# Multi-core computation with lightning.qubit and OpenMP support **URL:** \ **Category:** PennyLane Lightning\ **Created:** [July 18, 2025, 9:19am UTC](https://discuss.pennylane.ai/t/multi-core-computation-with-lightning-qubit-and-openmp-support/8776 "2025-07-18T09:19:28Z")\ **Posts on this page:** 16\ **Page:** 1
**Author:** ![minn-bj](https://avatars.discourse-cdn.com/v4/letter/m/7cd45c/32.png) [@minn-bj](https://discuss.pennylane.ai/u/minn-bj)\ **Post date:** [July 18, 2025, 9:19am UTC](https://discuss.pennylane.ai/t/multi-core-computation-with-lightning-qubit-and-openmp-support/8776/1 "2025-07-18T09:19:28Z")
Hello there, I searched a lot but can not find a proper explanation of how multi-core parallelization is possible in specific situations. I am using the JAX interface in combination with `catalyst.vmap` and `qml.qjit` as well as the `lightning.qubit` with OpenMP support. While testing different setups I recognized that the `qml.probs` measurement does not properly use multiple cores while the measurement in [How to optimize a QML model using Catalyst and quantum just-in-time (QJIT) compilation](https://pennylane.ai/qml/demos/tutorial_How_to_optimize_QML_model_using_JAX_catalyst_and_Optax) `qml.expval(qml.sum(*[qml.PauliZ(i) for i in range(n_wires)]))` When I am using `qml.probs` in that pennylane demo as measurement only one core runs with 100%. There are other pods opened by pennylane but they are only used in a range of ~2-8 %. Is there a way to truly parallelize over multiple cores with qml.probs or use multiple cores for gradient calculations? Which measurements can utilize multi-core parallelization? **A personal comment:** As far as I can say, such information can’t be found in the pennylane documentation. To investigate this topic without proper documentation is very time consuming and can be quite frustrating. I could imagine that many users are interested in this topic as the community starts to develop hybrid algorithms for real world scenarios that often need more computational resources. Some additional documentation regarding this topic could be interesting and beneficial for the community. Best regards minn-bj ---
**Author:** ![mlxd](https://yyz2.discourse-cdn.com/flex012/user_avatar/discuss.pennylane.ai/mlxd/32/526_2.png) [@mlxd](https://discuss.pennylane.ai/u/mlxd)\ **Post date:** [July 18, 2025, 6:05pm UTC](https://discuss.pennylane.ai/t/multi-core-computation-with-lightning-qubit-and-openmp-support/8776/2 "2025-07-18T18:05:32Z")
Hi @minn-bj Thank you for the feedback on the documentation for the multicore execution. We’ll chat internally about making this more visible in future. On the topic of `qml.probs`, since this is a simple loop over a vector, with limited compute per step, multithreading may actually hurt performance, rather than help. It is equivalent to a BLAS-1 like operation, which are inherently memory-bandwidth bound, and given the small size compared to, let’s say a BLAS-2 call (matrix-vector) or BLAS-3 (matrix-matrix) the serial execution (with implicit autovectorization) may be faster than spinning up a given thread. That being said, it is possible to explicitly force the use of OpenMP threading for the `qml.probs` kernels in LightningQubit by building with the options listed [here](https://docs.pennylane.ai/projects/lightning/en/latest/lightning_qubit/development/avx_kernels/kernel_tuning.html). In this case, you’ll need to install Lightning from source and provide an additional CMake argument `-DLQ_ENABLE_KERNEL_OMP=ON`. To do this onto an existing environment, you can ensure pip pulls the `sdist` release, and provide the CMake arguments directly as: ```bash # regular install of pennylane and catalyst python -m pip install pennylane pennylane-catalyst # adapted install of LightningQubit with the new build CMAKE_ARGS="-DLQ_ENABLE_KERNEL_OMP=ON" python -m pip \ install pennylane_lightning --no-binary "pennylane_lightning" \ --force-reinstall --no-cache-dir --verbose ``` While the above is definitely a mouthful, we tend to avoid turning this on by default for the above mentioned reasons. This will also enable multithreading support for all LightningQubit gates, which should thread over their application to the statevector. This can cause oversubscription when combined with multithreaded gradient workloads, so we designed the forward pass to be single threaded, allowing execution to be optimal for batched workloads. For HPC specific installs, we will generally validate workloads on the machine, and adapt the flags for said machines to yield the best runtime. Though, since this is a workload dependent and system dependent process, it isn’t something we’ve exposed much yet (though, we will aim to make this easier). I’d anticipate the above to yield better performance in regimes where `qml.probs` is evaluated on very large statevectors, and running on machines with high memory bandwidth, such as AMD Epycs or Xeons. In that way, the extra threading can help to saturate the available memory bus, and potentially yield better performance than serial, but it is still very likely to be workload dependent. As for where OpenMP acceleration is currently supported, the adjoint differentiation pipeline (turned on by default for LightningQubit) will scale across observables when using `qml.expval`. This scaling was used mostly with Hamiltionian expectation values, and offered a useful way to run parallel gradient evaluations. As this can spawn a lot of copies for large Hamiltonians, we also recommend the `batch_obs` keyword argument discussed [here](https://docs.pennylane.ai/projects/lightning/en/latest/lightning_qubit/device.html). You can also try out `lightning.kokkos` (`pip install pennylane-lightning-kokkos`), which should offer full OpenMP execution on all representative operations by default on Linux machines, though it will have less efficient use of vectorization than `lightning.qubit. Feel free to try out the above, and let us know if it helps any, and we’ll do our best to improve the documentation of the above. ---
