[CK_TILE] Update the file structure

include/ck_tile/ops/gemm/warp/warp_gemm.hpp → include/ck_tile/device/ops/gemm/warp/warp_gemm.hpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
 #include "ck_tile/core.hpp"
-#include "ck_tile/ops/gemm/warp/warp_gemm_impl.hpp"
-#include "ck_tile/ops/gemm/warp/warp_gemm_attribute_mfma.hpp"
+#include "ck_tile/device/ops/gemm/warp/warp_gemm_impl.hpp"
+#include "ck_tile/device/ops/gemm/warp/warp_gemm_attribute_mfma.hpp"
 
 namespace ck_tile {
 

include/ck_tile/ops/gemm/warp/warp_gemm_attribute_mfma.hpp → include/ck_tile/device/ops/gemm/warp/warp_gemm_attribute_mfma.hpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
 #include "ck_tile/core.hpp"
-#include "ck_tile/ops/gemm/warp/warp_gemm_attribute_mfma_impl.hpp"
+#include "ck_tile/device/ops/gemm/warp/warp_gemm_attribute_mfma_impl.hpp"
 
 namespace ck_tile {
 

include/ck_tile/ops/gemm/warp/warp_gemm_attribute_mfma_impl.hpp → include/ck_tile/device/ops/gemm/warp/warp_gemm_attribute_mfma_impl.hpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 

include/ck_tile/ops/gemm/warp/warp_gemm_dispatcher.hpp → include/ck_tile/device/ops/gemm/warp/warp_gemm_dispatcher.hpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
 #include "ck_tile/core.hpp"
-#include "ck_tile/ops/gemm/warp/warp_gemm.hpp"
+#include "ck_tile/device/ops/gemm/warp/warp_gemm.hpp"
 
 namespace ck_tile {
 

include/ck_tile/ops/gemm/warp/warp_gemm_impl.hpp → include/ck_tile/device/ops/gemm/warp/warp_gemm_impl.hpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 

include/ck_tile/device/ops/layernorm2d.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck_tile/device/ops/layernorm2d/kernel/layernorm2d_fwd_kernel.hpp"
+#include "ck_tile/device/ops/layernorm2d/pipeline/layernorm2d_fwd_pipeline_default_policy.hpp"
+#include "ck_tile/device/ops/layernorm2d/pipeline/layernorm2d_fwd_pipeline_one_pass.hpp"
+#include "ck_tile/device/ops/layernorm2d/pipeline/layernorm2d_fwd_pipeline_problem.hpp"
+#include "ck_tile/device/ops/layernorm2d/pipeline/layernorm2d_fwd_pipeline_two_pass.hpp"
+#include "ck_tile/device/ops/layernorm2d/pipeline/layernorm2d_fwd_traits.hpp"
+#include "ck_tile/device/ops/common/generic_2d_block_shape.hpp"
+#include "ck_tile/device/ops/common/tensor_layout.hpp"

include/ck_tile/ops/layernorm2d/kernel/layernorm2d_fwd_kernel.hpp → include/ck_tile/device/ops/layernorm2d/kernel/layernorm2d_fwd_kernel.hpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
 #include "ck_tile/core.hpp"
-#include "ck_tile/ops/common.hpp"
-#include "ck_tile/ops/layernorm2d/pipeline/layernorm2d_fwd_traits.hpp"
+#include "ck_tile/device/ops/common.hpp"
+#include "ck_tile/device/ops/layernorm2d/pipeline/layernorm2d_fwd_traits.hpp"
 
 namespace ck_tile {
 

include/ck_tile/ops/layernorm2d/pipeline/layernorm2d_fwd_pipeline_default_policy.hpp → include/ck_tile/device/ops/layernorm2d/pipeline/layernorm2d_fwd_pipeline_default_policy.hpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
 #include "ck_tile/core.hpp"
-#include "ck_tile/ops/norm_reduce/block/block_norm_reduce_problem.hpp"
-#include "ck_tile/ops/norm_reduce/block/block_norm_reduce.hpp"
+#include "ck_tile/device/ops/norm_reduce/block/block_norm_reduce_problem.hpp"
+#include "ck_tile/device/ops/norm_reduce/block/block_norm_reduce.hpp"
 
 namespace ck_tile {
 

include/ck_tile/ops/layernorm2d/pipeline/layernorm2d_fwd_pipeline_one_pass.hpp → include/ck_tile/device/ops/layernorm2d/pipeline/layernorm2d_fwd_pipeline_one_pass.hpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
 #include "ck_tile/core.hpp"
-#include "ck_tile/ops/layernorm2d/pipeline/layernorm2d_fwd_pipeline_default_policy.hpp"
-#include "ck_tile/ops/layernorm2d/pipeline/layernorm2d_fwd_traits.hpp"
+#include "ck_tile/device/ops/layernorm2d/pipeline/layernorm2d_fwd_pipeline_default_policy.hpp"
+#include "ck_tile/device/ops/layernorm2d/pipeline/layernorm2d_fwd_traits.hpp"
 #include <string>
 #include <type_traits>
 

include/ck_tile/ops/layernorm2d/pipeline/layernorm2d_fwd_pipeline_problem.hpp → include/ck_tile/device/ops/layernorm2d/pipeline/layernorm2d_fwd_pipeline_problem.hpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 

include/ck_tile/ops/layernorm2d/pipeline/layernorm2d_fwd_pipeline_two_pass.hpp → include/ck_tile/device/ops/layernorm2d/pipeline/layernorm2d_fwd_pipeline_two_pass.hpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
 #include "ck_tile/core.hpp"
-#include "ck_tile/ops/layernorm2d/pipeline/layernorm2d_fwd_pipeline_default_policy.hpp"
+#include "ck_tile/device/ops/layernorm2d/pipeline/layernorm2d_fwd_pipeline_default_policy.hpp"
 #include <string>
 #include <type_traits>
 

include/ck_tile/ops/layernorm2d/pipeline/layernorm2d_fwd_traits.hpp → include/ck_tile/device/ops/layernorm2d/pipeline/layernorm2d_fwd_traits.hpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 

include/ck_tile/device/ops/reduce.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck_tile/device/ops/reduce/block/block_reduce.hpp"
+#include "ck_tile/device/ops/reduce/block/block_reduce2d.hpp"
+#include "ck_tile/device/ops/reduce/block/block_reduce2d_default_policy.hpp"
+#include "ck_tile/device/ops/reduce/block/block_reduce2d_problem.hpp"
+#include "ck_tile/device/ops/common/generic_2d_block_shape.hpp"
+#include "ck_tile/device/ops/common/tensor_layout.hpp"

include/ck_tile/ops/reduce/block/block_reduce.hpp → include/ck_tile/device/ops/reduce/block/block_reduce.hpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 

include/ck_tile/ops/reduce/block/block_reduce2d.hpp → include/ck_tile/device/ops/reduce/block/block_reduce2d.hpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 

include/ck_tile/ops/reduce/block/block_reduce2d_default_policy.hpp → include/ck_tile/device/ops/reduce/block/block_reduce2d_default_policy.hpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
 #include "ck_tile/core.hpp"
-#include "ck_tile/ops/reduce/block/block_reduce2d_problem.hpp"
-#include "ck_tile/ops/reduce/block/block_reduce2d.hpp"
+#include "ck_tile/device/ops/reduce/block/block_reduce2d_problem.hpp"
+#include "ck_tile/device/ops/reduce/block/block_reduce2d.hpp"
 
 namespace ck_tile {
 

include/ck_tile/ops/reduce/block/block_reduce2d_problem.hpp → include/ck_tile/device/ops/reduce/block/block_reduce2d_problem.hpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 

