InductorError: CppCompileError: C++ compile error on a function with a single item call #158060
Description
🐛 Describe the bug
This was discovered in #157499, which tries to turn on capture_scalar_outputs by default. Specifically PYTORCH_OPINFO_SAMPLE_INPUT_INDEX=0 python test/inductor/test_torchinductor_opinfo.py TestInductorOpInfoCUDA.test_comprehensive_item_cuda_float64 fails. I made a minimal repro below.
import torch
torch._dynamo.config.capture_scalar_outputs = True
@torch.compile(backend="inductor")
def f(x):
return x.item()
x = torch.tensor(42, device='cuda', dtype=torch.float32) # passes if `dtype=torch.int32`
res = f(x)Error logs
Traceback (most recent call last):
File "/home/ryanguo99/scratch/comp.py", line 11, in <module>
res = f(x)
^^^^
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/eval_frame.py", line 797, in compile_wrapper
raise e.remove_dynamo_frames() from None # see TORCHDYNAMO_VERBOSE=1
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_inductor/compile_fx.py", line 952, in _compile_fx_inner
raise InductorError(e, currentframe()).with_traceback(
File "/home/ryanguo99/repos/pytorch/torch/_inductor/compile_fx.py", line 936, in _compile_fx_inner
mb_compiled_graph = fx_codegen_and_compile(
^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_inductor/compile_fx.py", line 1622, in fx_codegen_and_compile
return scheme.codegen_and_compile(gm, example_inputs, inputs_to_check, graph_kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_inductor/compile_fx.py", line 1485, in codegen_and_compile
compiled_module = graph.compile_to_module()
^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_inductor/graph.py", line 2289, in compile_to_module
return self._compile_to_module()
^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_inductor/graph.py", line 2299, in _compile_to_module
mod = self._compile_to_module_lines(wrapper_code)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_inductor/graph.py", line 2367, in _compile_to_module_lines
mod = PyCodeCache.load_by_key_path(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_inductor/codecache.py", line 3237, in load_by_key_path
mod = _reload_python_module(key, path, set_sys_modules=in_toplevel)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_inductor/runtime/compile_tasks.py", line 31, in _reload_python_module
exec(code, mod.__dict__, mod.__dict__)
File "/tmp/torchinductor_ryanguo99/a4/ca4ycqgdw5jjzhco5gtkixdcywxykmibbkvchgzr7cdklcchmaiw.py", line 48, in <module>
async_compile.wait(globals())
File "/home/ryanguo99/repos/pytorch/torch/_inductor/async_compile.py", line 547, in wait
self._wait_futures(scope)
File "/home/ryanguo99/repos/pytorch/torch/_inductor/async_compile.py", line 567, in _wait_futures
kernel = result.result()
^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_inductor/codecache.py", line 3971, in result
return self.result_fn()
^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_inductor/codecache.py", line 2714, in future
result = get_result()
^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_inductor/codecache.py", line 2510, in load_fn
future.result()
File "/home/ryanguo99/.conda/envs/comfyui/lib/python3.11/concurrent/futures/_base.py", line 456, in result
return self.__get_result()