**Author:** ![minn-bj](https://avatars.discourse-cdn.com/v4/letter/m/7cd45c/32.png) [@minn-bj](https://discuss.pennylane.ai/u/minn-bj)\ **Post date:** [July 21, 2025, 8:51am UTC](https://discuss.pennylane.ai/t/multi-core-computation-with-lightning-qubit-and-openmp-support/8776/3 "2025-07-21T08:51:56Z")
Hey @mlxd, Thank you for your detailed answer. I will try to follow your instructions :). I have a further question regarding the potential loss of performance due to multi-core execution. > Blockquote > I’d anticipate the above to yield better performance in regimes where `qml.probs` is evaluated on very large statevectors, and running on machines with high memory bandwidth, such as AMD Epycs or Xeons. In that way, the extra threading can help to saturate the available memory bus, and potentially yield better performance than serial, but it is still very likely to be workload dependent. I want to execute a 16 Qubit quantum circuit with an intermediate number of gates, a batch size of ~ 1000 and roughly 6-10 read out qubits for `qml.probs`. From my perspective this could be an attractive use case for multi-core execution if the forward and backward pass can be parallelized. Would you agree or did I misunderstood something and are both passes in fact parallelizable? My Hardware: 2x Intel Xeon Gold 6132 CPU and ~100 GB of Ram (in principle up to 300 GB if necessary). In addition I have access to 2 Nvidia V100 16GB. What would be your best guess to utilize this hardware and optimize performance? (I already gave lightning.gpu a chance but it is surprisingly slow compared to the lightning.qubit). Is that behavior expected even with large batch sizes? Best regards minn-bj ---
**Author:** ![minn-bj](https://avatars.discourse-cdn.com/v4/letter/m/7cd45c/32.png) [@minn-bj](https://discuss.pennylane.ai/u/minn-bj)\ **Post date:** [July 21, 2025, 3:00pm UTC](https://discuss.pennylane.ai/t/multi-core-computation-with-lightning-qubit-and-openmp-support/8776/4 "2025-07-21T15:00:05Z")
## Here are some benchmark results for a circuit with 36 Gates. Only the device changes **N=12 Qubits** **Batch\_size=1000** **Single Core JIT lightning.qubit:** - Forward: 0.16 sec - Update params: 6.3 sec - 100% usage **JIT lightning.gpu:** - Forward: 3.9 sec - Update params: 146.8 sec - 50% usage - warning: 2025-07-21 14:21:21.104025: W external/xla/xla/service/gpu/nvptx\_compiler.cc:760] The NVIDIA driver’s CUDA version is 12.4 which is older than the ptxas CUDA version (12.9.86). Because the driver is older than the ptxas version, XLA is disabling parallel compilation, which may slow down compilation. You should update your NVIDIA driver or use the NVIDIA-provided CUDA forward compatibility packages. **N=12 Qubits** **Batch\_size=500** **Single Core JIT lightning.qubit:** - Forward: 0.06 sec - Update params: 2.8 sec - 100% usage **JIT lightning.gpu:** - Forward: 1.9 sec - Update params: 79.0 sec - 50% usage - warning: 2025-07-21 14:21:21.104025: W external/xla/xla/service/gpu/nvptx\_compiler.cc:760] The NVIDIA driver’s CUDA version is 12.4 which is older than the ptxas CUDA version (12.9.86). Because the driver is older than the ptxas version, XLA is disabling parallel compilation, which may slow down compilation. You should update your NVIDIA driver or use the NVIDIA-provided CUDA forward compatibility packages. ---
**Author:** ![CatalinaAlbornoz](https://yyz2.discourse-cdn.com/flex012/user_avatar/discuss.pennylane.ai/catalinaalbornoz/32/1196_2.png) [@CatalinaAlbornoz](https://discuss.pennylane.ai/u/CatalinaAlbornoz)\ **Post date:** [July 22, 2025, 5:20pm UTC](https://discuss.pennylane.ai/t/multi-core-computation-with-lightning-qubit-and-openmp-support/8776/5 "2025-07-22T17:20:18Z")
Hi @minn-bj , I noticed a few things. On one hand, 16 qubits is below our recommendation for `lightning.gpu`. It’s normal to see `lightning.qubit` performing better than `lightning.gpu` when using less than 20 qubits. You can find this info and more in our [performance page](https://pennylane.ai/performance#which-simulator). On the other hand you have a warning saying that your NVIDIA driver is old meaning there could be slowdowns. This could further explain the performance difference. Finally, 12 qubits with 36 gates shouldn’t be a computationally intensive problem, but having thousands of datapoints is. Every time you try to load new data you have overheads so this looks to be the core of your speed problems. Quantum computers are known to be bad at handling large amounts of input data and parameters, so the solution here is to limit yourself to only a few datapoints. I hope this helps! ---
**Author:** ![minn-bj](https://avatars.discourse-cdn.com/v4/letter/m/7cd45c/32.png) [@minn-bj](https://discuss.pennylane.ai/u/minn-bj)\ **Post date:** [July 23, 2025, 8:08am UTC](https://discuss.pennylane.ai/t/multi-core-computation-with-lightning-qubit-and-openmp-support/8776/6 "2025-07-23T08:08:04Z")
Hey @CatalinaAlbornoz, > Blockquote > On one hand, 16 qubits is below our recommendation for `lightning.gpu`. It’s normal to see `lightning.qubit` performing better than `lightning.gpu` when using less than 20 qubits. You can find this info and more in our [performance page](https://pennylane.ai/performance#which-simulator). Yeah I already thought that this could be on issue. My hope was that lightning.gpu potentially parallelizes the computation efficiently since the batch size is quite large. > Blockquote > On the other hand you have a warning saying that your NVIDIA driver is old meaning there could be slowdowns. This could further explain the performance difference. I am already working on this, I just thought you could have some experience with this. > Blockquote > Finally, 12 qubits with 36 gates shouldn’t be a computationally intensive problem, but having thousands of datapoints is. Every time you try to load new data you have overheads so this looks to be the core of your speed problems. Quantum computers are known to be bad at handling large amounts of input data and parameters, so the solution here is to limit yourself to only a few datapoints. Unfortunately, limiting the input data is (at the moment) not really possible. Anyway, I was wondering how the default qubit (torch interface and executed on a GPU) can be quite fast (Forward: 0.064 sec, Update params: 0.22 sec) but the other gpu accelerators like lightning.gpu and kokkos gpu are much slower. My guess was that this is related to the adjoint (lightning.gpu and lighting.kokkos) diff\_method instead of the backprop diff\_method (default.qubit). Is that correct? Is there a faster alternative to the mentioned default.qubit on a gpu for the scenario explained above (maybe a jax version where one can use backprop and jit the execution on a GPU)? It is quite hard to guess which simulator is best for my scenario since I don’t really know what happens inside of the different simulators. I also recognized that the update time (gradient calc.) increases approx. linearly with the batch size. Is this behavior expected? ---