include/ck_tile/host.hpp

@@ -3,17 +3,68 @@
 
 #pragma once
 
-#include "ck_tile/host/arg_parser.hpp"
-#include "ck_tile/host/check_err.hpp"
-#include "ck_tile/host/convolution_host_tensor_descriptor_helper.hpp"
-#include "ck_tile/host/convolution_parameter.hpp"
-#include "ck_tile/host/device_memory.hpp"
-#include "ck_tile/host/fill.hpp"
-#include "ck_tile/host/hip_check_error.hpp"
-#include "ck_tile/host/host_tensor.hpp"
-#include "ck_tile/host/joinable_thread.hpp"
-#include "ck_tile/host/kernel_launch.hpp"
-#include "ck_tile/host/ranges.hpp"
+#include "ck_tile/host/ops/add_rmsnorm2d_rdquant/kernel/add_rmsnorm2d_rdquant_fwd_kernel.hpp"
+#include "ck_tile/host/ops/add_rmsnorm2d_rdquant/pipeline/add_rmsnorm2d_rdquant_fwd_pipeline_default_policy.hpp"
+#include "ck_tile/host/ops/add_rmsnorm2d_rdquant/pipeline/add_rmsnorm2d_rdquant_fwd_pipeline_one_pass.hpp"
+#include "ck_tile/host/ops/add_rmsnorm2d_rdquant/pipeline/add_rmsnorm2d_rdquant_fwd_pipeline_problem.hpp"
+#include "ck_tile/host/ops/add_rmsnorm2d_rdquant/pipeline/add_rmsnorm2d_rdquant_fwd_pipeline_three_pass.hpp"
+#include "ck_tile/host/ops/add_rmsnorm2d_rdquant.hpp"
+#include "ck_tile/host/ops/elementwise/unary_element_wise_operation.hpp"
+#include "ck_tile/host/ops/elementwise.hpp"
+#include "ck_tile/host/ops/flatmm/block/flatmm_32x512x128_1x4x1_16x16x32.hpp"
+#include "ck_tile/host/ops/flatmm/block/flatmm_sn_32x128x512_1x4x1_16x16x32.hpp"
+#include "ck_tile/host/ops/flatmm/block/flatmm_sn_32x128x512_1x4x1_16x16x32_itl.hpp"
+#include "ck_tile/host/ops/flatmm/block/flatmm_uk_config.hpp"
+#include "ck_tile/host/ops/flatmm.hpp"
+#include "ck_tile/host/ops/fused_moe/kernel/fused_moegemm_kernel.hpp"
+#include "ck_tile/host/ops/fused_moe/kernel/fused_moegemm_shape.hpp"
+#include "ck_tile/host/ops/fused_moe/kernel/fused_moegemm_tile_partitioner.hpp"
+#include "ck_tile/host/ops/fused_moe/kernel/moe_sorting_kernel.hpp"
+#include "ck_tile/host/ops/fused_moe/pipeline/fused_moegemm_pipeline_flatmm_ex.hpp"
+#include "ck_tile/host/ops/fused_moe/pipeline/fused_moegemm_pipeline_flatmm_policy.hpp"
+#include "ck_tile/host/ops/fused_moe/pipeline/fused_moegemm_pipeline_flatmm_uk.hpp"
+#include "ck_tile/host/ops/fused_moe/pipeline/fused_moegemm_pipeline_problem.hpp"
+#include "ck_tile/host/ops/fused_moe/pipeline/fused_moegemm_traits.hpp"
+#include "ck_tile/host/ops/fused_moe/pipeline/moe_sorting_pipeline.hpp"
+#include "ck_tile/host/ops/fused_moe/pipeline/moe_sorting_policy.hpp"
+#include "ck_tile/host/ops/fused_moe/pipeline/moe_sorting_problem.hpp"
+#include "ck_tile/host/ops/fused_moe.hpp"
+#include "ck_tile/host/ops/image_to_column/kernel/image_to_column_kernel.hpp"
+#include "ck_tile/host/ops/image_to_column/pipeline/block_image_to_column_problem.hpp"
+#include "ck_tile/host/ops/image_to_column/pipeline/tile_image_to_column_shape.hpp"
+#include "ck_tile/host/ops/image_to_column.hpp"
+#include "ck_tile/host/ops/norm_reduce/block/block_norm_reduce.hpp"
+#include "ck_tile/host/ops/norm_reduce/block/block_norm_reduce_problem.hpp"
+#include "ck_tile/host/ops/norm_reduce/thread/thread_welford.hpp"
+#include "ck_tile/host/ops/norm_reduce.hpp"
+#include "ck_tile/host/ops/permute/kernel/generic_permute_kernel.hpp"
+#include "ck_tile/host/ops/permute/pipeline/generic_petmute_problem.hpp"
+#include "ck_tile/host/ops/permute.hpp"
+#include "ck_tile/host/ops/rmsnorm2d/kernel/rmsnorm2d_fwd_kernel.hpp"
+#include "ck_tile/host/ops/rmsnorm2d/pipeline/rmsnorm2d_fwd_pipeline_default_policy.hpp"
+#include "ck_tile/host/ops/rmsnorm2d/pipeline/rmsnorm2d_fwd_pipeline_one_pass.hpp"
+#include "ck_tile/host/ops/rmsnorm2d/pipeline/rmsnorm2d_fwd_pipeline_problem.hpp"
+#include "ck_tile/host/ops/rmsnorm2d/pipeline/rmsnorm2d_fwd_pipeline_two_pass.hpp"
+#include "ck_tile/host/ops/rmsnorm2d.hpp"
+#include "ck_tile/host/ops/smoothquant/kernel/moe_smoothquant_kernel.hpp"
+#include "ck_tile/host/ops/smoothquant/kernel/smoothquant_kernel.hpp"
+#include "ck_tile/host/ops/smoothquant/pipeline/smoothquant_pipeline_default_policy.hpp"
+#include "ck_tile/host/ops/smoothquant/pipeline/smoothquant_pipeline_one_pass.hpp"
+#include "ck_tile/host/ops/smoothquant/pipeline/smoothquant_pipeline_problem.hpp"
+#include "ck_tile/host/ops/smoothquant/pipeline/smoothquant_pipeline_two_pass.hpp"
+#include "ck_tile/host/ops/smoothquant.hpp"
+#include "ck_tile/host/ops/softmax/block/block_softmax_2d.hpp"
+#include "ck_tile/host/ops/softmax/block/block_softmax_2d_problem.hpp"
+#include "ck_tile/host/ops/softmax.hpp"
+#include "ck_tile/host/ops/topk/block/block_topk_stream_2d.hpp"
+#include "ck_tile/host/ops/topk/block/block_topk_stream_2d_problem.hpp"
+#include "ck_tile/host/ops/topk.hpp"
+#include "ck_tile/host/ops/topk_softmax/kernel/topk_softmax_kernel.hpp"
+#include "ck_tile/host/ops/topk_softmax/pipeline/topk_softmax_warp_per_row_pipeline.hpp"
+#include "ck_tile/host/ops/topk_softmax/pipeline/topk_softmax_warp_per_row_policy.hpp"
+#include "ck_tile/host/ops/topk_softmax/pipeline/topk_softmax_warp_per_row_problem.hpp"
+#include "ck_tile/host/ops/topk_softmax.hpp"
+#include "ck_tile/host/reference/naive_attention.hpp"
 #include "ck_tile/host/reference/reference_batched_dropout.hpp"
 #include "ck_tile/host/reference/reference_batched_elementwise.hpp"
 #include "ck_tile/host/reference/reference_batched_gemm.hpp"
@@ -32,5 +83,18 @@
 #include "ck_tile/host/reference/reference_rowwise_quantization2d.hpp"
 #include "ck_tile/host/reference/reference_softmax.hpp"
 #include "ck_tile/host/reference/reference_topk.hpp"
-#include "ck_tile/host/stream_config.hpp"
-#include "ck_tile/host/timer.hpp"
+#include "ck_tile/host/reference.hpp"
+#include "ck_tile/host/util/arg_parser.hpp"
+#include "ck_tile/host/util/check_err.hpp"
+#include "ck_tile/host/util/convolution_host_tensor_descriptor_helper.hpp"
+#include "ck_tile/host/util/convolution_parameter.hpp"
+#include "ck_tile/host/util/device_memory.hpp"
+#include "ck_tile/host/util/fill.hpp"
+#include "ck_tile/host/util/hip_check_error.hpp"
+#include "ck_tile/host/util/host_tensor.hpp"
+#include "ck_tile/host/util/joinable_thread.hpp"
+#include "ck_tile/host/util/kernel_launch.hpp"
+#include "ck_tile/host/util/ranges.hpp"
+#include "ck_tile/host/util/stream_config.hpp"
+#include "ck_tile/host/util/timer.hpp"
+#include "ck_tile/host/util.hpp"

