[SWE-Paddle] Add task PaddlePaddle__Paddle-73850

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 |

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.

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

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。

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<T>(out);
++    return;
++  }
+   // get broadcast dim
+   std::vector<int64_t> x_bst_dims_vec;
+   std::vector<int64_t> 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<T>(out);
++    return;
++  }
+   // get broadcast dim
+   std::vector<int64_t> x_bst_dims_vec;
+   std::vector<int64_t> 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 <typename T, typename Context>
+@@ -38,6 +38,17 @@ void TriangularSolveGradKernel(const Context& dev_ctx,
+                                bool unitriangular,
+                                DenseTensor* dx,
+                                DenseTensor* dy) {
++  if (out.numel() == 0) {
++    if (dx) {
++      phi::Full<T, Context>(
++          dev_ctx, phi::IntArray(common::vectorize(dx->dims())), 0, dx);
++    }
++    if (dy) {
++      phi::Full<T, Context>(
++          dev_ctx, phi::IntArray(common::vectorize(dy->dims())), 0, dy);
++    }
++    return;
++  }
+   std::vector<int64_t> x_bst_dims_vec;
+   std::vector<int64_t> y_bst_dims_vec;
+   std::tie(x_bst_dims_vec, y_bst_dims_vec) =

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()

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