- Add warnings to clang for unsupported cases where unwinding is not possible

  because of streaming-mode changes without SVE available.

- Fixed incorrect labels in sme-vg-to-stack.ll

Author: kmclaughlin-arm

clang/include/clang/Basic/DiagnosticGroups.td

@@ -1412,7 +1412,7 @@ def MultiGPU: DiagGroup<"multi-gpu">;
 // libc and the CRT to be skipped.
 def AVRRtlibLinkingQuirks : DiagGroup<"avr-rtlib-linking-quirks">;
 
-// A warning group related to AArch64 SME function attribues.
+// A warning group related to AArch64 SME function attributes.
 def AArch64SMEAttributes : DiagGroup<"aarch64-sme-attributes">;
 
 // A warning group for things that will change semantics in the future.

clang/include/clang/Basic/DiagnosticSemaKinds.td

@@ -3730,6 +3730,12 @@ def warn_gnu_inline_cplusplus_without_extern : Warning<
   "'gnu_inline' attribute without 'extern' in C++ treated as externally"
   " available, this changed in Clang 10">,
   InGroup<DiagGroup<"gnu-inline-cpp-without-extern">>;
+def warn_sme_streaming_mode_change_no_sve : Warning<
+  "function requires a streaming-mode change, unwinding is not possible without 'sve'">,
+  InGroup<AArch64SMEAttributes>;
+def warn_sme_locally_streaming_no_sve : Warning<
+  "unwinding is not possible for locally-streaming functions without 'sve'">,
+   InGroup<AArch64SMEAttributes>;
 def err_attribute_vecreturn_only_vector_member : Error<
   "the vecreturn attribute can only be used on a class or structure with one member, which must be a vector">;
 def err_attribute_vecreturn_only_pod_record : Error<

clang/lib/Sema/SemaChecking.cpp

@@ -3505,6 +3505,9 @@ bool Sema::ParseSVEImmChecks(
 }
 
 static ArmStreamingType getArmStreamingFnType(const FunctionDecl *FD) {
+  const QualType QT = FD->getType();
+  if (QT.isNull())
+    return ArmNonStreaming;
   if (FD->hasAttr<ArmLocallyStreamingAttr>())
     return ArmStreaming;
   if (const Type *Ty = FD->getType().getTypePtrOrNull()) {
@@ -7978,8 +7981,11 @@ void Sema::checkCall(NamedDecl *FDecl, const FunctionProtoType *Proto,
     // If the callee has an AArch64 SME attribute to indicate that it is an
     // __arm_streaming function, then the caller requires SME to be available.
     FunctionProtoType::ExtProtoInfo ExtInfo = Proto->getExtProtoInfo();
-    if (ExtInfo.AArch64SMEAttributes & FunctionType::SME_PStateSMEnabledMask) {
-      if (auto *CallerFD = dyn_cast<FunctionDecl>(CurContext)) {
+    auto *CallerFD = dyn_cast<FunctionDecl>(CurContext);
+    bool IsCalleeStreaming =
+        ExtInfo.AArch64SMEAttributes & FunctionType::SME_PStateSMEnabledMask;
+    if (IsCalleeStreaming) {
+      if (CallerFD) {
         llvm::StringMap<bool> CallerFeatureMap;
         Context.getFunctionFeatureMap(CallerFeatureMap, CallerFD);
         if (!CallerFeatureMap.contains("sme"))
@@ -7989,18 +7995,28 @@ void Sema::checkCall(NamedDecl *FDecl, const FunctionProtoType *Proto,
       }
     }
 
-    // If the call requires a streaming-mode change and has scalable vector
-    // arguments or return values, then warn the user that the streaming and
-    // non-streaming vector lengths may be different.
-    const auto *CallerFD = dyn_cast<FunctionDecl>(CurContext);
-    if (CallerFD && (!FD || !FD->getBuiltinID()) && AnyScalableArgsOrRet) {
-      bool IsCalleeStreaming =
-          ExtInfo.AArch64SMEAttributes & FunctionType::SME_PStateSMEnabledMask;
+    if (CallerFD && (!FD || !FD->getBuiltinID())) {
       bool IsCalleeStreamingCompatible =
           ExtInfo.AArch64SMEAttributes &
           FunctionType::SME_PStateSMCompatibleMask;
       ArmStreamingType CallerFnType = getArmStreamingFnType(CallerFD);
-      if (!IsCalleeStreamingCompatible &&
+
+      // SME functions may require SVE to be available for unwinding, as the
+      // value of VG needs to be preserved across streaming-mode changes.
+      if (!Context.getTargetInfo().hasFeature("sve")) {
+        if (CallerFD->hasAttr<ArmLocallyStreamingAttr>())
+          Diag(Loc, diag::warn_sme_locally_streaming_no_sve);
+
+          if ((CallerFnType == ArmStreaming ||
+               CallerFnType == ArmStreamingCompatible) &&
+              (!IsCalleeStreaming && !IsCalleeStreamingCompatible))
+            Diag(Loc, diag::warn_sme_streaming_mode_change_no_sve);
+      }
+
+      // If the call requires a streaming-mode change and has scalable vector
+      // arguments or return values, then warn the user that the streaming and
+      // non-streaming vector lengths may be different.
+      if (AnyScalableArgsOrRet && !IsCalleeStreamingCompatible &&
           (CallerFnType == ArmStreamingCompatible ||
            ((CallerFnType == ArmStreaming) ^ IsCalleeStreaming)))
         Diag(Loc, diag::warn_sme_streaming_pass_return_vl_to_non_streaming);

clang/test/CodeGen/aarch64-sme-inline-streaming-attrs.c

@@ -1,7 +1,7 @@
-// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -S -o /dev/null -target-feature +sme -verify -DTEST_NONE %s
-// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -S -o /dev/null -target-feature +sme -verify -DTEST_COMPATIBLE %s
-// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -S -o /dev/null -target-feature +sme -verify -DTEST_STREAMING %s
-// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -S -o /dev/null -target-feature +sme -verify -DTEST_LOCALLY %s
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -S -o /dev/null -target-feature +sve -target-feature +sme -verify -DTEST_NONE %s
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -S -o /dev/null -target-feature +sve -target-feature +sme -verify -DTEST_COMPATIBLE %s
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -S -o /dev/null -target-feature +sve -target-feature +sme -verify -DTEST_STREAMING %s
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -S -o /dev/null -target-feature +sve -target-feature +sme -verify -DTEST_LOCALLY %s
 
 #define __ai __attribute__((always_inline))
 __ai void inlined_fn(void) {}

clang/test/CodeGen/aarch64-sme-intrinsics/aarch64-sme-attrs.cpp

@@ -1,5 +1,5 @@
-// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme \
-// RUN:   -S -disable-O0-optnone -Werror -emit-llvm -o - %s \
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sve \
+// RUN:   -target-feature +sme -S -disable-O0-optnone -Werror -emit-llvm -o - %s \
 // RUN: | opt -S -passes=mem2reg \
 // RUN: | opt -S -passes=inline \
 // RUN: | FileCheck %s
@@ -278,18 +278,18 @@ int test_variadic_template() __arm_inout("za") {
               preserves_za_decl);
 }
 
-// CHECK: attributes #[[SM_ENABLED]] = { mustprogress noinline nounwind "aarch64_pstate_sm_enabled" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+sme" }
-// CHECK: attributes #[[NORMAL_DECL]] = { "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+sme" }
-// CHECK: attributes #[[SM_ENABLED_DECL]] = { "aarch64_pstate_sm_enabled" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+sme" }
-// CHECK: attributes #[[SM_COMPATIBLE]] = { mustprogress noinline nounwind "aarch64_pstate_sm_compatible" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+sme" }
-// CHECK: attributes #[[SM_COMPATIBLE_DECL]] = { "aarch64_pstate_sm_compatible" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+sme" }
-// CHECK: attributes #[[SM_BODY]] = { mustprogress noinline nounwind "aarch64_pstate_sm_body" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+sme" }
-// CHECK: attributes #[[ZA_SHARED]] = { mustprogress noinline nounwind "aarch64_inout_za" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+sme" }
-// CHECK: attributes #[[ZA_SHARED_DECL]] = { "aarch64_inout_za" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+sme" }
-// CHECK: attributes #[[ZA_PRESERVED]] = { mustprogress noinline nounwind "aarch64_preserves_za" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+sme" }
-// CHECK: attributes #[[ZA_PRESERVED_DECL]] = { "aarch64_preserves_za" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+sme" }
-// CHECK: attributes #[[ZA_NEW]] = { mustprogress noinline nounwind "aarch64_new_za" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+sme" }
-// CHECK: attributes #[[NORMAL_DEF]] = { mustprogress noinline nounwind "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+sme" }
+// CHECK: attributes #[[SM_ENABLED]] = { mustprogress noinline nounwind vscale_range(1,16) "aarch64_pstate_sm_enabled" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+fp-armv8,+fullfp16,+neon,+sme,+sve" }
+// CHECK: attributes #[[NORMAL_DECL]] = { "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+fp-armv8,+fullfp16,+neon,+sme,+sve" }
+// CHECK: attributes #[[SM_ENABLED_DECL]] = { "aarch64_pstate_sm_enabled" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+fp-armv8,+fullfp16,+neon,+sme,+sve" }
+// CHECK: attributes #[[SM_COMPATIBLE]] = { mustprogress noinline nounwind vscale_range(1,16) "aarch64_pstate_sm_compatible" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+fp-armv8,+fullfp16,+neon,+sme,+sve" }
+// CHECK: attributes #[[SM_COMPATIBLE_DECL]] = { "aarch64_pstate_sm_compatible" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+fp-armv8,+fullfp16,+neon,+sme,+sve" }
+// CHECK: attributes #[[SM_BODY]] = { mustprogress noinline nounwind vscale_range(1,16) "aarch64_pstate_sm_body" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+fp-armv8,+fullfp16,+neon,+sme,+sve" }
+// CHECK: attributes #[[ZA_SHARED]] = { mustprogress noinline nounwind vscale_range(1,16) "aarch64_inout_za" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+fp-armv8,+fullfp16,+neon,+sme,+sve" }
+// CHECK: attributes #[[ZA_SHARED_DECL]] = { "aarch64_inout_za" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+fp-armv8,+fullfp16,+neon,+sme,+sve" }
+// CHECK: attributes #[[ZA_PRESERVED]] = { mustprogress noinline nounwind vscale_range(1,16) "aarch64_preserves_za" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+fp-armv8,+fullfp16,+neon,+sme,+sve" }
+// CHECK: attributes #[[ZA_PRESERVED_DECL]] = { "aarch64_preserves_za" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+fp-armv8,+fullfp16,+neon,+sme,+sve" }
+// CHECK: attributes #[[ZA_NEW]] = { mustprogress noinline nounwind vscale_range(1,16) "aarch64_new_za" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+fp-armv8,+fullfp16,+neon,+sme,+sve" }
+// CHECK: attributes #[[NORMAL_DEF]] = { mustprogress noinline nounwind vscale_range(1,16) "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bf16,+fp-armv8,+fullfp16,+neon,+sme,+sve" }
 // CHECK: attributes #[[SM_ENABLED_CALL]] = { "aarch64_pstate_sm_enabled" }
 // CHECK: attributes #[[SM_COMPATIBLE_CALL]] = { "aarch64_pstate_sm_compatible" }
 // CHECK: attributes #[[SM_BODY_CALL]] = { "aarch64_pstate_sm_body" }

clang/test/Sema/aarch64-sme-func-attrs-without-target-feature.cpp

@@ -38,7 +38,7 @@ void non_streaming_def(void (*streaming_fn_ptr)(void) __arm_streaming,
 void streaming_compatible_def2(void (*streaming_fn_ptr)(void) __arm_streaming,
                                void (*streaming_compatible_fn_ptr)(void) __arm_streaming_compatible)
                                 __arm_streaming_compatible {
-  non_streaming_decl(); // OK
+  non_streaming_decl(); // expected-warning {{function requires a streaming-mode change, unwinding is not possible without 'sve'}}
   streaming_compatible_decl(); // OK
   streaming_compatible_fn_ptr(); // OK
   streaming_decl(); // expected-error {{call to a streaming function requires 'sme'}}

clang/test/Sema/aarch64-streaming-mode-changes-no-sve.c

@@ -0,0 +1,45 @@
+// RUN: %clang_cc1  -triple aarch64-none-linux-gnu -target-feature +sme \
+// RUN:  -target-feature -sve -Waarch64-sme-attributes -fsyntax-only -verify %s
+
+// REQUIRES: aarch64-registered-target
+
+#include "arm_sme.h"
+
+int non_streaming_decl(void);
+int streaming_decl(void) __arm_streaming;
+int streaming_compatible_decl(void) __arm_streaming_compatible;
+
+// Streaming-mode changes which would require spilling VG, unsupported without SVE
+
+int streaming_caller_no_sve(void) __arm_streaming {
+  // expected-warning@+1 {{function requires a streaming-mode change, unwinding is not possible without 'sve'}}
+  return non_streaming_decl();
+}
+
+int sc_caller_non_streaming_callee(void) __arm_streaming_compatible {
+  // expected-warning@+1 {{function requires a streaming-mode change, unwinding is not possible without 'sve'}}
+  return non_streaming_decl();
+}
+
+__arm_locally_streaming int locally_streaming_no_sve(void) {
+  // expected-warning@+1 {{unwinding is not possible for locally-streaming functions without 'sve'}}
+  return streaming_decl();
+}
+
+// No warnings expected
+
+int normal_caller_streaming_callee(void) {
+  return streaming_decl();
+}
+
+int normal_caller_streaming_compatible_callee(void) {
+  return streaming_compatible_decl();
+}
+
+int sc_caller_streaming_callee(void) __arm_streaming_compatible {
+  return streaming_decl();
+}
+
+int sc_caller_sc_callee(void) __arm_streaming_compatible {
+  return streaming_compatible_decl();
+}

llvm/test/CodeGen/AArch64/sme-vg-to-stack.ll

@@ -1030,15 +1030,15 @@ define void @streaming_compatible_no_sve() #4 {
 ; NO-SVE-CHECK-NEXT:    .cfi_offset b15, -96
 ; NO-SVE-CHECK-NEXT:    bl __arm_sme_state
 ; NO-SVE-CHECK-NEXT:    and x19, x0, #0x1
-; NO-SVE-CHECK-NEXT:    tbnz w19, #0, .LBB6_2
+; NO-SVE-CHECK-NEXT:    tbnz w19, #0, .LBB8_2
 ; NO-SVE-CHECK-NEXT:  // %bb.1:
 ; NO-SVE-CHECK-NEXT:    smstart sm
-; NO-SVE-CHECK-NEXT:  .LBB6_2:
+; NO-SVE-CHECK-NEXT:  .LBB8_2:
 ; NO-SVE-CHECK-NEXT:    bl streaming_callee
-; NO-SVE-CHECK-NEXT:    tbnz w19, #0, .LBB6_4
+; NO-SVE-CHECK-NEXT:    tbnz w19, #0, .LBB8_4
 ; NO-SVE-CHECK-NEXT:  // %bb.3:
 ; NO-SVE-CHECK-NEXT:    smstop sm
-; NO-SVE-CHECK-NEXT:  .LBB6_4:
+; NO-SVE-CHECK-NEXT:  .LBB8_4:
 ; NO-SVE-CHECK-NEXT:    .cfi_def_cfa wsp, 96
 ; NO-SVE-CHECK-NEXT:    ldp x29, x30, [sp, #64] // 16-byte Folded Reload
 ; NO-SVE-CHECK-NEXT:    ldr x19, [sp, #80] // 8-byte Folded Reload