Automerge: [clang][SME] Emit error for OpenMP captured regions in SME functions (#124590)

Currently, these generate incorrect code, as streaming/SME attributes
are not propagated to the outlined function. As we've yet to work on
mixing OpenMP and streaming functions (and determine how they should
interact with OpenMP's runtime), we think it is best to disallow this
for now.

Author: MacDue

clang/include/clang/AST/Decl.h

@@ -5139,6 +5139,12 @@ static constexpr StringRef getOpenMPVariantManglingSeparatorStr() {
 bool IsArmStreamingFunction(const FunctionDecl *FD,
                             bool IncludeLocallyStreaming);
 
+/// Returns whether the given FunctionDecl has Arm ZA state.
+bool hasArmZAState(const FunctionDecl *FD);
+
+/// Returns whether the given FunctionDecl has Arm ZT0 state.
+bool hasArmZT0State(const FunctionDecl *FD);
+
 } // namespace clang
 
 #endif // LLVM_CLANG_AST_DECL_H

clang/include/clang/Basic/DiagnosticSemaKinds.td

@@ -3870,6 +3870,9 @@ def err_sme_definition_using_za_in_non_sme_target : Error<
   "function using ZA state requires 'sme'">;
 def err_sme_definition_using_zt0_in_non_sme2_target : Error<
   "function using ZT0 state requires 'sme2'">;
+def err_sme_openmp_captured_region : Error<
+  "OpenMP captured regions are not yet supported in "
+  "%select{streaming functions|functions with ZA state|functions with ZT0 state}0">;
 def warn_sme_streaming_pass_return_vl_to_non_streaming : Warning<
   "%select{returning|passing}0 a VL-dependent argument %select{from|to}0 a function with a different"
   " streaming-mode is undefined behaviour when the streaming and non-streaming vector lengths are different at runtime">,

clang/lib/AST/Decl.cpp

@@ -5845,3 +5845,17 @@ bool clang::IsArmStreamingFunction(const FunctionDecl *FD,
 
   return false;
 }
+
+bool clang::hasArmZAState(const FunctionDecl *FD) {
+  const auto *T = FD->getType()->getAs<FunctionProtoType>();
+  return (T && FunctionType::getArmZAState(T->getAArch64SMEAttributes()) !=
+                   FunctionType::ARM_None) ||
+         (FD->hasAttr<ArmNewAttr>() && FD->getAttr<ArmNewAttr>()->isNewZA());
+}
+
+bool clang::hasArmZT0State(const FunctionDecl *FD) {
+  const auto *T = FD->getType()->getAs<FunctionProtoType>();
+  return (T && FunctionType::getArmZT0State(T->getAArch64SMEAttributes()) !=
+                   FunctionType::ARM_None) ||
+         (FD->hasAttr<ArmNewAttr>() && FD->getAttr<ArmNewAttr>()->isNewZT0());
+}

clang/lib/Sema/SemaARM.cpp

@@ -624,20 +624,6 @@ static bool checkArmStreamingBuiltin(Sema &S, CallExpr *TheCall,
   return true;
 }
 
-static bool hasArmZAState(const FunctionDecl *FD) {
-  const auto *T = FD->getType()->getAs<FunctionProtoType>();
-  return (T && FunctionType::getArmZAState(T->getAArch64SMEAttributes()) !=
-                   FunctionType::ARM_None) ||
-         (FD->hasAttr<ArmNewAttr>() && FD->getAttr<ArmNewAttr>()->isNewZA());
-}
-
-static bool hasArmZT0State(const FunctionDecl *FD) {
-  const auto *T = FD->getType()->getAs<FunctionProtoType>();
-  return (T && FunctionType::getArmZT0State(T->getAArch64SMEAttributes()) !=
-                   FunctionType::ARM_None) ||
-         (FD->hasAttr<ArmNewAttr>() && FD->getAttr<ArmNewAttr>()->isNewZT0());
-}
-
 static ArmSMEState getSMEState(unsigned BuiltinID) {
   switch (BuiltinID) {
   default:

clang/lib/Sema/SemaStmt.cpp

@@ -4568,9 +4568,27 @@ buildCapturedStmtCaptureList(Sema &S, CapturedRegionScopeInfo *RSI,
   return false;
 }
 
+static std::optional<int>
+isOpenMPCapturedRegionInArmSMEFunction(Sema const &S, CapturedRegionKind Kind) {
+  if (!S.getLangOpts().OpenMP || Kind != CR_OpenMP)
+    return {};
+  if (const FunctionDecl *FD = S.getCurFunctionDecl(/*AllowLambda=*/true)) {
+    if (IsArmStreamingFunction(FD, /*IncludeLocallyStreaming=*/true))
+      return /* in streaming functions */ 0;
+    if (hasArmZAState(FD))
+      return /* in functions with ZA state */ 1;
+    if (hasArmZT0State(FD))
+      return /* in fuctions with ZT0 state */ 2;
+  }
+  return {};
+}
+
 void Sema::ActOnCapturedRegionStart(SourceLocation Loc, Scope *CurScope,
                                     CapturedRegionKind Kind,
                                     unsigned NumParams) {
+  if (auto ErrorIndex = isOpenMPCapturedRegionInArmSMEFunction(*this, Kind))
+    Diag(Loc, diag::err_sme_openmp_captured_region) << *ErrorIndex;
+
   CapturedDecl *CD = nullptr;
   RecordDecl *RD = CreateCapturedStmtRecordDecl(CD, Loc, NumParams);
 
