[flang][runtime] Enable PRINT of integer32 for device.

flang/include/flang/Common/real.h

@@ -13,6 +13,7 @@
 // The various representations are distinguished by their binary precisions
 // (number of explicit significand bits and any implicit MSB in the fraction).
 
+#include "flang/Runtime/api-attrs.h"
 #include <cinttypes>
 
 namespace Fortran::common {
@@ -119,6 +120,7 @@ template <int BINARY_PRECISION> class RealDetails {
   }
 
 public:
+  RT_OFFLOAD_VAR_GROUP_BEGIN
   static constexpr int binaryPrecision{BINARY_PRECISION};
   static constexpr int bits{BitsForBinaryPrecision(binaryPrecision)};
   static constexpr bool isImplicitMSB{binaryPrecision != 64 /*x87*/};
@@ -138,6 +140,7 @@ template <int BINARY_PRECISION> class RealDetails {
 
   static constexpr int maxHexadecimalConversionDigits{
       MaxHexadecimalConversionDigits(binaryPrecision)};
+  RT_OFFLOAD_VAR_GROUP_END
 
   static_assert(binaryPrecision > 0);
   static_assert(exponentBits > 1);

flang/include/flang/Common/uint128.h

@@ -20,6 +20,7 @@
 #endif
 
 #include "leading-zero-bit-count.h"
+#include "flang/Runtime/api-attrs.h"
 #include <cstdint>
 #include <type_traits>
 
@@ -260,7 +261,9 @@ template <bool IS_SIGNED = false> class Int128 {
       return LeadingZeroBitCount(high_);
     }
   }
+  RT_VAR_GROUP_BEGIN
   static constexpr std::uint64_t topBit{std::uint64_t{1} << 63};
+  RT_VAR_GROUP_END
 #if FLANG_LITTLE_ENDIAN
   std::uint64_t low_{0}, high_{0};
 #elif FLANG_BIG_ENDIAN

flang/include/flang/Decimal/binary-floating-point.h

@@ -14,6 +14,7 @@
 
 #include "flang/Common/real.h"
 #include "flang/Common/uint128.h"
+#include "flang/Runtime/api-attrs.h"
 #include <cinttypes>
 #include <climits>
 #include <cstring>
@@ -47,9 +48,11 @@ class BinaryFloatingPointNumber : public common::RealDetails<BINARY_PRECISION> {
 
   using RawType = common::HostUnsignedIntType<bits>;
   static_assert(CHAR_BIT * sizeof(RawType) >= bits);
+  RT_OFFLOAD_VAR_GROUP_BEGIN
   static constexpr RawType significandMask{(RawType{1} << significandBits) - 1};
 
-  constexpr BinaryFloatingPointNumber() {} // zero
+  constexpr RT_API_ATTRS BinaryFloatingPointNumber() {} // zero
+  RT_OFFLOAD_VAR_GROUP_END
   constexpr BinaryFloatingPointNumber(
       const BinaryFloatingPointNumber &that) = default;
   constexpr BinaryFloatingPointNumber(
@@ -58,37 +61,41 @@ class BinaryFloatingPointNumber : public common::RealDetails<BINARY_PRECISION> {
       const BinaryFloatingPointNumber &that) = default;
   constexpr BinaryFloatingPointNumber &operator=(
       BinaryFloatingPointNumber &&that) = default;
-  constexpr explicit BinaryFloatingPointNumber(RawType raw) : raw_{raw} {}
+  constexpr explicit RT_API_ATTRS BinaryFloatingPointNumber(RawType raw)
+      : raw_{raw} {}
 
-  RawType raw() const { return raw_; }
+  RT_API_ATTRS RawType raw() const { return raw_; }
 
-  template <typename A> explicit constexpr BinaryFloatingPointNumber(A x) {
+  template <typename A>
+  explicit constexpr RT_API_ATTRS BinaryFloatingPointNumber(A x) {
     static_assert(sizeof raw_ <= sizeof x);
     std::memcpy(reinterpret_cast<void *>(&raw_),
         reinterpret_cast<const void *>(&x), sizeof raw_);
   }
 
