llvm-runtime-resources-darwin 0.0.0 → 23.1.0

package/include/cpuid.h

@@ -0,0 +1,451 @@
+/*===---- cpuid.h - X86 cpu model detection --------------------------------===
+ *
+ * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+ * See https://llvm.org/LICENSE.txt for license information.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ *
+ *===-----------------------------------------------------------------------===
+ */
+
+#ifndef __CPUID_H
+#define __CPUID_H
+
+#if !defined(__x86_64__) && !defined(__i386__)
+#error this header is for x86 only
+#endif
+
+/* Responses identification request with %eax 0 */
+/* AMD:     "AuthenticAMD" */
+#define signature_AMD_ebx 0x68747541
+#define signature_AMD_edx 0x69746e65
+#define signature_AMD_ecx 0x444d4163
+/* CENTAUR: "CentaurHauls" */
+#define signature_CENTAUR_ebx 0x746e6543
+#define signature_CENTAUR_edx 0x48727561
+#define signature_CENTAUR_ecx 0x736c7561
+/* CYRIX:   "CyrixInstead" */
+#define signature_CYRIX_ebx 0x69727943
+#define signature_CYRIX_edx 0x736e4978
+#define signature_CYRIX_ecx 0x64616574
+/* HYGON:   "HygonGenuine" */
+#define signature_HYGON_ebx 0x6f677948
+#define signature_HYGON_edx 0x6e65476e
+#define signature_HYGON_ecx 0x656e6975
+/* INTEL:   "GenuineIntel" */
+#define signature_INTEL_ebx 0x756e6547
+#define signature_INTEL_edx 0x49656e69
+#define signature_INTEL_ecx 0x6c65746e
+/* TM1:     "TransmetaCPU" */
+#define signature_TM1_ebx 0x6e617254
+#define signature_TM1_edx 0x74656d73
+#define signature_TM1_ecx 0x55504361
+/* TM2:     "GenuineTMx86" */
+#define signature_TM2_ebx 0x756e6547
+#define signature_TM2_edx 0x54656e69
+#define signature_TM2_ecx 0x3638784d
+/* NSC:     "Geode by NSC" */
+#define signature_NSC_ebx 0x646f6547
+#define signature_NSC_edx 0x79622065
+#define signature_NSC_ecx 0x43534e20
+/* NEXGEN:  "NexGenDriven" */
+#define signature_NEXGEN_ebx 0x4778654e
+#define signature_NEXGEN_edx 0x72446e65
+#define signature_NEXGEN_ecx 0x6e657669
+/* RISE:    "RiseRiseRise" */
+#define signature_RISE_ebx 0x65736952
+#define signature_RISE_edx 0x65736952
+#define signature_RISE_ecx 0x65736952
+/* SIS:     "SiS SiS SiS " */
+#define signature_SIS_ebx 0x20536953
+#define signature_SIS_edx 0x20536953
+#define signature_SIS_ecx 0x20536953
+/* UMC:     "UMC UMC UMC " */
+#define signature_UMC_ebx 0x20434d55
+#define signature_UMC_edx 0x20434d55
+#define signature_UMC_ecx 0x20434d55
+/* VIA:     "VIA VIA VIA " */
+#define signature_VIA_ebx 0x20414956
+#define signature_VIA_edx 0x20414956
+#define signature_VIA_ecx 0x20414956
+/* VORTEX:  "Vortex86 SoC" */
+#define signature_VORTEX_ebx 0x74726f56
+#define signature_VORTEX_edx 0x36387865
+#define signature_VORTEX_ecx 0x436f5320
+
+/* Features in %ecx for leaf 1 */
+#define bit_SSE3        0x00000001
+#define bit_PCLMULQDQ   0x00000002
+#define bit_PCLMUL      bit_PCLMULQDQ   /* for gcc compat */
+#define bit_DTES64      0x00000004
+#define bit_MONITOR     0x00000008
+#define bit_DSCPL       0x00000010
+#define bit_VMX         0x00000020
+#define bit_SMX         0x00000040
+#define bit_EIST        0x00000080
+#define bit_TM2         0x00000100
+#define bit_SSSE3       0x00000200
+#define bit_CNXTID      0x00000400
+#define bit_FMA         0x00001000
+#define bit_CMPXCHG16B  0x00002000
+#define bit_xTPR        0x00004000
+#define bit_PDCM        0x00008000
+#define bit_PCID        0x00020000
+#define bit_DCA         0x00040000
+#define bit_SSE41       0x00080000
+#define bit_SSE4_1      bit_SSE41       /* for gcc compat */
+#define bit_SSE42       0x00100000
