[API-Compat] Add paddle.compat.min/max and new PHI kernel (min/max_with_index) #74547
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
你的PR提交成功，感谢你对开源项目的贡献! |
|
/re-run all-failed |
Codecov Report❌ Patch coverage is Additional details and impacted files@@ Coverage Diff @@
## develop #74547 +/- ##
==========================================
Coverage ? 78.19%
==========================================
Files ? 5
Lines ? 188
Branches ? 0
==========================================
Hits ? 147
Misses ? 41
Partials ? 0 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
Enigmatisms
force-pushed
the
min_max_api_new
branch
3 times, most recently
from
ff2ded8 to
6414203
Compare
|
/re-run all-failed |
Enigmatisms
force-pushed
the
min_max_api_new
branch
from
75b797d to
10eb128
Compare
|
/re-run CI-Build |
Attempting to fix integral type gradient computation (rejection)
…ed implementation
removed split API for independence.
Enigmatisms
force-pushed
the
min_max_api_new
branch
from
b2df5fe to
17f080e
Compare
|
/re-run all-failed |
2 similar comments
|
/re-run all-failed |
|
/re-run all-failed |
zyfncg
previously approved these changes
Aug 22, 2025
Enigmatisms
dismissed stale reviews from XiaoguangHu01, zyfncg, and zhangbo9674
via
0fbbb99
|
/re-run all-failed |
2 similar comments
|
/re-run all-failed |
|
/re-run all-failed |
zyfncg
approved these changes
Aug 25, 2025
Comment on lines
+580
to
+584
| - For case 1: a single value Tensor (0-dim) | ||
| - For case 2: a named tuple MinMaxRetType(values: Tensor, indices: Tensor), `values` has the same data type as the `input`, | ||
| while indices is always an int64 Tensor, with exactly the same shape as `values`. | ||
| MinMaxRetType can be used (indexed, packed, unpacked) in the same way as a regular tuple | ||
| - For case 3: see `paddle.minimum` |
There was a problem hiding this comment.
Suggested change
| - For case 1: a single value Tensor (0-dim) | |
| - For case 2: a named tuple MinMaxRetType(values: Tensor, indices: Tensor), `values` has the same data type as the `input`, | |
| while indices is always an int64 Tensor, with exactly the same shape as `values`. | |
| MinMaxRetType can be used (indexed, packed, unpacked) in the same way as a regular tuple | |
| - For case 3: see `paddle.minimum` | |
| - For case 1. A single value Tensor (0-dim) | |
| - For case 2. A named tuple MinMaxRetType(values: Tensor, indices: Tensor), `values` has the same data type as the `input`, | |
| while indices is always an int64 Tensor, with exactly the same shape as `values`. | |
| MinMaxRetType can be used (indexed, packed, unpacked) in the same way as a regular tuple | |
| - For case 3. See `paddle.minimum`(:ref:`api_paddle_minimum`) |
- Returns内容里，引号
:前面的内容会自动渲染成return type，所以尽量不要用引号
- 加一下case3 api的跳转引用
| ) -> Tensor | MinMaxRetType: | ||
| """ | ||
|
|
||
| Computes the minimum of tensor elements. There are mainly 3 cases (functionalities): |
There was a problem hiding this comment.
Suggested change
| Computes the minimum of tensor elements. There are mainly 3 cases (functionalities): | |
| Computes the minimum of tensor elements. There are mainly 3 cases (functionalities): | |
Comment on lines
+551
to
+552
| Special warning: the gradient behavior is NOT well-documented by PyTorch, the actual behavior should be: | ||
| 1. Case 1: the same as `min` |
There was a problem hiding this comment.
Suggested change
| Special warning: the gradient behavior is NOT well-documented by PyTorch, the actual behavior should be: | |
| 1. Case 1: the same as `min` | |
| Special warning: the gradient behavior is NOT well-documented by PyTorch, the actual behavior should be: | |
| 1. Case 1: the same as `min` |
| ) -> Tensor | MinMaxRetType: | ||
| """ | ||
|
|
||
| Computes the maximum of tensor elements. There are mainly 3 cases (functionalities): |
There was a problem hiding this comment.
Suggested change
| Computes the maximum of tensor elements. There are mainly 3 cases (functionalities): | |
| Computes the maximum of tensor elements. There are mainly 3 cases (functionalities): | |
Comment on lines
+703
to
+704
| Special warning: the gradient behavior is NOT well-documented by PyTorch, the actual behavior should be: | ||
| 1. Case 1: the same as `max` |
There was a problem hiding this comment.
Suggested change
| Special warning: the gradient behavior is NOT well-documented by PyTorch, the actual behavior should be: | |
| 1. Case 1: the same as `max` | |
| Special warning: the gradient behavior is NOT well-documented by PyTorch, the actual behavior should be: | |
| 1. Case 1: the same as `max` |
Comment on lines
+731
to
+736
| Returns: | ||
| - For case 1: a single value Tensor (0-dim) | ||
| - For case 2: a named tuple MinMaxRetType(values: Tensor, indices: Tensor), `values` has the same data type as the `input`, | ||
| while indices is always an int64 Tensor, with exactly the same shape as `values`. | ||
| MinMaxRetType can be used (indexed, packed, unpacked) in the same way as a regular tuple | ||
| - For case 3: see `paddle.maximum` |
There was a problem hiding this comment.
Suggested change
| Returns: | |
| - For case 1: a single value Tensor (0-dim) | |
| - For case 2: a named tuple MinMaxRetType(values: Tensor, indices: Tensor), `values` has the same data type as the `input`, | |
| while indices is always an int64 Tensor, with exactly the same shape as `values`. | |
| MinMaxRetType can be used (indexed, packed, unpacked) in the same way as a regular tuple | |
| - For case 3: see `paddle.maximum` | |
| Returns: | |
| - For case 1. A single value Tensor (0-dim) | |
| - For case 2. A named tuple MinMaxRetType(values: Tensor, indices: Tensor), `values` has the same data type as the `input`, | |
| while indices is always an int64 Tensor, with exactly the same shape as `values`. | |
| MinMaxRetType can be used (indexed, packed, unpacked) in the same way as a regular tuple | |
| - For case 3. See `paddle.maximum` (:ref:`api_paddle_maximum`) |
|
/re-run all-failed |
1 similar comment
|
/re-run all-failed |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
PR Category
Operator Mechanism
PR Types
New features
Description
本 PR 为 #74495 的 reopen 版本，rebase 到了一个更新的版本，并且解决了与 #74506 的冲突。#74495 当时只用于CI发现问题，本 PR 尝试对其中“共用amin/amax backward op 但amin/amax 不支持某些整数类型“的问题进行了修复，基于SFINAE与python端检查。目前本 PR 在 #74506 未合入前会显得改动过多，实际上是包含了部分前序 PR 的改动，前序 PR merge 后应该可以自动 resolve。
本 PR 新增的 feature:
(min/max)_with_index_grad。注意amin/amax的行为：只对minimum/maximum index位置的结果传梯度，像是对梯度进行了一个 take_along_axis。paddle.compat.min,paddle.compat.max，与 torch 的行为进行对齐。torch.min/torch.max输入输出关系很复杂（一个API包含了太多功能）：minimum/maximum一致除上述【情况2】在 CUDA GPU 后端下会调用
(min/max)_with_index，其余情况都是由 python 调用_C_ops.xxx获得结果的。其中情况1/2/3 在CUDA GPU后端下应该都具有较好的性能（没有进行组合，调用单算子完成），而情况2在其他后端下使用 argmin/max 与 take_along_axis 组合（并且需要配合 squeeze_ 操作），不是最优性能方案，但应当具有较高的开发性价比。TODO
test_compat_minmax.py: 等能达到单测覆盖率要求的算子单测：coverage 覆盖率可能显示有误。Pcard-89620