-  constexpr int BiasedExponent() const {
+  constexpr RT_API_ATTRS int BiasedExponent() const {
     return static_cast<int>(
         (raw_ >> significandBits) & ((1 << exponentBits) - 1));
   }
-  constexpr int UnbiasedExponent() const {
+  constexpr RT_API_ATTRS int UnbiasedExponent() const {
     int biased{BiasedExponent()};
     return biased - exponentBias + (biased == 0);
   }
-  constexpr RawType Significand() const { return raw_ & significandMask; }
-  constexpr RawType Fraction() const {
+  constexpr RT_API_ATTRS RawType Significand() const {
+    return raw_ & significandMask;
+  }
+  constexpr RT_API_ATTRS RawType Fraction() const {
     RawType sig{Significand()};
     if (isImplicitMSB && BiasedExponent() > 0) {
       sig |= RawType{1} << significandBits;
     }
     return sig;
   }
 
-  constexpr bool IsZero() const {
+  constexpr RT_API_ATTRS bool IsZero() const {
     return (raw_ & ((RawType{1} << (bits - 1)) - 1)) == 0;
   }
-  constexpr bool IsNaN() const {
+  constexpr RT_API_ATTRS bool IsNaN() const {
     auto expo{BiasedExponent()};
     auto sig{Significand()};
     if constexpr (bits == 80) { // x87
@@ -102,35 +109,38 @@ class BinaryFloatingPointNumber : public common::RealDetails<BINARY_PRECISION> {
       return expo == maxExponent && sig != 0;
     }
   }
-  constexpr bool IsInfinite() const {
+  constexpr RT_API_ATTRS bool IsInfinite() const {
     if constexpr (bits == 80) { // x87
       return BiasedExponent() == maxExponent &&
           Significand() == ((significandMask >> 1) + 1);
     } else {
       return BiasedExponent() == maxExponent && Significand() == 0;
     }
   }
-  constexpr bool IsMaximalFiniteMagnitude() const {
+  constexpr RT_API_ATTRS bool IsMaximalFiniteMagnitude() const {
     return BiasedExponent() == maxExponent - 1 &&
         Significand() == significandMask;
   }
-  constexpr bool IsNegative() const { return ((raw_ >> (bits - 1)) & 1) != 0; }
+  constexpr RT_API_ATTRS bool IsNegative() const {
+    return ((raw_ >> (bits - 1)) & 1) != 0;
+  }
 
-  constexpr void Negate() { raw_ ^= RawType{1} << (bits - 1); }
+  constexpr RT_API_ATTRS void Negate() { raw_ ^= RawType{1} << (bits - 1); }
 
   // For calculating the nearest neighbors of a floating-point value
-  constexpr void Previous() {
+  constexpr RT_API_ATTRS void Previous() {
     RemoveExplicitMSB();
     --raw_;
     InsertExplicitMSB();
   }
-  constexpr void Next() {
+  constexpr RT_API_ATTRS void Next() {
     RemoveExplicitMSB();
     ++raw_;
     InsertExplicitMSB();
   }
 
-  static constexpr BinaryFloatingPointNumber Infinity(bool isNegative) {
+  static constexpr RT_API_ATTRS BinaryFloatingPointNumber Infinity(
+      bool isNegative) {
     RawType result{RawType{maxExponent} << significandBits};
     if (isNegative) {
       result |= RawType{1} << (bits - 1);
@@ -139,7 +149,8 @@ class BinaryFloatingPointNumber : public common::RealDetails<BINARY_PRECISION> {
   }
 
   // Returns true when the result is exact
-  constexpr bool RoundToBits(int keepBits, enum FortranRounding mode) {
+  constexpr RT_API_ATTRS bool RoundToBits(
+      int keepBits, enum FortranRounding mode) {
     if (IsNaN() || IsInfinite() || keepBits >= binaryPrecision) {
       return true;
     }
@@ -180,12 +191,12 @@ class BinaryFloatingPointNumber : public common::RealDetails<BINARY_PRECISION> {
   }
 
 private:
-  constexpr void RemoveExplicitMSB() {
+  constexpr RT_API_ATTRS void RemoveExplicitMSB() {
     if constexpr (!isImplicitMSB) {
       raw_ = (raw_ & (significandMask >> 1)) | ((raw_ & ~significandMask) >> 1);
     }
   }
-  constexpr void InsertExplicitMSB() {
+  constexpr RT_API_ATTRS void InsertExplicitMSB() {
     if constexpr (!isImplicitMSB) {
       constexpr RawType mask{significandMask >> 1};
       raw_ = (raw_ & mask) | ((raw_ & ~mask) << 1);

flang/include/flang/Decimal/decimal.h

@@ -12,6 +12,7 @@
 #ifndef FORTRAN_DECIMAL_DECIMAL_H_
 #define FORTRAN_DECIMAL_DECIMAL_H_
 
