UT coverage for guassian_mkldnn_op and batch_norm_mkldnn_op (#19011)

* integrations problem
test=develop

* add batch_norm_mkldnn_op backward-reuse test and guassian seed=0 test
test=develop

python/paddle/fluid/tests/unittests/mkldnn/mkldnn_op_test.py

@@ -19,13 +19,14 @@ import paddle.fluid.core as core
 import paddle.fluid as fluid
 
 
+def __assert_close(test_case, tensor, np_array, msg, atol=1e-4):
+    test_case.assertTrue(
+        np.allclose(
+            np.array(tensor), np_array, atol=atol), msg)
+
+
 def check_if_mkldnn_primitives_exist_in_bwd(test_case, op_type, x, out,
                                             out_grad, x_grad):
-    def __assert_close(tensor, np_array, msg, atol=1e-4):
-        test_case.assertTrue(
-            np.allclose(
-                np.array(tensor), np_array, atol=atol), msg)
-
     place = core.CPUPlace()
 
     var_dict = {'x': x, 'out': out, 'out@GRAD': out_grad, 'x@GRAD': x_grad}
@@ -69,7 +70,81 @@ def check_if_mkldnn_primitives_exist_in_bwd(test_case, op_type, x, out,
                       for name in ['x', 'out@GRAD']},
                 fetch_list=['x@GRAD', 'out'])
 
-        __assert_close(x_grad, out[0], 'x@GRAD')
+        __assert_close(test_case, x_grad, out[0], 'x@GRAD')
+
+
+def check_if_mkldnn_batchnorm_primitives_exist_in_bwd(
+        test_case, var_dict, place, shape, data_layout):
+
+    var_names = [
+        'x', 'scale', 'bias', 'mean', 'variance', 'y', 'saved_mean',
+        'saved_variance'
+    ]
+    ground_truth = {name: var_dict[name] for name in var_names}
+    program = fluid.Program()
+    with fluid.program_guard(program):
+        block = program.global_block()
+        for name in ground_truth:
+            block.create_var(
+                name=name, dtype='float32', shape=ground_truth[name].shape)
+        bn_op = block.append_op(
+            type="batch_norm",
+            inputs={
+                "X": block.var('x'),
+                "Scale": block.var('scale'),
+                "Bias": block.var('bias'),
+                "Mean": block.var('mean'),
+                "Variance": block.var('variance')
+            },
+            outputs={
+                "Y": block.var('y'),
+                "MeanOut": block.var('mean'),  # share memory
+                "VarianceOut": block.var('variance'),  # share memory
+                "SavedMean": block.var('saved_mean'),
+                "SavedVariance": block.var('saved_variance')
+            },
+            attrs={
+                "momentum": test_case.momentum,
+                "epsilon": test_case.epsilon,
+                "is_test": False,
+                "data_layout": data_layout,
+                "use_mkldnn": test_case.use_mkldnn,
+                "fuse_with_relu": test_case.fuse_with_relu,
+                "use_global_stats": test_case.use_global_stats
+            })
+        block.create_var(
+            name='y@GRAD', dtype='float32', shape=var_dict['y'].shape)
+
+        # generate backward op_desc
+        grad_op_desc_list, op_grad_to_var = core.get_grad_op_desc(
+            bn_op.desc, test_case.no_grad_set, [])
+        grad_op_desc = grad_op_desc_list[0]
+        new_op_desc = block.desc.append_op()
+        new_op_desc.copy_from(grad_op_desc)
+        for var_name in grad_op_desc.output_arg_names():
+            block.desc.var(var_name.encode("ascii"))
+        grad_op_desc.infer_var_type(block.desc)
+        grad_op_desc.infer_shape(block.desc)
+        for arg in grad_op_desc.output_arg_names():
+            grad_var = block.desc.find_var(arg.encode("ascii"))
+            grad_var.set_dtype(core.VarDesc.VarType.FP32)
+
+        exe = fluid.Executor(place)
+
+        # Do at least 2 iterations
+        for i in range(2):
+            out = exe.run(
+                program,
+                feed={
+                    name: var_dict[name]
+                    for name in
+                    ['x', 'scale', 'bias', 'mean', 'variance', 'y@GRAD']
+                },
+                fetch_list=test_case.fetch_list)
+            for id, name in enumerate(test_case.fetch_list):
+                __assert_close(test_case, var_dict[name], out[id], name)
+
+        print("MKLDNN op test forward passed: ", str(place), data_layout)
 
