[Cpp API Compatibility] Complement ScalarType

[youge325]

PR Category

Execute Infrastructure

PR Types

Improvements

Description

完善 ScalarType 类型，提供占位符，以对齐枚举类型

是否引起精度变化

否

[paddle-bot]

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.

[Copilot]

Pull request overview

This PR enhances the C10 compatibility layer’s ScalarType support by adding missing placeholder scalar wrapper types (quantized, bit-packed, and additional float formats) and wiring them into the scalar-type dispatch/mapping macros to better align with upstream enum coverage.

Changes:

• Add placeholder c10 scalar wrapper headers for qint8/qint32, quint8/quint4x2/quint2x4, bits*, and several Float* low-precision formats.
• Extend AT_FORALL_SCALAR_TYPES_WITH_COMPLEX_AND_QINTS to include these types (and ComplexHalf) so toString, elementSize, and type mapping specializations are generated consistently from the macro.
• Remove now-redundant manual toString/elementSize switch cases and scalar-type constants that are covered by the macro generation.

Reviewed changes

Copilot reviewed 11 out of 11 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
paddle/phi/api/include/compat/c10/util/quint8.h	Adds placeholder c10::quint8 wrapper type and exports to at/torch.
paddle/phi/api/include/compat/c10/util/quint4x2.h	Adds placeholder c10::quint4x2 wrapper type and exports to at/torch.
paddle/phi/api/include/compat/c10/util/quint2x4.h	Adds placeholder c10::quint2x4 wrapper type and exports to at/torch.
paddle/phi/api/include/compat/c10/util/qint8.h	Adds placeholder c10::qint8 wrapper type and exports to at/torch.
paddle/phi/api/include/compat/c10/util/qint32.h	Adds placeholder c10::qint32 wrapper type and exports to at/torch.
paddle/phi/api/include/compat/c10/util/Float8_e8m0fnu.h	Adds placeholder c10::Float8_e8m0fnu wrapper type and exports to at/torch.
paddle/phi/api/include/compat/c10/util/Float8_e5m2fnuz.h	Adds placeholder c10::Float8_e5m2fnuz wrapper type and exports to at/torch.
paddle/phi/api/include/compat/c10/util/Float8_e4m3fnuz.h	Adds placeholder c10::Float8_e4m3fnuz wrapper type and exports to at/torch.
paddle/phi/api/include/compat/c10/util/Float4_e2m1fn_x2.h	Adds placeholder c10::Float4_e2m1fn_x2 wrapper type and exports to at/torch.
paddle/phi/api/include/compat/c10/util/bits.h	Adds placeholder c10::bits* wrapper types and exports to at/torch.
paddle/phi/api/include/compat/c10/core/ScalarType.h	Updates scalar-type macro lists and derives toString/elementSize/type mappings from the expanded macro coverage.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[youge325]

/re-run all-failed

[ShigureNyako]

这版把 ScalarType 的枚举覆盖、toString/elementSize 宏化整理到一起，方向是对的；我也看了当前唯一失败的 IXUCA，日志里是 Paddle-iluvatar patch apply 失败，看起来和这次 ScalarType 改动没有直接关系。

不过我这边先不 approve，主要是 public C++ surface 还有两处没完全对齐 upstream：

1. quantized wrapper 目前缺少 upstream 的 using underlying = ...，下游如果走量化 dispatch，source level 还是会断。
2. Float4_e2m1fn_x2 的公开形状也和 upstream 不一致，至少字段名与比较运算还没补齐。

这两个点都属于兼容层的“半迁移”问题。建议先把我行内评论里的差异补齐，再继续推进；否则会出现 ScalarType/类型名已经暴露出来了，但真实 C++ API 还不能按 PyTorch 方式使用的情况。

[ShigureNyako]

这里建议把 quantized wrapper 的 public surface 再向 upstream 对齐一下。PyTorch 的 torch/headeronly/util/qint8.h / qint32.h / quint8.h / quint4x2.h / quint2x4.h 都定义了 using underlying = ...，而且 AT_DISPATCH_CASE_QINT 这类宏会直接依赖 typename scalar_t::underlying。

当前实现只有 val_，下游如果拿这些类型走 quantized dispatch，source level 还是会再次卡住。建议至少把 underlying 别名补上；如果想完全贴近 upstream，alignas(...) 也最好一起保持一致。

[ShigureNyako]

这个占位类型最好也直接复刻 upstream 的最小 public surface。PyTorch 的 torch/headeronly/util/Float4_e2m1fn_x2.h 使用的是 uint8_t val_，并提供了 operator==/!=。

这里改成 x 且没有比较运算，会让依赖这些公开成员/比较接口的代码继续不兼容。既然这个 PR 的目标是 C++ API compatibility，建议至少把字段名和比较运算补齐到 upstream 的形状。