.