+#define bit_SSE4_2      bit_SSE42       /* for gcc compat */
+#define bit_x2APIC      0x00200000
+#define bit_MOVBE       0x00400000
+#define bit_POPCNT      0x00800000
+#define bit_TSCDeadline 0x01000000
+#define bit_AESNI       0x02000000
+#define bit_AES         bit_AESNI       /* for gcc compat */
+#define bit_XSAVE       0x04000000
+#define bit_OSXSAVE     0x08000000
+#define bit_AVX         0x10000000
+#define bit_F16C        0x20000000
+#define bit_RDRND       0x40000000
+
+/* Features in %edx for leaf 1 */
+#define bit_FPU         0x00000001
+#define bit_VME         0x00000002
+#define bit_DE          0x00000004
+#define bit_PSE         0x00000008
+#define bit_TSC         0x00000010
+#define bit_MSR         0x00000020
+#define bit_PAE         0x00000040
+#define bit_MCE         0x00000080
+#define bit_CX8         0x00000100
+#define bit_CMPXCHG8B   bit_CX8         /* for gcc compat */
+#define bit_APIC        0x00000200
+#define bit_SEP         0x00000800
+#define bit_MTRR        0x00001000
+#define bit_PGE         0x00002000
+#define bit_MCA         0x00004000
+#define bit_CMOV        0x00008000
+#define bit_PAT         0x00010000
+#define bit_PSE36       0x00020000
+#define bit_PSN         0x00040000
+#define bit_CLFSH       0x00080000
+#define bit_DS          0x00200000
+#define bit_ACPI        0x00400000
+#define bit_MMX         0x00800000
+#define bit_FXSR        0x01000000
+#define bit_FXSAVE      bit_FXSR        /* for gcc compat */
+#define bit_SSE         0x02000000
+#define bit_SSE2        0x04000000
+#define bit_SS          0x08000000
+#define bit_HTT         0x10000000
+#define bit_TM          0x20000000
+#define bit_PBE         0x80000000
+
+/* Features in %ebx for leaf 7 sub-leaf 0 */
+#define bit_FSGSBASE    0x00000001
+#define bit_SGX         0x00000004
+#define bit_BMI         0x00000008
+#define bit_HLE         0x00000010
+#define bit_AVX2        0x00000020
+#define bit_SMEP        0x00000080
+#define bit_BMI2        0x00000100
+#define bit_ENH_MOVSB   0x00000200
+#define bit_INVPCID     0x00000400
+#define bit_RTM         0x00000800
+#define bit_MPX         0x00004000
+#define bit_AVX512F     0x00010000
+#define bit_AVX512DQ    0x00020000
+#define bit_RDSEED      0x00040000
+#define bit_ADX         0x00080000
+#define bit_AVX512IFMA  0x00200000
+#define bit_CLFLUSHOPT  0x00800000
+#define bit_CLWB        0x01000000
+#define bit_AVX512PF    0x04000000
+#define bit_AVX512ER    0x08000000
+#define bit_AVX512CD    0x10000000
+#define bit_SHA         0x20000000
+#define bit_AVX512BW    0x40000000
+#define bit_AVX512VL    0x80000000
+
+/* Features in %ecx for leaf 7 sub-leaf 0 */
+#define bit_PREFTCHWT1       0x00000001
+#define bit_AVX512VBMI       0x00000002
+#define bit_PKU              0x00000004
+#define bit_OSPKE            0x00000010
+#define bit_WAITPKG          0x00000020
+#define bit_AVX512VBMI2      0x00000040
+#define bit_SHSTK            0x00000080
+#define bit_GFNI             0x00000100
+#define bit_VAES             0x00000200
+#define bit_VPCLMULQDQ       0x00000400
+#define bit_AVX512VNNI       0x00000800
+#define bit_AVX512BITALG     0x00001000
+#define bit_AVX512VPOPCNTDQ  0x00004000
+#define bit_RDPID            0x00400000
+#define bit_CLDEMOTE         0x02000000
+#define bit_MOVDIRI          0x08000000
+#define bit_MOVDIR64B        0x10000000
+#define bit_ENQCMD           0x20000000
+
+/* Features in %edx for leaf 7 sub-leaf 0 */
+#define bit_AVX5124VNNIW        0x00000004
+#define bit_AVX5124FMAPS        0x00000008
+#define bit_UINTR               0x00000020
+#define bit_AVX512VP2INTERSECT  0x00000100
