I want to implement my own version of Tensor.trace()
by following along the native/README.md.
As an initial test, I just duplicated the code corresponding to trace
in
aten/src/ATen/native/native_functions.yaml
,
aten/src/ATen/native/ReduceOps.cpp
,
aten/src/ATen/native/cuda/TriangularOps.cu
and
tools/autograd/derivatives.yaml
,
and added the prefix ‘my_
’ to the relevant declarations. This means I now got the following:
native_functions.yaml
:
- func: my_trace(Tensor self) -> Tensor
variants: method, function
dispatch:
CPU: my_trace_cpu
CUDA: my_trace_cuda
- func: my_trace_backward(Tensor grad, int[] sizes) -> Tensor
variants: function
device_check: NoCheck
device_guard: False
derivatives.yaml
:
- name: my_trace(Tensor self) -> Tensor
self: my_trace_backward(grad, self.sizes())
ReduceOps.cpp
:
Tensor my_trace_cpu(const Tensor& self) { ... }
TriangularOps.cu
:
Tensor my_trace_cuda(const Tensor& self) { ... }
There still seems to be something missing somewhere, because during my build I get the following error:
FAILED: bin/conv_to_nnpack_transform_test
: && /usr/lib/ccache/c++ -Wno-deprecated -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -fopenmp -DNDEBUG -DUSE_KINETO -DLIBKINETO_NOCUPTI -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -O2 -fPIC -Wno-narrowing -Wall -Wextra -Werror=return-type -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wno-sign-compare -Wno-unused-parameter -Wno-unused-variable -Wno-unused-function -Wno-unused-result -Wno-unused-local-typedefs -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -Wno-psabi -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Werror=cast-function-type -Wno-stringop-overflow -DHAVE_AVX_CPU_DEFINITION -DHAVE_AVX2_CPU_DEFINITION -O3 -DNDEBUG -DNDEBUG -rdynamic -Wl,-Bsymbolic-functions caffe2/CMakeFiles/conv_to_nnpack_transform_test.dir/transforms/conv_to_nnpack_transform_test.cc.o -o bin/conv_to_nnpack_transform_test -Wl,-rpath,/home/me/pytorch/build/lib: lib/libgtest_main.a -Wl,--no-as-needed,"/home/me/pytorch/build/lib/libtorch.so" -Wl,--as-needed -Wl,--no-as-needed,"/home/me/pytorch/build/lib/libtorch_cpu.so" -Wl,--as-needed lib/libprotobuf.a lib/libc10.so -lmkl_intel_lp64 -lmkl_gnu_thread -lmkl_core -fopenmp /usr/lib/x86_64-linux-gnu/libpthread.so -lm /usr/lib/x86_64-linux-gnu/libdl.so lib/libdnnl.a -ldl lib/libgtest.a -pthread && :
/usr/bin/ld: /home/me/pytorch/build/lib/libtorch_cpu.so: undefined reference to `at::native::my_trace_backward(at::Tensor const&, c10::ArrayRef<long>)'
collect2: error: ld returned 1 exit status
[1291/1581] Building CXX object test_tensorexpr/CMakeFiles/test_tensorexpr.dir/test_simplify.cpp.o
ninja: build stopped: subcommand failed.
The error message speaks of libtorch, but not the README.md. What did I miss?