[BUG] Example 63 requires C++20 support

Which component has the problem?

CUTLASS C++

Bug Report

Describe the bug
Compilation using NVHPC 25.9 fails with the following error:

/e/project1/jureap5/reuter8/raps/develop/external/cutlass/examples/63_hopper_gemm_with_weight_prefetch/collective/sm90_mma_tma_gmma_ss_warpspeciali
zed_with_prefetch.hpp: In instantiation of ‘constexpr const auto cutlass::gemm::collective::CollectiveMma<cutlass::gemm::MainloopSm90TmaGmmaWarpSpe
cializedWithPrefetch<13, cute::tuple<cute::C<1>, cute::C<1>, cute::C<1> >, cutlass::gemm::KernelTmaWarpSpecializedFP8FastAccumWithPrefetchAndSplitD
MA>, cute::tuple<cute::C<64>, cute::C<64>, cute::C<128> >, cutlass::float_e4m3_t, cute::tuple<long int, cute::C<1>, long int>, cutlass::float_e5m2_
t, cute::tuple<long int, cute::C<1>, long int>, cute::TiledMMA<cute::MMA_Atom<cute::SM90::GMMA::MMA_64x64x32_F32E4M3E5M2_SS_TN<cute::SM90::GMMA::Sc
aleIn::One, cute::SM90::GMMA::ScaleIn::One> >, cute::Layout<cute::tuple<cute::C<1>, cute::C<1>, cute::C<1> > >, cute::tuple<cute::Underscore, cute:
:Underscore, cute::Underscore> >, cute::SM90_TMA_LOAD, cute::ComposedLayout<cute::Swizzle<3, 4, 3>, cute::smem_ptr_flag_bits<8>, cute::Layout<cute:
:tuple<cute::C<8>, cute::C<128> >, cute::tuple<cute::C<128>, cute::C<1> > > >, void, cute::identity, cute::SM90_TMA_LOAD, cute::ComposedLayout<cute
::Swizzle<3, 4, 3>, cute::smem_ptr_flag_bits<8>, cute::Layout<cute::tuple<cute::C<8>, cute::C<128> >, cute::tuple<cute::C<128>, cute::C<1> > > >, v
oid, cute::identity>::prefetch_smem_size’:
/e/project1/jureap5/reuter8/raps/develop/external/cutlass/examples/63_hopper_gemm_with_weight_prefetch/collective/sm90_mma_tma_gmma_ss_warpspeciali
zed_with_prefetch.hpp:164:30:   required from ‘struct cutlass::gemm::collective::CollectiveMma<cutlass::gemm::MainloopSm90TmaGmmaWarpSpecializedWit
hPrefetch<13, cute::tuple<cute::C<1>, cute::C<1>, cute::C<1> >, cutlass::gemm::KernelTmaWarpSpecializedFP8FastAccumWithPrefetchAndSplitDMA>, cute::
tuple<cute::C<64>, cute::C<64>, cute::C<128> >, cutlass::float_e4m3_t, cute::tuple<long int, cute::C<1>, long int>, cutlass::float_e5m2_t, cute::tu
ple<long int, cute::C<1>, long int>, cute::TiledMMA<cute::MMA_Atom<cute::SM90::GMMA::MMA_64x64x32_F32E4M3E5M2_SS_TN<cute::SM90::GMMA::ScaleIn::One,
 cute::SM90::GMMA::ScaleIn::One> >, cute::Layout<cute::tuple<cute::C<1>, cute::C<1>, cute::C<1> > >, cute::tuple<cute::Underscore, cute::Underscore
, cute::Underscore> >, cute::SM90_TMA_LOAD, cute::ComposedLayout<cute::Swizzle<3, 4, 3>, cute::smem_ptr_flag_bits<8>, cute::Layout<cute::tuple<cute
::C<8>, cute::C<128> >, cute::tuple<cute::C<128>, cute::C<1> > > >, void, cute::identity, cute::SM90_TMA_LOAD, cute::ComposedLayout<cute::Swizzle<3
, 4, 3>, cute::smem_ptr_flag_bits<8>, cute::Layout<cute::tuple<cute::C<8>, cute::C<128> >, cute::tuple<cute::C<128>, cute::C<1> > > >, void, cute::
identity>’
  164 |   static constexpr auto prefetch_smem_size = cute::cosize_v<PrefetchSmemLayoutA>;
