🐛 Describe the bug
torch.compile produces a graph break when a class is defined within a method that is compiled. Example:
import torch
import torch._dynamo as dynamo
dynamo.reset()
def toy_example(inp):
class A:
def __init__():
pass
def something():
pass
return inp + 2
compiled_fn = torch.compile(toy_example, backend="inductor")
inp = torch.arange(2)
r1 = compiled_fn(inp)
print(f"r1 = {r1}")
This creates a graph break with the log:
V0618 09:39:34.526000 136241583071872 torch/_dynamo/convert_frame.py:776] [0/0] torchdynamo start compiling toy_example /content/load_build_class.py:6, stack (elided 6 frames):
V0618 09:39:34.526000 136241583071872 torch/_dynamo/convert_frame.py:776] [0/0] File "/content/load_build_class.py", line 17, in <module>
V0618 09:39:34.526000 136241583071872 torch/_dynamo/convert_frame.py:776] [0/0] r1 = compiled_fn(inp)
V0618 09:39:34.526000 136241583071872 torch/_dynamo/convert_frame.py:776] [0/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/eval_frame.py", line 433, in _fn
V0618 09:39:34.526000 136241583071872 torch/_dynamo/convert_frame.py:776] [0/0] return fn(*args, **kwargs)
V0618 09:39:34.526000 136241583071872 torch/_dynamo/convert_frame.py:776] [0/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 1116, in __call__
V0618 09:39:34.526000 136241583071872 torch/_dynamo/convert_frame.py:776] [0/0] return self._torchdynamo_orig_callable(
V0618 09:39:34.526000 136241583071872 torch/_dynamo/convert_frame.py:776] [0/0]
I0618 09:39:34.552000 136241583071872 torch/_dynamo/logging.py:56] [0/0] Step 1: torchdynamo start tracing toy_example /content/load_build_class.py:6
V0618 09:39:34.552000 136241583071872 torch/fx/experimental/symbolic_shapes.py:2541] [0/0] create_env
V0618 09:39:34.568000 136241583071872 torch/_dynamo/symbolic_convert.py:786] [0/0] [__trace_source] TRACE starts_line /content/load_build_class.py:7 in toy_example (toy_example.A)
V0618 09:39:34.568000 136241583071872 torch/_dynamo/symbolic_convert.py:786] [0/0] [__trace_source] class A:
V0618 09:39:34.576000 136241583071872 torch/_dynamo/symbolic_convert.py:809] [0/0] [__trace_bytecode] TRACE LOAD_BUILD_CLASS None []
V0618 09:39:34.576000 136241583071872 torch/_dynamo/symbolic_convert.py:825] [0/0] empty checkpoint
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] Graph break: skip: from user code at:
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] File "/content/load_build_class.py", line 7, in toy_example
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] class A:
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] Traceback (most recent call last):
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 948, in __call__
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] result = self._inner_convert(
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 472, in __call__
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] return _compile(
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] File "/usr/local/lib/python3.10/dist-packages/torch/_utils_internal.py", line 84, in wrapper_function
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] return StrobelightCompileTimeProfiler.profile_compile_time(
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] File "/usr/local/lib/python3.10/dist-packages/torch/_strobelight/compile_time_profiler.py", line 129, in profile_compile_time
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] return func(*args, **kwargs)
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] File "/usr/lib/python3.10/contextlib.py", line 79, in inner
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] return func(*args, **kwds)
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 817, in _compile
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] guarded_code = compile_inner(code, one_graph, hooks, transform)
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/utils.py", line 231, in time_wrapper
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] r = func(*args, **kwargs)
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 636, in compile_inner
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] out_code = transform_code_object(code, transform)
