added stream-k policies for 3 tile and 4 tile

include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp

@@ -1471,6 +1471,16 @@ struct BlockToCTileMap_GemmStreamK_v2
             {
                 sk_tiles = bigEnough ? (grid_size + num_tiles % grid_size) : num_tiles;
             }
+            else if(streamk_sel == 3)
+            {
+                sk_tiles = (num_tiles > (2 * grid_size)) ? (2 * grid_size + num_tiles % grid_size)
+                                                         : num_tiles;
+            }
+            else if(streamk_sel == 4)
+            {
+                sk_tiles = (num_tiles > (3 * grid_size)) ? (3 * grid_size + num_tiles % grid_size)
+                                                         : num_tiles;
+            }
             sk_num_blocks = sk_tiles;
             // if(sk_tiles < sk_num_blocks)
             // {

profiler/include/profiler/profile_gemm_universal_streamk_impl.hpp

@@ -153,7 +153,7 @@ bool profile_gemm_universal_streamk_impl(int do_verification,
     for(auto& op_ptr : op_ptrs)
     {
         std::vector<int> grid_size_list   = {38, 76, 114, 152, 190, 228, 266, 304, 342, 380};
-        std::vector<int> streamk_sel_list = {0, 1, 2};
+        std::vector<int> streamk_sel_list = {0, 1, 2, 3, 4};
 
         if(Grid_size == -1)
         {