      |                              ^~~~~~~~~~~~~~~~~~
/e/project1/jureap5/reuter8/raps/develop/external/cutlass/include/cutlass/gemm/device/gemm_universal_adapter.h:630:137:   recursively required by s
ubstitution of ‘template<class GemmKernel> struct cutlass::gemm::detail::IsCutlass3GemmKernel<GemmKernel, std::void_t<typename GemmKernel::ProblemS
hape> > [with GemmKernel = cutlass::gemm::kernel::GemmUniversal<cute::tuple<int, int, int, int>, cutlass::gemm::collective::CollectiveMma<cutlass::
gemm::MainloopSm90TmaGmmaWarpSpecializedWithPrefetch<13, cute::tuple<cute::C<1>, cute::C<1>, cute::C<1> >, cutlass::gemm::KernelTmaWarpSpecializedF
P8FastAccumWithPrefetchAndSplitDMA>, cute::tuple<cute::C<64>, cute::C<64>, cute::C<128> >, cutlass::float_e4m3_t, cute::tuple<long int, cute::C<1>,
 long int>, cutlass::float_e5m2_t, cute::tuple<long int, cute::C<1>, long int>, cute::TiledMMA<cute::MMA_Atom<cute::SM90::GMMA::MMA_64x64x32_F32E4M
3E5M2_SS_TN<cute::SM90::GMMA::ScaleIn::One, cute::SM90::GMMA::ScaleIn::One> >, cute::Layout<cute::tuple<cute::C<1>, cute::C<1>, cute::C<1> > >, cut
e::tuple<cute::Underscore, cute::Underscore, cute::Underscore> >, cute::SM90_TMA_LOAD, cute::ComposedLayout<cute::Swizzle<3, 4, 3>, cute::smem_ptr_
flag_bits<8>, cute::Layout<cute::tuple<cute::C<8>, cute::C<128> >, cute::tuple<cute::C<128>, cute::C<1> > > >, void, cute::identity, cute::SM90_TMA
_LOAD, cute::ComposedLayout<cute::Swizzle<3, 4, 3>, cute::smem_ptr_flag_bits<8>, cute::Layout<cute::tuple<cute::C<8>, cute::C<128> >, cute::tuple<c
ute::C<128>, cute::C<1> > > >, void, cute::identity>, cutlass::epilogue::collective::CollectiveEpilogue<cutlass::epilogue::Sm90TmaWarpSpecialized<1
, 1, 32, false, false>, cute::tuple<cute::C<64>, cute::C<64>, cute::C<128> >, cute::tuple<cute::C<64>, cute::C<64> >, cutlass::float_e4m3_t, cute::
tuple<cute::C<1>, long int, long int>, cutlass::float_e4m3_t, cute::tuple<cute::C<1>, long int, long int>, cutlass::epilogue::fusion::FusionCallbac
ks<cutlass::epilogue::Sm90TmaWarpSpecialized<1, 1, 32, false, false>, cutlass::epilogue::fusion::LinearCombination<cutlass::float_e4m3_t, float, cu
tlass::float_e4m3_t, float, cutlass::FloatRoundStyle::round_to_nearest>, cute::tuple<cute::C<64>, cute::C<64>, cute::C<128> >, cute::tuple<cute::C<
64>, cute::C<64> > >, cute::SM90_TMA_LOAD, cute::ComposedLayout<cute::Swizzle<2, 4, 3>, cute::smem_ptr_flag_bits<8>, cute::Layout<cute::tuple<cute:
:C<64>, cute::C<8> >, cute::tuple<cute::C<1>, cute::C<64> > > >, cute::AutoVectorizingCopyWithAssumedAlignment<128>, cute::SM90_TMA_STORE, cute::Co
mposedLayout<cute::Swizzle<2, 4, 3>, cute::smem_ptr_flag_bits<8>, cute::Layout<cute::tuple<cute::C<64>, cute::C<8> >, cute::tuple<cute::C<1>, cute:
:C<64> > > >, cute::AutoVectorizingCopyWithAssumedAlignment<128>, cute::Copy_Atom<cute::SM90_U32x4_STSM_N, cutlass::half_t>, void>, void>]’
  630 | class GemmUniversalAdapter<
      |

