From 9b7a4cf25edae3fdc912d6dffe3512d5b549a0a0 Mon Sep 17 00:00:00 2001 From: co63oc <4617245+co63oc@users.noreply.github.com> Date: Tue, 11 Aug 2026 16:05:20 +0800 Subject: [PATCH] [SWE-Paddle] Add task PaddlePaddle__Paddle-73850 --- .../PaddlePaddle__Paddle-73850/README.md | 51 ++++++++++++++ .../environment/README.md | 56 +++++++++++++++ .../PaddlePaddle__Paddle-73850/instruction.md | 43 ++++++++++++ .../PaddlePaddle__Paddle-73850/proposal.md | 58 ++++++++++++++++ .../solution/code.patch | 68 +++++++++++++++++++ .../tests/test.patch | 21 ++++++ .../PaddlePaddle__Paddle-73850/tests/test.sh | 8 +++ 7 files changed, 305 insertions(+) create mode 100644 swe-paddle/tasks/PaddlePaddle__Paddle-73850/README.md create mode 100644 swe-paddle/tasks/PaddlePaddle__Paddle-73850/environment/README.md create mode 100644 swe-paddle/tasks/PaddlePaddle__Paddle-73850/instruction.md create mode 100644 swe-paddle/tasks/PaddlePaddle__Paddle-73850/proposal.md create mode 100644 swe-paddle/tasks/PaddlePaddle__Paddle-73850/solution/code.patch create mode 100644 swe-paddle/tasks/PaddlePaddle__Paddle-73850/tests/test.patch create mode 100644 swe-paddle/tasks/PaddlePaddle__Paddle-73850/tests/test.sh diff --git a/swe-paddle/tasks/PaddlePaddle__Paddle-73850/README.md b/swe-paddle/tasks/PaddlePaddle__Paddle-73850/README.md new file mode 100644 index 000000000..462883934 --- /dev/null +++ b/swe-paddle/tasks/PaddlePaddle__Paddle-73850/README.md @@ -0,0 +1,51 @@ +# PaddlePaddle__Paddle-73850 + +This directory converts Paddle PR #73850 into a SWE-Paddle community task candidate. + +## Source + +| Field | Value | +| --- | --- | +| Repo | `PaddlePaddle/Paddle` | +| PR | [73850](https://github.com/PaddlePaddle/Paddle/pull/73850) | +| PR title | `[0-size Tensor No.118] Add 0-size Tensor support for paddle.linalg.triangular_solve` | +| Base commit | `917f720a58b3ed5aeb8a1ac0022fdbd76f3b2b4b` | +| Gold commit | `0a23433eddfd286cbdb8746240eaf662cd027c69` | +| Merged at | `2025-07-08` | +| Task type | `bug_fix` | +| Resource | CPU | +| Scope | C++ phi kernel + Python test | + +## Summary + +Fix `paddle.linalg.triangular_solve` to correctly handle 0-size tensors in both forward and backward passes. The forward kernel adds an early return when `x.numel() == 0 || y.numel() == 0`, and the backward kernel fills gradients with 0 when `out.numel() == 0`. + +## Why This Is A Good SWE-Paddle Candidate + +- It is derived from a merged Paddle bug-fix PR rather than a synthetic issue. +- The target behavior is isolated to the triangular_solve phi kernels (CPU/GPU) and grad kernel impl. +- The failure is deterministic: the base revision fails when processing 0-size tensors in triangular_solve. +- The task has clear regression coverage for existing non-zero-size behavior. +- The task runs on CPU and does not require distributed execution, external services, or additional datasets. + +## Files + +- `proposal.md`: candidate proposal for maintainer triage. +- `instruction.md`: self-contained problem statement for the coding agent. +- `solution/code.patch`: gold implementation patch. +- `tests/test.patch`: tests exposing the target behavior. +- `tests/test.sh`: minimal target test command. +- `environment/README.md`: environment and reproduction notes. + +## Verification + +```bash +bash tests/test.sh +``` + +Expected behavior: + +| Revision state | Existing behavior (P2P) | triangular_solve F2P | +| --- | ---: | ---: | +| Base + `tests/test.patch` | PASS | FAIL | +| Base + test patch + solution patch | PASS | PASS | diff --git a/swe-paddle/tasks/PaddlePaddle__Paddle-73850/environment/README.md b/swe-paddle/tasks/PaddlePaddle__Paddle-73850/environment/README.md new file mode 100644 index 000000000..e43fcc4db --- /dev/null +++ b/swe-paddle/tasks/PaddlePaddle__Paddle-73850/environment/README.md @@ -0,0 +1,56 @@ +# Environment Notes + +## Expected Environment + +- Repository: `PaddlePaddle/Paddle` +- Base commit: `917f720a58b3ed5aeb8a1ac0022fdbd76f3b2b4b` +- Gold commit: `0a23433eddfd286cbdb8746240eaf662cd027c69` +- Resource: CPU +- GPU required: no +- Patch type: C++ kernel (phi kernels) + Python test +- Python dependencies: PaddlePaddle (source build), NumPy, pytest + +The verifier should execute against the Paddle source revision represented by the selected patch state. Since the patch modifies C++ kernel files, a source rebuild is required after applying the solution patch. + +## Build Instructions + +1. Check out `PaddlePaddle/Paddle` at the base commit. +2. Build Paddle from source (CPU-only build is sufficient): +```bash +mkdir build && cd build +cmake .. -DWITH_GPU=OFF -DWITH_TESTING=ON -DCMAKE_BUILD_TYPE=Release +make -j$(nproc) +``` +3. Install the built Paddle package. + +## Run Order + +1. Check out `PaddlePaddle/Paddle` at the base commit. +2. Build and install Paddle from source. +3. Apply `tests/test.patch`. +4. Run the P2P tests; existing non-zero-size behavior should pass. +5. Run the 0-size tensor tests; the target cases should fail before the fix. +6. Apply `solution/code.patch`. +7. Rebuild Paddle from source so the modified C++ kernels take effect: +```bash +cd build +make -j$(nproc) +pip install --force-reinstall python/dist/paddlepaddle-*.whl +cd .. +``` +8. Run `bash tests/test.sh`; all target tests should pass. + +## Minimal Test Command + +```bash +bash tests/test.sh +``` + +## Expected Matrix + +| Revision state | P2P | triangular_solve F2P | +| --- | ---: | ---: | +| Base + test patch | PASS | FAIL | +| Base + test patch + solution patch | PASS | PASS | + +No GPU, distributed runtime, external service, or additional dataset is required. diff --git a/swe-paddle/tasks/PaddlePaddle__Paddle-73850/instruction.md b/swe-paddle/tasks/PaddlePaddle__Paddle-73850/instruction.md new file mode 100644 index 000000000..9a2376de4 --- /dev/null +++ b/swe-paddle/tasks/PaddlePaddle__Paddle-73850/instruction.md @@ -0,0 +1,43 @@ +# 修复 `paddle.linalg.triangular_solve` 对 0-size Tensor 的处理 + +## 详细描述 + +当 `paddle.linalg.triangular_solve(x, y, ...)