 
 def format_reorder(out, size):

python/paddle/fluid/tests/unittests/mkldnn/test_batch_norm_mkldnn_op.py

@@ -22,6 +22,7 @@ import paddle.fluid as fluid
 from paddle.fluid.tests.unittests.op_test import OpTest
 from paddle.fluid.framework import grad_var_name
 from paddle.fluid.tests.unittests.test_batch_norm_op import TestBatchNormOpInference, TestBatchNormOpTraining, _reference_training, _reference_grad
+from mkldnn_op_test import check_if_mkldnn_batchnorm_primitives_exist_in_bwd
 
 
 class TestMKLDNNBatchNormOpTraining(TestBatchNormOpTraining):
@@ -43,6 +44,36 @@ class TestMKLDNNBatchNormOpTraining(TestBatchNormOpTraining):
         return y, mean_out, variance_out, saved_mean, saved_variance, x_grad, scale_grad, bias_grad
 
 
+class TestMKLDNNBatchNormOpExistedPrimitives(TestMKLDNNBatchNormOpTraining):
+    def init_test_case(self):
+        TestMKLDNNBatchNormOpTraining.init_test_case(self)
+        self.fetch_list = ['y', 'x@GRAD']
+
+    def test_forward_backward(self):
+        place = core.CPUPlace()
+        shape = [2, 3, 4, 5]
+        scale_shape = [3]
+        data_layout = "NCHW"
+        # initialize the ground-truth
+        np.random.seed(123)
+        x = np.random.random_sample(shape).astype(np.float32)
+        scale = np.random.random_sample(scale_shape).astype(np.float32)
+        bias = np.random.random_sample(scale_shape).astype(np.float32)
+        mean, variance = self.set_mean_variance(scale_shape, x, data_layout)
+        y_grad = np.random.random_sample(shape).astype(np.float32)
+
+        y, mean_out, variance_out, saved_mean, saved_variance, x_grad, scale_grad, bias_grad = self.ref_forward_backward(
+            x, y_grad, scale, bias, mean, variance, self.epsilon, self.momentum,
+            shape, data_layout)
+        var_dict = locals()
+        var_dict['y@GRAD'] = y_grad
+        var_dict['x@GRAD'] = x_grad
+        var_dict['scale@GRAD'] = scale_grad
+        var_dict['bias@GRAD'] = bias_grad
+        check_if_mkldnn_batchnorm_primitives_exist_in_bwd(self, var_dict, place,
+                                                          shape, data_layout)
+
+
 class TestMKLDNNBatchNormOpInference(TestBatchNormOpInference):
     def init_kernel_type(self):
         self.use_mkldnn = True
@@ -50,7 +81,6 @@ class TestMKLDNNBatchNormOpInference(TestBatchNormOpInference):
     def test_check_output(self):
         place = core.CPUPlace()
         data_format = "NCHW"
-
         self.check_with_place(place, data_format, self.dtype, [2, 3, 4, 5])
 
 
@@ -62,7 +92,6 @@ class TestMKLDNNBatchNormOpWithReluInference(TestBatchNormOpInference):
     def test_check_output(self):
         place = core.CPUPlace()
         data_format = "NCHW"
-
         self.check_with_place(place, data_format, self.dtype, [2, 3, 4, 5])
 
 

python/paddle/fluid/tests/unittests/mkldnn/test_gaussian_random_mkldnn_op.py

@@ -19,7 +19,22 @@ import unittest
 from paddle.fluid.tests.unittests.test_gaussian_random_op import TestGaussianRandomOp
 
 
-class TestMKLDNN(TestGaussianRandomOp):
+class TestMKLDNNGaussianRandomOpSeed10(TestGaussianRandomOp):
+    def init_kernel_type(self):
+        self.use_mkldnn = True
+
+
+class TestMKLDNNGaussianRandomOpSeed0(TestGaussianRandomOp):
+    def setUp(self):
+        TestGaussianRandomOp.setUp(self)
+        self.attrs = {
+            "shape": [1000, 784],
+            "mean": .0,
+            "std": 1.,
+            "seed": 0,
+            "use_mkldnn": self.use_mkldnn
+        }
+
     def init_kernel_type(self):
         self.use_mkldnn = True