Rotating buffer PR CI fix (#2257)

* Revert "Revert "[CK_tile] Add rotating buffer feature for universal gemm (#2200)" (#2256)"

This reverts commit bbdaf79a52.

* fix regression

example/ck_tile/18_flatmm/flatmm_basic.cpp

@@ -11,6 +11,7 @@
 
 #include "ck_tile/host.hpp"
 #include "flatmm_basic.hpp"
+#include "run_flatmm_example.inc"
 
 template <typename ADataType,
           typename BDataType,
@@ -115,9 +116,47 @@ float flatmm_calc(const ck_tile::FlatmmHostArgs& args, const ck_tile::stream_con
                       << std::endl;
         }
 
-        float ave_time = ck_tile::launch_kernel(
-            s, ck_tile::make_kernel<blocks.x, kBlockPerCu>(Kernel{}, grids, blocks, 0, kargs));
+        float ave_time{0};
+        if(s.flush_cache_)
+        {
+            std::cout << "Flushing cache..." << std::endl;
+            static constexpr ck_tile::index_t APackedSize =
+                std::is_same_v<BDataType, ck_tile::pk_int4_t> ? 2 : 1;
+            static constexpr ck_tile::index_t BPackedSize =
+                std::is_same_v<BDataType, ck_tile::pk_int4_t> ? 2 : 1;
 
+            ck_tile::HostTensor<ADataType> a_m(ck_tile::host_tensor_descriptor(
+                args.M, args.K, args.stride_A, is_row_major(ALayout{})));
+            ck_tile::HostTensor<BDataType> b_n(ck_tile::host_tensor_descriptor(
+                args.K, args.N, args.stride_B, is_row_major(BLayout{})));
+
+            auto size_a_buffer = a_m.get_element_space_size_in_bytes() / APackedSize;
+            auto size_b_buffer = b_n.get_element_space_size_in_bytes() / BPackedSize;
+
+            ck_tile::RotatingMemWrapper<ADataType, BDataType> rotating_mem(
+                kargs.a_ptr, kargs.b_shuffle_ptr, s.rotating_count_, size_a_buffer, size_b_buffer);
+            rotating_mem.Print();
+
+            auto run_flush_cache = [&]() {
+                // flush icache
+                ck_tile::flush_icache();
+                // rotating mem
+                rotating_mem.Next();
+                // clear c mem
+                if(args.k_batch > 1)
+                    hipGetErrorString(hipMemsetAsync(
+                        args.c_ptr, 0, args.M * args.N * sizeof(CDataType), s.stream_id_));
+            };
+            ave_time = ck_tile::launch_kernel_preprocess(
+                s,
+                run_flush_cache,
+                ck_tile::make_kernel<blocks.x, kBlockPerCu>(Kernel{}, grids, blocks, 0, kargs));
+        }
+        else
+        {
+            ave_time = ck_tile::launch_kernel(
+                s, ck_tile::make_kernel<blocks.x, kBlockPerCu>(Kernel{}, grids, blocks, 0, kargs));
+        }
         return ave_time;
     };
     if(args.k_batch == 1)
@@ -132,8 +171,6 @@ float flatmm_calc(const ck_tile::FlatmmHostArgs& args, const ck_tile::stream_con
     }
 }
 
-#include "run_flatmm_example.inc"
-
 int run_flatmm_example(int argc, char* argv[])
 {
     auto [result, arg_parser] = create_args(argc, argv);

example/ck_tile/18_flatmm/flatmm_basic.hpp

@@ -133,4 +133,11 @@ auto create_args(int argc, char* argv[])
 }
 
 // host API
+template <typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CDataType,
+          typename ALayout,
+          typename BLayout,
+          typename CLayout>
 float flatmm_calc(const ck_tile::FlatmmHostArgs& args, const ck_tile::stream_config& s);

example/ck_tile/18_flatmm/run_flatmm_example.inc

@@ -122,7 +122,7 @@ float invoke_flatmm(ck_tile::DeviceMem& a_dev_buf,
 
     float ave_time =
         flatmm_calc<ADataType, BDataType, AccDataType, CDataType, ALayout, BLayout, CLayout>(
-            args, ck_tile::stream_config{nullptr, true, 1, n_warmup, n_repeat});
+            args, ck_tile::stream_config{nullptr, true, 1, n_warmup, n_repeat, true, true, 50});
 
     std::size_t flop = std::size_t(2) * M * N * K;
     std::size_t num_byte =