` 的输入 `x` 或 `y` 中存在大小为 `0` 的 dimension，即 `x.numel() == 0` 或 `y.numel() == 0` 时，当前实现会直接进入底层算子，导致报错或产生错误结果。 + +典型表现包括： + +- 底层算子在处理 0-size tensor 时出现 shape 推断异常或计算错误 +- 调用失败并抛出与 shape 相关的错误 + +例如： + +```python +import numpy as np +import paddle + +paddle.disable_static() +x = paddle.to_tensor(np.random.random([0, 2, 2]).astype('float32')) +y = paddle.to_tensor(np.random.random([0, 2, 1]).astype('float32')) +out = paddle.linalg.triangular_solve(x, y, upper=False, left=True, unitriangular=False) +# 期望: out shape 为 [0, 2, 1]，正常返回空 tensor +``` + +上述调用中 `x` 的 shape 为 `[0, 2, 2]`，`y` 的 shape 为 `[0, 2, 1]`，不包含任何元素。按照 API semantics，当输入 tensor 的 numel 为 0 时，不存在需要求解的方程组，因此该调用应正常完成并返回正确 shape 的空 tensor。 + +此外，反向传播也需要正确处理 0-size tensor 的情况。当 `out.numel() == 0` 时，`dx` 和 `dy` 应被填充为 0。 + +当前 C++ kernel 层在进入底层计算之前，没有对 0-size tensor 输入进行显式的早期返回处理。需要在 forward kernel 中检测 `x.numel() == 0 || y.numel() == 0` 并直接分配输出后返回；在 backward kernel 中检测 `out.numel() == 0` 并将梯度填充为 0。 + +## 验收说明 + +- 当输入 tensor 的 numel 为 0 时，`paddle.linalg.triangular_solve` 前向应正常完成，返回正确 shape 的空 tensor +- 返回的 out tensor 应保持与输入相同的 dtype +- 反向传播时，当 `out.numel() == 0`，`dx` 和 `dy` 应被正确填充为 0 +- 非 0-size tensor 输入下的 triangular_solve 行为不得退化 + +## 技术要求 + +- 熟悉 C++ 和 Paddle phi kernel 开发 +- 了解 Tensor shape、0-size Tensor 和 kernel 执行路径 +- 了解 triangular_solve 算子的前向和反向语义 +- 了解 Paddle phi kernel 的 CPU/GPU 实现结构 diff --git a/swe-paddle/tasks/PaddlePaddle__Paddle-73850/proposal.md b/swe-paddle/tasks/PaddlePaddle__Paddle-73850/proposal.md new file mode 100644 index 000000000..3a476bced --- /dev/null +++ b/swe-paddle/tasks/PaddlePaddle__Paddle-73850/proposal.md @@ -0,0 +1,58 @@ +# SWE-Paddle Task Proposal: PaddlePaddle__Paddle-73850 + +## 1. 来源信息 + +- Instance ID: `PaddlePaddle__Paddle-73850` +- PR 链接: https://github.com/PaddlePaddle/Paddle/pull/73850 +- PR 标题: `[0-size Tensor No.118] Add 0-size Tensor support for paddle.linalg.triangular_solve` +- Base commit: `917f720a58b3ed5aeb8a1ac0022fdbd76f3b2b4b` +- Gold commit: `0a23433eddfd286cbdb8746240eaf662cd027c69` +- Merged at: 2025-07-08 +- 你的身份: contributor + +## 2. 问题一句话 + +`paddle.linalg.triangular_solve` 在动态图模式下对 0-size tensor（任意维度含有 0）的输入缺少显式处理，前向进入底层算子时出错，反向未正确填充梯度为 0，需要补齐 0-size tensor 支持。 + +## 3. 为什么适合作为 SWE-Paddle 样本 + +- **真实性**：该问题来自 Paddle 的「0-size Tensor 机制建设」系列任务，是真实研发需求，目标是为 `triangular_solve` 算子补齐 0-size tensor 支持。 +- **代表性**：覆盖 C++ phi kernel 层面的算子边界处理，涉及 forward kernel 的早期返回和 backward kernel 的梯度填充，是 Paddle 算子机制增强的典型样本。 +- **边界清楚**：目标仅限 0-size tensor 输入的 forward/backward 早期返回逻辑；正向非零尺寸输入不应受影响。 +- **非平凡性**：修复需要在 CPU kernel、GPU kernel 和 grad kernel impl 三处分别添加 0-size 判断逻辑，涉及 `numel() == 0` 检查、输出分配和梯度填充（使用 `phi::Full`），不是简单机械修改。 +- **回归护栏明确**：目标 F2P 可覆盖 0-size tensor 输入的 `triangular_solve` 前向和反向调用；同文件已有的 `TestTriangularSolveOp` 等标准算子测试用例可作为 P2P 护栏。 + +## 4. 