Exporting a kernel into QPU compatible OpenQASM2 file

[ashsaki]

Suppose I have a kernel with high-level gates (that may not adhere to target QPU's basis gates and qubit connectivity). I want to get a QPU-compatible version of the kernel in OpenQASM2 format.

For example, I was trying to compile the following code for IQM Adonis QPU.

#include <cudaq.h>
#include <fstream>

struct adonis_kernel {
  // Adonis QPU.
  //       QB1
  //        |
  // QB2 - QB3 - QB4
  //        |
  //       QB5

  void operator()() __qpu__ {
    cudaq::qvector q(5);
    x(q[0]);

    for (int i = 0; i < 4; i++) {
      x<cudaq::ctrl>(q[i], q[i + 1]);
    }
    x<cudaq::ctrl>(q[0], q[2], q[1]);
    auto result = mz(q);
  }
};

int main() {
  // auto counts = cudaq::sample(adonis_kernel{});
  // counts.dump();
}

The nvq++ compiler flags I used:

nvq++ --emulate --target iqm --iqm-machine Adonis iqm_simple.cpp -o iqm_simple.x -v --save-temps

The -v shows that the .cpp code is converted to a quake (.qke) IR. Then, cudaq-opt runs on the .qke file, and the cudaq-opt use a pipeline of passes. Then, cudaq-translate is applied.

I tried to manually run the cudaq-translate with --convert-to=qasm2, But, it threw me an error.

From browsing the source code, one can see pipelines of optimization passes and tools for this conversion.

• How can one use the capabilities available in cudaq to generate QPU compatible OpenQASM representation of the a kernel?

  -- While I used IQM as an example for trial purpose, it will be helpful if the solution can can take user-defined basis gates and qubit connectivity.
  -- Sample or Observe the kernel is not a necessity now for me.

[annagrin]

Thank you ashsaki for the example, that discovered an issue with our current pipeline (some missing opts in conversion to OpenQASM2).

Here is a workaround that works for this concrete example, please let us know if you have any other failing ones:

cudaq-quake iqm_simple.cpp | cudaq-opt --pass-pipeline="builtin.module(func.func(memtoreg{quantum=0}),cc-loop-unroll)" | cudaq-translate --convert-to=openqasm2

[annagrin]

Filed an issue: #2176

[ashsaki]

Thanks!

To add to your answer, I found the following is needed to address basis conversion and routing on a restricted qubit connectivity.

cudaq-opt -pass-pipeline='builtin.module(func.func(multicontrol-decomposition),basis-conversion{basis=h,rz,x,x(1)},func.func(add-dealloc,combine-quantum-alloc,canonicalize,factor-quantum-alloc,memtoreg,qubit-mapping{device=path(10)},regtomem),basis-conversion{basis=h,rz,x,x(1)})'  out.qke > optimized.qke

• The qubit connectivity can be user-defined following a certain format (e.g., IQM Adonis).
• I found that basis cannot take any arbitrary universal set yet.