[0-size Tensor No.11、262、287] Add 0-size Tensor support for argsort/sort API. #72872

enkilee · 2025-05-22T07:11:43Z

PR Category

Execute Infrastructure

PR Types

Improvements

Description

No.11、287 argsort:
在xpu_test中加入单测。

develop的代码经测试没问题

aistudio@jupyter-115720-9209054:~/PaddleAPITest$ python engine.py --api_config_file="/home/aistudio/PaddleAPITest/tester/api_config/big_and_0size/argsort.txt" --paddle_only=True
2025-05-27 02:56:52.431372 test begin: paddle.argsort(Tensor([0, 1],"float64"), axis=0, )
W0527 02:57:01.197355  6586 gpu_resources.cc:114] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 12.0, Runtime API Version: 11.8
[Pass] paddle.argsort(Tensor([0, 1],"float64"), axis=0, )
2025-05-27 02:57:01.205345 test begin: paddle.argsort(Tensor([0, 1],"float64"), axis=0, descending=True, )
[Pass] paddle.argsort(Tensor([0, 1],"float64"), axis=0, descending=True, )
2025-05-27 02:57:01.207407 test begin: paddle.argsort(Tensor([0, 1],"float64"), axis=0, descending=True, stable=True, )
[Pass] paddle.argsort(Tensor([0, 1],"float64"), axis=0, descending=True, stable=True, )
2025-05-27 02:57:01.208636 test begin: paddle.argsort(Tensor([0, 1],"float64"), axis=0, stable=True, )
[Pass] paddle.argsort(Tensor([0, 1],"float64"), axis=0, stable=True, )
2025-05-27 02:57:01.209786 test begin: paddle.argsort(Tensor([0, 3, 4],"float64"), axis=0, descending=True, stable=True, )
[Pass] paddle.argsort(Tensor([0, 3, 4],"float64"), axis=0, descending=True, stable=True, )
2025-05-27 02:57:01.210981 test begin: paddle.argsort(Tensor([0, 3, 4],"float64"), axis=0, stable=True, )
[Pass] paddle.argsort(Tensor([0, 3, 4],"float64"), axis=0, stable=True, )
2025-05-27 02:57:01.212088 test begin: paddle.argsort(Tensor([0, 3, 4],"float64"), axis=1, )
[Pass] paddle.argsort(Tensor([0, 3, 4],"float64"), axis=1, )
2025-05-27 02:57:01.213198 test begin: paddle.argsort(Tensor([0, 3, 4],"float64"), axis=1, descending=True, )
[Pass] paddle.argsort(Tensor([0, 3, 4],"float64"), axis=1, descending=True, )
2025-05-27 02:57:01.216692 test begin: paddle.argsort(Tensor([0, 3, 4],"float64"), descending=True, )
[Pass] paddle.argsort(Tensor([0, 3, 4],"float64"), descending=True, )
2025-05-27 02:57:01.217781 test begin: paddle.argsort(Tensor([0, 30],"float64"), axis=1, descending=True, stable=True, )
[Pass] paddle.argsort(Tensor([0, 30],"float64"), axis=1, descending=True, stable=True, )
2025-05-27 02:57:01.218922 test begin: paddle.argsort(Tensor([0, 30],"float64"), axis=1, stable=True, )
[Pass] paddle.argsort(Tensor([0, 30],"float64"), axis=1, stable=True, )
2025-05-27 02:57:01.220044 test begin: paddle.argsort(Tensor([0, 4, 2],"int64"), axis=0, stable=True, )
[Pass] paddle.argsort(Tensor([0, 4, 2],"int64"), axis=0, stable=True, )
2025-05-27 02:57:01.221209 test begin: paddle.argsort(Tensor([0, 4, 2],"int64"), axis=1, stable=True, )
[Pass] paddle.argsort(Tensor([0, 4, 2],"int64"), axis=1, stable=True, )
2025-05-27 02:57:01.222335 test begin: paddle.argsort(Tensor([0, 4, 2],"int64"), axis=2, stable=True, )
[Pass] paddle.argsort(Tensor([0, 4, 2],"int64"), axis=2, stable=True, )
2025-05-27 02:57:01.233197 test begin: paddle.argsort(Tensor([0, 4],"int64"), axis=0, stable=True, )
[Pass] paddle.argsort(Tensor([0, 4],"int64"), axis=0, stable=True, )

