Skip to content

[2026春季][T1-1-9] DiFanrui#89

Open
DiFanrui wants to merge 1 commit into
InfiniTensor:masterfrom
DiFanrui:2026-spring-DiFanrui-T1-1-9
Open

[2026春季][T1-1-9] DiFanrui#89
DiFanrui wants to merge 1 commit into
InfiniTensor:masterfrom
DiFanrui:2026-spring-DiFanrui-T1-1-9

Conversation

@DiFanrui

@DiFanrui DiFanrui commented Jun 20, 2026

Copy link
Copy Markdown

实现 T1-1-9 赛题的 5 个算子:scatter_add multilabel_margin_loss frac fractional_max_pool2d fractional_max_pool3d

测试结果

全仓库: 871 passed, 0 failed, 88 skipped

tests/test_frac.py ...........          4 passed
tests/test_scatter_add.py ...........  8 passed
tests/test_multilabel_margin_loss.py ................ 6 passed
tests/test_fractional_max_pool.py(2d + 3d) .........  13 passed

──────────────────────────────────────────────
5 个新增算子                          31 passed
全仓库                                871 passed, 0 failed

NVIDIA 平台截图

1-1-9-NV

国产平台

天数

天数设备信息 1-1-9-天数

沐曦

沐曦设备信息 1-1-9-沐曦

摩尔

摩尔设备信息 1-1-9-摩尔

…, fractional_max_pool2d/3d

All five operators implemented with NineToothed kernels:
- frac: element_wise arrangement with where/floor/ceil for trunc
- scatter_add: vectorized ntl.atomic_add scatter
- multilabel_margin_loss: O(C^2) offsets-based + reduce_sum kernel
- fractional_max_pool2d/3d: output-driven vectorized with tail mask
                           and NaN semantics matching PyTorch 2.12

31 tests pass, covering batched/unbatched, return_indices, NaN,
output_size=1, output_ratio, negative dim, repeated destination,
and >128 element cases.

Co-Authored-By: Claude <noreply@anthropic.com>
@DiFanrui DiFanrui changed the title Implement T1-1-9 operators: frac, scatter_add, multilabel_margin_loss… [2026春季][T1-1-9] DiFanrui Jun 20, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant