change file extension to hip.hpp and hip.cpp

src/include/gridwise_implicit_gemm_convolution_1_chwn_csrk_khwn.cuh → src/include/gridwise_implicit_gemm_convolution_1_chwn_csrk_khwn.hip.hpp

 #pragma once
-#include "common.cuh"
-#include "ConstantTensorDescriptor.cuh"
-#include "ConstantMatrixDescriptor.cuh"
-#include "blockwise_4d_tensor_op.cuh"
-#include "blockwise_2d_tensor_op.cuh"
-#include "threadwise_4d_tensor_op.cuh"
-#include "blockwise_gemm.cuh"
+#include "common.hip.hpp"
+#include "ConstantTensorDescriptor.hip.hpp"
+#include "ConstantMatrixDescriptor.hip.hpp"
+#include "blockwise_4d_tensor_op.hip.hpp"
+#include "blockwise_2d_tensor_op.hip.hpp"
+#include "threadwise_4d_tensor_op.hip.hpp"
+#include "blockwise_gemm.hip.hpp"
 
 template <unsigned GridSize,
           unsigned BlockSize,
@@ -199,8 +199,7 @@ gridwise_implicit_gemm_convolution_1_chwn_csrk_khwn(const Float* const __restric
     threadwise_4d_tensor_set_zero(out_hkwn_thread_desc, p_out_thread);
 
     const Float* p_in_global_block_begin =
-        p_in_global +
-        in_chwn_global_desc.Get1dIndex(
+        p_in_global + in_chwn_global_desc.Get1dIndex(
                           0, hi_block_data_begin, wi_block_data_begin, n_block_data_begin);
 
     const Float* p_wei_global_block_begin =
@@ -258,8 +257,7 @@ gridwise_implicit_gemm_convolution_1_chwn_csrk_khwn(const Float* const __restric
         out_hkwn_thread_desc,
         p_out_thread,
         out_khwn_global_desc,
-        p_out_global +
-            out_khwn_global_desc.Get1dIndex(k_block_data_begin + k_thread_data_begin,
+        p_out_global + out_khwn_global_desc.Get1dIndex(k_block_data_begin + k_thread_data_begin,
                                                        ho_block_data_begin + ho_thread_data_begin,
                                                        wo_block_data_begin + wo_thread_data_begin,
                                                        n_block_data_begin + n_thread_data_begin),

src/include/gridwise_implicit_gemm_convolution_1_chwn_csrk_khwn_padded.cuh → src/include/gridwise_implicit_gemm_convolution_1_chwn_csrk_khwn_padded.hip.hpp

 #pragma once
-#include "common.cuh"
-#include "ConstantTensorDescriptor.cuh"
-#include "ConstantMatrixDescriptor.cuh"
-#include "blockwise_4d_tensor_op.cuh"
-#include "blockwise_2d_tensor_op.cuh"
-#include "threadwise_4d_tensor_op.cuh"
-#include "blockwise_gemm.cuh"
+#include "common.hip.hpp"
+#include "ConstantTensorDescriptor.hip.hpp"
+#include "ConstantMatrixDescriptor.hip.hpp"
+#include "blockwise_4d_tensor_op.hip.hpp"
+#include "blockwise_2d_tensor_op.hip.hpp"
+#include "threadwise_4d_tensor_op.hip.hpp"
+#include "blockwise_gemm.hip.hpp"
 
 template <unsigned GridSize,
           unsigned BlockSize,
@@ -283,8 +283,7 @@ __global__ void gridwise_implicit_gemm_convolution_1_chwn_csrk_khwn_padded(
         out_hkwn_thread_desc,
         p_out_thread,
         out_khwn_global_desc,
-        p_out_global +
-            out_khwn_global_desc.Get1dIndex(k_block_data_begin + k_thread_data_begin,
+        p_out_global + out_khwn_global_desc.Get1dIndex(k_block_data_begin + k_thread_data_begin,
                                                        ho_block_data_begin + ho_thread_data_begin,
                                                        wo_block_data_begin + wo_thread_data_begin,
                                                        n_block_data_begin + n_thread_data_begin),

src/include/gridwise_implicit_gemm_convolution_1_chwn_csrk_khwn_padded_lds_pipeline.cuh → src/include/gridwise_implicit_gemm_convolution_1_chwn_csrk_khwn_padded_lds_pipeline.hip.hpp

