[Clang]Throw frontend error for target feature mismatch when using flatten attribute

clang/docs/ReleaseNotes.rst

@@ -256,6 +256,9 @@ Bug Fixes in This Version
   cast chain. (#GH149967).
 - Fixed a crash with incompatible pointer to integer conversions in designated
   initializers involving string literals. (#GH154046)
+- Clang now emits a frontend error when a function marked with the `flatten` attribute
+  calls another function that requires target features not enabled in the caller. This
+  prevents a fatal error in the backend.
 - Fixed scope of typedefs present inside a template class. (#GH91451)
 
 Bug Fixes to Compiler Builtins

clang/include/clang/Basic/DiagnosticFrontendKinds.td

@@ -287,6 +287,10 @@ def err_function_needs_feature : Error<
   "always_inline function %1 requires target feature '%2', but would "
   "be inlined into function %0 that is compiled without support for '%2'">;
 
+def err_flatten_function_needs_feature : Error<
+  "flatten function %0 calls %1 which requires target feature '%2', but the "
+  "caller is compiled without support for '%2'">;
+
 let CategoryName = "Codegen ABI Check" in {
 def err_function_always_inline_attribute_mismatch : Error<
   "always_inline function %1 and its caller %0 have mismatching %2 attributes">;

clang/lib/CodeGen/CGCall.cpp

@@ -5261,9 +5261,12 @@ RValue CodeGenFunction::EmitCall(const CGFunctionInfo &CallInfo,
     // since otherwise we could be making a conditional call after a check for
     // the proper cpu features (and it won't cause code generation issues due to
     // function based code generation).
-    if (TargetDecl->hasAttr<AlwaysInlineAttr>() &&
-        (TargetDecl->hasAttr<TargetAttr>() ||
-         (CurFuncDecl && CurFuncDecl->hasAttr<TargetAttr>())))
+    if ((TargetDecl->hasAttr<AlwaysInlineAttr>() &&
+         (TargetDecl->hasAttr<TargetAttr>() ||
+          (CurFuncDecl && CurFuncDecl->hasAttr<TargetAttr>()))) ||
+        (CurFuncDecl && CurFuncDecl->hasAttr<FlattenAttr>() &&
+         (CurFuncDecl->hasAttr<TargetAttr>() ||
+          TargetDecl->hasAttr<TargetAttr>())))
       checkTargetFeatures(Loc, FD);
   }
 

clang/lib/CodeGen/CodeGenFunction.cpp

@@ -2827,6 +2827,9 @@ void CodeGenFunction::checkTargetFeatures(SourceLocation Loc,
   if (!FD)
     return;
 
+  bool IsAlwaysInline = TargetDecl->hasAttr<AlwaysInlineAttr>();
+  bool IsFlatten = FD && FD->hasAttr<FlattenAttr>();
+
   // Grab the required features for the call. For a builtin this is listed in
   // the td file with the default cpu, for an always_inline function this is any
   // listed cpu and any listed features.
@@ -2869,25 +2872,39 @@ void CodeGenFunction::checkTargetFeatures(SourceLocation Loc,
       if (F.getValue())
         ReqFeatures.push_back(F.getKey());
     }
-    if (!llvm::all_of(ReqFeatures, [&](StringRef Feature) {
-      if (!CallerFeatureMap.lookup(Feature)) {
-        MissingFeature = Feature.str();
-        return false;
-      }
-      return true;
-    }) && !IsHipStdPar)
-      CGM.getDiags().Report(Loc, diag::err_function_needs_feature)
-          << FD->getDeclName() << TargetDecl->getDeclName() << MissingFeature;
+    if (!llvm::all_of(ReqFeatures,
+                      [&](StringRef Feature) {
+                        if (!CallerFeatureMap.lookup(Feature)) {
+                          MissingFeature = Feature.str();
+                          return false;
+                        }
+                        return true;
+                      }) &&
+        !IsHipStdPar) {
+      if (IsAlwaysInline)
+        CGM.getDiags().Report(Loc, diag::err_function_needs_feature)
+            << FD->getDeclName() << TargetDecl->getDeclName() << MissingFeature;
+      else if (IsFlatten)
+        CGM.getDiags().Report(Loc, diag::err_flatten_function_needs_feature)
+            << FD->getDeclName() << TargetDecl->getDeclName() << MissingFeature;
+    }
+
   } else if (!FD->isMultiVersion() && FD->hasAttr<TargetAttr>()) {
     llvm::StringMap<bool> CalleeFeatureMap;
     CGM.getContext().getFunctionFeatureMap(CalleeFeatureMap, TargetDecl);
 
     for (const auto &F : CalleeFeatureMap) {
-      if (F.getValue() && (!CallerFeatureMap.lookup(F.getKey()) ||
-                           !CallerFeatureMap.find(F.getKey())->getValue()) &&
-          !IsHipStdPar)
-        CGM.getDiags().Report(Loc, diag::err_function_needs_feature)
-            << FD->getDeclName() << TargetDecl->getDeclName() << F.getKey();
+      if (F.getValue() &&
+          (!CallerFeatureMap.lookup(F.getKey()) ||
+           !CallerFeatureMap.find(F.getKey())->getValue()) &&
+          !IsHipStdPar) {
+        if (IsAlwaysInline)
+          CGM.getDiags().Report(Loc, diag::err_function_needs_feature)
+              << FD->getDeclName() << TargetDecl->getDeclName() << F.getKey();
+        else if (IsFlatten)
+          CGM.getDiags().Report(Loc, diag::err_flatten_function_needs_feature)
+              << FD->getDeclName() << TargetDecl->getDeclName() << F.getKey();
+      }
     }
   }
 }

clang/test/CodeGen/target-features-error-3.c

@@ -0,0 +1,12 @@
+// RUN: %clang_cc1 %s -triple=x86_64-linux-gnu -emit-llvm -verify -o /dev/null
+
+typedef double __v2df __attribute__((__vector_size__(16)));
+
+__v2df __attribute__((target("sse4.1"))) foo() {
+    __v2df v = {0.0, 0.0};
+    return __builtin_ia32_roundpd(v, 2);
+}
+
+__v2df __attribute__((flatten)) bar() {
+    return foo(); // expected-error {{flatten function 'bar' calls 'foo' which requires target feature 'sse4.1', but the caller is compiled without support for 'sse4.1'}}
+}

clang/test/CodeGen/target-features-error-4.c

@@ -0,0 +1,12 @@
+// RUN: %clang_cc1 %s -triple=x86_64-linux-gnu -emit-llvm -verify -o /dev/null
+
+typedef double __v2df __attribute__((__vector_size__(16)));
+
+__v2df __attribute__((target("sse4.1"))) foo() {
+    __v2df v = {0.0, 0.0};
+    return __builtin_ia32_roundpd(v, 2);
+}
+
+__v2df __attribute__((target("no-sse4.1"), flatten)) bar() {
+    return foo(); // expected-error {{flatten function 'bar' calls 'foo' which requires target feature 'sse4.1', but the caller is compiled without support for 'sse4.1'}}
+}

clang/test/CodeGen/target-features-error-5.c

@@ -0,0 +1,12 @@
+// RUN: %clang_cc1 %s -triple=x86_64-linux-gnu -target-feature +sse4.1 -emit-llvm -verify -o /dev/null
+
+typedef double __v2df __attribute__((__vector_size__(16)));
+
+__v2df foo() {
+    __v2df v = {0.0, 0.0};
+    return __builtin_ia32_roundpd(v, 2);
+}
+
+__v2df __attribute__((target("no-sse4.1"), flatten)) bar() {
+    return foo(); // expected-error {{flatten function 'bar' calls 'foo' which requires target feature 'sse4.1', but the caller is compiled without support for 'sse4.1'}}
+}

clang/test/CodeGen/target-features-no-error-2.c

@@ -0,0 +1,12 @@
+// RUN: %clang_cc1 %s -triple=x86_64-linux-gnu -emit-llvm -verify -o /dev/null
+
+typedef double __v2df __attribute__((__vector_size__(16)));
+
+__v2df __attribute__((target("sse4.1"))) foo() {
+    __v2df v = {0.0, 0.0};
+    return __builtin_ia32_roundpd(v, 2);
+}
+
+__v2df __attribute__((target("sse4.1"), flatten)) bar() {
+    return foo(); // expected-no-diagnostics
+}