【newir】Concat optest (#57043)

* add reference of lbfgs

* add reference of lbfgs

* modify concat kernel choose

* modify ci

paddle/fluid/ir/transforms/pd_op_to_kernel_pass.cc

@@ -569,6 +569,29 @@ phi::KernelKey GetKernelKey(
         kernel_key_parser.key_set.backend_set =
             kernel_key_parser.key_set.backend_set |
             paddle::experimental::BackendSet(data_op_backend);
+      } else if (op->operand_source(i).GetDefiningOp()->name() ==
+                 "builtin.combine") {
+        auto combine_op = op->operand_source(i).GetDefiningOp();
+        for (size_t j = 0; j < combine_op->num_operands(); ++j) {
+          if (combine_op->operand_source(j).GetDefiningOp()->name() ==
+              "pd.data") {
+            auto data_op = combine_op->operand_source(j).GetDefiningOp();
+            auto data_place = data_op->attributes()
+                                  .at("place")
+                                  .dyn_cast<dialect::PlaceAttribute>()
+                                  .data();
+
+            auto data_op_backend =
+                paddle::experimental::ParseBackend(data_place);
+            if (data_op_backend == phi::Backend::UNDEFINED) {
+              data_op_backend = paddle::experimental::ParseBackend(place);
+            }
+            kernel_key_parser.key_set.backend_set =
+                kernel_key_parser.key_set.backend_set |
+                paddle::experimental::BackendSet(data_op_backend);
+            break;
+          }
+        }
       }
     }
 

python/paddle/_C_ops.py

@@ -16,7 +16,7 @@ from paddle.base import core
 
 __all__ = []
 
-UNIFIED_APIS = ['mean']
+UNIFIED_APIS = ['mean', 'concat', 'add_n', 'scale']
 
 for name in dir(core.eager.ops):
     globals()[name] = getattr(core.eager.ops, name)

python/paddle/new_ir_utils.py

@@ -32,12 +32,14 @@ class IrGuard:
         paddle.base.framework.set_flags(old_flag)
 
     def __enter__(self):
+        paddle.enable_static()
         paddle.framework.set_flags({"FLAGS_enable_new_ir_api": True})
         self._switch_to_new_ir()
 
     def __exit__(self, exc_type, exc_val, exc_tb):
         paddle.framework.set_flags({"FLAGS_enable_new_ir_api": False})
         self._switch_to_old_ir()
+        paddle.disable_static()
 
     def _switch_to_new_ir(self):
         if paddle.ir.core._use_new_ir_api():

python/paddle/tensor/manipulation.py

@@ -34,6 +34,7 @@ from ..framework import (
     core,
     dygraph_only,
     in_dynamic_mode,
+    in_new_ir_mode,
 )
 from .creation import _complex_to_real_dtype, _real_to_complex_dtype, zeros
 
@@ -1131,11 +1132,11 @@ def concat(x, axis=0, name=None):
         if not isinstance(input, Variable):
             input = [t for t in input if t.shape.count(0) == 0]
         return _C_ops.concat(input, axis)
-    else:
-        if paddle.ir.core._use_new_ir_api():
+    elif in_new_ir_mode():
         if not isinstance(input, paddle.ir.Value):
             input = [t for t in input if t.shape.count(0) == 0]
-            return paddle._ir_ops.concat(input, axis)
+        return _C_ops.concat(input, axis)
+    else:
         check_type(input, 'input', (list, tuple, Variable), 'concat')
         if not isinstance(input, Variable):
             for id, x in enumerate(input):

python/paddle/tensor/math.py

@@ -15,6 +15,7 @@
 math functions
 """
 