 #pragma once
-#include "common.cuh"
-#include "ConstantTensorDescriptor.cuh"
-#include "ConstantMatrixDescriptor.cuh"
-#include "blockwise_4d_tensor_op.cuh"
-#include "blockwise_2d_tensor_op.cuh"
-#include "threadwise_4d_tensor_op.cuh"
-#include "blockwise_gemm.cuh"
+#include "common.hip.hpp"
+#include "ConstantTensorDescriptor.hip.hpp"
+#include "ConstantMatrixDescriptor.hip.hpp"
+#include "blockwise_4d_tensor_op.hip.hpp"
+#include "blockwise_2d_tensor_op.hip.hpp"
+#include "threadwise_4d_tensor_op.hip.hpp"
+#include "blockwise_gemm.hip.hpp"
 
 template <unsigned GridSize,
           unsigned BlockSize,
@@ -339,8 +339,7 @@ __global__ void gridwise_implicit_gemm_convolution_1_chwn_csrk_khwn_padded_lds_p
         out_hkwn_thread_desc,
         p_out_thread,
         out_khwn_global_desc,
-        p_out_global +
-            out_khwn_global_desc.Get1dIndex(k_block_data_begin + k_thread_data_begin,
+        p_out_global + out_khwn_global_desc.Get1dIndex(k_block_data_begin + k_thread_data_begin,
                                                        ho_block_data_begin + ho_thread_data_begin,
                                                        wo_block_data_begin + wo_thread_data_begin,
                                                        n_block_data_begin + n_thread_data_begin),

src/include/gridwise_implicit_gemm_convolution_1_nchw_kcsr_nkhw.cuh → src/include/gridwise_implicit_gemm_convolution_1_nchw_kcsr_nkhw.hip.hpp

 #pragma once
-#include "common.cuh"
-#include "ConstantTensorDescriptor.cuh"
-#include "ConstantMatrixDescriptor.cuh"
-#include "blockwise_4d_tensor_op.cuh"
-#include "threadwise_4d_tensor_op.cuh"
-#include "blockwise_gemm.cuh"
+#include "common.hip.hpp"
+#include "ConstantTensorDescriptor.hip.hpp"
+#include "ConstantMatrixDescriptor.hip.hpp"
+#include "blockwise_4d_tensor_op.hip.hpp"
+#include "threadwise_4d_tensor_op.hip.hpp"
+#include "blockwise_gemm.hip.hpp"
 
 template <unsigned GridSize,
           unsigned BlockSize,
@@ -160,8 +160,7 @@ gridwise_implicit_gemm_convolution_1_nchw_kcsr_nkhw(const Float* const __restric
         //   convert [N,C,Hi,Wi] to [C,Hi,Wi,N]
         blockwise_4d_tensor_copy_reorder_by_get_dst_from_src<BlockSize>(
             in_nchw_global_desc,
-            p_in_global +
-                in_nchw_global_desc.Get1dIndex(n_block_data_begin,
+            p_in_global + in_nchw_global_desc.Get1dIndex(n_block_data_begin,
                                                          c_block_data_begin,
                                                          hi_block_data_begin,
                                                          wi_block_data_begin),
@@ -245,8 +244,7 @@ gridwise_implicit_gemm_convolution_1_nchw_kcsr_nkhw(const Float* const __restric
         out_hkwn_thread_desc,
         p_out_thread,
         out_nkhw_global_desc,
-        p_out_global +
-            out_nkhw_global_desc.Get1dIndex(n_block_data_begin,
+        p_out_global + out_nkhw_global_desc.Get1dIndex(n_block_data_begin,
                                                        k_block_data_begin + k_thread_data_begin,
                                                        ho_block_data_begin + ho_thread_data_begin,
                                                        wo_block_data_begin + wo_thread_data_begin),
@@ -263,8 +261,7 @@ gridwise_implicit_gemm_convolution_1_nchw_kcsr_nkhw(const Float* const __restric
         out_nkhw_thread_desc,
         p_out_thread,
         out_nkhw_global_desc,
-        p_out_global +
-            out_nkhw_global_desc.Get1dIndex(n_block_data_begin,
+        p_out_global + out_nkhw_global_desc.Get1dIndex(n_block_data_begin,
                                                        k_block_data_begin + k_thread_data_begin,
                                                        ho_block_data_begin + ho_thread_data_begin,
                                                        wo_block_data_begin + wo_thread_data_begin),

src/include/gridwise_implicit_gemm_convolution_1_nchw_srck_nkhw.cuh → src/include/gridwise_implicit_gemm_convolution_1_nchw_srck_nkhw.hip.hpp

 #pragma once
-#include "common.cuh"
-#include "ConstantTensorDescriptor.cuh"
-#include "ConstantMatrixDescriptor.cuh"
-#include "blockwise_4d_tensor_op.cuh"
-#include "threadwise_4d_tensor_op.cuh"
-#include "blockwise_gemm.cuh"
+#include "common.hip.hpp"
+#include "ConstantTensorDescriptor.hip.hpp"
+#include "ConstantMatrixDescriptor.hip.hpp"
+#include "blockwise_4d_tensor_op.hip.hpp"
+#include "threadwise_4d_tensor_op.hip.hpp"
+#include "blockwise_gemm.hip.hpp"
 
 template <unsigned GridSize,
           unsigned BlockSize,
@@ -166,8 +166,7 @@ gridwise_implicit_gemm_convolution_1_nchw_srck_nkhw(const Float* const __restric
         //   convert [N,C,Hi,Wi] to [C,Hi,Wi,N]
         blockwise_4d_tensor_copy_reorder_by_get_dst_from_src<BlockSize>(
             in_nchw_global_desc,
-            p_in_global +
-                in_nchw_global_desc.Get1dIndex(n_block_data_begin,
+            p_in_global + in_nchw_global_desc.Get1dIndex(n_block_data_begin,
                                                          c_block_data_begin,
                                                          hi_block_data_begin,
                                                          wi_block_data_begin),
@@ -180,9 +179,8 @@ gridwise_implicit_gemm_convolution_1_nchw_srck_nkhw(const Float* const __restric
 #if 1
         // weight: global mem to LDS,
         //   format is [S,R,C,K], no conversion needed
-        blockwise_wei_copy.Run(
-            p_wei_global +
-                wei_srck_global_desc.Get1dIndex(0, 0, c_block_data_begin, k_block_data_begin),
+        blockwise_wei_copy.Run(p_wei_global + wei_srck_global_desc.Get1dIndex(
+                                                  0, 0, c_block_data_begin, k_block_data_begin),
                                p_wei_block);
 #endif
 
@@ -219,8 +217,7 @@ gridwise_implicit_gemm_convolution_1_nchw_srck_nkhw(const Float* const __restric
         out_hkwn_thread_desc,
         p_out_thread,
         out_nkhw_global_desc,
-        p_out_global +
-            out_nkhw_global_desc.Get1dIndex(n_block_data_begin + n_thread_data_begin,
+        p_out_global + out_nkhw_global_desc.Get1dIndex(n_block_data_begin + n_thread_data_begin,
                                                        k_block_data_begin + k_thread_data_begin,
                                                        ho_block_data_begin + ho_thread_data_begin,
                                                        wo_block_data_begin + wo_thread_data_begin),

src/include/gridwise_implicit_gemm_convolution_2_cnhw_csrk_knhw.cuh → src/include/gridwise_implicit_gemm_convolution_2_cnhw_csrk_knhw.hip.hpp