No.262:
根据log日式显示，
改了argsort，就等于改了sort

【开源任务】 Paddle API 0-size 机制建设 #72637

Pcard-67164

paddle-bot · 2025-05-22T07:11:48Z

你的PR提交成功，感谢你对开源项目的贡献!
请关注后续CI自动化测试结果，详情请参考Paddle-CI手册。
Your PR has been submitted. Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

DanielSun11 · 2025-05-23T07:04:20Z

修改cpu和gpu的kernel之后需要添加对应的单测。PR中补充paddleAPITest的测试结果

enkilee · 2025-05-23T08:01:42Z

修改cpu和gpu的kernel之后需要添加对应的单测。PR中补充paddleAPITest的测试结果

test_argsort_op.py中已经有0-size Tensor的单测了

paddle/phi/kernels/cpu/argsort_kernel.cc

paddle/phi/kernels/gpu/argsort_kernel.cu

paddle/phi/kernels/gpu/argsort_grad_kernel.cu

DanielSun11 · 2025-05-26T11:53:12Z

请补充下paddleAPITest的测试结果。
请一并报名任务287 paddle.Tensor.argsort

paddle/phi/kernels/xpu/argsort_kernel.cc

DanielSun11 · 2025-05-27T02:58:49Z

XPU的前向ArgsortKernel似乎没有添加对0 size Tensor的支持，请检查

enkilee · 2025-05-27T03:13:39Z

XPU的前向ArgsortKernel似乎没有添加对0 size Tensor的支持，请检查

if (rank == 0) {
phi::Copy(dev_ctx, input, dev_ctx.GetPlace(), false, output);
phi::funcs::set_constant(dev_ctx, indices, static_cast<int64_t>(0));
return;
}

DanielSun11 · 2025-05-27T03:29:19Z

XPU的前向ArgsortKernel似乎没有添加对0 size Tensor的支持，请检查

if (rank == 0) { phi::Copy(dev_ctx, input, dev_ctx.GetPlace(), false, output); phi::funcs::set_constant(dev_ctx, indices, static_cast<int64_t>(0)); return; }

auto rank = in_dims.size();这里rank指的是有几个维度，rank为0是0-dim。0-size指的是某个维度为0，进而所有维度的乘积也就是numel为0 请注意区分二者的区别。0-size Tensor的rank不一定为0，比如shape为[1,0,2]的Tensor，rank为3，numel = 1*0*2 = 0

enkilee · 2025-05-27T04:01:05Z

XPU的前向ArgsortKernel似乎没有添加对0 size Tensor的支持，请检查

if (rank == 0) { phi::Copy(dev_ctx, input, dev_ctx.GetPlace(), false, output); phi::funcs::set_constant(dev_ctx, indices, static_cast<int64_t>(0)); return; }

auto rank = in_dims.size();这里rank指的是有几个维度，rank为0是0-dim。0-size指的是某个维度为0，进而所有维度的乘积也就是numel为0 请注意区分二者的区别。0-size Tensor的rank不一定为0，比如shape为[1,0,2]的Tensor，rank为3，numel = 102 = 0

