[SWE-Paddle] Add task PaddlePaddle__Paddle-73387

swe-paddle/tasks/PaddlePaddle__Paddle-73387/README.md

@@ -0,0 +1,51 @@
+# PaddlePaddle__Paddle-73387
+
+This directory converts Paddle PR #73387 into a SWE-Paddle community task candidate.
+
+## Source
+
+| Field | Value |
+| --- | --- |
+| Repo | `PaddlePaddle/Paddle` |
+| PR | [73387](https://github.com/PaddlePaddle/Paddle/pull/73387) |
+| PR title | `[0-size Tensor Job2 No.58] Add 0-size Tensor support for gather_tree` |
+| Base commit | `57d91535621b3b793c2bd1e6d5dcc2801ed893fd` |
+| Gold commit | `71179f5ae909c4577479beb77321f87b9b0b00ae` |
+| Merged at | `2025-06-19` |
+| Task type | `bug_fix` |
+| Resource | CPU |
+| Scope | C++ Operator Kernel |
+
+## Summary
+
+Fix `gather_tree` operator to correctly handle 0-size tensors by adding early-return logic in CPU/GPU kernels and skipping shape equality checks in InferMeta when input is 0-size.
+
+## 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 C++ operator kernel level and requires rebuilding Paddle from source.
+- The failure is deterministic: the base revision fails when processing 0-size tensors due to shape mismatch checks and kernel execution on empty data.
+- 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 (C++ kernel and InferMeta changes).
+- `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) | gather_tree F2P |
+| --- | ---: | ---: |
+| Base + `tests/test.patch` | PASS | FAIL |
+| Base + `tests/test.patch` + `solution/code.patch` | PASS | PASS |

swe-paddle/tasks/PaddlePaddle__Paddle-73387/environment/README.md

@@ -0,0 +1,52 @@
+# Environment Notes
+
+## Expected Environment
+
+- Repository: `PaddlePaddle/Paddle`
+- Base commit: `57d91535621b3b793c2bd1e6d5dcc2801ed893fd`
+- Gold commit: `71179f5ae909c4577479beb77321f87b9b0b00ae`
+- Resource: CPU
+- GPU required: no
+- Patch type: C++ kernel (CPU/GPU backends) + InferMeta + Symbolic Shape
+- Python dependencies: PaddlePaddle (source build), NumPy
+
+The verifier should execute against the Paddle source revision represented by the selected patch state. A source build is required since the patch modifies C++ kernel code, InferMeta, and symbolic shape inference.
+
+## Build Instructions
+
+1. Check out `PaddlePaddle/Paddle` at the base commit.
+2. Apply `tests/test.patch`.
+3. 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)
+   ```
+4. 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 case should fail before the fix.
+6. Apply `solution/code.patch`.
+7. Rebuild Paddle from source.
+8. Reinstall Paddle package.
+9. Run `bash tests/test.sh`; all target tests should pass.