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/bytecode_transformation.py", line 1185, in transform_code_object
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] transformations(instructions, code_options)
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 178, in _fn
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] return fn(*args, **kwargs)
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 582, in transform
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] tracer.run()
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/symbolic_convert.py", line 2462, in run
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] super().run()
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/symbolic_convert.py", line 904, in run
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] while self.step():
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/symbolic_convert.py", line 816, in step
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] self.dispatch_table[inst.opcode](self, inst)
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/symbolic_convert.py", line 641, in _missing
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] unimplemented(f"missing: {opname}")
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/exc.py", line 221, in unimplemented
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] raise Unsupported(msg)
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] torch._dynamo.exc.Unsupported: missing: LOAD_BUILD_CLASS
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks]
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] from user code:
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] File "/content/load_build_class.py", line 7, in toy_example
V0618 09:39:34.577000 136241583071872 torch/_dynamo/convert_frame.py:989] [0/0] [__graph_breaks] class A:
Versions
PyTorch version: 2.5.0.dev20240617+cu121
Is debug build: False
CUDA used to build PyTorch: 12.1
ROCM used to build PyTorch: N/A
OS: Ubuntu 22.04.3 LTS (x86_64)
GCC version: (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version: 14.0.0-1ubuntu1.1
CMake version: version 3.27.9
Libc version: glibc-2.35
Python version: 3.10.12 (main, Nov 20 2023, 15:14:05) [GCC 11.4.0] (64-bit runtime)
Python platform: Linux-6.1.85+-x86_64-with-glibc2.35
Is CUDA available: True
CUDA runtime version: 12.2.140
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration: GPU 0: Tesla T4
Nvidia driver version: 535.104.05
cuDNN version: Probably one of the following:
/usr/lib/x86_64-linux-gnu/libcudnn.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_adv_infer.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_adv_train.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_infer.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_train.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_ops_infer.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_ops_train.so.8.9.6
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: 46 bits physical, 48 bits virtual
Byte Order: Little Endian
CPU(s): 2
On-line CPU(s) list: 0,1
Vendor ID: GenuineIntel
Model name: Intel(R) Xeon(R) CPU @ 2.00GHz
CPU family: 6
Model: 85
Thread(s) per core: 2
Core(s) per socket: 1
Socket(s): 1
Stepping: 3
BogoMIPS: 4000.39
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology nonstop_tsc cpuid tsc_known_freq pni pclmulqdq ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch invpcid_single ssbd ibrs ibpb stibp fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm mpx avx512f avx512dq rdseed adx smap clflushopt clwb avx512cd avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves arat md_clear arch_capabilities
Hypervisor vendor: KVM
Virtualization type: full
L1d cache: 32 KiB (1 instance)
L1i cache: 32 KiB (1 instance)
L2 cache: 1 MiB (1 instance)
L3 cache: 38.5 MiB (1 instance)
NUMA node(s): 1
NUMA node0 CPU(s): 0,1
Vulnerability Gather data sampling: Not affected
Vulnerability Itlb multihit: Not affected
Vulnerability L1tf: Mitigation; PTE Inversion
Vulnerability Mds: Vulnerable; SMT Host state unknown
Vulnerability Meltdown: Vulnerable
Vulnerability Mmio stale data: Vulnerable
Vulnerability Reg file data sampling: Not affected
Vulnerability Retbleed: Vulnerable
Vulnerability Spec rstack overflow: Not affected
