[BUG] Compile error from including `"cute/tensor.hpp"` when building with MSVC with `/permissive-` option

Which component has the problem?

CUTLASS C++

Bug Report

Describe the bug
When compiling for C++20 with MSVC and including "cute/tensor.hpp", I observed the following compile error:

D:\source\cutlass\include\cute\stride.hpp(299,46): error C3545: 'Ints': parameter pack expects a non-type template argument [D:\test\cute_cpp_test\build\cute_cpp_test_program.vcxproj]
  (compiling source file '../main.cc')
      D:\source\cutlass\include\cute\stride.hpp(299,46):
      the template instantiation context (the oldest one first) is
          D:\source\cutlass\include\cute\atom\mma_traits.hpp(70,8):
          while compiling class template partial specialization 'cute::MMA_Traits<cute::UniversalFMA<D,A,B,C>,>'
          D:\source\cutlass\include\cute\atom\mma_traits.hpp(86,35):
          while processing the default template argument of 'cute::Layout '
          D:\source\cutlass\include\cute\layout.hpp(98,51):
          see reference to alias template instantiation 'cute::LayoutLeft::Apply<Shape>' being compiled
          with
          [
              Shape=cute::tuple<cute::_1,cute::_1>
          ]
          D:\source\cutlass\include\cute\stride.hpp(369,26):
          see reference to function template instantiation 'auto cute::compact_major<cute::LayoutLeft,Shape,cute::C<1>,0x0>(const Shape &,const Current &)' being compiled
          with
          [
              Shape=cute::tuple<cute::_1,cute::_1>,
              Current=cute::C<1>
          ]
          D:\source\cutlass\include\cute\stride.hpp(357,27):
          see reference to function template instantiation 'auto cute::detail::compact<cute::LayoutLeft,Shape,Current>(const Shape &,const Current &)' being compiled
          with
          [
              Shape=cute::tuple<cute::_1,cute::_1>,
              Current=cute::C<1>
          ]

D:\source\cutlass\include\cute\atom\mma_traits.hpp(86,19): error C2976: 'cute::Layout': too few template arguments [D:\test\cute_cpp_test\build\cute_cpp_test_program.vcxproj]
  (compiling source file '../main.cc')
      D:\source\cutlass\include\cute\layout.hpp(99,8):
      see declaration of 'cute::Layout'

D:\source\cutlass\include\cute\atom\mma_traits_sm61.hpp(51,19): error C2976: 'cute::Layout': too few template arguments [D:\test\cute_cpp_test\build\cute_cpp_test_program.vcxproj]
  (compiling source file '../main.cc')
      D:\source\cutlass\include\cute\layout.hpp(99,8):
      see declaration of 'cute::Layout'

D:\source\cutlass\include\cute\atom\mma_traits_sm61.hpp(68,19): error C2976: 'cute::Layout': too few template arguments [D:\test\cute_cpp_test\build\cute_cpp_test_program.vcxproj]
  (compiling source file '../main.cc')
      D:\source\cutlass\include\cute\layout.hpp(99,8):
      see declaration of 'cute::Layout'

D:\source\cutlass\include\cute\atom\copy_traits_sm100.hpp(53,21): error C2976: 'cute::Layout': too few template arguments [D:\test\cute_cpp_test\build\cute_cpp_test_program.vcxproj]
  (compiling source file '../main.cc')
      D:\source\cutlass\include\cute\layout.hpp(99,8):
      see declaration of 'cute::Layout'

There is also a similar error when compiling for C++17 with MSVC with the /permissive- option.

Steps/Code to reproduce bug

Create the following files:

CMakeLists.txt:

cmake_minimum_required(VERSION 3.28)

project(cute_cpp_test CXX)

set(USE_CXX_20 OFF)

if (USE_CXX_20)
    # C++20 does not work
    set(CMAKE_CXX_STANDARD 20)
else()
    # C++17 and /permissive- also does not work
    set(CMAKE_CXX_STANDARD 17)
    add_compile_options("/permissive-")
endif()

add_executable(cute_cpp_test_program ${CMAKE_CURRENT_SOURCE_DIR}/main.cc)

set(CUTLASS_INCLUDE_DIR "D:/source/cutlass/include")
set(CUDA_INCLUDE_DIR "C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.4/include")

target_include_directories(cute_cpp_test_program PRIVATE ${CUTLASS_INCLUDE_DIR} ${CUDA_INCLUDE_DIR})

main.cc:

#include "cute/tensor.hpp"

int main() {
    return 0;
}

From the directory containing those files:

cmake -B ./build -S . -G "Visual Studio 18 2026"
cmake --build ./build --config Debug

Expected behavior
The code builds successfully.

Environment details (please complete the following information):

• Environment location: Local Windows 11 Machine
• Compiler: MSVC 19.50.35724.0
• CUTLASS commit: 3476ddb7bd6ca4161a0169103ceaa20ce0eb891f

Additional context
I encountered this issue while trying to update an existing codebase to C++20.

The docs here seem to imply that CUTLASS should be usable from C++20 code.

Applications should list /include within their include paths. They must be compiled as C++17 or greater.

Upon further testing based on the suggestion here, it appears that this is also an issue for C++17.

Reproduce the issue with the provided CMakeLists.txt and main.cc using MSVC, C++17 with /permissive-, and C++20. Start in include/cute/stride.hpp and follow the reported instantiation through include/cute/layout.hpp and include/cute/atom/mma_traits.hpp. Done means including cute/tensor.hpp compiles successfully in both configurations.