**Author:** ![mlxd](https://yyz2.discourse-cdn.com/flex012/user_avatar/discuss.pennylane.ai/mlxd/32/526_2.png) [@mlxd](https://discuss.pennylane.ai/u/mlxd)\ **Post date:** [July 23, 2025, 1:49pm UTC](https://discuss.pennylane.ai/t/multi-core-computation-with-lightning-qubit-and-openmp-support/8776/7 "2025-07-23T13:49:34Z")
Hi @minn-bj To answer your above questions — for your use cases, with moderate qubit count and batch dimensions, the `default.qubit` pipelines will most always be faster. The Lightning devices are generally optimised for HPC (large qubit counts, very deep circuits), and so do not have batch execution built in to the backends for gate, as would be the case in a classical ML context. We do this as for each new batch dimension an additional statevector is required, and since the `lightning.gpu` device is built for HPC workloads where memory availability is the largest constraint, we iterate over the batch dimension to avoid creating intermediate statevectors where possible. The adjoint diff gradient pipeline requires some copies here, but they should be limited, and still more efficient than a backpropagation pipeline’s memory use beyond a 20 qubit threshold. We may consider batching support for smaller qubit registers in future, but now this isn’t part of our plans. In this scenario, if you want to execute in the 10-16 qubit regime with batch dimensions, using `default.qubit` with the jax or torch backends and the backpropagation gradient pipeline would likely be the best pipeline for your needs, as this should offer the best performance. ---
**Author:** ![minn-bj](https://avatars.discourse-cdn.com/v4/letter/m/7cd45c/32.png) [@minn-bj](https://discuss.pennylane.ai/u/minn-bj)\ **Post date:** [July 25, 2025, 9:00am UTC](https://discuss.pennylane.ai/t/multi-core-computation-with-lightning-qubit-and-openmp-support/8776/8 "2025-07-25T09:00:05Z")
I just gave it a try, thanks again. Currently it uses multiple cores and works properly without jit and vmap. > Blockquote > #regular install of pennylane and catalyst > python -m pip install pennylane pennylane-catalyst > #adapted install of LightningQubit with the new build > CMAKE\_ARGS=“-DLQ\_ENABLE\_KERNEL\_OMP=ON” python -m pip > install pennylane\_lightning --no-binary “pennylane\_lightning” > –force-reinstall --no-cache-dir --verbose Unfortunately, its not possible to jit or vmap the computation since some issue with the compiler occures. Do you have some more information on this specific error? Error: ```auto --------------------------------------------------------------------------- CalledProcessError Traceback (most recent call last) File ~/.local/lib/python3.11/site-packages/catalyst/compiler.py:444, in Compiler.run_from_ir(self, ir, module_name, workspace) 443 print(f"[SYSTEM] {' '.join(cmd)}", file=self.options.logfile) --> 444 result = subprocess.run(cmd, check=True, capture_output=True, text=True) 445 if self.options.verbose or os.getenv("ENABLE_DIAGNOSTICS"): File /opt/conda/lib/python3.11/subprocess.py:571, in run(input, capture_output, timeout, check, *popenargs, **kwargs) 570 if check and retcode: --> 571 raise CalledProcessError(retcode, process.args, 572 output=stdout, stderr=stderr) 573 return CompletedProcess(process.args, retcode, stdout, stderr) CalledProcessError: Command '['/opt/conda/bin/catalyst', '-o', '/tmp/grad.loss_fn4z94rjgu/grad.loss_fn.ll', '--module-name', 'grad.loss_fn', '--workspace', '/tmp/grad.loss_fn4z94rjgu', '-verify-each=false', '--catalyst-pipeline', 'EnforceRuntimeInvariantsPass(split-multiple-tapes;builtin.module(apply-transform-sequence);inline-nested-module),HLOLoweringPass(canonicalize;func.func(chlo-legalize-to-hlo);stablehlo-legalize-to-hlo;func.func(mhlo-legalize-control-flow);func.func(hlo-legalize-to-linalg);func.func(mhlo-legalize-to-std);func.func(hlo-legalize-sort);convert-to-signless;canonicalize;scatter-lowering;hlo-custom-call-lowering;cse;func.func(linalg-detensorize{aggressive-mode});detensorize-scf;canonicalize),QuantumCompilationPass(annotate-function;lower-mitigation;lower-gradients;adjoint-lowering),BufferizationPass(one-shot-bufferize{dialect-filter=memref};inline;gradient-preprocess;gradient-bufferize;scf-bufferize;convert-tensor-to-linalg;convert-elementwise-to-linalg;arith-bufferize;empty-tensor-to-alloc-tensor;func.func(bufferization-bufferize);func.func(tensor-bufferize);catalyst-bufferize;func.func(linalg-bufferize);func.func(tensor-bufferize);quantum-bufferize;func-bufferize;func.func(finalizing-bufferize);canonicalize;gradient-postprocess;func.func(buffer-hoisting);func.func(buffer-loop-hoisting);func.func(buffer-deallocation);convert-arraylist-to-memref;convert-bufferization-to-memref;canonicalize;cp-global-memref),MLIRToLLVMDialect(expand-realloc;convert-gradient-to-llvm;memrefcpy-to-linalgcpy;func.func(convert-linalg-to-loops);convert-scf-to-cf;expand-strided-metadata;lower-affine;arith-expand;convert-complex-to-standard;convert-complex-to-llvm;convert-math-to-llvm;convert-math-to-libm;convert-arith-to-llvm;memref-to-llvm-tbaa;finalize-memref-to-llvm{use-generic-functions};convert-index-to-llvm;convert-catalyst-to-llvm;convert-quantum-to-llvm;emit-catalyst-py-interface;canonicalize;reconcile-unrealized-casts;gep-inbounds;register-inactive-callback),', '/tmp/grad.loss_fn4z94rjgu/tmpjp6o4e1l.mlir']' returned non-zero exit status 1. The above exception was the direct cause of the following exception: CompileError Traceback (most recent call last) Cell In[7], line 135 131 grad = catalyst.grad(loss_fn, method="fd") 133 ######################################################## --> 135 print(grad(params, data)['weights'].shape) 138 def update_step(params, opt_state, opt, data): 139 grads = grad(params, data) File ~/.local/lib/python3.11/site-packages/pennylane/logging/decorators.py:61, in log_string_debug_func..wrapper_entry(*args, **kwargs) 54 s_caller = "::L".join( 55 [str(i) for i in inspect.getouterframes(inspect.currentframe(), 2)[1][1:3]] 56 ) 57 lgr.debug( 58 f"Calling {f_string} from {s_caller}", 59 **_debug_log_kwargs, 60 ) ---> 61 return func(*args, **kwargs) File ~/.local/lib/python3.11/site-packages/catalyst/jit.py:594, in QJIT. __call__ (self, *args, **kwargs) 590 kwargs = {"static_argnums": self.compile_options.static_argnums, **kwargs} 592 return self.user_function(*args, **kwargs) --> 594 requires_promotion = self.jit_compile(args, **kwargs) 596 # If we receive tracers as input, dispatch to the JAX integration. 