[Feature] Add collect_results support for Ascend NPU

[xuuyangg]

Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily get feedback. If you do not understand some items, don't worry, just make the pull request and seek help from maintainers.

Motivation

1. The use of the 'set_npu_compile' function can lead to failures in launching multiprocessing on Ascend NPU. Please refer to the provided reference for more details. [Bug] Multiprocessing fails to launch on Ascend NPU device when import mmengine.dist #1307 https://www.hiascend.com/document/detail/zh/canncommercial/63RC2/modeldevpt/ptmigr/ptmigr_0058.html
2. collect_resulst for Ascend NPU added.

Modification

add collect_resulst_npu.

Use case

"""test.py"""

import os
import torch
import torch_npu
import mmengine.dist as dist
import torch.distributed as torch_dist

torch.npu.set_device(int(os.environ['LOCAL_RANK']))
torch_dist.init_process_group(backend="hccl")


if dist.get_rank() == 0:
    data = [torch.tensor(1, dtype=torch.float32).npu(), torch.tensor(2,dtype=torch.float32).npu()]
else:
    data = [torch.tensor(3, dtype=torch.float32).npu(), torch.tensor(4,dtype=torch.float32).npu()]
size = 4
output = dist.collect_results(data, size, device='npu')

if dist.get_rank() == 0:
    print(output)

command

torchrun --nproc_per_node 2 test.py

Checklist

1. Pre-commit or other linting tools are used to fix the potential lint issues.
2. The modification is covered by complete unit tests. If not, please add more unit test to ensure the correctness.
3. If the modification has potential influence on downstream projects, this PR should be tested with downstream projects, like MMDet or MMCls.
4. The documentation has been modified accordingly, like docstring or example tutorials.

[CLAassistant]

All committers have signed the CLA.