 #pragma once
-#include "common.cuh"
-#include "ConstantTensorDescriptor.cuh"
-#include "ConstantMatrixDescriptor.cuh"
-#include "blockwise_4d_tensor_op.cuh"
-#include "blockwise_2d_tensor_op.cuh"
-#include "threadwise_2d_tensor_op.cuh"
-#include "blockwise_gemm.cuh"
+#include "common.hip.hpp"
+#include "ConstantTensorDescriptor.hip.hpp"
+#include "ConstantMatrixDescriptor.hip.hpp"
+#include "blockwise_4d_tensor_op.hip.hpp"
+#include "blockwise_2d_tensor_op.hip.hpp"
+#include "threadwise_2d_tensor_op.hip.hpp"
+#include "blockwise_gemm.hip.hpp"
 
 // define B = flatten(N, Hi, Wi)
 template <unsigned GridSize,

src/include/gridwise_implicit_gemm_convolution_2_cnhw_csrk_knhw_lds_double_buffer.cuh → src/include/gridwise_implicit_gemm_convolution_2_cnhw_csrk_knhw_lds_double_buffer.hip.hpp

 #pragma once
-#include "common.cuh"
-#include "ConstantTensorDescriptor.cuh"
-#include "ConstantMatrixDescriptor.cuh"
-#include "blockwise_4d_tensor_op.cuh"
-#include "blockwise_2d_tensor_op.cuh"
-#include "threadwise_2d_tensor_op.cuh"
-#include "blockwise_gemm.cuh"
+#include "common.hip.hpp"
+#include "ConstantTensorDescriptor.hip.hpp"
+#include "ConstantMatrixDescriptor.hip.hpp"
+#include "blockwise_4d_tensor_op.hip.hpp"
+#include "blockwise_2d_tensor_op.hip.hpp"
+#include "threadwise_2d_tensor_op.hip.hpp"
+#include "blockwise_gemm.hip.hpp"
 
 // define B = flatten(N, Hi, Wi)
 template <unsigned GridSize,

src/include/gridwise_implicit_gemm_convolution_2_cnhw_srck_knhw.cuh → src/include/gridwise_implicit_gemm_convolution_2_cnhw_srck_knhw.hip.hpp

 #pragma once
-#include "common.cuh"
-#include "ConstantTensorDescriptor.cuh"
-#include "ConstantMatrixDescriptor.cuh"
-#include "blockwise_4d_tensor_op.cuh"
-#include "blockwise_2d_tensor_op.cuh"
-#include "threadwise_2d_tensor_op.cuh"
-#include "blockwise_gemm.cuh"
+#include "common.hip.hpp"
+#include "ConstantTensorDescriptor.hip.hpp"
+#include "ConstantMatrixDescriptor.hip.hpp"
+#include "blockwise_4d_tensor_op.hip.hpp"
+#include "blockwise_2d_tensor_op.hip.hpp"
+#include "threadwise_2d_tensor_op.hip.hpp"
+#include "blockwise_gemm.hip.hpp"
 
 // define B = flatten(N, Hi, Wi)
 template <unsigned GridSize,

src/include/gridwise_implicit_gemm_convolution_2_cnhw_srck_knhw_lds_pipeline.cuh → src/include/gridwise_implicit_gemm_convolution_2_cnhw_srck_knhw_lds_pipeline.hip.hpp

 #pragma once
-#include "common.cuh"
-#include "ConstantTensorDescriptor.cuh"
-#include "ConstantMatrixDescriptor.cuh"
-#include "blockwise_4d_tensor_op.cuh"
-#include "blockwise_2d_tensor_op.cuh"
-#include "threadwise_2d_tensor_op.cuh"
-#include "blockwise_gemm.cuh"
+#include "common.hip.hpp"
+#include "ConstantTensorDescriptor.hip.hpp"
+#include "ConstantMatrixDescriptor.hip.hpp"
+#include "blockwise_4d_tensor_op.hip.hpp"
+#include "blockwise_2d_tensor_op.hip.hpp"
+#include "threadwise_2d_tensor_op.hip.hpp"
+#include "blockwise_gemm.hip.hpp"
 
