Fluid bytecode c++ executor. #10220
Fluid bytecode c++ executor. #10220
Conversation
gongweibao
requested review from
jacquesqiao,
JiayiFeng,
typhoonzero and
panyx0718
| bool read_from_file(const std::string& file, char** buf, int64_t* buf_len) { | ||
| FILE* f = fopen(file.c_str(), "rb"); | ||
| if (NULL == f) { | ||
| printf("open %s error\n", file.c_str()); |
There was a problem hiding this comment.
fprintf(stderr? or LOG(ERROR)?
There was a problem hiding this comment.
change to fprintf(stderr.
Thanks.
|
|
||
| using namespace paddle; // NOLINT | ||
|
|
||
| framework::ProgramDesc* load_desc(const std::string& file) { |
| return (new framework::ProgramDesc(proto)); | ||
| } | ||
|
|
||
| DEFINE_string(start_up_proto, "", "start up proto file"); |
| std::transform(place_str.begin(), place_str.end(), place_str.begin(), | ||
| [](unsigned char ch) { return toupper(ch); }); | ||
|
|
||
| framework::Executor* exe = nullptr; |
| @@ -0,0 +1,126 @@ | |||
| // Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved. | |||
There was a problem hiding this comment.
should this file by call execute_program_desc?
| @@ -333,6 +335,12 @@ def run(self, | |||
| fetch_var_name=fetch_var_name) | |||
|
|
|||
| self._feed_data(program, feed, feed_var_name, scope) | |||
|
|
|||
| # TODO(gongwb): does a program should be saved in run function? | |||
| if len(save_program_to_file) > 0: | |||
There was a problem hiding this comment.
I'm not sure we should do the save here...
There was a problem hiding this comment.
The programdesc is modified in run function.And sometimes a user can't get chance to save it.
There was a problem hiding this comment.
Maybe run() shouldn't modify the programdesc? Or we should save after run is called?
|
Agree to put this to |
| @@ -277,7 +277,8 @@ def run(self, | |||
| fetch_var_name='fetch', | |||
| scope=None, | |||
| return_numpy=True, | |||
| use_program_cache=False): | |||
| use_program_cache=False, | |||
| save_program_to_file=""): | |||
There was a problem hiding this comment.
I still don't want to put "save_program_to_file" as a argument to Executor. Saving program should be in a separate API, such as "save_inference_model".
|
I build debug-version Python and so close this PR. |
Fix #10221