+#define bit_SERIALIZE           0x00004000
+#define bit_TSXLDTRK            0x00010000
+#define bit_PCONFIG             0x00040000
+#define bit_IBT                 0x00100000
+#define bit_AMXBF16             0x00400000
+#define bit_AVX512FP16          0x00800000
+#define bit_AMXTILE             0x01000000
+#define bit_AMXINT8             0x02000000
+
+/* Features in %eax for leaf 7 sub-leaf 1 */
+#define bit_SHA512        0x00000001
+#define bit_SM3           0x00000002
+#define bit_SM4           0x00000004
+#define bit_RAOINT        0x00000008
+#define bit_AVXVNNI       0x00000010
+#define bit_AVX512BF16    0x00000020
+#define bit_CMPCCXADD     0x00000080
+#define bit_AMXFP16       0x00200000
+#define bit_HRESET        0x00400000
+#define bit_AVXIFMA       0x00800000
+
+/* Features in %edx for leaf 7 sub-leaf 1 */
+#define bit_AVXVNNIINT8   0x00000010
+#define bit_AVXNECONVERT  0x00000020
+#define bit_AMXCOMPLEX    0x00000100
+#define bit_AVXVNNIINT16  0x00000400
+#define bit_PREFETCHI     0x00004000
+#define bit_USERMSR       0x00008000
+#define bit_AVX10         0x00080000
+#define bit_APXF          0x00200000
+
+/* Features in %eax for leaf 13 sub-leaf 1 */
+#define bit_XSAVEOPT    0x00000001
+#define bit_XSAVEC      0x00000002
+#define bit_XSAVES      0x00000008
+
+/* Features in %eax for leaf 0x14 sub-leaf 0 */
+#define bit_PTWRITE     0x00000010
+
+/* Features in %ecx for leaf 0x80000001 */
+#define bit_LAHF_LM     0x00000001
+#define bit_ABM         0x00000020
+#define bit_LZCNT       bit_ABM        /* for gcc compat */
+#define bit_SSE4a       0x00000040
+#define bit_PRFCHW      0x00000100
+#define bit_XOP         0x00000800
+#define bit_LWP         0x00008000
+#define bit_FMA4        0x00010000
+#define bit_TBM         0x00200000
+#define bit_MWAITX      0x20000000
+
+/* Features in %edx for leaf 0x80000001 */
+#define bit_MMXEXT      0x00400000
+#define bit_LM          0x20000000
+#define bit_3DNOWP      0x40000000
+#define bit_3DNOW       0x80000000
+
+/* Features in %ebx for leaf 0x80000008 */
+#define bit_CLZERO      0x00000001
+#define bit_RDPRU       0x00000010
+#define bit_WBNOINVD    0x00000200
+
+#ifdef __i386__
+#define __cpuid(__leaf, __eax, __ebx, __ecx, __edx) \
+    __asm("cpuid" : "=a"(__eax), "=b" (__ebx), "=c"(__ecx), "=d"(__edx) \
+                  : "0"(__leaf))
+
+#define __cpuid_count(__leaf, __count, __eax, __ebx, __ecx, __edx) \
+    __asm("cpuid" : "=a"(__eax), "=b" (__ebx), "=c"(__ecx), "=d"(__edx) \
+                  : "0"(__leaf), "2"(__count))
+#else
+/* x86-64 uses %rbx as the base register, so preserve it. */
+#define __cpuid(__leaf, __eax, __ebx, __ecx, __edx)                            \
+  __asm("  xchg{q|}  {%%|}rbx,%q1\n"                                           \
+        "  cpuid\n"                                                            \
+        "  xchg{q|}  {%%|}rbx,%q1"                                             \
+        : "=a"(__eax), "=r"(__ebx), "=c"(__ecx), "=d"(__edx)                   \
+        : "0"(__leaf))
+
+#define __cpuid_count(__leaf, __count, __eax, __ebx, __ecx, __edx)             \
+  __asm("  xchg{q|}  {%%|}rbx,%q1\n"                                           \
+        "  cpuid\n"                                                            \
+        "  xchg{q|}  {%%|}rbx,%q1"                                             \
+        : "=a"(__eax), "=r"(__ebx), "=c"(__ecx), "=d"(__edx)                   \
+        : "0"(__leaf), "2"(__count))
+#endif
+
+/// Queries the processor to determine the highest supported \c CPUID leaf.
