3.6.0 update (#2005)

* 3.6.0 update

* doc and swap stuff

---------

Co-authored-by: yuzhai <yuzhai@nvidia.com>
Co-authored-by: Haicheng Wu <haichengw@nvidia.com>

test/unit/transform/CMakeLists.txt

@@ -33,7 +33,7 @@ add_custom_target(
   cutlass_test_unit_transform
   DEPENDS
   cutlass_test_unit_transform_threadblock
-  cutlass_test_unit_transform_filter_format
+  cutlass_test_unit_transform_kernel
 )
 
 add_custom_target(

test/unit/transform/device/sm90_sparse_gemm_compressor_legacy.hpp

@@ -45,6 +45,7 @@
 #include "cute/tensor.hpp"                 // cute::Tensor, cute::make_tensor, cute::print_tensor
 #include "cutlass/arch/arch.h"             // cutlass::arch::Sm90
 #include "cutlass/cutlass.h"               // cutlass::Status
+#include "cutlass/detail/collective.hpp"
 #include "cutlass/detail/layout.hpp"       // cutlass::TagToStrideA_t
 #include "cutlass/fast_math.h"             // cutlass::ceil_div, cutlass::round_up
 #include "cutlass/kernel_hardware_info.h"  // cutlass::KernelHardwareInfo
@@ -219,9 +220,7 @@ public:
   // * EltA
   using ElementA = ElementA_;
   using ElementAUint = cute::uint_bit_t<cute::sizeof_bits_v<ElementA>>;
-  static constexpr bool IsRuntimeDataTypeA = cute::is_same_v<ElementA, cutlass::type_erased_dynamic_float8_t> ||
-                                             cute::is_same_v<ElementA, cutlass::type_erased_dynamic_float6_t> ||
-                                             cute::is_same_v<ElementA, cutlass::type_erased_dynamic_float4_t>;
+  static constexpr bool IsRuntimeDataTypeA = cutlass::gemm::collective::detail::is_sm10x_runtime_f8f6f4<ElementA>();
   using ArrayElementA = cute::conditional_t<IsRuntimeDataTypeA,
                                             cute::uint_bit_t<cute::sizeof_bits_v<ElementA>>,
                                             ElementA>;

test/unit/transform/device/testbed_sparse_gemm_compressor.hpp

@@ -60,6 +60,7 @@
 #include "cutlass/util/packed_stride.hpp"                            // cutlass::make_cute_packed_stride
 #include "cutlass/util/reference/host/tensor_compare.h"              // cutlass::reference::host::TensorEquals
 #include "cutlass/util/reference/host/tensor_fill.h"  // cutlass::reference::host::TensorFillRandomUniform, TensorFillIdentity, TensorFillRandomGaussian, BlockFillSequential, TensorFill
+#include "cutlass/detail/collective.hpp"
 
 #include "sm90_sparse_gemm_compressor_legacy.hpp"     // Legacy host compressor
 #include "../../common/cutlass_unit_test.h"           // CUTLASS UT, EXPECT_TRUE

test/unit/transform/kernel/CMakeLists.txt

@@ -27,6 +27,6 @@
 # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 cutlass_test_unit_add_executable(
-  cutlass_test_unit_transform_filter_format
+  cutlass_test_unit_transform_kernel
   filter_format_transformer.cu
 )