+
+## Minimal Test Command
+
+```bash
+bash tests/test.sh
+```
+
+## Expected Matrix
+
+| Revision state | P2P | gather_tree 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-73387/instruction.md

@@ -0,0 +1,57 @@
+# 修复 `gather_tree` 对 0-size Tensor 的处理
+
+## 详细描述
+
+当 `gather_tree(ids, parents)` 的输入 `ids` 和 `parents` 为 0-size Tensor（即 `out.numel() == 0`）时，当前实现存在以下问题：
+
+1. **InferMeta 层**：`GatherTreeMeta` 函数会强制检查 `ids_dims == parents_dims`，但当输入为 0-size 时，这个检查可能不必要或导致错误。
+2. **符号推导层**：`GatherTreeOpInferSymbolicShape` 在遍历维度时，会对 0-size 维度添加等式约束，这在 0-size 情况下是不合适的。
+3. **Kernel 层**：CPU 和 GPU kernel 在分配输出内存后，直接进入计算逻辑，对 0-size 输入会访问无效内存或执行无意义的计算。
+
+典型表现包括：
+
+- InferMeta 阶段抛出 shape 不匹配的异常
+- Kernel 执行时出现段错误或未定义行为
+- 0-size Tensor 输入无法通过 `gather_tree` 算子
+
+例如：
+
+```python
+import numpy as np
+import paddle
+
+paddle.disable_static()
+
+# 0-size tensor 输入
+ids = np.random.randint(0, high=10, size=(0, 2, 2)).astype('int64')
+parents = np.random.randint(0, high=2, size=(0, 2, 2)).astype('int64')
+
+ids_tensor = paddle.to_tensor(ids)
+parents_tensor = paddle.to_tensor(parents)
+
+out = paddle.nn.functional.gather_tree(ids_tensor, parents_tensor)
+# 期望返回 shape 为 (0, 2, 2) 的空 Tensor
+```
+
+上述调用中 `ids` 和 `parents` 的 shape 为 `[0, 2, 2]`，`out.numel() == 0`。按照 API semantics，当输入为 0-size 时，该调用应正常完成并返回正确 shape 的空 Tensor。
+
+需要在以下位置进行修改：
+- InferMeta 层：当 `ids` 的 numel 为 0 时，跳过 shape 相等性检查
+- 符号推导层：在遍历维度时，跳过 0-size 维度的等式约束添加
+- Kernel 层：在分配输出内存后，检查 `out->numel() == 0` 并直接返回
+
+## 验收说明
+
+- 当输入为 0-size Tensor 时，`gather_tree` 应正常完成，返回正确 shape 的空 Tensor
+- 输出的 shape 应与输入 `ids` 一致
+- 非 0-size Tensor 输入下的 `gather_tree` 行为不得退化
+- 梯度计算也应正常工作（0-size Tensor 的梯度也为空 Tensor）
+
+## 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 `gather_tree` 算子的输入输出语义
+- 了解 Paddle CPU/GPU kernel 的多 backend 实现模式
+- 了解 InferMeta 和符号推导机制
+- 需要从源码编译 Paddle 以验证修改

swe-paddle/tasks/PaddlePaddle__Paddle-73387/proposal.md

@@ -0,0 +1,62 @@
+# SWE-Paddle Task Proposal: PaddlePaddle__Paddle-73387
+
+## 1. 来源信息
+
+- Instance ID: `PaddlePaddle__Paddle-73387`
+- PR 链接: https://github.com/PaddlePaddle/Paddle/pull/73387
+- PR 标题: `[0-size Tensor Job2 No.58] Add 0-size Tensor support for gather_tree`
+- Base commit: `57d91535621b3b793c2bd1e6d5dcc2801ed893fd`
+- Gold commit: `71179f5ae909c4577479beb77321f87b9b0b00ae`
+- Merged at: 2025-06-19
+- 你的身份: contributor
+
+## 2. 问题一句话
+
+`gather_tree` 在输入为 0-size Tensor 时，InferMeta 的 shape 检查、符号推导的维度约束和 kernel 的计算逻辑均未处理 0-size 边界情况，需要在三个层面分别添加 0-size 早期返回或跳过逻辑。
+
+## 3. 为什么适合作为 SWE-Paddle 样本
+
+- **真实性**: 来自 Paddle「0-size Tensor 机制建设」系列任务，是真实研发需求。
+- **代表性**: 覆盖 C++ kernel 层面的 0-size Tensor 边界处理，涉及 CPU/GPU 双端 kernel、InferMeta 和符号推导三个层面的修改。
+- **边界清楚**: 目标仅限输入为 0-size 时的三个层面的早期返回/跳过逻辑；正向非零尺寸输入不应受影响。
+- **非平凡性**: 修复需要在三个层面分别处理：
+  - InferMeta: 当 `common::product(ids_dims) != 0` 时才进行 shape 检查
+  - 符号推导: 当 `ids_shape[i] == 0` 时跳过等式约束添加
+  - Kernel: 当 `out->numel() == 0` 时直接返回
+  涉及对 kernel 执行流程、InferMeta 和符号推导机制的理解。