597 if any(isinstance(arg, jax.core.Tracer) for arg in tree_flatten(args)[0]): File ~/.local/lib/python3.11/site-packages/pennylane/logging/decorators.py:61, in log_string_debug_func..wrapper_entry(*args, **kwargs) 54 s_caller = "::L".join( 55 [str(i) for i in inspect.getouterframes(inspect.currentframe(), 2)[1][1:3]] 56 ) 57 lgr.debug( 58 f"Calling {f_string} from {s_caller}", 59 **_debug_log_kwargs, 60 ) ---> 61 return func(*args, **kwargs) File ~/.local/lib/python3.11/site-packages/catalyst/jit.py:670, in QJIT.jit_compile(self, args, **kwargs) 667 self.jaxpr, self.out_type, self.out_treedef, self.c_sig = self.capture(args, **kwargs) 669 self.mlir_module = self.generate_ir() --> 670 self.compiled_function, _ = self.compile() 672 self.fn_cache.insert(self.compiled_function, args, self.out_treedef, self.workspace) 674 elif self.compiled_function is not cached_fn.compiled_fn: 675 # Restore active state from cache. File ~/.local/lib/python3.11/site-packages/catalyst/debug/instruments.py:145, in instrument..wrapper(*args, **kwargs) 142 @functools.wraps(fn) 143 def wrapper(*args, **kwargs): 144 if not InstrumentSession.active: --> 145 return fn(*args, **kwargs) 147 with ResultReporter(stage_name, has_finegrained) as reporter: 148 self = args[0] File ~/.local/lib/python3.11/site-packages/pennylane/logging/decorators.py:61, in log_string_debug_func..wrapper_entry(*args, **kwargs) 54 s_caller = "::L".join( 55 [str(i) for i in inspect.getouterframes(inspect.currentframe(), 2)[1][1:3]] 56 ) 57 lgr.debug( 58 f"Calling {f_string} from {s_caller}", 59 **_debug_log_kwargs, 60 ) ---> 61 return func(*args, **kwargs) File ~/.local/lib/python3.11/site-packages/catalyst/jit.py:799, in QJIT.compile(self) 793 shared_object, llvm_ir = self.compiler.run_from_ir( 794 self.overwrite_ir, 795 str(self.mlir_module.operation.attributes["sym_name"]).replace('"', ""), 796 self.workspace, 797 ) 798 else: --> 799 shared_object, llvm_ir = self.compiler.run(self.mlir_module, self.workspace) 801 compiled_fn = CompiledFunction( 802 shared_object, func_name, restype, self.out_type, self.compile_options 803 ) 805 return compiled_fn, llvm_ir File ~/.local/lib/python3.11/site-packages/pennylane/logging/decorators.py:61, in log_string_debug_func..wrapper_entry(*args, **kwargs) 54 s_caller = "::L".join( 55 [str(i) for i in inspect.getouterframes(inspect.currentframe(), 2)[1][1:3]] 56 ) 57 lgr.debug( 58 f"Calling {f_string} from {s_caller}", 59 **_debug_log_kwargs, 60 ) ---> 61 return func(*args, **kwargs) File ~/.local/lib/python3.11/site-packages/catalyst/compiler.py:486, in Compiler.run(self, mlir_module, *args, **kwargs) 470 @debug_logger 471 def run(self, mlir_module, *args, **kwargs): 472 """Compile an MLIR module to a shared object. 473 474 .. note:: (...) 483 (str): filename of shared object 484 """ --> 486 return self.run_from_ir( 487 mlir_module.operation.get_asm( 488 binary=False, print_generic_op_form=False, assume_verified=True 489 ), 490 str(mlir_module.operation.attributes["sym_name"]).replace('"', ""), 491 *args, 492 **kwargs, 493 ) File ~/.local/lib/python3.11/site-packages/pennylane/logging/decorators.py:61, in log_string_debug_func..wrapper_entry(*args, **kwargs) 54 s_caller = "::L".join( 55 [str(i) for i in inspect.getouterframes(inspect.currentframe(), 2)[1][1:3]] 56 ) 57 lgr.debug( 58 f"Calling {f_string} from {s_caller}", 59 **_debug_log_kwargs, 60 ) ---> 61 return func(*args, **kwargs) File ~/.local/lib/python3.11/site-packages/catalyst/compiler.py:451, in Compiler.run_from_ir(self, ir, module_name, workspace) 449 print(result.stderr.strip(), file=self.options.logfile) 450 except subprocess.CalledProcessError as e: # pragma: nocover --> 451 raise CompileError(f"catalyst failed with error code {e.returncode}: {e.stderr}") from e 453 if os.path.exists(output_ir_name): 454 with open(output_ir_name, "r", encoding="utf-8") as f: CompileError: catalyst failed with error code 1: Compilation failed: grad.loss_fn:221:7: error: 'quantum.device' op requires attribute 'device_name' "quantum.device"(%0) {kwargs = "{'shots': 0, 'mcmc': False, 'num_burnin': 0, 'kernel_name': None}", lib = "/home/jovyan/.local/lib/python3.11/site-packages/pennylane_lightning/liblightning_qubit_catalyst.so", name = "LightningSimulator"} : (i64) -> () ^ grad.loss_fn:221:7: note: see current operation: "quantum.device"(%1) <{kwargs = "{'shots': 0, 'mcmc': False, 'num_burnin': 0, 'kernel_name': None}", lib = "/home/jovyan/.local/lib/python3.11/site-packages/pennylane_lightning/liblightning_qubit_catalyst.so"}> {name = "LightningSimulator"} : (i64) -> () grad.loss_fn: grad.loss_fn:1:8: error: expected 'module asm' module @grad.loss_fn { ^ Failed to parse module as LLVM or MLIR source ``` ---
**Author:** ![CatalinaAlbornoz](https://yyz2.discourse-cdn.com/flex012/user_avatar/discuss.pennylane.ai/catalinaalbornoz/32/1196_2.png) [@CatalinaAlbornoz](https://discuss.pennylane.ai/u/CatalinaAlbornoz)\ **Post date:** [July 25, 2025, 9:37pm UTC](https://discuss.pennylane.ai/t/multi-core-computation-with-lightning-qubit-and-openmp-support/8776/9 "2025-07-25T21:37:55Z")
Thanks for sharing this @minn-bj . We’ll take a look and get back to you next week. ---
**Author:** ![CatalinaAlbornoz](https://yyz2.discourse-cdn.com/flex012/user_avatar/discuss.pennylane.ai/catalinaalbornoz/32/1196_2.png) [@CatalinaAlbornoz](https://discuss.pennylane.ai/u/CatalinaAlbornoz)\ **Post date:** [July 28, 2025, 2:54pm UTC](https://discuss.pennylane.ai/t/multi-core-computation-with-lightning-qubit-and-openmp-support/8776/10 "2025-07-28T14:54:42Z")
Hi @minn-bj , Can you please share a minimal reproducible example so that we can try to reproduce this issue? It looks like there’s an issue with the loss function. ---
**Author:** ![minn-bj](https://avatars.discourse-cdn.com/v4/letter/m/7cd45c/32.png) [@minn-bj](https://discuss.pennylane.ai/u/minn-bj)\ **Post date:** [July 29, 2025, 8:07am UTC](https://discuss.pennylane.ai/t/multi-core-computation-with-lightning-qubit-and-openmp-support/8776/11 "2025-07-29T08:07:26Z")
Hey @CatalinaAlbornoz. here is a minimal example and the corresponding error msg. The error msg is different since the script wants to execute the circuit, not the grad. Thank you for your support. ```auto import os os.environ["JAX_PLATFORMS"] = "cpu" os.environ["OMP_PROC_BIND"] = 'spread' os.environ["OMP_PLACES"] = "threads" import pennylane as qml import jax import catalyst n_wires = 8 batch_size=100 data=qml.numpy.random.rand(batch_size, n_wires) dev = qml.device("lightning.qubit", wires=n_wires, batch_obs=True) @qml.qnode(dev, interface='jax') def circuit(data): @qml.for_loop(0,n_wires,1) def encode(i): qml.RY(data[i], wires=i) encode() return qml.probs(wires=dev.wires)#[qml.expval(qml.PauliZ(i)) for i in range(n_wires)] circuit = catalyst.vmap(circuit, in_axes=(0)) circuit = qml.qjit(circuit, verbose=True) circuit(data) ``` error msg: ```auto --------------------------------------------------------------------------- CalledProcessError Traceback (most recent call last) File ~/.local/lib/python3.11/site-packages/catalyst/compiler.py:444, in Compiler.run_from_ir(self, ir, module_name, workspace) 443 print(f"[SYSTEM] {' '.join(cmd)}", file=self.options.logfile) --> 444 result = subprocess.run(cmd, check=True, capture_output=True, text=True) 445 if self.options.verbose or os.getenv("ENABLE_DIAGNOSTICS"): File /opt/conda/lib/python3.11/subprocess.py:571, in run(input, capture_output, timeout, check, *popenargs, **kwargs) 570 if check and retcode: --> 571 raise CalledProcessError(retcode, process.args, 572 output=stdout, stderr=stderr) 573 return CompletedProcess(process.args, retcode, stdout, stderr) CalledProcessError: Command '['/opt/conda/bin/catalyst', '-o', '/tmp/vmap.circuit45ibaj4b/vmap.circuit.ll', '--module-name', 