+
 import numpy as np
 
 import paddle
@@ -34,6 +35,8 @@ from ..framework import (
     convert_np_dtype_to_dtype_,
     core,
     in_dynamic_mode,
+    in_dynamic_or_new_ir_mode,
+    in_new_ir_mode,
 )
 from .creation import _complex_to_real_dtype
 from .layer_function_generator import generate_layer_fn, templatedoc
@@ -264,6 +267,10 @@ def scale(x, scale=1.0, bias=0.0, bias_after_scale=True, act=None, name=None):
             return _C_ops.scale(x, scale, float(bias), bias_after_scale)
         out = _C_ops.scale(x, scale, float(bias), bias_after_scale)
         return dygraph_utils._append_activation_in_dygraph(out, act)
+    elif in_new_ir_mode():
+        if act is None:
+            return _C_ops.scale(x, scale, float(bias), bias_after_scale)
+        raise ValueError("act is not implement in new ir of scale api.")
     else:
         check_variable_and_dtype(
             x,
@@ -1956,14 +1963,11 @@ def add_n(inputs, name=None):
             [[8. , 10., 12.],
              [14., 16., 18.]])
     """
-    if in_dynamic_mode():
+    if in_dynamic_or_new_ir_mode():
         if isinstance(inputs, Variable):
             inputs = [inputs]
         return _C_ops.add_n(inputs)
     else:
-        if paddle.ir.core._use_new_ir_api():
-            return paddle._ir_ops.add_n(inputs)
-
         helper = LayerHelper('add_n', **locals())
         check_type(inputs, 'inputs', (Variable, tuple, list), 'add_n')
         if isinstance(inputs, (list, tuple)):

test/legacy_test/eager_op_test.py

@@ -2921,7 +2921,6 @@ class OpTest(unittest.TestCase):
 
         if user_defined_grads is None and self.is_compared_with_fp32():
             self.enable_cal_ref_output()
-
             numeric_grads = self._get_gradient(
                 inputs_to_check,
                 place,
@@ -3164,7 +3163,6 @@ class OpTest(unittest.TestCase):
                 # delete the inputs which no need to calculate grad
                 for no_grad_val in no_grad_set:
                     del inputs[no_grad_val]
-
                 grad_inputs = paddle.grad(
                     outputs=paddle.utils.flatten(outputs),
                     inputs=paddle.utils.flatten(inputs),
@@ -3433,7 +3431,7 @@ class OpTest(unittest.TestCase):
             (
                 static_inputs,
                 attrs,
-                input_dict,
+                inputs_dict,
                 feed,
             ) = self.get_ir_input_attr_dict_and_feed(stop_gradient=False)
             # prepare args
@@ -3480,14 +3478,27 @@ class OpTest(unittest.TestCase):
             fetch_list = getattr(self, "fetch_list", [])
 
             outputs_valid = outputs
-            grad_inputs = inputs_to_check
+            loss_inputs = []
+            for input_name in inputs_to_check:
+                loss_inputs.append(inputs_dict[input_name])
+
             if user_defined_grad_outputs is None:
                 if len(outputs_valid) == 1:
                     for outputs_valid_key in outputs_valid:
                         loss = paddle.mean(outputs_valid[outputs_valid_key][0])
+                else:
+                    avg_sum = []
+                    for cur_loss in outputs_valid:
+                        cur_avg_loss = paddle.mean(outputs_valid[cur_loss][0])
+                        avg_sum.append(cur_avg_loss)
+                    loss_sum = paddle.add_n(avg_sum)
+                    loss = paddle.scale(
+                        loss_sum, scale=1.0 / float(len(avg_sum))
+                    )
+
                 grad_inputs = ir_grad(
                     outputs=paddle.utils.flatten(loss),
-                    inputs=paddle.utils.flatten(static_inputs),
+                    inputs=paddle.utils.flatten(loss_inputs),
                     grad_outputs=None,
                 )
             else:

test/legacy_test/test_mean_op.py

@@ -75,6 +75,7 @@ class TestMeanOp_ZeroDim(OpTest):
 
 class TestMeanOpError(unittest.TestCase):
     def test_errors(self):
+        paddle.enable_static()
         with program_guard(Program(), Program()):
             # The input type of mean_op must be Variable.
             input1 = 12
@@ -88,6 +89,7 @@ class TestMeanOpError(unittest.TestCase):
                 name='input3', shape=[-1, 4], dtype="float16"
             )
             paddle.nn.functional.softmax(input3)
+        paddle.disable_static()
 
 
 @unittest.skipIf(