+/// This intrinsic is only available on x86 and x64.
+///
+/// \headerfile <cpuid.h>
+///
+/// This intrinsic corresponds to the <c> CPUID </c> instruction.
+///
+/// \param __leaf
+///    \a __leaf can be either 0x0 or 0x8000000. If \a __leaf == 0x0, the
+///    highest supported value for basic \c CPUID information is returned.
+///    If \a __leaf == 0x8000000, the highest supported value for extended
+///    \c CPUID information is returned.
+/// \param __sig
+///    If the \a __sig pointer is non-null, the first four bytes of the
+///    signature (as found in the \c EBX register) are returned in the
+///    location pointed to by \a __sig.
+/// \returns Returns 0 if \c CPUID is supported; otherwise returns the value
+///    that \c CPUID returns in the \c EAX register.
+static __inline unsigned int __get_cpuid_max (unsigned int __leaf,
+                                              unsigned int *__sig)
+{
+    unsigned int __eax, __ebx, __ecx, __edx;
+#ifdef __i386__
+    int __cpuid_supported;
+
+    __asm("  pushf{l|d}\n"
+          "  pop{l|}   {%%|}eax\n"
+          "  mov{l|}   {%%eax,%%ecx|ecx,eax}\n"
+          "  xor{l|}   {$0x00200000,%%eax|eax,0x00200000}\n"
+          "  push{l|}  {%%|}eax\n"
+          "  popf{l|d}\n"
+          "  pushf{l|d}\n"
+          "  pop{l|}   {%%|}eax\n"
+          "  mov{l|}   {$0,%0|%0,0}\n"
+          "  cmp{l|}   {%%eax,%%ecx|ecx,eax}\n"
+          "  je     1f\n"
+          "  mov{l|}   {$1,%0|%0,1}\n"
+          "1:"
+          : "=r"(__cpuid_supported)
+          :
+          : "eax", "ecx");
+    if (!__cpuid_supported)
+        return 0;
+#endif
+
+    __cpuid(__leaf, __eax, __ebx, __ecx, __edx);
+    if (__sig)
+        *__sig = __ebx;
+    return __eax;
+}
+
+/// For the requested \c CPUID leaf, queries the processor for information
+/// about the CPU type and CPU features (such as processor vendor, supported
+/// instruction sets, CPU capabilities, cache sizes, CPU model and family, and
+/// other hardware details). This intrinsic is only available on x86 and x64.
+///
+/// \headerfile <cpuid.h>
+///
+/// This intrinsic corresponds to the <c> CPUID </c> instruction.
+///
+/// \param __leaf
+///    An unsigned integer that identifies the level (also called "leaf") at
+///    which the \c CPUID instruction will be executed.
+/// \param __eax
+///    A pointer to an integer that corresponds to the \c EAX register where
+///    \c CPUID stores output results.
+/// \param __ebx
+///    A pointer to an integer that corresponds to the \c EBX register where
+///    \c CPUID stores output results.
+/// \param __ecx
+///    A pointer to an integer that corresponds to the \c ECX register where
+///    \c CPUID stores output results.
+/// \param __edx
+///    A pointer to an integer that corresponds to the \c EDX register where
+///    \c CPUID stores output results.
+/// \returns Returns 1 if the requested \c CPUID leaf is supported; otherwise
+///    returns 0.
+static __inline int __get_cpuid (unsigned int __leaf, unsigned int *__eax,
+                                 unsigned int *__ebx, unsigned int *__ecx,
+                                 unsigned int *__edx)
+{
+    unsigned int __max_leaf = __get_cpuid_max(__leaf & 0x80000000, 0);
+
+    if (__max_leaf == 0 || __max_leaf < __leaf)
+        return 0;
+
+    __cpuid(__leaf, *__eax, *__ebx, *__ecx, *__edx);
+    return 1;
+}
+
+/// For the requested \c CPUID leaf and subleaf, queries the processor for
+/// information about the CPU type and CPU features (such as processor vendor,
+/// supported instruction sets, CPU capabilities, cache sizes, CPU model and
+/// family, and other hardware details). This intrinsic is only available on
+/// x86 and x64.
+///
+/// \headerfile <cpuid.h>
+///
+/// This intrinsic corresponds to the <c> CPUID </c> instruction.
+///
+/// \param __leaf
+///    An unsigned integer that identifies the level (also called "leaf") at
+///    which the \c CPUID instruction will be executed.
