Partitioned execution of TI DL network
authorYuan Zhao <yuanzhao@ti.com>
Wed, 9 May 2018 22:31:04 +0000 (17:31 -0500)
committerYuan Zhao <yuanzhao@ti.com>
Thu, 10 May 2018 20:55:52 +0000 (15:55 -0500)
commit27db5c36c95b9d14ac24d2e84ff7c7c03b6ba13d
tree49cb5db96f622f6ed5db75dbee390db623ca49f3
parent1f017cf9bc3a3f173a9677d075aee8454c216400
Partitioned execution of TI DL network

- Enable pipelined and partitioned execution of TIDL network
  by setting the same ArgInfo:
  - as the output of ExecutionObject for first  layersGroup
  - as the input  of ExecutionObject for second layersGroup
  See ssd_multibox for example.
- Enable internal input mode that intermediate results between
  different layersGroup no longer need to be saved in user
  application.
- Fixes for correct partitioned execution of TIDL network,
  e.g. dataQ needs to be passed between layersGroups
- MCT-969, MCT-974
examples/make.common
examples/ssd_multibox/main.cpp
tinn_api/inc/configuration.h
tinn_api/inc/execution_object.h
tinn_api/inc/executor.h
tinn_api/src/configuration.cpp
tinn_api/src/execution_object.cpp
tinn_api/src/executor.cpp
tinn_api/src/ocl_device.cpp