Initial project scaffold

kernels/cuda/CMakeLists.txt

@@ -0,0 +1,29 @@
+cmake_minimum_required(VERSION 3.25)
+project(kernel_lab LANGUAGES CXX CUDA)
+
+set(CMAKE_CXX_STANDARD 17)
+set(CMAKE_CUDA_STANDARD 17)
+set(CMAKE_POSITION_INDEPENDENT_CODE ON)
+
+if(NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
+  set(CMAKE_CUDA_ARCHITECTURES 120 CACHE STRING "Target CUDA architectures")
+endif()
+
+find_package(Torch REQUIRED)
+
+add_library(kernel_lab_extension SHARED
+  binding/binding.cpp
+  src/vector_add.cu
+  src/row_softmax.cu
+  src/tiled_matmul.cu
+  src/online_softmax.cu
+  src/flash_attention_fwd.cu
+)
+
+target_include_directories(kernel_lab_extension PRIVATE include)
+target_link_libraries(kernel_lab_extension PRIVATE "${TORCH_LIBRARIES}")
+target_compile_features(kernel_lab_extension PRIVATE cxx_std_17)
+set_target_properties(kernel_lab_extension PROPERTIES
+  PREFIX ""
+  CUDA_SEPARABLE_COMPILATION ON
+)