 // define B = N*Hi*Wi
 template <unsigned GridSize,
@@ -220,9 +220,8 @@ __global__ void gridwise_implicit_gemm_convolution_2_cnhw_srck_knhw_lds_pipeline
 #if 1
         // preload next data
         // input: global mem to LDS,
-        blockwise_in_copy.Run(
-            p_in_global +
-                in_cb_global_desc.Get1dIndex(c_block_data_begin + CPerBlock, b_block_data_begin),
+        blockwise_in_copy.Run(p_in_global + in_cb_global_desc.Get1dIndex(
+                                                c_block_data_begin + CPerBlock, b_block_data_begin),
                               p_in_block_next);
 #endif
 

src/include/gridwise_winograd_convolution.cuh → src/include/gridwise_winograd_convolution.hip.hpp

 #pragma once
-#include "ConstantTensorDescriptor.cuh"
-#include "blockwise_winograd_transform.cuh"
-#include "threadwise_winograd_transform.cuh"
+#include "ConstantTensorDescriptor.hip.hpp"
+#include "blockwise_winograd_transform.hip.hpp"
+#include "threadwise_winograd_transform.hip.hpp"
 
 template <class Float,
           class InGlobalDesc,
@@ -189,16 +189,15 @@ __global__ void gridwise_winograd_convolution(const Float* const __restrict__ p_
                 S,
                 R,
                 OutTileSizeH,
-                OutTileSizeW>(
-                in_transform_thread_block_desc,
-                p_in_transform_block +
-                    in_transform_block_desc.Get1dIndex(n_thread_data_begin,
+                OutTileSizeW>(in_transform_thread_block_desc,
+                              p_in_transform_block + in_transform_block_desc.Get1dIndex(
+                                                         n_thread_data_begin,
                                                          c_thread_data,
                                                          y_thread_data_begin * InTileSizeH,
                                                          x_thread_data_begin * InTileSizeW),
                               wei_transform_thread_block_desc,
-                p_wei_transform_block +
-                    wei_transform_block_desc.Get1dIndex(k_thread_data_begin, c_thread_data, 0, 0),
+                              p_wei_transform_block + wei_transform_block_desc.Get1dIndex(
+                                                          k_thread_data_begin, c_thread_data, 0, 0),
                               out_transform_thread_desc,
                               p_out_transform_thread);
         }

src/include/tensor.hpp

@@ -22,7 +22,8 @@ std::ostream& LogRange(std::ostream& os, Range&& r, std::string delim)
     return os;
 }
 
-typedef enum {
+typedef enum
+{
     Half  = 0,
     Float = 1,
 } DataType_t;

src/include/threadwise_2d_tensor_op.cuh → src/include/threadwise_2d_tensor_op.hip.hpp

 #pragma once
-#include "ConstantTensorDescriptor.cuh"
+#include "ConstantTensorDescriptor.hip.hpp"
 
 template <class Float, class Desc, class F>
 __device__ void threadwise_2d_tensor_pointwise_operation_unary(Desc, Float* __restrict__ p, F f)

src/include/threadwise_4d_tensor_op.cuh → src/include/threadwise_4d_tensor_op.hip.hpp

 #pragma once
-#include "ConstantTensorDescriptor.cuh"
+#include "ConstantTensorDescriptor.hip.hpp"
 
 template <class Float, class Desc, class F>
 __device__ void threadwise_4d_tensor_pointwise_operation_unary(Desc, Float* __restrict__ p, F f)

src/include/threadwise_direct_convolution.cuh → src/include/threadwise_direct_convolution.hip.hpp

 #pragma once
-#include "ConstantTensorDescriptor.cuh"
+#include "ConstantTensorDescriptor.hip.hpp"
 
 // optimized for scenario if p_in, p_wei, p_out are in register
 template <class Float, class InDesc, class WeiDesc, class OutDesc>