beam_search_op cannot run on GPU

[ktlichkid]

For now, beam_search_op will cause segmentation fault on GPU, as current implementation of this op has no kernel and doesn't support memory auto-transform.

[ktlichkid]

I0418 09:20:34.403944 16261 executor.cc:334] CUDAPlace(0) Op(beam_search), inputs:{ids[top_k_0.tmp_1[2, 50]({{0, 1, 2}{0, 1, 2}})], pre_ids[array_read_0.tmp_0[2, 1]({{0, 1, 2}{0, 1, 2}})], scores[top_k_0.tmp_0[2, 50]({{0, 1, 2}{0, 1, 2}})]}, outputs:{selected_ids[beam_search_0.tmp_10], selected_scores[beam_search_0.tmp_00]}.
*** Aborted at 1524043234 (unix time) try "date -d @1524043234" if you are using GNU date ***
PC: @ 0x0 (unknown)
*** SIGSEGV (@0x7f729a5ccb40) received by PID 16261 (TID 0x7f75f7ca4700) from PID 18446744072004356928; stack trace: ***
@ 0x7f75f7881390 (unknown)
@ 0x7f7560975a0b paddle::operators::BeamSearch::NextItemSet()
@ 0x7f75609760b6 paddle::operators::BeamSearch::SelectTopBeamSizeItems()
@ 0x7f7560976adf paddle::operators::BeamSearch::operator()()
@ 0x7f7560979bf2 paddle::operators::BeamSearchOp::RunImpl()
@ 0x7f756062e76e paddle::framework::Executor::RunPreparedContext()
@ 0x7f756062fb12 paddle::framework::Executor::Run()
@ 0x7f7560bd7b8b paddle::operators::WhileOp::RunImpl()
@ 0x7f756062e76e paddle::framework::Executor::RunPreparedContext()
@ 0x7f756062fb12 paddle::framework::Executor::Run()
@ 0x7f756059ea83 ZZN8pybind1112cpp_function10initializeIZNS0_C4IvN6paddle9framework8ExecutorEIRKNS4_11ProgramDescEPNS4_5ScopeEibbEINS_4nameENS_9is_methodENS_7siblingEEEEMT0_FT_DpT1_EDpRKT2_EUlPS5_S8_SA_ibbE_vISO_S8_SA_ibbEISB_SC_SD_EEEvOSF_PFSE_SH_ESN_ENUlRNS_6detail13function_callEE1_4_FUNESV
@ 0x7f756059b644 pybind11::cpp_function::dispatcher()
@ 0x4c37ed PyEval_EvalFrameEx
@ 0x4b9ab6 PyEval_EvalCodeEx
@ 0x4c16e7 PyEval_EvalFrameEx
@ 0x4c136f PyEval_EvalFrameEx
@ 0x4b9ab6 PyEval_EvalCodeEx
@ 0x4eb30f (unknown)
@ 0x4e5422 PyRun_FileExFlags
@ 0x4e3cd6 PyRun_SimpleFileExFlags
@ 0x493ae2 Py_Main
@ 0x7f75f74c6830 __libc_start_main
@ 0x4933e9 _start
@ 0x0 (unknown)
Segmentation fault