Operator torch._ops.aten.unfold.default is not Aten Canonical

[ari-ruokamo]

🐛 Describe the bug

I'm experimenting in exporting various MSS models to Executorch. Following the example export scenario in Executorch Documentation, the export terminates in error: 'Operator torch._ops.aten.unfold.default is not Aten Canonical'.

Sample reference code input:

# Instatiate model & load weights
model = MyMSSModel(**config.model)
model = load(model, checkpoint_path)
model.eval()

# Dummy one second tensor
model_args = (torch.rand(1, 2, 44100),)

# 1. torch.export: Defines the program with the ATen operator set.
aten_dialect = export(model, model_args)

# 2. to_edge: Make optimizations for Edge devices
edge_program = to_edge(aten_dialect) ### <-- FAIL

# 3. to_executorch: Convert the graph to an ExecuTorch program
executorch_program = edge_program.to_executorch()

Export output:

Traceback (most recent call last):
  File "/home/arijr/miniconda3/envs/executorch/lib/python3.10/runpy.py", line 196, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/home/arijr/miniconda3/envs/executorch/lib/python3.10/runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "/home/arijr/projects/SCNet/scnet/export-executorch.py", line 29, in <module>
    edge_program = to_edge(aten_dialect)
  File "/home/arijr/miniconda3/envs/executorch/lib/python3.10/site-packages/executorch/exir/program/_program.py", line 1019, in to_edge
    edge_programs[name] = _generate_edge_program(name, config, program)
  File "/home/arijr/miniconda3/envs/executorch/lib/python3.10/site-packages/executorch/exir/program/_program.py", line 680, in _generate_edge_program
    EXIRATenDialectVerifier(ops_set_to_not_decompose)(program.graph_module)
  File "/home/arijr/miniconda3/envs/executorch/lib/python3.10/site-packages/executorch/exir/verification/verifier.py", line 74, in __call__
    return self._check_graph_module(*args, **kwargs)
  File "/home/arijr/miniconda3/envs/executorch/lib/python3.10/site-packages/torch/_export/verifier.py", line 222, in _check_graph_module
    _check_valid_op(node.target)
  File "/home/arijr/miniconda3/envs/executorch/lib/python3.10/site-packages/torch/_export/verifier.py", line 205, in _check_valid_op
    self.check_valid_op(op)
  File "/home/arijr/miniconda3/envs/executorch/lib/python3.10/site-packages/executorch/exir/verification/verifier.py", line 116, in check_valid_op
    raise SpecViolationError(
**torch._export.verifier.SpecViolationError: Operator torch._ops.aten.unfold.default is not Aten Canonical.**

Versions

Collecting environment information...
PyTorch version: 2.4.0+cpu
Is debug build: False
CUDA used to build PyTorch: Could not collect
ROCM used to build PyTorch: N/A

OS: Ubuntu 24.04 LTS (x86_64)
GCC version: (Ubuntu 13.2.0-23ubuntu4) 13.2.0
Clang version: Could not collect
CMake version: version 3.30.3
Libc version: glibc-2.39

Python version: 3.10.0 (default, Mar 3 2022, 09:58:08) [GCC 7.5.0] (64-bit runtime)
Python platform: Linux-6.8.0-40-generic-x86_64-with-glibc2.39
Is CUDA available: False
CUDA runtime version: 12.2.91
CUDA_MODULE_LOADING set to: N/A
GPU models and configuration:
GPU 0: NVIDIA GeForce RTX 3090
GPU 1: NVIDIA GeForce RTX 3090
GPU 2: NVIDIA GeForce RTX 3090

Nvidia driver version: 535.183.01
cuDNN version: Probably one of the following:
/usr/lib/x86_64-linux-gnu/libcudnn.so.8.3.3
/usr/lib/x86_64-linux-gnu/libcudnn_adv_infer.so.8.3.3
/usr/lib/x86_64-linux-gnu/libcudnn_adv_train.so.8.3.3
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_infer.so.8.3.3
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_train.so.8.3.3
/usr/lib/x86_64-linux-gnu/libcudnn_ops_infer.so.8.3.3
/usr/lib/x86_64-linux-gnu/libcudnn_ops_train.so.8.3.3
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True

