support size() to avoid fallback to public format

Author: pinzhenx

scripts/cpu/gen-dense-cpu-ops.py

@@ -64,6 +64,7 @@
     'aten::convolution_overrideable(Tensor input, Tensor weight, Tensor? bias, int[] stride, int[] padding, int[] dilation, bool transposed, int[] output_padding, int groups) -> Tensor',
     'aten::convolution_backward_overrideable(Tensor grad_output, Tensor input, Tensor weight, int[] stride, int[] padding, int[] dilation, bool transposed, int[] output_padding, int groups, bool[3] output_mask) -> (Tensor grad_input, Tensor grad_weight, Tensor grad_bias)',
     'aten::resize_(Tensor(a!) self, int[] size, *, MemoryFormat? memory_format=None) -> Tensor(a!)',
+    'aten::size.int(Tensor self, int dim) -> int',
     'aten::clone(Tensor self, *, MemoryFormat? memory_format=None) -> Tensor',
     'aten::gelu(Tensor self) -> Tensor',
     'aten::gelu_backward(Tensor grad, Tensor self) -> Tensor',

torch_ipex/csrc/cpu/DevOPs.cpp

@@ -1395,6 +1395,14 @@ at::Tensor& AtenIpexCPUDev::dil_resize_(at::Tensor& self, at::IntArrayRef size,
   return self;
 }
 
+int64_t AtenIpexCPUDev::dil_size(const at::Tensor & self, int64_t dim) {
+  DEBUG("AtenIpexCPUDev::dil_size\n");
+  CHECK_DNNL_OP_PRE_COND(self);
+
+  dim = at::maybe_wrap_dim(dim, self.dim(), false);
+  return self.sizes()[dim];
+}
+
 at::Tensor AtenIpexCPUDev::dil_clone(const at::Tensor& self, c10::optional<c10::MemoryFormat> optional_memory_format) {
   DEBUG("AtenIpexCPUDev::dil_clone\n");
   CHECK_DNNL_OP_PRE_COND(self);

torch_ipex/csrc/cpu/DevOPs.h

@@ -66,6 +66,7 @@ class AtenIpexCPUDev {
   static at::Tensor dil_sigmoid_backward(const at::Tensor& grad_output, const at::Tensor& output);
   static at::Tensor dil_reshape(const at::Tensor& self, at::IntArrayRef size);
   static at::Tensor& dil_resize_(at::Tensor& self, at::IntArrayRef size, c10::optional<c10::MemoryFormat> memory_format);
+  static int64_t dil_size(const at::Tensor & self, int64_t dim);
   static at::Tensor dil_clone(const at::Tensor& self, c10::optional<c10::MemoryFormat> optional_memory_format);
   static at::Tensor dil_transpose(const at::Tensor & self, int64_t dim0, int64_t dim1);
   static at::Tensor& dil_cat_out(at::Tensor& result, at::TensorList tensors, int64_t dim);