include/ck_tile/host/ops/add_rmsnorm2d_rdquant.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck_tile/host/ops/add_rmsnorm2d_rdquant/kernel/add_rmsnorm2d_rdquant_fwd_kernel.hpp"
+#include "ck_tile/host/ops/add_rmsnorm2d_rdquant/pipeline/add_rmsnorm2d_rdquant_fwd_pipeline_default_policy.hpp"
+#include "ck_tile/host/ops/add_rmsnorm2d_rdquant/pipeline/add_rmsnorm2d_rdquant_fwd_pipeline_one_pass.hpp"
+#include "ck_tile/host/ops/add_rmsnorm2d_rdquant/pipeline/add_rmsnorm2d_rdquant_fwd_pipeline_problem.hpp"
+#include "ck_tile/host/ops/add_rmsnorm2d_rdquant/pipeline/add_rmsnorm2d_rdquant_fwd_pipeline_three_pass.hpp"

include/ck_tile/ops/add_rmsnorm2d_rdquant/kernel/add_rmsnorm2d_rdquant_fwd_kernel.hpp → include/ck_tile/host/ops/add_rmsnorm2d_rdquant/kernel/add_rmsnorm2d_rdquant_fwd_kernel.hpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
 #include "ck_tile/core.hpp"
-#include "ck_tile/ops/common.hpp"
+#include "ck_tile/device/ops/common.hpp"
 
 namespace ck_tile {