[AArch64] Implement __builtin_cpu_supports, compiler-rt tests. (#82378)

The patch complements https://github.com/llvm/llvm-project/pull/68919
and adds AArch64 support for builtin
`__builtin_cpu_supports("feature1+...+featureN")`
which return true if all specified CPU features in argument are
detected. Also compiler-rt aarch64 native run tests for features
detection mechanism were added and 'cpu_model' check was fixed after its
refactor merged https://github.com/llvm/llvm-project/pull/75635 Original
RFC was https://reviews.llvm.org/D153153

clang/lib/Basic/Targets/AArch64.cpp

@@ -667,7 +667,13 @@ StringRef AArch64TargetInfo::getFeatureDependencies(StringRef Name) const {
 }
 
 bool AArch64TargetInfo::validateCpuSupports(StringRef FeatureStr) const {
-  return llvm::AArch64::parseArchExtension(FeatureStr).has_value();
+  // CPU features might be separated by '+', extract them and check
+  llvm::SmallVector<StringRef, 8> Features;
+  FeatureStr.split(Features, "+");
+  for (auto &Feature : Features)
+    if (!llvm::AArch64::parseArchExtension(Feature.trim()).has_value())
+      return false;
+  return true;
 }
 
 bool AArch64TargetInfo::hasFeature(StringRef Feature) const {

clang/lib/Basic/Targets/AArch64.h

@@ -165,7 +165,7 @@ public:
                             DiagnosticsEngine &Diags) override;
   ParsedTargetAttr parseTargetAttr(StringRef Str) const override;
   bool supportsTargetAttributeTune() const override { return true; }
-
+  bool supportsCpuSupports() const override { return true; }
   bool checkArithmeticFenceSupported() const override { return true; }
 
   bool hasBFloat16Type() const override;

clang/lib/CodeGen/CGBuiltin.cpp

@@ -10638,6 +10638,9 @@ Value *CodeGenFunction::EmitAArch64BuiltinExpr(unsigned BuiltinID,
       BuiltinID <= clang::AArch64::LastSMEBuiltin)
     return EmitAArch64SMEBuiltinExpr(BuiltinID, E);
 
+  if (BuiltinID == Builtin::BI__builtin_cpu_supports)
+    return EmitAArch64CpuSupports(E);
+
   unsigned HintID = static_cast<unsigned>(-1);
   switch (BuiltinID) {
   default: break;
@@ -14025,6 +14028,19 @@ Value *CodeGenFunction::EmitX86CpuInit() {
   return Builder.CreateCall(Func);
 }
 
+Value *CodeGenFunction::EmitAArch64CpuSupports(const CallExpr *E) {
+  const Expr *ArgExpr = E->getArg(0)->IgnoreParenCasts();
+  StringRef ArgStr = cast<StringLiteral>(ArgExpr)->getString();
+  llvm::SmallVector<StringRef, 8> Features;
+  ArgStr.split(Features, "+");
+  for (auto &Feature : Features) {
+    Feature = Feature.trim();
+    if (Feature != "default")
+      Features.push_back(Feature);
+  }
+  return EmitAArch64CpuSupports(Features);
+}
+
 llvm::Value *
 CodeGenFunction::EmitAArch64CpuSupports(ArrayRef<StringRef> FeaturesStrs) {
   uint64_t FeaturesMask = llvm::AArch64::getCpuSupportsMask(FeaturesStrs);

clang/lib/CodeGen/CodeGenFunction.h

@@ -5013,10 +5013,10 @@ private:
   llvm::Value *EmitAArch64CpuInit();
   llvm::Value *
   FormAArch64ResolverCondition(const MultiVersionResolverOption &RO);
+  llvm::Value *EmitAArch64CpuSupports(const CallExpr *E);
   llvm::Value *EmitAArch64CpuSupports(ArrayRef<StringRef> FeatureStrs);
 };
 
-
 inline DominatingLLVMValue::saved_type
 DominatingLLVMValue::save(CodeGenFunction &CGF, llvm::Value *value) {
   if (!needsSaving(value)) return saved_type(value, false);

clang/test/CodeGen/aarch64-cpu-supports-target.c

@@ -0,0 +1,52 @@
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -emit-llvm -o - %s | FileCheck %s
+
+int check_all_feature() {
+  if (__builtin_cpu_supports("rng+flagm+flagm2+fp16fml+dotprod+sm4"))
+    return 1;
+  else if (__builtin_cpu_supports("rdm+lse+fp+simd+crc+sha1+sha2+sha3"))
+    return 2;
+  else if (__builtin_cpu_supports("aes+pmull+fp16+dit+dpb+dpb2+jscvt"))
+    return 3;
+  else if (__builtin_cpu_supports("fcma+rcpc+rcpc2+rcpc3+frintts+dgh"))
+    return 4;
+  else if (__builtin_cpu_supports("i8mm+bf16+ebf16+rpres+sve+sve-bf16"))
+    return 5;
+  else if (__builtin_cpu_supports("sve-ebf16+sve-i8mm+f32mm+f64mm"))
+    return 6;
+  else if (__builtin_cpu_supports("sve2+sve2-aes+sve2-pmull128"))
+    return 7;
+  else if (__builtin_cpu_supports("sve2-bitperm+sve2-sha3+sve2-sm4"))
+    return 8;
+  else if (__builtin_cpu_supports("sme+memtag+memtag2+memtag3+sb"))
+    return 9;
+  else if (__builtin_cpu_supports("predres+ssbs+ssbs2+bti+ls64+ls64_v"))
+    return 10;
+  else if (__builtin_cpu_supports("ls64_accdata+wfxt+sme-f64f64"))
+    return 11;
+  else if (__builtin_cpu_supports("sme-i16i64+sme2"))
+    return 12;
+  else
+    return 0;
+}
+
+// CHECK-LABEL: define dso_local i32 @neon_code() #1
+int __attribute__((target("simd"))) neon_code() { return 1; }
+
+// CHECK-LABEL: define dso_local i32 @sve_code() #2
+int __attribute__((target("sve"))) sve_code() { return 2; }
+
+// CHECK-LABEL: define dso_local i32 @code() #0
+int code() { return 3; }
+
+// CHECK-LABEL: define dso_local i32 @test_versions() #0
+int test_versions() {
+  if (__builtin_cpu_supports("sve"))
+    return sve_code();
+  else if (__builtin_cpu_supports("simd"))
+    return neon_code();
+  else
+    return code();
+}
+// CHECK: attributes #0 = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+// CHECK: attributes #1 = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+neon" }
+// CHECK: attributes #2 = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+fullfp16,+neon,+sve" }

clang/test/CodeGen/aarch64-cpu-supports.c

@@ -0,0 +1,54 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --check-globals --version 2
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -emit-llvm -o - %s | FileCheck %s
+
+// CHECK: @__aarch64_cpu_features = external dso_local global { i64 }
+// CHECK-LABEL: define dso_local i32 @main
+// CHECK-SAME: () #[[ATTR0:[0-9]+]] {
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
+// CHECK-NEXT:    store i32 0, ptr [[RETVAL]], align 4
+// CHECK-NEXT:    [[TMP0:%.*]] = load i64, ptr @__aarch64_cpu_features, align 8
+// CHECK-NEXT:    [[TMP1:%.*]] = and i64 [[TMP0]], 70368744177664
+// CHECK-NEXT:    [[TMP2:%.*]] = icmp eq i64 [[TMP1]], 70368744177664
+// CHECK-NEXT:    [[TMP3:%.*]] = and i1 true, [[TMP2]]
+// CHECK-NEXT:    br i1 [[TMP3]], label [[IF_THEN:%.*]], label [[IF_END:%.*]]
+// CHECK:       if.then:
+// CHECK-NEXT:    store i32 1, ptr [[RETVAL]], align 4
+// CHECK-NEXT:    br label [[RETURN:%.*]]
+// CHECK:       if.end:
+// CHECK-NEXT:    [[TMP4:%.*]] = load i64, ptr @__aarch64_cpu_features, align 8
+// CHECK-NEXT:    [[TMP5:%.*]] = and i64 [[TMP4]], 9070970929152
+// CHECK-NEXT:    [[TMP6:%.*]] = icmp eq i64 [[TMP5]], 9070970929152
+// CHECK-NEXT:    [[TMP7:%.*]] = and i1 true, [[TMP6]]
+// CHECK-NEXT:    br i1 [[TMP7]], label [[IF_THEN1:%.*]], label [[IF_END2:%.*]]
+// CHECK:       if.then1:
+// CHECK-NEXT:    store i32 2, ptr [[RETVAL]], align 4
+// CHECK-NEXT:    br label [[RETURN]]
+// CHECK:       if.end2:
+// CHECK-NEXT:    [[TMP8:%.*]] = load i64, ptr @__aarch64_cpu_features, align 8
+// CHECK-NEXT:    [[TMP9:%.*]] = and i64 [[TMP8]], 166633186212708352
+// CHECK-NEXT:    [[TMP10:%.*]] = icmp eq i64 [[TMP9]], 166633186212708352
+// CHECK-NEXT:    [[TMP11:%.*]] = and i1 true, [[TMP10]]
+// CHECK-NEXT:    br i1 [[TMP11]], label [[IF_THEN3:%.*]], label [[IF_END4:%.*]]
+// CHECK:       if.then3:
+// CHECK-NEXT:    store i32 3, ptr [[RETVAL]], align 4
+// CHECK-NEXT:    br label [[RETURN]]
+// CHECK:       if.end4:
+// CHECK-NEXT:    store i32 0, ptr [[RETVAL]], align 4
+// CHECK-NEXT:    br label [[RETURN]]
+// CHECK:       return:
+// CHECK-NEXT:    [[TMP12:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK-NEXT:    ret i32 [[TMP12]]
+//
+int main(void) {
+  if (__builtin_cpu_supports("sb"))
+    return 1;
+
+  if (__builtin_cpu_supports("sve2-pmull128+memtag"))
+    return 2;
+
+  if (__builtin_cpu_supports("sme2+ls64_v+wfxt"))
+    return 3;
+
+  return 0;
+}

clang/test/Preprocessor/has_builtin_cpuid.c

@@ -13,8 +13,3 @@
 #   error "ARM/PPC shouldn't have __builtin_cpu_init"
 # endif
 #endif
-#if __has_builtin(__builtin_cpu_supports)
-# ifdef ARM
-#   error "ARM shouldn't have __builtin_cpu_supports"
-# endif
-#endif

clang/test/Sema/aarch64-cpu-supports.c

@@ -0,0 +1,26 @@
+// RUN: %clang_cc1 -fsyntax-only -triple aarch64-linux-gnu -verify %s
+
+int test_aarch64_features(void) {
+  char * ssbs2;
+  // expected-error@+1 {{expression is not a string literal}}
+  if (__builtin_cpu_supports(ssbs2))
+    return 1;
+  // expected-error@+1 {{invalid cpu feature string}}
+  if (__builtin_cpu_supports(""))
+    return 2;
+  // expected-error@+1 {{invalid cpu feature string}}
+  if (__builtin_cpu_supports("pmull128"))
+    return 3;
+  // expected-error@+1 {{invalid cpu feature string}}
+  if (__builtin_cpu_supports("sve2,rpres"))
+    return 4;
+  // expected-error@+1 {{invalid cpu feature string}}
+  if (__builtin_cpu_supports("dgh+sve2-pmull"))
+    return 5;
+  // expected-error@+1 {{invalid cpu feature string}}
+  if (__builtin_cpu_supports("default"))
+    return 6;
+  if (__builtin_cpu_supports(" ssbs + bti "))
+    return 7;
+  return 0;
+}

clang/test/Sema/builtin-cpu-supports.c

@@ -27,7 +27,7 @@ int main(void) {
   (void)__builtin_cpu_supports("x86-64-v4");
   (void)__builtin_cpu_supports("x86-64-v5"); // expected-error {{invalid cpu feature string for builtin}}
 #else
-  if (__builtin_cpu_supports("aes")) // expected-error {{builtin is not supported on this target}}
+  if (__builtin_cpu_supports("neon")) // expected-error {{invalid cpu feature string for builtin}}
     a("vsx");
 
   if (__builtin_cpu_is("cortex-x3")) // expected-error {{builtin is not supported on this target}}

compiler-rt/test/builtins/Unit/aarch64_cpu_features_test.c

@@ -0,0 +1,17 @@
+// REQUIRES: aarch64-target-arch
+// REQUIRES: native-run
+// RUN: %clang_builtins %s %librt -o %t && %run %t
+// REQUIRES: librt_has_aarch64
+int main(void) {
+  if (__builtin_cpu_supports("fp+simd+pmull+sha2+crc")) {
+    if (__builtin_cpu_supports("fp") && __builtin_cpu_supports("simd") &&
+        __builtin_cpu_supports("pmull") && __builtin_cpu_supports("sha2") &&
+        __builtin_cpu_supports("crc")) {
+      return 0;
+    } else {
+      // Something wrong in feature detection
+      return 1;
+    }
+  }
+  return 0;
+}

compiler-rt/test/builtins/Unit/cpu_model_test.c

@@ -1,6 +1,6 @@
 // REQUIRES: x86-target-arch
 // RUN: %clang_builtins %s %librt -o %t && %run %t
-// REQUIRES: librt_has_cpu_model
+// REQUIRES: librt_has_x86
 
 // FIXME: XFAIL the test because it is expected to return non-zero value.
 // XFAIL: *