@@ -4602,6 +4620,9 @@ void Sema::ActOnCapturedRegionStart(SourceLocation Loc, Scope *CurScope,
                                     CapturedRegionKind Kind,
                                     ArrayRef<CapturedParamNameType> Params,
                                     unsigned OpenMPCaptureLevel) {
+  if (auto ErrorIndex = isOpenMPCapturedRegionInArmSMEFunction(*this, Kind))
+    Diag(Loc, diag::err_sme_openmp_captured_region) << *ErrorIndex;
+
   CapturedDecl *CD = nullptr;
   RecordDecl *RD = CreateCapturedStmtRecordDecl(CD, Loc, Params.size());
 

clang/test/Sema/aarch64-sme-attrs-openmp-captured-region.c

@@ -0,0 +1,81 @@
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme -target-feature +sme2 -fopenmp -fsyntax-only -verify %s
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme -target-feature +sme2 -fopenmp -fsyntax-only -verify=expected-cpp -x c++ %s
+
+int compute(int);
+
+void streaming_openmp_captured_region(int * out) __arm_streaming {
+  // expected-error@+2 {{OpenMP captured regions are not yet supported in streaming functions}}
+  // expected-cpp-error@+1 {{OpenMP captured regions are not yet supported in streaming functions}}
+  #pragma omp parallel for num_threads(32)
+  for (int ci = 0; ci < 8; ci++) {
+    out[ci] = compute(ci);
+  }
+}
+
+__arm_locally_streaming void locally_streaming_openmp_captured_region(int * out) {
+  // expected-error@+2 {{OpenMP captured regions are not yet supported in streaming functions}}
+  // expected-cpp-error@+1 {{OpenMP captured regions are not yet supported in streaming functions}}
+  #pragma omp parallel for num_threads(32)
+  for (int ci = 0; ci < 8; ci++) {
+    out[ci] = compute(ci);
+  }
+}
+
+void za_state_captured_region(int * out) __arm_inout("za") {
+  // expected-error@+2 {{OpenMP captured regions are not yet supported in functions with ZA state}}
+  // expected-cpp-error@+1 {{OpenMP captured regions are not yet supported in functions with ZA state}}
+  #pragma omp parallel for num_threads(32)
+  for (int ci = 0; ci < 8; ci++) {
+    out[ci] = compute(ci);
+  }
+}
+
+__arm_new("za") void new_za_state_captured_region(int * out) {
+  // expected-error@+2 {{OpenMP captured regions are not yet supported in functions with ZA state}}
+  // expected-cpp-error@+1 {{OpenMP captured regions are not yet supported in functions with ZA state}}
+  #pragma omp parallel for num_threads(32)
+  for (int ci = 0; ci < 8; ci++) {
+    out[ci] = compute(ci);
+  }
+}
+
+void zt0_state_openmp_captured_region(int * out) __arm_inout("zt0") {
+  // expected-error@+2 {{OpenMP captured regions are not yet supported in functions with ZT0 state}}
+  // expected-cpp-error@+1 {{OpenMP captured regions are not yet supported in functions with ZT0 state}}
+  #pragma omp parallel for num_threads(32)
+  for (int ci = 0; ci < 8; ci++) {
+    out[ci] = compute(ci);
+  }
+}
+
+__arm_new("zt0") void new_zt0_state_openmp_captured_region(int * out) {
+  // expected-error@+2 {{OpenMP captured regions are not yet supported in functions with ZT0 state}}
+  // expected-cpp-error@+1 {{OpenMP captured regions are not yet supported in functions with ZT0 state}}
+  #pragma omp parallel for num_threads(32)
+  for (int ci = 0; ci < 8; ci++) {
+    out[ci] = compute(ci);
+  }
+}
+
+/// OpenMP directives that don't create a captured region are okay:
+
+void streaming_function_openmp(int * out) __arm_streaming __arm_inout("za", "zt0") {
+  #pragma omp unroll full
+  for (int ci = 0; ci < 8; ci++) {
+    out[ci] = compute(ci);
+  }
+}
+
+__arm_locally_streaming void locally_streaming_openmp(int * out) __arm_inout("za", "zt0") {
+  #pragma omp unroll full
+  for (int ci = 0; ci < 8; ci++) {
+    out[ci] = compute(ci);
+  }
+}
+
+__arm_new("za", "zt0") void arm_new_openmp(int * out) {
+  #pragma omp unroll full
+  for (int ci = 0; ci < 8; ci++) {
+    out[ci] = compute(ci);
+  }
+}