[clang][FMV][AArch64] Improve streaming mode compatibility. (#100181)

* Allow arm-streaming if all the functions versions adhere to it.
* Allow arm-streaming-compatible if all the functions versions adhere to
it.

When the caller needs to toggle the streaming mode all the function
versions of the callee must adhere to the same mode, otherwise the call
will yield a runtime error.

Imagine the versions of the callee live in separate TUs. The version
that is visible to the caller will determine the calling convention used
when generating code for the callsite. Therefore we cannot support
mixing streaming with non-streaming function versions. Imagine TU1 has a
streaming caller and calls foo._sme which is streaming-compatible. The
codegen for the callsite will not switch off the streaming mode. Then in
TU2 we have a version which is non-streaming and could potentially be
called in streaming mode. Similarly if the caller is non-streaming and
the called version is streaming-compatible the codegen for the callsite
will not switch on the streaming mode, but other versions may be
streaming.

clang/include/clang/Basic/DiagnosticSemaKinds.td

@@ -3812,8 +3812,6 @@ def warn_sme_locally_streaming_has_vl_args_returns : Warning<
   InGroup<AArch64SMEAttributes>, DefaultIgnore;
 def err_conflicting_attributes_arm_state : Error<
   "conflicting attributes for state '%0'">;
-def err_sme_streaming_cannot_be_multiversioned : Error<
-  "streaming function cannot be multi-versioned">;
 def err_unknown_arm_state : Error<
   "unknown state '%0'">;
 def err_missing_arm_state : Error<

clang/lib/Sema/SemaDecl.cpp

@@ -11009,6 +11009,9 @@ static bool AttrCompatibleWithMultiVersion(attr::Kind Kind,
   switch (Kind) {
   default:
     return false;
+  case attr::ArmLocallyStreaming:
+    return MVKind == MultiVersionKind::TargetVersion ||
+           MVKind == MultiVersionKind::TargetClones;
   case attr::Used:
     return MVKind == MultiVersionKind::Target;
   case attr::NonNull:
@@ -11145,7 +11148,21 @@ bool Sema::areMultiversionVariantFunctionsCompatible(
     FunctionType::ExtInfo OldTypeInfo = OldType->getExtInfo();
     FunctionType::ExtInfo NewTypeInfo = NewType->getExtInfo();
 
-    if (OldTypeInfo.getCC() != NewTypeInfo.getCC())
+    const auto *OldFPT = OldFD->getType()->getAs<FunctionProtoType>();
+    const auto *NewFPT = NewFD->getType()->getAs<FunctionProtoType>();
+
+    bool ArmStreamingCCMismatched = false;
+    if (OldFPT && NewFPT) {
+      unsigned Diff =
+          OldFPT->getAArch64SMEAttributes() ^ NewFPT->getAArch64SMEAttributes();
+      // Arm-streaming, arm-streaming-compatible and non-streaming versions
+      // cannot be mixed.
+      if (Diff & (FunctionType::SME_PStateSMEnabledMask |
+                  FunctionType::SME_PStateSMCompatibleMask))
+        ArmStreamingCCMismatched = true;
+    }
+
+    if (OldTypeInfo.getCC() != NewTypeInfo.getCC() || ArmStreamingCCMismatched)
       return Diag(DiffDiagIDAt.first, DiffDiagIDAt.second) << CallingConv;
 
     QualType OldReturnType = OldType->getReturnType();
@@ -11165,9 +11182,8 @@ bool Sema::areMultiversionVariantFunctionsCompatible(
     if (!CLinkageMayDiffer && OldFD->isExternC() != NewFD->isExternC())
       return Diag(DiffDiagIDAt.first, DiffDiagIDAt.second) << LanguageLinkage;
 
-    if (CheckEquivalentExceptionSpec(
-            OldFD->getType()->getAs<FunctionProtoType>(), OldFD->getLocation(),
-            NewFD->getType()->getAs<FunctionProtoType>(), NewFD->getLocation()))
+    if (CheckEquivalentExceptionSpec(OldFPT, OldFD->getLocation(), NewFPT,
+                                     NewFD->getLocation()))
       return true;
   }
   return false;

clang/lib/Sema/SemaDeclAttr.cpp

@@ -3034,9 +3034,6 @@ bool Sema::checkTargetVersionAttr(SourceLocation LiteralLoc, Decl *D,
       return Diag(LiteralLoc, diag::warn_unsupported_target_attribute)
              << Unsupported << None << CurFeature << TargetVersion;
   }
-  if (IsArmStreamingFunction(cast<FunctionDecl>(D),
-                             /*IncludeLocallyStreaming=*/false))
-    return Diag(LiteralLoc, diag::err_sme_streaming_cannot_be_multiversioned);
   return false;
 }
 
@@ -3133,10 +3130,6 @@ bool Sema::checkTargetClonesAttrString(
           HasNotDefault = true;
         }
       }
-      if (IsArmStreamingFunction(cast<FunctionDecl>(D),
-                                 /*IncludeLocallyStreaming=*/false))
-        return Diag(LiteralLoc,
-                    diag::err_sme_streaming_cannot_be_multiversioned);
     } else {
       // Other targets ( currently X86 )
       if (Cur.starts_with("arch=")) {

clang/test/CodeGen/aarch64-fmv-streaming.c

@@ -0,0 +1,107 @@
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme -emit-llvm -o - %s | FileCheck %s
+
+
+// CHECK-LABEL: define {{[^@]+}}@n_callee._Msve
+// CHECK-SAME: () #[[ATTR0:[0-9]+]] {
+//
+// CHECK-LABEL: define {{[^@]+}}@n_callee._Msimd
+// CHECK-SAME: () #[[ATTR1:[0-9]+]] {
+//
+__arm_locally_streaming __attribute__((target_clones("sve", "simd"))) void n_callee(void) {}
+// CHECK-LABEL: define {{[^@]+}}@n_callee._Msme2
+// CHECK-SAME: () #[[ATTR2:[0-9]+]] {
+//
+__attribute__((target_version("sme2"))) void n_callee(void) {}
+// CHECK-LABEL: define {{[^@]+}}@n_callee.default
+// CHECK-SAME: () #[[ATTR3:[0-9]+]] {
+//
+__attribute__((target_version("default"))) void n_callee(void) {}
+
+
+// CHECK-LABEL: define {{[^@]+}}@s_callee._Msve
+// CHECK-SAME: () #[[ATTR4:[0-9]+]] {
+//
+// CHECK-LABEL: define {{[^@]+}}@s_callee._Msimd
+// CHECK-SAME: () #[[ATTR5:[0-9]+]] {
+//
+__attribute__((target_clones("sve", "simd"))) void s_callee(void) __arm_streaming {}
+// CHECK-LABEL: define {{[^@]+}}@s_callee._Msme2
+// CHECK-SAME: () #[[ATTR6:[0-9]+]] {
+//
+__arm_locally_streaming __attribute__((target_version("sme2"))) void s_callee(void) __arm_streaming {}
+// CHECK-LABEL: define {{[^@]+}}@s_callee.default
+// CHECK-SAME: () #[[ATTR7:[0-9]+]] {
+//
+__attribute__((target_version("default"))) void s_callee(void) __arm_streaming {}
+
+
+// CHECK-LABEL: define {{[^@]+}}@sc_callee._Msve
+// CHECK-SAME: () #[[ATTR8:[0-9]+]] {
+//
+// CHECK-LABEL: define {{[^@]+}}@sc_callee._Msimd
+// CHECK-SAME: () #[[ATTR9:[0-9]+]] {
+//
+__attribute__((target_clones("sve", "simd"))) void sc_callee(void) __arm_streaming_compatible {}
+// CHECK-LABEL: define {{[^@]+}}@sc_callee._Msme2
+// CHECK-SAME: () #[[ATTR10:[0-9]+]] {
+//
+__arm_locally_streaming __attribute__((target_version("sme2"))) void sc_callee(void) __arm_streaming_compatible {}
+// CHECK-LABEL: define {{[^@]+}}@sc_callee.default
+// CHECK-SAME: () #[[ATTR11:[0-9]+]] {
+//
+__attribute__((target_version("default"))) void sc_callee(void) __arm_streaming_compatible {}
+
+
+// CHECK-LABEL: define {{[^@]+}}@n_caller
+// CHECK-SAME: () #[[ATTR3:[0-9]+]] {
+// CHECK:    call void @n_callee()
+// CHECK:    call void @s_callee() #[[ATTR12:[0-9]+]]
+// CHECK:    call void @sc_callee() #[[ATTR13:[0-9]+]]
+//
+void n_caller(void) {
+  n_callee();
+  s_callee();
+  sc_callee();
+}
+
+
+// CHECK-LABEL: define {{[^@]+}}@s_caller
+// CHECK-SAME: () #[[ATTR7:[0-9]+]] {
+// CHECK:    call void @n_callee()
+// CHECK:    call void @s_callee() #[[ATTR12]]
+// CHECK:    call void @sc_callee() #[[ATTR13]]
+//
+void s_caller(void) __arm_streaming {
+  n_callee();
+  s_callee();
+  sc_callee();
+}
+
+
+// CHECK-LABEL: define {{[^@]+}}@sc_caller
+// CHECK-SAME: () #[[ATTR11:[0-9]+]] {
+// CHECK:    call void @n_callee()
+// CHECK:    call void @s_callee() #[[ATTR12]]
+// CHECK:    call void @sc_callee() #[[ATTR13]]
+//
+void sc_caller(void) __arm_streaming_compatible {
+  n_callee();
+  s_callee();
+  sc_callee();
+}
+
+
+// CHECK: attributes #[[ATTR0:[0-9]+]] = {{.*}} "aarch64_pstate_sm_body"
+// CHECK: attributes #[[ATTR1:[0-9]+]] = {{.*}} "aarch64_pstate_sm_body"
+// CHECK: attributes #[[ATTR2:[0-9]+]] = {{.*}}
+// CHECK: attributes #[[ATTR3]] = {{.*}}
+// CHECK: attributes #[[ATTR4:[0-9]+]] = {{.*}} "aarch64_pstate_sm_enabled"
+// CHECK: attributes #[[ATTR5:[0-9]+]] = {{.*}} "aarch64_pstate_sm_enabled"
+// CHECK: attributes #[[ATTR6:[0-9]+]] = {{.*}} "aarch64_pstate_sm_body" "aarch64_pstate_sm_enabled"
+// CHECK: attributes #[[ATTR7]] = {{.*}} "aarch64_pstate_sm_enabled"
+// CHECK: attributes #[[ATTR8:[0-9]+]] = {{.*}} "aarch64_pstate_sm_compatible"
+// CHECK: attributes #[[ATTR9:[0-9]+]] = {{.*}} "aarch64_pstate_sm_compatible"
+// CHECK: attributes #[[ATTR10]] = {{.*}} "aarch64_pstate_sm_body" "aarch64_pstate_sm_compatible"
+// CHECK: attributes #[[ATTR11]] = {{.*}} "aarch64_pstate_sm_compatible"
+// CHECK: attributes #[[ATTR12]] = {{.*}} "aarch64_pstate_sm_enabled"
+// CHECK: attributes #[[ATTR13]] = {{.*}} "aarch64_pstate_sm_compatible"

clang/test/Sema/aarch64-fmv-streaming.c

@@ -0,0 +1,46 @@
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme -Waarch64-sme-attributes -fsyntax-only -verify %s
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme -Waarch64-sme-attributes -fsyntax-only -verify=expected-cpp -x c++ %s
+
+__attribute__((target_clones("sve", "simd"))) void ok_arm_streaming(void) __arm_streaming {}
+__arm_locally_streaming __attribute__((target_version("sme2"))) void ok_arm_streaming(void) __arm_streaming {}
+__attribute__((target_version("default"))) void ok_arm_streaming(void) __arm_streaming {}
+
+__attribute__((target_clones("sve", "simd"))) void ok_arm_streaming_compatible(void) __arm_streaming_compatible {}
+__arm_locally_streaming __attribute__((target_version("sme2"))) void ok_arm_streaming_compatible(void) __arm_streaming_compatible {}
+__attribute__((target_version("default"))) void ok_arm_streaming_compatible(void) __arm_streaming_compatible {}
+
+__arm_locally_streaming __attribute__((target_clones("sve", "simd"))) void ok_no_streaming(void) {}
+__attribute__((target_version("sme2"))) void ok_no_streaming(void) {}
+__attribute__((target_version("default"))) void ok_no_streaming(void) {}
+
+__attribute__((target_clones("sve", "simd"))) void bad_mixed_streaming(void) {}
+// expected-cpp-error@+2 {{multiversioned function declaration has a different calling convention}}
+// expected-error@+1 {{multiversioned function declaration has a different calling convention}}
+__attribute__((target_version("sme2"))) void bad_mixed_streaming(void) __arm_streaming {}
+// expected-cpp-error@+2 {{multiversioned function declaration has a different calling convention}}
+// expected-error@+1 {{multiversioned function declaration has a different calling convention}}
+__attribute__((target_version("default"))) void bad_mixed_streaming(void) __arm_streaming_compatible {}
+// expected-cpp-error@+2 {{multiversioned function declaration has a different calling convention}}
+// expected-error@+1 {{multiversioned function declaration has a different calling convention}}
+__arm_locally_streaming __attribute__((target_version("dotprod"))) void bad_mixed_streaming(void) __arm_streaming {}
+
+void n_caller(void) {
+  ok_arm_streaming();
+  ok_arm_streaming_compatible();
+  ok_no_streaming();
+  bad_mixed_streaming();
+}
+
+void s_caller(void) __arm_streaming {
+  ok_arm_streaming();
+  ok_arm_streaming_compatible();
+  ok_no_streaming();
+  bad_mixed_streaming();
+}
+
+void sc_caller(void) __arm_streaming_compatible {
+  ok_arm_streaming();
+  ok_arm_streaming_compatible();
+  ok_no_streaming();
+  bad_mixed_streaming();
+}

clang/test/Sema/aarch64-sme-func-attrs.c

@@ -455,48 +455,6 @@ void unimplemented_spill_fill_za(void (*share_zt0_only)(void) __arm_inout("zt0")
   share_zt0_only();
 }
 
-// expected-cpp-error@+2 {{streaming function cannot be multi-versioned}}
-// expected-error@+1 {{streaming function cannot be multi-versioned}}
-__attribute__((target_version("sme2")))
-void cannot_work_version(void) __arm_streaming {}
-// expected-cpp-error@+5 {{function declared 'void ()' was previously declared 'void () __arm_streaming', which has different SME function attributes}}
-// expected-cpp-note@-2 {{previous declaration is here}}
-// expected-error@+3 {{function declared 'void (void)' was previously declared 'void (void) __arm_streaming', which has different SME function attributes}}
-// expected-note@-4 {{previous declaration is here}}
-__attribute__((target_version("default")))
-void cannot_work_version(void) {}
-
-
-// expected-cpp-error@+2 {{streaming function cannot be multi-versioned}}
-// expected-error@+1 {{streaming function cannot be multi-versioned}}
-__attribute__((target_clones("sme2")))
-void cannot_work_clones(void) __arm_streaming {}
-
-
-__attribute__((target("sme2")))
-void just_fine_streaming(void) __arm_streaming {}
-__attribute__((target_version("sme2")))
-void just_fine(void) { just_fine_streaming(); }
-__attribute__((target_version("default")))
-void just_fine(void) {}
-
-
-__arm_locally_streaming
-__attribute__((target_version("sme2")))
-void incompatible_locally_streaming(void) {}
-// expected-error@-1 {{attribute 'target_version' multiversioning cannot be combined with attribute '__arm_locally_streaming'}}
-// expected-cpp-error@-2 {{attribute 'target_version' multiversioning cannot be combined with attribute '__arm_locally_streaming'}}
-__attribute__((target_version("default")))
-void incompatible_locally_streaming(void) {}
-
-
-void fmv_caller() {
-    cannot_work_version();
-    cannot_work_clones();
-    just_fine();
-    incompatible_locally_streaming();
-}
-
 void sme_streaming_with_vl_arg(__SVInt8_t a) __arm_streaming { }
 
 __SVInt8_t sme_streaming_returns_vl(void) __arm_streaming { __SVInt8_t r; return r; }