[BUG]: /usr/local/include/cuda/std/__type_traits/integral_constant.h(26): error: "integral_constant" has already been declared in the current scope

[linuxlonelyeagle]

Is this a duplicate?

• I confirmed there appear to be no duplicate issues for this bug and that I agree to the Code of Conduct

Type of Bug

Compile-time Error

Component

libcu++

Describe the bug

I found it cann't include "cuda/ptx", so I install cccl from github.I found following issue.I hope someone can help me.Thanks all.

my code

#include "cuda_runtime.h"
#include "cuda/barrier"
#include <cuda/std/atomic>
#include <cuda/ptx>

using barrier = cuda::barrier<cuda::thread_scope_block>;
using ptx = cuda::ptx;

int main() {
  return 0;
}

when i compile it.

~/test # nvcc tma.cu  -arch=sm_80                                                              2 ↵ root@dbc0cf650200
nv/usr/local/include/cuda/std/__type_traits/integral_constant.h(26): error: "integral_constant" has already been declared in the current scope
  struct _CCCL_TYPE_VISIBILITY_DEFAULT integral_constant
                                       ^

/usr/local/include/cuda/std/__type_traits/integral_constant.h(26): error: incomplete type is not allowed
  struct _CCCL_TYPE_VISIBILITY_DEFAULT integral_constant
                                       ^

/usr/local/include/cuda/std/__type_traits/integral_constant.h(28): error: expected an expression
    static constexpr const _Tp value = __v;
    ^

/usr/local/include/cuda/std/__type_traits/integral_constant.h(28): error: expected a "}"
    static constexpr const _Tp value = __v;
                                          ^

/usr/local/include/cuda/std/__type_traits/integral_constant.h(29): error: identifier "_Tp" is undefined
    typedef _Tp value_type;
            ^

/usr/local/include/cuda/std/__type_traits/integral_constant.h(30): error: argument list for variable template "cuda::std::__4::integral_constant" is missing
    typedef integral_constant type;
            ^

/usr/local/include/cuda/std/__type_traits/integral_constant.h(31): error: conversion function must be a nonstatic member function
    __attribute__ ((__visibility__("hidden"))) __attribute__ ((__always_inline__)) constexpr operator value_type() const noexcept
                                                                                             ^

/usr/local/include/cuda/std/__type_traits/integral_constant.h(31): error: a type qualifier is not allowed on a nonmember function
    __attribute__ ((__visibility__("hidden"))) __attribute__ ((__always_inline__)) constexpr operator value_type() const noexcept
                                                                                                                   ^

/usr/local/include/cuda/std/__type_traits/integral_constant.h(33): error: identifier "value" is undefined
      return value;
             ^

/usr/local/include/cuda/std/__type_traits/integral_constant.h(35): error: "operator()" must be a member function
    __attribute__ ((__visibility__("hidden"))) __attribute__ ((__always_inline__)) constexpr value_type operator()() const noexcept
                                                                                                        ^

/usr/local/include/cuda/std/__type_traits/integral_constant.h(35): error: a type qualifier is not allowed on a nonmember function
    __attribute__ ((__visibility__("hidden"))) __attribute__ ((__always_inline__)) constexpr value_type operator()() const noexcept
                                                                                                                     ^

/usr/local/include/cuda/std/__type_traits/integral_constant.h(37): error: identifier "value" is undefined
      return value;
             ^

/usr/local/include/cuda/std/__type_traits/integral_constant.h(42): error: "cuda::std::integral_constant" is ambiguous
  constexpr const _Tp integral_constant<_Tp, __v>::value;
                      ^

/usr/local/include/cuda/std/__type_traits/integral_constant.h(42): error: "constexpr" is not valid here
  constexpr const _Tp integral_constant<_Tp, __v>::value;
  ^

/usr/local/include/cuda/std/__type_traits/integral_constant.h(44): error: "cuda::std::integral_constant" is ambiguous
  typedef integral_constant<bool, true> true_type;
          ^

/usr/local/include/cuda/std/__type_traits/integral_constant.h(45): error: "cuda::std::integral_constant" is ambiguous
  typedef integral_constant<bool, false> false_type;
          ^

/usr/local/include/cuda/std/__type_traits/integral_constant.h(48): error: expected an "="
  using _BoolConstant __attribute__ ((deprecated)) _CCCL_NODEBUG_ALIAS = integral_constant<bool, _Val>;
                                                   ^

/usr/local/include/cuda/std/__type_traits/integral_constant.h(48): error: identifier "_CCCL_NODEBUG_ALIAS" is undefined
  using _BoolConstant __attribute__ ((deprecated)) _CCCL_NODEBUG_ALIAS = integral_constant<bool, _Val>;
                                                   ^

/usr/local/include/cuda/std/__type_traits/integral_constant.h(48): error: expected a ";"
  using _BoolConstant __attribute__ ((deprecated)) _CCCL_NODEBUG_ALIAS = integral_constant<bool, _Val>;
                                                                       ^

/usr/local/include/cuda/std/__type_traits/integral_constant.h(51): error: "integral_constant" is ambiguous
  using bool_constant = integral_constant<bool, __b>;
                        ^

