experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Improve help documentation for split flags.

Browse Source 
Make it easier to locally find out how to pass flags to the different backends.

PiperOrigin-RevId: 357281536
Change-Id: I4fbdf200f8f65f5aec1588f5d7285dc8622549cd
This commit is contained in:
Jacques Pienaar 2021-02-12 15:38:46 -08:00 committed by TensorFlower Gardener
parent 156b5ce928
commit be0ec80ce4
2 changed files with 12 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
tensorflow/compiler/mlir/init_mlir.cc
View File

@ -17,6 +17,17 @@ limitations under the License.
#include "tensorflow/core/platform/init_main.h"
static llvm::cl::extrahelp FlagSplittingHelp(R"(
The command line parsing is split between the two flag parsing libraries used by
TensorFlow and LLVM:
* Flags before the first '--' are parsed by tensorflow::InitMain while those
post are parsed by LLVM's command line parser.
* If there is no separator, then no flags are parsed by InitMain and only
LLVM command line parser used.
The above help options reported are for LLVM's parser, run with `--help --` for
TensorFlow's help.
)");
namespace tensorflow {
InitMlir::InitMlir(int *argc, char ***argv) : init_llvm_(*argc, *argv) {

3
tensorflow/compiler/mlir/init_mlir.h
View File

@ -21,12 +21,11 @@ limitations under the License.
namespace tensorflow {
// Initializer to perform both InitLLVM and TF"s InitMain initialization.
// Initializer to perform both InitLLVM and TF's InitMain initialization.
// InitMain also performs flag parsing and '--' is used to separate flags passed
// to it: Flags before the first '--' are parsed by InitMain and argc and argv
// progressed to the flags post. If there is no separator, then no flags are
// parsed by InitMain and argc/argv left unadjusted.
// TODO(jpienaar): The way help flag is handled could be improved.
class InitMlir {
public:
InitMlir(int *argc, char ***argv);