/e/project1/jureap5/reuter8/raps/develop/external/cutlass/include/cutlass/gemm/device/gemm_universal_adapter.h:630:137:   required by substitution
of ‘template<class GemmKernel_> class cutlass::gemm::device::GemmUniversalAdapter<GemmKernel_, typename std::enable_if<(! cutlass::gemm::detail::IsCutlass3GemmKernel<typename cutlass::GetUnderlyingKernel<T>::type>::value), void>::type> [with GemmKernel_ = cutlass::gemm::kernel::GemmUniversal<cute::tuple<int, int, int, int>, cutlass::gemm::collective::CollectiveMma<cutlass::gemm::MainloopSm90TmaGmmaWarpSpecializedWithPrefetch<13, cute::tuple<cute::C<1>, cute::C<1>, cute::C<1> >, cutlass::gemm::KernelTmaWarpSpecializedFP8FastAccumWithPrefetchAndSplitDMA>, cute::tuple<cute::C<64>, cute::C<64>, cute::C<128> >, cutlass::float_e4m3_t, cute::tuple<long int, cute::C<1>, long int>, cutlass::float_e5m2_t, cute::tuple<long int, cute::C<1>, long int>, cute::TiledMMA<cute::MMA_Atom<cute::SM90::GMMA::MMA_64x64x32_F32E4M3E5M2_SS_TN<cute::SM90::GMMA::ScaleIn::One, cute::SM90::GMMA::ScaleIn::One> >, cute::Layout<cute::tuple<cute::C<1>, cute::C<1>, cute::C<1> > >, cute::tuple<cute::Underscore, cute::Underscore, cute::Underscore> >, cute::SM90_TMA_LOAD, cute::ComposedLayout<cute::Swizzle<3, 4, 3>, cute::smem_ptr_flag_bits<8>, cute::Layout<cute::tuple<cute::C<8>, cute::C<128> >, cute::tuple<cute::C<128>, cute::C<1> > > >, void, cute::identity, cute::SM90_TMA_LOAD, cute::ComposedLayout<cute::Swizzle<3, 4, 3>, cute::smem_ptr_flag_bits<8>, cute::Layout<cute::tuple<cute::C<8>, cute::C<128> >, cute::tuple<cute::C<128>, cute::C<1> > > >, void, cute::identity>, cutlass::epilogue::collective::CollectiveEpilogue<cutlass::epilogue::Sm90TmaWarpSpecialized<1, 1, 32, false, false>, cute::tuple<cute::C<64>, cute::C<64>, cute::C<128> >, cute::tuple<cute::C<64>, cute::C<64> >, cutlass::float_e4m3_t, cute::tuple<cute::C<1>, long int, long int>, cutlass::float_e4m3_t, cute::tuple<cute::C<1>, long int, long int>, cutlass::epilogue::fusion::FusionCallbacks<cutlass::epilogue::Sm90TmaWarpSpecialized<1, 1, 32, false, false>, cutlass::epilogue::fusion::LinearCombination<cutlass::float_e4m3_t, float, cutlass::float_e4m3_t, float, cutlass::FloatRoundStyle::round_to_nearest>, cute::tuple<cute::C<64>, cute::C<64>, cute::C<128> >, cute::tuple<cute::C<64>, cute::C<64> > >, cute::SM90_TMA_LOAD, cute::ComposedLayout<cute::Swizzle<2, 4, 3>, cute::smem_ptr_flag_bits<8>, cute::Layout<cute::tuple<cute::C<64>, cute::C<8> >, cute::tuple<cute::C<1>, cute::C<64> > > >, cute::AutoVectorizingCopyWithAssumedAlignment<128>, cute::SM90_TMA_STORE, cute::ComposedLayout<cute::Swizzle<2, 4, 3>, cute::smem_ptr_flag_bits<8>, cute::Layout<cute::tuple<cute::C<64>, cute::C<8> >, cute::tuple<cute::C<1>, cute::C<64> > > >, cute::AutoVectorizingCopyWithAssumedAlignment<128>, cute::Copy_Atom<cute::SM90_U32x4_STSM_N, cutlass::half_t>, void>, void>]’
/e/project1/jureap5/reuter8/raps/develop/external/cutlass/examples/63_hopper_gemm_with_weight_prefetch/63_hopper_gemm_with_weight_prefetch.cu:199:0:   required from here
  199 | using EpilogueOutputOp  = typename Gemm::EpilogueOutputOp;
/e/project1/jureap5/reuter8/raps/develop/external/cutlass/examples/63_hopper_gemm_with_weight_prefetch/collective/sm90_mma_tma_gmma_ss_warpspecialized_with_prefetch.hpp:159:63: error: non-type template parameters of class type only available with ‘-std=c++20’ or ‘-std=gnu++20’
  159 |   using PrefetchSmemLayoutA = decltype(make_layout(make_shape(
      |                                                     ~~~~~~~~~~^
/e/project1/jureap5/reuter8/raps/develop/external/cutlass/examples/63_hopper_gemm_with_weight_prefetch/collective/sm90_mma_tma_gmma_ss_warpspecialized_with_prefetch.hpp:159:63: error: non-type template parameters of class type only available with ‘-std=c++20’ or ‘-std=gnu++20’
gmake[2]: *** [examples/63_hopper_gemm_with_weight_prefetch/CMakeFiles/63_hopper_gemm_with_weight_prefetch.dir/build.make:80: examples/63_hopper_gemm_with_weight_prefetch/CMakeFiles/63_hopper_gemm_with_weight_prefetch.dir/63_hopper_gemm_with_weight_prefetch.cu.o] Error 1
gmake[1]: *** [CMakeFiles/Makefile2:61836: examples/63_hopper_gemm_with_weight_prefetch/CMakeFiles/63_hopper_gemm_with_weight_prefetch.dir/all] Error 2
gmake[1]: *** Waiting for unfinished jobs....