+- **回归护栏明确**: 目标 F2P 可覆盖 0-size Tensor 输入的 `gather_tree` 算子测试；同文件中已有的 `TestGatherTreeOp` 等标准测试用例可作为 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: `[operator_kernel, gather_tree, 0-size_tensor, cpu_kernel, gpu_kernel, infermeta, symbolic_shape]`
+
+## 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_gather_tree_op.py`（`TestGatherTreeOp_ZeroSize`、`TestGatherTreeOp_ZeroSize2`）
+- P2P 候选: 同文件中已有的 `TestGatherTreeOp`、`TestGatherTreeOp_batch_size_1` 等标准 gather_tree 算子测试用例。
+- 修复前预期: `base_commit` + `tests/test.patch` 后，0-size Tensor 输入的 `gather_tree` 算子测试失败（InferMeta shape 检查或 kernel 执行报错）。
+- 修复后预期: 继续应用 `solution/code.patch` 并重新编译后，0-size Tensor 输入正常返回空 Tensor，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: 含 C++ kernel 修改（CPU/GPU 双端）+ InferMeta + 符号推导，需要重新编译 Paddle。
+- 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「gather_tree 对 0-size Tensor 输入的行为异常」，不指出具体 `out->numel() == 0` 分支逻辑或具体代码位置。
+- 环境风险: 中。任务涉及 C++ kernel 修改，需要源码编译 Paddle，编译时间较长。
+- 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 `gather_tree` 算子的 0-size 边界处理，涉及三个层面的修改，测试明确指向 `TestGatherTreeOp_ZeroSize` 和 `TestGatherTreeOp_ZeroSize2`，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（`TestGatherTreeOp_ZeroSize` 和 `TestGatherTreeOp_ZeroSize2`）在 `base_commit` 编译后确实失败。

swe-paddle/tasks/PaddlePaddle__Paddle-73387/solution/code.patch

@@ -0,0 +1,65 @@
+diff --git a/paddle/fluid/pir/dialect/operator/interface/infer_symbolic_shape/binary_infer_sym.cc b/paddle/fluid/pir/dialect/operator/interface/infer_symbolic_shape/binary_infer_sym.cc
+index 16b7a8186c4ab..5c08b36179ad7 100644
+--- a/paddle/fluid/pir/dialect/operator/interface/infer_symbolic_shape/binary_infer_sym.cc
++++ b/paddle/fluid/pir/dialect/operator/interface/infer_symbolic_shape/binary_infer_sym.cc
+@@ -1058,6 +1058,9 @@ bool GatherTreeOpInferSymbolicShape(
+                         "shape of Input(Ids)."));
+   size_t rank = ids_shape.size();
+   for (size_t i = 0; i < rank; ++i) {
++    if (ids_shape[i] == 0) {
++      continue;
++    }
+     infer_context->AddEqualCstr(ids_shape[i], parents_shape[i]);
+   }
+
+diff --git a/paddle/phi/infermeta/binary.cc b/paddle/phi/infermeta/binary.cc
+index k911dab1a6d0f..013a8e9488b6b 100644
+--- a/paddle/phi/infermeta/binary.cc
++++ b/paddle/phi/infermeta/binary.cc
+@@ -2138,11 +2138,13 @@ void GatherTreeMeta(const MetaTensor& ids,
+                     MetaTensor* out) {
+   auto ids_dims = ids.dims();
+   auto parents_dims = parents.dims();
+-  PADDLE_ENFORCE_EQ(ids_dims == parents_dims,
+-                    true,
+-                    common::errors::InvalidArgument(
+-                        "The shape of Input(Parents) must be same with the "
+-                        "shape of Input(Ids)."));
++  if (common::product(ids_dims) != 0) {
++    PADDLE_ENFORCE_EQ(ids_dims == parents_dims,
++                      true,
++                      common::errors::InvalidArgument(
++                          "The shape of Input(Parents) must be same with the "
++                          "shape of Input(Ids)."));
++  }
+   out->set_dims(ids_dims);
+   out->set_dtype(ids.dtype());
