Some target tests don't work with translation to OpenQASM 2.0

[kshyatt-aws]

Required prerequisites

• Consult the security policy. If reporting a security vulnerability, do not report the bug using this form. Use the process described in the policy to report the issue.
• Make sure you've read the documentation. Your issue may be addressed there.
• Search the issue tracker to verify that this hasn't already been reported. +1 or comment there if it has.
• If possible, make a PR with a failing test to give us a starting point to work on!

Describe the bug

Creating a new backend which uses OpenQASM 2.0 as the IR breaks several of the targettests/execution tests, including:

• custom_operation_basic.cpp - custom ops aren't supported in OQ2
• custom_operation_adj.cpp
• custom_operation_ctrl.cpp
  These 3 fail with:

+ /usr/local/llvm/bin/FileCheck /workspaces/cuda-quantum/targettests/execution/custom_operation_adj.cpp
loc("-":30:1): error: 'cc.global' op unable to translate op to OpenQASM 2.0
terminate called after throwing an instance of 'std::runtime_error'
  what():  Could not successfully translate to qasm2.
FileCheck error: '<stdin>' is empty.
FileCheck command line:  /usr/local/llvm/bin/FileCheck /workspaces/cuda-quantum/targettests/execution/custom_operation_adj.cpp

Also failing are:

• graph_coloring-1.cpp
• graph_coloring.cpp
  with:

+ /usr/local/llvm/bin/FileCheck /workspaces/cuda-quantum/targettests/execution/graph_coloring-1.cpp
loc(callsite("graph_coloring-1.cpp":42:3 at "graph_coloring-1.cpp":86:3)): error: cannot convert operation to OpenQASM 2.0.
terminate called after throwing an instance of 'std::runtime_error'
  what():  Could not successfully translate to qasm2.
FileCheck error: '<stdin>' is empty.
FileCheck command line:  /usr/local/llvm/bin/FileCheck /workspaces/cuda-quantum/targettests/execution/graph_coloring-1.cpp

and

loc(callsite(callsite("graph_coloring.cpp":41:3 at "graph_coloring.cpp":59:3) at "graph_coloring.cpp":105:3)): error: cannot create adjoint for this operation.
terminate called after throwing an instance of 'std::runtime_error'
  what():  Could not successfully translate to qasm2.
FileCheck error: '<stdin>' is empty.

• int8_t.cpp
• int8_t_free_func.cpp
• load_value.cpp
• swap_gate.cpp
• variable_size_qreg.cpp
  with

+ /workspaces/cuda-quantum/build/targettests/execution/Output/int8_t.cpp.tmp
+ /usr/local/llvm/bin/FileCheck /workspaces/cuda-quantum/targettests/execution/int8_t.cpp
loc("int8_t.cpp":26:5): error: 'cc.alloca' op unable to translate op to OpenQASM 2.0
terminate called after throwing an instance of 'std::runtime_error'
  what():  Could not successfully translate to qasm2.
FileCheck error: '<stdin>' is empty.

• qspan_slices.cpp
  with

+ /usr/local/llvm/bin/FileCheck /workspaces/cuda-quantum/targettests/execution/qspan_slices.cpp
loc(callsite("qspan_slices.cpp":27:19 at "qspan_slices.cpp":35:3)): error: 'quake.subveq' op unable to translate op to OpenQASM 2.0
terminate called after throwing an instance of 'std::runtime_error'
  what():  Could not successfully translate to qasm2.
FileCheck error: '<stdin>' is empty.

• sudoku_2x2.cpp
• sudoku_2x2-1.cpp
• sudoku_2x2-reg_name.cpp
• sudoku_2x2-bit_names.cpp
  with

loc(callsite("sudoku_2x2-1.cpp":27:3 at "sudoku_2x2-1.cpp":49:5)): error: cannot convert operation to OpenQASM 2.0.
terminate called after throwing an instance of 'std::runtime_error'
  what():  Could not successfully translate to qasm2.
FileCheck error: '<stdin>' is empty.

and

• phase_estimation.cpp
  with

+ CUDAQ_DUMP_JIT_IR=1
+ /workspaces/cuda-quantum/build/targettests/execution/Output/phase_estimation.cpp.tmp
/workspaces/cuda-quantum/build/targettests/execution/Output/phase_estimation.cpp.script: line 2: 40075 Aborted                 CUDAQ_DUMP_JIT_IR=1 /workspaces/cuda-quantum/build/targettests/execution/Output/phase_estimation.cpp.tmp &> phase_estimation.cpp.ir

(didn't get much info from this about what the failure cause was).

My lowering pass is:

platform-lowering-config: "const-prop-complex,canonicalize,cse,lift-array-value,state-prep,unitary-synthesis,canonicalize,apply-op-specialization,aggressive-early-inlining,unrolling-pipeline,func.func(lower-to-cfg),canonicalize,func.func(multicontrol-decomposition),func.func(memtoreg{quantum=0})"

Steps to reproduce the bug

Create a new REST backend with the above lowering config in its backend.yml and use that as the target in the targettests/execution tests.

Expected behavior

Tests should ideally pass or offer a more descriptive error message in some cases.

Is this a regression? If it is, put the last known working version (or commit) here.

Not a regression

Environment

• CUDA Quantum version: 0.8, CUDA 11.8
• Python version: N/A
• C++ compiler: gcc 11.4.0
• Operating system: Linux 01c0ebb026db 6.6.12-linuxkit #1 SMP Thu Feb 8 06:36:34 UTC 2024 aarch64 aarch64 aarch64 GNU/Linux (the CUDA-Quantum dev container on a Mac M2)

Suggestions

No response

[kshyatt-aws]

cc @khalatepradnya

[kshyatt-aws]

Some more testing reveals that exp_pauli.cpp also fails with:

+ /usr/local/llvm/bin/FileCheck /workspaces/cuda-quantum/targettests/execution/exp_pauli.cpp
loc("exp_pauli.cpp":30:30): error: 'cc.string_literal' op unable to translate op to OpenQASM 2.0
terminate called after throwing an instance of 'std::runtime_error'
  what():  Could not successfully translate to qasm2.

and test-6.cpp with:

+ /usr/local/llvm/bin/FileCheck /workspaces/cuda-quantum/targettests/execution/test-6.cpp
loc("test-6.cpp":42:17): error: 'quake.mx' op unable to translate op to OpenQASM 2.0
terminate called after throwing an instance of 'std::runtime_error'
  what():  Could not successfully translate to qasm2.
FileCheck error: '<stdin>' is empty.

(maybe related to #2219)?

[kshyatt-aws]

The error: 'cc.alloca' op unable to translate op to OpenQASM 2.0 has been resolved by adding func.func(const-prop-complex,canonicalize,cse,lift-array-alloc),globalize-array-values to the pipeline lowering config, thank you to @khalatepradnya for the suggestion. The sudoku examples are still failing at this gate:

z<cudaq::ctrl>(qubits[0], qubits[1], qubits[2], qubits[3]);

Because gate modfiiers aren't supported in OQ2. However, I'm wondering if another compiler pass could lower this to gates which are supported (e.g. SWAP, CX, and friends)?