[C++17] Support __GCC_[CON|DE]STRUCTIVE_SIZE (#89446)

These macros are used by STL implementations to support implementation
of std::hardware_destructive_interference_size and
std::hardware_constructive_interference_size

Fixes #60174

---------

Co-authored-by: Louis Dionne <[email protected]>

Author: AaronBallman

clang-tools-extra/test/pp-trace/pp-trace-pragma-general.cpp

@@ -21,6 +21,12 @@ void foo() {
 
 // CHECK: ---
 // CHECK-NEXT: - Callback: PragmaDirective
+// CHECK-NEXT:   Loc: "<built-in>:{{.+}}:1"
+// CHECK-NEXT:   Introducer: PIK_HashPragma
+// CHECK-NEXT: - Callback: PragmaDirective
+// CHECK-NEXT:   Loc: "<built-in>:{{.+}}:1"
+// CHECK-NEXT:   Introducer: PIK_HashPragma
+// CHECK-NEXT: - Callback: PragmaDirective
 // CHECK-NEXT:   Loc: "{{.*}}{{[/\\]}}pp-trace-pragma-general.cpp:3:1"
 // CHECK-NEXT:   Introducer: PIK_HashPragma
 // CHECK-NEXT: - Callback: PragmaDiagnosticPush

clang-tools-extra/test/pp-trace/pp-trace-pragma-ms.cpp

@@ -18,6 +18,12 @@
 
 // CHECK: ---
 // CHECK-NEXT: - Callback: PragmaDirective
+// CHECK-NEXT:   Loc: "<built-in>:{{.+}}:1"
+// CHECK-NEXT:   Introducer: PIK_HashPragma
+// CHECK-NEXT: - Callback: PragmaDirective
+// CHECK-NEXT:   Loc: "<built-in>:{{.+}}:1"
+// CHECK-NEXT:   Introducer: PIK_HashPragma
+// CHECK-NEXT: - Callback: PragmaDirective
 // CHECK-NEXT:   Loc: "{{.*}}{{[/\\]}}pp-trace-pragma-ms.cpp:3:1"
 // CHECK-NEXT:   Introducer: PIK_HashPragma
 // CHECK-NEXT: - Callback: PragmaComment
@@ -67,7 +73,7 @@
 // CHECK-NEXT:   Introducer: PIK_HashPragma
 // CHECK-NEXT: - Callback: PragmaMessage
 // CHECK-NEXT:   Loc: "{{.*}}{{[/\\]}}pp-trace-pragma-ms.cpp:13:9"
-// CHECK-NEXT:   Namespace: 
+// CHECK-NEXT:   Namespace:
 // CHECK-NEXT:   Kind: PMK_Message
 // CHECK-NEXT:   Str: message argument
 // CHECK-NEXT: - Callback: PragmaDirective

clang-tools-extra/test/pp-trace/pp-trace-pragma-opencl.cpp

@@ -6,6 +6,12 @@
 
 // CHECK: ---
 // CHECK-NEXT: - Callback: PragmaDirective
+// CHECK-NEXT:   Loc: "<built-in>:{{.+}}:1"
+// CHECK-NEXT:   Introducer: PIK_HashPragma
+// CHECK-NEXT: - Callback: PragmaDirective
+// CHECK-NEXT:   Loc: "<built-in>:{{.+}}:1"
+// CHECK-NEXT:   Introducer: PIK_HashPragma
+// CHECK-NEXT: - Callback: PragmaDirective
 // CHECK-NEXT:   Loc: "{{.*}}{{[/\\]}}pp-trace-pragma-opencl.cpp:3:1"
 // CHECK-NEXT:   Introducer: PIK_HashPragma
 // CHECK-NEXT: - Callback: PragmaOpenCLExtension

clang/docs/LanguageExtensions.rst

@@ -5572,3 +5572,25 @@ but the expression has no runtime effects.
 Type- and value-dependent expressions are not supported yet.
 
 This facility is designed to aid with testing name lookup machinery.
+
+Predefined Macros
+=================
+
+`__GCC_DESTRUCTIVE_SIZE` and `__GCC_CONSTRUCTIVE_SIZE`
+------------------------------------------------------
+Specify the mimum offset between two objects to avoid false sharing and the
+maximum size of contiguous memory to promote true sharing, respectively. These
+macros are predefined in all C and C++ language modes, but can be redefined on
+the command line with ``-D`` to specify different values as needed or can be
+undefined on the command line with ``-U`` to disable support for the feature.
+
+**Note: the values the macros expand to are not guaranteed to be stable. They
+are are affected by architectures and CPU tuning flags, can change between
+releases of Clang and will not match the values defined by other compilers such
+as GCC.**
+
+Compiling different TUs depending on these flags (including use of
+``std::hardware_constructive_interference`` or
+``std::hardware_destructive_interference``)  with different compilers, macro
+definitions, or architecture flags will lead to ODR violations and should be
+avoided.

clang/docs/ReleaseNotes.rst

@@ -95,6 +95,18 @@ C++14 Feature Support
 - Sized deallocation is enabled by default in C++14 onwards. The user may specify
   ``-fno-sized-deallocation`` to disable it if there are some regressions.
 
+C++17 Feature Support
+^^^^^^^^^^^^^^^^^^^^^
+- Clang now exposes ``__GCC_DESTRUCTIVE_SIZE`` and ``__GCC_CONSTRUCTIVE_SIZE``
+  predefined macros to support standard library implementations of
+  ``std::hardware_destructive_interference_size`` and
+  ``std::hardware_constructive_interference_size``, respectively. These macros
+  are predefined in all C and C++ language modes. The values the macros
+  expand to are not stable between releases of Clang and do not need to match
+  the values produced by GCC, so these macros should not be used from header
+  files because they may not be stable across multiple TUs (the values may vary
+  based on compiler version as well as CPU tuning). #GH60174
+
 C++20 Feature Support
 ^^^^^^^^^^^^^^^^^^^^^
 

clang/include/clang/Basic/TargetInfo.h

@@ -40,6 +40,7 @@
 #include <cassert>
 #include <optional>
 #include <string>
+#include <utility>
 #include <vector>
 
 namespace llvm {
@@ -1792,6 +1793,15 @@ class TargetInfo : public TransferrableTargetInfo,
   /// Whether to support HIP image/texture API's.
   virtual bool hasHIPImageSupport() const { return true; }
 
+  /// The first value in the pair is the minimum offset between two objects to
+  /// avoid false sharing (destructive interference). The second value in the
+  /// pair is maximum size of contiguous memory to promote true sharing
+  /// (constructive interference). Neither of these values are considered part
+  /// of the ABI and can be changed by targets at any time.
+  virtual std::pair<unsigned, unsigned> hardwareInterferenceSizes() const {
+    return std::make_pair(64, 64);
+  }
+
 protected:
   /// Copy type and layout related info.
   void copyAuxTarget(const TargetInfo *Aux);

clang/lib/Basic/Targets/ARM.h

@@ -225,6 +225,10 @@ class LLVM_LIBRARY_VISIBILITY ARMTargetInfo : public TargetInfo {
   bool hasBitIntType() const override { return true; }
 
   const char *getBFloat16Mangling() const override { return "u6__bf16"; };
+
+  std::pair<unsigned, unsigned> hardwareInterferenceSizes() const override {
+    return std::make_pair(getTriple().isArch64Bit() ? 256 : 64, 64);
+  }
 };
 
 class LLVM_LIBRARY_VISIBILITY ARMleTargetInfo : public ARMTargetInfo {

clang/lib/Basic/Targets/AVR.h

@@ -175,6 +175,10 @@ class LLVM_LIBRARY_VISIBILITY AVRTargetInfo : public TargetInfo {
   std::optional<std::string> handleAsmEscapedChar(char EscChar) const override;
   StringRef getABI() const override { return ABI; }
 
+  std::pair<unsigned, unsigned> hardwareInterferenceSizes() const override {
+    return std::make_pair(32, 32);
+  }
+
 protected:
   std::string CPU;
   StringRef ABI;

clang/lib/Basic/Targets/BPF.h

@@ -113,6 +113,10 @@ class LLVM_LIBRARY_VISIBILITY BPFTargetInfo : public TargetInfo {
     StringRef CPUName(Name);
     return isValidCPUName(CPUName);
   }
+
+  std::pair<unsigned, unsigned> hardwareInterferenceSizes() const override {
+    return std::make_pair(32, 32);
+  }
 };
 } // namespace targets
 } // namespace clang

clang/lib/Basic/Targets/M68k.h

@@ -56,6 +56,10 @@ class LLVM_LIBRARY_VISIBILITY M68kTargetInfo : public TargetInfo {
   BuiltinVaListKind getBuiltinVaListKind() const override;
   bool setCPU(const std::string &Name) override;
   CallingConvCheckResult checkCallingConvention(CallingConv CC) const override;
+
+  std::pair<unsigned, unsigned> hardwareInterferenceSizes() const override {
+    return std::make_pair(32, 32);
+  }
 };
 
 } // namespace targets

clang/lib/Basic/Targets/Mips.h

@@ -431,6 +431,10 @@ class LLVM_LIBRARY_VISIBILITY MipsTargetInfo : public TargetInfo {
 
   bool validateTarget(DiagnosticsEngine &Diags) const override;
   bool hasBitIntType() const override { return true; }
+
+  std::pair<unsigned, unsigned> hardwareInterferenceSizes() const override {
+    return std::make_pair(32, 32);
+  }
 };
 } // namespace targets
 } // namespace clang

clang/lib/Basic/Targets/PPC.h

@@ -423,6 +423,10 @@ class LLVM_LIBRARY_VISIBILITY PPC32TargetInfo : public PPCTargetInfo {
     // This is the ELF definition
     return TargetInfo::PowerABIBuiltinVaList;
   }
+
+  std::pair<unsigned, unsigned> hardwareInterferenceSizes() const override {
+    return std::make_pair(32, 32);
+  }
 };
 
 // Note: ABI differences may eventually require us to have a separate
@@ -503,6 +507,10 @@ class LLVM_LIBRARY_VISIBILITY PPC64TargetInfo : public PPCTargetInfo {
       return CCCR_Warning;
     }
   }
+
+  std::pair<unsigned, unsigned> hardwareInterferenceSizes() const override {
+    return std::make_pair(128, 128);
+  }
 };
 
 class LLVM_LIBRARY_VISIBILITY AIXPPC32TargetInfo :

clang/lib/Basic/Targets/RISCV.h

@@ -122,6 +122,10 @@ class RISCVTargetInfo : public TargetInfo {
   void fillValidTuneCPUList(SmallVectorImpl<StringRef> &Values) const override;
   bool supportsTargetAttributeTune() const override { return true; }
   ParsedTargetAttr parseTargetAttr(StringRef Str) const override;
+
+  std::pair<unsigned, unsigned> hardwareInterferenceSizes() const override {
+    return std::make_pair(32, 32);
+  }
 };
 class LLVM_LIBRARY_VISIBILITY RISCV32TargetInfo : public RISCVTargetInfo {
 public:

clang/lib/Basic/Targets/Sparc.h

@@ -140,6 +140,10 @@ class LLVM_LIBRARY_VISIBILITY SparcTargetInfo : public TargetInfo {
     CPU = getCPUKind(Name);
     return CPU != CK_GENERIC;
   }
+
+  std::pair<unsigned, unsigned> hardwareInterferenceSizes() const override {
+    return std::make_pair(32, 32);
+  }
 };
 
 // SPARC v8 is the 32-bit mode selected by Triple::sparc.

clang/lib/Basic/Targets/SystemZ.h

@@ -220,6 +220,10 @@ class LLVM_LIBRARY_VISIBILITY SystemZTargetInfo : public TargetInfo {
   int getEHDataRegisterNumber(unsigned RegNo) const override {
     return RegNo < 4 ? 6 + RegNo : -1;
   }
+
+  std::pair<unsigned, unsigned> hardwareInterferenceSizes() const override {
+    return std::make_pair(256, 256);
+  }
 };
 } // namespace targets
 } // namespace clang

clang/lib/Frontend/InitPreprocessor.cpp

@@ -1308,6 +1308,16 @@ static void InitializePredefinedMacros(const TargetInfo &TI,
     Builder.defineMacro("__GCC_ATOMIC_TEST_AND_SET_TRUEVAL", "1");
   }
 
+  // GCC defines these macros in both C and C++ modes despite them being needed
+  // mostly for STL implementations in C++.
+  auto [Destructive, Constructive] = TI.hardwareInterferenceSizes();
+  Builder.defineMacro("__GCC_DESTRUCTIVE_SIZE", Twine(Destructive));
+  Builder.defineMacro("__GCC_CONSTRUCTIVE_SIZE", Twine(Constructive));
+  // We need to use push_macro to allow users to redefine these macros from the
+  // command line with -D and not issue a -Wmacro-redefined warning.
+  Builder.append("#pragma push_macro(\"__GCC_DESTRUCTIVE_SIZE\")");
+  Builder.append("#pragma push_macro(\"__GCC_CONSTRUCTIVE_SIZE\")");
+
   auto addLockFreeMacros = [&](const llvm::Twine &Prefix) {
     // Used by libc++ and libstdc++ to implement ATOMIC_<foo>_LOCK_FREE.
 #define DEFINE_LOCK_FREE_MACRO(TYPE, Type)                                     \

clang/test/AST/ast-dump-macro-json.c

@@ -132,7 +132,7 @@ void BLAP(foo, __COUNTER__)(void);
 // CHECK-NEXT:   "spellingLoc": {
 // CHECK-NEXT:    "offset": {{[0-9]+}},
 // CHECK-NEXT:    "file": "<scratch space>",
-// CHECK-NEXT:    "line": 3,
+// CHECK-NEXT:    "line": 5,
 // CHECK-NEXT:    "col": 1,
 // CHECK-NEXT:    "tokLen": 4
 // CHECK-NEXT:   },
@@ -169,7 +169,7 @@ void BLAP(foo, __COUNTER__)(void);
 // CHECK-NEXT:   "spellingLoc": {
 // CHECK-NEXT:    "offset": {{[0-9]+}},
 // CHECK-NEXT:    "file": "<scratch space>",
-// CHECK-NEXT:    "line": 5,
+// CHECK-NEXT:    "line": 7,
 // CHECK-NEXT:    "col": 1,
 // CHECK-NEXT:    "tokLen": 4
 // CHECK-NEXT:   },

clang/test/Analysis/html_diagnostics/relevant_lines/multifile.c

@@ -11,4 +11,4 @@ int f(int coin) {
 // RUN: rm -rf %t.output
 // RUN: %clang_analyze_cc1 -analyze -analyzer-checker=core -analyzer-output html -o %t.output %s
 // RUN: cat %t.output/* | FileCheck %s --match-full-lines
-// CHECK: var relevant_lines = {"1": {"3": 1, "4": 1, "5": 1, "6": 1}, "3": {"3": 1, "4": 1, "5": 1, "6": 1, "7": 1}};
+// CHECK: var relevant_lines = {"1": {"3": 1, "4": 1, "5": 1, "6": 1}, "4": {"3": 1, "4": 1, "5": 1, "6": 1, "7": 1}};

clang/test/Lexer/update_consecutive_macro_address_space.c

@@ -1,14 +1,15 @@
 // RUN: %clang -cc1 -print-stats %s 2>&1 | FileCheck %s
-// CHECK: 6 local SLocEntries allocated
+// CHECK: 7 local SLocEntries allocated
 //
-// Verify that the macro arg expansion is split to two file ids, we have 6 file
-// ids rather than 5:
+// Verify that the macro arg expansion is split to two file ids, we have 7 file
+// ids rather than 6:
 //   0: invalid file id
 //   1: main file
 //   2: builtin file
-//   3: macro expansion for X
-//   4: macro arg expansions for 1
-//   5: macro arg expansions for == 2
+//   3: scratch space for __GCC_[CON|DE]STRUCTIVE_SIZE macros
+//   4: macro expansion for X
+//   5: macro arg expansions for 1
+//   6: macro arg expansions for == 2
 #define X(x) (int)(x);
 void func() {
   X(1

