Full codegen for logdet

[miladm]

Full codegen for logdet, slogdet

• SLogDet compilation is blocked on PyTorch shape inference landing in torch/csrc/lazy/core/shape_inference.h
  • Added support for slogdet in LazyTensor shape inference pytorch#77904
• SLogDet compilation is blocked on upstream codegen issues
  • Codegen doesn't support ops with multiple output tensors (e.g. slogdet) #3596

---

Generate LazyIr.h

class Logdet : public XlaNode {
 public:
  static torch::lazy::OpKind ClassOpKind() {
    return torch::lazy::OpKind(at::aten::logdet);
  }

  Logdet(const torch_xla::XlaValue& self, std::vector<torch::lazy::Shape>&& shapes)

      : XlaNode(torch::lazy::OpKind(at::aten::logdet),
              {self}, std::move(shapes),
              [&]() { return LogdetOutputShape(self); },
              /* num_outputs */ 1,
              torch::lazy::MHash())
  {  }

  std::string ToString() const override {
    std::stringstream ss;
    ss << XlaNode::ToString();

    return ss.str();
  }

  bool CanBeReused(const torch_xla::XlaValue& self) const {
    return false;
    }

  torch_xla::XlaOpVector Lower(LoweringContext* loctx) const override;
};

---

Generated XLANativeFunctions.cpp :

    at::Tensor XLANativeFunctions::logdet(const at::Tensor & self) {

        XLA_FN_COUNTER("xla::");
        auto common_device = torch_xla::bridge::GetXlaDevice(self);
        TORCH_INTERNAL_ASSERT(common_device);

        torch_xla::XLATensorPtr lazy_self = torch_xla::bridge::GetXlaTensorOrCreateForWrappedNumber(self, *common_device);
        torch::lazy::NodePtr node = torch::lazy::ReuseNode<Logdet>(lazy_self->GetIrValue());
        if (!node) {

            auto shapes = torch::lazy::compute_shape_logdet(self);
            TORCH_INTERNAL_ASSERT(shapes.size() == 1);
            if(torch::lazy::symbolicShapeEnabled()){
                std::vector<torch::jit::IValue> inputs = { self };
                char* schema_str = "aten::logdet(Tensor self) -> Tensor";
                applySymbolicShapesOnLT(schema_str, inputs, shapes);
            }

            node = torch::lazy::MakeNode<Logdet>(lazy_self->GetIrValue(), std::move(shapes));
            CacheNode(node);
        }

        auto result = torch_xla::bridge::AtenFromXlaTensor(
                torch_xla::XLATensor::Create(std::move(node), *common_device));
        return result;
    };

[miladm]

As discussed in #3596, slogdet causes codegen blockers. Commenting out slogdet codegen to unblock logdet. I will address slogdet as a future PR once unblocked.

[wonjoolee95]

nit: maybe should we move this back to the original place so this PR doesn't show slogdet as being modified?

[wonjoolee95]

@miladm Thanks! One thing -- this PR is showing it changed tensorflow version and was wondering why?

nit: can we also change the PR title to remove slogdet to prevent possible confusion?

[miladm]

This PR should be in a good shape now. Build passed. First test passed. Will merge after the second test passes.