【Hackathon 10th Spring No.52】[WIP][Build] Enable C++20 toolchain baseline - part 4#79260
Open
gouzil wants to merge 7 commits into
Open
【Hackathon 10th Spring No.52】[WIP][Build] Enable C++20 toolchain baseline - part 4#79260gouzil wants to merge 7 commits into
gouzil wants to merge 7 commits into
Conversation
Contributor
There was a problem hiding this comment.
Pull request overview
This PR advances Paddle’s C++20 toolchain baseline by switching the primary CMake/CUDA/CINN build standard to C++20, updating CI/docker toolchains accordingly, and applying a small set of source/third-party compatibility fixes needed for C++20 builds.
Changes:
- Switch core build flags/toolchain gates to C++20 (and require newer GCC/Clang/AppleClang).
- Refresh CI docker images/workflows and Windows default toolchain settings to match the new baseline.
- Fix C++20 incompatibilities (e.g.,
std::result_ofremoval) in both in-tree code and a third-party threadpool patch.
Reviewed changes
Copilot reviewed 17 out of 17 changed files in this pull request and generated 2 comments.
Show a summary per file
| File | Description |
|---|---|
| tools/dockerfile/ci_dockerfile.sh | Update generated CI Dockerfiles (CPU/CUDA baselines, cuDNN/TRT/GCC pins). |
| patches/threadpool/ThreadPool.h.patch | Patch third-party ThreadPool for C++20 (result_of → invoke_result_t). |
| paddle/pir/src/core/type_util.cc | C++20-related lambda signature tweak (also touches shape-compat logic). |
| paddle/pir/include/pass/analysis_manager.h | Add <type_traits> and resolve template overload ambiguities under C++20. |
| paddle/phi/core/memory/allocation/retry_allocator.cc | Fix atomic memory order spelling for C++20 builds. |
| paddle/fluid/framework/new_executor/workqueue/workqueue.h | Replace std::result_of with std::invoke_result_t. |
| paddle/fluid/framework/io/fs.cc | Avoid deprecated shared_ptr::unique() in C++20. |
| paddle/fluid/framework/barrier.h | Replace std::result_of with std::invoke_result_t and add header include. |
| paddle/common/ddim.h | Replace std::result_of with std::invoke_result_t and add header include. |
| cmake/flags.cmake | Enforce C++20 and raise minimum compiler versions. |
| cmake/external/threadpool.cmake | Apply threadpool C++20 patch during external project step. |
| cmake/cuda.cmake | Move CUDA compilation standard to C++20 (required). |
| cmake/cinn.cmake | Update CINN fallback -std message/flags to C++20. |
| ci/windows/config_env.bat | Update default Windows generator to VS2019. |
| ci/windows/build.bat | Update VS vcvars path to VS2019. |
| .github/workflows/Night_ALL_Coverage.yml | Point coverage image build to new Dockerfile name. |
| .github/workflows/docker.yml | Update docker workflow env to new Dockerfile names. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| dims.begin(), dims.end(), dims.front(), [](auto fold, auto dim) { | ||
| return ShapedTypeInterface::IsDynamic(dim) ? fold : dim; | ||
| }); | ||
| return std::all_of(dims.begin(), dims.begin(), [&](auto dim) { |
| if %errorlevel% NEQ 0 exit /b 1 | ||
| if exist .git\index.lock del .git\index.lock 2>NUL | ||
| if not defined GENERATOR echo GENERATOR="Visual Studio 15 2017 Win64">> %GITHUB_ENV% | ||
| if not defined GENERATOR echo GENERATOR="Visual Studio 16 2019">> %GITHUB_ENV% |
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
CI报告基于以下代码生成（30分钟更新一次）: 1 Required任务 : 20/40 通过
2 失败详情🔴 Linux-NPU / Test — PR问题（置信度: 高）分析器: ci_analyze_paddle
关键日志:
修复建议:
关联变更: 🔴 Coverage build — 环境问题（置信度: 高）分析器: ci_analyze_paddle
关键日志:
修复建议:
关联变更: 未关联 PR 源码变更 🔴 Windows-GPU / Build and test — PR问题（置信度: 高）分析器: ci_analyze_paddle
关键日志:
修复建议:
关联变更: 🔴 Check approval — 需要 Approval（置信度: 高）该 Job 需要人工 Approval，完成审批后 CI 才会继续执行。 🔴 其余 8 个失败任务 — 未知（置信度: 低）受本轮时间与默认深度分析数量限制，以下任务未继续下载单 job 深度日志，标记为分析省略/待后续深挖：
|
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
PaddlePaddle-bot
left a comment
PaddlePaddle-bot
left a comment
There was a problem hiding this comment.