'vmap.circuit', '--workspace', '/tmp/vmap.circuit45ibaj4b', '-verify-each=false', '--catalyst-pipeline', 'EnforceRuntimeInvariantsPass(split-multiple-tapes;builtin.module(apply-transform-sequence);inline-nested-module),HLOLoweringPass(canonicalize;func.func(chlo-legalize-to-hlo);stablehlo-legalize-to-hlo;func.func(mhlo-legalize-control-flow);func.func(hlo-legalize-to-linalg);func.func(mhlo-legalize-to-std);func.func(hlo-legalize-sort);convert-to-signless;canonicalize;scatter-lowering;hlo-custom-call-lowering;cse;func.func(linalg-detensorize{aggressive-mode});detensorize-scf;canonicalize),QuantumCompilationPass(annotate-function;lower-mitigation;lower-gradients;adjoint-lowering),BufferizationPass(one-shot-bufferize{dialect-filter=memref};inline;gradient-preprocess;gradient-bufferize;scf-bufferize;convert-tensor-to-linalg;convert-elementwise-to-linalg;arith-bufferize;empty-tensor-to-alloc-tensor;func.func(bufferization-bufferize);func.func(tensor-bufferize);catalyst-bufferize;func.func(linalg-bufferize);func.func(tensor-bufferize);quantum-bufferize;func-bufferize;func.func(finalizing-bufferize);canonicalize;gradient-postprocess;func.func(buffer-hoisting);func.func(buffer-loop-hoisting);func.func(buffer-deallocation);convert-arraylist-to-memref;convert-bufferization-to-memref;canonicalize;cp-global-memref),MLIRToLLVMDialect(expand-realloc;convert-gradient-to-llvm;memrefcpy-to-linalgcpy;func.func(convert-linalg-to-loops);convert-scf-to-cf;expand-strided-metadata;lower-affine;arith-expand;convert-complex-to-standard;convert-complex-to-llvm;convert-math-to-llvm;convert-math-to-libm;convert-arith-to-llvm;memref-to-llvm-tbaa;finalize-memref-to-llvm{use-generic-functions};convert-index-to-llvm;convert-catalyst-to-llvm;convert-quantum-to-llvm;emit-catalyst-py-interface;canonicalize;reconcile-unrealized-casts;gep-inbounds;register-inactive-callback),', '--verbose', '/tmp/vmap.circuit45ibaj4b/tmpfpiv5329.mlir']' returned non-zero exit status 1. The above exception was the direct cause of the following exception: CompileError Traceback (most recent call last) Cell In[1], line 36 33 circuit = catalyst.vmap(circuit, in_axes=(0)) 34 circuit = qml.qjit(circuit, verbose=True) ---> 36 circuit(data) File ~/.local/lib/python3.11/site-packages/pennylane/logging/decorators.py:61, in log_string_debug_func..wrapper_entry(*args, **kwargs) 54 s_caller = "::L".join( 55 [str(i) for i in inspect.getouterframes(inspect.currentframe(), 2)[1][1:3]] 56 ) 57 lgr.debug( 58 f"Calling {f_string} from {s_caller}", 59 **_debug_log_kwargs, 60 ) ---> 61 return func(*args, **kwargs) File ~/.local/lib/python3.11/site-packages/catalyst/jit.py:594, in QJIT. __call__ (self, *args, **kwargs) 590 kwargs = {"static_argnums": self.compile_options.static_argnums, **kwargs} 592 return self.user_function(*args, **kwargs) --> 594 requires_promotion = self.jit_compile(args, **kwargs) 596 # If we receive tracers as input, dispatch to the JAX integration. 597 if any(isinstance(arg, jax.core.Tracer) for arg in tree_flatten(args)[0]): File ~/.local/lib/python3.11/site-packages/pennylane/logging/decorators.py:61, in log_string_debug_func..wrapper_entry(*args, **kwargs) 54 s_caller = "::L".join( 55 [str(i) for i in inspect.getouterframes(inspect.currentframe(), 2)[1][1:3]] 56 ) 57 lgr.debug( 58 f"Calling {f_string} from {s_caller}", 59 **_debug_log_kwargs, 60 ) ---> 61 return func(*args, **kwargs) File ~/.local/lib/python3.11/site-packages/catalyst/jit.py:670, in QJIT.jit_compile(self, args, **kwargs) 667 self.jaxpr, self.out_type, self.out_treedef, self.c_sig = self.capture(args, **kwargs) 669 self.mlir_module = self.generate_ir() --> 670 self.compiled_function, _ = self.compile() 672 self.fn_cache.insert(self.compiled_function, args, self.out_treedef, self.workspace) 674 elif self.compiled_function is not cached_fn.compiled_fn: 675 # Restore active state from cache. File ~/.local/lib/python3.11/site-packages/catalyst/debug/instruments.py:145, in instrument..wrapper(*args, **kwargs) 142 @functools.wraps(fn) 143 def wrapper(*args, **kwargs): 144 if not InstrumentSession.active: --> 145 return fn(*args, **kwargs) 147 with ResultReporter(stage_name, has_finegrained) as reporter: 148 self = args[0] File ~/.local/lib/python3.11/site-packages/pennylane/logging/decorators.py:61, in log_string_debug_func..wrapper_entry(*args, **kwargs) 54 s_caller = "::L".join( 55 [str(i) for i in inspect.getouterframes(inspect.currentframe(), 2)[1][1:3]] 56 ) 57 lgr.debug( 58 f"Calling {f_string} from {s_caller}", 59 **_debug_log_kwargs, 60 ) ---> 61 return func(*args, **kwargs) File ~/.local/lib/python3.11/site-packages/catalyst/jit.py:799, in QJIT.compile(self) 793 shared_object, llvm_ir = self.compiler.run_from_ir( 794 self.overwrite_ir, 795 str(self.mlir_module.operation.attributes["sym_name"]).replace('"', ""), 796 self.workspace, 797 ) 798 else: --> 799 shared_object, llvm_ir = self.compiler.run(self.mlir_module, self.workspace) 801 compiled_fn = CompiledFunction( 802 shared_object, func_name, restype, self.out_type, self.compile_options 803 ) 805 return compiled_fn, llvm_ir File ~/.local/lib/python3.11/site-packages/pennylane/logging/decorators.py:61, in log_string_debug_func..wrapper_entry(*args, **kwargs) 54 s_caller = "::L".join( 55 [str(i) for i in inspect.getouterframes(inspect.currentframe(), 2)[1][1:3]] 56 ) 57 lgr.debug( 58 f"Calling {f_string} from {s_caller}", 59 **_debug_log_kwargs, 60 ) ---> 61 return func(*args, **kwargs) File ~/.local/lib/python3.11/site-packages/catalyst/compiler.py:486, in Compiler.run(self, mlir_module, *args, **kwargs) 470 @debug_logger 471 def run(self, mlir_module, *args, **kwargs): 472 """Compile an MLIR module to a shared object. 473 474 .. note:: (...) 483 (str): filename of shared object 484 """ --> 486 return self.run_from_ir( 487 mlir_module.operation.get_asm( 488 binary=False, print_generic_op_form=False, assume_verified=True 489 ), 490 str(mlir_module.operation.attributes["sym_name"]).replace('"', ""), 491 *args, 492 **kwargs, 493 ) File ~/.local/lib/python3.11/site-packages/pennylane/logging/decorators.py:61, in log_string_debug_func..wrapper_entry(*args, **kwargs) 54 s_caller = "::L".join( 55 [str(i) for i in inspect.getouterframes(inspect.currentframe(), 2)[1][1:3]] 56 ) 57 lgr.debug( 58 f"Calling {f_string} from {s_caller}", 59 **_debug_log_kwargs, 60 ) ---> 61 return func(*args, **kwargs) File ~/.local/lib/python3.11/site-packages/catalyst/compiler.py:451, in Compiler.run_from_ir(self, ir, module_name, workspace) 449 print(result.stderr.strip(), file=self.options.logfile) 450 except subprocess.CalledProcessError as e: # pragma: nocover --> 451 raise CompileError(f"catalyst failed with error code {e.returncode}: {e.stderr}") from e 453 if os.path.exists(output_ir_name): 454 with open(output_ir_name, "r", encoding="utf-8") as f: CompileError: catalyst failed with error code 1: Compilation