Add profiler code

Author: EikanWang

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

@@ -140,6 +140,8 @@ class AtenIpexCPUDefault {{
 #include <ATen/CPUGenerator.h>
 #include <c10/util/Exception.h>
 #include <c10/util/Logging.h>
+#include <torch/csrc/autograd/function.h>
+#include <torch/csrc/autograd/record_function.h>
 
 #include "aten_ipex_bridge.h"
 #include "utils.h"
@@ -515,9 +517,16 @@ def is_conv_overrideable_func(fname):
             # Gen definition code for cpp file
             code = '{} {{\n'.format(cpp_func_str_cpp)
 
+            # Gen profile info
             code += '#if defined(_DEBUG)\n'
             code += '  printf("{}::{}\\n");\n'.format(_IPEX_OP_FUNC_NS, cpp_sig.def_name)
             code += '#endif\n'
+            profiler_inputs = []
+            for param in cpp_sig.input_params:
+                if param.core_type in ['Tensor', 'Scalar']:
+                    profiler_inputs.append(param.name)
+            code += '  RECORD_FUNCTION("{ns}::{name}", std::vector<c10::IValue>({{{input_names}}}), torch::autograd::Node::peek_at_next_sequence_nr());\n'.format(ns=_IPEX_OP_FUNC_NS, name=cpp_sig.def_name, input_names=', '.join(profiler_inputs))
+
 
             if is_conv_overrideable_func(cpp_sig.def_name):
                 code += '  return AtenIpexCPUDev::dil_{}({});\n'.format(cpp_sig.def_name, ', '.join([param.name for param in cpp_sig.input_params]))