Vulnerability Spec store bypass: Vulnerable
Vulnerability Spectre v1: Vulnerable: __user pointer sanitization and usercopy barriers only; no swapgs barriers
Vulnerability Spectre v2: Vulnerable; IBPB: disabled; STIBP: disabled; PBRSB-eIBRS: Not affected; BHI: Vulnerable (Syscall hardening enabled)
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Vulnerable
Versions of relevant libraries:
[pip3] numpy==1.26.4
[pip3] pytorch-triton==3.0.0+45fff310c8
[pip3] torch==2.5.0.dev20240617+cu121
[pip3] torchaudio==2.4.0.dev20240617+cu121
[pip3] torchsummary==1.5.1
[pip3] torchtext==0.18.0
[pip3] torchvision==0.19.0.dev20240617+cu121
[pip3] triton==2.3.0
[conda] Could not collect
cc @ezyang @chauhang @penguinwu @voznesenskym @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @chenyang78 @kadeng @amjames @rec @anijain2305
🐛 Describe the bug
torch.compile produces a graph break when a class is defined within a method that is compiled. Example:
This creates a graph break with the log:
Versions
PyTorch version: 2.5.0.dev20240617+cu121
Is debug build: False
CUDA used to build PyTorch: 12.1
ROCM used to build PyTorch: N/A
OS: Ubuntu 22.04.3 LTS (x86_64)
GCC version: (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version: 14.0.0-1ubuntu1.1
CMake version: version 3.27.9
Libc version: glibc-2.35
Python version: 3.10.12 (main, Nov 20 2023, 15:14:05) [GCC 11.4.0] (64-bit runtime)
Python platform: Linux-6.1.85+-x86_64-with-glibc2.35
Is CUDA available: True
CUDA runtime version: 12.2.140
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration: GPU 0: Tesla T4
Nvidia driver version: 535.104.05
cuDNN version: Probably one of the following:
/usr/lib/x86_64-linux-gnu/libcudnn.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_adv_infer.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_adv_train.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_infer.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_train.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_ops_infer.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_ops_train.so.8.9.6
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: 46 bits physical, 48 bits virtual
Byte Order: Little Endian
CPU(s): 2
On-line CPU(s) list: 0,1
Vendor ID: GenuineIntel
Model name: Intel(R) Xeon(R) CPU @ 2.00GHz
CPU family: 6
Model: 85
Thread(s) per core: 2
Core(s) per socket: 1
Socket(s): 1
Stepping: 3
BogoMIPS: 4000.39
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology nonstop_tsc cpuid tsc_known_freq pni pclmulqdq ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch invpcid_single ssbd ibrs ibpb stibp fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm mpx avx512f avx512dq rdseed adx smap clflushopt clwb avx512cd avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves arat md_clear arch_capabilities
Hypervisor vendor: KVM
Virtualization type: full
L1d cache: 32 KiB (1 instance)
L1i cache: 32 KiB (1 instance)
L2 cache: 1 MiB (1 instance)
L3 cache: 38.5 MiB (1 instance)
NUMA node(s): 1
NUMA node0 CPU(s): 0,1
Vulnerability Gather data sampling: Not affected
Vulnerability Itlb multihit: Not affected
Vulnerability L1tf: Mitigation; PTE Inversion
Vulnerability Mds: Vulnerable; SMT Host state unknown
Vulnerability Meltdown: Vulnerable
Vulnerability Mmio stale data: Vulnerable
Vulnerability Reg file data sampling: Not affected
Vulnerability Retbleed: Vulnerable
Vulnerability Spec rstack overflow: Not affected
Vulnerability Spec store bypass: Vulnerable
Vulnerability Spectre v1: Vulnerable: __user pointer sanitization and usercopy barriers only; no swapgs barriers
Vulnerability Spectre v2: Vulnerable; IBPB: disabled; STIBP: disabled; PBRSB-eIBRS: Not affected; BHI: Vulnerable (Syscall hardening enabled)
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Vulnerable
Versions of relevant libraries:
[pip3] numpy==1.26.4
[pip3] pytorch-triton==3.0.0+45fff310c8
[pip3] torch==2.5.0.dev20240617+cu121
[pip3] torchaudio==2.4.0.dev20240617+cu121
[pip3] torchsummary==1.5.1
[pip3] torchtext==0.18.0
[pip3] torchvision==0.19.0.dev20240617+cu121
[pip3] triton==2.3.0
[conda] Could not collect
cc @ezyang @chauhang @penguinwu @voznesenskym @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @chenyang78 @kadeng @amjames @rec @anijain2305