[BUG] CUTE SGEMM Example Doesn't Compile

[abhishekvijeev]

Hi,

The CUTE example in examples/cute/tutorial/sgemm_1.cu doesn't compile. I receive the following error:

src/main.cu(104): error: identifier "X" is undefined
    Tensor gA = local_tile(mA, cta_tiler, cta_coord, Step<_1, X,_1>{});
                                                              ^

./cutlass/include/cute/container/tuple.hpp(175): error: incomplete type is not allowed
  struct tuple : detail::TupleBase<make_index_sequence<sizeof...(T)>, T...>
                 ^
          detected during instantiation of class "cute::tuple<T...> [with T=<cute::_1, <error-type>, cute::_1>]" at line 104 of src/main.cu

src/main.cu(105): error: identifier "X" is undefined
    Tensor gB = local_tile(mB, cta_tiler, cta_coord, Step< X,_1,_1>{});
                                                           ^

./cutlass/include/cute/container/tuple.hpp(175): error: incomplete type is not allowed
  struct tuple : detail::TupleBase<make_index_sequence<sizeof...(T)>, T...>
                 ^
          detected during instantiation of class "cute::tuple<T...> [with T=<<error-type>, cute::_1, cute::_1>]" at line 105 of src/main.cu

src/main.cu(106): error: identifier "X" is undefined
    Tensor gC = local_tile(mC, cta_tiler, cta_coord, Step<_1,_1, X>{});
                                                                 ^

./cutlass/include/cute/container/tuple.hpp(175): error: incomplete type is not allowed
  struct tuple : detail::TupleBase<make_index_sequence<sizeof...(T)>, T...>
                 ^
          detected during instantiation of class "cute::tuple<T...> [with T=<cute::_1, cute::_1, <error-type>>]" at line 106 of src/main.cu

src/main.cu(138): error: identifier "X" is undefined
    Tensor tCsA = local_partition(sA, tC, threadIdx.x, Step<_1, X>{});
                                                                ^

./cutlass/include/cute/container/tuple.hpp(175): error: incomplete type is not allowed
  struct tuple : detail::TupleBase<make_index_sequence<sizeof...(T)>, T...>
                 ^
          detected during instantiation of class "cute::tuple<T...> [with T=<cute::_1, <error-type>>]" at line 138 of src/main.cu

src/main.cu(140): error: identifier "X" is undefined
    Tensor tCsB = local_partition(sB, tC, threadIdx.x, Step< X,_1>{});
                                                             ^

./cutlass/include/cute/container/tuple.hpp(175): error: incomplete type is not allowed
  struct tuple : detail::TupleBase<make_index_sequence<sizeof...(T)>, T...>
                 ^
          detected during instantiation of class "cute::tuple<T...> [with T=<<error-type>, cute::_1>]" at line 140 of src/main.cu

Has CUTE's API changed such that the placeholder X no longer works?

Thanks!

[thakkarV]

Can you please include your version and toolchain info and repro steps?

[abhishekvijeev]

Sure. I'm using the main branch of CUTLASS and have created a repo with instructions to reproduce the issue:

https://github.com/abhishekvijeev/cute_gemm_test/tree/main

Here's the output of nvcc --version:

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2024 NVIDIA Corporation
Built on Fri_Jun_14_16:34:21_PDT_2024
Cuda compilation tools, release 12.6, V12.6.20
Build cuda_12.6.r12.6/compiler.34431801_0

Please let me know if any additional information is needed.

Thanks!

[thakkarV]

Your cmake setup is incorrect. You need to add the cutlass dir as a subdir in your cmake config

[abhishekvijeev]

Ah I see, thank you!

[abhishekvijeev]

Hi,

I added the cutlass directory as a subdirectory using add_subdirectory(./cutlass) - that doesn't seem to fix the issue. Did you mean something else?

Thanks!

[abhishekvijeev]

Oddly enough, using Underscore instead of X works fine!