+/// \param __subleaf
+///    An unsigned integer that identifies the sublevel (also called
+///    "subleaf") at which the \c CPUID instruction will be executed.
+/// \param __eax
+///    A pointer to an integer that corresponds to the \c EAX register where
+///    \c CPUID stores output results.
+/// \param __ebx
+///    A pointer to an integer that corresponds to the \c EBX register where
+///    \c CPUID stores output results.
+/// \param __ecx
+///    A pointer to an integer that corresponds to the \c ECX register where
+///    \c CPUID stores output results.
+/// \param __edx
+///    A pointer to an integer that corresponds to the \c EDX register where
+///    \c CPUID stores output results.
+/// \returns Returns 1 if the requested \c CPUID leaf is supported; otherwise
+///    returns 0.
+static __inline int __get_cpuid_count (unsigned int __leaf,
+                                       unsigned int __subleaf,
+                                       unsigned int *__eax, unsigned int *__ebx,
+                                       unsigned int *__ecx, unsigned int *__edx)
+{
+    unsigned int __max_leaf = __get_cpuid_max(__leaf & 0x80000000, 0);
+
+    if (__max_leaf == 0 || __max_leaf < __leaf)
+        return 0;
+
+    __cpuid_count(__leaf, __subleaf, *__eax, *__ebx, *__ecx, *__edx);
+    return 1;
+}
+
+// In some configurations, __cpuidex is defined as a builtin (primarily
+// -fms-extensions) which will conflict with the __cpuidex definition below.
+#if !(__has_builtin(__cpuidex))
+// In some cases, offloading will set the host as the aux triple and define the
+// builtin. Given __has_builtin does not detect builtins on aux triples, we need
+// to explicitly check for some offloading cases.
+#if !defined(__NVPTX__) && !defined(__AMDGPU__) && !defined(__SPIRV__)
+/// Executes the \c CPUID instruction with the specified leaf and subleaf
+/// values, and returns the results from the CPU's registers. This intrinsic
+/// is only available on x86 and x64.
+///
+/// \headerfile <cpuid.h>
+///
+/// This intrinsic corresponds to the <c> CPUID </c> instruction.
+///
+/// \param __cpu_info
+///    An output array of four integers:
+///    <ul>
+///    <li>\a __cpuInfo[0] receives the value of the \c EAX register.</li>
+///    <li>\a __cpuInfo[1] receives the value of the \c EBX register.</li>
+///    <li>\a __cpuInfo[2] receives the value of the \c ECX register.</li>
+///    <li>\a __cpuInfo[3] receives the value of the \c EDX register.</li>
+///    </ul>
+/// \param __leaf
+///    An unsigned integer that identifies the level (also called the "leaf")
+///    at which the \c CPUID instruction will be executed.
+/// \param __subleaf
+///    An unsigned integer that identifies the sublevel (also called the
+///    "subleaf") at which the \c CPUID instruction will be executed.
+static __inline void __cpuidex(int __cpu_info[4], int __leaf, int __subleaf) {
+  __cpuid_count(__leaf, __subleaf, __cpu_info[0], __cpu_info[1], __cpu_info[2],
+                __cpu_info[3]);
+}
+#endif
+#endif
+
+#endif /* __CPUID_H */

package/include/crc32intrin.h

@@ -0,0 +1,102 @@
+/*===---- crc32intrin.h - SSE4.2 Accumulate CRC32 intrinsics ---------------===
+ *
+ * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+ * See https://llvm.org/LICENSE.txt for license information.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ *
+ *===-----------------------------------------------------------------------===
+ */
+
+#ifndef __CRC32INTRIN_H
+#define __CRC32INTRIN_H
+
+/// We only declare crc32 as a constexpr if we are compiling C++ code
+#if defined(__cplusplus) && (__cplusplus >= 201103L)
+#define __DEFAULT_FN_ATTRS                                                     \
+  __attribute__((__always_inline__, __nodebug__, __target__("crc32"))) constexpr
+#else
+#define __DEFAULT_FN_ATTRS                                                     \
+  __attribute__((__always_inline__, __nodebug__, __target__("crc32")))
+#endif
+
+/// Adds the unsigned integer operand to the CRC-32C checksum of the
+///    unsigned char operand.
+///
+/// \headerfile <x86intrin.h>
+///
+/// This intrinsic corresponds to the <c> CRC32B </c> instruction.
+///
+/// \param __C
+///    An unsigned integer operand to add to the CRC-32C checksum of operand
+///    \a  __D.
