[clang] Support fixed point types in C++

[PiJoules]

This initially just adds support for mangling.

[github-actions]

⚠️ C/C++ code formatter, clang-format found issues in your code. ⚠️

You can test this locally with the following command:

git-clang-format --diff 2f3b7d33f421b723728262a6978041d93f1f8c5c d92d6d2ba64d9dffad0b5f3aa6f784f8fa2076e6 -- clang/test/CodeGenCXX/fixed-point-mangle.cpp clang/lib/AST/ItaniumMangle.cpp clang/lib/Parse/ParseExpr.cpp clang/lib/Parse/ParseExprCXX.cpp clang/lib/Parse/ParseTentative.cpp clang/test/Frontend/fixed_point_errors.cpp

View the diff from clang-format here.

diff --git a/clang/lib/AST/ItaniumMangle.cpp b/clang/lib/AST/ItaniumMangle.cpp
index 830a35b98..99d6caef4 100644
--- a/clang/lib/AST/ItaniumMangle.cpp
+++ b/clang/lib/AST/ItaniumMangle.cpp
@@ -3044,7 +3044,8 @@ void CXXNameMangler::mangleType(const BuiltinType *T) {
   // UNSUPPORTED:    ::= De # IEEE 754r decimal floating point (128 bits)
   // UNSUPPORTED:    ::= Df # IEEE 754r decimal floating point (32 bits)
   //                 ::= Dh # IEEE 754r half-precision floating point (16 bits)
-  //                 ::= DF <number> _ # ISO/IEC TS 18661 binary floating point type _FloatN (N bits);
+  //                 ::= DF <number> _ # ISO/IEC TS 18661 binary floating point
+  //                 type _FloatN (N bits);
   //                 ::= Di # char32_t
   //                 ::= Ds # char16_t
   //                 ::= Dn # std::nullptr_t (i.e., decltype(nullptr))

[rjmccall]

This patch is primarily adding support for fixed-point types to C++, which seems like it ought to be in the PR title. With that said, I don't have any objection to either the goal or the implementation.

[AaronBallman]

This is introducing an extension from C into C++ -- was there an RFC or something discussed about whether we wish to support this in C++? This breaks code in system headers entirely unrelated to fixed point:

#include <functional>
#include <iostream>

int main(int argc, char* argv[]) {
    std:: plus<int> x;
    int a = x((int)30, (int)5);
    int b = 35;
    std::cout << a << "vs" << b << std::endl;
}

when compiled against MSVC 2022 headers in C++23 mode gives:

F:\source\llvm-project>llvm\out\build\x64-Debug\bin\clang.exe -std=c++23 -fsyntax-only "C:\Users\aballman\OneDrive - Intel Corporation\Desktop\test.cpp"
In file included from C:\Users\aballman\OneDrive - Intel Corporation\Desktop\test.cpp:1:
In file included from C:\Program Files\Microsoft Visual Studio\2022\Professional\VC\Tools\MSVC\14.37.32822\include\functional:19:
In file included from C:\Program Files\Microsoft Visual Studio\2022\Professional\VC\Tools\MSVC\14.37.32822\include\unordered_map:11:
In file included from C:\Program Files\Microsoft Visual Studio\2022\Professional\VC\Tools\MSVC\14.37.32822\include\xhash:14:
C:\Program Files\Microsoft Visual Studio\2022\Professional\VC\Tools\MSVC\14.37.32822\include\vector:3263:16: error:
      compile with '-ffixed-point' to enable fixed point types
 3263 |         _Vbase _Accum = 0;
      |                ^
C:\Program Files\Microsoft Visual Studio\2022\Professional\VC\Tools\MSVC\14.37.32822\include\vector:3263:23: error:
      expected unqualified-id
 3263 |         _Vbase _Accum = 0;
      |                       ^
C:\Program Files\Microsoft Visual Studio\2022\Professional\VC\Tools\MSVC\14.37.32822\include\vector:3268:13: error:
      compile with '-ffixed-point' to enable fixed point types
 3268 |             _Accum |= _Tmp ? _Mask : _Vbase{0};
      |             ^
C:\Program Files\Microsoft Visual Studio\2022\Professional\VC\Tools\MSVC\14.37.32822\include\vector:3268:20: error:
      expected unqualified-id
 3268 |             _Accum |= _Tmp ? _Mask : _Vbase{0};
      |                    ^
C:\Program Files\Microsoft Visual Studio\2022\Professional\VC\Tools\MSVC\14.37.32822\include\vector:3270:53: error:
      expected '(' for function-style cast or type construction
 3270 |                 this->_Emplace_back_unchecked(_Accum);
      |                                               ~~~~~~^
C:\Program Files\Microsoft Visual Studio\2022\Professional\VC\Tools\MSVC\14.37.32822\include\vector:3271:17: error:
      compile with '-ffixed-point' to enable fixed point types
 3271 |                 _Accum = 0;
      |                 ^
C:\Program Files\Microsoft Visual Studio\2022\Professional\VC\Tools\MSVC\14.37.32822\include\vector:3271:24: error:
      expected unqualified-id
 3271 |                 _Accum = 0;
      |                        ^
C:\Program Files\Microsoft Visual Studio\2022\Professional\VC\Tools\MSVC\14.37.32822\include\vector:3277:49: error:
      expected '(' for function-style cast or type construction
 3277 |             this->_Emplace_back_unchecked(_Accum);
      |                                           ~~~~~~^
C:\Program Files\Microsoft Visual Studio\2022\Professional\VC\Tools\MSVC\14.37.32822\include\vector:3286:16: error:
      compile with '-ffixed-point' to enable fixed point types
 3286 |         _Vbase _Accum    = 0;
      |                ^
C:\Program Files\Microsoft Visual Studio\2022\Professional\VC\Tools\MSVC\14.37.32822\include\vector:3286:26: error:
      expected unqualified-id
 3286 |         _Vbase _Accum    = 0;
      |                          ^
C:\Program Files\Microsoft Visual Studio\2022\Professional\VC\Tools\MSVC\14.37.32822\include\vector:3291:13: error:
      compile with '-ffixed-point' to enable fixed point types
 3291 |             _Accum |= _Tmp ? _Mask : _Vbase{0};
      |             ^
C:\Program Files\Microsoft Visual Studio\2022\Professional\VC\Tools\MSVC\14.37.32822\include\vector:3291:20: error:
      expected unqualified-id
 3291 |             _Accum |= _Tmp ? _Mask : _Vbase{0};
      |                    ^
C:\Program Files\Microsoft Visual Studio\2022\Professional\VC\Tools\MSVC\14.37.32822\include\vector:3293:46: error:
      expected '(' for function-style cast or type construction
 3293 |                 this->_Myvec.push_back(_Accum);
      |                                        ~~~~~~^
C:\Program Files\Microsoft Visual Studio\2022\Professional\VC\Tools\MSVC\14.37.32822\include\vector:3294:17: error:
      compile with '-ffixed-point' to enable fixed point types
 3294 |                 _Accum = 0;
      |                 ^
C:\Program Files\Microsoft Visual Studio\2022\Professional\VC\Tools\MSVC\14.37.32822\include\vector:3294:24: error:
      expected unqualified-id
 3294 |                 _Accum = 0;
      |                        ^
C:\Program Files\Microsoft Visual Studio\2022\Professional\VC\Tools\MSVC\14.37.32822\include\vector:3300:42: error:
      expected '(' for function-style cast or type construction
 3300 |             this->_Myvec.push_back(_Accum);
      |                                    ~~~~~~^
16 errors generated.

because Microsoft is already using this reserved identifier as part of their implementation in <algorithm> and <vector>.

This needs to be reverted temporarily until this issue can be resolved.

[PiJoules]

This is introducing an extension from C into C++ -- was there an RFC or something discussed about whether we wish to support this in C++?

The original request for this I believe was on one of my earlier patches in 2018 for adding fixed point support in clang, but limiting it only to C because mangling for the types did not exist for the C++ ABI. Unfortunately it's been so long that phabricator doesn't seem to show this in my history. The earliest request for this was in itanium-cxx-abi/cxx-abi#56 and this was mentioned in a status update for fixed point types in https://lists.llvm.org/pipermail/llvm-dev/2021-March/149216.html. We have some users who would like to see fixed point support in C++ without having to go through helper classes.

I think the revert was appropriate since fixed point types should only be opt-in and gated by the -ffixed-point flag, but it looks like I accidentally just kept it unconditionally on for C++. This shouldn't affect the MSVC headers otherwise.