明白了，学习，谢谢

luotao1 · 2025-05-29T03:08:43Z

@DanielSun11 可以review了

DanielSun11

对于paddle.as_real的处理不够完善。
paddle.as_rea 报的是InvalidArgument，此时说明是参数检查报的错，但是并不代表不需要修改。请根据 https://www.paddlepaddle.org.cn/documentation/docs/zh/3.0-beta/api/paddle/as_real_cn.html#as-real 检查下参数是否合法，以及验证一下torch是否支持同样的shape。如果0 size Tensor的输入是合法的但是报了InvalidArgument，此时需要修改

enkilee · 2025-05-29T07:03:06Z

对于paddle.as_real的处理不够完善。 paddle.as_rea 报的是InvalidArgument，此时说明是参数检查报的错，但是并不代表不需要修改。请根据 https://www.paddlepaddle.org.cn/documentation/docs/zh/3.0-beta/api/paddle/as_real_cn.html#as-real 检查下参数是否合法，以及验证一下torch是否支持同样的shape。如果0 size Tensor的输入是合法的但是报了InvalidArgument，此时需要修改

没太明白，一切以APITest工具为准吗？可是错误是3月份的log,5月份的并没有as_real的错，而且直接调用，3月份的错误代码现在执行并不会报错

import paddle
x = paddle.to_tensor([10, 0, 10, 20], dtype = "complex128")
s = paddle.as_real(x)
print(s)

W0529 07:01:43.434468 19996 gpu_resources.cc:114] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 12.0, Runtime API Version: 11.8
Tensor(shape=[4, 2], dtype=float64, place=Place(gpu:0), stop_gradient=True,
       [[10., 0. ],
        [0. , 0. ],
        [10., 0. ],
        [20., 0. ]])

DanielSun11 · 2025-05-29T07:09:39Z

10, 0, 10, 20]

paddleAPITest的config中 paddle.as_real(Tensor([0, 10, 10, 20],"complex128"), ) ，[0, 10, 10, 20]是Tensor的shape，并不是tensor初始化的tuple，请重新测一下。测试结果以paddleAPITest为准，但是仍然需要添加单测来验证一些极端case。

enkilee · 2025-05-29T07:18:58Z

10, 0, 10, 20]
paddleAPITest的config中 paddle.as_real(Tensor([0, 10, 10, 20],"complex128"), ) ，[0, 10, 10, 20]是Tensor的shape，并不是tensor初始化的tuple，请重新测一下。测试结果以paddleAPITest为准，但是仍然需要添加单测来验证一些极端case。

好的，那我12重开PR

wanghuancoder

这个还需要改GPU、CPU的Kernel

enkilee · 2025-05-29T07:58:38Z

as_real按照3月份log的shape测试了下，好像也没有问题：

import paddle
x = paddle.randn(shape=[0, 4], dtype='complex64')
z = paddle.as_real(x)
print(z)
y = paddle.randn(shape=[10, 0, 10, 20], dtype='complex128')
z = paddle.as_real(y)
print(z)

result:

Tensor(shape=[0, 4, 2], dtype=float32, place=Place(gpu:0), stop_gradient=True,
       [])
Tensor(shape=[10, 0, 10, 20, 2], dtype=float64, place=Place(gpu:0), stop_gradient=True,
       [[],
        [],
        [],
        [],
        [],
        [],
        [],
        [],
        [],
        []])

DanielSun11

LGTM

fix

1398bff

paddle-bot bot added the contributor External developers label May 22, 2025

luotao1 mentioned this pull request May 22, 2025

【开源任务】 Paddle API 0-size 机制建设 #72637

Open

luotao1 added the HappyOpenSource Pro 进阶版快乐开源活动，更具挑战性的任务 label May 22, 2025

luotao1 assigned luotao1, wanghuancoder and DanielSun11 May 22, 2025

enkilee added 3 commits May 23, 2025 10:39

fix

56a9f18

fix

fde1a1e

fix

b5ba165

enkilee changed the title ~~[0-size Tensor No.12] Add 0-size Tensor support for argsort API.~~ [0-size Tensor No.11] Add 0-size Tensor support for argsort API. May 23, 2025

fix

159309a

enkilee changed the title ~~[0-size Tensor No.11] Add 0-size Tensor support for argsort API.~~ [0-size Tensor No.11、12] Add 0-size Tensor support for argsort API. May 23, 2025

