Add option to enable max ILP scheduling strategy

CMakeLists.txt

@@ -29,6 +29,8 @@ if(USE_BITINT_EXTENSION_INT4)
     message("CK compiled with USE_BITINT_EXTENSION_INT4 set to ${USE_BITINT_EXTENSION_INT4}")
 endif()
 
+option(USE_MAX_ILP_SCHED_STRATEGY, "Whether to enable scheduling strategy to maximize ILP for a single wave." OFF)
+
 ## Threads
 set(THREADS_PREFER_PTHREAD_FLAG ON)
 find_package(Threads REQUIRED)

library/src/tensor_operation_instance/gpu/CMakeLists.txt

@@ -3,6 +3,10 @@ function(add_instance_library INSTANCE_NAME)
     add_library(${INSTANCE_NAME} OBJECT ${ARGN})
     target_compile_features(${INSTANCE_NAME} PUBLIC)
     set_target_properties(${INSTANCE_NAME} PROPERTIES POSITION_INDEPENDENT_CODE ON)
+    if(USE_MAX_ILP_SCHED_STRATEGY)
+        set_target_properties(${INSTANCE_NAME} PROPERTIES COMPILE_FLAGS "-mllvm -amdgpu-enable-max-ilp-scheduling-strategy")
+    endif(USE_MAX_ILP_SCHED_STRATEGY)
+
     clang_tidy_check(${INSTANCE_NAME})
 endfunction(add_instance_library INSTANCE_NAME)