^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/.conda/envs/comfyui/lib/python3.11/concurrent/futures/_base.py", line 401, in __get_result
raise self._exception
File "/home/ryanguo99/.conda/envs/comfyui/lib/python3.11/concurrent/futures/thread.py", line 58, in run
result = self.fn(*self.args, **self.kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_inductor/codecache.py", line 2540, in _worker_compile_cpp
builder.build()
File "/home/ryanguo99/repos/pytorch/torch/_inductor/cpp_builder.py", line 1712, in build
run_compile_cmd(build_cmd, cwd=_build_tmp_dir)
File "/home/ryanguo99/repos/pytorch/torch/_inductor/cpp_builder.py", line 402, in run_compile_cmd
_run_compile_cmd(cmd_line, cwd)
File "/home/ryanguo99/repos/pytorch/torch/_inductor/cpp_builder.py", line 397, in _run_compile_cmd
raise exc.CppCompileError(cmd, output) from e
torch._inductor.exc.InductorError: CppCompileError: C++ compile error
Command:
g++ /tmp/torchinductor_ryanguo99/h7/ch7dz3bknb6c6x6ktdzmbfykruashnh4wujzmz4az2lidms6rs4l.main.cpp -D TORCH_INDUCTOR_CPP_WRAPPER -D STANDALONE_TORCH_HEADER -D C10_USING_CUSTOM_GENERATED_MACROS -D CPU_CAPABILITY_AVX512 -shared -fPIC -O3 -DNDEBUG -fno-trapping-math -funsafe-math-optimizations -ffinite-math-only -fno-signed-zeros -fno-math-errno -fexcess-precision=fast -fno-finite-math-only -fno-unsafe-math-optimizations -ffp-contract=off -fno-tree-loop-vectorize -march=native -Wall -std=c++17 -Wno-unused-variable -Wno-unknown-pragmas -pedantic -fopenmp -include /tmp/torchinductor_ryanguo99/precompiled_headers/celw6u5lto4enac6gwgc7wzodghtjir5ewe5bk76ucgat424h3oa.h -I/home/ryanguo99/.conda/envs/comfyui/include/python3.11 -I/home/ryanguo99/repos/pytorch/torch/include -I/home/ryanguo99/repos/pytorch/torch/include/torch/csrc/api/include -mavx512f -mavx512dq -mavx512vl -mavx512bw -mfma -o /tmp/torchinductor_ryanguo99/h7/ch7dz3bknb6c6x6ktdzmbfykruashnh4wujzmz4az2lidms6rs4l.main.so -ltorch -ltorch_cpu -ltorch_python -lgomp -L/home/ryanguo99/.conda/envs/comfyui/lib -L/home/ryanguo99/repos/pytorch/torch/lib
Output:
/tmp/torchinductor_ryanguo99/h7/ch7dz3bknb6c6x6ktdzmbfykruashnh4wujzmz4az2lidms6rs4l.main.cpp: In function ‘void kernel(double*)’:
/tmp/torchinductor_ryanguo99/h7/ch7dz3bknb6c6x6ktdzmbfykruashnh4wujzmz4az2lidms6rs4l.main.cpp:8:29: error: ‘zuf0’ was not declared in this scope
8 | auto tmp0 = zuf0;
| ^~~~
In file included from /home/ryanguo99/repos/pytorch/torch/include/ATen/cpu/vec/vec512/vec512.h:14,
from /home/ryanguo99/repos/pytorch/torch/include/ATen/cpu/vec/vec.h:4,
from /home/ryanguo99/repos/pytorch/torch/include/ATen/cpu/vec/functional_base.h:6,
from /home/ryanguo99/repos/pytorch/torch/include/ATen/cpu/vec/functional.h:3,
from /home/ryanguo99/repos/pytorch/torch/include/torch/csrc/inductor/cpp_prefix.h:44,
from /tmp/torchinductor_ryanguo99/precompiled_headers/celw6u5lto4enac6gwgc7wzodghtjir5ewe5bk76ucgat424h3oa.h:1:
/home/ryanguo99/repos/pytorch/torch/include/ATen/cpu/vec/vec512/vec512_int.h: In instantiation of ‘at::vec::CPU_CAPABILITY::Vectorized<T> at::vec::CPU_CAPABILITY::shift_512_8(const at::vec::CPU_CAPABILITY::Vectorized<T>&, const at::vec::CPU_CAPABILITY::Vectorized<T>&) [with bool left_shift = true; T = signed char; typename std::enable_if<(is_same_v<T, signed char> || is_same_v<T, unsigned char>), int>::type <anonymous> = 0]’:
/home/ryanguo99/repos/pytorch/torch/include/ATen/cpu/vec/vec512/vec512_int.h:2067:27: required from here
/home/ryanguo99/repos/pytorch/torch/include/ATen/cpu/vec/vec512/vec512_int.h:1859:7: warning: overflow in conversion from ‘int’ to ‘char’ changes value from ‘128’ to ‘'\37777777600'’ [-Woverflow]
1859 | 0x80,
| ^~~~
/home/ryanguo99/repos/pytorch/torch/include/ATen/cpu/vec/vec512/vec512_int.h:1861:7: warning: overflow in conversion from ‘int’ to ‘char’ changes value from ‘128’ to ‘'\37777777600'’ [-Woverflow]
1861 | 0x80,
| ^~~~
/home/ryanguo99/repos/pytorch/torch/include/ATen/cpu/vec/vec512/vec512_int.h:1863:7: warning: overflow in conversion from ‘int’ to ‘char’ changes value from ‘128’ to ‘'\37777777600'’ [-Woverflow]
1863 | 0x80,
| ^~~~
/home/ryanguo99/repos/pytorch/torch/include/ATen/cpu/vec/vec512/vec512_int.h:1865:7: warning: overflow in conversion from ‘int’ to ‘char’ changes value from ‘128’ to ‘'\37777777600'’ [-Woverflow]
1865 | 0x80,
| ^~~~
/home/ryanguo99/repos/pytorch/torch/include/ATen/cpu/vec/vec512/vec512_int.h:1867:7: warning: overflow in conversion from ‘int’ to ‘char’ changes value from ‘128’ to ‘'\37777777600'’ [-Woverflow]
1867 | 0x80,
| ^~~~
/home/ryanguo99/repos/pytorch/torch/include/ATen/cpu/vec/vec512/vec512_int.h:1869:7: warning: overflow in conversion from ‘int’ to ‘char’ changes value from ‘128’ to ‘'\37777777600'’ [-Woverflow]
1869 | 0x80,
| ^~~~
/home/ryanguo99/repos/pytorch/torch/include/ATen/cpu/vec/vec512/vec512_int.h:1871:7: warning: overflow in conversion from ‘int’ to ‘char’ changes value from ‘128’ to ‘'\37777777600'’ [-Woverflow]
1871 | 0x80,
| ^~~~
/home/ryanguo99/repos/pytorch/torch/include/ATen/cpu/vec/vec512/vec512_int.h:1873:7: warning: overflow in conversion from ‘int’ to ‘char’ changes value from ‘128’ to ‘'\37777777600'’ [-Woverflow]
1873 | 0x80,
| ^~~~
/home/ryanguo99/repos/pytorch/torch/include/ATen/cpu/vec/vec512/vec512_int.h:1875:7: warning: overflow in conversion from ‘int’ to ‘char’ changes value from ‘128’ to ‘'\37777777600'’ [-Woverflow]
1875 | 0x80,
| ^~~~
/home/ryanguo99/repos/pytorch/torch/include/ATen/cpu/vec/vec512/vec512_int.h:1877:7: warning: overflow in conversion from ‘int’ to ‘char’ changes value from ‘128’ to ‘'\37777777600'’ [-Woverflow]
1877 | 0x80,
| ^~~~
/home/ryanguo99/repos/pytorch/torch/include/ATen/cpu/vec/vec512/vec512_int.h:1879:7: warning: overflow in conversion from ‘int’ to ‘char’ changes value from ‘128’ to ‘'\37777777600'’ [-Woverflow]
1879 | 0x80,
| ^~~~
/home/ryanguo99/repos/pytorch/torch/include/ATen/cpu/vec/vec512/vec512_int.h:1881:7: warning: overflow in conversion from ‘int’ to ‘char’ changes value from ‘128’ to ‘'\37777777600'’ [-Woverflow]
1881 | 0x80,
| ^~~~
/home/ryanguo99/repos/pytorch/torch/include/ATen/cpu/vec/vec512/vec512_int.h:1883:7: warning: overflow in conversion from ‘int’ to ‘char’ changes value from ‘128’ to ‘'\37777777600'’ [-Woverflow]
1883 | 0x80,
| ^~~~
/home/ryanguo99/repos/pytorch/torch/include/ATen/cpu/vec/vec512/vec512_int.h:1885:7: warning: overflow in conversion from ‘int’ to ‘char’ changes value from ‘128’ to ‘'\37777777600'’ [-Woverflow]
1885 | 0x80,
| ^~~~
......
Versions
main fcc682b, python 3.11
cc @chauhang @penguinwu @voznesenskym @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @aakhundov