CPU:
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Address sizes: 48 bits physical, 48 bits virtual
Byte Order: Little Endian
CPU(s): 32
On-line CPU(s) list: 0-31
Vendor ID: AuthenticAMD
Model name: AMD Ryzen 9 5950X 16-Core Processor
CPU family: 25
Model: 33
Thread(s) per core: 2
Core(s) per socket: 16
Socket(s): 1
Stepping: 2
Frequency boost: enabled
CPU(s) scaling MHz: 58%
CPU max MHz: 5083,3979
CPU min MHz: 2200,0000
BogoMIPS: 6799,81
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 hw_pstate ssbd mba ibrs ibpb stibp vmmcall fsgsbase bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local user_shstk clzero irperf xsaveerptr rdpru wbnoinvd arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif v_spec_ctrl umip pku ospke vaes vpclmulqdq rdpid overflow_recov succor smca fsrm debug_swap
L1d cache: 512 KiB (16 instances)
L1i cache: 512 KiB (16 instances)
L2 cache: 8 MiB (16 instances)
L3 cache: 64 MiB (2 instances)
NUMA node(s): 1
NUMA node0 CPU(s): 0-31
Vulnerability Gather data sampling: Not affected
Vulnerability Itlb multihit: Not affected
Vulnerability L1tf: Not affected
Vulnerability Mds: Not affected
Vulnerability Meltdown: Not affected
Vulnerability Mmio stale data: Not affected
Vulnerability Reg file data sampling: Not affected
Vulnerability Retbleed: Not affected
Vulnerability Spec rstack overflow: Mitigation; Safe RET
Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2: Mitigation; Retpolines; IBPB conditional; IBRS_FW; STIBP always-on; RSB filling; PBRSB-eIBRS Not affected; BHI Not affected
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Not affected

Versions of relevant libraries:
[pip3] executorch==0.3.0a0+7d77d78
[pip3] numpy==2.1.1
[pip3] torch==2.4.0+cpu
[pip3] torchaudio==2.4.0+cpu
[pip3] torchsr==1.0.4
[pip3] torchvision==0.19.0
[conda] executorch 0.3.0a0+7d77d78 pypi_0 pypi
[conda] numpy 2.1.1 pypi_0 pypi
[conda] torch 2.4.0+cpu pypi_0 pypi
[conda] torchaudio 2.4.0+cpu pypi_0 pypi
[conda] torchsr 1.0.4 pypi_0 pypi
[conda] torchvision 0.19.0 pypi_0 pypi
(

[guangy10]

All existing core ATen ops can be found here: https://github.com/pytorch/pytorch/blob/main/aten/src/ATen/native/native_functions.yaml

cc: @SS-JIA for core aten opset

[ari-ruokamo]

Knowing this might not be the solution but I tried disabling the dialect op check using,

edge_program = to_edge(aten_dialect, compile_config=EdgeCompileConfig(_check_ir_validity=False))
the to_edge() phase completes, but the next phase terminates in an error:

Traceback (most recent call last):
  File "/home/arijr/miniconda3/envs/executorch/lib/python3.10/runpy.py", line 196, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/home/arijr/miniconda3/envs/executorch/lib/python3.10/runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "/home/arijr/projects/SCNet/scnet/export-executorch.py", line 33, in <module>
    executorch_program = edge_program.to_executorch()
  File "/home/arijr/miniconda3/envs/executorch/lib/python3.10/site-packages/executorch/exir/program/_program.py", line 1196, in to_executorch
    new_gm_res = p(new_gm)
  File "/home/arijr/miniconda3/envs/executorch/lib/python3.10/site-packages/torch/fx/passes/infra/pass_base.py", line 41, in __call__
    res = self.call(graph_module)
  File "/home/arijr/miniconda3/envs/executorch/lib/python3.10/site-packages/executorch/exir/passes/__init__.py", line 423, in call
    raise RuntimeError(f"Missing out variants: {missing_out_vars}")
RuntimeError: Missing out variants: {'aten::_unsafe_index_put'}

Any tips & assistance how to mitigate these issues is highly appreciated - thanks!

[ari-ruokamo]

An example/dummy forward-function:

    def forward(self, x):
        L = x.shape[-1]
        x = x.reshape(-1, L)
        x = torch.stft(x, **self.stft_config, return_complex=True)

        # NOP

        x = torch.istft(x, **self.stft_config)

        return x

Export trials:

1. Export with config (_check_ir_validity=True) makes the to_edge() compilation fail at the first stft-funtion,

--> **torch._export.verifier.SpecViolationError: Operator torch._ops.aten.unfold.default is not Aten Canonical.**

2. Export with config (_check_ir_validity=False) make the edge_program.to_executorch() fail at the latter istft-function with an error,

--> RuntimeError: Missing out variants: {'aten::_unsafe_index_put'}

[divideconcept]

Does this mean it's impossible to export any model that uses stft/istft ? Is that planned ?

