Coverage for  / home / jenkins / .local / lib / python3.10 / site-packages / hyper_parallel / core / shard / ops / parallel_argsort.py: 97%

1# Copyright 2026 Huawei Technologies Co., Ltd 

2# 

3# Licensed under the Apache License, Version 2.0 (the "License"); 

4# you may not use this file except in compliance with the License. 

5# You may obtain a copy of the License at 

6# 

7# http://www.apache.org/licenses/LICENSE-2.0 

8# 

9# Unless required by applicable law or agreed to in writing, software 

10# distributed under the License is distributed on an "AS IS" BASIS, 

11# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 

12# See the License for the specific language governing permissions and 

13# limitations under the License. 

14# ============================================================================ 

15""" 

16Distributed implementation for Argsort operator. 

17""" 

18 

19from typing import Tuple 

20 

21from .parallel_ops import DistributedOp 

22 

23 

24def _normalize_argsort_args(x, dim=-1, descending=False, stable=False): 

25 """Normalize torch.argsort arguments to (args, kwargs). 

26 

27 torch.argsort(input, dim=-1, descending=False, *, stable=False) 

28 Only `stable` is keyword-only; all other params are positional. 

29 """ 

30 return (x, dim, descending), {'stable': stable} 

31 

32 

33class ArgsortDistributedOp(DistributedOp): 

34 """Distributed implementation for torch.argsort.""" 

35 

36 _MS_PRIMITIVE_OP_NAMES = frozenset({'ArgSort'}) 

37 

38 def preprocess(self, args: tuple, kwargs: dict) -> tuple: 

39 """ 

40 Preprocess arguments for Argsort operator. 

41 

42 Args: 

43 args (tuple): Input arguments, first element is the input tensor. 

44 kwargs (dict): Keyword arguments (dim, descending, stable). 

45 

46 Returns: 

47 tuple: (local_args, local_kwargs, cache_values) 

48 """ 

49 args, kwargs = _normalize_argsort_args(*args, **kwargs) 

50 input_tensor = args[0] 

51 dim = args[1] 

52 descending = args[2] 

53 stable = kwargs['stable'] 

54 

55 if self.op_name in self._MS_PRIMITIVE_OP_NAMES: 

56 local_args = (input_tensor.to_local(), dim, descending, stable) 

57 local_kwargs = {} 

58 else: 

59 local_args = (input_tensor.to_local(),) 

60 local_kwargs = {'dim': dim, 'descending': descending, 'stable': stable} 

61 

62 cache_values = [input_tensor.layout, dim] 

63 return local_args, local_kwargs, cache_values 

64 

65 def infer_layout(self, cache_values: list) -> Tuple[tuple, None]: 

66 """ 

67 Infer output layout for Argsort operator. 

68 

69 Rules: 

70 1. Input must not have Partial status. 

71 2. dim must be an integer within the valid range [-ndim, ndim-1]. 

72 3. The sort dimension must not be sharded. 

73 4. Output layout is identical to the input layout. 

74 

75 Args: 

76 cache_values (list): [input_layout, dim] where dim is the sort dimension. 

77 

78 Returns: 

79 tuple: ((output_layout,), None) 

80 

81 Raises: 

82 ValueError: If input has Partial status, dim is out of range, or the sort 

83 dimension is sharded. 

84 """ 

85 layout = cache_values[0] 

86 dim = cache_values[1] 

87 

88 if not self._allow_partial_inputs: 

89 self._check_partial_inputs([layout]) 

90 

91 if not isinstance(dim, int): 

92 raise ValueError( 

93 f"For {self.op_name}, dimension should be int, but got {type(dim)}" 

94 ) 

95 

96 alias_map = layout.alias_tensor_map 

97 ndim = len(alias_map) 

98 

99 if dim < -ndim or dim >= ndim: 

100 raise ValueError( 

101 f"For {self.op_name}, dimension out of range " 

102 f"(expected to be in range of [{-ndim}, {ndim - 1}], but got {dim})" 

103 ) 

104 

105 if dim < 0: 

106 dim += ndim 

107 

108 if alias_map[dim] != "None": 

109 raise ValueError( 

110 f"For {self.op_name}, sorting along a sharded dimension " 

111 f"(dim {dim} mapped to {alias_map[dim]}) is not supported. " 

112 f"Please redistribute the tensor to Replicate on this dimension before sorting." 

113 ) 

114 

115 return ((layout,), None)