flash_c_api 放在third_party目录下

编译指令：
source /opt/dtk/env.sh

source /usr/local/bin/fastpt -T

export CPLUS_INCLUDE_PATH=/opt/dtk/roctracer/include:$CPLUS_INCLUDE_PATH
export AMDGPU_TARGETS="gfx906;gfx926;gfx928;gfx936"
export FASTPT_USE_ASM=1

CXX=hipcc CC=hipcc python setup.py bdist_wheel