+ }
+diff --git a/paddle/phi/kernels/cpu/gather_tree_kernel.cc b/paddle/phi/kernels/cpu/gather_tree_kernel.cc
+index a7e85a731b824..4e3bc2be0f2ff 100644
+--- a/paddle/phi/kernels/cpu/gather_tree_kernel.cc
++++ b/paddle/phi/kernels/cpu/gather_tree_kernel.cc
+@@ -28,6 +28,9 @@ void GatherTreeKernel(const Context &dev_ctx,
+   const auto *parents_data = parents.data<T>();
+
+   T *out_data = dev_ctx.template Alloc<T>(out);
++  if (out && out->numel() == 0) {
++    return;
++  }
+
+   auto &ids_dims = ids.dims();
+   int64_t max_length = ids_dims[0];
+diff --git a/paddle/phi/kernels/gpu/gather_tree_kernel.cu b/paddle/phi/kernels/gpu/gather_tree_kernel.cu
+index 76aa7f4b00f2a..4ea564105ff5f 100644
+--- a/paddle/phi/kernels/gpu/gather_tree_kernel.cu
++++ b/paddle/phi/kernels/gpu/gather_tree_kernel.cu
+@@ -63,6 +63,9 @@ void GatherTreeKernel(const Context &dev_ctx,
+   const auto *ids_data = ids.data<T>();
+   const auto *parents_data = parents.data<T>();
+   T *out_data = dev_ctx.template Alloc<T>(out);
++  if (out && out->numel() == 0) {
++    return;
++  }
+
+   PADDLE_ENFORCE_NOT_NULL(ids_data,
+                           common::errors::InvalidArgument(

swe-paddle/tasks/PaddlePaddle__Paddle-73387/tests/test.patch

@@ -0,0 +1,38 @@
+diff --git a/test/legacy_test/test_gather_tree_op.py b/test/legacy_test/test_gather_tree_op.py
+index 5379ec37c774a..289a82c4c2fa6 100644
+--- a/test/legacy_test/test_gather_tree_op.py
++++ b/test/legacy_test/test_gather_tree_op.py
+@@ -150,6 +150,33 @@ def test_parents_ndim():
+         paddle.disable_static()
+
+
++class TestGatherTreeOp_ZeroSize(OpTest):
++    def init_shape(self):
++        self.ids_shape = (0, 2, 2)
++        self.parents_shape = (0, 2, 2)
++
++    def setUp(self):
++        self.op_type = "gather_tree"
++        self.python_api = paddle.nn.functional.gather_tree
++        self.init_shape()
++        ids_shape = self.ids_shape
++        parents_shape = self.parents_shape
++        max_length, batch_size, beam_size = ids_shape
++        ids = np.random.randint(0, high=10, size=ids_shape)
++        parents = np.random.randint(0, high=beam_size, size=parents_shape)
++        self.inputs = {"Ids": ids, "Parents": parents}
++        self.outputs = {'Out': ids}
++
++    def test_check_output(self):
++        self.check_output(check_pir=True)
++
++
++class TestGatherTreeOp_ZeroSize2(TestGatherTreeOp_ZeroSize):
++    def init_shape(self):
++        self.ids_shape = (0, 2, 2)
++        self.parents_shape = (1, 2, 2)
++
++
+ if __name__ == "__main__":
+     paddle.enable_static()
+     unittest.main()

swe-paddle/tasks/PaddlePaddle__Paddle-73387/tests/test.sh

@@ -0,0 +1,9 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+# P2P tests (pass-to-pass)
+python -m pytest test/legacy_test/test_gather_tree_op.py::TestGatherTreeOp -q
+
+# F2P tests (fail-to-pass)
+python -m pytest test/legacy_test/test_gather_tree_op.py::TestGatherTreeOp_ZeroSize -q
+python -m pytest test/legacy_test/test_gather_tree_op.py::TestGatherTreeOp_ZeroSize2 -q