failed: vmap.circuit:108:7: error: 'quantum.device' op requires attribute 'device_name' "quantum.device"(%0) {kwargs = "{'shots': 0, 'mcmc': False, 'num_burnin': 0, 'kernel_name': None}", lib = "/home/jovyan/.local/lib/python3.11/site-packages/pennylane_lightning/liblightning_qubit_catalyst.so", name = "LightningSimulator"} : (i64) -> () ^ vmap.circuit:108:7: note: diagnostic emitted with trace: Stack dump without symbol names (ensure you have llvm-symbolizer in your PATH or set the environment var `LLVM_SYMBOLIZER_PATH` to point to it): 0 catalyst 0x000000000ab8194b llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) + 43 1 catalyst 0x000000000a9a95d0 2 catalyst 0x000000000a9a9761 mlir::emitError(mlir::Location, llvm::Twine const&) + 17 3 catalyst 0x000000000aa00e19 mlir::Operation::emitError(llvm::Twine const&) + 25 4 catalyst 0x000000000aa01164 mlir::Operation::emitOpError(llvm::Twine const&) + 52 5 catalyst 0x000000000aa026ac mlir::OpState::emitOpError(llvm::Twine const&) + 12 6 catalyst 0x000000000646704f catalyst::quantum::DeviceInitOp::verifyInvariantsImpl() + 175 7 catalyst 0x00000000064371ff 8 catalyst 0x0000000006440fbc 9 catalyst 0x000000000aa304cf 10 catalyst 0x000000000aa31b56 11 catalyst 0x000000000aa2f8f0 12 catalyst 0x000000000aa31b56 13 catalyst 0x000000000aa2f8f0 14 catalyst 0x000000000aa31b56 15 catalyst 0x000000000aa2f8f0 16 catalyst 0x000000000aa32bc4 mlir::verify(mlir::Operation*, bool) + 36 17 catalyst 0x0000000009aaeed2 mlir::parseAsmSourceFile(llvm::SourceMgr const&, mlir::Block*, mlir::ParserConfig const&, mlir::AsmParserState*, mlir::AsmParserCodeCompleteContext*) + 5458 18 catalyst 0x0000000003ddff9d 19 catalyst 0x0000000003de83f7 QuantumDriverMain(catalyst::driver::CompilerOptions const&, catalyst::driver::CompilerOutput&, mlir::DialectRegistry&) + 1095 20 catalyst 0x0000000003ded623 QuantumDriverMainFromCL(int, char**) + 10771 21 libc.so.6 0x00007f941b64cd90 22 libc.so.6 0x00007f941b64ce40 __libc_start_main + 128 23 catalyst 0x0000000003dc75ae _start + 46 vmap.circuit:108:7: note: see current operation: "quantum.device"(%1) <{kwargs = "{'shots': 0, 'mcmc': False, 'num_burnin': 0, 'kernel_name': None}", lib = "/home/jovyan/.local/lib/python3.11/site-packages/pennylane_lightning/liblightning_qubit_catalyst.so"}> {name = "LightningSimulator"} : (i64) -> () vmap.circuit: vmap.circuit:1:8: error: expected 'module asm' module @vmap.circuit { ^ Failed to parse module as LLVM or MLIR source ``` ---
**Author:** ![CatalinaAlbornoz](https://yyz2.discourse-cdn.com/flex012/user_avatar/discuss.pennylane.ai/catalinaalbornoz/32/1196_2.png) [@CatalinaAlbornoz](https://discuss.pennylane.ai/u/CatalinaAlbornoz)\ **Post date:** [July 30, 2025, 9:40pm UTC](https://discuss.pennylane.ai/t/multi-core-computation-with-lightning-qubit-and-openmp-support/8776/12 "2025-07-30T21:40:52Z")
Thanks for adding your code @minn-bj . We will investigate and get back to you. ---
**Author:** ![maliasadi](https://yyz2.discourse-cdn.com/flex012/user_avatar/discuss.pennylane.ai/maliasadi/32/3939_2.png) [@maliasadi](https://discuss.pennylane.ai/u/maliasadi)\ **Post date:** [July 31, 2025, 8:08pm UTC](https://discuss.pennylane.ai/t/multi-core-computation-with-lightning-qubit-and-openmp-support/8776/13 "2025-07-31T20:08:04Z")
Hi @minn-bj, Thanks for providing the example! You should be able to QJIT compile and execute it on Lightning devices using proper versions of PennyLane, Catalyst, and Lightning. From the error, it appears you’re using a development version of Catalyst with some incompatibilities in the compiler. If you installed Catalyst from source, please ensure you’ve carefully followed [the installation guidelines](https://docs.pennylane.ai/projects/catalyst/en/stable/dev/installation.html); you’ll need to run `make all` after installing the dependencies. Otherwise, I suggest creating a **new** environment and trying the following commands to pull and install the official versions of these packages from PyPI: ```bash python -m pip install pennylane pennylane-catalyst CMAKE_ARGS="-DLQ_ENABLE_KERNEL_OMP=ON" python -m pip install pennylane_lightning --no-binary "pennylane_lightning" --force-reinstall --no-cache-dir --verbose ``` You can also run this to check the installed versions: ```bash python -c "import pennylane as qml; qml.about()" ``` Note that PennyLane and Lightning v0.41 and v0.42 require Catalyst v0.11 and v0.12, respectively. If these suggestions don’t resolve your issue, feel free to report back with the versions you’ve tried. ---
**Author:** ![minn-bj](https://avatars.discourse-cdn.com/v4/letter/m/7cd45c/32.png) [@minn-bj](https://discuss.pennylane.ai/u/minn-bj)\ **Post date:** [August 5, 2025, 3:35pm UTC](https://discuss.pennylane.ai/t/multi-core-computation-with-lightning-qubit-and-openmp-support/8776/14 "2025-08-05T15:35:06Z")
Hello @maliasadi , thanks for your reply. I followed your instructions but still have the same issue. If you need more information than the qml.about() and the error msg. feel free to approach me. qml.about() ```auto Name: PennyLane Version: 0.41.1 Summary: PennyLane is a cross-platform Python library for quantum computing, quantum machine learning, and quantum chemistry. Train a quantum computer the same way as a neural network. Home-page: https://github.com/PennyLaneAI/pennylane Author: Author-email: License: Apache License 2.0 Location: /home/jovyan/.local/lib/python3.11/site-packages Requires: appdirs, autograd, autoray, cachetools, diastatic-malt, networkx, numpy, packaging, pennylane-lightning, requests, rustworkx, scipy, tomlkit, typing-extensions Required-by: PennyLane-Catalyst, PennyLane_Lightning, PennyLane_Lightning_GPU, PennyLane_Lightning_Kokkos Platform info: Linux-5.15.0-142-generic-x86_64-with-glibc2.35 Python version: 3.11.10 Numpy version: 2.3.2 Scipy version: 1.16.1 Installed devices: - nvidia.custatevec (PennyLane-Catalyst-0.11.0) - nvidia.cutensornet (PennyLane-Catalyst-0.11.0) - oqc.cloud (PennyLane-Catalyst-0.11.0) - softwareq.qpp (PennyLane-Catalyst-0.11.0) - lightning.gpu (PennyLane_Lightning_GPU-0.41.1) - default.clifford (PennyLane-0.41.1) - default.gaussian (PennyLane-0.41.1) - default.mixed (PennyLane-0.41.1) - default.qubit (PennyLane-0.41.1) - default.qutrit (PennyLane-0.41.1) - default.qutrit.mixed (PennyLane-0.41.1) - default.tensor (PennyLane-0.41.1) - null.qubit (PennyLane-0.41.1) - reference.qubit (PennyLane-0.41.1) - lightning.qubit (PennyLane_Lightning-0.41.1) - lightning.kokkos (PennyLane_Lightning_Kokkos-0.41.1) ``` Error: ```auto [LIB] Running compiler driver in /tmp/vmap.circuitudc846z1 [SYSTEM] /opt/conda/bin/catalyst -o /tmp/vmap.circuitudc846z1/vmap.circuit.ll --module-name vmap.circuit --workspace /tmp/vmap.circuitudc846z1 -verify-each=false --catalyst-pipeline