[ari-ruokamo]

Here's the graph output of the dummy forward-function. Where does the export compiler find a torch._ops.aten.unfold.default in there?

ExportedProgram:
    class GraphModule(torch.nn.Module):
        def forward(self, x: "f32[1, 2, 895088]"):
            # File: /home/arijr/miniconda3/envs/executorch/lib/python3.10/site-packages/torch/nn/functional.py:4552 in pad, code: return torch._C._nn.pad(input, pad, mode, value)
            pad: "f32[1, 2, 896000]" = torch.ops.aten.pad.default(x, [0, 912]);  x = None
            
            # File: /home/arijr/projects/Mss/Mss.py:331 in forward, code: x = x.reshape(-1, L)
            view: "f32[2, 896000]" = torch.ops.aten.view.default(pad, [-1, 896000]);  pad = None
            
            # File: /home/arijr/projects/Mss/Mss.py:333 in forward, code: x = torch.stft(x, **self.stft_config, return_complex=True)
            view_1: "f32[1, 2, 896000]" = torch.ops.aten.view.default(view, [1, 2, 896000]);  view = None
            pad_1: "f32[1, 2, 900096]" = torch.ops.aten.pad.default(view_1, [2048, 2048], 'reflect');  view_1 = None
            view_2: "f32[2, 900096]" = torch.ops.aten.view.default(pad_1, [2, 900096]);  pad_1 = None
            stft: "c64[2, 2049, 876]" = torch.ops.aten.stft.default(view_2, 4096, 1024, 4096, None, True, None, True);  view_2 = None
            
            # File: /home/arijr/projects/Mss/Mss.py:338 in forward, code: x = torch.istft(x, **self.stft_config)
            istft: "f32[2, 896000]" = torch.ops.aten.istft.default(stft, 4096, 1024, 4096, None, True, True);  stft = None
            return (istft,)

[ari-ruokamo]

RuntimeError: view_as_complex_copy does not support automatic differentiation for outputs with complex dtype.

Moved to a macOS with configuration:

OS: macOS 14.6.1 (arm64)
Clang version: 16.0.0 (clang-1600.0.26.3)
CMake version: version 3.30.5

Python version: 3.10.0 (default, Mar  3 2022, 03:54:28) [Clang 12.0.0 ] (64-bit runtime)
Python platform: macOS-14.6.1-arm64-arm-64bit
Is XNNPACK available: True

CPU:
Apple M2 Max

Versions of relevant libraries:
[pip3] executorch==0.4.0a0+6a085ff
[pip3] numpy==1.21.3
[pip3] torch==2.5.0
[pip3] torchaudio==2.5.0
[pip3] torchsr==1.0.4
[pip3] torchvision==0.20.0
[conda] executorch                0.4.0a0+6a085ff          pypi_0    pypi
[conda] numpy                     1.21.3                   pypi_0    pypi
[conda] torch                     2.5.0                    pypi_0    pypi
[conda] torchaudio                2.5.0                    pypi_0    pypi
[conda] torchsr                   1.0.4                    pypi_0    pypi
[conda] torchvision               0.20.0                   pypi_0    pypi

Running a dummy model export on macOS the export completes successfully with,

• Fixed model input tensor
• EdgeCompileConfig(_check_ir_validity=False)

However, when exporting the real model with the exact same export-script, export terminates at to_edge():

export(...):

ExportedProgram:
    class GraphModule(torch.nn.Module):

<clip>

             # File: /Users/ariruokamo/projects/Mss/Mss.py:392 in forward, code: x = torch.view_as_complex(x.contiguous())
            clone_19: "f32[4, 2049, 880, 2]" = torch.ops.aten.clone.default(permute_19, memory_format = torch.contiguous_format);  permute_19 = None
            view_as_complex: "c64[4, 2049, 880]" = torch.ops.aten.view_as_complex.default(clone_19);  clone_19 = None

to_edge(...):

