diff --git a/example/ck_tile/03_gemm/gemm_utils.hpp b/example/ck_tile/03_gemm/gemm_utils.hpp
index 588b66ca43..e9b2798172 100644
--- a/example/ck_tile/03_gemm/gemm_utils.hpp
+++ b/example/ck_tile/03_gemm/gemm_utils.hpp
@@ -252,7 +252,7 @@ struct GemmConfigComputeV5 : public GemmConfigBase
 
     static constexpr bool DoubleSmemBuffer               = false;
     static constexpr ck_tile::index_t Pipeline           = CK_TILE_PIPELINE_COMPUTE_V5;
-    static constexpr ck_tile::index_t NumWaNumWaveGroups = 2;
+    static constexpr ck_tile::index_t NumWaveGroups = 2;
 };
 
 template <typename PrecType>
diff --git a/example/ck_tile/03_gemm/universal_gemm_invoker.hpp b/example/ck_tile/03_gemm/universal_gemm_invoker.hpp
index 19855c7f72..d0762e4970 100644
--- a/example/ck_tile/03_gemm/universal_gemm_invoker.hpp
+++ b/example/ck_tile/03_gemm/universal_gemm_invoker.hpp
@@ -102,6 +102,7 @@ struct UniversalInvoker
                                                  TilePartitioner::NPerBlock,
                                                  GemmConfig::M_Warp,
                                                  GemmConfig::N_Warp,
+                                                 GemmConfig::K_Warp,
                                                  GemmConfig::M_Warp_Tile,
                                                  GemmConfig::N_Warp_Tile,
                                                  GemmConfig::K_Warp_Tile,
diff --git a/include/ck_tile/ops/epilogue/cshuffle_epilogue.hpp b/include/ck_tile/ops/epilogue/cshuffle_epilogue.hpp
index 628af0e0b3..b4036cae04 100644
--- a/include/ck_tile/ops/epilogue/cshuffle_epilogue.hpp
+++ b/include/ck_tile/ops/epilogue/cshuffle_epilogue.hpp
@@ -40,6 +40,7 @@ template <typename ADataType_,
           index_t kN_,
           index_t MWave_,
           index_t NWave_,
+          index_t KWave_,
           index_t MPerXdl_,
           index_t NPerXdl_,
           index_t KPerXdl_,
@@ -59,7 +60,7 @@ struct CShuffleEpilogueProblem
     using DsLayout                                         = remove_cvref_t<DsLayout_>;
     using ELayout                                          = remove_cvref_t<ELayout_>;
     using CDElementwise                                    = remove_cvref_t<CDElementwise_>;
-    static constexpr index_t kBlockSize                    = MWave_ * NWave_ * get_warp_size();
+    static constexpr index_t kBlockSize                    = MWave_ * NWave_ * KWave_ * get_warp_size();
     static constexpr index_t kMPerBlock                    = kM_;
     static constexpr index_t kNPerBlock                    = kN_;
     static constexpr index_t MWave                         = MWave_;