EnforceRuntimeInvariantsPass(split-multiple-tapes;builtin.module(apply-transform-sequence);inline-nested-module),HLOLoweringPass(canonicalize;func.func(chlo-legalize-to-hlo);stablehlo-legalize-to-hlo;func.func(mhlo-legalize-control-flow);func.func(hlo-legalize-to-linalg);func.func(mhlo-legalize-to-std);func.func(hlo-legalize-sort);convert-to-signless;canonicalize;scatter-lowering;hlo-custom-call-lowering;cse;func.func(linalg-detensorize{aggressive-mode});detensorize-scf;canonicalize),QuantumCompilationPass(annotate-function;lower-mitigation;lower-gradients;adjoint-lowering),BufferizationPass(one-shot-bufferize{dialect-filter=memref};inline;gradient-preprocess;gradient-bufferize;scf-bufferize;convert-tensor-to-linalg;convert-elementwise-to-linalg;arith-bufferize;empty-tensor-to-alloc-tensor;func.func(bufferization-bufferize);func.func(tensor-bufferize);catalyst-bufferize;func.func(linalg-bufferize);func.func(tensor-bufferize);quantum-bufferize;func-bufferize;func.func(finalizing-bufferize);canonicalize;gradient-postprocess;func.func(buffer-hoisting);func.func(buffer-loop-hoisting);func.func(buffer-deallocation);convert-arraylist-to-memref;convert-bufferization-to-memref;canonicalize;cp-global-memref),MLIRToLLVMDialect(expand-realloc;convert-gradient-to-llvm;memrefcpy-to-linalgcpy;func.func(convert-linalg-to-loops);convert-scf-to-cf;expand-strided-metadata;lower-affine;arith-expand;convert-complex-to-standard;convert-complex-to-llvm;convert-math-to-llvm;convert-math-to-libm;convert-arith-to-llvm;memref-to-llvm-tbaa;finalize-memref-to-llvm{use-generic-functions};convert-index-to-llvm;convert-catalyst-to-llvm;convert-quantum-to-llvm;emit-catalyst-py-interface;canonicalize;reconcile-unrealized-casts;gep-inbounds;register-inactive-callback), --verbose /tmp/vmap.circuitudc846z1/tmpb1ay7f59.mlir --------------------------------------------------------------------------- CalledProcessError Traceback (most recent call last) File ~/.local/lib/python3.11/site-packages/catalyst/compiler.py:444, in Compiler.run_from_ir(self, ir, module_name, workspace) 443 print(f"[SYSTEM] {' '.join(cmd)}", file=self.options.logfile) --> 444 result = subprocess.run(cmd, check=True, capture_output=True, text=True) 445 if self.options.verbose or os.getenv("ENABLE_DIAGNOSTICS"): File /opt/conda/lib/python3.11/subprocess.py:571, in run(input, capture_output, timeout, check, *popenargs, **kwargs) 570 if check and retcode: --> 571 raise CalledProcessError(retcode, process.args, 572 output=stdout, stderr=stderr) 573 return CompletedProcess(process.args, retcode, stdout, stderr) CalledProcessError: Command '['/opt/conda/bin/catalyst', '-o', '/tmp/vmap.circuitudc846z1/vmap.circuit.ll', '--module-name', 'vmap.circuit', '--workspace', '/tmp/vmap.circuitudc846z1', '-verify-each=false', '--catalyst-pipeline', 'EnforceRuntimeInvariantsPass(split-multiple-tapes;builtin.module(apply-transform-sequence);inline-nested-module),HLOLoweringPass(canonicalize;func.func(chlo-legalize-to-hlo);stablehlo-legalize-to-hlo;func.func(mhlo-legalize-control-flow);func.func(hlo-legalize-to-linalg);func.func(mhlo-legalize-to-std);func.func(hlo-legalize-sort);convert-to-signless;canonicalize;scatter-lowering;hlo-custom-call-lowering;cse;func.func(linalg-detensorize{aggressive-mode});detensorize-scf;canonicalize),QuantumCompilationPass(annotate-function;lower-mitigation;lower-gradients;adjoint-lowering),BufferizationPass(one-shot-bufferize{dialect-filter=memref};inline;gradient-preprocess;gradient-bufferize;scf-bufferize;convert-tensor-to-linalg;convert-elementwise-to-linalg;arith-bufferize;empty-tensor-to-alloc-tensor;func.func(bufferization-bufferize);func.func(tensor-bufferize);catalyst-bufferize;func.func(linalg-bufferize);func.func(tensor-bufferize);quantum-bufferize;func-bufferize;func.func(finalizing-bufferize);canonicalize;gradient-postprocess;func.func(buffer-hoisting);func.func(buffer-loop-hoisting);func.func(buffer-deallocation);convert-arraylist-to-memref;convert-bufferization-to-memref;canonicalize;cp-global-memref),MLIRToLLVMDialect(expand-realloc;convert-gradient-to-llvm;memrefcpy-to-linalgcpy;func.func(convert-linalg-to-loops);convert-scf-to-cf;expand-strided-metadata;lower-affine;arith-expand;convert-complex-to-standard;convert-complex-to-llvm;convert-math-to-llvm;convert-math-to-libm;convert-arith-to-llvm;memref-to-llvm-tbaa;finalize-memref-to-llvm{use-generic-functions};convert-index-to-llvm;convert-catalyst-to-llvm;convert-quantum-to-llvm;emit-catalyst-py-interface;canonicalize;reconcile-unrealized-casts;gep-inbounds;register-inactive-callback),', '--verbose', '/tmp/vmap.circuitudc846z1/tmpb1ay7f59.mlir']' returned non-zero exit status 1. The above exception was the direct cause of the following exception: CompileError Traceback (most recent call last) Cell In[2], line 1 ----> 1 circuit(data) File ~/.local/lib/python3.11/site-packages/pennylane/logging/decorators.py:61, in log_string_debug_func..wrapper_entry(*args, **kwargs) 54 s_caller = "::L".join( 55 [str(i) for i in inspect.getouterframes(inspect.currentframe(), 2)[1][1:3]] 56 ) 57 lgr.debug( 58 f"Calling {f_string} from {s_caller}", 59 **_debug_log_kwargs, 60 ) ---> 61 return func(*args, **kwargs) File ~/.local/lib/python3.11/site-packages/catalyst/jit.py:594, in QJIT. __call__ (self, *args, **kwargs) 590 kwargs = {"static_argnums": self.compile_options.static_argnums, **kwargs} 592 return self.user_function(*args, **kwargs) --> 594 requires_promotion = self.jit_compile(args, **kwargs) 596 # If we receive tracers as input, dispatch to the JAX integration. 597 if any(isinstance(arg, jax.core.Tracer) for arg in tree_flatten(args)[0]): File ~/.local/lib/python3.11/site-packages/pennylane/logging/decorators.py:61, in log_string_debug_func..wrapper_entry(*args, **kwargs) 54 s_caller = "::L".join( 55 [str(i) for i in inspect.getouterframes(inspect.currentframe(), 2)[1][1:3]] 56 ) 57 lgr.debug( 58 f"Calling {f_string} from {s_caller}", 59 **_debug_log_kwargs, 60 ) ---> 61 return func(*args, **kwargs) File ~/.local/lib/python3.11/site-packages/catalyst/jit.py:670, in QJIT.jit_compile(self, args, **kwargs) 667 self.jaxpr, self.out_type, self.out_treedef, self.c_sig = self.capture(args, **kwargs) 669 self.mlir_module = self.generate_ir() --> 670 self.compiled_function, _ = self.compile() 672 self.fn_cache.insert(self.compiled_function, args, self.out_treedef, self.workspace) 674 elif self.compiled_function is not cached_fn.compiled_fn: 675 # Restore active state from cache. File ~/.local/lib/python3.11/site-packages/catalyst/debug/instruments.py:145, in instrument..wrapper(*args, **kwargs) 142 @functools.wraps(fn) 143 def wrapper(*args, **kwargs): 144 if not InstrumentSession.active: --> 145 return fn(*args, **kwargs) 147 with ResultReporter(stage_name, has_finegrained) as reporter: 148 self = args[0] File ~/.local/lib/python3.11/site-packages/pennylane/logging/decorators.py:61, in