[None][perf] Replace DSV4 O_b split-K with CuTe DSL

Replace the dedicated C++/THOP O_b split-K GEMM with a CuTe DSL implementation built on the persistent SM100 block-scaled GEMM. Consume packed K128 UE8M0 scales directly, issue runtime-N tcgen05 MMA, and preserve BF16 split-major partials for the existing mHC fused reduction.

Tune the packed-scale, dynamic-N, scheduler, and epilogue paths against current DeepGEMM. Dispatch SK2 through M=128, use CuTe SK1 for supported M>=256 shapes, and retain DeepGEMM as the unsupported-shape fallback. Add tactic, dispatch, packed-scale, and split-partial correctness coverage while removing the obsolete C++ kernel and registration.

Signed-off-by: Mingyang Hao <200044211+mingyangHao@users.noreply.github.com>

Author: mingyangHao

cpp/tensorrt_llm/kernels/mhcKernels/CMakeLists.txt

@@ -19,7 +19,7 @@
 # this directory (bench_*.cu, probe_*.cu, profile_*.cu, test_*.cu) are
 # standalone binaries with their own main() and must NOT be compiled into the
 # trtllm shared library.
-set(SRC_CU dsv4Fp8SplitKGemm.cu mhcKernels.cu mhcFusedHcKernel.cu)
+set(SRC_CU mhcKernels.cu mhcFusedHcKernel.cu)
 add_library(mhcKernels_src OBJECT ${SRC_CU})
 
 set_property(TARGET mhcKernels_src PROPERTY POSITION_INDEPENDENT_CODE ON)
@@ -28,7 +28,7 @@ target_compile_options(mhcKernels_src
                        PRIVATE $<$<COMPILE_LANGUAGE:CUDA>:--use_fast_math>)
 
 # Expose the DeepGEMM SM100 MMA/TMA building blocks used by the fused
-# post-mapping kernels and DSV4 O_b split-K GEMM.
+# post-mapping kernels.
 target_include_directories(
   mhcKernels_src
   PRIVATE ${CMAKE_BINARY_DIR}/_deps/deepgemm-src/deep_gemm/include)