From PR: pytorch/pytorch#42579
CircleCI error
The pytorch_xla_linux_bionic_py3_6_clang9_build failed in the PyTorch PR to implement torch.maximum and torch.minimum functions.
Below is the error message of the xla build CI.
Aug 14 10:05:46 Generated 1499 wrappers for /var/lib/jenkins/workspace/xla/scripts/../../torch/csrc/autograd/generated/RegistrationDeclarations.h
Aug 14 10:05:46 AtenXlaType function missed override: Tensor max(const Tensor& self, const Tensor& other); // max(Tensor,Tensor)->Tensor
Aug 14 10:05:46 AtenXlaType function missed override: Tensor min(const Tensor& self, const Tensor& other); // min(Tensor,Tensor)->Tensor
Aug 14 10:05:46 Traceback (most recent call last):
Aug 14 10:05:46 File "/var/lib/jenkins/workspace/xla/scripts/gen.py", line 1137, in <module>
Aug 14 10:05:46 generate(args)
Aug 14 10:05:46 File "/var/lib/jenkins/workspace/xla/scripts/gen.py", line 1107, in generate
Aug 14 10:05:46 assert check_overrides(overrides, overridden)
Aug 14 10:05:46 AssertionError
Binary Ops change
pytorch/pytorch#42579 tries to promote torch.maximum and torch.minimum over those hidden overloaded versions of torch.max and torch.min.
The aten/src/ATen/native/BinaryOps.cpp of this PR
Tensor& maximum_out(Tensor& result, const Tensor& self, const Tensor& other) {
TORCH_CHECK(!self.is_complex() && !other.is_complex(), "maximum does not support complex inputs.");
auto iter = TensorIterator::binary_op(result, self, other,
/*check_mem_overlap=*/true);
maximum_stub(iter.device_type(), iter);
return result;
}
Tensor maximum(const Tensor& self, const Tensor& other) {
TORCH_CHECK(!self.is_complex() && !other.is_complex(), "maximum does not support complex inputs.");
Tensor result;
auto iter = TensorIterator::binary_op(result, self, other);
maximum_stub(iter.device_type(), iter);
return iter.output();
}
// binary max, alias for maximum
Tensor& max_out(Tensor& result, const Tensor& self, const Tensor& other) {
return at::maximum_out(result, self, other);
}
Tensor max(const Tensor& self, const Tensor& other) {
return at::maximum(self, other);
}While max.other and min.other are removed in the native_functions.yaml. My bad in the previous description,
Please refer to @mruberry 's comment below.
@ailzhang / @dlibenzi / @JackCaoG.
Cc @mruberry
From PR: pytorch/pytorch#42579
CircleCI error
The
pytorch_xla_linux_bionic_py3_6_clang9_buildfailed in the PyTorch PR to implementtorch.maximumandtorch.minimumfunctions.Below is the error message of the xla build CI.
Binary Ops change
pytorch/pytorch#42579 tries to promote
torch.maximumandtorch.minimumover those hidden overloaded versions oftorch.maxandtorch.min.The
aten/src/ATen/native/BinaryOps.cppof this PRWhileMy bad in the previous description,max.otherandmin.otherare removed in the native_functions.yaml.Please refer to @mruberry 's comment below.
@ailzhang / @dlibenzi / @JackCaoG.
Cc @mruberry