🤖 Paddle-CI-Agent | pr_review |
2026-06-08 10:43:26
📋 Review 摘要
PR 概述：将主构建链路推进到 C++20 baseline，并修复一批 C++20 下的源码/third_party 编译问题。
变更范围：cmake/、ci/windows/、patches/、PIR/AP/Framework/Phi 局部 C++ 代码
影响面 Tag：[Environment Adaptation] [Execute Infrastructure] [CINN] [Operator Mechanism]
问题
| 级别 | 文件 | 概述 |
|---|---|---|
| 🟡 建议 | PR 描述 | 当前描述声称更新 Linux CI/Docker baseline，但本次 diff 未包含 .github/workflows/docker.yml、tools/dockerfile/ci_dockerfile.sh 或 Dockerfile 变更，描述与代码不一致。 |
历史 Findings 修复情况
| Finding | 问题 | 状态 |
|---|---|---|
| F1 | PATCH_COMMAND 使用 && shell 语法存在可移植性隐患 |
|
| F2 | VS2019 generator 缺少 64 位架构指定 | |
| F3 | 零上下文 patch（--unidiff-zero）的脆弱性 |
|
| F4 | type_util.cc 中 std::all_of(dims.begin(), dims.begin()) 仍未修复 |
|
| F5 | Windows CUDA flag 与 CMAKE_CUDA_STANDARD 17 仍不一致 |
📝 PR 规范检查
标题格式检查：当前标题为 【Hackathon 10th Spring No.52】[WIP][Build] Enable C++20 toolchain baseline - part 4，包含 Hackathon 前缀和 [WIP] 标记，[Build] 不在官方 Tag 枚举内。
标题建议（可直接复制）：
[Environment Adaptation] Enable C++20 toolchain baseline (part 4)
PR 描述建议（点击展开，可直接复制）
### PR Category
Environment Adaptation
### PR Types
Devs
### Description
本 PR 建立 Paddle 的 C++20 工具链基线（第 4 部分），主要改动如下：
- 将主 C++、CUDA（`cmake/cuda.cmake`）、CINN（`cmake/cinn.cmake`）标准入口切换到 C++20，并启用 `CMAKE_CXX_STANDARD_REQUIRED ON`。
- 更新工具链 gate（`cmake/flags.cmake`）：GCC >= 12、Clang >= 14、AppleClang >= 14。
- 更新 Linux CI/Docker baseline（`tools/dockerfile/ci_dockerfile.sh`、`.github/workflows/docker.yml`）：
- CPU 镜像切到 GCC 13（`Dockerfile.gcc13_ubuntu20_cpu`）。
- Coverage 镜像切到 CUDA 12.0 + Ubuntu 22.04 + GCC 12（`Dockerfile.cuda120_cudnn8_gcc12_ubuntu22_coverage`）。
- Build/CE framework 镜像切到 CUDA 12.3 + cuDNN 9 + TensorRT 8 + GCC 12.2（`Dockerfile.cuda123_cudnn9_gcc122_trt8`）。
- Distribute 镜像路径同步 GCC 12.2 命名。
- 将 Windows 默认构建工具链从 VS2017 更新到 VS2019（`ci/windows/build.bat`、`ci/windows/config_env.bat`）。
- 增加 `patches/threadpool/ThreadPool.h.patch`，修复第三方 threadpool 在 C++20 下 `std::result_of` 问题。
- 修复本机 C++20 baseline build 暴露的最小源码 hard error：
- `std::result_of` → `std::invoke_result_t`（`paddle/common/ddim.h`、`paddle/fluid/framework/barrier.h`、`paddle/fluid/framework/new_executor/workqueue/workqueue.h`）
- `shared_ptr::unique()` → `use_count() == 1`（`paddle/fluid/framework/io/fs.cc`）
- `std::memory_order::memory_order_relaxed` → `std::memory_order_relaxed`（`paddle/phi/core/memory/allocation/retry_allocator.cc`）
- C++20 模板重载约束歧义修复（`paddle/pir/include/pass/analysis_manager.h`）
- `std::accumulate` lambda 引用绑定问题（`paddle/pir/src/core/type_util.cc`）
**未覆盖范围**：HIP、XPU、IPU、NPU、DCU、JIT/custom-op/device SDK 的 C++ 标准收敛留给后续 PR；CUDA 11.x、GCC 8.2、部分外部 SDK 路径作为 legacy/exception 保留。
### 是否引起精度变化
否总体评价
整体改动聚焦 C++20 baseline 编译问题修复，本轮未发现新的阻塞性代码缺陷。合入前建议同步解决历史 build 配置问题，并让 PR 描述与当前 diff 保持一致。
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
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
Environment Adaptation
PR Types
Devs
Description
本 PR 建立 Paddle 的 C++20 工具链基线，将主 CMake/CUDA/CINN 构建入口切换到 C++20，并同步更新常规 CI 镜像、Windows 默认工具链和本机 baseline build 暴露的最小源码兼容问题。
本 PR 不一次性迁移所有 JIT、custom-op、HIP/XPU/IPU/NPU/DCU 等外部 SDK 或设备路径中的分散
-std=c++17标志。主要改动
patches/threadpool/ThreadPool.h.patch，修复第三方 threadpool 在 C++20 下使用std::result_of的问题。std::result_of->std::invoke_result_tshared_ptr::unique()->use_count() == 1std::memory_order_relaxed写法std::accumulatelambda 引用绑定问题：std::accumulate的效果描述为acc = std::move(acc) + *i或acc = binary_op(std::move(acc), *i)，因此 binary op 的第一个参数会收到累加器的右值。[](int64_t& a, int64_t& b)要求第一个参数是非常量左值引用，不能绑定std::move(acc)产生的右值；标准的引用初始化规则也规定，非 const/volatile 左值引用在不能按左值引用规则初始化时是 ill-formed。因此在 GCC 13/libstdc++ + nvcc 路径下会报引用绑定错误。[](int64_t a, int64_t b)只复制两个int64_t标量维度值，不复制 Tensor/vector 数据，也不修改输入元素；该 lambda 原本只做a * b，所以计算语义不变。[accumulate]：https://eel.is/c++draft/algorithms#accumulate-2；C++ working draft[dcl.init.ref]：https://eel.is/c++draft/dcl.init.ref#5.2。std::accumulate对 C++20std::move(acc)行为的说明：https://en.cppreference.com/w/cpp/algorithm/accumulate。未覆盖范围 / 后续工作
相关链接
是否引起精度变化
否