[cherry-pick]support multi_layer of bilstm,*test=kunlun (#42076)

* modify xpu.cmake,*test=kunlun (#41832)

* modify xpu.cmake,*test=kunlun

* modify xpu.cmake,*test=kunlun

* modify xpu.cmake,*test=kunlun

* modify xpu.cmake,*test=kunlun

* support bilstm,*test=kunlun

* [cherry-pick]support multi_layer of bilstm,*test=kunlun

paddle/fluid/platform/device/xpu/xpu2_op_list.h

@@ -296,6 +296,7 @@ XPUOpMap& get_kl2_ops() {
                                  pOpKernelType(vartype::BOOL, XPUPlace()),
                                  pOpKernelType(vartype::FP32, XPUPlace())})},
       {"rnn", XPUKernelSet({pOpKernelType(vartype::FP32, XPUPlace())})},
+      {"rnn_grad", XPUKernelSet({pOpKernelType(vartype::FP32, XPUPlace())})},
       {"roi_align", XPUKernelSet({pOpKernelType(vartype::FP32, XPUPlace())})},
       {"roi_align_grad",
        XPUKernelSet({pOpKernelType(vartype::FP32, XPUPlace())})},

python/paddle/fluid/tests/unittests/xpu/test_rnn_op_xpu.py

@@ -48,6 +48,10 @@ class XPUTestRNNOp(XPUOpTestWrapper):
             self.place = paddle.XPUPlace(0)
             self.sequence_length = np.ones(
                 (self.batch_size, ), dtype=np.int32) * self.seq_length
+            #self.sequence_length = np.array(
+            #    [12, 11, 10, 9, 8], dtype=np.int32)
+            self.num_layers = 1
+            self.is_bidirec = False
             self.set_attrs()
             self.mode = "LSTM"
             self.is_test = False
@@ -61,6 +65,10 @@ class XPUTestRNNOp(XPUOpTestWrapper):
                 high=0.1,
                 size=(self.seq_length, self.batch_size,
                       self.input_size)).astype(self.dtype)
+            input[11][1:][:] = 0
+            input[10][2:][:] = 0
+            input[9][3:][:] = 0
+            input[8][4:][:] = 0
 
             rnn1 = LSTM(
                 self.input_size,
@@ -117,7 +125,7 @@ class XPUTestRNNOp(XPUOpTestWrapper):
 
         def set_xpu(self):
             self.__class__.use_xpu = True
-            self.__class__.no_need_check_grad = True
+            self.__class__.no_need_check_grad = False
             self.__class__.op_type = self.in_type
 
         def test_check_output(self):
@@ -125,11 +133,20 @@ class XPUTestRNNOp(XPUOpTestWrapper):
                 self.place, atol=0.01,
                 no_check_set=['Reserve', 'DropoutState'])
 
+        def test_grad(self):
+            if not self.is_test:
+                var_name_list = self.get_weight_names()
+                grad_check_list = ['Input', 'init_h', 'init_c']
+                grad_check_list.extend(var_name_list)
+                self.check_grad_with_place(self.place,
+                                           set(grad_check_list),
+                                           ['Out', 'last_hidden', 'last_cell'])
+
         def init_size(self):
-            self.seq_length = 1
-            self.batch_size = 1
-            self.input_size = 5
-            self.hidden_size = 16
+            self.seq_length = 12
+            self.batch_size = 5
+            self.input_size = 3
+            self.hidden_size = 2
 
         def get_weight_names(self):
             weight_names = []
@@ -142,38 +159,24 @@ class XPUTestRNNOp(XPUOpTestWrapper):
             return weight_names
 
         def set_attrs(self):
-            self.num_layers = 1
-            self.is_bidirec = False
+            self.num_layers = 2
+            self.is_bidirec = True
 
-    class TestRNNOp1(TestRNNOp):
-        def init_size(self):
-            self.seq_length = 2
-            self.batch_size = 4
-            self.input_size = 10
-            self.hidden_size = 32
+    class TestRNNOp0(TestRNNOp):
+        def set_attrs(self):
+            self.sequence_length = None
 
+    class TestRNNOp1(TestRNNOp):
         def set_attrs(self):
             self.num_layers = 1
             self.is_bidirec = False
 
     class TestRNNOp2(TestRNNOp):
-        def init_size(self):
-            self.seq_length = 5
-            self.batch_size = 16
-            self.input_size = 30
-            self.hidden_size = 64
-
         def set_attrs(self):
             self.num_layers = 1
             self.is_bidirec = True
 
     class TestRNNOp3(TestRNNOp):
-        def init_size(self):
-            self.seq_length = 10
-            self.batch_size = 64
-            self.input_size = 50
-            self.hidden_size = 64
-
         def set_attrs(self):
             self.num_layers = 2
             self.is_bidirec = False
@@ -188,6 +191,17 @@ class XPUTestRNNOp(XPUOpTestWrapper):
             self.num_layers = 2
             self.is_bidirec = True
 
+    class TestRNNOp6(TestRNNOp):
+        def set_attrs(self):
+            self.num_layers = 2
+            self.is_bidirec = True
+            self.sequence_length = None
+
+    class TestRNNOp7(TestRNNOp):
+        def set_attrs(self):
+            self.num_layers = 3
+            self.is_bidirec = True
+
 
 support_types = get_xpu_op_support_types('rnn')
 for stype in support_types: