[SYCL][NVPTX][AMDGCN] Move devicelib cmath to header

clang/include/clang/Basic/Attr.td

@@ -1590,6 +1590,13 @@ def SYCLDevice : InheritableAttr {
   let Documentation = [SYCLDeviceDocs];
 }
 
+def SYCLDeviceOnly : InheritableAttr {
+  let Spellings = [GNU<"sycl_device_only">];
+  let Subjects = SubjectList<[Function]>;
+  let LangOpts = [SYCLIsDevice, SilentlyIgnoreSYCLIsHost];
+  let Documentation = [Undocumented];
+}
+
 def SYCLGlobalVar : InheritableAttr {
   let Spellings = [GNU<"sycl_global_var">];
   let Subjects = SubjectList<[GlobalStorageNonLocalVar], ErrorDiag>;

clang/lib/AST/Decl.cpp

@@ -3729,6 +3729,11 @@ unsigned FunctionDecl::getBuiltinID(bool ConsiderWrapperFunctions) const {
       !(BuiltinID == Builtin::BIprintf || BuiltinID == Builtin::BImalloc))
     return 0;
 
+  if (Context.getLangOpts().isSYCL() && hasAttr<SYCLDeviceOnlyAttr>() &&
+      BuiltinID != Builtin::BIprintf) {
+    return 0;
+  }
+
   // As AMDGCN implementation of OpenMP does not have a device-side standard
   // library, none of the predefined library functions except printf and malloc
   // should be treated as a builtin i.e. 0 should be returned for them.

clang/lib/CodeGen/CGBuiltin.cpp

@@ -2544,10 +2544,7 @@ RValue CodeGenFunction::EmitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID,
       GenerateIntrinsics =
           ConstWithoutErrnoOrExceptions && ErrnoOverridenToFalseWithOpt;
   }
-  bool IsSYCLDeviceWithoutIntrinsics =
-      getLangOpts().SYCLIsDevice &&
-      (getTarget().getTriple().isNVPTX() || getTarget().getTriple().isAMDGCN());
-  if (GenerateIntrinsics && !IsSYCLDeviceWithoutIntrinsics) {
+  if (GenerateIntrinsics) {
     switch (BuiltinIDIfNoAsmLabel) {
     case Builtin::BIacos:
     case Builtin::BIacosf:
@@ -3647,7 +3644,7 @@ RValue CodeGenFunction::EmitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID,
   case Builtin::BI__builtin_modf:
   case Builtin::BI__builtin_modff:
   case Builtin::BI__builtin_modfl:
-    if (Builder.getIsFPConstrained() || IsSYCLDeviceWithoutIntrinsics)
+    if (Builder.getIsFPConstrained())
       break; // TODO: Emit constrained modf intrinsic once one exists.
     return RValue::get(emitModfBuiltin(*this, E, Intrinsic::modf));
   case Builtin::BI__builtin_isgreater:

clang/lib/Sema/SemaDeclAttr.cpp

@@ -7199,6 +7199,9 @@ ProcessDeclAttribute(Sema &S, Scope *scope, Decl *D, const ParsedAttr &AL,
   case ParsedAttr::AT_SYCLDevice:
     S.SYCL().handleSYCLDeviceAttr(D, AL);
     break;
+  case ParsedAttr::AT_SYCLDeviceOnly:
+    handleSimpleAttribute<SYCLDeviceOnlyAttr>(S, D, AL);
+    break;
   case ParsedAttr::AT_SYCLScope:
     S.SYCL().handleSYCLScopeAttr(D, AL);
     break;

clang/lib/Sema/SemaOverload.cpp

@@ -1629,6 +1629,14 @@ static bool IsOverloadOrOverrideImpl(Sema &SemaRef, FunctionDecl *New,
     }
   }
 
+  // Allow overloads with SYCLDeviceOnlyAttr
+  if (SemaRef.getLangOpts().isSYCL()) {
+    if (Old->hasAttr<SYCLDeviceOnlyAttr>() !=
+        New->hasAttr<SYCLDeviceOnlyAttr>()) {
+      return true;
+    }
+  }
+
   // The signatures match; this is not an overload.
   return false;
 }
@@ -10960,6 +10968,16 @@ bool clang::isBetterOverloadCandidate(
            S.CUDA().IdentifyPreference(Caller, Cand2.Function);
   }
 
+  // In SYCL device compilation mode prefer the overload with the
+  // SYCLDeviceOnly attribute.
+  if (S.getLangOpts().isSYCL() && S.getLangOpts().SYCLIsDevice &&
+      Cand1.Function && Cand2.Function) {
+    if (Cand1.Function->hasAttr<SYCLDeviceOnlyAttr>() !=
+        Cand2.Function->hasAttr<SYCLDeviceOnlyAttr>()) {
+      return Cand1.Function->hasAttr<SYCLDeviceOnlyAttr>();
+    }
+  }
+
   // General member function overloading is handled above, so this only handles
   // constructors with address spaces.
   // This only handles address spaces since C++ has no other
@@ -11314,6 +11332,15 @@ OverloadingResult OverloadCandidateSet::BestViableFunctionImpl(
   if (S.getLangOpts().CUDA)
     CudaExcludeWrongSideCandidates(S, Candidates);
 
+  // In SYCL host compilation remove candidates marked SYCLDeviceOnly.
+  if (S.getLangOpts().isSYCL() && !S.getLangOpts().SYCLIsDevice) {
+    auto IsDeviceCand = [&](const OverloadCandidate *Cand) {
+      return Cand->Viable && Cand->Function &&
+             Cand->Function->hasAttr<SYCLDeviceOnlyAttr>();
+    };
+    llvm::erase_if(Candidates, IsDeviceCand);
+  }
+
   Best = end();
   for (auto *Cand : Candidates) {
     Cand->Best = false;

clang/test/Misc/pragma-attribute-supported-attributes-list.test

@@ -189,6 +189,7 @@
 // CHECK-NEXT: SYCLDeviceGlobal (SubjectMatchRule_record)
 // CHECK-NEXT: SYCLDeviceHas (SubjectMatchRule_function)
 // CHECK-NEXT: SYCLDeviceIndirectlyCallable (SubjectMatchRule_function)
+// CHECK-NEXT: SYCLDeviceOnly (SubjectMatchRule_function)
 // CHECK-NEXT: SYCLGlobalVariableAllowed (SubjectMatchRule_record)
 // CHECK-NEXT: SYCLIntelDisableLoopPipelining (SubjectMatchRule_function)
 // CHECK-NEXT: SYCLIntelInitiationInterval (SubjectMatchRule_function)

libdevice/cmath_wrapper.cpp

@@ -8,8 +8,7 @@
 
 #include "device_math.h"
 
-#if defined(__SPIR__) || defined(__SPIRV__) || defined(__NVPTX__) ||           \
-    defined(__AMDGCN__)
+#if defined(__SPIR__) || defined(__SPIRV__)
 
 DEVICE_EXTERN_C_INLINE
 int abs(int x) { return __devicelib_abs(x); }
@@ -210,4 +209,4 @@ DEVICE_EXTERN_C_INLINE
 float rintf(float x) { return __ocml_rint_f32(x); }
 #endif // __AMDGCN__
 
-#endif // __SPIR__ || __SPIRV__ || __NVPTX__ || __AMDGCN__
+#endif // __SPIR__ || __SPIRV__

libdevice/cmath_wrapper_fp64.cpp

@@ -9,8 +9,7 @@
 
 #include "device_math.h"
 
-#if defined(__SPIR__) || defined(__SPIRV__) || defined(__NVPTX__) ||           \
-    defined(__AMDGCN__)
+#if defined(__SPIR__) || defined(__SPIRV__)
 
 // All exported functions in math and complex device libraries are weak
 // reference. If users provide their own math or complex functions(with
@@ -507,4 +506,4 @@ double _Sinh(double x, double y) { // compute y * sinh(x), |y| <= 1
   }
 }
 #endif // defined(_WIN32)
-#endif // __SPIR__ || __SPIRV__ || __NVPTX__ || __AMDGCN__
+#endif // __SPIR__ || __SPIRV__

libdevice/fallback-cmath.cpp

@@ -8,8 +8,7 @@
 
 #include "device_math.h"
 
-#if defined(__SPIR__) || defined(__SPIRV__) || defined(__NVPTX__) ||           \
-    defined(__AMDGCN__)
+#if defined(__SPIR__) || defined(__SPIRV__)
 
 // To support fallback device libraries on-demand loading, please update the
 // DeviceLibFuncMap in llvm/tools/sycl-post-link/sycl-post-link.cpp if you add
@@ -215,4 +214,4 @@ float __devicelib_asinhf(float x) { return __spirv_ocl_asinh(x); }
 DEVICE_EXTERN_C_INLINE
 float __devicelib_atanhf(float x) { return __spirv_ocl_atanh(x); }
 
-#endif // __SPIR__ || __SPIRV__ || __NVPTX__ || __AMDGCN__
+#endif // __SPIR__ || __SPIRV__

sycl/include/sycl/stl_wrappers/cmath

@@ -16,6 +16,10 @@
 
 // *** <sycl/builtins.hpp> ***
 
+#if defined(__NVPTX__) || defined(__AMDGCN__)
+#include "cmath-fallback.h"
+#endif
+
 #include <sycl/detail/defines_elementary.hpp>
 
 #ifdef __SYCL_DEVICE_ONLY__