Traceback (most recent call last):
  File "/opt/miniconda3/envs/executorch/lib/python3.10/runpy.py", line 196, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/opt/miniconda3/envs/executorch/lib/python3.10/runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "/Users/ariruokamo/projects/Mss/export-executorch.py", line 63, in <module>
    executorch_program = edge_program.to_executorch()
  File "/opt/miniconda3/envs/executorch/lib/python3.10/site-packages/executorch/exir/program/_program.py", line 1322, in to_executorch
    new_gm_res = p(new_gm)
  File "/opt/miniconda3/envs/executorch/lib/python3.10/site-packages/torch/fx/passes/infra/pass_base.py", line 41, in __call__
    res = self.call(graph_module)
  File "/opt/miniconda3/envs/executorch/lib/python3.10/site-packages/executorch/exir/pass_base.py", line 572, in call
    result = self.call_submodule(graph_module, tuple(inputs))
  File "/opt/miniconda3/envs/executorch/lib/python3.10/site-packages/executorch/exir/pass_base.py", line 658, in call_submodule
    res = super().call_submodule(graph_module, inputs)
  File "/opt/miniconda3/envs/executorch/lib/python3.10/site-packages/executorch/exir/pass_base.py", line 535, in call_submodule
    interpreter.run(*inputs_data)
  File "/opt/miniconda3/envs/executorch/lib/python3.10/site-packages/torch/fx/interpreter.py", line 146, in run
    self.env[node] = self.run_node(node)
  File "/opt/miniconda3/envs/executorch/lib/python3.10/site-packages/executorch/exir/pass_base.py", line 375, in run_node
    return super().run_node(n)
  File "/opt/miniconda3/envs/executorch/lib/python3.10/site-packages/torch/fx/interpreter.py", line 203, in run_node
    return getattr(self, n.op)(n.target, args, kwargs)
  File "/opt/miniconda3/envs/executorch/lib/python3.10/site-packages/executorch/exir/pass_base.py", line 607, in call_function
    return self.callback.call_operator(
  File "/opt/miniconda3/envs/executorch/lib/python3.10/site-packages/executorch/exir/passes/spec_prop_pass.py", line 96, in call_operator
    meta["spec"] = pytree.tree_map(make_spec, op(*args_data, **kwargs_data))
  File "/opt/miniconda3/envs/executorch/lib/python3.10/site-packages/executorch/exir/dialects/edge/_ops.py", line 333, in __call__
    return self._op(*args, **kwargs)
  File "/opt/miniconda3/envs/executorch/lib/python3.10/site-packages/torch/_ops.py", line 716, in __call__
    return self._op(*args, **kwargs)
RuntimeError: view_as_complex_copy does not support automatic differentiation for outputs with complex dtype.

While executing %aten_view_as_complex_copy_default : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.view_as_complex_copy.default](args = (%aten_clone_default_79,), kwargs = {})
Original traceback:
  File "/Users/ariruokamo/projects/Mss/Mss.py", line 392, in forward
    x = torch.view_as_complex(x.contiguous())

This is a little confusing, view_as_complex() get the exact same tensor as in the dummy version. Any help appreciated!

[valkjsaaa]

I have the same issue. Thanks for posting this.

[slycheese]

Adding to the list here... I'm having the exact same issue.

aten_dialect = export() works fine, but
edge_program = to_edge(aten_dialect)

gives the verification error,
_'Operator torch.ops.aten.unfold.default is not Aten Canonical'.

The model I'm trying to export also uses torch.stft

As noted above by @ari-ruokamo, it's possible (at least for my model) to force my way to a lowered model by disabling the verification when lowering to edge, but the subsequent model crashes when I try to execute it via executor_runner (unsurprisingly!).

This is probably not useful, but just in case, this is the output when the model fails to run:

I 00:00:00.001028 executorch:executor_runner.cpp:82] Model file ../tmp/compute_features.pte is loaded.
I 00:00:00.001050 executorch:executor_runner.cpp:91] Using method forward
I 00:00:00.001061 executorch:executor_runner.cpp:138] Setting up planned buffer 0, size 5165040.
E 00:00:00.002096 executorch:tensor_parser_portable.cpp:49] Invalid or unsupported ScalarType 9
E 00:00:00.002103 executorch:method.cpp:386] Failed parsing tensor at index 29: 0x23
F 00:00:00.002111 executorch:executor_runner.cpp:160] In function main(), assert failed (method.ok()): Loading of method forward failed with status 0x23

[slycheese]

#5381 (comment)

Here's the graph output of the dummy forward-function. Where does the export compiler find a torch._ops.aten.unfold.default in there?

