Fix build errors (except PyTorch depedency)

example/ck_tile/18_paged_attention/CMakeLists.txt

@@ -5,4 +5,14 @@ string(REGEX REPLACE "^[0-9]+_" "" TRIMMED_DIR_NAME "${DIR_NAME}")
 # add prefix "tile_example_" to the processed directory name
 set(EXAMPLE_NAME "tile_example_${TRIMMED_DIR_NAME}")
 
-add_executable(${EXAMPLE_NAME} EXCLUDE_FROM_ALL paged_attention.cpp)
\ No newline at end of file
+add_executable(${EXAMPLE_NAME} EXCLUDE_FROM_ALL paged_attention.cpp attention.cpp)
+target_include_directories(${EXAMPLE_NAME} AFTER PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/include)
+target_compile_definitions(${EXAMPLE_NAME} PRIVATE USE_ROCM)
+target_compile_options(${EXAMPLE_NAME} PRIVATE
+  -Wno-undefined-reinterpret-cast
+  -Wno-unused-variable
+  -Wno-unused-parameter
+  -Wno-old-style-cast
+  -Wno-deprecated-copy
+  -Wno-shadow
+)
\ No newline at end of file

example/ck_tile/18_paged_attention/attention.cu → example/ck_tile/18_paged_attention/attention.cpp

@@ -14,15 +14,13 @@
  * limitations under the License.
  */
 
-#include <torch/all.h>
-#include <ATen/cuda/CUDAContext.h>
-#include <c10/cuda/CUDAGuard.h>
 #include <hip/hip_bf16.h>
 #include "cuda_compat.h"
 
 #include <algorithm>
-#include "../attention/dtype_fp8.cuh"
-#include "../quantization/fp8/amd/quant_utils.cuh"
+#include <cfloat>
+#include "attention/dtype_fp8.cuh"
+#include "quantization/fp8/amd/quant_utils.cuh"
 
 #if defined(__HIPCC__) && (defined(__gfx90a__) || defined(__gfx940__) || \
                            defined(__gfx941__) || defined(__gfx942__))