add op C++ benchmark test framework(matmul, elementwise_add op benchmark test）

[wenming2014]

add op C++ benchmark test framework(matmul, elementwise_add op benchmark test）（#271）

1. OpBenchmarkTester is the basic class to test the benchmark of op. It can test the op benchmark with default strategy or specific compute and schedule. You should feed it registered op_name, input_shapes and out_types.
   e.g.：
   default:
   TEST(test_matmul, default) {
   int M = 1024;
   int N = 1024;
   int K = 1024;
   std::vector<std::vector> input_shapes{{M, K}, {K, N}};
   std::string op_name = "matmul";
   hlir::framework::NodeAttr attrs;
   OpBenchmarkTester matmul_tester(op_name, input_shapes);
   std::vector type{Float(32)};
   matmul_tester.TestOp("matmul_default", attrs, type);
   }

2. if you want to define specfic stragegy, you can inherit the OpBenchmarkTester class and overide the "CreateSpecificStrategy" funtion which depicts the out tensors' compute and stages. Then invoke TestOp function with the fourth arg set to be 'false' which indicates not to use the default stragegy.
   e.g.
   class MatmulTester : public OpBenchmarkTester {
   public:
   MatmulTester(const std::string &op_name,
   const std::vector<std::vector> &input_shapes,
   const common::Target &target = common::DefaultHostTarget(),
   int repeat = 10,
   float diff = 1e-5)
   : OpBenchmarkTester(op_name, input_shapes, target, repeat, diff) {}

   std::vectorir::Tensor CreateSpecificStrategy(const std::vectorir::Tensor &inputs,
   poly::StageMap *stages) override;
   };
   TEST(test_matmul, tile) {
   int M = 1024;
   int N = 1024;
   int K = 1024;
   std::vector<std::vector> input_shapes{{M, K}, {K, N}};
   std::string op_name = "matmul";
   hlir::framework::NodeAttr attrs;
   MatmulTileTester matmul_tester(op_name, input_shapes);
   std::vector type{Float(32)};
   matmul_tester.TestOp("matmul_tile", attrs, type, false);
   }

3. the benchmark result will be showed in the log temporily as follows:
   I1109 07:31:30.700934 154627 test_utils.cc:27] Testing elementwise_add_default
   I1109 07:31:30.701002 154627 test_utils.cc:34] repeat times: 10, kernel run time: 0.0043922 ms

[paddle-bot-old]

Thanks for your contribution! Please wait for the result of CI firstly.

[haozech]

LGTM

[haozech]

LGTM