log_string_debug_func..wrapper_entry(*args, **kwargs) 54 s_caller = "::L".join( 55 [str(i) for i in inspect.getouterframes(inspect.currentframe(), 2)[1][1:3]] 56 ) 57 lgr.debug( 58 f"Calling {f_string} from {s_caller}", 59 **_debug_log_kwargs, 60 ) ---> 61 return func(*args, **kwargs) File ~/.local/lib/python3.11/site-packages/catalyst/jit.py:799, in QJIT.compile(self) 793 shared_object, llvm_ir = self.compiler.run_from_ir( 794 self.overwrite_ir, 795 str(self.mlir_module.operation.attributes["sym_name"]).replace('"', ""), 796 self.workspace, 797 ) 798 else: --> 799 shared_object, llvm_ir = self.compiler.run(self.mlir_module, self.workspace) 801 compiled_fn = CompiledFunction( 802 shared_object, func_name, restype, self.out_type, self.compile_options 803 ) 805 return compiled_fn, llvm_ir File ~/.local/lib/python3.11/site-packages/pennylane/logging/decorators.py:61, in log_string_debug_func..wrapper_entry(*args, **kwargs) 54 s_caller = "::L".join( 55 [str(i) for i in inspect.getouterframes(inspect.currentframe(), 2)[1][1:3]] 56 ) 57 lgr.debug( 58 f"Calling {f_string} from {s_caller}", 59 **_debug_log_kwargs, 60 ) ---> 61 return func(*args, **kwargs) File ~/.local/lib/python3.11/site-packages/catalyst/compiler.py:486, in Compiler.run(self, mlir_module, *args, **kwargs) 470 @debug_logger 471 def run(self, mlir_module, *args, **kwargs): 472 """Compile an MLIR module to a shared object. 473 474 .. note:: (...) 483 (str): filename of shared object 484 """ --> 486 return self.run_from_ir( 487 mlir_module.operation.get_asm( 488 binary=False, print_generic_op_form=False, assume_verified=True 489 ), 490 str(mlir_module.operation.attributes["sym_name"]).replace('"', ""), 491 *args, 492 **kwargs, 493 ) File ~/.local/lib/python3.11/site-packages/pennylane/logging/decorators.py:61, in log_string_debug_func..wrapper_entry(*args, **kwargs) 54 s_caller = "::L".join( 55 [str(i) for i in inspect.getouterframes(inspect.currentframe(), 2)[1][1:3]] 56 ) 57 lgr.debug( 58 f"Calling {f_string} from {s_caller}", 59 **_debug_log_kwargs, 60 ) ---> 61 return func(*args, **kwargs) File ~/.local/lib/python3.11/site-packages/catalyst/compiler.py:451, in Compiler.run_from_ir(self, ir, module_name, workspace) 449 print(result.stderr.strip(), file=self.options.logfile) 450 except subprocess.CalledProcessError as e: # pragma: nocover --> 451 raise CompileError(f"catalyst failed with error code {e.returncode}: {e.stderr}") from e 453 if os.path.exists(output_ir_name): 454 with open(output_ir_name, "r", encoding="utf-8") as f: CompileError: catalyst failed with error code 1: Compilation failed: vmap.circuit:108:7: error: 'quantum.device' op requires attribute 'device_name' "quantum.device"(%0) {kwargs = "{'shots': 0, 'mcmc': False, 'num_burnin': 0, 'kernel_name': None}", lib = "/home/jovyan/.local/lib/python3.11/site-packages/pennylane_lightning/liblightning_qubit_catalyst.so", name = "LightningSimulator"} : (i64) -> () ^ vmap.circuit:108:7: note: diagnostic emitted with trace: Stack dump without symbol names (ensure you have llvm-symbolizer in your PATH or set the environment var `LLVM_SYMBOLIZER_PATH` to point to it): 0 catalyst 0x000000000ab8194b llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) + 43 1 catalyst 0x000000000a9a95d0 2 catalyst 0x000000000a9a9761 mlir::emitError(mlir::Location, llvm::Twine const&) + 17 3 catalyst 0x000000000aa00e19 mlir::Operation::emitError(llvm::Twine const&) + 25 4 catalyst 0x000000000aa01164 mlir::Operation::emitOpError(llvm::Twine const&) + 52 5 catalyst 0x000000000aa026ac mlir::OpState::emitOpError(llvm::Twine const&) + 12 6 catalyst 0x000000000646704f catalyst::quantum::DeviceInitOp::verifyInvariantsImpl() + 175 7 catalyst 0x00000000064371ff 8 catalyst 0x0000000006440fbc 9 catalyst 0x000000000aa304cf 10 catalyst 0x000000000aa31b56 11 catalyst 0x000000000aa2f8f0 12 catalyst 0x000000000aa31b56 13 catalyst 0x000000000aa2f8f0 14 catalyst 0x000000000aa31b56 15 catalyst 0x000000000aa2f8f0 16 catalyst 0x000000000aa32bc4 mlir::verify(mlir::Operation*, bool) + 36 17 catalyst 0x0000000009aaeed2 mlir::parseAsmSourceFile(llvm::SourceMgr const&, mlir::Block*, mlir::ParserConfig const&, mlir::AsmParserState*, mlir::AsmParserCodeCompleteContext*) + 5458 18 catalyst 0x0000000003ddff9d 19 catalyst 0x0000000003de83f7 QuantumDriverMain(catalyst::driver::CompilerOptions const&, catalyst::driver::CompilerOutput&, mlir::DialectRegistry&) + 1095 20 catalyst 0x0000000003ded623 QuantumDriverMainFromCL(int, char**) + 10771 21 libc.so.6 0x00007fe8c2803d90 22 libc.so.6 0x00007fe8c2803e40 __libc_start_main + 128 23 catalyst 0x0000000003dc75ae _start + 46 vmap.circuit:108:7: note: see current operation: "quantum.device"(%1) <{kwargs = "{'shots': 0, 'mcmc': False, 'num_burnin': 0, 'kernel_name': None}", lib = "/home/jovyan/.local/lib/python3.11/site-packages/pennylane_lightning/liblightning_qubit_catalyst.so"}> {name = "LightningSimulator"} : (i64) -> () vmap.circuit: vmap.circuit:1:8: error: expected 'module asm' module @vmap.circuit { ^ Failed to parse module as LLVM or MLIR source ``` ---
**Author:** ![maliasadi](https://yyz2.discourse-cdn.com/flex012/user_avatar/discuss.pennylane.ai/maliasadi/32/3939_2.png) [@maliasadi](https://discuss.pennylane.ai/u/maliasadi)\ **Post date:** [August 5, 2025, 4:29pm UTC](https://discuss.pennylane.ai/t/multi-core-computation-with-lightning-qubit-and-openmp-support/8776/15 "2025-08-05T16:29:11Z")
@minn-bj Could you please confirm whether you have only one installation of Catalyst in your python enviroment? Based on the error message, it looks like your program isn’t running inside an isolated environment (like a Conda or virtual environment). The logs show references to both `/home/jovyan/.local/lib/python3.11/` and `/opt/conda/lib/python3.11/`, but there should be only one reference; from a single Python installation. To fix this, please uninstall PennyLane and Catalyst from both your Conda base environment and your user-level Python (`.local/bin` and `.local/lib`). Then, create a clean environment using either `venv` or `conda`, and pip install these packages inside there. Once you have a single, clean installation of PennyLane and Catalyst, you can compile using the `verbose` compilation option: `@qml.qjit(verbose=True)` in your code. This will show you which Python binary is being used during compilation and execution, and it should point to only one location. Let us know if you need help with any of these steps. ---
**Author:** ![minn-bj](https://avatars.discourse-cdn.com/v4/letter/m/7cd45c/32.png) [@minn-bj](https://discuss.pennylane.ai/u/minn-bj)\ **Post date:** [August 8, 2025, 7:51am UTC](https://discuss.pennylane.ai/t/multi-core-computation-with-lightning-qubit-and-openmp-support/8776/16 "2025-08-08T07:51:27Z")
Hey @maliasadi, thanks you where totally right. I could fix the issue. Many thanks and best regards. minn-bj