+/// \param __D
+///    An unsigned 8-bit integer operand used to compute the CRC-32C checksum.
+/// \returns The result of adding operand \a __C to the CRC-32C checksum of
+///    operand \a __D.
+static __inline__ unsigned int __DEFAULT_FN_ATTRS
+_mm_crc32_u8(unsigned int __C, unsigned char __D) {
+  return __builtin_ia32_crc32qi(__C, __D);
+}
+
+/// Adds the unsigned integer operand to the CRC-32C checksum of the
+///    unsigned short operand.
+///
+/// \headerfile <x86intrin.h>
+///
+/// This intrinsic corresponds to the <c> CRC32W </c> instruction.
+///
+/// \param __C
+///    An unsigned integer operand to add to the CRC-32C checksum of operand
+///    \a __D.
+/// \param __D
+///    An unsigned 16-bit integer operand used to compute the CRC-32C checksum.
+/// \returns The result of adding operand \a __C to the CRC-32C checksum of
+///    operand \a __D.
+static __inline__ unsigned int __DEFAULT_FN_ATTRS
+_mm_crc32_u16(unsigned int __C, unsigned short __D) {
+  return __builtin_ia32_crc32hi(__C, __D);
+}
+
+/// Adds the first unsigned integer operand to the CRC-32C checksum of
+///    the second unsigned integer operand.
+///
+/// \headerfile <x86intrin.h>
+///
+/// This intrinsic corresponds to the <c> CRC32L </c> instruction.
+///
+/// \param __C
+///    An unsigned integer operand to add to the CRC-32C checksum of operand
+///    \a __D.
+/// \param __D
+///    An unsigned 32-bit integer operand used to compute the CRC-32C checksum.
+/// \returns The result of adding operand \a __C to the CRC-32C checksum of
+///    operand \a __D.
+static __inline__ unsigned int __DEFAULT_FN_ATTRS
+_mm_crc32_u32(unsigned int __C, unsigned int __D) {
+  return __builtin_ia32_crc32si(__C, __D);
+}
+
+#ifdef __x86_64__
+/// Adds the unsigned integer operand to the CRC-32C checksum of the
+///    unsigned 64-bit integer operand.
+///
+/// \headerfile <x86intrin.h>
+///
+/// This intrinsic corresponds to the <c> CRC32Q </c> instruction.
+///
+/// \param __C
+///    An unsigned integer operand to add to the CRC-32C checksum of operand
+///    \a __D.
+/// \param __D
+///    An unsigned 64-bit integer operand used to compute the CRC-32C checksum.
+/// \returns The result of adding operand \a __C to the CRC-32C checksum of
+///    operand \a __D.
+static __inline__ unsigned long long __DEFAULT_FN_ATTRS
+_mm_crc32_u64(unsigned long long __C, unsigned long long __D) {
+  return __builtin_ia32_crc32di(__C, __D);
+}
+#endif /* __x86_64__ */
+
+#undef __DEFAULT_FN_ATTRS
+
+#endif /* __CRC32INTRIN_H */

package/include/cuda_wrappers/__utility/declval.h

@@ -0,0 +1,28 @@
+#ifndef __CUDA_WRAPPERS_UTILITY_DECLVAL_H__
+#define __CUDA_WRAPPERS_UTILITY_DECLVAL_H__
+
+#include_next <__utility/declval.h>
+
+// The stuff below is the exact copy of the <__utility/declval.h>,
+// but with __device__ attribute applied to the functions, so it works on a GPU.
+
+_LIBCPP_BEGIN_NAMESPACE_STD
+
+// Suppress deprecation notice for volatile-qualified return type resulting
+// from volatile-qualified types _Tp.
+_LIBCPP_SUPPRESS_DEPRECATED_PUSH
+template <class _Tp> __attribute__((device)) _Tp &&__declval(int);
+template <class _Tp> __attribute__((device)) _Tp __declval(long);
+_LIBCPP_SUPPRESS_DEPRECATED_POP
+
+template <class _Tp>
+__attribute__((device)) _LIBCPP_HIDE_FROM_ABI decltype(std::__declval<_Tp>(0))
+declval() _NOEXCEPT {
+  static_assert(!__is_same(_Tp, _Tp),
+                "std::declval can only be used in an unevaluated context. "
+                "It's likely that your current usage is trying to extract a "
+                "value from the function.");
+}
+
+_LIBCPP_END_NAMESPACE_STD
+#endif // __CUDA_WRAPPERS_UTILITY_DECLVAL_H__