ExportedProgram:
    class GraphModule(torch.nn.Module):
        def forward(self, x: "f32[1, 2, 895088]"):
            # File: /home/arijr/miniconda3/envs/executorch/lib/python3.10/site-packages/torch/nn/functional.py:4552 in pad, code: return torch._C._nn.pad(input, pad, mode, value)
            pad: "f32[1, 2, 896000]" = torch.ops.aten.pad.default(x, [0, 912]);  x = None
            
            # File: /home/arijr/projects/Mss/Mss.py:331 in forward, code: x = x.reshape(-1, L)
            view: "f32[2, 896000]" = torch.ops.aten.view.default(pad, [-1, 896000]);  pad = None
            
            # File: /home/arijr/projects/Mss/Mss.py:333 in forward, code: x = torch.stft(x, **self.stft_config, return_complex=True)
            view_1: "f32[1, 2, 896000]" = torch.ops.aten.view.default(view, [1, 2, 896000]);  view = None
            pad_1: "f32[1, 2, 900096]" = torch.ops.aten.pad.default(view_1, [2048, 2048], 'reflect');  view_1 = None
            view_2: "f32[2, 900096]" = torch.ops.aten.view.default(pad_1, [2, 900096]);  pad_1 = None
            stft: "c64[2, 2049, 876]" = torch.ops.aten.stft.default(view_2, 4096, 1024, 4096, None, True, None, True);  view_2 = None
            
            # File: /home/arijr/projects/Mss/Mss.py:338 in forward, code: x = torch.istft(x, **self.stft_config)
            istft: "f32[2, 896000]" = torch.ops.aten.istft.default(stft, 4096, 1024, 4096, None, True, True);  stft = None
            return (istft,)

FWIW, my forced-to-edge graph does have aten.unfold.default operations. Here's the edge graph I got:

graph():
    %c_l__args___0__modules__compute_stft___window : [num_users=1] = placeholder[target=c_l__args___0__modules__compute_stft___window]
    %c_l__args___0__modules__compute_fbanks___f_central : [num_users=1] = placeholder[target=c_l__args___0__modules__compute_fbanks___f_central]
    %c_l__args___0__modules__compute_fbanks___all_freqs_mat : [num_users=1] = placeholder[target=c_l__args___0__modules__compute_fbanks___all_freqs_mat]
    %c_l__args___0__modules__compute_fbanks___band : [num_users=1] = placeholder[target=c_l__args___0__modules__compute_fbanks___band]
    %_lifted_tensor_constant0 : [num_users=1] = placeholder[target=_lifted_tensor_constant0]
    %_lifted_tensor_constant1 : [num_users=1] = placeholder[target=_lifted_tensor_constant1]
    %_lifted_tensor_constant2 : [num_users=1] = placeholder[target=_lifted_tensor_constant2]
    %_lifted_tensor_constant3 : [num_users=1] = placeholder[target=_lifted_tensor_constant3]
    %_lifted_tensor_constant4 : [num_users=1] = placeholder[target=_lifted_tensor_constant4]
    %wav : [num_users=1] = placeholder[target=wav]
    %aten_view_copy_default : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.view_copy.default](args = (%wav, [1, 1, 207952]), kwargs = {})
    %aten_constant_pad_nd_default : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.constant_pad_nd.default](args = (%aten_view_copy_default, [200, 200], 0.0), kwargs = {})
    %aten_view_copy_default_1 : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.view_copy.default](args = (%aten_constant_pad_nd_default, [1, 208352]), kwargs = {})
    **%aten_unfold_copy_default** : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.**unfold_copy.default**](args = (%aten_view_copy_default_1, -1, 400, 160), kwargs = {})
    %aten_mul_tensor : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.mul.Tensor](args = (%**aten_unfold_copy_default**, %c_l__args___0__modules__compute_stft___window), kwargs = {})
    %aten__fft_r2c_default : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten._fft_r2c.default](args = (%aten_mul_tensor, [2], 0, True), kwargs = {})
    %aten_permute_copy_default : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.permute_copy.default](args = (%aten__fft_r2c_default, [0, 2, 1]), kwargs = {})
    %aten_view_as_real_copy_default : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.view_as_real_copy.default](args = (%aten_permute_copy_default,), kwargs = {})
    %aten_permute_copy_default_1 : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.permute_copy.default](args = (%aten_view_as_real_copy_default, [0, 2, 1, 3]), kwargs = {})
    %aten_pow_tensor_scalar : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.pow.Tensor_Scalar](args = (%aten_permute_copy_default_1, 2), kwargs = {})
    %aten_sum_dim_int_list : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.sum.dim_IntList](args = (%aten_pow_tensor_scalar, [-1]), kwargs = {})
    %aten_pow_tensor_scalar_1 : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.pow.Tensor_Scalar](args = (%aten_sum_dim_int_list, 1), kwargs = {})
    %aten_repeat_default : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.repeat.default](args = (%c_l__args___0__modules__compute_fbanks___f_central, [201, 1]), kwargs = {})
    %aten_permute_copy_default_2 : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.permute_copy.default](args = (%aten_repeat_default, [1, 0]), kwargs = {})
    %aten_repeat_default_1 : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.repeat.default](args = (%c_l__args___0__modules__compute_fbanks___band, [201, 1]), kwargs = {})
    %aten_permute_copy_default_3 : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.permute_copy.default](args = (%aten_repeat_default_1, [1, 0]), kwargs = {})
    %aten_sub_tensor : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.sub.Tensor](args = (%c_l__args___0__modules__compute_fbanks___all_freqs_mat, %aten_permute_copy_default_2), kwargs = {})
    %aten_div_tensor : [num_users=2] = call_function[target=executorch.exir.dialects.edge._ops.aten.div.Tensor](args = (%aten_sub_tensor, %aten_permute_copy_default_3), kwargs = {})
    %aten_add_tensor : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.add.Tensor](args = (%aten_div_tensor, %_lifted_tensor_constant0), kwargs = {})
    %aten_neg_default : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.neg.default](args = (%aten_div_tensor,), kwargs = {})
    %aten_add_tensor_1 : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.add.Tensor](args = (%aten_neg_default, %_lifted_tensor_constant1), kwargs = {})
    %aten_full_default : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.full.default](args = ([1], 0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cpu, pin_memory: False})
    %aten_minimum_default : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.minimum.default](args = (%aten_add_tensor, %aten_add_tensor_1), kwargs = {})
    %aten_maximum_default : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.maximum.default](args = (%aten_full_default, %aten_minimum_default), kwargs = {})
    %aten_permute_copy_default_4 : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.permute_copy.default](args = (%aten_maximum_default, [1, 0]), kwargs = {})
    %aten_view_copy_default_2 : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.view_copy.default](args = (%aten_pow_tensor_scalar_1, [1300, 201]), kwargs = {})
    %aten_mm_default : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.mm.default](args = (%aten_view_copy_default_2, %aten_permute_copy_default_4), kwargs = {})
    %aten_view_copy_default_3 : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.view_copy.default](args = (%aten_mm_default, [1, 1300, 60]), kwargs = {})
    %aten_clamp_default : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.clamp.default](args = (%aten_view_copy_default_3, 1e-10), kwargs = {})
    %aten_log10_default : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.log10.default](args = (%aten_clamp_default,), kwargs = {})
    %aten__to_copy_default : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten._to_copy.default](args = (%_lifted_tensor_constant2,), kwargs = {dtype: torch.float32})
    %aten_mul_tensor_1 : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.mul.Tensor](args = (%aten_log10_default, %aten__to_copy_default), kwargs = {})
    %aten_sub_tensor_1 : [num_users=2] = call_function[target=executorch.exir.dialects.edge._ops.aten.sub.Tensor](args = (%aten_mul_tensor_1, %_lifted_tensor_constant3), kwargs = {})
    %aten_amax_default : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.amax.default](args = (%aten_sub_tensor_1, [-2, -1]), kwargs = {})
    %aten_sub_tensor_2 : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.sub.Tensor](args = (%aten_amax_default, %_lifted_tensor_constant4), kwargs = {})
    %aten_view_copy_default_4 : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.view_copy.default](args = (%aten_sub_tensor_2, [1, 1, 1]), kwargs = {})
    %aten_maximum_default_1 : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.maximum.default](args = (%aten_sub_tensor_1, %aten_view_copy_default_4), kwargs = {})
    return (aten_maximum_default_1,)