CUDA quantum fails to render nested kernel calls

[orclassiq]

Required prerequisites

• 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.

Describe the bug

Calling cudaq::kernel_builder::to_quake with nested kernel calls fails because the serializer can't find the innermost kernel symbol.

Steps to reproduce the bug

The output of the following program:

#include <cudaq.h>

#include <iostream>

int main() {
  auto [kernel1, qubit1] = cudaq::make_kernel<cudaq::qubit>();
  std::cout << "kernel1.name " << kernel1.name() << std::endl;
  std::cout << kernel1.to_quake();

  auto [kernel2, qubit2] = cudaq::make_kernel<cudaq::qubit>();
  kernel2.call(kernel1, qubit2);
  std::cout << "kernel2.name " << kernel2.name() << std::endl;
  std::cout << kernel2.to_quake();

  auto kernel3 = cudaq::make_kernel();
  auto qreg3 = kernel3.qalloc(1);
  auto qubit3 = qreg3[0];
  kernel3.call(kernel2, qubit3);
  std::cout << "kernel3.name " << kernel3.name() << std::endl;
  std::cout << kernel3.to_quake();
}

Is:

kernel1.name __nvqppBuilderKernel_367535629127
module {
  func.func @__nvqpp__mlirgen____nvqppBuilderKernel_367535629127(%arg0: !quake.ref) {
    return
  }
}
kernel2.name __nvqppBuilderKernel_093606261879
module {
  func.func @__nvqpp__mlirgen____nvqppBuilderKernel_093606261879(%arg0: !quake.ref) {
    call @__nvqpp__mlirgen____nvqppBuilderKernel_367535629127(%arg0) : (!quake.ref) -> ()
    return
  }
  func.func @__nvqpp__mlirgen____nvqppBuilderKernel_367535629127(%arg0: !quake.ref) {
    return
  }
}
kernel3.name __nvqppBuilderKernel_202375922897
loc("-":3:5): error: 'func.call' op '__nvqpp__mlirgen____nvqppBuilderKernel_367535629127' does not reference a valid function
terminate called after throwing an instance of 'std::runtime_error'
  what():  cudaq::builder failed to JIT compile the Quake representation.
Aborted

The code successfully serializes the first two kernels, and but fails to serialize the third.
The missing symbol is the same symbol used by the other kernels to describe the innermost kernel.

Note, the code is compiled without any additional flags on the latest docker image (ghcr.io/nvidia/cuda-quantum@sha256:92e55d523b5124a22868472ef5bf532a7647d6173f5d12a371dd0f46f8570330).

Expected behavior

The above program shouldn't fail. Instead, the last "section" should result in an output similar to the following text:

kernel1.name __nvqppBuilderKernel_367535629127
module {
  func.func @__nvqpp__mlirgen____nvqppBuilderKernel_367535629127(%arg0: !quake.ref) {
    return
  }
}
kernel2.name __nvqppBuilderKernel_093606261879
module {
  func.func @__nvqpp__mlirgen____nvqppBuilderKernel_093606261879(%arg0: !quake.ref) {
    call @__nvqpp__mlirgen____nvqppBuilderKernel_367535629127(%arg0) : (!quake.ref) -> ()
    return
  }
  func.func @__nvqpp__mlirgen____nvqppBuilderKernel_367535629127(%arg0: !quake.ref) {
    return
  }
}
kernel3.name __nvqppBuilderKernel_202375922897
module {
  func.func @__nvqpp__mlirgen____nvqppBuilderKernel_202375922897() {
    %0 = quake.alloca !quake.veq<1>
    %1 = quake.extract_ref %0[0] : (!quake.veq<1>) -> !quake.ref
    call @__nvqpp__mlirgen____nvqppBuilderKernel_093606261879(%1) : (!quake.ref) -> ()
    return
  }
  func.func @__nvqpp__mlirgen____nvqppBuilderKernel_093606261879(%arg0: !quake.ref) {
    call @__nvqpp__mlirgen____nvqppBuilderKernel_367535629127(%arg0) : (!quake.ref) -> ()
    return
  }
  func.func @__nvqpp__mlirgen____nvqppBuilderKernel_367535629127(%arg0: !quake.ref) {
    return
  }
}

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

Not a regression

Environment

• Docker image: I've used the latest version of the cuda-quantum image: ghcr.io/nvidia/cuda-quantum@sha256:92e55d523b5124a22868472ef5bf532a7647d6173f5d12a371dd0f46f8570330.
  I couldn't find the correct the exact CUDA Quantum version used in this image (either the version or the commit hash).
• CUDA Quantum version: I've used the builtin version of the CUDA Quantum library in the docker image
• C++ compiler: I've used the builtin nvq++ in the docker image
• Operating system: I've used the docker image

Suggestions

No response

[amccaskey]

Thanks for filing this @orclassiq. I just issued a PR for the fix (#334).