Coverage for  / home / jenkins / .local / lib / python3.10 / site-packages / hyper_multicore / scheduler / builder.py: 100%

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""" 

16Standalone RuntimeConfig builder. 

17 

18build_runtime_config() is the generic framework entry point for converting a 

19ComputeGraph + TaskSplitValue into a binary RuntimeConfigC. It omits 

20graph-specific post-processing (add_terminate, revise_task_queue, etc.) — call 

21build_config_for_rank() from a model's gen_runtime_data.py for full generation. 

22""" 

23from hyper_multicore._core_spec import DEFAULT_CORE_SPEC 

24from hyper_multicore.scheduler.config import ( 

25 RuntimeConfigC, 

26 TaskSplitValue, 

27 init_task_split_value, 

28 mega_moe_event_capacity, 

29) 

30from hyper_multicore.scheduler.graph import ComputeGraph 

31from hyper_multicore.scheduler.runtime import allocate_runtime_config 

32 

33 

34def allocate_graph_config(graph: ComputeGraph, tsv: TaskSplitValue, 

35 num_cube_cores: int = DEFAULT_CORE_SPEC.num_cube_cores) -> RuntimeConfigC: 

36 """Allocate graph tasks plus the terminal before any fill operation. 

37 

38 Args: 

39 graph: Graph after split propagation. 

40 tsv: Topology whose experts determine event storage. 

41 num_cube_cores: Available Cube workers whose group lists need scratch. 

42 

43 Returns: 

44 Zero-initialized runtime with sufficient task, queue and event slots. 

45 """ 

46 required = 1 + sum(op.task_num for op in graph.topological_sort()) 

47 return allocate_runtime_config( 

48 required, mega_moe_event_capacity(tsv.all_expert_num, tsv.ep), tsv.single_rank_expert_num, 

49 num_workers=2 * num_cube_cores, 

50 ) 

51 

52 

53def build_runtime_config(graph: ComputeGraph, tsv: TaskSplitValue, 

54 rank_id: int = 0, 

55 num_cube_cores: int = DEFAULT_CORE_SPEC.num_cube_cores) -> RuntimeConfigC: 

56 """ 

57 Build RuntimeConfigC for a single rank from a ComputeGraph. 

58 

59 Args: 

60 graph: A fully-built ComputeGraph after propagate_splits() has been called. 

61 tsv: TaskSplitValue with TP/EP/seq topology. 

62 rank_id: Target rank (written to tsv.rank_id before filling). 

63 num_cube_cores: Number of Cube workers. 

64 

65 Returns: 

66 RuntimeConfigC ready to be serialized with serialize_runtime_config(cfg). 

67 """ 

68 cfg = allocate_graph_config(graph, tsv, num_cube_cores) 

69 cfg.num_workers = 2 * num_cube_cores 

70 

71 init_task_split_value(tsv) 

72 tsv.rank_id = rank_id 

73 

74 for op in graph.topological_sort(): 

75 op.fill_config.fill(cfg, op, tsv) 

76 

77 cfg.task_num = sum(op.task_num for op in graph.topological_sort()) 

78 cfg.atomic_add_values[0] = 1 

79 return cfg