Steps/Code to reproduce bug

Compile Cutlass v4.3.5 using the following commands:

export FC=nvfortran
export CC=nvc
export CXX=nvc++
cmake -DCMAKE_INSTALL_PREFIX=$installdir \
          -DCMAKE_C_COMPILER=$CC \
          -DCMAKE_CXX_COMPILER=$CXX \
          -DCMAKE_C_COMPILER_FLAGS="-O2 -fPIC" \
          -DCMAKE_CXX_COMPILER_FLAGS="-O2 -fPIC" \
          -DCUTLASS_NVCC_ARCHS="90" \
      -S $srcdir -B $builddir

cmake --build $builddir -j ${NCPUS:-32}

Expected behavior
Successful build. I managed to work around it by hacking the build flags to use -std=c++20 in $builddir/examples/63_hopper_gemm_with_weight_prefetch/CMakeFiles/63_hopper_gemm_with_weight_prefetch.dir/flags.make

Environment details (please complete the following information):
JSC JUPITER Supercomputer. NVHPC 25.9.

Additional context
Add any other context about the problem here.

Start with examples/63_hopper_gemm_with_weight_prefetch/63_hopper_gemm_with_weight_prefetch.cu and collective/sm90_mma_tma_gmma_ss_warpspecialized_with_prefetch.hpp, then inspect the generated flags.make from the reported CMake build. Reproduce with NVHPC 25.9 and verify that the example builds without manually adding -std=c++20; the C++20 requirement should be handled by the project configuration.