Does anyone know what the issue might be? In my case, I initially attempted to override the native implementation of avg_pool2d using the following code
namespace {
TORCH_LIBRARY_IMPL(aten, CPU, m) {
m.impl(
"aten::avg_pool2d", TORCH_FN((&torch_xtorch::cpu::avg_pool2d_impl))
);
}
}
After that, I verified it on x86, and there were no issues. However, when validating on the RISC-V platform, a deadlock occurred. Here is the call stack; has anyone encountered a similar issue?"
#0 futex_wait (private=0, expected=2, futex_word=0x5adf58) at ../sysdeps/nptl/futex-internal.h:146
#1 __GI___lll_lock_wait (futex=futex@entry=0x5adf58, private=0) at ./nptl/lowlevellock.c:49
#2 0x0000003ff7e139f0 in lll_mutex_lock_optimized (mutex=0x5adf58) at ./nptl/pthread_mutex_lock.c:48
#3 ___pthread_mutex_lock (mutex=0x5adf58) at ./nptl/pthread_mutex_lock.c:93
#4 0x0000003ff0e1ac00 in c10::SingleElementType<(c10::TypeKind)6, c10::ListType>::equals(c10::Type const&) const () from /home/debian/my_venv/torch1/lib/python3.11/site-packages/torch/lib/libtorch_cpu.so
#5 0x0000003ff1043236 in c10::findSchemaDifferences[abi:cxx11](c10::FunctionSchema const&, c10::FunctionSchema const&) () from /home/debian/my_venv/torch1/lib/python3.11/site-packages/torch/lib/libtorch_cpu.so
#6 0x0000003ff1008472 in c10::impl::(anonymous namespace)::checkSchema(c10::OperatorName const&, c10::FunctionSchema const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, c10::KernelFunction const&, c10::FunctionSchema const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&) () from /home/debian/my_venv/torch1/lib/python3.11/site-packages/torch/lib/libtorch_cpu.so
#7 0x0000003ff100cea4 in c10::impl::OperatorEntry::registerKernel(c10::Dispatcher const&, std::optional<c10::DispatchKey>, c10::KernelFunction, std::optional<c10::impl::CppSignature>, std::unique_ptr<c10::FunctionSchema, std::default_delete<c10::FunctionSchema> >, std::__cxx11::basih.so cpu.so
#12 0x0000003fece593a2 in torch::detail::TorchLibraryInit::TorchLibraryInit(torch::Library::Kind, void (*)(torch::Library&), char const*, std::opti:impl::CppSignature>, std::unique_ptr<c10::FunctionSchema, std::default_delete<c10::FunctionSchema> >, std::__cxx11::basic_string<char, sonal<c10::DispatchKey>, char const*, unsigned int) () from /home/debian/pro/xtorch/build/libxtorch.so
#13 0x0000003fece525aa in __static_initialization_and_destruction_0(int, int) () from /home/debian/pro/xtorch/build/libxtorch.so lib/python3.11/site-packages/torch/lib/libtorch_cpu.so
#14 0x0000003fece525e6 in _GLOBAL__sub_I_AveragePool.cpp () from /home/debian/pro/xtorch/build/libxtorch.so rayRef<long>, c10::ArrayRef<long>, c10::ArrayRef<long>, bool, bool, std::optional<long>, at::Tensor&), &torch_xtorch::cpu::shl_avg_pool2d
#15 0x0000003ff7fdfe9e in call_init (env=0x302fc60, argv=0x3ffffff2e8, argc=2, l=<optimized out>) at ./elf/dl-init.c:74