clang/test/Preprocessor/hardware_interference.cpp

@@ -0,0 +1,17 @@
+// RUN: %clang_cc1 -E -dM -D__GCC_CONSTRUCTIVE_SIZE=1000 -D__GCC_DESTRUCTIVE_SIZE=1001 %s -verify -Weverything | FileCheck %s
+// RUN: %clang_cc1 -D__GCC_CONSTRUCTIVE_SIZE=1000 -D__GCC_DESTRUCTIVE_SIZE=1001 %s -verify -Weverything
+// RUN: %clang_cc1 -E -dM -U__GCC_CONSTRUCTIVE_SIZE -U__GCC_DESTRUCTIVE_SIZE %s -verify -Weverything | FileCheck --check-prefix DISABLED %s
+// expected-no-diagnostics
+
+// Validate that we can set a new value on the command line without issuing any
+// diagnostics and that we can disabled the macro on the command line without
+// issuing any diagnostics.
+
+// CHECK: #define __GCC_CONSTRUCTIVE_SIZE 1000
+// CHECK: #define __GCC_DESTRUCTIVE_SIZE 1001
+// DISABLED-NOT: __GCC_CONSTRUCTIVE_SIZE
+// DISABLED-NOT: __GCC_DESTRUCTIVE_SIZE
+
+int main() {
+  return 0;
+}

clang/test/Preprocessor/init-aarch64.c

@@ -119,6 +119,8 @@
 // AARCH64-NEXT: #define __FP_FAST_FMA 1
 // AARCH64-NEXT: #define __FP_FAST_FMAF 1
 // AARCH64-NEXT: #define __GCC_ASM_FLAG_OUTPUTS__ 1
+// AARCH64-NEXT: #define __GCC_CONSTRUCTIVE_SIZE {{.+}}
+// AARCH64-NEXT: #define __GCC_DESTRUCTIVE_SIZE {{.+}}
 // AARCH64-NEXT: #define __GCC_HAVE_SYNC_COMPARE_AND_SWAP_1 1
 // AARCH64-NEXT: #define __GCC_HAVE_SYNC_COMPARE_AND_SWAP_16 1
 // AARCH64-NEXT: #define __GCC_HAVE_SYNC_COMPARE_AND_SWAP_2 1
@@ -220,11 +222,11 @@
 // AARCH64-NEXT: #define __LONG_MAX__ 9223372036854775807L
 // AARCH64-NEXT: #define __LONG_WIDTH__ 64
 // AARCH64-NEXT: #define __LP64__ 1
-// AARCH64-NEXT: #define __MEMORY_SCOPE_DEVICE 1 
-// AARCH64-NEXT: #define __MEMORY_SCOPE_SINGLE 4 
-// AARCH64-NEXT: #define __MEMORY_SCOPE_SYSTEM 0 
-// AARCH64-NEXT: #define __MEMORY_SCOPE_WRKGRP 2 
-// AARCH64-NEXT: #define __MEMORY_SCOPE_WVFRNT 3 
+// AARCH64-NEXT: #define __MEMORY_SCOPE_DEVICE 1
+// AARCH64-NEXT: #define __MEMORY_SCOPE_SINGLE 4
+// AARCH64-NEXT: #define __MEMORY_SCOPE_SYSTEM 0
+// AARCH64-NEXT: #define __MEMORY_SCOPE_WRKGRP 2
+// AARCH64-NEXT: #define __MEMORY_SCOPE_WVFRNT 3
 // AARCH64-NEXT: #define __NO_INLINE__ 1
 // AARCH64-NEXT: #define __NO_MATH_ERRNO__ 1
 // AARCH64-NEXT: #define __OBJC_BOOL_IS_BOOL 0