Frontend: sink vendor definitions from Basic to Frontend

[compnerd]

The vendor specific macro definitions are based on the vendor rather than the target specific information. This ensures that __APPLE__ is always defined for *-apple-* targets. Take the opportunity to do likewise with the __AMD__ vendor macro.

[compnerd]

CC: @kubamracek

[llvmbot]

@llvm/pr-subscribers-backend-amdgpu

@llvm/pr-subscribers-clang

Author: Saleem Abdulrasool (compnerd)

Changes

The vendor specific macro definitions are based on the vendor rather than the target specific information. This ensures that __APPLE__ is always defined for *-apple-* targets. Take the opportunity to do likewise with the __AMD__ vendor macro.

---

Full diff: https://github.com/llvm/llvm-project/pull/80364.diff

3 Files Affected:

• (modified) clang/lib/Basic/Targets/AMDGPU.cpp (-1)
• (modified) clang/lib/Basic/Targets/OSTargets.cpp (-1)
• (modified) clang/lib/Frontend/InitPreprocessor.cpp (+5)

diff --git a/clang/lib/Basic/Targets/AMDGPU.cpp b/clang/lib/Basic/Targets/AMDGPU.cpp
index 141501e8a4d9a..c6d68f591a5a8 100644
--- a/clang/lib/Basic/Targets/AMDGPU.cpp
+++ b/clang/lib/Basic/Targets/AMDGPU.cpp
@@ -266,7 +266,6 @@ ArrayRef<Builtin::Info> AMDGPUTargetInfo::getTargetBuiltins() const {
 
 void AMDGPUTargetInfo::getTargetDefines(const LangOptions &Opts,
                                         MacroBuilder &Builder) const {
-  Builder.defineMacro("__AMD__");
   Builder.defineMacro("__AMDGPU__");
 
   if (isAMDGCN(getTriple()))
diff --git a/clang/lib/Basic/Targets/OSTargets.cpp b/clang/lib/Basic/Targets/OSTargets.cpp
index 899aefa6173ac..64b4514ba55bb 100644
--- a/clang/lib/Basic/Targets/OSTargets.cpp
+++ b/clang/lib/Basic/Targets/OSTargets.cpp
@@ -23,7 +23,6 @@ void getDarwinDefines(MacroBuilder &Builder, const LangOptions &Opts,
                       const llvm::Triple &Triple, StringRef &PlatformName,
                       VersionTuple &PlatformMinVersion) {
   Builder.defineMacro("__APPLE_CC__", "6000");
-  Builder.defineMacro("__APPLE__");
   Builder.defineMacro("__STDC_NO_THREADS__");
 
   // AddressSanitizer doesn't play well with source fortification, which is on
diff --git a/clang/lib/Frontend/InitPreprocessor.cpp b/clang/lib/Frontend/InitPreprocessor.cpp
index 877e205e2e9bf..27621ee332419 100644
--- a/clang/lib/Frontend/InitPreprocessor.cpp
+++ b/clang/lib/Frontend/InitPreprocessor.cpp
@@ -804,6 +804,11 @@ static void InitializePredefinedMacros(const TargetInfo &TI,
     }
   }
 
+  if (TI.getTriple().getVendor() == Triple::AMD)
+    Builder.defineMacro("__AMD__");
+  if (TI.getTriple().getVendor() == Triple::Apple)
+    Builder.defineMacro("__APPLE__");
+
   // Define macros for the C11 / C++11 memory orderings
   Builder.defineMacro("__ATOMIC_RELAXED", "0");
   Builder.defineMacro("__ATOMIC_CONSUME", "1");

[jroelofs]

I think this is a behavior change for non-darwin triples, but only in a "tree falls in the forest" kind of way.

@fpetrogalli you were talking about this a while back. I don't remember the verdict.

[compnerd]

Yes, and something that I think that is desirable in the context of Swift as well (where we are looking at some of the embedded system usage where the target triple may be something like arm64-apple-none-macho).

[kubamracek]

👍