Add non-contiguous test.

Author: baoqiwen

tester/api_config/config_analyzer.py

@@ -9,8 +9,10 @@
 import numpy
 import paddle
 import torch
+import random
 
 USE_CACHED_NUMPY = os.getenv("USE_CACHED_NUMPY", "False").lower() == "true"
+TEST_NON_CONTIGUOUS = os.getenv("TEST_NON_CONTIGUOUS", "False").lower() == "true"
 cached_numpy = {}
 
 not_zero_apis = frozenset(
@@ -84,6 +86,7 @@ def __init__(self, shape, dtype, place=None):
         self.numpy_tensor = None
         self.paddle_tensor = None
         self.torch_tensor = None
+        self.shuffle_dims = None
 
     def __deepcopy__(self, memo):
         cls = self.__class__
@@ -2642,6 +2645,13 @@ def get_paddle_tensor(self, api_config):
             self.paddle_tensor.stop_gradient = False
             if self.dtype == "bfloat16":
                 self.paddle_tensor = paddle.cast(self.paddle_tensor, dtype="bfloat16")
+        if TEST_NON_CONTIGUOUS:
+            if not self.shuffle_dims:
+                ndim = self.paddle_tensor.dim()
+                self.shuffle_dims = list(range(ndim))
+                random.shuffle(self.shuffle_dims)
+            print("paddle shuffle:", self.shuffle_dims)
+            return paddle.transpose(self.paddle_tensor, self.shuffle_dims)
         return self.paddle_tensor
 
     def get_torch_tensor(self, api_config):
@@ -2669,6 +2679,13 @@ def get_torch_tensor(self, api_config):
             )
             if self.dtype == "bfloat16":
                 self.torch_tensor = self.torch_tensor.to(dtype=torch.bfloat16)
+        if TEST_NON_CONTIGUOUS:
+            if not self.shuffle_dims:
+                ndim = self.torch_tensor.dim()
+                self.shuffle_dims = list(range(ndim))
+                random.shuffle(self.shuffle_dims)
+            print("torch shuffle:", self.shuffle_dims)
+            return torch.permute(self.torch_tensor, self.shuffle_dims)
         return self.torch_tensor
 
     def clear_tensor(self):