任务类型和标签 + +- 任务类型：`bug_fix` +- 执行后端：`cpu` +- 设备范围：`cpu_only` +- 模块标签：`[phi_kernel, triangular_solve, 0-size_tensor, forward, backward, cpu, gpu]` + +## 5. 验证思路 + +- 目标测试命令：`bash tests/test.sh` +- 目标测试文件： + - `test/legacy_test/test_triangular_solve_op.py`（`TestTriangularSolveOp_ZeroSize`） +- P2P 候选：同文件中已有的 `TestTriangularSolveOp` 等标准 triangular_solve 算子测试用例。 +- 修复前预期：`base_commit` + `tests/test.patch` 后，0-size tensor 输入在 `triangular_solve` 的前向/反向调用中失败（进入底层算子时出错）。 +- 修复后预期：继续应用 `solution/code.patch` 后，0-size tensor 输入返回正确的空 tensor（shape 与预期一致），反向梯度填充为 0，P2P 存量测试仍然通过。 + +## 6. 环境与资源 + +- 是否能提供 Docker: 无 +- Dockerfile 或镜像地址: 暂无 +- Paddle 来源: `PaddlePaddle/Paddle` source checkout at `base_commit`，需要源码编译。 +- OS / Python / CUDA / cuDNN / 其他关键依赖: Linux CPU + Python + numpy；编译需要 CMake、GCC；不要求 CUDA/cuDNN（CPU 编译即可验证）。 +- 硬件: CPU 即可（编译和测试均不需要 GPU）。 +- patch 类型: C++ kernel 修改（`paddle/phi/kernels/cpu/triangular_solve_kernel.cc`、`paddle/phi/kernels/gpu/triangular_solve_kernel.cu`、`paddle/phi/kernels/impl/triangular_solve_grad_kernel_impl.h`），需要重新编译。 +- 最小测试命令: `bash tests/test.sh` +- 是否有 oracle 日志: 无 + +## 7. 风险自查 + +- 泄露风险：正式 `instruction.md` 只描述「triangular_solve 对 0-size tensor 输入的行为异常」，不指出具体 `numel() == 0` 分支逻辑或具体代码位置。 +- 环境风险：中。任务需要 Paddle 源码环境，patch 涉及 C++ 文件，需要重新编译。 +- flaky 风险：低。测试使用固定的 0-size tensor 构造，不依赖随机数差异或多设备同步。 +- 拆分风险：低。该 PR 目标集中在 triangular_solve 的 forward/backward kernel 的 0-size 处理，测试明确指向零尺寸分支，适合作为一个独立样本。 +- 其他不确定点：完整任务包阶段应确认新增 F2P（`TestTriangularSolveOp_ZeroSize`）在 `base_commit` 上确实失败，并选择同文件中已有的 `TestTriangularSolveOp` 等标准测试用例作为在 base 与修复后都稳定通过的 P2P nodeid。 diff --git a/swe-paddle/tasks/PaddlePaddle__Paddle-73850/solution/code.patch b/swe-paddle/tasks/PaddlePaddle__Paddle-73850/solution/code.patch new file mode 100644 index 000000000..c6fc95ac0 --- /dev/null +++ b/swe-paddle/tasks/PaddlePaddle__Paddle-73850/solution/code.patch @@ -0,0 +1,68 @@ +diff --git a/paddle/phi/kernels/cpu/triangular_solve_kernel.cc b/paddle/phi/kernels/cpu/triangular_solve_kernel.cc +index 68af8bc2b1..de3ae7ef06 100644 +--- a/paddle/phi/kernels/cpu/triangular_solve_kernel.cc ++++ b/paddle/phi/kernels/cpu/triangular_solve_kernel.cc +@@ -32,6 +32,10 @@ void TriangularSolveKernel(const Context& dev_ctx, + bool transpose, + bool unitriangular, + DenseTensor* out) { ++ if (x.numel() == 0 || y.numel() == 0) { ++ dev_ctx.template Alloc(out); ++ return; ++ } + // get broadcast dim + std::vector x_bst_dims_vec; + std::vector y_bst_dims_vec; +diff --git a/paddle/phi/kernels/gpu/triangular_solve_kernel.cu b/paddle/phi/kernels/gpu/triangular_solve_kernel.cu +index 2c2bb299f8..5311cea5fc 100644 +--- a/paddle/phi/kernels/gpu/triangular_solve_kernel.cu ++++ b/paddle/phi/kernels/gpu/triangular_solve_kernel.cu +@@ -33,6 +33,10 @@ void TriangularSolveKernel(const Context& dev_ctx, + bool transpose, + bool unitriangular, + DenseTensor* out) { ++ if (x.numel() == 0 || y.numel() == 0) { ++ dev_ctx.template Alloc(out); ++ return; ++ } + // get broadcast dim + std::vector x_bst_dims_vec; + std::vector y_bst_dims_vec; +diff --git a/paddle/phi/kernels/impl/triangular_solve_grad_kernel_impl.h b/paddle/phi/kernels/impl/triangular_solve_grad_kernel_impl.h +index 483694335a..ad656b7a6c 100644 +--- a/paddle/phi/kernels/impl/triangular_solve_grad_kernel_impl.h ++++ b/paddle/phi/kernels/impl/triangular_solve_grad_kernel_impl.h +@@ -16,6 +16,7 @@ + + #include "paddle/phi/core/tensor_utils.h" + #include "paddle/phi/kernels/empty_kernel.h" ++#include "paddle/phi/kernels/full_kernel.h" + #include "paddle/phi/kernels/funcs/blas/blas.h" + #include "paddle/phi/kernels/funcs/common_shape.h" + #include "paddle/phi/kernels/funcs/complex_functors.h" +@@ -24,7 +25,6 @@ + #include "paddle/phi/kernels/funcs/tril_triu_compute.h" + #include "paddle/phi/kernels/triangular_solve_grad_kernel.h" + #include "paddle/phi/kernels/triangular_solve_kernel.h" +- + namespace phi { + + template +@@ -38,6 +38,17 @@ void TriangularSolveGradKernel(const Context& dev_ctx, + bool unitriangular, + DenseTensor* dx, + DenseTensor* dy) { ++ if (out.numel() == 0) { ++ if (dx) { ++ phi::Full( ++ dev_ctx, phi::IntArray(common::vectorize(dx->dims())), 0, dx); ++ } ++ if (dy) { ++ phi::Full( ++ dev_ctx, phi::IntArray(common::vectorize(dy->dims())), 0, dy); ++ } ++ return; ++ } + std::vector x_bst_dims_vec; + std::vector y_bst_dims_vec; + std::tie(x_bst_dims_vec, y_bst_dims_vec) = diff --git a/swe-paddle/tasks/PaddlePaddle__Paddle-73850/tests/test.patch b/swe-paddle/tasks/PaddlePaddle__Paddle-73850/tests/test.patch new file mode 100644 index 000000000..de3836361 --- /dev/null +++ b/swe-paddle/tasks/PaddlePaddle__Paddle-73850/tests/test.patch @@ -0,0 +1,21 @@ +diff --git a/test/legacy_test/test_triangular_solve_op.py b/test/legacy_test/test_triangular_solve_op.py +index 0475eaf944..b1e194a813 100644 +--- a/test/legacy_test/test_triangular_solve_op.py ++++ b/test/legacy_test/test_triangular_solve_op.py +@@ -840,5 +840,16 @@ class TestTriangularSolveOpError(unittest.TestCase): + ) + + ++class TestTriangularSolveOp_ZeroSize(TestTriangularSolveOp): ++ def config(self): ++ self.__class__.exist_fp64_check_grad = True ++ self.x_shape = [0, 2, 2] ++ self.y_shape = [0, 2, 1] ++ self.upper = False ++ self.transpose = False ++ self.unitriangular = False ++ self.dtype = "float32" ++ ++ + if __name__ == "__main__": + unittest.main() diff --git a/swe-paddle/tasks/PaddlePaddle__Paddle-73850/tests/test.sh b/swe-paddle/tasks/PaddlePaddle__Paddle-73850/tests/test.sh new file mode 100644 index 000000000..399303d6d --- /dev/null +++ b/swe-paddle/tasks/PaddlePaddle__Paddle-73850/tests/test.sh @@ -0,0 +1,8 @@ +#!/usr/bin/env bash +set -euo pipefail + +# P2P tests (pass-to-pass) +python -m pytest test/legacy_test/test_triangular_solve_op.py::TestTriangularSolveOp -q + +# F2P tests (fail-to-pass) +python -m pytest test/legacy_test/test_triangular_solve_op.py::TestTriangularSolveOp_ZeroSize -q