/usr/local/include/cuda/std/__type_traits/integral_constant.h(56): error: expected a declaration
  } } }
      ^

/usr/local/include/cuda/__ptx/ptx_dot_variants.h(133): error: identifier "dot_sem" is undefined
  template <dot_sem __sem>
            ^

/usr/local/include/cuda/__ptx/ptx_dot_variants.h(134): error: "sem_t" has already been declared in the current scope
  using sem_t = 
        ^

/usr/local/include/cuda/__ptx/ptx_dot_variants.h(134): error: identifier "dot_sem" is undefined
                                 ::integral_constant<dot_sem, __sem>;
                                                     ^

/usr/local/include/cuda/__ptx/ptx_dot_variants.h(134): error: variable "cuda::std::__4::integral_constant [with _Tp=<error-type>, __v=<error>]" is not a type name
                       cuda::std::__4
                       ^

/usr/local/include/cuda/__ptx/ptx_dot_variants.h(135): error: name followed by "::" must be a class or namespace name
  using sem_acq_rel_t = sem_t<dot_sem::acq_rel>;
                              ^

/usr/local/include/cuda/__ptx/ptx_dot_variants.h(136): error: name followed by "::" must be a class or namespace name
  using sem_acquire_t = sem_t<dot_sem::acquire>;
                              ^

/usr/local/include/cuda/__ptx/ptx_dot_variants.h(137): error: name followed by "::" must be a class or namespace name
  using sem_relaxed_t = sem_t<dot_sem::relaxed>;
                              ^

/usr/local/include/cuda/__ptx/ptx_dot_variants.h(138): error: name followed by "::" must be a class or namespace name
  using sem_release_t = sem_t<dot_sem::release>;
                              ^

/usr/local/include/cuda/__ptx/ptx_dot_variants.h(139): error: name followed by "::" must be a class or namespace name
  using sem_sc_t = sem_t<dot_sem::sc>;
                         ^

/usr/local/include/cuda/__ptx/ptx_dot_variants.h(140): error: name followed by "::" must be a class or namespace name
  using sem_weak_t = sem_t<dot_sem::weak>;
                           ^

/usr/local/include/cuda/__ptx/ptx_dot_variants.h(150): error: variable "cuda::std::__4::integral_constant [with _Tp=dot_space, __v=__spc]" is not a type name
                         cuda::std::__4
                         ^

/usr/local/include/cuda/__ptx/ptx_dot_variants.h(160): error: variable "cuda::std::__4::integral_constant [with _Tp=dot_scope, __v=__scope]" is not a type name
                         cuda::std::__4
                         ^

/usr/local/include/cuda/__ptx/ptx_dot_variants.h(172): error: variable "cuda::std::__4::integral_constant [with _Tp=dot_op, __v=__op]" is not a type name
                     cuda::std::__4
                     ^

/usr/local/include/cuda/__ptx/ptx_dot_variants.h(196): error: variable "cuda::std::__4::integral_constant [with _Tp=dot_cta_group, __v=__cta_group]" is not a type name
                       cuda::std::__4
                       ^

/usr/local/include/cuda/__ptx/ptx_dot_variants.h(204): error: variable "cuda::std::__4::integral_constant [with _Tp=dot_kind, __v=__kind]" is not a type name
                         cuda::std::__4
                         ^

/usr/local/include/cuda/__ptx/ptx_dot_variants.h(222): error: variable "cuda::std::__4::integral_constant [with _Tp=int, __v=n]" is not a type name
               cuda::std::__4
               ^

/usr/local/include/cuda/std/__type_traits/enable_if.h(23): error: expected a ";"
  namespace cuda { namespace std { inline namespace __4 {
  ^

tma.cu(6): warning #12-D: parsing restarts here after previous syntax error
  using barrier = cuda::barrier<cuda::thread_scope_block>;
                                                         ^

Remark: The warnings can be suppressed with "-diag-suppress <warning-number>"

tma.cu(7): error: namespace "cuda" has no member "ptx"
  using ptx = cuda::ptx;

How to Reproduce

The above is enough.

Expected behavior

compile complete.

Reproduction link

No response

Operating System

No response

nvidia-smi output

Sat Mar 15 14:13:32 2025
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.54.15 Driver Version: 550.54.15 CUDA Version: 12.4 |
|-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 NVIDIA H100 PCIe On | 00000000:A1:00.0 Off | 0 |
| N/A 28C P0 77W / 350W | 485MiB / 81559MiB | 0% Default |
| | | Disabled |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=========================================================================================|
+-----------------------------------------------------------------------------------------+

NVCC version

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2023 NVIDIA Corporation
Built on Mon_Apr__3_17:16:06_PDT_2023
Cuda compilation tools, release 12.1, V12.1.105
Build cuda_12.1.r12.1/compiler.32688072_0

[miscco]

This is a bug in your setup:

#include "cuda/barrier"
#include <cuda/std/atomic>
#include <cuda/ptx>

I assume that you are trying to get CCCL from github, but you are once including from system path via <cuda/...> and once from local path via "cuda/barrier".

That most likely pulls in two different versions of CCCL, the one from the CTK installed and the one from github.