[amdgpu] Part1 add codegen (taichi-dev#6469)

Issue: taichi-dev#6434

### Brief Summary
This part contains `CHI IR->LLVM IR` part. Similar to part0, the
skeleton of this part is similar to `cuda`. Some important notes are
following.
1. `ocml` is the math library like 'libdevice' in `nvidia`. It provides
most of the functions we need. Remains such as `sgn`, and `abs` have
been completed by hand codes
2. The kernel parameter passing part is different from the `CUDA`.
`extra_args` is the only way that could be accepted by `device kernel.`
3. still uses jargon from `nvidia`.(e.g. `sm`, `block` and `grid`)

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

cmake/TaichiCore.cmake

@@ -241,6 +241,9 @@ if(TI_WITH_LLVM)
     if (TI_WITH_AMDGPU)
         llvm_map_components_to_libnames(llvm_amdgpu_libs AMDGPU)
         add_subdirectory(taichi/rhi/amdgpu)
+        add_subdirectory(taichi/codegen/amdgpu)
+
+        target_link_libraries(${CORE_LIBRARY_NAME} PRIVATE amdgpu_codegen)
         target_link_libraries(${CORE_LIBRARY_NAME} PRIVATE amdgpu_rhi)
     endif()
 

taichi/codegen/amdgpu/CMakeLists.txt

@@ -0,0 +1,17 @@
+# ./taichi/codegen/amdgpu/CMakeLists.txt
+
+add_library(amdgpu_codegen)
+target_sources(amdgpu_codegen
+  PRIVATE
+    codegen_amdgpu.cpp
+  )
+
+target_include_directories(amdgpu_codegen
+  PRIVATE
+    ${PROJECT_SOURCE_DIR}
+    ${PROJECT_SOURCE_DIR}/external/eigen
+    ${LLVM_INCLUDE_DIRS}
+  )
+
+target_link_libraries(amdgpu_codegen PRIVATE taichi_util)
+# target_link_libraries(amdgpu_codegen PRIVATE amdgpu_runtime)