+#include "flang/Runtime/api-attrs.h"
 #include <stddef.h>
 
 #ifdef __cplusplus
@@ -65,27 +66,27 @@ enum DecimalConversionFlags {
 
 #ifdef __cplusplus
 template <int PREC>
-ConversionToDecimalResult ConvertToDecimal(char *, size_t,
+RT_API_ATTRS ConversionToDecimalResult ConvertToDecimal(char *, size_t,
     DecimalConversionFlags, int digits, enum FortranRounding rounding,
     BinaryFloatingPointNumber<PREC> x);
 
-extern template ConversionToDecimalResult ConvertToDecimal<8>(char *, size_t,
-    enum DecimalConversionFlags, int, enum FortranRounding,
+extern template RT_API_ATTRS ConversionToDecimalResult ConvertToDecimal<8>(
+    char *, size_t, enum DecimalConversionFlags, int, enum FortranRounding,
     BinaryFloatingPointNumber<8>);
-extern template ConversionToDecimalResult ConvertToDecimal<11>(char *, size_t,
-    enum DecimalConversionFlags, int, enum FortranRounding,
+extern template RT_API_ATTRS ConversionToDecimalResult ConvertToDecimal<11>(
+    char *, size_t, enum DecimalConversionFlags, int, enum FortranRounding,
     BinaryFloatingPointNumber<11>);
-extern template ConversionToDecimalResult ConvertToDecimal<24>(char *, size_t,
-    enum DecimalConversionFlags, int, enum FortranRounding,
+extern template RT_API_ATTRS ConversionToDecimalResult ConvertToDecimal<24>(
+    char *, size_t, enum DecimalConversionFlags, int, enum FortranRounding,
     BinaryFloatingPointNumber<24>);
-extern template ConversionToDecimalResult ConvertToDecimal<53>(char *, size_t,
-    enum DecimalConversionFlags, int, enum FortranRounding,
+extern template RT_API_ATTRS ConversionToDecimalResult ConvertToDecimal<53>(
+    char *, size_t, enum DecimalConversionFlags, int, enum FortranRounding,
     BinaryFloatingPointNumber<53>);
-extern template ConversionToDecimalResult ConvertToDecimal<64>(char *, size_t,
-    enum DecimalConversionFlags, int, enum FortranRounding,
+extern template RT_API_ATTRS ConversionToDecimalResult ConvertToDecimal<64>(
+    char *, size_t, enum DecimalConversionFlags, int, enum FortranRounding,
     BinaryFloatingPointNumber<64>);
-extern template ConversionToDecimalResult ConvertToDecimal<113>(char *, size_t,
-    enum DecimalConversionFlags, int, enum FortranRounding,
+extern template RT_API_ATTRS ConversionToDecimalResult ConvertToDecimal<113>(
+    char *, size_t, enum DecimalConversionFlags, int, enum FortranRounding,
     BinaryFloatingPointNumber<113>);
 
 template <int PREC> struct ConversionToBinaryResult {
@@ -94,20 +95,20 @@ template <int PREC> struct ConversionToBinaryResult {
 };
 
 template <int PREC>
-ConversionToBinaryResult<PREC> ConvertToBinary(const char *&,
+RT_API_ATTRS ConversionToBinaryResult<PREC> ConvertToBinary(const char *&,
     enum FortranRounding = RoundNearest, const char *end = nullptr);
 