DanielSun11 reviewed May 23, 2025

View reviewed changes

paddle/phi/kernels/cpu/argsort_kernel.cc Outdated Show resolved Hide resolved

paddle/phi/kernels/gpu/argsort_kernel.cu Outdated Show resolved Hide resolved

paddle/phi/kernels/gpu/argsort_grad_kernel.cu Outdated Show resolved Hide resolved

fix

d356842

enkilee changed the title ~~[0-size Tensor No.11、12] Add 0-size Tensor support for argsort API.~~ [0-size Tensor No.11、12、262] Add 0-size Tensor support for argsort API. May 26, 2025

enkilee changed the title ~~[0-size Tensor No.11、12、262] Add 0-size Tensor support for argsort API.~~ [0-size Tensor No.11、12、262] Add 0-size Tensor support for argsort/as_real/sort API. May 26, 2025

co63oc mentioned this pull request May 26, 2025

[0-size Tensor No.93] Add 0-size Tensor support for isin #72920

Merged

DanielSun11 requested changes May 26, 2025

View reviewed changes

paddle/phi/kernels/xpu/argsort_kernel.cc Outdated Show resolved Hide resolved

enkilee and others added 2 commits May 27, 2025 09:52

Merge branch 'PaddlePaddle:develop' into 0size-11-argsort

cab15d8

fix

1d1fec8

enkilee changed the title ~~[0-size Tensor No.11、12、262] Add 0-size Tensor support for argsort/as_real/sort API.~~ [0-size Tensor No.11、12、262、287] Add 0-size Tensor support for argsort/as_real/sort API. May 27, 2025

fix

9e0e973

This comment was marked as abuse.

Sign in to view

DanielSun11 self-requested a review May 29, 2025 06:36

DanielSun11 requested changes May 29, 2025

View reviewed changes

enkilee changed the title ~~[0-size Tensor No.11、12、262、287] Add 0-size Tensor support for argsort/as_real/sort API.~~ [0-size Tensor No.11、262、287] Add 0-size Tensor support for argsort/as_real/sort API. May 29, 2025

wanghuancoder reviewed May 29, 2025

View reviewed changes

enkilee changed the title ~~[0-size Tensor No.11、262、287] Add 0-size Tensor support for argsort/as_real/sort API.~~ [0-size Tensor No.11、262、287] Add 0-size Tensor support for argsort/sort API. May 29, 2025

DanielSun11 approved these changes May 30, 2025

View reviewed changes

wanghuancoder approved these changes May 30, 2025

View reviewed changes

wanghuancoder merged commit cf56cbd into PaddlePaddle:develop May 30, 2025
98 of 100 checks passed

[0-size Tensor No.11、262、287] Add 0-size Tensor support for argsort/sort API. #72872

[0-size Tensor No.11、262、287] Add 0-size Tensor support for argsort/sort API. #72872

Uh oh!

Conversation

enkilee commented May 22, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

PR Category

PR Types

Description

Uh oh!

paddle-bot bot commented May 22, 2025

Uh oh!

DanielSun11 commented May 23, 2025

Uh oh!

enkilee commented May 23, 2025

Uh oh!

Uh oh!

Uh oh!

Uh oh!

DanielSun11 commented May 26, 2025

Uh oh!

Uh oh!

DanielSun11 commented May 27, 2025

Uh oh!

enkilee commented May 27, 2025

Uh oh!

DanielSun11 commented May 27, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

enkilee commented May 27, 2025

Uh oh!

luotao1 commented May 29, 2025

Uh oh!

This comment was marked as abuse.

Uh oh!

DanielSun11 left a comment

Choose a reason for hiding this comment

Uh oh!

enkilee commented May 29, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

DanielSun11 commented May 29, 2025

Uh oh!

enkilee commented May 29, 2025

Uh oh!

wanghuancoder left a comment

Choose a reason for hiding this comment

Uh oh!

enkilee commented May 29, 2025

Uh oh!

DanielSun11 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

enkilee commented May 22, 2025 •

edited

Loading

DanielSun11 commented May 27, 2025 •

edited

Loading

enkilee commented May 29, 2025 •

edited

Loading