-extern template ConversionToBinaryResult<8> ConvertToBinary<8>(
+extern template RT_API_ATTRS ConversionToBinaryResult<8> ConvertToBinary<8>(
     const char *&, enum FortranRounding, const char *end);
-extern template ConversionToBinaryResult<11> ConvertToBinary<11>(
+extern template RT_API_ATTRS ConversionToBinaryResult<11> ConvertToBinary<11>(
     const char *&, enum FortranRounding, const char *end);
-extern template ConversionToBinaryResult<24> ConvertToBinary<24>(
+extern template RT_API_ATTRS ConversionToBinaryResult<24> ConvertToBinary<24>(
     const char *&, enum FortranRounding, const char *end);
-extern template ConversionToBinaryResult<53> ConvertToBinary<53>(
+extern template RT_API_ATTRS ConversionToBinaryResult<53> ConvertToBinary<53>(
     const char *&, enum FortranRounding, const char *end);
-extern template ConversionToBinaryResult<64> ConvertToBinary<64>(
+extern template RT_API_ATTRS ConversionToBinaryResult<64> ConvertToBinary<64>(
     const char *&, enum FortranRounding, const char *end);
-extern template ConversionToBinaryResult<113> ConvertToBinary<113>(
+extern template RT_API_ATTRS ConversionToBinaryResult<113> ConvertToBinary<113>(
     const char *&, enum FortranRounding, const char *end);
 } // namespace Fortran::decimal
 extern "C" {
@@ -116,21 +117,21 @@ extern "C" {
 #define NS(x) x
 #endif /* C++ */
 
-struct NS(ConversionToDecimalResult)
+RT_API_ATTRS struct NS(ConversionToDecimalResult)
     ConvertFloatToDecimal(char *, size_t, enum NS(DecimalConversionFlags),
         int digits, enum NS(FortranRounding), float);
-struct NS(ConversionToDecimalResult)
+RT_API_ATTRS struct NS(ConversionToDecimalResult)
     ConvertDoubleToDecimal(char *, size_t, enum NS(DecimalConversionFlags),
         int digits, enum NS(FortranRounding), double);
-struct NS(ConversionToDecimalResult)
+RT_API_ATTRS struct NS(ConversionToDecimalResult)
     ConvertLongDoubleToDecimal(char *, size_t, enum NS(DecimalConversionFlags),
         int digits, enum NS(FortranRounding), long double);
 
-enum NS(ConversionResultFlags)
+RT_API_ATTRS enum NS(ConversionResultFlags)
     ConvertDecimalToFloat(const char **, float *, enum NS(FortranRounding));
-enum NS(ConversionResultFlags)
+RT_API_ATTRS enum NS(ConversionResultFlags)
     ConvertDecimalToDouble(const char **, double *, enum NS(FortranRounding));
-enum NS(ConversionResultFlags) ConvertDecimalToLongDouble(
+RT_API_ATTRS enum NS(ConversionResultFlags) ConvertDecimalToLongDouble(
     const char **, long double *, enum NS(FortranRounding));
 #undef NS
 #ifdef __cplusplus

flang/include/flang/Runtime/api-attrs.h

@@ -102,7 +102,7 @@
  * to appear as part of a C++ decl-specifier.
  */
 #ifndef RT_CONST_VAR_ATTRS
-#if defined(__CUDACC__) || defined(__CUDA__)
+#if (defined(__CUDACC__) || defined(__CUDA__)) && defined(__CUDA_ARCH__)
 #define RT_CONST_VAR_ATTRS __constant__
 #else
 #define RT_CONST_VAR_ATTRS

flang/include/flang/Runtime/io-api.h

@@ -58,6 +58,14 @@ extern "C" {
 
 #define IONAME(name) RTNAME(io##name)
 
+#ifndef IODECL
+#define IODECL(name) RT_API_ATTRS IONAME(name)
+#endif
+
+#ifndef IODEF
+#define IODEF(name) RT_API_ATTRS IONAME(name)
+#endif
+
 // These functions initiate data transfer statements (READ, WRITE, PRINT).
 // Example: PRINT *, 666 is implemented as the series of calls:
 //   Cookie cookie{BeginExternalListOutput(DefaultOutputUnit,
@@ -139,7 +147,7 @@ enum Iostat IONAME(CheckUnitNumberInRange128)(common::int128_t unit,
     const char *sourceFile = nullptr, int sourceLine = 0);
 
 // External synchronous I/O initiation
-Cookie IONAME(BeginExternalListOutput)(ExternalUnit = DefaultOutputUnit,
+Cookie IODECL(BeginExternalListOutput)(ExternalUnit = DefaultOutputUnit,
     const char *sourceFile = nullptr, int sourceLine = 0);
 Cookie IONAME(BeginExternalListInput)(ExternalUnit = DefaultInputUnit,
     const char *sourceFile = nullptr, int sourceLine = 0);
@@ -253,7 +261,7 @@ bool IONAME(InputDescriptor)(Cookie, const Descriptor &);
 // Formatted (including list directed) I/O data items
 bool IONAME(OutputInteger8)(Cookie, std::int8_t);
 bool IONAME(OutputInteger16)(Cookie, std::int16_t);
-bool IONAME(OutputInteger32)(Cookie, std::int32_t);
+bool IODECL(OutputInteger32)(Cookie, std::int32_t);
 bool IONAME(OutputInteger64)(Cookie, std::int64_t);
 bool IONAME(OutputInteger128)(Cookie, common::int128_t);
 bool IONAME(InputInteger)(Cookie, std::int64_t &, int kind = 8);
@@ -357,7 +365,7 @@ bool IONAME(InquireInteger64)(
 // returned is guaranteed to only be one of the problems that the
 // EnableHandlers() call has indicated should be handled in compiled code
 // rather than by terminating the image.
-enum Iostat IONAME(EndIoStatement)(Cookie);
+enum Iostat IODECL(EndIoStatement)(Cookie);
 
 } // extern "C"
 } // namespace Fortran::runtime::io

flang/include/flang/Runtime/iostat.h

@@ -11,6 +11,7 @@
 
 #ifndef FORTRAN_RUNTIME_IOSTAT_H_
 #define FORTRAN_RUNTIME_IOSTAT_H_
+#include "flang/Runtime/api-attrs.h"
 #include "flang/Runtime/magic-numbers.h"
 namespace Fortran::runtime::io {
 
@@ -88,7 +89,7 @@ enum Iostat {
   IostatNonExternalDefinedUnformattedIo,
 };
 
-const char *IostatErrorString(int);
+const RT_API_ATTRS char *IostatErrorString(int);
 
 } // namespace Fortran::runtime::io
 #endif // FORTRAN_RUNTIME_IOSTAT_H_

flang/include/flang/Runtime/memory.h

@@ -128,20 +128,30 @@ inline RT_API_ATTRS bool operator!=(std::nullptr_t, const OwningPtr<X> &x) {
 
 template <typename A> class SizedNew {
 public:
-  explicit SizedNew(const Terminator &terminator) : terminator_{terminator} {}
+  explicit RT_API_ATTRS SizedNew(const Terminator &terminator)
+      : terminator_{terminator} {}
+
+  // Disable warnings about calling constructors for host-only
+  // classes (those thare are not supposed to be constructed)
+  // on the device.
+  RT_DIAG_PUSH
+  RT_DIAG_DISABLE_CALL_HOST_FROM_DEVICE_WARN
   template <typename... X>
-  [[nodiscard]] OwningPtr<A> operator()(std::size_t bytes, X &&...x) {
+  [[nodiscard]] RT_API_ATTRS OwningPtr<A> operator()(
+      std::size_t bytes, X &&...x) {
     return OwningPtr<A>{new (AllocateMemoryOrCrash(terminator_, bytes))
             A{std::forward<X>(x)...}};
   }
+  RT_DIAG_POP
 
 private:
   const Terminator &terminator_;
 };
 
 template <typename A> struct New : public SizedNew<A> {
   using SizedNew<A>::SizedNew;
-  template <typename... X> [[nodiscard]] OwningPtr<A> operator()(X &&...x) {
+  template <typename... X>
+  [[nodiscard]] RT_API_ATTRS OwningPtr<A> operator()(X &&...x) {
     return SizedNew<A>::operator()(sizeof(A), std::forward<X>(x)...);
   }
 };

flang/include/flang/Runtime/type-code.h

@@ -12,7 +12,6 @@
 #include "flang/Common/Fortran.h"
 #include "flang/Common/optional.h"
 #include "flang/ISO_Fortran_binding_wrapper.h"
-#include <optional>
 #include <utility>
 
 namespace Fortran::runtime {